recent advances in electronics and ...recent advances in electronics and communication systems...

RECENT ADVANCES in ELECTRONICS and COMMUNICATION SYSTEMS

Proceedings of the 2013 International Conference on Electronics and Communication Systems

(ECS 2013)

Rhodes Island, Greece July 16‐19, 2013


Proceedings of the 2013 International Conference on Electronics and Communication Systems (ECS 2013)


Copyright © 2013, by the editors

All the copyright of the present book belongs to the editors. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior written permission of the editors.

All papers of the present volume were peer reviewed by no less that two independent reviewers. Acceptance was granted when both reviewers' recommendations were positive.

ISBN: 978‐1‐61804‐201‐9



(ECS 2013)


Organizing Committee General Chairs (EDITORS)

Professor Charles A. Long Professor Emeritus University of Wisconsin Stevens Point, Wisconsin, USA

Professor Nikos E. Mastorakis Military Institutes of University Education (ASEI) Hellenic Naval Academy Sector of Electrical Engineering and Computer Science Piraeus, Greece ‐also with‐ Technical University of Sofia 1000 Sofia, Bulgaria

Professor Valeri Mladenov Technical University of Sofia 1000 Sofia, Bulgaria

Senior Program Chair

Professor Philippe Dondon ENSEIRB Rue A Schweitzer 33400 Talence France

Program Chairs

Professor Filippo Neri Dipartimento di Informatica e Sistemistica University of Naples "Federico II" Naples, Italy

Professor Hamid Reza Karimi Department of Engineering Faculty of Engineering and Science University of Agder, N‐4898 Grimstad Norway

Professor Sandra Sendra Instituto de Inv. para la Gestión Integrada de Zonas Costeras (IGIC) Universidad Politécnica de Valencia Spain

Tutorials Chair

Professor Pradip Majumdar Department of Mechanical Engineering Northern Illinois University Dekalb, Illinois, USA

Special Session Chair

Professor Pavel Varacha Tomas Bata University in Zlin Faculty of Applied Informatics Department of Informatics and Artificial Intelligence Zlin, Czech Republic

Workshops Chair

Professor Ryszard S. Choras Institute of Telecommunications University of Technology & Life Sciences Bydgoszcz, Poland

Local Organizing Chair

Assistant Professor Klimis Ntalianis, Tech. Educ. Inst. of Athens (TEI), Athens, Greece

Publication Chair

Professor Gen Qi Xu Department of Mathematics Tianjin University Tianjin, China

Publicity Committee

Professor Reinhard Neck Department of Economics Klagenfurt University Klagenfurt, Austria

Professor Myriam Lazard Institut Superieur d' Ingenierie de la Conception Saint Die, France

International Liaisons

Professor Ka‐Lok Ng Department of Bioinformatics Asia University Taichung, Taiwan

Professor Olga Martin Applied Sciences Faculty Politehnica University of Bucharest Romania

Professor Vincenzo Niola Departement of Mechanical Engineering for Energetics University of Naples "Federico II" Naples, Italy

Professor Eduardo Mario Dias Electrical Energy and Automation Engineering Department Escola Politecnica da Universidade de Sao Paulo Brazil

Steering Committee Professor Aida Bulucea, University of Craiova, Romania Professor Dana Simian, Univ. of Sibiu, Sibiu, Romania Professor Zoran Bojkovic, Univ. of Belgrade, Serbia Professor Metin Demiralp, Istanbul Technical University, Turkey Professor F. V. Topalis, Nat. Tech. Univ. of Athens, Greece Professor Imre Rudas, Obuda University, Budapest, Hungary

Program Committee Prof. Lotfi Zadeh (IEEE Fellow,University of Berkeley, USA) Prof. Leon Chua (IEEE Fellow,University of Berkeley, USA) Prof. Michio Sugeno (RIKEN Brain Science Institute (RIKEN BSI), Japan) Prof. Dimitri Bertsekas (IEEE Fellow, MIT, USA) Prof. Demetri Terzopoulos (IEEE Fellow, ACM Fellow, UCLA, USA) Prof. Georgios B. Giannakis (IEEE Fellow, University of Minnesota, USA) Prof. George Vachtsevanos (Georgia Institute of Technology, USA) Prof. Abraham Bers (IEEE Fellow, MIT, USA) Prof. David Staelin (IEEE Fellow, MIT, USA) Prof. Brian Barsky (IEEE Fellow, University of Berkeley, USA) Prof. Aggelos Katsaggelos (IEEE Fellow, Northwestern University, USA) Prof. Josef Sifakis (Turing Award 2007, CNRS/Verimag, France) Prof. Hisashi Kobayashi (Princeton University, USA) Prof. Kinshuk (Fellow IEEE, Massey Univ. New Zeland), Prof. Leonid Kazovsky (Stanford University, USA) Prof. Narsingh Deo (IEEE Fellow, ACM Fellow, University of Central Florida, USA) Prof. Kamisetty Rao (Fellow IEEE, Univ. of Texas at Arlington,USA) Prof. Anastassios Venetsanopoulos (Fellow IEEE, University of Toronto, Canada) Prof. Steven Collicott (Purdue University, West Lafayette, IN, USA) Prof. Nikolaos Paragios (Ecole Centrale Paris, France) Prof. Nikolaos G. Bourbakis (IEEE Fellow, Wright State University, USA) Prof. Stamatios Kartalopoulos (IEEE Fellow, University of Oklahoma, USA) Prof. Irwin Sandberg (IEEE Fellow, University of Texas at Austin, USA), Prof. Michael Sebek (IEEE Fellow, Czech Technical University in Prague, Czech Republic) Prof. Hashem Akbari (University of California, Berkeley, USA) Prof. Yuriy S. Shmaliy, (IEEE Fellow, The University of Guanajuato, Mexico) Prof. Lei Xu (IEEE Fellow, Chinese University of Hong Kong, Hong Kong) Prof. Paul E. Dimotakis (California Institute of Technology Pasadena, USA) Prof. M. Pelikan (UMSL, USA) Prof. Patrick Wang (MIT, USA) Prof. Wasfy B Mikhael (IEEE Fellow, University of Central Florida Orlando,USA) Prof. Sunil Das (IEEE Fellow, University of Ottawa, Canada) Prof. Panos Pardalos (University of Florida, USA) Prof. Nikolaos D. Katopodes (University of Michigan, USA) Prof. Bimal K. Bose (Life Fellow of IEEE, University of Tennessee, Knoxville, USA) Prof. Janusz Kacprzyk (IEEE Fellow, Polish Academy of Sciences, Poland) Prof. Sidney Burrus (IEEE Fellow, Rice University, USA) Prof. Biswa N. Datta (IEEE Fellow, Northern Illinois University, USA) Prof. Mihai Putinar (University of California at Santa Barbara, USA) Prof. Wlodzislaw Duch (Nicolaus Copernicus University, Poland) Prof. Tadeusz Kaczorek (IEEE Fellow, Warsaw University of Tehcnology, Poland) Prof. Michael N. Katehakis (Rutgers, The State University of New Jersey, USA) Prof. Pan Agathoklis (Univ. of Victoria, Canada) Prof. P. Demokritou (Harvard University, USA) Prof. P. Razelos (Columbia University, USA) Dr. Subhas C. Misra (Harvard University, USA) Prof. Martin van den Toorn (Delft University of Technology, The Netherlands) Prof. Malcolm J. Crocker (Distinguished University Prof., Auburn University,USA) Prof. S. Dafermos (Brown University, USA) Prof. Urszula Ledzewicz, Southern Illinois University , USA. Prof. Dimitri Kazakos, Dean, (Texas Southern University, USA) Prof. Ronald Yager (Iona College, USA) Prof. Athanassios Manikas (Imperial College, London, UK)

Prof. Keith L. Clark (Imperial College, London, UK) Prof. Argyris Varonides (Univ. of Scranton, USA) Prof. S. Furfari (Direction Generale Energie et Transports, Brussels, EU) Prof. Constantin Udriste, University Politehnica of Bucharest , ROMANIA Dr. Michelle Luke (Univ. Berkeley, USA) Prof. Patrice Brault (Univ. Paris‐sud, France) Dr. Christos E. Vasios (MIT, USA) Prof. Jim Cunningham (Imperial College London, UK) Prof. Philippe Ben‐Abdallah (Ecole Polytechnique de l'Universite de Nantes, France) Prof. Photios Anninos (Medical School of Thrace, Greece) Prof. Ichiro Hagiwara, (Tokyo Institute of Technology, Japan) Prof. Metin Demiralp ( Istanbul Technical University / Turkish Academy of Sciences, Istanbul, Turkey) Prof. Andris Buikis (Latvian Academy of Science. Latvia) Prof. Akshai Aggarwal (University of Windsor, Canada) Prof. George Vachtsevanos (Georgia Institute of Technology, USA) Prof. Ulrich Albrecht (Auburn University, USA) Prof. Imre J. Rudas (Obuda University, Hungary) Prof. Alexey L Sadovski (IEEE Fellow, Texas A&M University, USA) Prof. Amedeo Andreotti (University of Naples, Italy) Prof. Ryszard S. Choras (University of Technology and Life Sciences Bydgoszcz, Poland) Prof. Remi Leandre (Universite de Bourgogne, Dijon, France) Prof. Moustapha Diaby (University of Connecticut, USA) Prof. Brian McCartin (New York University, USA) Prof. Elias C. Aifantis (Aristotle Univ. of Thessaloniki, Greece) Prof. Anastasios Lyrintzis (Purdue University, USA) Prof. Charles Long (Prof. Emeritus University of Wisconsin, USA) Prof. Marvin Goldstein (NASA Glenn Research Center, USA) Prof. Costin Cepisca (University POLITEHNICA of Bucharest, Romania) Prof. Kleanthis Psarris (University of Texas at San Antonio, USA) Prof. Ron Goldman (Rice University, USA) Prof. Ioannis A. Kakadiaris (University of Houston, USA) Prof. Richard Tapia (Rice University, USA) Prof. F.‐K. Benra (University of Duisburg‐Essen, Germany) Prof. Milivoje M. Kostic (Northern Illinois University, USA) Prof. Helmut Jaberg (University of Technology Graz, Austria) Prof. Ardeshir Anjomani (The University of Texas at Arlington, USA) Prof. Heinz Ulbrich (Technical University Munich, Germany) Prof. Reinhard Leithner (Technical University Braunschweig, Germany) Prof. Elbrous M. Jafarov (Istanbul Technical University, Turkey) Prof. M. Ehsani (Texas A&M University, USA) Prof. Sesh Commuri (University of Oklahoma, USA) Prof. Nicolas Galanis (Universite de Sherbrooke, Canada) Prof. S. H. Sohrab (Northwestern University, USA) Prof. Rui J. P. de Figueiredo (University of California, USA) Prof. Valeri Mladenov (Technical University of Sofia, Bulgaria) Prof. Hiroshi Sakaki (Meisei University, Tokyo, Japan) Prof. Zoran S. Bojkovic (Technical University of Belgrade, Serbia) Prof. K. D. Klaes, (Head of the EPS Support Science Team in the MET Division at EUMETSAT, France) Prof. Emira Maljevic (Technical University of Belgrade, Serbia) Prof. Kazuhiko Tsuda (University of Tsukuba, Tokyo, Japan) Prof. Milan Stork (University of West Bohemia , Czech Republic) Prof. C. G. Helmis (University of Athens, Greece) Prof. Lajos Barna (Budapest University of Technology and Economics, Hungary) Prof. Nobuoki Mano (Meisei University, Tokyo, Japan)

Prof. Nobuo Nakajima (The University of Electro‐Communications, Tokyo, Japan) Prof. Victor‐Emil Neagoe (Polytechnic University of Bucharest, Romania) Prof. E. Protonotarios (National Technical University of Athens, Greece) Prof. P. Vanderstraeten (Brussels Institute for Environmental Management, Belgium) Prof. Annaliese Bischoff (University of Massachusetts, Amherst, USA) Prof. Virgil Tiponut (Politehnica University of Timisoara, Romania) Prof. Andrei Kolyshkin (Riga Technical University, Latvia) Prof. Fumiaki Imado (Shinshu University, Japan) Prof. Sotirios G. Ziavras (New Jersey Institute of Technology, USA) Prof. Constantin Volosencu (Politehnica University of Timisoara, Romania) Prof. Marc A. Rosen (University of Ontario Institute of Technology, Canada) Prof. Alexander Zemliak (Puebla Autonomous University, Mexico) Prof. Thomas M. Gatton (National University, San Diego, USA) Prof. Leonardo Pagnotta (University of Calabria, Italy) Prof. Yan Wu (Georgia Southern University, USA) Prof. Daniel N. Riahi (University of Texas‐Pan American, USA) Prof. Alexander Grebennikov (Autonomous University of Puebla, Mexico) Prof. Bennie F. L. Ward (Baylor University, TX, USA) Prof. Guennadi A. Kouzaev (Norwegian University of Science and Technology, Norway) Prof. Eugene Kindler (University of Ostrava, Czech Republic) Prof. Geoff Skinner (The University of Newcastle, Australia) Prof. Hamido Fujita (Iwate Prefectural University(IPU), Japan) Prof. Francesco Muzi (University of L'Aquila, Italy) Prof. Les M. Sztandera (Philadelphia University, USA) Prof. Claudio Rossi (University of Siena, Italy) Prof. Christopher J. Koroneos (Aristotle University of Thessaloniki, Greece) Prof. Sergey B. Leonov (Joint Institute for High Temperature Russian Academy of Science, Russia) Prof. Arpad A. Fay (University of Miskolc, Hungary) Prof. Lili He (San Jose State University, USA) Prof. M. Nasseh Tabrizi (East Carolina University, USA) Prof. Alaa Eldin Fahmy (University Of Calgary, Canada) Prof. Ion Carstea (University of Craiova, Romania) Prof. Paul Dan Cristea (University "Politehnica" of Bucharest, Romania) Prof. Gh. Pascovici (University of Koeln, Germany) Prof. Pier Paolo Delsanto (Politecnico of Torino, Italy) Prof. Radu Munteanu (Rector of the Technical University of Cluj‐Napoca, Romania) Prof. Ioan Dumitrache (Politehnica University of Bucharest, Romania) Prof. Corneliu Lazar (Technical University Gh.Asachi Iasi, Romania) Prof. Nicola Pitrone (Universita degli Studi Catania, Italia) Prof. Miquel Salgot (University of Barcelona, Spain) Prof. Amaury A. Caballero (Florida International University, USA) Prof. Maria I. Garcia‐Planas (Universitat Politecnica de Catalunya, Spain) Prof. Petar Popivanov (Bulgarian Academy of Sciences, Bulgaria) Prof. Alexander Gegov (University of Portsmouth, UK) Prof. Lin Feng (Nanyang Technological University, Singapore) Prof. Colin Fyfe (University of the West of Scotland, UK) Prof. Zhaohui Luo (Univ of London, UK) Prof. Mikhail Itskov (RWTH Aachen University, Germany) Prof. George G. Tsypkin (Russian Academy of Sciences, Russia) Prof. Wolfgang Wenzel (Institute for Nanotechnology, Germany) Prof. Weilian Su (Naval Postgraduate School, USA) Prof. Phillip G. Bradford (The University of Alabama, USA) Prof. Ray Hefferlin (Southern Adventist University, TN, USA) Prof. Gabriella Bognar (University of Miskolc, Hungary)

Prof. Hamid Abachi (Monash University, Australia) Prof. Karlheinz Spindler (Fachhochschule Wiesbaden, Germany) Prof. Josef Boercsoek (Universitat Kassel, Germany) Prof. Eyad H. Abed (University of Maryland, Maryland, USA) Prof. F. Castanie (TeSA, Toulouse, France) Prof. Robert K. L. Gay (Nanyang Technological University, Singapore) Prof. Andrzej Ordys (Kingston University, UK) Prof. Harris Catrakis (Univ of California Irvine, USA) Prof. T Bott (The University of Birmingham, UK) Prof. Petr Filip (Institute of Hydrodynamics, Prague, Czech Republic) Prof. T.‐W. Lee (Arizona State University, AZ, USA) Prof. Le Yi Wang (Wayne State University, Detroit, USA) Prof. George Stavrakakis (Technical University of Crete, Greece) Prof. John K. Galiotos (Houston Community College, USA) Prof. M. Petrakis (National Observatory of Athens, Greece) Prof. Philippe Dondon (ENSEIRB, Talence, France) Prof. Dalibor Biolek (Brno University of Technology, Czech Republic) Prof. Oleksander Markovskyy (National Technical University of Ukraine, Ukraine) Prof. Suresh P. Sethi (University of Texas at Dallas, USA) Prof. Hartmut Hillmer(University of Kassel, Germany) Prof. Bram Van Putten (Wageningen University, The Netherlands) Prof. Alexander Iomin (Technion ‐ Israel Institute of Technology, Israel) Prof. Roberto San Jose (Technical University of Madrid, Spain) Prof. Minvydas Ragulskis (Kaunas University of Technology, Lithuania) Prof. Arun Kulkarni (The University of Texas at Tyler, USA) Prof. Joydeep Mitra (New Mexico State University, USA) Prof. Vincenzo Niola (University of Naples Federico II, Italy) Prof. Ion Chryssoverghi (National Technical University of Athens, Greece) Prof. Dr. Aydin Akan (Istanbul University, Turkey) Prof. Sarka Necasova (Academy of Sciences, Prague, Czech Republic) Prof. C. D. Memos (National Technical University of Athens, Greece) Prof. S. Y. Chen, (Zhejiang University of Technology, China and University of Hamburg, Germany) Prof. Duc Nguyen (Old Dominion University, Norfolk, USA) Prof. Tuan Pham (James Cook University, Townsville, Australia) Prof. Habil M. Patzold (Agder University College,Norway ) Prof. Jiri Klima (Technical Faculty of CZU in Prague, Czech Republic) Prof. Rossella Cancelliere (University of Torino, Italy) Prof. L.Kohout (Florida State University, Tallahassee, Florida, USA) Prof. D' Attelis (Univ. Buenos Ayres, Argentina) Prof. Dr‐Eng. Christian Bouquegneau (Faculty Polytechnique de Mons, Belgium) Prof. Wladyslaw Mielczarski (Technical University of Lodz, Poland) Prof. Ibrahim Hassan (Concordia University, Montreal, Quebec, Canada) Prof. Stavros J.Baloyannis (Medical School, Aristotle University of Thessaloniki, Greece) Prof. James F. Frenzel (University of Idaho, USA) Prof. Mirko Novak (Czech Technical University in Prague,Czech Republic) Prof. Zdenek Votruba (Czech Technical University in Prague,Czech Republic) Prof. Vilem Srovnal,(Technical University of Ostrava, Czech Republic) Prof. J. M. Giron‐Sierra (Universidad Complutense de Madrid, Spain) Prof. Zeljko Panian (University of Zagreb, Croatia) Prof. Walter Dosch (University of Luebeck, Germany) Prof. Rudolf Freund (Vienna University of Technology, Austria) Prof. Erich Schmidt (Vienna University of Technology, Austria) Prof. Alessandro Genco (University of Palermo, Italy) Prof. Martin Lopez Morales (Technical University of Monterey, Mexico)

Prof. Ralph W. Oberste‐Vorth (Marshall University, USA) Prof. Vladimir Damgov (Bulgarian Academy of Sciences, Bulgaria) Prof. Menelaos Karanasos (Brunel University, UK) Prof. P.Borne (Ecole Central de Lille, France)

Additional Reviewers Vaclav Skala Ahmad Azarnik Keffala Mohamed Rochdi Valentine Baranov Hugo Rodrigues Dumitru Cazacu Cornelia Aida Bulucea Jiri Hrebicek Sanjib Kumar Datta Maria Dobritoiu Umer Asgher Anna Adamik Jorge Magalhaes Mendes Gilberto Perez Lechuga Jelena Vasiljevic Deolinda Dias Rasteiro Ozlem Turker Bayrak Ahmed El Kashlan Najat Ouaaline Armando Silva Afonso Klimis Ntalianis Igor Astrov Filippo Neri Valeriu Prepelita Mihaiela Iliescu Calin Ciufudean Milan Stork Anel Tanovic Ioannis Gonos Lamberto Tronchin Dan Lacrama Sudhir Dawra Kala Ouarda Shahram Javadi Babak Bashari Rad Yilun Shang Alina Badulescu Corina Carranca Christos Volos Elena Zaitseva Daniela Litan Farhad Mehran Ahmet Ertek Fernando Reinaldo Ribeiro Ayman Batisha Madalina Xenia Calbureanu Popescu Cornelia Gyorodi Dzenana Donko

Dario Ferreira Anabela Gomes Andrey N. Dmitriev Gabriel Badescu Catalin Popescu Vassiliki T. Kontargyri H. Kijima

Table of Contents

Keynote Lecture 1: Shifting the Expansion Point in Direct Power Series of Multivaried Functions

15

Metin Demiralp

Plenary Lecture 1: Intelligent Prediction of Vehicle Dynamics Using Structured Neural Networks

17

Stratis Kanarachos

Plenary Lecture 2: Multitime Solitons and Their Maple Animation 18

Constantin Udriste

Plenary Lecture 3: Invariant Subspaces and Structural Properties of 2‐D Control Systems 20

Valeriu Prepelita

Tutorial 1: Projective Geometry and Duality for Graphics, Games and Visualization 22

Vaclav Skala

Tutorial 2: Nonlinear Distortion in Wireless Transmitters 23

D. Budimir

Analytic Model of MLAR1 25

Yahia Hasan Jazyah

The Effect of GPS Error on MLAR1 36

Yahia Hasan Jazyah

Performance Evaluation of VoIP Codecs over Network Coding in Wireless Mesh Networks 43

Erik Pertovt, Kemal Alic, Aleš Švigelj, Mihael Mohorcic

Delay Factors Modelling for Real‐Time Traffic Information Systems 50

Marius Minea, Iulian Bădescu

Optimizing Cloud Security by Applying New Innovative Filter 56

Mehdi Darbandi

Studying Security Criteria’s of Cloud and VM Platforms and Present New Innovative Solution for It

65

Mehdi Darbandi

Joint Evaluation of I /Q Imbalance and Reconfigurable RF Filter Nonlinearity in LTE Transmitters

73

M. Bozic, A. Anastasijevic, K. Rabbi, N. Mohottige, D. Budimir


13

An Efficient Delay Estimation Model for High Speed VLSI Interconnects 77

M. Kavicharan, N. S. Murthy, N. Bheema Rao

Design of 8‐bit Dynamic CMOS Priority Resolvers Based on Active‐High and Active‐Low Logic

82

Preeti Panchal, C. Vinitha, Rashi Srivastava, P. Balasubramanian, N. E. Mastorakis

Study of the Cell Towers Radiation Levels in Residential Areas 87

Sabah Hawar Saeid

A Closed‐Form Delay Estimation Model for Current‐Mode High Speed VLSI Interconnects 91

M. Kavicharan, N. S. Murthy, N. Bheema Rao

AC‐DC & DC‐DC Converters for DC Motor Drives ‐ Review of Basic Topologies 96

G. Ch. Ioannidis, C. S. Psomopoulos, S. D. Kaminaris, P. Pachos, H. Villiotis, S. Tsiolis, P. Malatestas, G. A. Vokas, S. N. Manias

Secure Communication for Cognitive Networks Based on MIMO and Spreading LDPC Codes

104

Yang Xiao, Kaiyao Wang

Linear Precoding for the Downlink of Cognitive Radio Network 113

Pengpeng Lan, Yang Xiao, Jinfeng Kou

Cooperative Spatial Multiplexing for CR Users Sharing a Common Channel with Primary Users

118

Yang Xiao, Jinfeng Kou

A Spatial Coding Approach for MIMO Cognitive Radio Networks’ Bandwidth Sharing 123

Yanru Qiao, Yang Xiao

Optimizing Power Allocation in a Cellular DS/FFH‐CDMA System under Rayleigh Fading 127

P. Varzakas

Statistics of the Channel Capacity in a Cellular DS/FFH‐CDMA Rayleigh Fading System 131

P. Varzakas

Modeling the Value Chain with Object‐Valued Petri Nets 136

Jaroslav Zacek, Frantisek Hunka, Zdenek Melis

Generalized Net Model for Telecommunication Processes in Telecare Services 142Mikhail Matveev, Velin Andonov, Maria Milanova

Design of a Dynamic CMOS Incrementer/Decrementer and a Parallel Cascading Architecture 146

B. Archanadevi, V. Anbumani, T. Malathy, P. Balasubramanian, N. E. Mastorakis

Authors Index 151


14

Keynote Lecture 1

Shifting the Expansion Point in Direct Power Series of Multivaried Functions

Professor Metin Demiralp Istanbul Technical University

TURKEY E‐mail: [email protected]

Abstract: Taylor series play an important role in the representation of the analytic functions. They are in fact certain infinite linear combinations of some power functions which are linearly independent despite their functional dependence. This series are valid not only for univariate functions, their multivariate versions are also capable of representing the target function. However, the expressions of multivariate functions become complicated because of the high number of terms when the number of the independent variables increases. We generally need to use high number of indices and multiple sums for the representation. This may necessitate the multivariate algebra and its tools. On the other hand the use of the direct powers based on direct product, or Kronecker product at the remembrance of the relevant scientist, then everything can be handled just by using the ordinary linear algebra and its tools in a one index notation. The direct power series of a function of n independent variables, x1,..., xn can be given as follows

where the vector x has the elements from x1 to xn inclusive while Fj stands for a constant rectangular matrix of n×n type. The elements of Fj

j are related to the partial derivatives of the left hand side function f. The symbol is for the direct power which can be recursively defined as follows

To initialize this recursion we take the zeroth direct power of a vector just as the scalar 1 by following the well‐known convention of algebra. This produces the vector x at the first direct power and therefore contains n elements. The number of the elements in the direct product increases as the degree of the direct product increases. Hence, x 2, in other words, the direct square of the vector x, is composed of all possible binary products of the elements of x by considering different ordering of the same product factors separately even though they are same because of the commutativity. The number of the elements in this case is n2. These


15

considerations can be extended to the case of general direct product with, say, jth degree. Then all possible j‐factor products of the elements of x form the resulting vector whose number of elements is nj. The coefficient matrices Fjs are not unique because of the existence of same j‐factor products as the elements residing at different locations in the jth direct power of x. There are certain rectangular matrices under the action of which the jth direct power of x van‐ ishes. This brings the possibility of introducing certain flexibility parameters to the co‐ efficient matrices. These flexibilities can be used to give certain specific natures to the coefficient matrices. The replacement of x by (y + a) in (1) and the rearrangement of the resulting ex‐ pression in direct power series of the variable vector y produces another repreresentation for the same original function. However, now, all coefficient matrices become the function of the constant vector a. All derivatives now should be somehow related to the partial derivatives of the original function evaluated at not origin but the point characterized by a. This is apparently a shift of the expansion point and the construction is not so straight‐ forward as maybe considered as the very first glance because of the multivariate structure of the direct product operation. The noncommutativity in the representation may produce certain complications. However they can be handled by using some permutation matrices. Presentation will focus on these issues in certain detail as much as possible. Brief Biography of the Speaker: Metin Demiralp was born in Turkiye (Turkey) on 4 May 1948. His education from elementary school to university was entirely in Turkey. He got his BS, MS degrees and PhD from the same institution, ˙Istanbul Technical University. He was originally chemical engineer, however, through theoretical chemistry, applied mathematics, and computational science years he was mostly working on methodology for computational sciences and he is continuing to do so. He has a group (Group for Science and Methods of Computing) in Informatics Institute of ˙Istanbul Technical University (he is the founder of this institute). He collaborated with the Prof. Herschel A. Rabitz's group at Princeton University (NJ, USA) at summer and winter semester breaks during the period 1985‐2003 after his 14 month long postdoctoral visit to the same group in 1979‐1980. He was also (and still is) in collaboration with a neuroscience group at the Psychology Department in the University of Michigan at Ann Arbour in last three years (with certain publications in journals and proceedings). Metin Demiralp has more than 100 papers in well known and prestigious scientific journals, and, more than 230 contributions together with various keynote, plenary, and, tutorial talks to the proceedings of various international conferences. He gave many invited talks in various prestigious scientific meetings and academic institutions. He has a good scientific reputation in his country and he was one of the principal members of Turkish Academy of Sciences since 1994. He has resigned on June 2012 because of the governmental decree changing the structure of the academy and putting politicial influence possibility by bringing a member assignation system. Metin Demiralp is also a member of European Mathematical Society. He has also two important awards of turkish scientific establishments. The important recent foci in research areas of Metin Demiralp can be roughly listed as follows: Probabilistic Evolution Method in Explicit ODE Solutions and in Quantum and Liouville Mechanics, Fluctuation Expansions in Matrix Representations, High Dimensional Model Representations, Space Extension Methods, Data Processing via Multivariate Analytical Tools, Multivariate Numerical Integration via New Efficient Approaches, Matrix Decompositions, Multiway Array Decompositions, Enhanced Multivariate Product Representations, Quantum Optimal Control.


16

Plenary Lecture 1

Intelligent Prediction of Vehicle Dynamics Using Structured Neural Networks

Professor Stratis Kanarachos, Frederick University

Department of Mechanical Engineering Lemesos, Cyprus

E‐mail: [email protected]

Abstract: The majority of traffic accidents today are caused by human errors in judgment and driving reaction. With the advance of microelectronics driver assistance systems (DAS) have been developed and became one of the principal priorities for most vehicle manufacturers. Obstacle avoidance or collision avoidance systems fall into this category and have naturally addressed the interest of many researchers during the last decade. These systems warn the driver or intervene using the braking/steering system based on the output of a decision algorithm. This algorithm calculates at each time instant a threat factor which represents the probability to get involved in a crash in the near future. This paper is focused on the intelligent prediction of the vehicle dynamics in the near future using a structured neural network approach. A nonlinear vehicle model with four degrees has been utilized. As will be shown, by means of simulation in Matlab, it is possible to predict the response of the vehicle with minimal computational burden. Therefore, the threat assessment can be enhanced and the performance of the decision algorithm improved. False alarms which determine the driver acceptance as well as liability issues are largely dependent on its performance. Brief Biography of the Speaker: Dr S. Kanarachos holds a Diploma (5 years) and a PhD in Mechanical Engineering (2004) from the National Technical University of Athens (Hellas). He has worked as a consultant on product development (2001‐2005) and researcher at the Mechanical Engineering Department of NTUA. From 2005‐2007, he served the Department of Mechanical Engineering at Frederick Institute of Technology (Cyprus) as a Lecturer and from 2007‐2012 the Department of Mechanical Engineering at Frederick University (Cyprus) as Assistant Professor specializing in the fields of Computational Dynamics & Numerical Optimization. From 2012 and onwards he works at the Integrated Vehicle Safety expertise group of TNO (The Netherlands) and from 2013 he is the head of the Vehicle Dynamics group. His research interests include: flexible multibody dynamics, finite element & multibody dynamics code coupling, metamodeling using structured neural networks and model reduction. He is the author of15 publications in highly rated ISI journals and 50 in conference proceedings (of IEEE, WSEAS, etc.). He has participated as the principal researcher, coordinator or scientific supervisor in more than twenty research projects funded by National or European Framework Programmes (GSRT, RPF, FP6, FP7). He serves as a reviewer in several Scientific journals and conferences and he is member of the editorial board of the International Journal of Vehicle Systems Modelling and Testing. He has served as an evaluator of research projects for National & European Framework Programmes.


17

Plenary Lecture 2

Multitime Solitons and Their Maple Animation

Professor Constantin Udriste University Politehnica of Bucharest

Department of Mathematics Informatics Romania

E‐mail: [email protected] Abstract: The development of multitime PDE concepts is now in vogue in physics ( multitime Maxwell PDEs), in electronics (widely‐separated time scales, difference‐frequency time scales, etc) and in mathematics (multitime solitons, introduced and analyzed in our research group). Indeed, to handle frequency‐modulation effectively, it is necessary to use a novel concept, known as warped time, within a multitime partial differential equation framework. Generally, the purpose of a multi‐dimensional model is to represent efficiently phenomena including widely separated time scales (for example, control of composite systems via the multi time‐scale approach). An m‐soliton is a special wave whose temporal evolution is m‐dimensional. Of course, a multitime simulation requires special integrators and animations. For a two‐time soliton, we prefer to adopt Maple Animation in 3d, creating spectacular evolutions. Brief Biography of the Speaker: Important Career Positions: Emeritus Professor, Consultant Professor, Dean, Director, Chair, Full Professor 1990‐, University Politehnica of Bucharest, Department of Mathematics‐Informatics. Number of PhD Students: 25 in due time and 21 Doctors in Mathematics. Membership of Associations: AMS, 1987; Tensor Society, 1985; Balkan Society of Geometers, President, 1994; Publications: over 50 books; 300 papers; 300 communications. Honors: D. Hurmuzescu Prize, Romanian Academy, 1985; Award MEI, 1988; Correspondent Member, Academia Peloritana, Messina, 1997; Prize COPIRO ‐ 2000 for Exact Sciences; Premio Anassiloos International 2002, Arte Cultura Scienze, Italy; Titular Member, Academy of Romanian Scientists, 2007; Honorary Member, World Scientific and Engineering Academy and Society, 2008‐; Stefan Hepites Prize, Academy of Romania, 2010. Organizer: Chair‐Committee: American Conference on Applied Mathematics (Math '08) and Management, Marketing and Finances (MMF '08), Cambridge, Massachusetts, USA, March 24‐26, 2008. International Program Committee: The Applied Computing Conference (ACC‐08), Istanbul, Turkey, May 27‐30, 2008; The International Conference of Differential Geometry and Dynamical Systems (DGDS‐2009), October 8 ‐ 11, 2009, University Politehnica of Bucharest, Bucharest, Romania; European Computing Conference (ECC‐09), 115‐119, Tbilisi, Georgia, June


18

26‐28, 2009; The International Conference of Differential Geometry and Dynamical Systems ( DGDS‐2010 ), 25‐28 August 2010, University Politehnica of Bucharest , Romania. Fields of Interest: Differential Geometry, Optimizations on Riemannian Manifolds, Magnetic Dynamical Systems, Geometric Dynamics.


19

Plenary Lecture 3

Invariant Subspaces and Structural Properties of 2‐D Control Systems

Professor Valeriu Prepelita University Politehnica of Bucharest

Department of Mathematics Informatics Romania

E‐mail: [email protected] Abstract: Since its birth about sixty years ago, Systems Theory has developed into a scientific and engineering discipline connected with all aspects of modern society. At the beginning it was studied as Control Theory by mathematicians and engineers, but soon Systems Theory extended to the study and the applications of various domains such as economics, business, political science, sociology, medicine, biology, psychology, ecology etc. The past three decades have seen a continually growing interest of many researchers in the theory of two‐dimensional (2D), which became a distinct and important branch of the Systems Theory. Two‐dimensional models were developed in a series of papers by Roesser [15], Fornasini and Marchesini [4], Attasi [1], Eising [3] and others [6], [7], [12], [13] . The reasons for the interest in this domain are on one side the richness in potential application fields and on the other side the richness and significance of the theoretical approaches. The application fields include circuits, control and signal processing, image processing (which is the core of this approach), computer tomography, seismology, control of multipass processes [16], [17], iterative learning control [8] etc. The invariant subspaces with respect to linear transformations represent the fundamentals of the Geometric Approach, which is one of the main trends in Systems and Control Theory. Geometric Approach provides a very clear concept of minimality and explicit geometric conditions for controllability, observability, constructibility, pole assignability, tracking or regulation etc. These concepts are used to obtain efficient and elegant solutions of important problems of controller synthesis such as decoupling and pole‐assignment problems or the compensator and regulator synthesis for linear time‐invariant multivariable systems. The history of the Geometric Approach starts in 1969 when Basile and Marro [2] introduced and studied the basic geometric tools named by them controlled and conditioned invariant subspaces. They applied these techniques to disturbance rejection or unknown‐ input observability. In 1970 Wonham and Morse [18] applied a maximal controlled invariant method to decoupling and noninteracting control problems and later on Wonham's book [19] imposed the name of "(A,B)‐invariant" instead of "(A,B)‐controlled invariant". Basile and Marro opened new prospects to many applications (disturbance rejection, noninteraction etc.) by the robust controlled invariant and the emphasis of the duality [2]. The Maro’s monograph [11] presents the various aspects of the Geometric Approach, from the fundamental concepts, the structure of the linear systems, invariant subspaces, up to applications to the regulator problem,


20

noninteraction, feedback and robustness etc. The LQ problem was studied in a geometric framework by Silverman, Hautus and Willems. An important series of researchers among which Anderson, Akashi, Bhattacharyya, Kucera, Malabre, Molinari, Pearson, Francis and Schumacher developed the theory and applications of the Geometric Approach. The range of the applications of the Geometric Approach was extended to various areas, including, for instance, Markovian representations (Lindquist, Picci and Ruckebusch [10]) or modeling and estimation of linear stochastic systems [9]). All these books and papers refers to the 'classical' 1D systems. In the present lecture some aspects of the Geometric Approach are extended to a class of 2D systems described by a partial differential state equation. The state space representation of these systems is characterized by five matrices: two drift matrices and , an input‐state matrix , a state‐output matrix and a input‐output matrix . These systems represent the continuous counterpart of Attasi's 2D discrete‐time model. The behavior of these 2D systems is described and their general response formula is obtained. The concepts of complete controllability and complete observability are introduced and they are characterized by means of two suitable 2D controllability and observability Gramians. In the case of time‐invariant 2D some controllability and observability criteria are derived. The controllability and observability matrices are constructed (by extending the usual 1D ones). The first is used to characterize the space of the controlable states as the minimal ‐ invariant subspace which contains the columns of the matrix B and to obtain necessary and sufficient conditions of controlability for 2D systems. An algorithm is presented which compute the minimal ‐invariant subspace included in , (i.e. the subspace of the controllable states of the system ) and which extends the 1D algorithm from [9]. The observability Gramian and the observability matrix are employed to obtain the description of the space of non‐observable states as the maximal ‐invariant subspace contained in and to derive some observability criteria. An algorithm is proposed which compute this invariant subspace. These invariant subspaces can be used to obtain the Kalman canonical decomposition of the state space and to reduce the system to a minimal realization. Brief Biography of the Speaker: Prof. Valeriu Prepelita graduated from the Faculty of Mathematics‐Mechanics of the University of Bucharest in 1964. He obtained Ph.D. in Mathematics at the University of Bucharest in 1974. He is currently Professor at the Faculty of Applied Sciences, the University Politehnica of Bucharest, Director of the Department Mathematics‐ Informatics. His research and teaching activities have covered a large area of domains such as Systems Theory and Control, Multidimensional Systems, Functions of a Complex Variables, Linear and Multilinear Algebra, Special Functions, Ordinary Differential Equations, Partial Differential Equations, Operational Calculus, Probability Theory and Stochastic Processes, Operational Research, Mathematical Programming, Mathematics of Finance. Professor Valeriu Prepelita is author of more than 110 published papers in refereed journals or conference proceedings and author or co‐author of 15 books. He has participated in many national and international Grants. He is member of the Editorial Board of some journals, member in the Organizing Committee and the Scientific Committee of several international conferences, keynote lecturer or chairman of some sections of these conferences. He is a reviewer for five international journals. He received the Award for Distinguished Didactic and Scientific Activity of the Ministry of Education and Instruction of Romania.


21

Tutorial 1

Projective Geometry and Duality for Graphics, Games and Visualization

Professor Vaclav Skala University of West Bohemia, Plzen

and VSB‐Technical University, Ostrava Czech Republic

URL: http://www.VaclavSkala.eu Summary: The tutorial gives a practical overview of projective geometry and its applications in geometry, GPU computations and games. It will show how typical geometrical and computational problems can be solved easily if reformulated using the projective geometry. Presented algorithms are easy to understand, implement and they are robust as well. Brief Biography of the Tutor: Prof. Vaclav Skala is a professor at the University of West Bohemia in Plzen where he established computer graphics labs and he is currently the director of the Center of Computer Graphics and Visualization (http://Graphics.zcu.cz). He is also a professor at the VSB Technical University in Ostrava. He is concentrated mostly on fundamental algorithms for computer graphics and visualization. In 2009, prof.Skala he became a Fellow of Eurographics Association.


22

Tutorial 2

Nonlinear Distortion in Wireless Transmitters

Dr. D. Budimir Wireless Communications Research Group

Faculty of Science and Technology University of Westminster

London, UK E‐mail: [email protected]

Summary: The aim of this tutorial is to provide a thorough understanding of principles, techniques and the state‐of‐art of the compensation of nonlinear distortion in RF and microwave circuits and transmitters for mobile and wireless applications such as such as IEEE 802.11ac, 3G, 3.5G, 3.9G (LTE) and 4G (LTE‐A). During the tutorial attendees will be aware of the strengths and limitations of commercial available circuit and subsystem design tools and have a fully understanding of the problems and issues involved in their applications. At the end of the tutorial attendees will also be able to assess rigorously both the theoretical and commercial variability of using various simulators, circuit and system design tools. Brief Biography of the Tutor: D. Budimir (M’93, SM’02)received a Ph.D. degree in Electronic and Electrical Engineering, University of Leeds, Leeds, UK. In March 1994, he joined the Department of Electronic and Electrical Engineering at Kings College London, University of London. Since January 1997, he has been with the School of Electronics and Computer Science, University of Westminster, London, UK, where is now a Reader of wireless communicationsand leads the Wireless Communications Research Group. His research interests include analysis and design of hybrid and monolithic microwave integrated circuits, the design of amplifiers, filters and multiplexing networks for RF, microwave and millimetre‐wave applications and RF and microwave wireless system design. Dr Budimir authored or coauthored over 250 journal and conference papers in the field ofRF, microwave, and millimeter‐wave wireless systems and technologies. He is author of the book Generalized Filter Design by Computer Optimization (Artech House, 1998) and Software and Users ManualEPFIL‐Waveguide E‐plane Filter Design (Artech House, 2000), and a chapter in Encyclopaedia of RF and Microwave Engineering(Wiley, 2005). He is a Member of the EPSRCPeerReviewCollege, a senior Member of IEEE. He is also a regular referee for IET Electronic Letters, IET Microwaves, Antennas, and Propagation, IEEE Microwave and Wireless Components Letters, IEEE Transactions on Microwave Theory and Techniques, IEEE Antennas and Wireless Propagation Letters, IEEE Transactions on Antennas and Propagation, IEEE Transactions on Circuits and Systems II and Proceedings of the IEEE, and Proceedings of the IEEE, and International Journal of RF and Microwave Computer Aided


23

Engineering. He is a member of several International conference Technical Program Committees. He has given more than 18 invited presentations at workshops, conferences and seminars. He has supervised 13 PhD research students as main supervisor/director of studies to a successful completion and is currently supervising 7 PhD students as the main supervisor. He has won awards for his journal papers.


24

Abstract— The Mobile Ad hoc Network (MANET) has

become an increasingly popular area of research during the last

decade. As an emerging fourth generation wireless topology,

the main goal of the MANET will be to seamlessly connect

mobile devices anywhere and anytime. Recently, the

introduction of Ultra-WideBand (UWB) technology has

become a promising candidate to support MANET’s and

research in this area has been extensive over the last few years

due to its powerful capabilities such as the large Band Width

(BW) available and the high data rates possible. Routing

protocols enable mobile hosts (or nodes), such as laptops and

cellular telephones, the ability to communicate with each other

and the design of routing protocols has encouraged researchers

to investigate and develop new strategies that establish and

maintain connections between mobile nodes. The main aim of

the research presented in this paper was to design and

implement an efficient routing protocol for MANET based on

UWB technology. The Location Aided Routing (LAR)

protocol scheme 1, is first considered in order to satisfy the

requirements of the proposed routing protocol, and combines

the advantages of the requested zone in LAR, and the dynamic

extended zones in the modified schemes of LAR.

The main objectives of this research was to reduce the

power consumption of the mobile nodes (thus increasing the

lifetime of network), reduce the network overhead, and

increasing route reliability. Analytical model is presented to

prove the validation of the protocol in addition to simulation

results which show that the Modified LAR1 (MLAR1)

outperforms both the AODV and LAR1 protocols, it increases

the life time of network, produce less overhead, and also

provide the highest throughput.

Keywords—MANET, UWB, Routing protocol, LAR.

I. INTRODUCTION

IRELESS Local Area Networks (WLANs) provide

wireless access to different types of mobile hosts such

as personal digital assistants (PDA), laptops and cellular

phones. These nodes are equipped with short range

transmitters and receivers, and antennas which may be

omnidirectional (broadcast), highly-directional (point-to-

point), or some combination of the two [1].

In a WLAN environment, routing protocols then enable

nodes to relay data packets if they are within transmission

range, or if they can communicate directly. If they are away

from each other, intermediate nodes are required to establish a

multihop route between the sender and receiver. The wireless

routing protocols that provide this key functionality, in

general, are classified as either topological based or position

based.

Topological based routing protocols use the existing

information about links in network to flood (or forward)

packets. There are two main routing strategies classified

as topological based; proactive [2] that maintains routing

information for each node in the network and stores this

information in routing tables, such as Destination-

Sequenced Distance Vector (DSDV) [3], Cluster-head

Gateway Switch Routing (CGSR) [3], Wireless Routing

Protocol (WRP) [4], and Optimized Link State Routing

Protocol (OLSR) [5]. The second type is reactive routing

protocols which maintain a route on demand, such as Ad

hoc On-Demand Distance Vector (AODV) [6], Dynamic

Source Routing (DSR) [7], Temporally Ordered Routing

Algorithm (TORA) [4], and Associativity-Based Routing

(ABR) [3]

Position based routing protocols exploit positional

information to direct flooding towards the destination in

order to reduce overheads and power consumption,

Location Aided Routing Protocol (LAR) [8], GRID [9],

Compass [10], and Greedy Perimeter Stateless Routing

(GPSR) [11] are examples of position based routing

protocols.

Ultra-WideBand (UWB) [12][13] is a radio technology that

has been proposed for use in Personal Area Networks (PAN)

and appears in the IEEE 802.15.3a draft PAN standard. UWB

systems consume very low energy levels, and can be used in

short-range, high-bandwidth (BW) communications systems

where the BW > 500 MHz, at 20% of the center frequency.

UWB is defined to operate between 3.1–10.6 GHz and is

restricted to a maximum allowable power spectral density of

41.3 dBm / MHz corresponding to average transmitted power

of 0.5 mW. Therefore, UWB provides relatively short radio

range but given the spectrum available very high data rates in

excess of 100 Mbps can be achieved, (with bit rates of 55, 110

& 200 Mbps [14]).

UWB is best used for ad-hoc and sensor networks [15], it is

used as part of location systems and real time location systems

such as hospitals and healthcare, short broadcast time, “see-

through-the-wall” precision radar imaging technology [16],

and replacing cables between portable multimedia Consumer

Electronics (CE) devices and in the next-generation of

Analytic Model of MLAR1

Dr. Yahia Hasan Jazyah

[email protected]

W


25

Bluetooth Technology devices [17].

This paper proposes a new routing protocol for the UWB

MANET and is based on the conventional Location Aided

Routing protocol, scheme 1 (LAR1). The new approach

proposes a dynamic and static request zone at the same time,

in order to consider power as a metric when a route is selected.

Three regions of request zones are assigned; the first zone

represents a rectangle where the source and destination lie in

opposite corners of the rectangle, see Fig. 1.

The second and third zones, see Fig. 2, represent rectangles

with dimensions that are dependent on the dimension of the

first zone, (this is explained later in more detail).

In order to achieve reliability and increase the probability of

finding route, when each node within a request zone receives a

Route Request (RREQ), they respond by transmitting a Route

Reply (RREP) according to the following criteria:

In each region, a threshold value of residual battery power is

assigned

The threshold of region 1 (Th1) is greater than that of

region 2

The threshold value of region 2 (Th2) is greater than that of

region 3.

The basic operation of the protocol is as follows:

The initial RREQ is flooded over the request zone. When a

node within one of the three zones receives a RREQ, it checks

the header of RREQ for the threshold and dimensions of each

region (it should be noted here that positional information is

known by all nodes). If the node is within one of the zones,

and has a residual battery power above the threshold of that

zone, it forwards the RREQ and so on until RREQ reaches to

destination which replies to the first received RREQ using the

reverse path.

The rest of paper is organized as follows: section 2

summarizes related work, section 3 presents the proposed

routing protocol, section 4 presents simulation results, and the

summary in section 5.

II. RELATED WORK AND MOTIVATION

The new routing protocol exploits the advantages of LAR1

routing protocol besides power issue in Location-Aware

routing Protocol with Dynamic Adaptive of Request Zone

protocol (LARDAR). Below are more details about LAR1 and

LARDAR.

A. Location Aided Routing Protocol (LAR)

Location Aided Routing (LAR) [18] is an on-demand

routing protocol. LAR uses the modified Dijkstra's Algorithm

to find the shortest path; it relies on a flooding-based route

discovery procedure which causes a huge amount of routing

overhead. Destination lies in a circular region of certain radius

centred at a position at certain time, known as the Expected

Zone, which indicates which zone of the network should be

reached by RREQ packets. Global Positioning System (GPS)

enabled terminals to know its own position and speed, while

dissemination is performed by piggybacking location

information in all routing packets.

There are two proposed schemes of LAR: [19]

First assumption (LAR1) defines request zone that includes

sender and receiver on opposite corner of a rectangle as shown

in Fig. 1. The rectangle dimensions are estimated according to

the receiver average speed at a certain time. Nodes within this

zone respond to the RREQ of sender by forwarding the RREQ

to their nieghbors. This scheme reduces network overhead but

causes delay.

Another LAR scheme (LAR2) is proposed depending on the

calculated distance between source location and the estimated

position of destination. Each node receives the RREQ

calculates the distance toward destination, if the distance is

less than of the distance from the previous sender node to

destination, it forwards the packet. In this scheme,

intermediate receiving node may be the closest node to

Source

Destination

Region1

Region3 Th3

Th2

Th1

Th1>Th2>Th3

Figure. 2 Requested zones

Expected Zone

Request Zone

Source

Destination

Figure 1. LAR, scheme 1.

Region2


26

destination, and so the algorithm reaches a dead-end.

B. Location-Aware routing Protocol with Dynamic

Adaptive of Request Zone (LARDAR)

LARDAR [20] is a location-based routing protocol. It uses

a) the destination's position to form a triangle or rectangle

request zone in order to reduce the traffic, b) dynamic

adaptation of request zone technique in order to adapt the

precision of the estimated request zone and reduce the

searching range, and c) increasing-exclusive search approach

to redo route discovery by a progressive increasing search

angle basis when route discovery failed and reduce routing

overhead.

LARDAR utilizes the GPS information of destination

node’s location, timestamp (location information obtained

time), and velocity which can be calculated using the traveled

distance and the time needed for traveling that distance.

LARDAR improves the drawbacks of LAR and Distance

Routing Effect Algorithm for Mobility (DREAM) [2]. It

utilizes the location and expected zone in LAR, location

update frequency of DREAM, and uses energy aware and

geographical informed neighbour selection to reduce network

overhead which is the number of control packets transmitted.

In LARDAR, increasing the angle of the triangle-zone shape

depends on some factors; it can be improved by genetic

algorithm. When the expected zone is extended, source

forwards RREQ with the same sequence number, when

intermediate nodes receive the RREQ again, they discard it.

This could be a problem; these nodes could be the only ones

that can relay RREQ to the other nodes within the extended

zone.

LARDAR saves power and lengthen system lifetime.

C. Power-Aware Routing

Since mobile hosts depend only on local power supplied by

batteries, power aware is an important issue which must be

taken into account when designing wireless routing protocols.

Energy efficient issues, in general, are: Transmission Power

Control, Minimum Power Routing, Power-aware route

selection, and Battery-Cost-Aware Routing. [21]

Several routing techniques uses residual battery power as

metric to select next hop or detecting link failure, such as

Minimal Battery Cost Routing (MBCR) [22] which uses

battery power evenly depending on a cost function in order to

select next hop node, the cost function considers only the total

cost while the route can include a node with little energy while

the other nodes have a plenty of energy. Min-Max Battery

Cost Routing (MMBCR) [22] which considers nodes' lifetime

and avoid nodes that will exhaust soon when selecting next

hop, it prolongs the lifetime of an individual node by

introducing a new path cost, but it can set up the route with an

excessive hop count and then consume a lot of total

transmission energy because it takes into account the

remaining energy level of individual nodes instead of the total

energy. Max-Min Battery Capacity Routing (CMMBCR) [22]

which tries to make balance between MBCR and MMBCR, the

basic idea behind CMMBCR is that when all nodes on route

have remaining battery capacity above a threshold, a route

with minimum total transmission power among all routes is

chosen. It maximizes the lifetime of ad hoc mobile networks,

but algorithm is very complex. AODV with Break Avoidance

(AODV-BA) [23] that can avoid route breaks; intermediate

node that detects the breakage re-establishes a new route

before the route breaks. The detection of link breakage based

on threshold value which are: received radio, the residual

battery power, and the density. Energy-Aware AODV

(EAAODV) [24] where each node receives RREQ checks its

residual battery power, if it is above a threshold, it forwards

the RREQ, and otherwise, the node ignores it. And Power

Aware On-Demand (PAOD) [25] where the selection criterion

is based on two thresholds; each node compares its

residual battery power to these levels. When the

residual battery power reaches to a certain level

(link is going to be failed), an action is done.

III. ROUTING PROTOCOL

The new proposed routing protocol exploits functionality of

LAR1 to improve route reliability and decrease power

consumption.

A. Routing Strategy

Three regions are assigned, and each region has a certain

threshold value of power. Each node within its region and has

a residual battery power greater than the threshold responds to

RREQ by retransmitting it as shown in Fig. 2.

The threshold values should be within the maximum

available power and minimum received power which is

calculated as follows:

(1)

Where: Pr is received power, Pt is the transmitted power,

and N is thermal noise. Whereas thermal noise is calculated

according to the following equation [26]:

(2)

Where T is noise temperature in Kelvin of the input

termination and equal to 290 K which is the standard

temperature, K is Boltzmann constant (= 1.379 x 10-23

Joules/Kelvin), B is bandwidth of carrier signal in Hz, and F is

a constant called the noise factor [27].

When source node wants to send data packets to destination,

it forwards RREQ based on the current information about

location and velocity of destination. First it establishes the

expected zone based on the previous information stored in its

routing table about the destination, this information includes

the speed and position at certain time, source node assumes the

destination in the center of circle where the radius is calculated

according to the speed and time equation whereas the radius is

the speed times the period of traveling, this period is the

difference between the current time and the time when that

information is updated in routing tables. Then the source

establishes the first request zone as shown in Fig. 1, then it

N = T * k * B * F

Pr = Pt – N


27

calculates the dimensions of the other two zones where the

dimensions are proportional to the dimension of first zone as

shown in Fig. 2. And then the source forwards the RREQ

including in the header extra information which are the

dimensions of each region, threshold values assigned to each

region.

If RREP is not received within a certain time, RREQ is

reinitiated again; Fig. 3 shows the process of initiating RREQ.

Fig. 3 Initiating RREQ

When intermediate node receives RREQ for the first time, it

checks its position by GPS and RREQ’s header for zones’

information (dimensions) and threshold values, if node is

within the three regions, it compares its residual battery power

to the threshold assigned to its zone, if it is greater than the

threshold, it forwards the message to its neighbors. Fig. 4

shows the process, otherwise, It ignores the RREQ.

Fig. 4 Handling of RREQ

The process is keeping running until RREQ reaches its

destination.

When RREQ is received by destination, it forwards RREP

to source using the selected reverse path. Fig. 5 illustrates the

process.

Fig. 5 Handling of RREQ by destination

Each node has routing tables that stores routing information

which is updated whenever it receives control packet. When

RREP is received by intermediate node that is located on the

reverse path, it updates its routing table according to the

information included in the header of the received RREP, and

then it forwards the RREP towards the source. Fig. 6 shows

the process.

Fig. 6 Handling of RREP

IV. ANALYTICAL MODEL

Simulation representation is the common way to prove the

validation of wireless routing protocols for ad hoc networks.

Mathematical verification is considered a complementing

method for the evaluation of wireless routing protocols. The

following presents the mathematical model of the MLAR1

protocol.

The degree of a node x, d(x), is the number of nodes

directly connected to x., A node is said to be an isolated node

if it has a zero degree [28].

The probability P that each node has at least n0 neighbors is

given by: [29]

20

0

.!

)()n(

0

2

00

rn

en

rdP

(3)

Where P is the probability that a randomly selected node

has n0 neighbors, n0 is Number of node’s neighbors,

is

homogeneous node density, and r0: Transmission range of

node.

Broadcast forwarding probability [30] is the probability that

a node will forward an RREQ message to its neighbors

successfully. RREQs in wireless routing protocols may collide

or get lost due to channel errors or dynamic topology, and so,

this value is calculated and can be represented by the

following equation:

ecrs pppp .. (4)

Where spis broadcast forwarding probability, rp

is the

probability with which a node will forward a RREQ to its

neighbors, cpis the probability of not experiencing a collision

at the MAC layer, and is the probability that the RREQ is not

lost due to channel errors.

Notice here that rp is 1 in case of wireless routing

protocols that flood RREQ over the whole network, while

rp< 1 in case of position based routing protocols as not all

nodes are involved in rebroadcasting the RREQ.

Define Request Zone

Initiate RREQ

Wait certain time for RREP

If RREP is received

Establish route and send data packets

Else reinitiate RREQ

If RREP received

If node address appears in header of RREP

Update cache

Forward RREP

If not

Ignore it

If RREP received

If node address appears in header of

RREP

Update cache

Forward RREP

If not

Ignore it

If RREQ received for the first time

Check region

If within regions 1, 2, or 3

Check threshold

If greater than dedicated threshold

Node adds its address

Forward RREQ

If not

Ignore the message

If outside regions 1, 2, or 3

Ignore the message

If RREQ is already received

Ignore it

If RREQ is received

Initiate RREP and forward it using reverse path


28

In position based wireless routing protocols, not all nodes

that receive RREQ will forward it due to either it has received

it by another link or due to forwarding restrictions. In this

work, the network overhead is the number of relayed RREQ

packets, in other words it is a function of Expected Forward

Degree [29].

When source broadcast a RREQ to its neighbors, this RREQ

will be received on average by avgec dPP .. nodes and then

each neighbor node receives the RREQ will forward it by

probability of rp. And so the number of source’s neighbors

that will forward the received RREQ to their neighbors is

avgs dP . nodes. Keeping accumulating the number of

forwarding the received RREQ successfully each hop up to h

hops from the source (until RREQ is received by destination)

yields to the total number of network overhead which can be

represented by: [30]

1

1 1

1 .][)(

,1

h

i

i

j

f

i

savgavgs

avgs

potherwisejdPddP

hifdP

C

(5)

Where pC is the total expected network overhead, avgd

is

average degree, ][ jd f is expected forward degree of a node

at j hops, and h is number of hops from the source node.

Form the previous equation, the network route overhead is

directly related to the number of hops traversed by a RREQ,

probability of forwarding a RREQ ( rp), and the expected

forward degree of nodes. When rp = 1 (topology based

routing protocol such as AODV), RREQ is flooded over the

whole network, the result is huge network overhead, when rp

< 1 (position based routing protocol such as LAR1), the pC

will be decreased compared to the previous case, increasing

the restrictions on selecting the next hop node leads to less rp

which leads to less pC which means less network overhead as

the case in the MLAR1 protocol. This proves that the network

overhead of the MLAR1 protocol is less than the AODV and

LAR1.

The energy consumed by node is due to relaying RREQ (or

route discovery) and data packets. The following equation

represents the total energy consumed by a protocol: [30]

eavgerreqpeerrepdatotal RdTBCRTmBMBLG )(

(6)

Where Gtotal is the total energy consumed by a protocol, M

is total number of data packets, La is average path length, Bd is

size of data packets (bits), m is number of paths discovered by

a protocol, Brrep is size of RREP packet (bits), Te is

transmitted energy per bit, Re is received energy per bit, Cp is

the total expected network route overhead, Brreq is size of

RREQ packet (bits), and davg is average degree.

In the equation of total energy consumed by a protocol

(network), there are constant and variable terms. We have the

following constants that will not affect the comparison

between the three protocols, which are: total number of data

packets, average path length if we consider the same path is

selected for the three protocols for simplicity, size of data

packets, number of paths discovered by a protocol where all of

the three protocols are single path routing protocols,

transmitted energy per bit, and received energy per bit.

On the other hand, the following terms are variable and can

affect the energy consumed by the protocol, which are: size of

RREP Packet, size of RREQ packet, average degree, and the

total expected network overhead.

The average degree of AODV is highest value, LAR1 has

less value than AODV as nodes near the border of request

zone do not have connection to neighbor nodes outside the

request zone, but the MLAR1 has the least value due to the

restrictions of selecting neighbors of each nodes, whereas

nodes within the zone and have sufficient energy can be

considered as neighbor nodes. The total expected network

overhead of MLAR1 has the least value as discussed

previously, based on the previous two terms; the MLAR1

consumes less energy than the AODV and LAR1. The effect of

RREP packet size has a very low influence on the result cause

only one RREP is forwarded to source by destination in case

of MLAR1 despite the fact that the size of RREP’s header is

increased by adding additional fields, but the size of RREQ

packet, which is also extended, can affect the consumed

power; the average degree of the MLAR1 is less than the other

two protocols which mitigate the effect of this factor, besides

that the huge amount of transmitted bits is due to the data

packets rather than the RREQ packet, from the previous talk

we can conclude that the MLAR1 protocol consumes less

energy than AODV and LAR1 routing protocols.

The following are some results of the analytical model for

the AODV, LAR1, and the MLAR1 based on equation

5(Network overhead): Consider the following values: Ps = 1 in

case of AODV with reference to equation 4, if we consider a

node will forward the RREQ successfully to 5 neighboring

nodes, then Ps.davg = 5 (the number of neighbors close to the

source that will forward the received RREQ to their

neighbors), and davg = 5 as Ps = 1, df[j] = 2 (df[j] = (2j + 1) /

(2j – 1), table 1 shows the values of parameters of equation 4

of the AODV for three hops as a simple example.


29

Ps davg Ps.davg i (Ps)i+1

df[j]

1 5 5 1 1 2

1 5 5 2 1 5/3

1 5 5 3 1 7/5

Table 1 Routing overhead of AODV

Substituting the values from table 1 into the equation 4, then

Cp(AODV) =

1+5+(1*5*2)+(1*12*5*2*5/3)+(1*13*5*2*5/3*7/5)=50

By the same way we generate the table for LAR1, but if we

consider the total number of neighboring nodes that the source

can send them the RREQ is 5 as the case of AODV (same

scenario used) but 3 nodes are within the request zone for

example, then Ps = 3/5.


df[j]

3/5 5 3 1 9/25 2

3/5 5 3 2 27/125 5/3

3/5 5 3 3 81/625 7/5

Table 2 Routing overhead of LAR1

Then the Cp(LAR1) = 14

By the same way we generate the table for MLAR1, but if

we consider the total number of neighboring nodes that the

source can send them the RREQ is 5 as the case of AODV

(same scenario used) but 2 nodes satisfy the two conditions of

forwarding for example, then Ps = 2/5.


df[j]

2/5 5 2 1 4/25 2

2/5 5 2 2 8/125 5/3

2/5 5 2 3 16/625 7/5

Table 3 Routing overhead of MLAR1

Then the total routing overhead Cp(MLAR1) = 7

The results are as expected, Cp(MLAR1)< Cp(LAR1)<

Cp(AODV). If we consider LAR1 fails to find the route within

the request zone, it will resort to flooding the whole network as

the AODV, then the total routing over head is Cp(AODV) +

Cp(LAR1) = 64, and so Cp(MLAR1)< Cp(AODV)<

Cp(LAR1) which is similar to the simulation results from

QualNet (see figures 8, 9, 11, 13, 14, and 16).

The next step is to calculate the energy consumed by the

three protocols based on equation 6. We have that The

transmitted power is 0.5 mW, the bit rate is 55 Mbit/s, Eb = P /

BR where Eb is energy per bit, P is carrier power, and BR is

the bit rate, substituting to the previous equation yields the

transmitted energy per bit is 9.091*10-12 Joules. While the

received power was is 0.5 mW which yields the same energy

per bit which is 9.091*10-12 Joules.

Referring to the equation 6 (The total energy consumed by

nodes), the following values are assumed to calculate the total

energy consumed by nodes for the AODV (3-hops): M = 100

packets, La = 5 hops, Bd = 512 bits, m = 1 bath, Te =

9.091*10-12 Joules, Re = 9.091*10-12 Joules, Cp = 50, davg

= 5, Brreq = 143 bits, Brrep = 126 bits. And then, by

refereeing to equation 6, the total energy consumed by nodes

for 3-hops is 5.056*10-6 Joule.

Regarding the LAR1 protocol, Cp = 14, Brreq = 115 bits,

Brrep = 76 bits, substituting the previous values into equation

6 yields 4.749*10-6 Joule as the total energy consumed by

nodes for 3-hops.

When considering the MLAR1 routing protocol with Cp =

7, Brreq = 189 bits, Brrep = 150 bits, substituting the previous

values into equation 6 yields 4.740*10-6 Joule as the total

energy consumed by nodes for 3-hops.

The previous values of energy prove that Cp(MLAR1) <

Cp(LAR1) < Cp(AODV), if we consider the case that LAR1

could not find the route within the request zone, then it will

resort to flooding the whole network as the AODV, and then

the total energy consumed by nodes for LAR1 = Cp(LAR1) +

Cp(AODV) = 9.805*10-6 Joule, which is similar to the results

from QualNet simulator (see figures 7 and 12).

V. NETWORK ENVIRONMENT AND SIMULATION RESULTS

The proposed scenario considers a network of 20 hosts

where the positions of nodes are selected randomly based on

the standard normal distribution, the scenario represents an

outdoor network, each host has a certain level of battery power

which is selected randomly, and nodes in this network scenario

are static. and threshold values are selected to have reasonable

values regarding the maximum and minimum battery power of

nodes which ranges from 1.0 mAhr and 0.1 mAhr. Lower

values are selected rather than the real one whereas selecting

large values will not affect the results, but the statistics graphs

will not show the difference in power consumption, for

example, when power values of batteries are big while the

consumed power is low, the graphs will not show the effect of

modified routing protocol.

Using frequency of 3.5 GHz (within the UWB frequency

range) and substituting into equations 1 and 2 yields in

received power of 0.33mA, And so, the threshold values

should be within the range: 1.0 > Th > 0.33 (received current).

Each node is equipped with GPS system which imposes an

outdoor scenario, besides an UWB system that uses 0.5

milliWatt as transmitted power, frequency of channel is 3.5

GHz (range of frequency available is 3.1 – 10.6 GHz), SNR

sensitivity is 1.0; which is acceptable ratio to enable receivers

to detect incoming signal, temperature is 290 Kelvin which is


30

the standard temperature, bit rate is 55 Mbps which is the (the

minimum bit rate) and the range is up to 20 m as the minimum

bit rate is used which are the typical values [14].

Number of zones is three, where the dimensions of second

zone equal to 50% of the firs zone (requested zone), while the

dimensions of third zone equal to 30% of the requested zone.

Constant Bit Rate (CBR) is selected to be used as traffic

generator; it is a UDP-based client-server application which

sends data from a client to a server at a constant bit rate,

whereas 100 packets will be forwarded by source to

destination, each packet is 512 bytes in size; all these values

are applied to QualNet simulator.

A. Static Mode Results

This section shows the results of modified protocol in

compare to LAR1 and AODV in static mode (nodes are fixed).

As shown in Fig. 7, MLAR1 has the minimum power

consumption (system aggregation) in transmission mode and

LAR1 has the maximum amount, while AODV has an

intermediate value between them. This is an indicator that

MLAR1 protocol generate less RREQ packets than the other

two protocols because the only nodes which have residual

battery power over than the threshold can forward the RREQ

and data packets as well despite the fact that the flooded area

of MLAR1 is greater than that of LAR1, which indicates that

the MLAR1 increases the life time of the network as a whole

in case of transmission mode.

Fig.s 8 and 9 show the broadcast packet sent and received to

channel respectively. They represent the amount of received

and transmitted RREQ packets, which is a meter to network

overhead; whereas the MLAR1 has the minimum value of both

received and transmitted signals, LAR1 has the maximum

value, and AODV has an intermediate value. An explanation

of the result is that the requested zone of LAR1 does not cover

the destination and so source resorts to flooding over the

whole network as a method to find route to destination, this

means that RREQ is retransmitted more than once by the same

some nodes (nodes within the requested zone), AODV uses

flooding directly to cover the whole network, and so it ignores

the retransmission of RREQ as in LAR1, while in MLAR1, the

requested zone is extended to three regions instead of one, the

new requested zones cover the destination, and so no need to

resort to flooding, on the other hand, not all nodes forward the

RREQ, because only nodes which have residual battery power

over a predefined threshold value forward the RREQ. As a

result, MLAR1 forwards less number of RREQ, which affects

the number of received RREQ as well, and so not only it

reduces network overhead, but it also reduces power

consumption.

Fig. 7 Battery energy consumed in transmit mode (static mode).

Fig. 8 Broadcast packets sent to channel (static mode).


31

Fig. 9 Broadcast received (static mode).

Fig. 10 illustrates the throughput of the three routing

protocol, it shows that the throughput is equal for all of them;

data packets are transmitted and received by the same amount,

and so the difference is in control packets especially RREQ,

that means the lower amount of consumed energy in either

transmit or receive modes does not affect the throughput and it

is not cause of data packets, on the contrary, it is at the

expenses of RREQ; because not all node participate in

forwarding the RREQ, only the nodes which have residual

battery power greater than the threshold can forward it.

Fig. 11 indicates the amount of packets received clearly, in

other words, how much the system is reliable; the figure

illustrates that the three protocols have the same value of

reliability; because nodes are distributed randomly over the

network and power values are assigned randomly as well, the

selected route in all cases is the same, or at least the nodes on

the active route have enough battery power to recognize logic

1 as logic 1 and logic 0 as logic 0. The effect of reliability

appears in dynamic mode as nodes move, and so not the same

nodes are selected in the active route for different routing

protocols.

Fig. 10 Throughput (static mode).

Fig. 11 Unicast packets received clearly (static mode).


32

B. Dynamic Mode Results

This section shows the results of modified protocol in

compare to LAR1 and AODV in dynamic mode (nodes are

moving). Fig. 12 shows the energy consumed in transmit mode

when nodes are dynamic. We notice that the MLAR1 has the

maximum power consumption in transmission mode and

AODV has the minimum amount, while LAR1 has an

intermediate value between them; regarding MLAR1 protocol,

as nodes move randomly in every direction and the restrictions

of selecting nodes that will forward RREQ besides the

confined area of flooding, this makes it difficult to find

suitable route quickly, on the opposite side, AODV uses

flooding over the network, which makes it easy to find route to

destination, while LAR1 is only restricted by the confined

area. Nevertheless, the three results are close to each other as

discussed later on regarding fig. 5.15 (throughput).

Fig.s 13 and 14 show the broadcast packet sent and received

to channel respectively. It represents the amount of received

and transmitted RREQ packets, which is an indicator to

network overhead. These two figures confirm the previous

results whereas most of the time and power consumed in

transmitting and receiving are due forwarding and receiving

data packets rather that control packets.

The two figures show that the MLAR1 has the minimum

value of both received and transmitted signals, LAR1 has the

maximum value, and AODV has an intermediate value. An

explanation of the result is that the requested zone of LAR1

does not cover the destination and so source node resorts to

flooding over the whole network as a method to find route to

destination, this means that RREQ is retransmitted more than

once by the same some nodes (nodes within the requested zone

only), AODV uses flooding directly to cover the whole

network, and so it ignores the retransmission of RREQ as in

LAR1, another point is that RERR packet could be generated

as nodes may have not enough power to transmit and receive

all the time which forces the source to generate more RREQs

to find new route to destination. While in MLAR1, the

requested zone is extended to three regions instead of one, the

new requested zones cover the destination, and so no need to

resort to flooding, On the other hand, not all nodes forward the

RREQ because only nodes which have residual battery power

over a predefined threshold value forward the RREQ. As a

result, MLAR1 forwards less number of RREQ, which affects

the number of received RREQ as well, and so it does not only

reduce network overhead, but also reduces power consumption

Fig. 15 clarifies the throughput of the three routing protocol,

it shows that MLAR1 achieves the highest throughput of the

three protocols; that is, nodes which are selected in active

route have enough power to receive and transmit packets, and

so it achieves the highest throughput, and this is an explanation

of the previous results in the last four figures (why the MLAR1

consumes more energy in transmit and receive mode).

Fig. 12 Battery energy consumed in transmit mode (dynamic mode).

Fig. 13 Broadcast packets sent to channel (dynamic mode).


33

Fig. 14 Broadcast received (dynamic mode).

Fig. 15 Throughput (dynamic mode).

Fig. 16 Unicast packets received clearly (dynamic mode).

Fig. 16 represents the reliability of protocol; it indicates the

amount of packets received clearly. The figure illustrates that

the MLAR1 protocol has the highest value than the other two

protocols, which means that it is the most reliable compared to

LAR1 and AODV routing protocols. This is because of the

strategy of route selection; where nodes of higher power is

selected rather than low power nodes, which maintains the

route and enable active nodes to receive data packets clearly,

i.e. logic 1 is received as logic 1 and logic 0 is received as

logic 0.

VI. CONCLUSION AND FUTURE WORK

In this paper, a new routing protocol for UWB MANET is

proposed which exploits functionality of conventional LAR1

in order to improve route reliability and decrease power

consumption of nodes and overhead.

In the proposed protocol, only nodes located within the

assigned region respond to RREQ by detecting its position by

GPS, determine the zone they belong to and the its threshold

by detecting RREQ’s header; if the previous conditions are

valid then the node forward RREQ to its neighbors.

Analytical model is presented which prove the validity of

the MLAR1 protocol and the simulation results confirm that

the MLAR1 outperforms both the AODV and LAR1 routing

protocols

The MLAR1 protocol performs well in static and dynamic

modes, it outperforms both AODV and LAR1 protocol at all

levels (network life time, network overhead, reliability); the


34

key factor for the above results is the selection criteria of

nodes; where nodes of residual battery power above a

threshold can participate in active route, which guarantees

route reliability, network life time, and high throughput,

besides the extension of the confinement flooded area by

dividing the resulting requested zone into three zones of

different dimensions and assigning a threshold value to each

zone.

Results of dynamic mode are better than results of static

mode. In dynamic mode, nodes are moving in every and any

way, and so any protocol may stick if a node of low battery

power is selected in active route, running out the node’s power

will cause a link failure and affect the performance of routing

protocol, MLAR1 does not stick in such problem as nodes of

residual battery power above a threshold are always selected,

which guarantees a high performance even in dynamic mode.

REFERENCES

[1] M. S. Corson and J. Macker, "Mobile Ad hoc Networking (MANET):

Routing Protocol Performance Issues and Evaluation Considerations,"

1999.

[2] M. Mauve, J. Widmer, and H. Hartenstein, "A Survey on Position-

Based Routing in Mobile Ad Hoc Networks," IEEE Networks, pages

30–39, Nov. Dec. 2001.

[3] M. Abolhasan, T. Wysocki, and E. Dutkiewicz, "A review of routing

protocols for mobile ad hoc networks," Received 25 March 2003;

accepted 4 June 2003, Elsevier B.V.

[4] Yahia Jazyah and Martin Hope, “A Review of Routing Protocols for

UWB MANETs”, Computational Science and Its Applications – ICCSA

2010, Lecture Notes in Computer Science, March 2010, Volume

6018/2010, Page(s): 228-245, DOI: 10.1007/978-3-642-12179-1_21,

Print ISBN: 978-3-642-12178-4, Online ISBN: 978-3-642-12179-1

[5] T. Clausen, P. Jacquet, A. Laouiti, P. Muhlethaler, A. Qayyum, and L.

Viennot, “Optimized Link State Routing Protocol,” in Proceedings of

IEEE INMIC, 2001.

[6] C. Perkins, E. Belding-Royer, and S. Das, "Ad hoc On-Demand

Distance Vector (AODV) Routing," July 2003.

[7] D. B. Johnson, D. A. Maltz, and J. Broch. “DSR: The Dynamic Source

Routing Protocol for Multi-Hop Wireless Ad Hoc Networks. in Ad Hoc

Networking,” edited by Charles E. Perkins, Chapter 5, pp. 139-172,

Addison-Wesley, 2001.

[8] Y. B. Ko and N. H. Vaidya, "Location-Aided Routing (LAR) in mobile

ad hoc networks," Wireless Networks 6 (2000) 307–321 307.

[9] W. H. Liao, Y. C. Tseng, and J. P. Sheu, "GRID: A fully location-aware

routing protocols for mobile ad hoc networks," Proc. IEEE HICSS,

January 2000.

[10] E. Kranakis, H. Singh, and J. Urrutia, "Compass Routing on Geometric

Networks," Proc. 11th Canadian Conf. Computational Geometry, Aug.

1999.

[11] B. Karp and H. T. Kung, "GPSR: Greedy Perimeter Stateless Routing

for Wireless Networks,” MobiCom 2000.

[12] K. Kuladinith, A.s Timm-Giel and C. Görg, “Mobile Ad-Hoc

Communications in AEC Industry”, ITcon Vol. 9 (2004), August 2004

Page(s): 313 – 323 available at http://www.itcon.org/2004/22/

[13] E. R. Green and S. Roy, "System Architectures for High-rate Ultra-

wideband Communication Systems: A Review of Recent

Developments," Intel Labs, available

http://www.intel.com/technology/comms/uwb/download/w241_paper.pd

f

[14] L. Xia, A. Lo, I. Niemegeers and T. Baugé, "An Ultra-Wide Band Based

Ad Hoc Networking Scheme for Personnel Tracking in Emergencies",

this work was performed in the IST Framework 6 EUROPCOM project.

[15] Project SPEARHUW, "Enhanced Sensor and WLAN Networks using

UWB Communications", April 2005.

[16] M. G. D. Benedetto , G. Giancola, and D. Domenicali, "Ultra Wide

Band radio in Distributed Wireless Networks," 2006 IEEE.



1999.

[18] X. An, H. Shen, S. Kim, and K. Kwak, "Novel Location-Aided Routing

Scheme for Ultra-Wideband Ad-hoc Networks," Proceedings of ISCIT

2005.



[20] T. F. Shih and H. C. Yen, "Location-aware routing protocol with

dynamic adaptation of request zone for mobile ad hoc networks,"

Published online: 9 October 2006, Springer Science + Business Media,

LLC 2006. Wireless Network (2008) 14:321–333.

[21] M. Maleki, K. Dantu, and M. Pedram, "Power-aware Source Routing

Protocol for Mobile Ad Hoc Networks," Low Power Electronics and

Design, 2002. ISLPED '02. Proceedings of the 2002 International

Symposium on 2002 Page(s):72 – 75.

[22] B. Long and Q. Ding, "A Hybrid Energy-aware Multi-hop Routing

Algorithm for Distributed UWB Network," Communication

Technology, 2006. ICCT '06. International Conference on Nov. 2006

Page(s):1 - 5.

[23] M. Tauchi, T. Ideguchi, and Takashi Okuda, "Ad-hoc Routing Protocol

Avoiding Route Breaks Based on AODV," Proceedings of the 38th

Hawaii International Conference on System Sciences – 2005. 2005

IEEE.

[24] Z. Chenchen and Y. Zhen, "A New EAAODV Routing Protocol based

on Mobile Agent," Systems and Networks Communications, 2006.

ICSNC '06. International Conference on Oct. 2006 Page(s):4 – 4.

[25] A. Sameh and M. Kamel, "Deploying power-aware on-demand (PAOD)

schemes over routing protocols of mobile wireless ad hoc networks,"

INTERNATIONAL JOURNAL OF COMMUNICATION SYSTEMS

Int. J. Communication Systems. 2005. 18:333–346. Published online 27

January 2005 in Wiley InterScience.

[26] QualNet-4.5.1-UsersGuide, available at http://www.scalable-

networks.com/publications/documentation/index.php

[27] Maxim Integrated Products, “Three Methods of Noise Figure

Measurement”, November 21, 2003, available at http://pdfserv.maxim-

ic.com/

[28] T. H. Cormen, C. E. Leiserson, R. L. Rivest, and C. Stein, “Introduction

to Algorithms”, Second Edition, The MIT Press, McGraw-Hill Book

Company, 2001, Page(s): 497 – 498, 504 – 506.

[29] C. Bettstetter, “On the minimum node degree and connectivity of a

wireless multihop network,” in Proceedings of the 3rd ACM

International Symposium on Mobile Ad Hoc Networking and

Computing (MobiHoc ’02), pp. 80–91, Lausanne, Switzerland, June

2002.

[30] M. Saleem, S. A. Khayam, and M. Farooq, “On Performance Modeling

of Ad Hoc Routing Protocols”, EURASIP Journal on Wireless

Communications and Networking Volume 2010 (2010), Article ID

373759, February 2010.


35

Abstract— The Mobile Ad hoc Network (MANET) has

become an increasingly popular area of research during the last

decade. As an emerging fourth generation wireless topology,

the main goal of the MANET will be to seamlessly connect

mobile devices anywhere and anytime. Recently, the

introduction of Ultra-WideBand (UWB) technology has

become a promising candidate to support MANET’s and

research in this area has been extensive over the last few years

due to its powerful capabilities such as the large Band Width

(BW) available and the high data rates possible. Routing

protocols enable mobile hosts (or nodes), such as laptops and

cellular telephones, the ability to communicate with each other

and the design of routing protocols has encouraged researchers

to investigate and develop new strategies that establish and

maintain connections between mobile nodes. The main aim of

the research presented in this paper was to reduce the effect of

GPS error in determining nodes’ positions on the performance

of Modified Location Aided Routing Protocol (MLAR1).

Simulation results along with the analytical model show that

the MLAR1 reduce the effect of GPS error on the performance

in terms of consumed energy, throughput, routing overhead,

reliability of route, and the average end-to-end delay.

Keywords—MANET, UWB, Routing protocol, LAR.

I. INTRODUCTION

IRELESS Local Area Networks (WLANs) provide

wireless access to different types of mobile hosts such

as personal digital assistants (PDA), laptops and cellular

phones. These nodes are equipped with short range

transmitters and receivers, and antennas which may be

omnidirectional (broadcast), highly-directional (point-to-

point), or some combination of the two [1].

In a WLAN environment, routing protocols then enable

nodes to relay data packets if they are within transmission

range, or if they can communicate directly. If they are away

from each other, intermediate nodes are required to establish a

multihop route between the sender and receiver. The wireless

routing protocols that provide this key functionality, in

general, are classified as either topological based or position

based.

• Topological based routing protocols use the existing

information about links in network to flood (or forward)

packets. There are two main routing strategies classified

as topological based; proactive [2] that maintains routing

information for each node in the network and stores this

information in routing tables, such as Destination-

Sequenced Distance Vector (DSDV) [3], Cluster-head

Gateway Switch Routing (CGSR) [3], Wireless Routing

Protocol (WRP) [4], and Optimized Link State Routing

Protocol (OLSR) [5]. The second type is reactive routing

protocols which maintain a route on demand, such as Ad

hoc On-Demand Distance Vector (AODV) [6], Dynamic

Source Routing (DSR) [7], Temporally Ordered Routing

Algorithm (TORA) [4], and Associativity-Based Routing

(ABR) [3]

• Position based routing protocols exploit positional

information to direct flooding towards the destination in

order to reduce overheads and power consumption,

Location Aided Routing Protocol (LAR) [8], GRID [9],

Compass [10], and Greedy Perimeter Stateless Routing

(GPSR) [11] are examples of position based routing

protocols.

Ultra-WideBand (UWB) [12][13] is a radio technology that

has been proposed for use in Personal Area Networks (PAN)

and appears in the IEEE 802.15.3a draft PAN standard. UWB

systems consume very low energy levels, and can be used in

short-range, high-bandwidth (BW) communications systems

where the BW > 500 MHz, at 20% of the center frequency.

UWB is defined to operate between 3.1–10.6 GHz and is

restricted to a maximum allowable power spectral density of

41.3 dBm / MHz corresponding to average transmitted power

of 0.5 mW. Therefore, UWB provides relatively short radio

range but given the spectrum available very high data rates in

excess of 100 Mbps can be achieved, (with bit rates of 55, 110

& 200 Mbps [14]).

UWB is best used for ad-hoc and sensor networks [15], it is

used as part of location systems and real time location systems

such as hospitals and healthcare, short broadcast time, “see-

through-the-wall” precision radar imaging technology [16],

and replacing cables between portable multimedia Consumer

Electronics (CE) devices and in the next-generation of

Bluetooth Technology devices [17].

Global Positioning System (GPS) [18] is used in this

research as the means to determine the position of nodes; for

instance, NAVSTAR [18] (Navigation System with Timing

and Ranging) is a navigational system was originally run with

24 satellites orbiting the earth and their supporting receivers

are on the earth. The distance between the satellite and the

earth is approximately 12,000 miles above the surface, and the

satellites make two complete orbits every 24 hours. GPS

satellites transmit periodic digital radio locations and time

The Effect of GPS Error on MLAR1

Dr. Yahia Hasan Jazyah

[email protected]

W


36

information. Satellites are equipped with precise atomic clocks

and GPS can calculate the height, longitude and latitude.

GPS was used in the MLAR1 [19] to determine nodes’

positions, GPS has an error in determining node’s position up

to 15m, which is the worst case, this research paper study the

effect of GPS error in the worst case on the performance of

MLAR1, and shows that MLAR1 reduces that error’s effect on

the performance of the protocol.

The rest of paper is organized as follows: section 2

summarizes the principles of MLAR1, section 3 presents the

routing strategy of MLAR1 and power issue, section 4 presents

the analytical model of MLAR1 and MLAR1 with GPS error,

section 5 represents the simulation results of MLAR1 with and

without GPS error, and the summary in section 6.

II. MLAR1 OVERVIEW

MLAR1 is based on the conventional Location Aided

Routing protocol, scheme 1 (LAR1). The MLAR1 proposes a

dynamic and static request zone at the same time in order to

consider power as a metric when a route is selected.

Three regions of request zones are assigned; the first zone

represents a rectangle where the source and destination lie in

opposite corners of the rectangle, see Fig. 1.

The second and third zones, see Fig. 2, represent rectangles

with dimensions that are dependent on the dimension of the

first zone, (this is explained later in more detail).

In order to achieve reliability and increase the probability of

finding route, when each node within a request zone receives a

Route Request (RREQ), they respond by transmitting a Route

Reply (RREP) according to the following criteria:

• In each region, a threshold value of residual battery power is

assigned

• The threshold of region 1 (Th1) is greater than that of

region 2

• The threshold value of region 2 (Th2) is greater than that of

region 3.

The basic operation of the protocol is as follows:

The initial RREQ is flooded over the request zone. When a

node within one of the three zones receives a RREQ, it checks

the header of RREQ for the threshold and dimensions of each

region (it should be noted here that positional information is

known by all nodes). If the node is within one of the zones,

and has a residual battery power above the threshold of that

zone, it forwards the RREQ and so on until RREQ reaches to

destination which replies to the first received RREQ using the

reverse path.

III. ROUTING PROTOCOL

The new proposed routing protocol exploits functionality of

LAR1 to improve route reliability and decrease power

consumption.

A. Routing Strategy

Three regions are assigned, and each region has a certain

threshold value of power. Each node within its region and has

a residual battery power greater than the threshold responds to

RREQ by retransmitting it as shown in Fig. 2.

The threshold values should be within the maximum

available power and minimum received power which is

calculated as follows:

(1)

Where: Pr is received power, Pt is the transmitted power,

and N is thermal noise. Whereas thermal noise is calculated

according to the following equation [20]:

(2)

Where T is noise temperature in Kelvin of the input

termination and equal to 290 K which is the standard

temperature, K is Boltzmann constant (= 1.379 x 10-23

Joules/Kelvin), B is bandwidth of carrier signal in Hz, and F is

a constant called the noise factor [21].

Source

Destination

Region1

Region3 Th3

Th2

Th1

Th1>Th2>Th3

Figure. 2 Request zones

Expected Zone

Request Zone

Source

Destination

Figure 1. LAR, scheme 1.

N = T * k * B * F

Pr = Pt – N

Region2


37

When source node wants to send data packets to destination,

it forwards RREQ based on the current information about

location and velocity of destination. First it establishes the

expected zone based on the previous information stored in its

routing table about the destination, this information includes

the speed and position at certain time, source node assumes the

destination in the center of circle where the radius is calculated

according to the speed and time equation whereas the radius is

the speed times the period of traveling, this period is the

difference between the current time and the time when that

information is updated in routing tables. Then the source

establishes the first request zone as shown in Fig. 1, then it

calculates the dimensions of the other two zones where the

dimensions are proportional to the dimension of first zone as

shown in Fig. 2. And then the source forwards the RREQ

including in the header extra information which are the

dimensions of each region, threshold values assigned to each

region.

If RREP is not received within a certain time, RREQ is

reinitiated again.

When intermediate node receives RREQ for the first time, it

checks its position by GPS and RREQ’s header for zones’

information (dimensions) and threshold values, if node is

within the three regions, it compares its residual battery power

to the threshold assigned to its zone, if it is greater than the

threshold, it forwards the message to its neighbors.

The process is keeping running until RREQ reaches its

destination.

When RREQ is received by destination, it forwards RREP

to source using the selected reverse path.

Each node has routing tables that stores routing information

which is updated whenever it receives control packet. When

RREP is received by intermediate node that is located on the

reverse path, it updates its routing table according to the

information included in the header of the received RREP, and

then it forwards the RREP towards the source.

IV. ANALYTICAL MODEL

Simulation representation is the common way to prove the

validation of wireless routing protocols for ad hoc networks.

Mathematical verification is considered a complementing

method for the evaluation of wireless routing protocols. The

following presents the mathematical model of the MLAR1

protocol.

The degree of a node x, d(x), is the number of nodes

directly connected to x., A node is said to be an isolated node

if it has a zero degree [22].

The probability P that each node has at least n0 neighbors is

given by: [23]

20

0

.!

)()n(

0

2

00

rn

en

rdP

ρπρπ −==

(3)

Where P is the probability that a randomly selected node

has n0 neighbors, n0 is Number of node’s neighbors,ρ

is

homogeneous node density, and r0: Transmission range of

node.

Broadcast forwarding probability [24] is the probability that

a node will forward a RREQ message to its neighbors

successfully. RREQs in wireless routing protocols may collide

or get lost due to channel errors or dynamic topology, and so,

this value is calculated and can be represented by the

following equation:

ecrs pppp ..= (4)

Where spis broadcast forwarding probability, rp

is the

probability with which a node will forward a RREQ to its

neighbors, cpis the probability of not experiencing a collision

at the MAC layer, and is the probability that the RREQ is not

lost due to channel errors.

Notice here that rp is 1 in case of wireless routing

protocols that flood RREQ over the whole network, while

rp< 1 in case of position based routing protocols as not all

nodes are involved in rebroadcasting the RREQ.

In position based wireless routing protocols, not all nodes

that receive RREQ will forward it due to either it has received

it by another link or due to forwarding restrictions. In this

work, the network overhead is the number of relayed RREQ

packets, in other words it is a function of Expected Forward

Degree [23].

When source broadcast a RREQ to its neighbors, this RREQ

will be received on average by avgec dPP .. nodes and then

each neighbor node receives the RREQ will forward it by

probability of rp. And so the number of source’s neighbors

that will forward the received RREQ to their neighbors is

avgs dP . nodes. Keeping accumulating the number of

forwarding the received RREQ successfully each hop up to h

hops from the source (until RREQ is received by destination)

yields to the total number of network overhead which can be

represented by: [24]

+

=

=∑ ∏

−

= =

+1

1 1

1 .][)(

,1

h

i

i

j

f

i

savgavgs

avgs

potherwisejdPddP

hifdP

C (5)


38

Where pC is the total expected network overhead, avgd

is

average degree, ][ jd f is expected forward degree of a node

at j hops, and h is number of hops from the source node.

Form the previous equation, the network route overhead is

directly related to the number of hops traversed by a RREQ,

probability of forwarding a RREQ ( rp), and the expected

forward degree of nodes. When rp = 1 (topology based

routing protocol such as AODV), RREQ is flooded over the

whole network, the result is huge network overhead, when rp

< 1 (position based routing protocol such as LAR1), the pC

will be decreased compared to the previous case, increasing

the restrictions on selecting the next hop node leads to less rp

which leads to less pC which means less network overhead as

the case in the MLAR1 protocol. This proves that the network

overhead of the MLAR1 protocol is less than the AODV and

LAR1.

The energy consumed by node is due to relaying RREQ (or

route discovery) and data packets. The following equation

represents the total energy consumed by a protocol: [24]

( ) ( )eavgerreqpeerrepdatotal RdTBCRTmBMBLG ++++= )(

(6)

Where Gtotal is the total energy consumed by a protocol, M

is total number of data packets, La is average path length, Bd is

size of data packets (bits), m is number of paths discovered by

a protocol, Brrep is size of RREP packet (bits), Te is

transmitted energy per bit, Re is received energy per bit, Cp is

the total expected network route overhead, Brreq is size of

RREQ packet (bits), and davg is average degree.

In the equation of total energy consumed by a protocol

(network), there are constant and variable terms. We have the

following constants that will not affect the comparison

between the three protocols, which are: total number of data

packets, average path length if we consider the same path is

selected for the three protocols for simplicity, size of data

packets, number of paths discovered by a protocol where all of

the three protocols are single path routing protocols,

transmitted energy per bit, and received energy per bit.

On the other hand, the following terms are variable and can

affect the energy consumed by the protocol, which are: size of

RREP Packet, size of RREQ packet, average degree, and the

total expected network overhead.

When considering the effect of GPS error on the MLAR1

performance, the number of hops that the RREQ will be

relayed is increased due to the increment of zones' dimensions

(if we consider the worst case of 15m that the error in GPS

positioning will increase the flooded area of MLAR1), then:

h` = h + n (7)

Where: h`: number of hops after adding a positioning error

of GPS to the destination's position, h: number of hops before

adding a positioning error of GPS to the destination's position,

n: the number of extra hops that the RREQ will be relayed by

nodes.

And so, the routing overhead generated by MLAR1 when

the GPS error is added to the destination's position is:

∑ ∏+−

= =

++=nh

i

i

j

f

i

savgavgsp jdPddPC1`

1 1

1 ][)( (8)

Equation 8 represents the total expected routing overhead

when a GPS positioning error is added to destination's

position.

The number of hops is increased by n-hops, it is expected

that network overhead is increased and the consumed power as

well, but the throughput and route reliability are the same in

both cases, on the other hand, the difference in consumed

power is very little as shown in simulation results in the next

section.

The analytic model was designed to validate the MLAR1

protocol. The key factor of the analytic model was the forward

degree value, which is the number of neighboring nodes that

receive the RREQ and forwards it to their neighbors. An

increment in this value affects the routing overhead negatively,

which also adversely influences the energy consumption. The

analytic model proves that the MLAR1 protocol mitigates the

effect of GPS error in determining the destination’s position by

the two restrictions on the selection criteria of next hop node.

If we consider the total number of neighboring nodes that

the source can send them the RREQ is 5 but 2 nodes satisfy

the two conditions of forwarding for example, then Ps = 2/5.


df[j]

2/5 5 2 1 4/25 2

2/5 5 2 2 8/125 5/3

2/5 5 2 3 16/625 7/5

Table 1. Routing overhead of MLAR1

Referring to equation 5, the total routing overhead of

MLAR1, Cp(MLAR1) = 7. When considering the GPS error

(15m), the size of zones are extended but still the same 2

nodes satisfy the two forwarding conditions, which means

Cp(MLAR1+15m error) = 7, and that is an indication that

MLAR1 mitigates the effect of GPS error in determining the

destination’s position, which in turn reflects on the consumed

power in the same way as shown in next section.

V. NETWORK ENVIRONMENT AND SIMULATION RESULTS

The same network scenario in [18] is used, as summary:

number of hosts is 20, the battery power of nodes ranges from

1.0 mAhr and 0.1 mAhr, frequency is 3.5 GHz, SNR

sensitivity is 1.0; which is acceptable ratio to enable receivers


39

to detect incoming signal, temperature is 290 Kelvin which is

the standard temperature, bit rate is 55 Mbps which is (the

minimum bit rate) and the range is up to 20 m as the minimum

bit rate is used which are the typical values [14], and GPS

error is 15m.

Number of zones is three, where the dimensions of second

zone equal to 50% of the firs zone (requested zone), while the

dimensions of third zone equal to 30% of the requested zone.

Constant Bit Rate (CBR) is selected to be used as traffic

generator; it is a UDP-based client-server application which

sends data from a client to a server at a constant bit rate,

whereas 100 packets will be forwarded by source to

destination, each packet is 512 bytes in size; all these values

are applied to QualNet simulator.

Fig. 3 shows that when adding an error of 15 metres to the

destination’s position, the energy consumed in transmit mode

is similar for both cases of MLAR1 and the MLAR1 with

positioning errors. Adding an error of 15 metres means to

extend the first request zone by 15 m. The more nodes that are

involved in relaying the RREQ, the more energy is expected to

be consumed. However, the results taken from the simulator

indicate that the transmitted energy is the same for both cases;

the MLAR1 protocol is restricted by two conditions, the

request zones and the threshold values; the first zone has the

highest threshold value, so increasing the dimensions of first

zone will increase the restrictions on node selection as the next

hop, which mitigates the effect of increasing the zones’

dimensions. On the other hand, nodes of adequate energy are

located within the same positions (real positions) whatever the

dimensions of the request zones are and are distributed

normally over the network, and so, changing the sizes of

request zones does not affect the transmitted energy, and so

whatever the region’s size is, it will not affect the consumed

energy.

Figure 3. Energy consumed in transmit mode.

Figure 4. Throughput

Fig. 4 illustrates the throughput for both scenarios, and in

both cases the data rate is equivalent. Both scenarios select

nodes of high energy, which enables nodes in the active route

to send and receive packets clearly whatever the size of

request zones. Again, this because nodes of high energy that

lie within the transmission range between source and

destination are always selected.

Fig. 5 shows the broadcast packet sent to the channel. It

represents the amount of transmitted RREQ packets, which is

an indicator to the routing overhead.

Fig. 5 shows that both the MLAR1 with error and without

error have the same amount of broadcast packets sent to

channel. This is a logical result given that no node will forward

packets (RREQ) even when the size of the request zone is

extended due to an error in the destination’s position; this is

because of the forwarding restriction placed on the nodes by

means of threshold values which are applied on the nodes

within the request zones.

Figure 5. Broadcasts sent


40

Fig. 6 illustrates the unicast packets received clearly. This

metric is not affected by the GPS error in the destination’s

position. In both cases, the unicast packets are received

clearly, as nodes of residual battery capacity above the

threshold are always selected to forward data packets whatever

the size of request zones.

Figure 6. Unicast packets received clearly

Fig. 7 considers the average end-to-end delay. Both

scenarios achieve the same average end-to-end delay because

the same route is selected whatever the size of request zone.

Average end-to-end delay is the time needed to receive a

RREP by the source node after it sends a RREQ to the

destination. When an error in the destination’s position occurs,

the size of request zones is increased, but the real destination's

position is the same (destination does not move to the wrong

position) whatever the error in its position, this means that the

same route is selected when the error of 15m is added.

Figure 7. Average end-to-end delay

VI. CONCLUSION AND FUTURE WORK

In this paper, the effect of GPS error (15m), added to the

destination’s position, is studied with respect to the

performance of MLAR1.

Both the analytical model and simulation results show the

same results. Errors in determining the destination’s position

through the use of GPS increases the routing overhead, if

nodes of high energy are distributed normally over the

network. This is because more nodes participate in forwarding

the RREQ, and so more energy is consumed in transmitting

and receiving. Nevertheless, the MLAR1 routing protocol

controls the situation by confining the flooded area and

limiting the number of participating nodes in relaying the

RREQ.

REFERENCES



1999.

[2] M. Mauve, J. Widmer, and H. Hartenstein, "A Survey on Position-

Based Routing in Mobile Ad Hoc Networks," IEEE Networks, pages

30–39, Nov. Dec. 2001.

[3] M. Abolhasan, T. Wysocki, and E. Dutkiewicz, "A review of routing

protocols for mobile ad hoc networks," Received 25 March 2003;

accepted 4 June 2003, Elsevier B.V.

[4] M. E. Royer, S. Barbara, and C. K. Toh, “A Review of Current Routing

Protocols for Ad hoc Mobile Wireless Networks,” IEEE Personal

Communications, April 1999.

[5] T. Clausen, P. Jacquet, A. Laouiti, P. Muhlethaler, A. Qayyum, and L.

Viennot, “Optimized Link State Routing Protocol,” in Proceedings of

IEEE INMIC, 2001.

[6] C. Perkins, E. Belding-Royer, and S. Das, "Ad hoc On-Demand

Distance Vector (AODV) Routing," July 2003.

[7] D. B. Johnson, D. A. Maltz, and J. Broch. “DSR: The Dynamic Source

Routing Protocol for Multi-Hop Wireless Ad Hoc Networks. in Ad Hoc

Networking,” edited by Charles E. Perkins, Chapter 5, pp. 139-172,

Addison-Wesley, 2001.



[9] W. H. Liao, Y. C. Tseng, and J. P. Sheu, "GRID: A fully location-aware

routing protocols for mobile ad hoc networks," Proc. IEEE HICSS,

January 2000.

[10] E. Kranakis, H. Singh, and J. Urrutia, "Compass Routing on Geometric

Networks," Proc. 11th Canadian Conf. Computational Geometry, Aug.

1999.

[11] B. Karp and H. T. Kung, "GPSR: Greedy Perimeter Stateless Routing

for Wireless Networks,” MobiCom 2000.

[12] K. Kuladinith, A.s Timm-Giel and C. Görg, “Mobile Ad-Hoc

Communications in AEC Industry”, ITcon Vol. 9 (2004), August 2004

Page(s): 313 – 323 available at http://www.itcon.org/2004/22/

[13] E. R. Green and S. Roy, "System Architectures for High-rate Ultra-

wideband Communication Systems: A Review of Recent

Developments," Intel Labs, available

http://www.intel.com/technology/comms/uwb/download/w241_paper.pd

f

[14] L. Xia, A. Lo, I. Niemegeers and T. Baugé, "An Ultra-Wide Band Based

Ad Hoc Networking Scheme for Personnel Tracking in Emergencies",

this work was performed in the IST Framework 6 EUROPCOM project.

[15] Project SPEARHUW, "Enhanced Sensor and WLAN Networks using

UWB Communications", April 2005.

[16] M. G. D. Benedetto , G. Giancola, and D. Domenicali, "Ultra Wide

Band radio in Distributed Wireless Networks," 2006 IEEE.



1999.

[18] Xu Lin and Ivan Stojmenovic, "GPS based distributed routing

algorithms for wireless networks", research is partially supported by


41

NSERC, available at

http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.47.137

[19] Yahia Jazyah and Martin Hope, “A New Routing Protocol for UWB

MANET”, European Conference of Communications (ECCOM'10),

November 30- December 2, 2010, Puerto De La Cruz, Tenerife, Spain.

ISBN: 978-960-474-250-9

[20] QualNet-4.5.1-UsersGuide, available at http://www.scalable-

networks.com/publications/documentation/index.php

[21] Maxim Integrated Products, “Three Methods of Noise Figure

Measurement”, November 21, 2003, available at http://pdfserv.maxim-

ic.com/

[22] T. H. Cormen, C. E. Leiserson, R. L. Rivest, and C. Stein, “Introduction

to Algorithms”, Second Edition, The MIT Press, McGraw-Hill Book

Company, 2001, Page(s): 497 – 498, 504 – 506.

[23] C. Bettstetter, “On the minimum node degree and connectivity of a

wireless multihop network,” in Proceedings of the 3rd ACM

International Symposium on Mobile Ad Hoc Networking and

Computing (MobiHoc ’02), pp. 80–91, Lausanne, Switzerland, June

2002.

[24] M. Saleem, S. A. Khayam, and M. Farooq, “On Performance Modeling

of Ad Hoc Routing Protocols”, EURASIP Journal on Wireless

Communications and Networking Volume 2010 (2010), Article ID

373759, February 2010.


42

Performance Evaluation of VoIP Codecs over

Network Coding in Wireless Mesh Networks

Erik Pertovt, Kemal Alič, Aleš Švigelj, Mihael Mohorčič

Department of Communication Systems

Jožef Stefan Institute

Ljubljana, Slovenia

[email protected]

Abstract—Voice over Internet protocol (VoIP) is used for

transmitting voice signals in a packet-switched Internet protocol

(IP) networks in real time. For transmitting voice over a wireless

mesh networks (WMNs), the analog voice signal has to be

digitalized, encoded, and packetized. Codecs based on different

Quality of Service (QoS) requirements are used. One of the main

QoS requirements is that packets are transmitted through the

network in real time; one-way transmission time or End-to-End

(ETE) packet delay, and packet delay variation or jitter have to

be lower than thresholds. ETE delay depends on various

parameters; among them is also network delay. Various

mechanisms are used to lower the network delay in WMNs. A

promising mechanism, for improving the performance of

streaming services such as the case also in VoIP, is network

coding. In this paper, we evaluate the benefits of using wireless

network coding for VoIP in WMNs. Network coding procedure

in combination with various VoIP codecs is used to observe the

impact on network delay and jitter of the VoIP application. The

simulation results show that network coding can decrease

network delay and jitter. Moreover, results show that network

coding benefits are codec dependent.

Keywords—Voice over Internet protocol; network coding;

wireless mesh networks; performance evaluation

I. INTRODUCTION

Voice over Internet protocol (VoIP) is a paradigm dealing with delivery of voice application and multimedia sessions over the packet-switched broadband Internet protocol (IP) networks in real time [1]. Voice signal is divided into VoIP packets based on various used codecs, which consider different Quality of Service (QoS) requirements [2]. VoIP packets are transmitted with other IP packets over the network.

Typical representatives of wireless packet-switched IP networks are wireless mesh networks (WMNs), where nodes are connected to each other through multi-hop wireless links forming a wireless access/backbone network [3, 4]. In order to improve the WMNs performance, various mechanisms are used. Among the promising mechanisms, which experience an increasing attention, is also network coding [5]. Instead of using “classical” receive and forward mechanism for packets, network coding combines multiple received packets either from the same or from different traffic flows into one encoded

packet and then forwards it in order to increase the network capacity. In wireless networks, network coding exploits the broadcast nature of the wireless medium, where nodes can overhear packets, which are not destined to them, resulting in new coding opportunities, which enable combining even more packets together [6]. A practical network coding procedure, COPE, is proposed in [7] that encodes two or more packets in a single transmission based on the nodes knowledge on what information (which packets) neighboring nodes have. The procedure was tested in a real WMN deployment, which is of a particular importance [8].

VoIP application is highly exposed to QoS impairment in wireless IP networks, such as WMNs [9], even when using QoS enforcement [10]. VoIP QoS performance can be improved with various mechanisms. As the VoIP is a real-time application and requires specific QoS, the benefits of using the novel mechanisms for VoIP have to be tested.

In this paper, we investigate the benefits of using wireless network coding for VoIP application in WMNs. The VoIP application is presented with the emphasis on the one-way transmission time or End-to-End (ETE) delay and packet delay variation or jitter QoS requirements. We also show various codecs delay characteristics. Moreover, we present network coding for WMNs used to decrease the network delay. In addition, we perform extensive simulations to evaluate the performance of various VoIP codecs when using network coding in WMNs in the sense of a network delay and jitter.

II. VOICE OVER INTERNET PROTOCOL APPLICATION

As the internet is a packet-switched network, the voice of VoIP telephone call has to be packetized before being sent through the network. The packetization is the process of dividing the data of a stream into structured blocks, called packets [11]. The packetization of the voice has to consider the fact that real time delivery of packets has to be performed [12]. For this, different types of codecs (coder/decoder) exist. Codec is a coding/decoding device which samples a voice signal and transforms it into a digitalized form with a predefined bit rate [13]. It also compresses the data of the signal to reduce the bandwidth requirements of established call. Codec selection is a balance between the bandwidth efficiency and the quality


43

(compression level) of transmitted VoIP calls [14]. Some of the most frequently used (standard) codecs for VoIP packet transmissions are G.711, G.722, G.723, G.726, G.728 and G.729 [15].

VoIP application is a real-time application and IP network is not perfectly designed for such applications. The IP network is not as robust as public switched telephone network (PSTN) network in terms of the network reliability. There is no guarantee that packets are successfully delivered in sequential order to the destination, therefore, QoS is not guaranteed. Instead of that, best-effort transmission takes place in IP networks. If the network conditions are bad, a receiver will have difficulties understanding the speaker’s speech. In the worst case scenario, receiver will not be able to understand or hear the speaker at all. In these cases, the conversation through VoIP call is not possible. There are several QoS specifications in the sense of various parameters limitations to be followed. These limitations have to be taken into account in the case of using VoIP.

Parameters with major impact on the VoIP call QoS are: one-way transmission time or end-to-end (ETE) packet delay, packet delay variation or jitter, packet loss rate, bandwidth, out-of-order packet delivery and hardware capacity [16]. The stated parameters have to be under the required threshold values to prevent call degradation that can result in the high delay, the understanding difficulties, etc.

In the following, one-way transmission time and jitter influences on the VoIP QoS will be presented and the threshold values of these parameters, beyond which network should not go if supporting a certain QoS of VoIP application, will be given. The one-way transmission time and jitter parameters are then investigated with network coding in Section 4.

A. One-way Transmission Time

Group TIPHON [17] classifies VoIP application into different network QoS performance classes regarding the voice packets one-way delay [18]. In the case of speech transmission it is a “mouth-to-ear” delay; the delay between the time a packet is sent from the “speaker” and the time a packet is received at the “listener”. The classes are provided in Table I.

TABLE I. VOIP QOS CLASSES REGARDING THE ONE-WAY PACKET

DELAY.

3

(Wide-

band)

2 (NARROWBAND) 1

(Best

effort)

2H

(High)

2M

(Medium)

2A

(Acce-

ptable)

Relative Speech

Quality (one

way, non-

interactive

speech quality)

Better

than

G.711

Equivalent or better

than ITU-T

Recommendation

G.726 at 32 kbps

Equivalent

or better

than

GSM-FR

Not

defined

Not

defined

Delay < 100 ms < 100 ms < 150 ms < 400 ms < 400 ms

NOTE: The delay for best effort class is a target value.

In ITU-T G.114 [19] recommendations is stated that one-way delay should never exceed 400 ms for general network planning. As long as the ETE delay is kept below 150 ms, only a few VoIP sessions may get affected. From the user point of

view, delays up to 290 ms are satisfactory. Delays between 290 ms and 400 ms cause the dissatisfaction to some users. Delays above 400 ms can only be used if we suppose that the user is familiar with higher delay as, e.g., in a satellite communication.

VoIP application delay has different causes [20]: coding/ encoding, packetization, jitter buffer and network delay (or network latency). The one-way delay caused by the first three stated causes is described as a codec delay. It can be calculates as:

CodecDelay = CSI + CPP + CPP + JBS (1)

CPP is so-called pooling period of Central Processing Unit (CPU) or CPU pooling period and is ½ of the CSI, which stands for codec sample interval and is the time interval of a speech a codec takes and handles at once. JBS represents jitter buffer size. For example, CSI of codec G.711/10 is 10 ms. Thus, CPP is 5 ms and the recommended JBS for G.711/10 is 20 ms. CodecDelay results then in 40 ms.

Network delay occurs as the packet is sent through the network. However, it cannot be defined, as delays, presented above. In WMN, packet is sent through several wireless routers to be delivered to its destination. Different packets can be routed through the network with different speeds resulting in the variable delays of packets on the receiver.

B. Jitter

Jitter describes a non-constant packet delay at the receiver as the packet latency can vary when packets are sent across the IP network [14]. Jitter can occur when packets of the same stream are sent via different routes through the network. Beside this, it can occur as the traffic intensity of a network can vary through the time thus delaying packets differently. Expected jitter influences the size of a jitter buffer. Higher the jitter, greater the size of a jitter buffer needed to compensate the difference in the delay of packets of the same stream at the receiver. This buffer enables a continuous speech. The jitter buffer size is the same or a multiple (i.e., 1, 2, 3) value of CSI interval. In Table II, delays are represented for various codecs when taking also into account a jitter buffer delay besides the delays in (1), assuming the jitter buffer size of two CSI (i.e., 2 * CSI). Network delay is not considered here.

TABLE II. ONE-WAY CODEC DELAYS FOR VARIOUS CODECS.

One-way codec delay [ms]

G.711/10 40

G.711/20 80

G.711/30 120

G.723/30 120

G.723/60 240

G.729/20 80

G.729/40 160

Jitter is measured as difference in ETE delays between the two consecutive packets of the same VoIP call stream. The jitter values greater than 100 ms are causing delays which are above ITU-T organization recommendations. Jitter values from


44

100 ms to 200 ms can be still handled by some jitter buffers introducing some conversational problems. If the packet arrives at the VoIP device too late (i.e., out of the jitter buffer value), it is lost. In the context of a network, packet jitter is measured as the average of all jitter packets values.

III. NETWORK CODING FOR WIRELESS MESH NETWORKS

Network coding is the mechanism to improve the network performance. It experiences an increasing attention in the past few years in both, wired and wireless networks, mainly due to promising results from the initial research and testbed deployments [7, 8].

Network coding enables encoding multiple packets either from the same or from different traffic flows into one encoded packet for saving bandwidth and thus increasing the network capacity while maintaining the desired Quality of Service parameters. It can be also used to decrease the network delay, as will be demonstrated in Section 4. In wireless networks, network coding exploits the broadcast nature of the wireless medium, where nodes can overhear packets which are not destined to them, resulting in new coding opportunities [6]. These packets are later on needed for decoding process.

The network coding principle is presented in Fig. 1, where it is assumed that we have wireless nodes (e.g., wireless routers). Nodes S1 and S2 has to deliver packets m1 and m2 to nodes D1 and D2. Without network coding, packets are first sent to a relay node R and then forwarded to its corresponding destinations. Therefore, four transmissions are required to deliver packets. While with network coding, three transmissions are only required to deliver packets, as both packets are encoded into one packet (linear operation over the two packets) on node R, which is then forwarded to both destinations. Therefore, only one transmission is required by node R. The coding is possible as D1 knows m2 as it hears node S2 and can decode m1 from encoded packet sent from node R. Similar, D2 knows m1 as it hears node S1 and can decode m2 from encoded packet sent from node R. With this, one transmission has been saved.

One of the well-known network coding procedures for increasing the throughput of a WMN is COPE procedure, which is described in the following.

A. COPE Network Coding Procedure

COPE [7] is an intra-session network coding algorithm,

which exploits the broadcast nature of the wireless medium. It codes packets for one hop, where packet decoding is done. The coding process depends on the nodes knowledge on what information (which packets) neighboring nodes have. In case the node knows which information neighbors have (through listening to neighbor’s broadcasts (packets and ACKs) or receiving their updates) the coding process is straightforward and the decoding process will have a high success rate. Information arriving through particular massages and through listening to all the broadcast, is not sufficient and provides only few coding opportunities. In the case that the information on the packet presence at specific neighbor’s node is not available the coding needs to guess on the situation. The node estimates probability that the node A has packet P, by looking at the delivery probability between packet’s previous hop and node A. With all the needed information the node can code together as many packets as possible, as long as none of the packets have been created on this node, all the packets have different next hops and we know that there is a strong possibility that each next hop (all the neighboring nodes that we are encoding packets in for) will be able to decode the packet. The next hop can decode the packet if it has already received all except one of the packets coded together.

IV. PERFORMANCE EVALUATION OF VOIP USING NETWORK

CODING

We performed the evaluation of VoIP with network coding using network coding simulation model, which we presented in [21, 22]. The simulation model has been built using OPNET Modeler [23] simulation tool. In this chapter, we present and analyze the results obtained by simulation runs in the simulation model. We compare simulation results when using VoIP without network coding to the simulation results when using VoIP with COPE network coding procedure.

The performance of VoIP with network coding was tested in different network topologies and simulation results were collected for each of them. After analyzing the results, one network topology was chosen for the representation as an example, although the similar results were obtained by different topologies. The results from the presented network topology were chosen to present the VoIP performance using various codecs in a typical WMN with or without network coding.

A. Simulation Parameters

The parameters used in our case are numerous. In this chapter, we describe the main parameters that are used in different simulation runs.

In our analysis, we assume that all wireless network nodes are of the same type and have identical configuration, representing homogeneous network. Networks with different number of nodes and topologies are investigated, where each node is given a random location within a given area. A typical network topology for WMNs with 10 wireless nodes and 3 neighbors per each node, depicted in Fig. 2, has been selected and is further analyzed in this paper. The wireless nodes have been randomly positioned within the square size of 2000

Fig. 1 Presentation of wireless network coding principle.


45

meters per 2000 meters (i.e., 2000 m * 2000 m), which is the size of the simulated wireless environment.

Each node has 1 Mbit/s of channel bandwidth. Wireless connections established between neighbors, which are represented in the network topology as wireless links, are graphically presented in Fig. 2 with dashed lines between nodes. For the simulation purposes all the links are symmetrical are lossless, meaning that no packets get lost during transmissions. Lossless links means that network conditions have to be perfect or there are some good wireless connections in the network, which can be selected as wireless links. Moreover, packets on wireless links are delayed due to propagation through wireless medium.

In the simulation, VoIP application is simulated establishing VoIP calls between node pairs. VoIP call is simulated with two packet streams being sent between the two wireless nodes, which are representing the two speakers of a VoIP call. VoIP calls are established between each node pair in the network. For a network topology with 10 wireless nodes in Fig. 2, it results in 45 individual calls. Only one VoIP call is established at the same time in the network. Each simulated VoIP call lasts for 30 seconds. In Table III, the parameters for various used codecs are presented. The total size of VoIP packet and the number of VoIP packets sent each second (i.e., packet per second, PPS), are calculated for various codecs used in simulation, considering the fact that VoIP application is implemented in 802.11b WMN. The VoIP packet total size represents the size of voice payload data (i.e., the codec sample size, CSS) and 802.11b overhead in one VoIP packet. The CSS depends on the codec bit rate (CBR), which determines the number of bits per second that has to be sent to deliver a voice call, and, the codec sample interval (CSI), which is the time interval of a speech a codec takes and handles at once. PPS represents the number of packets that has to be transmitted every second in order to deliver the codec bit rate. It depends on the CSI. In addition, traffic load per second, produced by VoIP call every second, is calculated by multiplying PPS value and VoIP packet total size for each codec in Table III. The results of various codecs are then compared between them.

TABLE III. CODECS PARAMETERS.

Codec CBR

[kbit/s]

CSI

[ms]

CSS

[bytes]

PPS

[pps]

VoIP Packet

Total Size

[bytes]

Traffic load

per second

[bytes]

G.711/10 64 10 80 100 178 17800

G.711/20 64 20 160 50 258 12900

G.711/30 64 30 240 33 338 11154

G.723/30 6.4 30 24 33 122 4026

G.723/60 6.4 60 48 16 146 2336

G.729/20 8 20 20 50 118 5900

G.729/40 8 40 40 25 138 3450

Background traffic is simulated all the time during performing VoIP calls. It is simulated to evaluate its impact on the performance of VoIP calls. Background traffic load is generated as packet streams between all nodes with the same intensity using exponential distribution of inter-arrival times and constant packet lengths (i.e., 10 kbit). The background traffic load is increased through simulation runs until the VoIP packet delay in the network is not being increased due to this traffic and VoIP traffic can not be handled any more by the network to have a feasible speech communication. All network nodes are source nodes for generating background traffic with the same probabilities and select destination nodes using uniform probability distribution among all network nodes. Results are presented for six different intensities of traffic background loads (i.e., for six different total amounts of background traffic sent into the network), denoted by L1, L2, L3, L4, L5, and L6. L1 represents the lowest network traffic intensity used in the presented results, while L6 represents the highest intensity (i.e., the intensity, when network is already congested). It means that the used intensities increase as follows: L1 < L2 < L3 < L4 < L5 < L6. The network diameter is 3. The network average hop count is 2.

COPE network coding procedure [7] has been used for encoding packets for increasing the network throughput. The simulation cases without network coding are compared with the cases when COPE is used in the network to evaluate the impact of network coding on the performance of VoIP application in WMN.

Important modification has been made to the COPE procedure to increase packet delivery reliability at the network coding layer. Instead of using cumulative ACKs as described in the original paper each coded packet is immediately confirmed with the individual ACK packet. This allows us to shorten the round time and schedule possible retransmissions sooner. This is an important modification as it lowers the jitter and decreases the possibility of receiving packets with delay higher than expected by QoS parameters. The individual ACKs increase the overhead in the network and thus lower the network goodput.

Routing of packets through the network was done using static tables, which were calculated ahead of simulation runs. Routing tables are calculated using Dijkstra’s algorithm taking into account hop count distances between nodes.

The timeline of simulation was as follows. Every simulation run took 1400 seconds. The background traffic was

Fig. 2 Network topology with 10 nodes and 3 neighbors per node.


46

generated during the whole simulation run. The time of 5 seconds (warm up time) is required at the beginning of the simulation to have steady state conditions. Only one VoIP call between two wireless nodes was established at the same time in the network using one of the codecs in Table III. VoIP calls were generated consecutively with the 1s delay between them. In one simulation run, 45 individual calls were simulated and each VoIP call lasted 30 seconds resulting in 1350 seconds of VoIP calls simulation. At the end, 5 seconds are used for simulation control purposes as, e.g., the intensity of network congestion in the case of high background traffic, which is detected by receiving VoIP packets at the receivers also after the 1355

th second of simulation, up to the 1400

th second. In

each simulation run, the background traffic was increased.

B. Simulation Results

We have averaged the network delay of all calls established in one simulation scenario. In every scenario, a particular codec has been used for transmitting VoIP calls. To simulate different traffic densities in the network, we have created different amounts of background traffic. Then, we evaluated how background traffic affects the VoIP application performance with various codecs. The results are presented in Fig. 3.

From Fig. 3, we can see that delays are increasing with the increased background traffic in the network for all codecs, as expected. Codec G.711/10 has the highest average network delay, while G.723/30, G.723/60, G.729/20 and G.729/40 have lower delay. This is because of the specific traffic load per second a particular codec has, which is presented in Table III. Please note that we do not present the scenarios, where the network gets congested (i.e. delays goes towards infinity). Therefore, there is no mark for these scenarios on the graph in Fig. 3 (see curves going into “infinity”, out from the figure). Similar is also done in the figures, which are presented in the following. Background traffic loads (e.g., L4, L5, L6), when delays are very high, cause (in some cases) network congestion, when using a particular VoIP codec. It means that we are presenting the results, when network is highly loaded or is already congested, with the exception for L1.

We have done the same for jitter measurements. In Fig. 4, jitter is presented for various used codecs in dependency of background traffic load. We can see that average jitter is increasing with background traffic, but not so rapidly as network delay in Fig. 3.

After analyzing the VoIP performance without network coding, we have also performed simulations, when COPE network coding procedure has been used in the WMN network to increase the throughput of the network. The scope of that was to investigate the impact of network coding on the VoIP performance. In Fig. 5 and Fig. 6, network delays and jitters are presented for the cases, when network coding (i.e., COPE procedure) is used on wireless nodes in the network. The results are presented for the same scenarios as in Fig. 3 and Fig. 4.

When using COPE, average network delays are lower, when background traffic load is high, compared to the cases when network coding is not used in the network. This difference between the COPE and no-COPE is increasing with

L1 L2 L3 L4 L5 L60

50

100

Network delay (no coding)

Network load

Del

ay (

ms)

G.711/10

G.711/20

G.711/30

G.723/30

G.723/60

G.729/20

G.729/40

Fig. 3 Network delay when network coding is not used.

L1 L2 L3 L4 L5 L60

10

15

Jitter (no coding)

Network load

Jitt

er (

ms)

G.711/10

G.711/20

G.711/30

G.723/30

G.723/60

G.729/20

G.729/40

Fig. 4 Jitter when network coding is not used.

L1 L2 L3 L4 L5 L60

50

100

Network delay (COPE)

Network load

Del

ay (

ms)

G.711/10

G.711/20

G.711/30

G.723/30

G.723/60

G.729/20

G.729/40

Fig. 5 Network delay when COPE is used.


47

the increased background traffic, as expected. More packets are in the network, more coding opportunities arise and more packets can be encoded, thus saving more bandwidth at the transmission. Moreover, it can be seen from Fig. 5 that VoIP application using COPE, in most cases (not true for codec G.711/10 with L6), still performs well having L4, L5 and L6 background traffic in the network, while without network coding the VoIP application is degraded due to high delays of VoIP packets (represented with no marks on the graph). Here, we can conclude that network coding improves the performance of VoIP application in WMN, especially when the network is high loaded or overloaded to a certain point. For jitter values, the difference between COPE and no-COPE case is very small, so the improvement is, in most cases, negligible. It is worth noting that using COPE does not increase the value of jitter.

In addition, we have investigated the difference in the impact of using network coding with different VoIP codecs. We have compared the scenario of using COPE and the scenario when network coding is not used for various codecs in Fig. 7. The comparison of network delays and jitters, in dependency of different background traffic loads, using COPE procedure and without using network coding, is presented for various VoIP codecs, separately.

It can be seen that codecs, which require higher traffic load per second, benefit from network coding more than codecs with lower traffic load per second. Once more, this is due to the fact that more VoIP packets are encoded with other packets (because of the increased overall traffic load in the network), thus increasing more the capacity of the network with network coding.

We can conclude that VoIP application benefits from the use of network coding in WMN, as the network delay is decreased and the VoIP performance is improved when the network traffic is high or the network is already congested to a certain point. Moreover, network coding does not degrade jitter.

L1 L2 L3 L4 L5 L60

10

15

Jitter (COPE)

Network load

Jitt

er (

ms)

G.711/10

G.711/20

G.711/30

G.723/30

G.723/60

G.729/20

G.729/40

Fig. 6 Jitter when COPE is used.

L1 L2 L3 L4 L5 L60

50

100

G.711/10: Network delay

Network load

Del

ay (

ms)

no coding

COPE

L1 L2 L3 L4 L5 L60

10

15

G.711/10: Jitter

Network load

Jitt

er (

ms)

no coding

COPE

L1 L2 L3 L4 L5 L60

50

100


Network load

Del

ay (

ms)

no coding

COPE

L1 L2 L3 L4 L5 L60

10

15

G.711/20: Jitter

Network load

Jitt

er (

ms)

no coding

COPE

L1 L2 L3 L4 L5 L60

50

100


Network load

Del

ay (

ms)

no coding

COPE

L1 L2 L3 L4 L5 L60

10

15

G.711/30: Jitter

Network load

Jitt

er (

ms)

no coding

COPE

L1 L2 L3 L4 L5 L60

50

100


Network load

Del

ay (

ms)

no coding

COPE

L1 L2 L3 L4 L5 L60

10

15

G.723/30: Jitter

Network load

Jitt

er (

ms)

no coding

COPE

L1 L2 L3 L4 L5 L60

50

100


Network load

Del

ay (

ms)

no coding

COPE

L1 L2 L3 L4 L5 L60

10

15

G.723/60: Jitter

Network load

Jitt

er (

ms)

no coding

COPE

L1 L2 L3 L4 L5 L60

50

100


Network load

Del

ay (

ms)

no coding

COPE

L1 L2 L3 L4 L5 L60

10

15

G.729/20: Jitter

Network load

Jitt

er (

ms)

no coding

COPE

L1 L2 L3 L4 L5 L60

50

100


Network load

Del

ay (

ms)

no coding

COPE

L1 L2 L3 L4 L5 L60

10

15

G.729/40: Jitter

Network load

Jitt

er (

ms)

no coding

COPE

Fig. 7 Network delay and jitter for various codecs with and without

COPE.


48

V. CONCLUSION AND FURTHER WORK

This paper evaluates the use of network coding for VoIP application performance benefits in WMNs. We present the VoIP application using various codecs to transmit voice signal in a packet-switched IP networks in real time. Furthermore, we describe VoIP QoS requirements in the sense of one-way transmission time and jitter. Then, we describe the use of network coding in wireless mesh networks and present the well-known COPE procedure for network coding. After that, we compare the use of VoIP with and without using network coding in WMNs. The simulation results show that network coding can improve the VoIP performance in WMNs especially when the network is highly loaded or congested. Network coding decreases the network delay while the influence on jitter is small. The benefit of network coding depends on the used VoIP codec. Codecs, which require higher traffic loads per second, benefit from network coding more than codecs with lower traffic loads per second.

In further work, the use of VoIP with network coding in WMN should be also evaluated in more details. For example, the VoIP scenarios presented in the paper should be also analyzed at the level of individual calls; network delay, jitter and packet loss variation should be investigated throughout each call all the time.

REFERENCES

[1] M. E. Fernandes, C. R. C. Lima, J. Schimiguel, “Strategic management using VoIP technology: a case study in a call center company,” WSEAS Trans. Comm., vol. 10, no. 1, pp. 34-43, 2011.

[2] M. Voznak, “Speech bandwidth requirements in IPsec and TLS environment,” Proc. 13th WSEAS Int. Conf. Comp., pp. 217-220.

[3] D. Benyamina, N. Hallam, A. Hafid, “On optimizing the planning of multi-hop wireless networks using a multi objective evolutionary approach,” Int. J. Commun., vol. 4, no. 4, pp. 213-221, 2008.

[4] I. F. Akyildiz, X. Wang, “A survey on wireless mesh networks,” IEEE Comm. Mag., vol. 43, no. 9, pp. S23-S30, September 2005.

[5] R. Ahlswede, N. Cai, S.-y. Robert Li, R. W. Yeung. “Network information flow,” IEEE Trans. Inf. Theory, vol. 46, no. 4, pp. 1204-1216, July 2000.

[6] C. Fragouli, D. Katabi, S. Katti, A. Markopoulou, M. Medard, H. Rahul, “Wireless network coding Opportunities and challanges,” Proc. Milit. Comm. Conf., 2007.

[7] S. Katti, H. Rahul, W. Hu, D. Katabi, M. Médard, and J. Crowcroft, “XORs in the air: Practical wireless network coding,” IEEE/ACM Trans. Networking, vol. 16, no.3, pp. 497-510, June 2008.

[8] M. Conti, J. Crowcroft, A. Passarella, “Multi-hop ad hoc networks from theory to reality,” Nova Science Publisher, Inc. Commack, NY, USA, 2008.

[9] R. Libnik, G. Kandus, A. Svigelj, “Simulation model for performance evaluation of advancde SIP based mobility management techniques,” Int. J. Commun. , vol. 5, no. 1, pp. 26-35, 2011.

[10] S. Munir, I. Ahmad, “VoIP on wireless LAN: a comprehensive review,” 7th WSEAS Int. Conf. Elect., Hard., Opt. Comm., Cambridge, pp. 225-230, UK, February 2008.

[11] B. Ngamwongwattana, “Effects of packetization on VoIP performance” 5th Int. Conf. ECTI-CON, vol. 1, pp. 373-376, May 2008.

[12] B. Goode, “Voice over Internet protocol (VoIP),” Proc. IEEE, vol. 90, no. 9, pp. 1495-1517, September 2002.

[13] R. S. Naoum, M. Maswady, “Performance evaluation for VoIP over IP MPLS,” World Comp. Sc. and Inf. Tech. J. (WCSIT), vol. 2, no. 3, pp. 110-114, 2012.

[14] V. Skorpil, D. Novak, “Service for advanced communication networks,” WSEAS Trans. Comm., vol. 9, no. 10, pp. 636-645, October 2010.

[15] R. S. Ramakrishnan, V. Kumar, “Pefromance analysis of different codecs in VoIP using SIP,” Nat. Conf. Mobile Perv. Comput. (CoMPC), India, 2008.

[16] Y. Ogunjimi, “Practical analysis of voice quality problems of Voice over Internet protocol (VoIP),” Bachelor’s Thesis, USA, 2012.

[17] ETSI TIPHON TR 101 329-2 (2002), Quality of Service (QoS) Classes.

[18] ITU-T recommendation Y.1541, Internet protocol aspects – Quality of service and network performance, December 2011.

[19] ITU-T G.114 ITU-T recommendation G.114, One-way transmission time, May 2003.

[20] Cisco Inc., While Paper. Document ID: 5125, Understanding Delay in Packet Voice Networks, 2008.

[21] K. Alič, E. Pertovt, A. Švigelj, “Simulation environment for network coding,” Proc. Mosharaka Int. Conf. Comm. Net. Inf. Tech. (MICCNIT 2011), pp. 1-6, Dubai, UAE, December 2011.

[22] K. Alič, E. Pertovt, A. Švigelj, “Network coding simulation model in OPNET Modeler,” OPNETWORK 2012, pp. 1-7, Washington, USA, August 2012.

[23] OPNET web page. URL http://www.opnet.com/.


49

Delay Factors Modelling for Real-Time

Traffic Information Systems

Marius Minea Dept. Telematics and Electronics for Transports

Transports Faculty, University Politehnica of Bucharest Bucharest, Romania

[email protected]

Iulian Bădescu Dept. Telematics and Electronics for Transports

Transports Faculty, University Politehnica of Bucharest Bucharest, Romania

[email protected]

Abstract—The paper is presenting an overview of the factors that lead to critical delays which may occur in multi-modal real-time traffic and travel information systems (MRTTI) employing cellular data network communications. The analysis is taking into account MRTTI systems that use mobile devices to detect position and speed of a vehicle and receive traffic information furthermore used to adjust a previously defined route from origin to destination points. Information transmission delays are also investigated for route guidance systems used in critical applications, such as emergency vehicles management. The identification of the delay factors and some models regarding the delays are presented along with measurements on a real system.

Keywords — real-time traffic information, information propagation delay, jitter

I. INTRODUCTION

The systems used for on-board navigation are well known for their advantages and quite widely spread in road traffic amongst drivers. However, the large majority of such devices are simply GPS-navigation devices with no possibility to use real-time traffic triggered information. The number of such devices that employ traffic and incident actuation in order to adjust a route from an origin O to a destination D is still limited and relying on the coverage area of the system used for communication and to the existence of a traffic management system, used as primary information source. If a RDS-TMC1 system is used, then the communication is a simplex-type and information is flowing only from the RDS-TMC centre to the mobile device. In more sophisticated systems such as multi-modal real-time traffic and travel information systems, the mobile device detects the user’s position and computes a route to a chosen destination based on information requested to a central computer. This information may contain data regarding the actual status of a specific location (e.g., a parking place), along with traffic restrictions or route obstructions on the route to that destination, collected via several traffic and public transport management systems, or parking systems. Emergency vehicles routing to traffic incident is also relying on actuated information regarding current vehicle position and traffic ahead. In some situations, receiving correct information in due time may become critical, especially when such an emergency

1 RDS-TMC – a simplex system used for delivering traffic information to onboard navigation devices (Radio Data System – Traffic Message Channel)

vehicle also needs traffic signals actuation and prioritization. The paper identifies the main factors causing delay in this information chain and tries to investigate some appropriate mathematical models for evaluating them.

The paper is structured in the following parts:

Chapter II is dedicated to the description of the multi-modal real-time traffic and travel route guidance systems architecture and identification of the main delay factors in the information transmission chain. Chapter III is proposing some mathematical models for the main delay factors (with a focus on multimodal route guidance and emergency vehicles navigation systems). Chapter IV performs a brief analysis on the central layer time-shifting, taking into account the numerous processes it has to perform. Chapter V is considering some tests performed on a real system [5] and presents diagrams with recorded delays in information delivery times and table with thresholds accepted for multimodal route computing and information delivery. Finally, chapter VI offers some conclusions and possible future actions to improve the response time of MRTTI systems.

II. IDENTIFICATION OF DELAY FACTORS IN MRTTI

ARCHITECTURES

A. Information flowing in complex MRTTI route guidance systems

A complex MRTTI route guidance system is composed of several layers, each of those contributing to the global response time. The main elements of such a system are presented in figure 1. In such systems, the information is firstly analyzed locally in the mobile device and then additional data is requested from the central layer (the MRTTI centre). The cellular network introduces the delay stage D4. When a request from a mobile device is received, the central computer searches for actuated information for the interest zone. If such information is not available or is outdated, then it formulates its own request to information providers (transferring information induces delays in layers D2 and D1).

The central element of this architecture is composed by the MRTTI management centre, where information is collected from various sources and processed (format adaptation, map routing, additional information adding etc.). This layer introduces a typical delay noted as D3. When the information


50

in appropriated format of data is completed, the central computer re-sends actuated information to the mobile device.

Fig. 1. The layers and main delays introduced by MRTTI navigation systems

architecture

It can be observed that in some cases, when a request of information is issued by a mobile device, some of the presented layers have a double contribution at the global time-shifting recorded at the final user.

B. Information Flowing in Incident/Traffic Actuated Route Guidance Systems for Emergency Services

Unlike the above presented MRTTI systems, in an emergency traffic actuated route guidance system, the information is only flowing from the dispatcher computer (or operator) to the mobile on-board navigation device. The only information a mobile device sends is the position (and/or other on-board parameters, e.g., identity tag, speed or heading, fuel level etc.). Therefore, the delays that may induce errors in navigation calculations are only produced on the chain from the dispatcher to the mobile device, i.e. communication delays. Figure 2 presents these aspects of the analysis.

Layers 1 and 2 contribute less than in the previous case to the information delaying; due to the fact that related data is permanently resident in the system (such systems have priority compared to civil appliances, and in some cases they originate in the network the information regarding an incident2). The most significant part of the time shifting is identified to be produced by the communication system and central data processing. If the emergency route guidance is linked with a

2 The e-Call systems employ incident-triggered message sending via cellular network

traffic actuated controller, then the message has to arrive at the vehicle before the vehicle reaches the next traffic light.

Fig. 2. The layers and main delays introduced in emergency services route guidance systems

The process of route guidance in this case is simpler, because the critical information only flows from the dispatcher to the emergency vehicle, thus offering shorter information delivery delays.

III. MODELING DELAYS IN INFORMATION TRANSMISSION

CHAIN

A. Analysis of Delays in Real-Time Multimodal Systems

As stated before, the MRTTI systems employ a double-communication for each route and information-related request. Therefore, in the analysis of time shifting, several stages have to be considered twice in the calculation of the global delay. The most significant delays are introduced as following:

In the simple routing case, the local processing (D5, figure 1) – the mobile device uses its own sensors (usually satellite navigation sensing, such as GPS) to detect position. Let’s denote this delay . This includes (1): position detection delay (with presumed GPS hot-start timing), local information processing delay and route computing time

(1).

In a multi-modal routing case, the local processing (D5, figure 1) includes a multi-modal route computing, which induces a different , larger than the previous one, due to the fact that the processor has to compute several (multi-mode) routes from the actual position


51

(origin) to a selected destination. This selected destination may be inferred from a request to the central computer (for example, when selecting public transport, a person would like to know where is located the public transport vehicle stop, or when requesting for the nearest parking place – the mobile device asks for this information the central computer). Let us denote in this case the route computing time for multimodal route guidance and the local processing time (other than route calculation) in multi-mode transport routing. It is obvious that

and (2)

(3)

where τ is the local processing delay in layer D5, in the case of a multi-modal traffic information system.

The next stage where time shifting occurs is the mobile (cellular) communication network. Delays (D4 in figure 1) in information transmission, in this case, are dependent on the density of users and availability of modern generation installations (e.g., 4G - LTE3) equipment. The necessary time for the information to pass over the communication network is in the first case

(4)

where denotes the time shifting introduced by the mobile network, and are delays in the network on the up-link (request of information) and down-link (answer) paths, and represent random delay factors on the up-link and respectively on the down-link transmission paths, depending on the network coverage and accessibility in difficult reception areas (such as tunnels). Usually, and induce severe delays when the signal is lost (ranging from tens of seconds to minutes). The influence of the number of local users is represented in equation (4) by the random network congestion factor . For estimating the factor ν, in previous works [7] a model was represented by an adaptation of the Pollaczek-Khinchin formula, where the queuing delay becomes

∙

⋅ (5).

In equation (5), δ stands for data packet delay at queuing, represents the average service time at first

transmission, the average service time at second transmission (retransmission) and λ the arrival rate of messages at the queue. Considering a coefficient for the congestion of the network due to multiple simultaneous transmission requests, we can further develop the model such as

∙⋅ (6)

3 4G-LTE – fourth generation of GSM data transmission technology – Long Term Evolution, that allows for rapid transfer rates

where ∈ 0.8 0.95 is a random coefficient directly dependant on the number of instant cell users and is a geometrically distributed random variable. The service time is proportional to the number of the transmission attempts and inversely with the cluster (data packet) size. The coefficient is containing this information.

Relevant MRTTI data and electronic-format maps4 transmission through Internet Protocol require consistent, non-restrictive network bandwidth as well as the navigation computer hardware sufficient to support not only the communications of the agent but also other tasks being performed on the navigation computer. Delays can occur in at least two ways:

o Delays due to propagation, . This type of time shifting is related to the distance that IP packets must travel through the network, the number of hops that packets take through a network, the available bandwidth of that network, and other network traffic.

o Delays due to encryption and messages processing, . VPNs encrypt data in order to ensure privacy over an otherwise public network. Encryption of packet data takes time and can increase delay in a network. The VPN should be implemented with hardware acceleration in order to minimize delay. If a PC is used as one of the endpoints in a VPN connection, the PC may introduce a large delay in processing the encryption/decryption algorithm.

However, when transiting a VPN network, the total time a data packet is taking from source to destination (end-to-end delay) is composed of i) the processing delay, ii) the queuing delay, iii) the serialization delay and iv) the propagation delay. The total delay produced by the VPN communication is modelled with

2 (7)

where denotes a random coefficient depending on network load, a random coefficient depending on the variation of the delay (also called jitter5) and the average variation of delay on a defined period of time.

Delays produced by the third-party information providers. In MRTTI systems usually the information related to traffic congestions or events on a route is collected via third-party systems, such as (figure 1) traffic management systems. Except the time shifting produced solely by the third-party system operation, there are some delays that may also be modelled i.e. conversions of different data formats (presented as D1

4 Delays due to processing operations for formatting of maps, such as WFS (Web Feature Service) or WMS (Web Map Service) 5 The jitter represents an undesired deviation from true periodicity of an assumed periodic process (or signal, in electronics or telecommunications)


52

in figure 1). Usually, this process is carried on by a so-called Commonly Agreed Interface, or CAI. Let us denote these delays with .

Concluding and considering (1), (3), (4) and (7), the total end-to-end delay in simple route guidance systems (using RDS-TMC) may be written as

(8)

with being the RDS-TMC channel communication delay.

For MRTTI systems the total end-to-end delay on the information chain becomes:

(9).

B. Packet Delaying in IP Networks and VPN Networks

In order for a MRTTI system to offer good service and with a suitable security, VPN networks, equipped with QoS6 are to be employed in its architecture [1], [2]. VPN networks that are suitable for such applications may be those enabling QoS also suitable for providing business-class voice along with video support that meet the following requirements [2]:

end-to-end delay 150 ms

end-to-end jitter 50 ms

end-to-end packet loss less than 1 percent.

The usual predictor used for modeling end-to-end delays in current TCP7 networks is based on Jacobson’s [4] algorithm, where a TCP sender records a “round trip time (RTT)” when it receives a correct acknowledgement packet. Considering τ n the observation for the nth packet, we obtain a recursive model for τ :

1 1 (10)

where has recommended values around 1/8.

IV. ANALYSIS OF THE CENTRAL LAYER TIME-SHIFTING

The central layer (MRTTI dispatcher) has a substantial contribution – denoted as D3 – on the overall time-shifting produced by the system. In fact, if we consider a correct service response from layer 1 (Information Providers, D1), then the most important delay is produced by D3. This is due to the fact the CAI and the Regional Data Service Server (RDSS) have to convert, process, re-convert and set ready for broadcasting information in different formats (such as maps presentation to final user). Another option is that the RDSS not only provides dynamic traffic information as standalone data, but also generates dynamic vehicle routing itself by means of provision of waypoints for navigation to the traffic information service providers (figure 3). The most time-consuming service from the

6 QoS – Quality of Service, a concept for IP networks that manages the following elements: bandwidth (the amount of data that can be transmitted at once), delay (time to send data from source to destination), jitter (variation in delay) and reliability (packet losses). 7 TCP – Transfer Control Protocol

whole palette of services offered by the MRTTI system is, according to field measurements, the Dynamic Traffic Information Service (DTI). This service has to acquire real-time traffic information, to convert it regularly into appropriate formats and to constantly update relevant data to users that travel on a route that involves that information.

Fig. 3. The service chain that produces the most significant time-shifting in data processing – in the central RDSS core

It is difficult to obtain a very close mathematical model of the overall time-shifting produced by this segment D3, due to the fact the service itself has a high complexity [5]. Figure 4 shows the use case diagram for the main operations produced in this layer D3.

Fig. 4. Use-case diagram showing operations that produce consistent time-

shifting for the core service Dynamic Traffic Information

In figure 4 it should be considered only the main operations timings, in order to obtain a simple model of the process. The following is an application for the core service DTI, which only considers time-shifting for the main operations produced in RDSS:

Time necessary for local selections of the user ( . This includes operations like: method for declaring origin and destination of trip and specification of additional criteria;

Time shifting produced by importing dynamic traffic information ( - this is an operation with a high

RDSS

TISP

End User Serv iceWFS:getDynamicRoadTraffic

«device»End User Dev ice

Road Data

Dynamic road traffic information

Dynamic Road Traffic Data Road Traffic Data

Serv ice (dynamic) WFS

End User

Geo-Coding

Interactive selection on map

Specifies a point

Specifies additional criterias Get Dynamic Traffic

Information

Specify type of dynamic Information

Specify type of disturbance

Specify period of v alidity

Get disturbance notifications

Get camera pictures

RDSSTISP

End User Application

Prov ide Dynamic Road Data

Integrate Dynamic Data and Map Data

Other Serv ice Prov ider

Map Data

Specify an address (or PT stop)

Specifies an area

Specifies Origin and Destinationuse

«include»«include»

«precedes»

«include»

«include»

realize

use

use

«include»

provide

provide

provideprovide

«invokes»

«include»

«include»«include»

use

«precedes»

«precedes»

«precedes»

«precedes»


53

degree of incertitude, in terms of duration. It depends on the availability of data and correct operation of third-party information providers;

Time shifting produced by the end-user application operations, also including a certain degree of jitter,

- it includes integration of dynamic and map data on the electronic map [8]. This stage is difficult to compute due to the large number of possible solutions (mixed routes) – the response time will require more time than those for non-intermodal services.

A random component, depending on the hardware configuration and mobile operating system employed by the user mobile equipment, .

Because is not a part of the model (it is a time shifting produced by the user himself in choosing different criteria for his/hers route and transport mode selections, it will be ignored in the estimation of this sub-model. The following estimative sub-model (11) for the layer L3 time-shifting is proposed:

∙ θ θ (11)

where l represents the length of the route (in km), R denotes the number of alternative routes to compute and N the number of transport modes used in multi-modal public transport route alternative.

V. TESTING THE MODELS AND THE REAL SYSTEM

The RTTI system [6] has been implemented in six European cities and several tests have been performed, including performance testing. Since in such a system the focus in on the final user, the performance has to be tested over the full information chain, which is from the handheld to the backend system (with the different processing steps distinguished in this paper). However, because there are third party actors involved in this chain, the tests should only refer to the components that really belong on the own system’s information chain, that is the CAI and RDSS.

A. Test I

The purpose was to analyze the performance of the system on the CAI-RDSS-end user chain. The test has been performed for a number of 15 requests of public transport routes on a distance of 8.6 km in Vienna. The response times were biased as in figure 5.

Fig. 5. Delay of response recorded on a set of 15 tests performed for a 8.6

km route (Vienna, public transport). Horizontal axis: number of test, vertical axis: delay [s]. Maximum jitter: 1.04 s

B. Test 2:

A set of 15 requests for route with public transport, Vienna, route length: 18.3 km. It appears that in this case, the jitter recorded is much more significant: 1.85 s.

Fig. 6. Delay of response recorded on a set of 15 tests performed for a 18.3

km route (Vienna, public transport). Horizontal axis: number of test, vertical axis: delay [s]. Maximum jitter: 1.85 s

C. Simulated:

Fig. 7. Simulation for . Horizontal axis: number of event; vertical axis:

value in seconds

A simulation to determine the variation of has also been performed (figure 7), considering the following values: maximum jitter: 3.69 s; ∈ 0.06 0.17 , ∈ 2 4 , ∈ 2 5 , , ∈ . A comparison has been made with the

tests performed on the most complex service (request of multi-modal route from origin O to destination D), approximately the same length as in Test #2. The results are shown in Table I.

TABLE I. THRESOLDS FOR MULTIMODAL ROUTE REQUEST [5]

Testing period

[s]

Number of

request sent in time

frame

Average maximum threshold for

response time for each request [s]

Notes

1 1 3,8 1 2 4 1 5 5

2 5 4

Average distribution allows for lower maximum peaks thus lower average response time

2 10 4,8


5 20 4,5


5 50 5


0

2

4

6

1 3 5 7 9 11 13 15

Delay [s]

0

2

4

6

0 10 20

Delay [s]

0

2

4

6

0 10 20

Delay [s]


54

VI. CONCLUSIONS

In this paper, an analysis on the effects of multiple stages processing in MRTTI systems has been performed. The multi-modal real-time traffic and travel information systems have been split in different layers and effects in terms of information delaying have been analyzed. Because such systems are very complex and, in fact, they are composed of several sub-systems, modeling the time-shifting of information messages due to signals processing in the different functional components is a difficult task. In some stages, the information is delayed because of the lack of quick response of third-party subsystems, or communication networks. The authors have tried to identify the different effects of each layer, providing a mathematical model for the delay, where possible or appropriate. Considering the results of a research project [5], and the effective participation at the quality tests performed for the system implemented in Europe, some basic conclusions can be drawn:

The MRTTI applications used for emergency vehicles routing must have a closely controlled time response, in order to give an appropriate help in vehicle route guidance and quick response for traffic signals actuation.

It is extremely difficult, even impossible to create a model for the response of third-party systems that are used for collecting traffic/route information – this depends on too many factors: technologies employed, architectures of the systems, interfaces, prioritization of messages etc.; these third-party systems have the most significant influence in the variation (jitter) of response times.

The delays produced in Internet/VPN data communications can be mathematically modeled and have a less influence in delaying of traffic information messages, compared to the previous case. The delays usually range in the domain of seconds, if a correct QoS is ensured for the network [2], [4]. This time-shifting does not usually affect a normal, civil on-board route guidance appliance effect on navigation.

The delays produced by the mobile communication network, if GPRS, give also the possibility to create a mathematical model, and their amplitude is significant only if the network is too congested or if there are losses of GSM signal strength below a certain level. Compared to the MRTTI process speed, these delays are still in the acceptable domain of values.

It resulted from the tests that processing of data in the mobile device is delaying information delivery time because of: poor hardware processing power, low memory or graphics hardware. Also, the operating system has its contribution to the processing time. Regarding position sensing, the mobile device has to be in a good GPS signal coverage in order to collect rapidly geographical

position, otherwise routes calculation from the actual position becomes longer.

Longer routes computation creates longer response times, which also increase in case of public transport mode selection (the device has to multiply its calculations for several routes, employing several transport modes). Variations in response time also increase in this case and range usually in the order of seconds.

Future actions: the authors consider that information systems in traffic will become more and more necessary, taking into account the continuous increase in road traffic density. With the support of growing capacities and quality of service that the communication networks ensure, such systems will become part of the next-generation road vehicles, making driving safer and less stressing. New standards, related to intelligent transportation systems are going to be developed at the European level, also including specifications for MRTTI systems. Research in this area, combined with ad-hoc short range dedicated vehicular communications have a great potential to improve present road traffic safety and environmental protection.

ACKNOWLEDGMENT

The authors would like to thank all the project partners and the personnel that worked in the period of testing (16 persons, 5 days, 8 cars, 6 hours daily), for their fruitful cooperation. They hope that the concept of the system and the services provided will continue to be developed via new research projects in the next future.

REFERENCES

[1] A.S.Khan, B. Afzal, “MPLS VPNs with DiffServ – A QoS Performance

Study”, School of Information Science, Computer and Electrical Engineering, Halmstad University. Master’s Thesis in Computer Network Engineering

[2] * * * Cisco Systems “Quality of Service for Managed Multiservice VPNs: Network Requirements and Out-Tasking Options”, White Paper, 1992-2005 Cisco Systems

[3] A.C. Begen, M.A. Begen, Y. Altunbasak, “Predictive Modelling of Video Packet Delay in VPN Networks” IEEE Conference on Image Processing, 2006

[4] W. Jacobson, “Congestion Avoidance and Control”, ACM SIGCOMM, 1988

[5] * * * CIP ICT PSP 2008-2 Programme, Objective 2.2. Contract no. 238880, “Intelligent and Efficient Travel Management for European Cities – In-Time”, deliverable D.3.1.2, Vienna, Austria, 2011

[6] * * * - CIP ICT PSP 2008-2, Research Project 238880 “Intelligent and Efficient Travel Management for European Cities (In Time)”, 22 partners, DESCA FP 7, ICT PSP Support Programme European Union; Work Packages WP2, 3, 5. URL: www.in-time-project.eu; http://srvwebri.softeco.it/emiXerServer/

[7] M. Minea, I. Badescu, S. Dumitrescu. “Efficiency of Multimodal Real-time Traffic and Travel Information Services Employing Mobile Communications”, 10th International Conference on Telecommunications in Modern Satellite, Cable and Broadcasting Services - TELSIKS 2011, Nis, Republic of Serbia, October 5-8, 2011;

[8] M. Minea, Carmen Eleonora Stan, R.S. Timnea. “Integrated Platform for Road Traffic Safety Data Collection and Information Management”. IARIA ICCGI Conference, IEEE, Valencia 2010


55

Optimizing cloud security by applying new innovative filter

Mehdi Darbandi

Department of Electrical Engineering and Computer Science at Iran University of Science and Technology

[email protected]

Abstract—In this paper, at first authors discuss about principles of Cloud Computing and basic concepts of this new and ground-breaking technology. After this brief introduction, they study more about influences of this technology on different industries and products; they specially focus on Amazon products and their future decisions. After that, in the second section, they present new and innovative filter which can be used as an estimator in cloud platforms. Authors claim that if service providers use such innovative algorithms and equip their gateways and routers with such algorithm they are able to estimate and predict about lots of critical criteria’s. For example, they can estimate the presence of hackers or race them when they’re inside of the network and eliminate them finally or even they can be able to estimate and predict the amount of resources which are need in specific time to prevent from wasting of resources or sudden crashing. Authors of this paper proof their new algorithm by mathematical equations and several simulations at the end of their papers.

Keywords- Cloud platforms security, new innovative Kalman Filter, hacker tracing.

I. INTRODUCTION The importance of virtualized infrastructures and cloud computing is currently increasing rapidly. Virtual infrastructures allow servers, networks, and storage to be virtualized and shared between different users. Cloud computing generalizes and automates this approach such that users of a data center can request virtually any number of machines, networks, and storage while provisioning and scaling is fast and managed transparently by the provider [12]. The increasing complexity and multitenancy of such virtualized infrastructures can cause severe security problems due to possible misconfigurations, e.g. two different users have access to the same storage, and the abstraction of cloud computing hinders the verification of policy compliance. An automated mechanism is required to handle these scenarios and IBM built a prototype for retrieving the configuration of virtual

systems and performing certain security analysis on them[12]. Cloud computing has gained remarkable popularity in the recent years by a wide spectrum of consumers, ranging from small start-ups to governments [12]. However, its benefits in terms of flexibility, scalability, and low upfront investments, are shadowed by security challenges which inhibit its adoption. In particular, these highly flexible but complex cloud computing environments are prone to misconfigurations leading to security incidents, e.g., erroneous exposure of services due to faulty network security configurations. In recent years, Cloud Computing has gained remarkable popularity due to the economical and technical benefits provided by this new way of delivering computing resources, and the pervasive availability of high-speed networks. Businesses can offload their IT infrastructure into the cloud and benefit from the rapid provisioning and scalability. This allows an on-demand growth of IT resources in addition to a pay-as-you-go pricing scheme, which does not require a high up-front capital investment. These benefits are in particular attractive to small businesses, like start-ups, who often have traffic spikes or a steep growth rate, and who prefer to avoid intensive up-front capital investment in their IT infrastructure [12]. However, cloud computing is not limited to such small business. The US government, one of the largest consumers of information technology, is initiating a move of parts of its IT infrastructure into the cloud, in order to reduce costs and gain productivity. These general principles of cloud computing can be implemented on different abstraction levels. While Infrastructure as a Service, such as Amazon EC2, provides virtual machines, storage, and networks, higher abstractions include Platform as a Service as well as Software as a Service that provide the actual web-based applications to end-users [12]. Despite its benefits, Cloud Computing also induces unique challenges in terms of security. Multi-tenancy requires proper isolation of users, the abstraction of the cloud hinders compliance verification of the underlying


56

mailto:[email protected]

architecture, and the sheer complexity of such a system implies a high probability of misconfigurations endangering the overall security. While the benefits of cloud computing are clear and end-users demand such services, security is a major inhibitor of cloud computing adoption on all levels of abstraction. In numerous studies the security related problems have been pointed out. One of the top risks exposed in the study is the failure of isolation in the cloud computing environment [12]. Cloud computing environments are becoming increasingly complex, more tenants are sharing the same physical resources, and the flexibility and possibility of programmatic configurations can lead to unforeseen misconfigurations. For example, network-based storage volumes can be flexibly attached to virtual machines, and potentially a volume will be attached to a wrong virtual machine risking the exposure of sensitive data on that volume. Network security is also flexibly managed through a programmatic interface, which could lead to problems resulting in network services exposed wrongly to the public and opening not properly secured services to other peers [12]. Administrators of such virtual infrastructures must be able to easily understand the complex deployments and ensure that proper security is given. The dynamic and agility of such environments also provides a challenge in ensuring the security over its entire lifetime due to their constant changes [12]. In order to successfully address the problem of configuration complexity and potential misconfigurations in cloud computing environments, we narrowed down the problem domain to a specific case of multi-tier applications deployed in infrastructure clouds using a specific cloud provider as an example case. We will study existing literature in the broad domain of virtual machine security, which plays a fundamental part in the security of infrastructure clouds, and network security analysis with a focus on vulnerability assessment and reachability [12]. Based on the insights and inspirations obtained by performing the literature review, we will propose a novel approach in assessing the security of a multi-tier application deployed on the Amazon infrastructure cloud. By implementing our approach and then evaluating it regarding practicality and scalability, we will determine the practical usefulness for detecting misconfigurations even in large-scale deployments. The evaluation is performed both theoretical and practical. The theoretical evaluation is conducted by assuming complex configuration scenarios and analyzes the algorithm run-time using an

ideal computer. The practical evaluation is performed using the implementation on a sample multi-tier application deployed on Amazon EC2 [12]. The main contribution of this paper is a novel approach in the security evaluation of multi-tier virtual infrastructures, inspired by vulnerability assessment approaches for traditional computing environments and applied for the case of the Amazon infrastructure cloud. The security evaluation consists of an automated security audit process of the currently deployed configuration with regard to a given policy specifying the desired state of the configuration, and an abstract framework for evaluating the security impact of configuration changes. Besides the main contribution stated above, multiple minor contributions can be pointed out [12]. A comprehensive description of the underlying architecture of the Amazon infrastructure cloud is presented, which was publicly only available in incomplete and fragmented form. We provide a comparison of two methods for deploying multi-tier virtual infrastructures on Amazon with regard to the provided isolation levels [12]. Finally, a data model for representing the configuration of Amazon deployments is presented and integrated into a larger data model capable of representing configurations of different virtualization systems [12]. Cloud computing is a broad term combining several different types of service offerings. In general we distinguish between Software, Platform, and Infrastructure as a service, which are offered by the cloud provider [12]. The main focus of this paper lies on Infrastructure as a Service, also called Infrastructure Clouds, but for comparison reasons the other types of offerings are also briefly presented [12]. Storage can be provided in different ways varying among the multiple IaaS providers. Four different forms can be identified in the currently available providers: NAS-like, SAN-like, API-based data objects, and Virtual Machine storage. A virtual machine has typically a fixed-size data storage available, which is equivalent to a hard disk in a regular desktop or server computer. In some cases this type of storage is only intended to be used for temporary data and is itself non-persistent, i.e., after the machine terminates the data is lost [12]. NAS-like storage, like GoGrid Cloud Storage, is accessible from the VMs on a file-based level using standard protocols like CIFS. Amazon Elastic Block Store (EBS) is a SAN-like storage type, which appears to the VM as an additional block-device. An EBS volume can be attached to different VMs, but not to multiple VMs simultaneously, and the size can be adjusted presuming the file system on the block-device is resizable as well.


57

The last type of storage is accessible through an API and holds data objects up to a specific size, e.g., in the range of several gigabytes. This is a very scalable kind of storage, i.e., one can store an arbitrary amount of objects, and also provides the possibility of distributing these objects using a Content Distribution Network offered by the provider. Examples of this kind of storage are Amazon Simple Storage Service (S3) and RackSpace CloudFiles [12]

II. PERFORMANCE COMPARISON OF TWO STAGE KALMAN FILTERING TECHNIQUE FOR SURVEILLANCE PERMEATING TRACKING IN

CLOUD COMPUTING [17]:

1. Statement of the Problem: The problem of interest is described by the discretized equation set [13]:

xkkkkkk WUBXAX ++=+1 (1)

ukkkk WUCU +=+1 (2)

kkkk VXHZ += (3)

Where n

k RX ∈ is the system state, m

k RU ∈ and p

k RZ ∈ are the input and the measurement vectors, respectively. Matrices kA , kB , kC and kH are assumed to be known functions of the time interval k and are of appropriate dimensions. Matrix kC is assumed

nonsingular. The process noises x

kW , u

kW and the measurement noise kV are zero-mean white Gaussian sequences with the following covariance’s:

klxk

xl

xk QWWE δ=])([ ' , kl

xuk

ul

xk QWWE δ=])([ ' ,

kluk

ul

uk QWWE δ=])([ '

klklk RVVE δ=][ '

, 0][ ' =lx

k VWE and 0][ ' =l

uk VWE , where

' denotes transpose and klδ

denotes the Kronecker delta function. The initial states 0X and 0U are assumed to be uncorrelated with the

sequencesx

kW , u

kW and kV . The initial conditions are assumed to be Gaussian random variables with 00

ˆ][ XXE = , xPXXE 0

'00 ][ = , 00

ˆ][ UUE = , uPUUE 0

'00 ][ = ,

xuPUXE 0'00 ][ = .

Treating kX and kU as the augmented system state, the AUSKE is described by [13]:

)( |1111|11|1Aug

kkAugkk

Augk

Augkk

Augkk XHZKXX +++++++ −+= (4)

Augkk

Augk

Augkk XAX ||1 =+ (5)

1'1|11

'1|11 ])([)( −

++++++ += kAugkkk

Augk

Augkkk

Augk RHPHHPK (6)

kAug

kkkAug

kkk QAPAP +=+'

||1 )( (7)

kkAugk

Augkkk PHKIP |1111|1 )( +++++ −= (8)

Where

=

k

kAugk U

XX

,

=

uk

xkAug

k KK

K,

=

uk

xuk

xuk

xk

k PPPP

P')( ,

=

× knm

kkAugk C

BAA

0 ,

'

0

=

×mp

kAugk

HH

,

=

uk

xuk

xuk

xk

k QQQQ

Q')(

Where the superscript ‘Aug’ denotes the augmented system state, I denotes the identity matrix of any dimension and nm×0 is a nm× zero matrix. It is clear from (4)-(8) that the computational cost of the AUSKE increases with the augmented state dimension. The OPSKE formulation is based on the following equations [13]:

)ˆ(ˆˆ|1111|11|1 kkkkkkkkk XHZKXX +++++++ −+= (9)

kkkkk XAX ||1ˆˆ =+ (10)

1'1|11

'1|11 ])([ −

++++++ += kkx

kkkkx

kkk RHPHHPK (11) xkk

xkkk

xkk QAPAP +=+

'||1 )( (12)

xkkkk

xkk PHKIP |1111|1 )( +++++ −= (13)

1111 ++++ −= kkkk M]HKI[N (14) ]ˆ~

[ˆˆ|11111|11|1 kkkkk

ukkkkk UMHZKUU ++++++++ −+= (15)

kkkkk UCU ˆˆ1 =+ (16)

1|1

'1

'1|111

'1

'1|11 ]3[2 −

++++++++++ +×= zkkkk

ukkkkkk

ukk

uk PHMPMHHMPK (17)

ukkkk

uk

ukkk

ukk

uk

zkk

uk

ukkk

ukkkk

uk

ukk

ukk

PMHKKHMPKPK

KHMPMHKPP

|11111'

1'

1|11|11

1'

1'

1|1111|11|1

2)(2)(

)(3

+++++++++++

++++++++++

−′−′+

′+=

(18)

ukk

ukkk

ukk QCPCP +=+

'||1 (19)

1'

1|11|1 +++++ += kkx

kkkz

kk RHPHP (20) u

kkkkzu

kk PMHP |111|1 ++++ = (21)

11|1|1ˆˆ

++++ += kkkkkk UMXX 111|11|1

ˆˆ++++++ += kkkkkk UNXX

(22)

1001

11 ,....3,2,][

−

−+

=

=+=

CBMkCBMAM kkkkk

(23)

1111 ][ ++++ −= kkkk MHKIN (24)

2. Performance Evaluations [13]: To demonstrate the computational advantage of the OPSKE over the AUSKE, the number of arithmetic operations are considered, i.e., multiplications and summations. The arithmetic operations of a standard Kalman estimator with state dimension n and measurement dimension p , are listed in Table 1. It is clear from the equations (4)-(8) and Table 1, that the arithmetic operations required for the AUSKE which has state dimension mn + and measurement dimension


58

p , are ),( pmnM + for multiplications and ),( pmnS + for summations. Table 2 shows the arithmetic operations of the input estimation and the auxiliary matrices needed by the OPSKE which has state dimension n , measurement dimension p and input vector dimension m . Note that the number of the arithmetic operations of the AUSKE increases with the augmented state dimension, which makes the algorithm computationally inefficient. In contrast, the OPSKE based on the two-stage decoupling technique required fewer computations. The efficiency of the OPSKE is due to order reduction, i.e., implementing two less order n and m partitioned filters. This enables the proposed algorithm to have much better computational efficiency than the AUSKE. So, the arithmetic operations required (AOR) for the AUSKE are [13]:

)]()()(2)(2)(3[])(2)()(2)(2)(3[

),(),()(

23223

23223

mnmnppmnpmnmnpmnmnppmnpmnmn

pmnSpmnMAUSKEAOR

+−+−+++++++

++++++++++=

+++=

(25)

The arithmetic operations required for the input estimation and auxiliary matrices, by the OPSKE as shown in Table 2 and using equations (15)-(24) are

]222422[]224

2223[]223[

]2223[),,(),,(),(),(

)(

2222

33222

2223

23222

23223

23223

nmpnmmnnpnppnmpmppmmmmp

nmpnmmnnmpnmppmppmmmp

nnpnppnnnpnpnppnn

pmnSpmnMpnSpnMOPSKEAOR

OPOP

+++−+−+

++++−−−+

++++++

++++++

−−++++

+++++=

+++=

(26)

Using (25) and (26), the operational savings, denoted

by OPSKEAUSKEOS , of the OPSKE as compared to the

AUSKE are [13]:

nmpmnnppnmppnnmmnmpmnSpmnMpnS

pnMpmnSpmnMOPSKEAORAUSKEAOROS

OPOP

OPSKEAUSKE

22226417152

),,(),,(),(),(),(),(

)()(

223

2223

−−−++−

+−++−=

−−−

−+++=−=

(27)

And the operational savings of the OTSKE over the AUSKE are:

nmpmmnmpnmmnmOTSKEAORAUSKEAOROS OTSKE

AUSKE

22412124)()(

3222

3

−−−+++

+−=−=

(28)

Therefore, using (27) and (28) the operational savings of the OPSKE over the OTSKE are [13]:

22322

23

2224532)()(

pmmnnppnmppnnmmnmOPSKEAOROTSKEAOROS OPSKE

OTSKE

−+−++−+−+

+=−=

(29)

It is clear from (27) and (29) that for npm ≤and , the proposed scheme has computational advantage over the AUSKE and it is comparable to the OTSKE. The operational savings discussed here will be tested as an example in the simulation results section. To measure the relative operational savings of the OPSKE with respect to the arithmetic operation required by the AUSKE ( )(AUSKEAOR ), the percentage of the operational savings defined as below:

100)(×=

AUSKEAOROS

POSOPSKEAUSKEOPSKE

AUSKE

(30)

Using (27), (29) and (30), the operational savings and the percentage of the operational savings, of the OPSKE comparing to the OTSKE and the AUSKE for different values of n , m and p are shown in Table 3. It can be inferred from Table 3 that the OPSKE has better overall performance than the AUSKE (averaged 32%) and the OTSKE (averaged 7.3%) [13].

Table 1:Standard Kalman Estimator Arithmetic Operation Requirements [13] Variable Number of Multiplications, )p,n(M Number of summations, )p,n(S 1 11 ++ k|kX np2 np2 2 k|KX 1+

2n nn −2 3 x

kK 1+ 322 2 pnppn ++ nppnppn 22 322 −++

4 xk|KP 1+

32n 232 nn −

5 xk|KP 11 ++ pnn 23 +

223 npnn −+ Totals npnpnppnn 2223 23223 +++++ nnpnppnn −−+++ 23223 223

Table 2:Input Estimation and Auxiliary Matrices Arithmetic Operation Requirements for the OPSKE [13]

Variable Number of Multiplications )p,m,n(M OP

Number of summations )p,m,n(S OP

1 11 ++ k|kU mp2 mp2 2 k|KU 1+

2m mm −2


59

3 ukK 1+ mpppmppm ++++ 2322 2 mppmppm 22 322 −++

4 uk|KP 1+

32m 232 mm − 5 u

k|KP 11 ++ 223 mpmm ++

223 mpmm −+ 6 z

k|kP 1+ pn22 22 22 pnppn +−

7 k|kX 1+ mn mn

8 11 ++ k|kX mn nmn − 9 1+kM

232 nmmmn ++ nmnmmmn −++ 232

10 1+kN mn2 nmmn −2 11 11 ++ kk MH

nmp mpnmp −

Totals nmpnmmnnmpnm

ppmppmmmp++++++

+++++2223

23222

2242223

nmpnmmnnpnppnmpmppmmmmp

+++−+−+

++++−−−2222

33222

222422

Table 3:the Operational Savings and the Percentage of the Operational Savings of the OPSKE Compared to the AUSKE and the OTSKE [13]

The state vector dimensions

OPSKEAUSKEOS

OPSKEAUSKEPOS (%)

OPSKEOTSKEOS

OPSKEOTSKEPOS (%)

244 === p,m,n 1340 35.7 592 15.7 224 === p,m,n 578 33.7 102 5.9 124 === p,m,n 553 37.5 155 10.5 114 === p,m,n 242 27.5 23 2.6 334 === p,m,n 978 32.7 247 8.2 2210 === p,m,n 2954 25.1 132 1.12

Average ≅ 1107 32.0 ≅ 208 7.3

3. Simulation Results: To evaluate the proposed algorithm, an example of maneuvering target tracking problem which turns, in two-dimensional space is simulated such as permeating a hacker into a very important network or databases. In this simulation example, the performance of the OPSKE for the maneuvering target tracking has been compared with the traditional works that done in this concept, as an example of the AUSKE method. As mentioned before in the augmented state method the state vector includes the input vector i.e., acceleration and jerk parameter in maneuvering target tracking problem. The sampling interval is T=0.01 (sec) and target maneuver is applied at 9th second (900th sample). The initial conditions are selected similar for the AUSKE as well as the OPSKE. The state vectors are

[ ]' ykk

xkkk vyvxX = , [ ]' y

kyk

xk

xkk jujuU = ,

[ ]' yk

yk

xk

xk

ykk

xkk

Augk jujuvyvxX =

Where kx , xkv ,

xku and

xkj denote the position,

velocity, acceleration and jerk of the target along the x axis, respectively. We consider the target initial

conditions for the state and the acceleration vectors as below [13]:

[ ]' s/m m s/m m X 2512508021650 −= , [ ]' sec/g g sec/g g U 00000 = [ ] ' sec/ 0 0sec/ 0 0/ 25 1250/ 80 21650 ggggsmmsmmX Aug −=

The target begins to maneuver as [ ] ' sec/ 4.00sec/ 7.00900 ggggU −= for

sec)( 90(sec) 9 ≤≤ t . The system matrices are given by

=

1000100

0010001

T

T

Ak

,

=

2/006/2/00

002/006/2/

2

32

2

32

TTTT

TTTT

Bk

,

=

1000100

0010001

T

T

Ck

,

'

00100001

=kH


60

=

TTTT

TTTT

Q juk

2/002/3/00

002/002/3/

2

2

23

2

23

ασ

,

=

20/72/0072/252/00

0020/72/0072/252/

2

56

67

56

67

TTTT

TTTT

Q jxk ασ

=

6/8/0024/30/00

006/8/0024/30/

2

34

45

34

45

TTTT

TTTT

Q jxuk ασ

, 440 10 ×= IP x

,

441.0 ×= IPuo , 440 ×= IPxu

,

'

420

=

×

kAugk

HH

=

× k

kkAugk C

BAA

440 ,

=

uk

xuk

xuk

xk

k QQQQ

Q')( ,

=

uk

xuk

xuk

xk

k PPPP

P')( .

Where )(09.0 3−= msjσ the variance of the target is jerk and )(s 0123.0 -1=α is the reciprocal of the jerk time constant ατ /1= . The measurement standard deviations of x and y target positions are:

)( 1010 mx =σ , )( 20 my =σ . Thus, the measurement

covariance matrix is

=

400001000

kR for both methods

[13]. The Root Mean Square Error (RMSE) index is used for the results evaluation. Fig. 1 shows the actual value and the estimation of x and y and RMS errors of x and y positions estimations by the proposed OPSKE and the AUSKE. Fig. 2 shows the actual value and the estimations of

yx vv , and the RMS errors of the x and y velocities estimations by the proposed method compared with the augmented method. The actual value and the accelerations estimations in the x and y directions and their corresponding averaged RMS errors can be seen in Fig. 3.Fig. 4 displays the actual value and the estimated jerk parameters are evaluated by the OPSKE and the AUSKE methodologies [13].

10 15 20 25

-3

-2

-1

0

x 104

Time (sec)

x (m

)

Atcual positionOPSKE method estimationAUSKE method estimation

10 15 20 250

100

200

300

400

500

Time (sec)

Aver

aged

RM

SE o

f x (m

)

OPSKEAUSKE

10 15 20 25

0

0.5

1

1.5

2

x 104

Time (sec)

y (m

)

Atcual positionOPSKE method estimationAUSKE method estimation

10 15 20 25

0

50

100

150

200

250

300

Time (sec)

Aver

aged

RM

SE o

f y (m

)

OPSKEAUSKE

Fig. 1. The actual value and the estimation of the x, y positions and RMS errors estimations by the OPSKE and the AUSKE methods.


61

10 15 20 25

-5000

-4000

-3000

-2000

-1000

0

Time (sec)

v x(m/se

c)

Atcual velocityOPSKE method estimationAUSKE method estimation

10 15 20 250

100

200

300

Time (sec)

Aver

aged

RM

SE o

f v x (m

/sec)

OPSKEAUSKE

10 15 20 25

0

1000

2000

3000

Time (sec)

v y (m/se

c)

Atcual velocityOPSKE method estimationAUSKE method estimation

10 15 20 25

0

50

100

150

200

250

Time (sec)

Aver

aged

RM

SE o

f v y (m

/sec)

OPSKEAUSKE

Fig. 2. The actual value and the estimation of yx v ,v and RMS errors of x and y velocities estimations by the

OPSKE and the AUSKE methods.

10 15 20 25

-15

-10

-5

0

Time (sec)

u x (m/se

c2 )

Atcual accelerationOPSKE method estimationAUSKE method estimation

10 15 20 25

0

0.5

1

1.5

2

Time (sec)

Aver

aged

RM

SE o

f ux (m

/sec2 )

OPSKEAUSKE

10 15 20 25

0

2

4

6

Time (sec)

u y (m/se

c2 )

Atcual accelerationOPSKE method estimationAUSKE method estimation

10 15 20 25

0.2

0.4

0.6

0.8

1

1.2

Time (sec)

Aver

aged

RM

SE o

f uy (m

/sec2 )

OPSKEAUSKE

Fig. 3. The actual value and the estimation of acceleration in x and y directions and corresponding RMS errors by the proposed method compared with the augmented methods.


62

10 15 20 25

-0.6

-0.4

-0.2

0

time (sec)

j x (m/se

c3 )

Atcual jerkOPSKE method estimationAUSKE method estimation

10 15 20 250

0.1

0.2

0.3

0.4

0.5

0.6

time (sec)

Aver

aged

RM

SE o

f j x (m/se

c3 )

OPSKEAUSKE

10 15 20 25

0.2

0.4

0.6

0.8

time (sec)

j y (m/se

c3 )

Atcual jerkOPSKE method estimationAUSKE method estimation

10 15 20 250

0.1

0.2

0.3

0.4

0.5

0.6

time (sec)

Aver

aged

RM

SE o

f j y (m/se

c3 )

OPSKEAUSKE

Fig. 4. The actual value and the estimation of jerk parameters and RMS errors by the OPSKE method compared with the AUSKE method.


63

It is clear that the performance of the proposed OPSKE is as well as the results obtained by the AUSKE in the maneuvering target tracking problem. Note that in this example 4=n , 4=m and 2=p , and the operation savings for the OPSKE over the AUSKE and the OTSKE as shown in Table 3 are 1340 (or 35.7%) and 592 (or 15.7%), respectively.

III. CONCLUSION In this paper, first of all, authors discuss about different aspects of cloud computing and impacts of this technology on different industries and societies, they study about impacts and influences of this technology with focus to the Amazon products. After understanding this technology and attain more about applications of this technology, they reveal their new and novel algorithm, which is named as two-stage Kalman filtering. Authors claim that by using such algorithm we can estimate and predict about all important factors that are dealing with using of such networks.

References [1] Mehdi Darbandi “Applying Kalman Filtering in

solving SSM estimation problem by the means of EM algorithm with considering a practical example”; published by the Journal of Computing – Springer, 2012; USA.

[2] Mehdi Darbandi; “Comparison between miscellaneous platforms that present for cloud computing and accreting the security of these platforms by new filter”; published by the Journal of Computing – Springer, 2012; USA.

[3] Mehdi Darbandi; “New and novel technique in designing electromagnetic filter for eliminating EMI radiations and optimization performances”; published by the Journal of Computing - Springer, 2012; USA.

[4] Mehdi Darbandi; “Appraising the role of cloud computing in daily life and presenting new solutions for stabilization of this technology”; published by the Journal of Computing - Springer, 2012; USA.

[5] Mehdi Darbandi; “Cloud Computing make a revolution in economy and Information Technology”; published by the Journal of Computing - Springer, 2012; USA.

[6] Mehdi Darbandi; “Considering the high impact of gettering of silicon on fabrication of wafer designing and optimize the designing with new innovative solutions”; published by the Journal of Computing – Springer, 2012; USA.

[7] Mehdi Darbandi; “Developing concept of electromagnetic filter design by considering new parameters and use of mathematical analysis”; published by the Journal of Computing - Springer, 2012; USA.

[8] Mehdi Darbandi; “Is the cloud computing real or hype Affirmation momentous traits of this technology by proffering maiden scenarios”; published by the Journal of Computing – Springer, 2012; USA.

[9] Mehdi Darbandi; “Measurement and collation overriding traits of computer networks and ascertainment consequential exclusivities of cloud computing by the means of Bucy filtering”; published by the Journal of Computing - Springer, 2012; USA.

[10] Mehdi Darbandi; “Unabridged collation about multifarious computing methods and outreaching cloud computing based on innovative procedure”; published by the Journal of Computing - Springer, 2012; USA.

[11] Mehdi Darbandi; “Scrutiny about all security standards in cloud computing and present new novel standard for security of such networks”; published by the Journal of Computing - Springer, 2012; USA.

[12] MSc. Thesis of Sören Bleikertz; Norwegian University of Science and Technology; June 2010.

[13] A. Karsaz, H. Khaloozade, M. Darbandi; “Performance Comparison of the two-stage Kalman filtering Techniques for Target Tracking” Int. IEEE Conf. Harbin, China.


64

Studying security criteria’s of cloud and VM platforms and present new innovative solution

for it

Mehdi Darbandi

Department of Electrical Engineering and Computer Science at Iran University of Science and Technology

[email protected]

Abstract: This paper categorized as such: in the first section of this paper, authors present general information about basis of cloud computing. By the means of several examples they show the importance of security in such networks. They discuss about security problems in VMware and combine it with security of cloud computing. After that they present their new novel approach for intensifying security of such network, they purpose Sequence Unscented Kalman Filtering Algorithm for acquiring better security. At the end of this paper, authors proof workability and results of applying such algorithm to the cloud computing by the means of mathematical demonstration and MATLAB simulations.

Keywords: Cloud Computing, Internet technology, Kalman filter.

Introduction: Infrastructure clouds make significant use of virtualization and the clouds provide computational resources which are consumed by the means of virtual machines [10]. Due to this strong connection between these two technologies, security problems associated with virtual machines will have an impact on the overall security of infrastructure clouds. Therefore a review of existing literature on the topic of virtual machine security will

give us a useful foundation for analyzing the security of infrastructure clouds and provides an insight in the underlying security challenges [10]. Virtual machines provide a high degree of flexibility by allowing users to easily create, copy, snapshot, rollback, and migrate them. This flexibility results in major adoption of virtual machines by users for different purposes, e.g., for testing of software or configurations using snapshots and the rollback mechanism [10]. The authors of the paper extracted the following list of security issues related to virtual machines. Scaling represents the problem that users now have multiple virtual machines, e.g., for testing and development purposes, instead of a very few number of physical ones [10]. Therefore the total number of machines drastically increases within one organization and the workload on the security systems will increases accordingly. Diversity in operating systems, OS versions and patch levels increases the complexity in the security management of the infrastructure [10]. VMs typically result in a high diversity, because users have multiple snapshots of VMs and testers can have a collection of different VMs. Transience is another security issue induced by the flexibility of virtual machines [10]. It


65

mailto:[email protected]

mainly deals with the problem that VMs appear and disappear very rapidly in the network which makes security management, e.g., patch management and vulnerability scanning, very difficult [10]. The authors describe this as the missing of a steady state in the network, where the steady state means that all machines are patched and properly managed. The Mobility of a VM, i.e., the VM can easily be copied or migrated, imposes multiple security problems: all the hosts, the VM will be executed on, have to be part of the trusted computing base (TCB) [10]; sensitive information can leave a security perimeter or malware is introduced, and the theft of VMs can easily be done by simply copying a file. The traditional Software Lifecycle, i.e., a monotonic forward progress of the software state, is broken by virtual machine’s snapshot and rollback mechanisms, because the execution of the virtual machine can be forked and be rolled back [10]. In particular the rollback mechanism induces a lot of problems regarding freshness of randomness sources used for cryptographic protocols or critical patches are removed by a rollback. Limited Data Lifetime, e.g., for sensitive or cryptographic information, can be compromised due to the rollback mechanism and that the content of the virtual machine’s memory might be stored on the disk of the host due to paging, snapshots, or migration [10]. In traditional computing environments, the Identity of a machine is often deduced from properties like the MAC address, the location, or Ethernet port. Virtual machines however typically use dynamically created MAC addresses and they might migrate from one physical host to another, therefore properties like the location or Ethernet port will change, and make it difficult to assign an identity [10]. Now, author of this paper present some solutions and highlights some benefits about

these technology [10]. The role of the VMM is to isolate the VMs from each other and the correctness of enforcing this property is crucial, therefore a high assurance VMM is required [10]. Introducing an extended virtualization layer that overtakes functionality originally performed in the guest operating systems has a certain number of benefits. Users do not have to worry about security management, e.g., firewalling or anti-virus detection, if these mechanisms are provided by the virtualization layer and are operated by a central administration staff. Furthermore, these security services are now independent of the guest operating systems, which results in a higher flexibility because a high diversity of VMs can be securely managed. Regarding the security issue associated with software lifecycle and the rollback feature, the virtualization layer could provide mechanisms to store such sensitive information and to provide strong randomness [10]. The security of the Virtual Machine Monitor (VMM) is crucial, because it provides the necessary isolation between the hosted VMs and typically runs with the highest privileges on the system [10]. Introducing a new software layer, such as the one providing virtualization, inherently increases the complexity of the system, which also increases the possibility of software security vulnerabilities [10]. Such vulnerabilities in the VMM can lead to the break of isolation, i.e., a VM can access other VMs resources. Different solutions exist to mitigate security problems in the virtualization layer which are based on principles of building secure software [10]: formal verification, security by isolation and disaggregation, and reducing the trusted code size [10]. An interesting example for formal verification of software, which is also relevant for our topic


66

of VMM security, is the seL4 project, a formally verified L4 microkernel. The proof verifies that the implementation in the C programming languages matches the abstract specification of the system and implies that certain software vulnerabilities, like buffer overflows and null pointer dereferences, are absent in the implementation. Microkernel and VMM are very similar, therefore either a formally verified microkernel acting as a VMM can be used or adapting the formal proof for VMMs, although the size of existing VMMs make formal verification very difficult. Another approach of improving the security of the VMM is to reduce the complexity and trusted code base (TCB) by means of decomposition. An approach for extracting the domain builder functionality of the Xen dom0 into a separate domain. With a separate domain builder VM, the user-space of dom0 can be removed from the TCB, because no privileged functionality for VM construction and management need to be exposed to user-space applications, e.g., xend. However, in their current state the dom0 kernel is still part of the TCB due to required interaction with physical I/O devices. Besides the dom0 kernel, the Xen hypervisor and the domain builder are part of the TCB [10]. The recent prototype operating system Qubes OS implements, among other security features, disaggregation of Xen dom0 by establishing driver domains which are limited to a specific hardware resource by the means of IOMMU as implemented by Intel VT-d, i.e., the VMM monitors DMA requests and can possibly restrict them [10]. These driver domains can run with limited privileges and the overall complexity of dom0 can be reduced. Thereby software vulnerability in one of the drivers will not result in a break of isolation when running in a non-privilege driver domain compared to

running in dom0. IOMMU would also benefit the disaggregation of Xen using a domain builder VM, because the dom0 kernel could be removed from the TCB when I/O with physical devices is offloaded to driver domains. The virtualization architecture NOVA uses a minimal microkernel, with a size similar to the formally verified seL4, and provides virtualization functionality as user-land applications. Therefore the amount of high privilege code is reduced to a minimal microkernel-based hypervisor [10].

Sequence Unscented Kalman Filtering Algorithm: Since Unscented Kalman Filter (UKF) is proposed by Julier and Uhlman, it has absolved many researchers to study it, and many kinds of new algorithms [2-6] with different accuracies have come out. Unlike the Extend Kalman Filter (EKF), which is based on the linearizing the nonlinear function by using Taylor series expansions, UKF uses the true nonlinear models and approximates a distribution of the state random variable [11]. Furthermore, it only needs a minimal set carefully chosen sample points, by which the posterior mean and covariance can be accurate to the second order for any nonlinearity, avoiding Jacobian’s computation. If the priori random variable is Gaussian, the estimation of the posterior mean and covariance can could be accurate to the third order. It can be seen that in all the UKF algorithms it needs inversing the matrix in measurement update. The dimension of the inversed matrix is equal to that of measurement vector. If the dimension of the measurement is very large, so it could cost a great computing time. In order to decrease the computing cost and not to inverse the matrix, a sequence method is used to solve this problem [11]. In this paper, the sequence UKF is proposed. It deals with nonlinear stochastic system with


67

linear measurement. Based on RBUKF and traditional Kalman Filter (KF), it deduced the special algorithm for the sequence UKF in case of the covariance matrix of measurement noise is diagonal matrix or not. In theory it is proven that the sequence UKF has the same estimation accuracy with RBUKF, but has lower computational cost. Simulation results verify the high performance of sequence UKF.

UKF MECHANISM: UKF is used to solve the estimation problem for any nonlinear system. The considered nonlinear system is represented by [11]:

1 ( )( )

k k k

k k k

+ = + = +

x f x v

z h x w .

Where kx and kz denote the state vector with n -dimension and the measurement vector with m -dimension at step k , respectively. The

nonlinear mapping f( )⋅ and h( )⋅ are assumed to be continuously differentiable with respect

to kx .Moreover, N( ; , )k k kv v 0 Q denote the process noise with n -dimension.

N( ; , )k k kw w 0 R denote the measurement

noise with m -dimension. kv and kw are independent of each other. Like Kalman Filter (KF), UKF is also a minimum mean-square error estimator (MMSE). For system (1), the mechanism of MMSE is time-update and measurement-update as follows [11]: Time-update:

/ 1 1[f( )]k k kE− −=x x / 1 [ ]T

k k k kE− =P e e Measurement-update:

/ 1ˆ k k k k n−= +x x W υ ˆk k k= −υ z z

ˆ [h( )]k kE=z x

/ 1T

k k k k k k−= −P P W S W Where / 1k k k k−= −e x x , the weight matrix kW is chosen to minimize the trace of the updated

covariance kP . Its value is calculated from [11]: 1xz

k k k−=W P S

Wherexz

kP is the cross covariance

between ke and kυ , kS is the covariance of kυ . UKF is based on the mechanism above. By applying the unscented transformation (UT) to a number of chosen sigma

points, / 1k k−x , / 1k k−P , kS andxz

kP can be approximately expressed by the linear composition of the transformed sigma points. So UKF solves the nonlinear estimation problem using MMSE mechanism. When measurement equation in system is linear, it changes to system as follows [11]:

1 f( )k k k

k k k kx+ = +

= +

x x v

z H w To solve the estimation problem for system, it only needs transformed sigma points to

estimate / 1k k−x and / 1k k−P , and kS andxz

kP can be computed accurately. So the UKF algorithm can be reduced to RBUKF algorithm. Compared to UKF, RBUKF is not less computational cost, but also higher accuracies. The difference part between the UKF and RBUKF is measurement-update. For comparison, it only gives the measurement-update of RBUKF as follows [11]:

, | 1T

zz k k k k k k−= +P H P H R

, | 1T

k xz k k k−=P P H 1

, | 1 ,k xz k k zz k−

−=K P P | 1 | 1ˆ ( )k k k k k k k− −= + −x x K z z

| 1 ,T

k k k k zz k k−= −P P K P K

SEQUENCE UKF ALGORITHM


68

No matter UKF or RBUKF, it needs to inverse the matrix in measurement-update. If the dimension of the measurement vector is very large, it could be a great computational cost for

compute gain matrix kK . In order to avoid

inversing the matrix in computing kK , the sequence UKF will deal with ever component of measurement vector one by one instead of the vector at one time. This method needs not to inverse the matrix and can greatly decrease computational cost. For system, it deduced the sequence UKF as follows [11].

Theory I: For system, the measurement-update in UKF can be computed as follows:

1 1 1( )i i iT i i iT ik k k k k k kR− − −= +K P H H P H

1 1( )i i i i i ik k k k k kz− −= + −x x K H x

1 1i i i i ik k k k k

− −= −P P K H P 1, 2, ,i m= ⋅⋅⋅ Where

ikH is the i -th row in kH ,

ikz is the i -th

scalar in the measurement vector at step k , ikR is

the i -th element in diagonal of kR [11]. Proof: Rewrite the measurement equation in system, it gets [11]:

1 1 1

1 2 2k k k

k k kk

m m mk k k

z vz v

z v

= +

H

Hx

H

Because of the form of, the measurement

vector kz can be seen as ikz ( 1,2, ,i m= ⋅⋅⋅ ) one by

one to update the time-update equations. It

must be noticed that when ikz one by one

instead of kz updates the equations in measurement-update, it implies that the state

equation is invariant for everyikz at step k . So

the measurement-update in UKF at step k is equivalent to the filtering problem of new linear systems as follows [11]:

1i ik ki i i ik k k kz w

− =

= +

x x

H x ( 1, 2, , )i m= Where N( ;0, )i i i

k k kw w R , is the equivalent measurement noise. The initial value of the

filter is 0

/ 1k k k−=x x and0

/ 1k k k−=P P . So the equivalent linear filter can be easily derived from classic Kalman Filter equations as follows:

/ 1 1i i ik k

− −=x x / 1 / 1( )i i i i i i i i

k k k k k kz− −= + −x x K H x / 1 / 1 1( )i i i iT i i i iT i

k k k k k k kR− − −= +K P H H P H / 1 1i i i T

k k− −=P IP I

/ 1( )i i i i ik k k k

−= −P I K H P Or

1 / 1 1 1( ) ( ) ( )i i i iT i ik k k k kR− − − −= +P P H H

Substituting these recent equations results in the following, and substituting previous equations in each other, results in next equations. So it obtains the new equations for the measurement-update in UKF as follows:

1 1( )i i i i i ik k k k k kz− −= + −x x K H x

1 1 1( )i i iT i i iT ik k k k k k kR− − −= +K P H H P H

1( )i i i ik k k k

−= −P I K H P Or

1 1 1 1( ) ( ) ( )i i iT i ik k k k kR− − − −= +P P H H

By theory I and the UKF mechanism, it obtains the sequence UKF algorithm as follows: Calculate sigma points [11]:

, 1 , 1

, 1 , 1 1

, 1 , 1 1

0,

( ( ) ) 1, , ,

( ( ) ) 1, , 2 ,

i k i k

i k i k k i

i k i k k i L

i

L i L

L i L L

λ

λ

− −

− − −

− − − −

= = = + + =

= − + = +

χ x

χ x P

χ x P

Time-update:

/ 1 1f( )k k k− −=χ χ 2

( )/ 1 , / 1

0

nm

k k i i k ki

W− −=

= ∑x χ

2( )

/ 1 , / 1 | 1 , / 1 / 10

( )( )n

c Tk k i i k k k k i k k k k k

iW− − − − −

=

= − − +∑P χ x χ x Q

/ 1 / 1 ( 1, 2, , )i i ik k k k kz i m− −= =H x

Measurement-update:


69

0/ 1k k k−=P P

0/ 1k k k−=x x 1 1( )i i i i i i

k k k k k kz− −= + −x x K H x 1 1 1( )i i iT i i iT i

k k k k k k kR− − −= +K P H H P H 1( )i i i i

k k k k−= −P I K H P ( 1, 2, , )i m=

mk k=x x

m

k k=P P Where

( ) 20 ( ) (1 )cW Lλ λ α β= + + − + ,

( ) ( ) 1 [2( )]m ci iW W L λ= = + , 1, 2, , 2i L= .

ALGORITHM PERFORMANCE ANALYSIS: Filtering accuracy Analysis: For system, in order to comparison of RBUKF and the sequence UKF, it needs to get the filtering covariance respectively. Firstly, it is to

derived covariance kP of RBUKF. By

definition, it calculates covariance kP of RBUKF as follows:

ˆ ˆ[( )( ) ]Tk k k k kE= − −P x x x x

Substitute this equation in previous equations results in [11]:

/ 1 / 1 / 1

/ 1

T Tk k k k k k k k k k k

T Tk k k k k k

− − −

−

= − −

+

P P P H K K H P

K H P H K By previous equations, it obtains

1, / 1 , / 1

1/ 1 / 1( )

k xz k k zz k k

Tk k k k k k k

−− −

−− −

=

= +

K P P

P H P H R Substituting these recent equations results in:

1/ 1 / 1 / 1 / 1

1 1 1/ 1

( )

( )

T Tk k k k k k k k k k k k k k

Tk k k k k

−− − − −

− − −−

= − +

= +

P P P H H P H R H P

P H R H By inversing both sides of this equation, it obtains [11]:

1 1 11/

Tk k k k k k− − −

−= +P P H R H Because of:

1 2[ ]T T mT Tk k k k=H H H H 1 1 2 1

1 1 2 1 1

( ( ))

(( ) ( ) ( ) )

mk k k k

mk k k

diag R R Rdiag R R R

− −

− − −

=

=

R

Substituting these recent equations in each other, and doing matrices multiplication, it gets

1 1 11/

1 11/

1( )

Tk k k k k k

miT i i

k k k k ki

R

− − −−

− −−

=

= +

= +∑

P P H R H

P H H

Secondly, it calculates the covariance of the sequence UKF. By the measurement-update equation, it easily gets

1 1 1 1

2 1 1 1 1 1

1

0 1 1 1 1 1

1

( ) ( ) ( )

( ) ( ) ( )

( )

( ) ( ) ( )

m m mT m mk k k k k

m m T m mk k k k

mT m mk k k

mi T i i

k k k ki

RR

R

R

− − − −

− − − − − −

−

− − − − −

=

= +

= +

+

= +∑

P P H H

P H H

H H

P H H

Substituting previous equations in this equation results in the covariance of the sequence UKF as follows:

1 1 1/ 1

1

ˆ( ) ( ) ( ) ( )m

i T i ik k k k k k

iR− − −

−=

= = +∑P P H H

Compare these two recent equations, it can be seen that the covariance of the sequence UKF is equal to that of RUKF, which means that the accuracies of the two filters are the same in theory [11]. Computational Complexity Analysis: In the sequence UKF algorithm, because

1i i iT ik k k kR− +H P H is a scalar, so this algorithm has

successfully converted the inversion of m -dimension matrix into m times division. The computational complexity has been greatly decreased. In order to compare the computational complexity between the sequence UKF and RBUKF, for their time-update algorithms are the same, here only gives the comparison results of their measurement-update algorithms in table I. From this table, it can be seen that number of calculating times in RBUKF algorithm contains the third order of measurement dimension, while the sequence UKF has only second order component. So


70

when the measurement dimension is large, the computational cost in the sequence UKF will be much less [11]. TABLE I CALCULATION TIMES COMPARISON Algorithm

Num of ×÷ Num of+ −

The sequence UKF

2 2

3 2

(5 1) 22 4

3 3

m n m n

m m m

− +

+ + +

2 2

3

5 ( 4 )

2 2

mn m m nm m

+ −

+ +

RBUKF 24 4mn mn m+ + 24mn NUMERICAL SIMULATION: In order to show the efficiency of the sequence UKF, it is applied to an example system in comparison with the RBUKF. Estimation performance and computational complexity of the filters are evaluated with Monte Carlo simulations [11]. The numerical example considered in this section is a fifth-order nonlinear model given by system, with four-dimension measurement.

1

21,1, 2, 2, 3,

22,1, 2, 5,

23,1, 2, 3,

24,5, 4, 2,

25,4, 3, 1,

sin cos 0.5 0.1( )sin (sin ) 0.1

cos exp( ) 0.1( )sin cos ( ) 0.5sin cos ( ) 0.1

k

kk k k k

kk k k

kk k k

kk k k

kk k k

vx x x xvx x xvx x xvx x xvx x x

+ =

+ − + − ++ − +

+ − + −

x

0.1 0.2 0 0 00.15 0 0 0 0

0 0 0 0.1 0.50 0 1 0 1

k

− = −

H

The covariance matrices of kv and kw are:

50.0001k =Q I , 40.01k =R I . The initial conditions for the system and the filters are [11]:

1,0 2,0 3,0 4,0 5,0 0.5x x x x x= = = = = 1,0 2,0 3,0 4,0 5,0ˆ ˆ ˆ ˆ ˆ 2x x x x x= = = = =

And the initial covariance matrix is chosen as

20 5

ˆ 100=P I .

0 50 100 150 200 250 300-1

-0.5

0

0.5

1

1.5

t/step

X5

RBUKF

X5sequence UKF

Fig.1 Comparison of state 5x

0 50 100 150 200 250 3000

0.2

0.4

0.6

0.8

1

t/step

cora

via

nce

P o

f X5

sequence UKFRBUKF

Fig.2 Comparison covariance of 5x

0

100

200

300

400

500

600

700

± in seq-UKF ± in UKF ÷× in seq-UKF ÷× in RBUKF

N/ti

me

s

Fig.3 comparison of computational complexity For briefness, it only shows the filtering results

of state 5x . From fig. 1, it can be seen the filtering accuracy of the sequence UKF is same with that of RBUKF, but curve of the RBUKF is more fluctuant than that of the sequence UKF, which means that the sequence UKF


71

maybe has a better filtering performance in actually usage. Fig. 2 is the filtering covariance of the two filters. It can be seen that the filtering covariance is almost equal to each other. Fig. 3 is the comparison of the computational complexity between the two filters for the example. It shows that the number of multiplication and division in the sequence UKF is much smaller than that in RBUKF, so is it for the number of addition and subtraction. It verifies that the computational complexity of sequence is less than RBUKF. For the computational cost of RBUKF is less than traditional UKF, So the sequence UKF will have great advantage in comparison with UKF [11].

Conclusion: In this paper, at first authors discuss about different aspects of cloud computing. They discuss about literature and principles of this technology. After that, they highlight the security of such network and remark this important factor by the means of several examples. After that, authors discuss about security problems of VMware’s and cloud infrastructures and by purposing new generation of Kalman filter they attempt to overcome to this problem. Authors claim that by using of such algorithm like Kalman filter (that is based on knowing past and present states and predict future state according to them) they can estimate and predict about lots of important factors and avoid lots of crashes.

References: [1] Mehdi Darbandi “Applying Kalman

Filtering in solving SSM estimation problem by the means of EM algorithm with considering a practical example”; published by the Journal of Computing – Springer, 2012; USA.

[2] Mehdi Darbandi; “Comparison between miscellaneous platforms that present for

cloud computing and accreting the security of these platforms by new filter”; published by the Journal of Computing – Springer, 2012; USA.

[3] Mehdi Darbandi; “New and novel technique in designing electromagnetic filter for eliminating EMI radiations and optimization performances”; published by the Journal of Computing - Springer, 2012; USA.

[4] Mehdi Darbandi; “Developing concept of electromagnetic filter design by considering new parameters and use of mathematical analysis”; published by the Journal of Computing - Springer, 2012; USA.

[5] Mehdi Darbandi; “Is the cloud computing real or hype Affirmation momentous traits of this technology by proffering maiden scenarios”; published by the Journal of Computing – Springer, 2012; USA.

[6] Mehdi Darbandi; “Measurement and collation overriding traits of computer networks and ascertainment consequential exclusivities of cloud computing by the means of Bucy filtering”; published by the Journal of Computing - Springer, 2012; USA.

[7] Mehdi Darbandi; “Unabridged collation about multifarious computing methods and outreaching cloud computing based on innovative procedure”; published by the Journal of Computing - Springer, 2012; USA.

[8] Mehdi Darbandi; “Scrutiny about all security standards in cloud computing and present new novel standard for security of such networks”; published by the Journal of Computing - Springer, 2012; USA.

[9] Microsoft’s Accessible Technology Vision and Strategy; September 2011.

[10] MSc. Thesis of Sören Bleikertz; Norwegian University of Science and Technology; June 2010.

[11] Hui-ping Li, De-min Xu and Fu-bin Zhang ; “Sequence Unscented Kalman Filtering Algorithm”.

[12] Mehdi Darbandi; “Appraising the role of cloud computing in daily life and presenting new solutions for stabilization of this technology”; published by the Journal of Computing - Springer, 2012; USA.


72

Joint Evaluation of I /Q Imbalance and

Reconfigurable RF Filter Nonlinearity in

LTE Transmitters

M. Bozic, A. Anastasijevic, K. Rabbi, N. Mohottige and D. Budimir

Wireless Communications Research Group,

Faculty of Science and Technology

University of Westminster,

London, W1W 6UW, UK

Abstract—In this paper, a joint evaluation of I/Q imbalance

and reconfigurable bandpass filter nonlinearity in wireless

transmitters is described. An experimental analysis of complete

Orthogonal Frequency Division Multiplex (OFDM) transmitter is

presented for the purpose of quantifying these nonlinearities

using LTE R9 3 MHz 16 QAM and LTE R9 3 MHz 64 QAM

signals.

Keywords—LTE, Nonlinear distortion, reconfigurate filters;

I/Q imbalance; wireless transmitters.

I. INTRODUCTION

Wireless communication systems are developing to support

more users and provide higher data rates in a limited radio-

frequency (RF) spectrum. Protocols that support modern

wireless communication systems allow users to access a

variety of multimedia services, potentially providing it with a

speed of 100 Mb/s. In the transmitter chain, modulator, PA

and filter are the most challenging blocks. As nonlinear

elements, these components cause distortion to the transmitted

signals which significantly degrade the quality of the signal.

The Quadrature and In-phase carriers in the analog modulator

do not have exactly the same amplitudes and phase

differences. These discrepancies are called gain/phase

imbalance and can cause crosstalk between the I and Q

channels, which degrade quality of the signal [1]-[4].

This paper is an evaluation of the nonlinearity effects with

and without I/Q imbalance in reconfigurable RF circuits for

wireless transmitters. It is applied for analyzing 16 QAM

OFDM/64 QAM OFDM signal in wireless transmitter. It will

be shown that by introducing I/Q imbalance an additional

distortion appears in wireless transmitter. Experimental

analysis for LTE R9 16 QAM (16 QAM OFDM) and LTE R9

64 QAM (64 QAM OFDM) signals of a wireless transmitter

shows that out-of-band distortion increase correspondingly

with addition of the mentioned undesired effect. The paper is

organized as follows. At first, the effects of I/Q imbalance is

explained using baseband methodology. Experimental setup

and results are presented in section III. Finally, the conclusion

is given in section IV.

II. I/Q IMBALANCE

Nonlinearity of the components in transmitter chain

produces compression of signal amplitude and phase, degrades

the quality of the transmitted signal. Modulator is nonlinear

element which up-converts the baseband signal to RF.

Problem with modulator is that it has phase and gain

imbalances that affect transmitter’s performances. This

disturbs the ideal 90˚ degree phase relationship between I and

Q signals along with gain imbalance [5]-[7]. The output of

modulator can be represented as:

[ ] [ ] (1)

where is the imbalanced signal, is gain imbalance, 𝜃 is

phase imbalance and is 16/64 QAM OFDM input signal.


73

I PATH

Q PATH

90°- θI + θQ

TO RECEIVER

LPFSWITCH

AMPLIFIER

AMPLIFIER

LPFMODULATOR

RECONFIGURABLE

FILTER

Fig. 1. Wireless transmitter with undesired effects

III. RESULTS

A. Measurement setup

Measurement setup consists of signal generator Agilent

MXG N5182A, reconfigurable microstrip bandpass filter as

DUT, which is illustrated in Fig. 2, used to emulate wireless

transmitters. The layout of the reconfigurable circuit

(bandpass filter) is based on the combination of a bent single

λg/2 resonator and a pair of bent λg/4 short circuited resonators.

The filter is inductively coupled to the source. Compactness of

this filter is attained through the reduction of filter length and

width. In addition, two bent short stubs are shorted to common

ground in order to miniaturize the filter. This filter is designed

to have a 3 dB passband from 925 MHz – 960 MHz (LTE

band 8) with a mid-band frequency of 942.50 MHz. The

proposed filter layout comprises of lines of width 1.2 mm and

the open circuited stub leading to the gap is 2.4 mm wide.

Fig. 2. Reconfigurable bandpass filter as DUT.

General-purpose interface bus (GPIB) was used to connect

this generator with PC. The signals were created in Matlab and

download to MXG using Agilent Signal Studio Toolkit. The

signal named RF output was passed through DUT

(reconfigurable pin switch based bandpass filter). Finally, the

signal was captured with VSA 4406A for signal analysis. The

measurement setup is shown in Fig. 3.

MXG-N5182A

DUT

GPIB

PC

Agilent Signal Studio Toolkit

VSA 4406A

RF OUT

Matlab

Advanced Design System

Fig. 3. Measurement setup of wireless transmitter.

B. Measurement results

Experiment was conducted for two different cases of input

signals, LTE R9 3 MHz 16 QAM and LTE R9 3 MHz 64

QAM signals with and without I/Q imbalance. Spectrum of

the LTE R9 3 MHz 16 QAM signals with and without I/Q

imbalance are shown in Figs. 4a, 4b, 4c and 4d respectively.

The measured power spectrum of the LTE R9 3 MHz 64

QAM signals with/without I/Q imbalance are shown in Figs.

5a, 5b, 5c and 5d respectively.


74

Fig. 4a. The measured power spectra of the LTE R9 3 MHz

16 QAM signal at the input of the reconfigurable filter

without I/Q imbalance.

Fig. 4b. The measured power spectra of the LTE R9 3 MHz

16 QAM signal at the output of the reconfigurable filter

without I/Q imbalance.

Fig. 4c. The measured power spectra of the LTE R9 3 MHz

16 QAM signal at the input of the reconfigurable filter with

I/Q imbalance.

Fig. 4d. The measured power spectra of the LTE R9 3 MHz

16 QAM signal at the output of the reconfigurable filter

with I/Q imbalance.

Fig. 5a. Spectrum of the LTE R9 3 MHz 64 QAM signal at

the input of the reconfigurable filter without I/Q imbalance.

Fig. 5b. Spectrum of the LTE R9 3 MHz 64 QAM signal at

the output of the reconfigurable filter without I/Q

imbalance.


75

Fig. 5c. Spectrum of the LTE R9 3 MHz 64 QAM signal at

the input of the reconfigurable filter with I/Q imbalance.

Fig. 5d. Spectrum of the LTE R9 3 MHz 64 QAM signal at

the output of the reconfigurable filter with I/Q imbalance.

Quantitative measures of distortion is defined as:

Δ (2)

where represents output power of the transmitted signal

and represents distortion power. Measurement results

with and without I/Q imbalance are presented in Tables 1 and

2, respectively.

TABLE 1: DISTORTION OF RECONFIGURABLE BANDPASS FILTER IN

WIRELESS TRANSMITTER WITHOUT I/Q IMBALANCE

M QAM modulation

Input Power [dBm]

Δ [dB] without I/Q imbalance

16 16 0.46

64 16 1.93

TABLE 2: DISTORTION OF RECONFIGURABLE BANDPASS FILTER IN

WIRELESS TRANSMITTER WITH I/Q IMBALANCE M QAM

modulation Input Power [dBm]

Δ [dB] with I/Q imbalance

16 16 3.82

64 16 6.7

ACKNOWLEDGEMENT

Financial support was provided by the EU-Erasmus Mundus

Action 2 project EUROWEB.

REFERENCES

[1] D. Saffar, N. Boulejfen, F. Ghannouchi , M. Helaoui and A. Gharssalah,

“Behavioral Modeling of MIMO Transmitters Exhibiting Nonlinear Distortion and Hardware Impairements,” Proceedings of the 6th

European Microwave Integrated Circuits Conference

[2] Y.-D. Kim, E.-R. Jeong and Y. H. Lee, "Adaptive Compensation for Power Amplifier Nonlinearity in the Presence of

QuadraturModulation/Demodulation Errors," IEEE Transactions Signal

Process., vol. 55, no. 9, pp. 4717-4721, Sep. 2007

[3] F. H. Raab, P. Asbeck, S. Cripps, P. B. Kenington, Z. B. Popovic, N.

Pothecary, J. F. Sevic, and N. O. Sokal, “Power Amplifiers and

Transmitters for RF and Microwave“, IEEE Transactions on Microwave Theory and Techniques, vol. 50, no. 3, pp. 814-826, March 2002.

[4] P. B. Kenington, "High Linearity RF Amplifier Design", Artech House,

2000. [5] S.Bhattacharjee, Rawat, K.; Rawat, M.; Donglin Wang; Helaoui, M.;

Leung, H.; Ghannouchi, F.; , "Joint evaluation and mitigation of RF

impairments and nonlinear distortion in WiMAX Transceiver under Nakagami-m fading channel," Electrical and Computer Engineering

(CCECE), 2011 24th Canadian Conference on , vol., no., pp.000926-000929, 8-11 May 2011

[6] L.Ding, Z. Ma, D. Morgan, M. Zierdt, and G. Tong Zhou,

“Compensation of frequency-dependent gain/phase imbalance in predistortion linearization systems,” IEEE Transactions Circuits and

Systems I, vol 55, no. 1, pp. 390-397, Feb. 2008.

[7] L. Anttila, P. Handel, and M. Valkama, “Joint mitigation of power amplifier and IQ modulator impairments in broadband direct-conversion

transmitters,” IEEE Transactions Microwave Theory and Techniques,

vol. 58, no. 4, pp. 730-739, Apr. 2010.


76

An Efficient Delay Estimation Model for High Speed

VLSI Interconnects

M.Kavicharan

Dept. of ECE

National Institute of Technology

Warangal, India

[email protected]

N.S.Murthy

Dept. of ECE


Warangal, India

[email protected]

N. Bheema Rao

Dept. of ECE


Warangal, India

[email protected]

Abstract —In this paper a closed-form matrix rational model for

the computation of step and finite ramp responses of Resistance

Inductance Capacitance (RLC) interconnects in VLSI circuits is

presented. This model allows the numerical estimation of delay

and overshoot in lossy VLSI interconnects. The proposed method

is based on the U-transform, which provides rational function

approximation for obtaining passive interconnect model. With

the reduced order lossy interconnect transfer function, step and

finite ramp responses are obtained and line delay and signal

overshoot are estimated. The estimated delay and overshoot

values are compared with the Euder method, Pade method and

HSPICE W- element model. The 50% delay results are in good

agreement with those of HSPICE within 0.5% error while the

overshoot error is within 1% for a 2 mm long interconnect. For

global lines of length more than 5 mm in SOC (system on chip)

applications, the proposed method is found to be nearly four

times more accurate than existing methods.

Keywords – Delay; matrix rational model; ramp input; RLC

interconnects; transient analysis; transfer function; U-

approximation.

I. INTRODUCTION

As the physical dimensions in VLSI technologies scale

down, interconnect delay dominates the gate delay in

determining circuit performance [1]. In deep submicron VLSI

circuits it is necessary to have computationally economical

and accurate interconnect delay models. Thus for the design of

complex circuits, more accurate analytic models are needed to

predict the interconnect delay accurately.

Originally VLSI interconnects were modeled as RC lines

and single pole Elmore-based models [2]–[3] because of long channel device delay dominance over negligible interconnect delay. However for high speed interconnects, inductance effects are becoming progressively important and can no longer be ignored. Under these circumstances, the Elmore model fails since it does not consider the inductance effects [4]. It is necessary to use a second-order model, which includes the effect of inductance. Kahng et al. considered equivalent Elmore delay model based on the Resistance Inductance and Capacitance (RLC) of the interconnects [4] and [5]. Ismail et al. [6] proposed two pole model to capture far end time domain solution for single line interconnect.

A simplified voltage transfer function obtained using

Taylor series approximation for transient analysis [7]-[8] has

less accuracy in delay calculation. Nakhla et al.[9] use

modified nodal analysis (MNA) for obtaining far end and near

end responses of interconnects. Roy [10] extended [9] for

obtaining more accurate far end responses of coupled RLC

interconnects using delay algebraic equations.

A matrix rational-approximation model for SPICE

analysis of high-speed interconnects is presented in [11]-[12],

however, the approximations made to derive the models

contributed to inaccuracy. This has been extended using Pade

approximation model [13] to estimate the delay of

interconnects. All the above models still suffer from various

inaccuracies and need improvement for accurate delay

estimations.

In this paper, we present an improved analytic delay

model by extending the concepts developed in [11]-[13] for

on-chip RLC interconnects. The accuracy of Euder

approximation method [16] has been improved by using fourth

order MacLaurin series for RLC interconnects and compared

with Pade method, proposed method and HSPICE. The

proposed model is based on U-transform [14]-[15], which is

simple in structure and easier to implement. For a given

number of terms used in the transform, the U-approximant

requires less algebraic manipulations than the Pade scheme

and thus computationally less expensive. This U-transform is

used to solve the Telegraphers equation solution for the first

time.

The remainder of the paper is organized as follows. Section

II briefly describes the mathematical analysis to determine the linear transfer function of RLC interconnect to find the transient analysis. Section III develops the proposed U-model for single RLC line. For validation of the proposed model simulation results are compared with standard HSPICE and reported in sections IV. Conclusions and future scope appear at the end.


77

I. ANALYSIS OF RLC INTERCONNECT

The solution of interconnects are described by telegrapher’s

equations as

szIsLRszVx

,,

szsCVszIx

,,

(1)

where ‘s’ is the Laplace-transform variable, z is a

variable which represents position; szV , and szI , stand

for the voltage and current vectors of the transmission line,

respectively, in the frequency domain; and R, L and C are the

per unit length (p.u.l.) resistance, inductance, and capacitance

matrices, respectively.

The solution of (1) can be written as an exponential

matrix function as

sI

sVe

sdI

sdVd

,0

,0

,

,

(2)

where

0

0

Y

Z

and ‘d’ is the length of the transmission line, with

Z=R+sL and Y=sC. The exponential matrix of (2) can be

written in terms of cosh and sinh functions as

YZdYZdY

YZdYZYdd

e coshsinh

sinhcosh

0

1

0

where 1

0 )( YZYY

Equation (2) does not have a direct representation in the time

domain, so it is difficult to analytically predict the delay and

overshoot of transmission lines.

II. PROPOSED U MODEL

This model is based on a generalized U-transform [14].

For the power series expansion of a function f(x), where ‘x’ is

a complex variable

n

n

nxaxf

0

)(

The sequence sn is a partial sum of original series

1

0

n

k

k

kn xas

The closed form rational function approximation for an

exponential matrix is

i

knj

k

j

ijknj

k

j

ikn

j

nknxw

awxSu

0

0

1

0

(3)

where

1

2

!!

!1

jkn

kj

knja

jkn

jkj

kw

(4)

Thus ukn represents a table of rational functions, each element

of which is obtained from n + k terms of the original sequence

Sn, n = 1, 2,... and is an approximant of the function f(x)

specified above.

Calculation procedure for estimating delay and overshoot

using U-approximants are as follows.

(i) Use the Interconnect line parameters as per Table I.

(ii) Telegrapher’s equations are solved and the solution

can be written as exponential matrix.

(iii) This transfer function matrix parameters can be

approximated using the U-model.

(iv) In the proposed model calculate the coefficient of the

exponential function i.e., ai where

(v) Calculate wknj from the relation (4)

(vi) Calculate the inner sum of the Eq (3) numerator.

(vii) Total sum of the numerator is obtained

(viii) Calculate the total sum of the denominator of the U-

approximants

(ix) Calculate the U-approximants

(x) Make use of the U-approximants to get

approximated transfer function

(xi) Find the time domain response of approximated

transfer function using inverse Laplace transform to

estimate delay and overshoot of interconnect.

The basic idea of the matrix rational-approximation model

is to use predetermined coefficients to analytically obtain

rational functions for (2). To obtain a passive model, the

exponential function eФd

is approximated using Eq (3) and the

resultant model is used for obtaining time response.

A single RLC line is shown in Fig. 1.The line is driven by

a step input and 1-V finite ramp with rise time of 0.1 ns. This

represents a point-to-point interconnection driven by a

transistor (modelled as a resistance Rs) and connected to the

next gate (modelled as a capacitance Cl).

ni 0


78

Fig. 1. Circuit model of the single-line distributed RLC interconnect.

The frequency-domain solution at the far end is expressed

as

dYsCYRdCsR

VV

lsls

inf

sinhcosh1

1

00

(4)

where

YZ ,

Rs is the source resistance at the near end, Cl is the load

capacitance at the far end, and Vin is the input voltage. The

exact transfer function of distributed RLC transmission line

has cosh and sinh terms, which are multiplied with Yo and it’s

inverse. It is extremely difficult to find the time domain

response of this complex transfer function, so several

approximations are proposed in literature to find the time

domain response. An approximate transfer function has been

derived using U-transform. This transfer function is inverse

Laplace transformed to get time domain response for

estimation of delay and overshoot in single RLC interconnect.

III. SIMULATION RESULTS

The single RLC line is presented in this section to

demonstrate the validity and efficiency of the proposed

method. The results were obtained using MATLAB R2010a

operating on HP 64-bit Intel i5 processor with clock speed of

2.53 GHz and are also compared with HSPICE using the W-

element method.

The typical interconnect parameters [13] considered for

simulation of single RLC interconnect are given in table-I.

The Pade approximation, Eudes model and proposed U-

approximation are implemented in MATLAB for the same set

of input parameters and various approximation orders.

Table I: The values of Interconnects parameters [13]

The accuracy of proposed model validated using the

frequency response of cosh function as shown in Fig. 2. The

frequency response is obtained using pade (3/3) and proposed

U-model (3/3) are compared with the exact solution of

telegrapher’s equations. It is observed that, the proposed

method is better than Pade method and well matches with

exact cosh function for the order of 3/3 up to the frequency of

25 GHz. The far-end responses to a finite ramp input of single

interconnect is plotted in Fig. 3. The plots compared the responses of proposed, Pade and HSPICE W-element models.

0 0.5 1 1.5 2 2.5

x 1010

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Frequency(GHz)

Ab

so

lute

valu

e o

f co

sh

fu

ncti

on

Frequency response of cosh function

Exact cosh function

proposed u approximation order 3/3

Pade model order 3/3

Fig: 2. Frequency response of exact cosh function, proposed U-approximation

order 3/3 and Pade approximation order 3/3.

From Fig. 3, it is noticed that, the proposed U-

approximation and Pade method are very close as compared to

HSPICE. But Eudes model of order 4 has more overshoot as

compared to other methods.

Fig: 3.Transient analysis of single interconnect line, when length

=0.2cm, Rs=50Ω and Cl=50fF.

The MATLAB results of step response and finite ramp

response are plotted for the line length of 0.2 cm, source

resistance of 100Ω and load capacitance of 100fF. Step

response in Fig. 4 has less ringing in proposed method as

compared to Pade method, for the same approximation order

of 3/3, whereas Fig. 5 gives finite ramp response of single line

interconnect using U-model matches very well with the


79

HSPICE. But Eudes model needs more settling time as

compared to the proposed model.

0 0.5 1 1.5 2 2.5 3 3.5 4

x 10-10

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Frequency (GHz)

Voltage (

v)

Input Ramp

Eudes model order (4)

Pade model order (3/3)

Prposed model order (3/3)

Fig: 4.Step response of single line when length =0.2cm, Rs=100Ω and Cl=100fF.

0 0.5 1 1.5 2 2.5 3 3.5 4

x 10-10

-0.2

0

0.2

0.4

0.6

0.8

1

1.2

Step Response

Time (sec)

Vo

ltag

e (

v)

Eudes model order (4)

Pade model order (3/3)

Proposed model order (3/3)

Fig: 5.Ramp response of single line when length =0.2cm, Rs=100Ω, Cl=100fF.

Table II: Comparisons of 50% delay of HSPICE W Element, Eudes model Pade model and proposed model for various lengths, source Resistances and load Capacitances.

Table III: Comparisons of overshoot of HSPICE W Element, Eudes model, Pade model and proposed model for

various lengths, source Resistances and load Capacitances.

The Tables II and III give the comparisons of 50% delay

and overshoot values obtained using HSPICE for various

lengths, source Resistances and load Capacitances. These

tables include the percentage error values with respect to

HSPICE. From Table II the Eudes model of order 4 has worst

case error of 11.42%, whereas Pade and proposed models have

8.69% and 2.811%.

It can be observed that the methods implemented for

global lines have more error percentage than our proposed

method. Both Pade and proposed methods perform similarly

for smaller length interconnects while Eudes method has more

error percentage.

As noticed in Table III, the Eudes model has worst case

overshoot error percentage of 9%, but Pade model has an error

percentage up to 2% while the proposed model has error

within 1%. In the case of overshoot estimation our model is

best for all cases. For 2 mm range lines the proposed method

has delay and overshoot errors within 1% .

L (cm)

Rs (Ω)

Cl (fF)

HSPICE Eudes model order (4)

Pade model order 3/3

Proposed Model order (3/3)

50% delay (ps)

50% delay (ps)

(%Error)

50% delay (ps)

(%Error)

50% delay (ps)

(%Error)

0.2

50 50 79.8 79.1 (0.8%) 80.2 (0.5%) 80.2 (0.5%)

100 100 98.7 96.8 (1.92%) 98.6 (.1%) 98.65 (0.05%)

0.5

50 50 135.7 142.8 (5.23%) 137.8 (1.54%) 137.7 (1.4%)

100 100 156.6 162.6 (3.83%) 151.9 (3%) 155.3 (0.83%)

1.0

50 50 231.2 250.2 (8.21%) 211.1 (8.69%) 224.7 (2.811%)

100 100 255.6 284.8 (11.42%) 249.7 (2.3%) 252.5 (1.21%)

L (cm)

Rs (Ω)

Cl (fF)

HSPICE Eudes model order (4)

Pade Order 3/3 Proposed Model order (3/3)

Overshoot (V)

Overshoot (V)

(%Error)

Overshoot (V)

(%Error)

Overshoot (V)

(%Error)

0.2

50 50 1.14 1.14 (0%) 1.12 (1.7%) 1.13 (0.87% )

100 100 1.00 1.00 (0%) 1.00 (0%) 1.00 (0% )

0.5

50 50 1.15 1.24 (7.8%) 1.15 (0%) 1.14 ( 0.87%)

100 100 1.00 1.03 (3%) 1.00 (0%) 1.00 (0% )

1.0

50 50 1.00 1.09 (9%) 1.02 (2%) 1.01 (1% )

100 100 1.00 1.00 (0%) 1.00 (0%) 1.00 (0% )


80

CONCLUSION

This paper presents a U-transform based closed form model for delay and overshoot estimation of high speed VLSI interconnects in DSM

regime. A single line interconnect has been used for validating the proposed

model by comparing with the Eudes model, Pade method and HSPICE. In SOC (system on chip) applications, for global lines of lengths 2 mm and

above the proposed method is found to be more accurate than existing

methods. This method can be used to estimate the signal integrity characteristics of Carbon nano tubes.

REFERENCES

[1] Semiconductor Industry Association. The International Technology Roadmap for Semiconductors. 1999.

[2] W. C. Elmore, “The transient response of damped linear networks with particular regard to wideband amplifiers,”J. Appl. Phys., vol. 19, no.1, pp. 55–63, Jan. 1948.

[3] T.Sakurai, “Closed-form expressions for interconnection delay, cou-pling, and crosstalk in VLSI’s,”IEEE Trans. Electron Devices, vol. 40, no. 1, pp. 118–124, Jan. 1993.

[4] Kahng A B, Muddu S, “An analytical delay for RLC interconnects”, IEEE Trans. Computer-Aided Design of Integrated Circuits and Systems, 1997, 16(12): 1507−1514

[5] Kahng A B, Muddu S, “Two-pole analysis of interconnection trees”, Proc. of IEEE MCMC Conf.(MCMC-95), Santa Cruz, California, USA: 1995, 105−110.

[6] Xiaopeng Ji, Long Ge, Zhiquan Wang, “Analysis of on-chip distributed interconnects based on Pade expansion”, Journal of Control Theory and Applications, 2009, 7 (1) pp. 92–96.

[7] Ren Yinglei, Mao Junfa and Li Xiaochun, "Analytical delay models for RLC interconnects under ramp input", Frontiers of Electrical and Electronic Engineering in China, vol. 2, pp. 88-91, March 2006.

[8] Y. I. Ismail, E. G. Friedman, and J. L. Neves, “Equivalent Elmore delay for RLC trees,” IEEE Trans. Comput.-Aided Design Integr. Circuits Syst.,vol. 19, no. 1, pp. 83–97, Jan. 2000.

[9] N.Nakhla ,A.Dounavis,R.Achar,M.Nakhla,“DEPACT:Delay Extraction BasedPassive Compact Transmission-Line Macromodeling Algorithm,” IEEE Transactions on Advanced Packaging, vol. 28, issue 1, pp. 13-23, Feb 2005.

[10] S.Roy, A. Dounavis, “Efficient Delay and Crosstalk Modeling of RLC Interconnects using Delay Algebraic Equations”, IEEE Transactions on Very Large Scale Integration Systems, vol. 2, issue 2, pp. 342-345, Feb. 2011.

[11] A. Dounavis, R. Achar, and L. Xin,“Passive closed-form transmission-line model for general-purpose circuit simulators,” IEEE Trans. Microw. Theory Tech., vol. 47, no. 12, pp. 2450–2459, Dec. 1999.

[12] A. Dounavis, R. Achar, and M. Nakhla, “A general class of passive macromodels for lossy multiconductor transmission lines,”IEEE Trans. Microw. Theory Tech., vol. 49, no. 10, pp. 1686–1696, Oct. 2001.

[13] S. Roy and A. Dounavis, “Closed-Form Delay and Crosstalk Models for RLC On-Chip Interconnects using a Matrix Rational Approximation”, IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 28, issue 10, pp. 1481 - 1492, Oct.2009.

[14] Roy, D., Bhattacharya, R., and Bhowmick, S., “Rational approximants generated by the U-transform,”Comput. Phys. Commun. 78, 29 – 54 (1993).

[15] Tarun Kumar Sheel, Thesis “Rational approximants generated by Pade approximation and U-transform” University of Dhaka, Bangladesh MARCH 1997.

[16] T.Eudes, B. Ravelo, and A. Louis,“Transient response characterization of the high-speed interconnection RLCG-model for the signal integrity analysis,” PIER Journal, Vol. 112, pp. 183–197, 2011.


81

Design of 8-bit Dynamic CMOS Priority Resolvers

based on Active-High and Active-Low Logic

Preeti Panchal, C. Vinitha, Rashi Srivastava,

P. Balasubramanian

Department of Electronics & Communication Engineering

S. A. Engineering College, Chennai 600 077, India

[email protected]

N. E. Mastorakis

Division of Electrical Engineering & Computer Science

Military Institutions of University Education

Hellenic Naval Academy, Piraeus 18539, Greece

[email protected]

Abstract—A couple of new dynamic CMOS based designs of

an 8-bit priority resolver corresponding to active-high and

active-low logic are presented in this paper. The proposed

designs result from modifications to an 8-bit priority resolver

designed by Huang and Chang [15], which pertains to active-high

logic. Compared to Huang and Chang’s original 8-bit CMOS

priority resolver, the modified designs achieve 4× mean reduction

in power dissipation, and report average improvement in the

power-delay product by 43%. The simulation results were

obtained using Tanner tools (TSPICE), and correspond to a

0.25µm CMOS process technology.

Keywords—Priority encoder; Dynamic CMOS; Low power;

Power-delay product; Digital integrated circuit

I. INTRODUCTION

Priority resolvers, also called ‘priority encoders’, often find applications in data bus and comparators [1 – 3], fixed and floating point arithmetic units [4], incrementer/decrementer circuits [5] [6], interconnection network routers [7] [8] and content-addressable memories [9] [10]. A ‘priority resolver’ has similar number of inputs and outputs, and basically acts as an arbiter in providing priority-based access for devices connected at its inputs to the devices/peripherals connected at its outputs. The inputs and outputs of a priority resolver are assigned priorities to avoid any contention issues. It is usual practice to design a small-size priority resolver (4-bits or 8-bits) and then use this basic building block to realize larger size priority resolvers using a cascade. This procedure helps to contain signal propagation through a series network of pMOS or nMOS transistors. With respect to dynamic CMOS logic style, the outputs of the priority resolver are either ‘set’ or ‘reset’ during the precharge phase of the synchronizing clock. During the clock’s evaluation phase, the data token associated with the highest priority input is alone transferred to the output side.

Delgado-Frias and Nyathi [11] designed a priority encoder that permits sequential passage of priority token from the highest priority primary input to the lowest priority input – the disadvantage of this design being that the sequential passage of priority token encounters a delay of O(n), where ‘n’ represents the total number of primary inputs or outputs. To alleviate the linear increase in delay, Wang and Huang [12] put forward two 8-bit priority encoder designs, comprising two 4-bit encoder

blocks with the provision of an internal lookahead signal – one of the designs extensively utilizes pMOS transistors while the other design widely deploys nMOS transistors. Kun et al. [13] came up with the design idea of an 8-bit priority encoder module, eliminating the need for sub-modules and internal lookahead signalling. While Huang et al. [6] proposed a serial cascading architecture to realize higher order priority encoders, with the lookahead output of a 8-bit encoder module serving as the lookahead input for the succeeding encoder block, Kun et al. [13] proposed a parallel priority-based cascading topology to implement larger size priority encoders. Mohanraj et al. [14] presented a new 8-bit priority encoder design, which is in fact a refinement of Kun et al.’s encoder design by exploiting shared logic to reduce the number of devices needed for physical realization. Huang and Chang [15] introduced a new NOR-based priority encoder, where during the precharge phase of the clock, all the outputs are driven to ‘high’ state, and in the evaluation phase, based upon input request(s), the input that assumes a higher priority is enabled and its corresponding output is retained as ‘high’, while the other primary outputs are pulled to ‘low’. In this aspect, the Huang and Chang’s design is similar to Huang et al.’s pMOS-based priority encoder design. All the priority encoder designs surveyed thus far correspond to active-high logic, which implies input request(s) have to be ‘high’ so as to activate the priority resolver to produce the desired ‘high’ output.

The remainder of this paper is organized into three sections. Section 2 describes the operation of Huang and Chang’s basic 8-bit priority encoder. Modifications made to this circuit to realize new power optimized active-high and active-low priority resolver modules are discussed in Section 3. Lastly, Section 4 presents the simulation results and conclusions.

II. HUANG AND CHANG’S 8-BIT PRIORITY RESOLVER

The 8-bit dynamic CMOS priority resolver design according to Huang and Chang [15] is shown in Figure 1, which uses 76 transistors in total. IP1 to IP8 represent the primary inputs, while OP1 to OP8 specify the primary outputs. Here, IP1 (OP1) is the least significant primary input (output), and is assigned the highest priority amongst all the inputs (outputs). The equations governing the priority resolver functionality are given next, and the truth table is given in the Appendix. From equations (1) to (8), it may be evident that the


82

priority resolver is NOR-based, and Figure 1 implements these equations predominantly using nMOS transistors.

11 IPOP = (1)

( )212 IPIPOP += (2)

( )3213 IPIPIPOP ++= (3)

( )43214 IPIPIPIPOP +++= (4)

( )543215 IPIPIPIPIPOP ++++= (5)

( )6543216 IPIPIPIPIPIPOP +++++= (6)

( )76543217 IPIPIPIPIPIPIPOP ++++++= (7)

( )876543218 IPIPIPIPIPIPIPIPOP +++++++= (8)

Fig. 1. Huang and Chang’s 8-bit priority resolver

During the precharge phase (falling-edge) of clock signal, CLK, pMOS transistors ‘p1’ to ‘p8’ will turn-on, driving the primary outputs OP1 to OP8 to 1 (preset). At this time instant, nMOS transistors ‘clk1 to clk8’ will remain off. During the clock evaluation phase (rising-edge of CLK), pMOS transistors ‘p1’ to ‘p8’ will turn-off, and nMOS transistors ‘clk1’ to ‘clk8’ will turn-on. If the lookahead input is 1, then nMOS transistors ‘la1’ to ‘la8’ will also turn-on. Under this condition, let us assume the primary input data to be IP8 – IP1 = 00101000, i.e. IP6 = IP4 = 1, and the other inputs are 0. Given that IP4 and IP6 are equal to 1, nMOS transistors ‘M1’ and ‘M5’ will turn-off, and the other nMOS transistors ‘n1’ to ‘n12’ will turn-on. As a result, the pre-charged primary outputs OP1 to OP3 and OP5 to OP8 have a discharge path through at least one nMOS transistor which eventually drives down their output value to 0. It can be seen that outputs OP1 to OP3 and OP6 have a single discharge path via transistors ‘n1’, ‘n2’, ‘n3’ and ‘n8’ respectively, while output OP5 has two discharge paths via nMOS transistors ‘n4’ and ‘n7’. Primary outputs OP7 and OP8 both have three respective discharge paths through the set of nMOS transistors (‘n5’, ‘n9’, ‘n11’) and (‘n6’, ‘n10’, ‘n12’). Since IP4 is associated with a higher priority than IP6, and since nMOS transistors ‘M1’ to ‘M4’ remain off, output OP4 does not have a discharge path and hence OP4 remains ‘high’ (1), which is indeed the desired output. It is observed that since the primary outputs are all ‘set’ initially due to pre-charging of pMOS transistors, and more than one nMOS transistor could turn-on during the clock evaluation phase to reset the required outputs, with the lookahead and clocked nMOS transistors also turning-on, the switching activity (low-to-high, and high-to-low) of this 8-bit encoder module is inherently high, which could lead to increased power dissipation.

III. MODIFIED HUANG AND CHANG’S PRIORITY RESOLVER

– ACTIVE-HIGH AND ACTIVE-LOW CONFIGURATIONS

A. Modified Huang and Chang’s 8-bit Priority Resolver –

Active-High Configuration

From the operating principle of Huang and Chang’s 8-bit priority resolver discussed previously, it is clear that the nMOS transistors associated with lookahead-input and clock signals, viz. ‘la1’ to ‘la8’ and ‘clk1’ to ‘clk8’ turn-on during the leading clock edge, and thus they provide a discharge path for the unnecessary outputs to be reset. Therefore, there arises an opportunity for making many of these transistors redundant by having shared input-lookahead and clock-driven nMOS transistors, which may be designated with high aspect ratios for improved driving capability.

The modified Huang and Chang’s 8-bit priority resolver incorporating the proposed circuit optimization strategy is shown in Figure 2, which has only 62 transistors, i.e. a 18.4% reduction in device count has been achieved in comparison with the original priority resolver macro shown in Figure 1. This could in turn translate into reduced switching activity and subsequently help in minimizing total power dissipation, which was in fact confirmed through TSPICE simulations. The operation of the modified 8-bit priority resolver operation is similar to that portrayed in Figure 1, with the exception of nMOS transistors ‘lat’ and ‘clkt’ serving as replacement for the


83

groups of nMOS transistors ‘la1’ to ‘la8’ and ‘clk1’ to ‘clk8’ of Figure 1 respectively. Equations (1) to (8), mentioned earlier, hold well for the modified Huang and Chang’s 8-bit priority resolver that adheres to active-high logic.

Fig. 2. Modified Huang and Chang’s 8-bit priority resolver module, corresponding to active-high logic

B. Modified Huang and Chang’s 8-bit Priority Resolver –

Active-Low Configuration

This section deals with the modification of Huang and Chang’s 8-bit priority resolver macro to suit active-low logic. Active-low logic implies the priority resolver accepts binary 0 as input request(s) instead of binary 1 as is the case with active-high logic – the truth table is given in the Appendix. The equations which characterize the modified Huang and Chang’s 8-bit priority resolver conforming to active-low logic are given as follows.

11 IPOP = (9)

( )212 IPIPOP += (10)

( )3213 IPIPIPOP ++= (11)

( )43214 IPIPIPIPOP +++= (12)

( )543215 IPIPIPIPIPOP ++++= (13)

( )6543216 IPIPIPIPIPIPOP +++++= (14)

( )76543217 IPIPIPIPIPIPIPOP ++++++= (15)

( )876543218 IPIPIPIPIPIPIPIPOP +++++++=

(16)

OP8

IP8

OP7

IP7

OP6

IP6

OP5

IP5

OP4

IP4

OP3

IP3

OP2

IP2

OP1

IP1

LAin

CLK

p8

p7

p6

p5

p4

p3

p2

p1

clktlat

n8

n7

n9

n1

n2

n3

n4

n5

n6

M1

M7

M4M2 M3 M5 M6

Fig. 3. Modified Huang and Chang’s 8-bit priority resolver macro, pertaining to

active-low logic

The 8-bit priority resolver circuit that implements (9) – (16) is depicted in Figure 3. During the falling-edge of clock, CLK, pMOS transistors ‘p1’ to ‘p8’ are pre-charged as usual causing all the primary outputs OP1 to OP8 to become high (binary 1). At the rising clock edge, the pMOS transistors turn-off, and the nMOS transistor ‘clkt’ turns-on; with the input lookahead also getting enabled, nMOS transistor ‘lat’ turns-on. The operation of the above circuit is now briefed by considering an example


84

primary input data as IP8 – IP1 = 10011111, i.e. IP7 and IP6 are asserted low (binary 0, signifying valid input requests), while the rest of the inputs are asserted high (binary 1). Since IP1 to IP5 and IP8 is 1, nMOS transistors ‘n1’ to ‘n5’ and ‘n6’ turn-on, resulting in the discharge of the corresponding outputs OP1 to OP5 and OP8 to a low potential (binary 0); output OP8 has two additional discharge paths via nMOS transistors ‘n8’ and ‘n9’, because IP6 and IP7 are 0. As the nMOS transistor ‘n7’ is turned-on since IP6 is equal to 0, output OP7 discharges via ‘n7’, ‘lat’ and ‘clkt’ and eventually becomes 0, although the nMOS transistor ‘M7’ remains off. As a result of the parallel combination of nMOS transistors ‘M1’ to ‘M6’ remaining off, output OP6 continues to maintain its high potential (i.e. binary 1), which is indeed the desired output since IP6 (OP6) assumes higher priority than IP7 (OP7). Comparing this 8-bit priority resolver with the one portrayed in Figure 2, it is evident that the total device count is reduced from 62 to 60 transistors in case of the former compared to the latter. Thus the modified Huang and Chang’s 8-bit priority resolver corresponding to active-low logic has managed to achieve a 21% reduction in the number of devices in comparison with the original active-high design counterpart.

IV. SIMULATION RESULTS AND CONCLUSIONS

Five 8-bit dynamic CMOS based priority resolvers were designed in a full-custom fashion at the transistor level using Tanner tools, and was simulated using Tanner SPICE targeting a licensed 0.25µm bulk CMOS process. A set of random input patterns was applied at a clock frequency of 50MHz to verify the functionality of the different priority resolvers, and also to estimate the average power dissipation. The total power values estimated for various priority resolvers are given in Table 1.

TABLE I. TOTAL POWER DISSIPATION OF VARIOUS 8-BIT DYNAMIC

CMOS PRIORITY RESOLVERS

Encoder

Type

Power Dissipation

(mW)

Factor

Increase

Kun et al. [13] 66.8 2.3×

Mohanraj et al. [14] 62.7 2.2×

Huang and Chang [15] 178.2 6.2×

Modified Huang and Chang

(Active-High) – This work 85.1 2.9×

Modified Huang and Chang

(Active-Low) – This work 28.9 -

From the simulation results tabulated, it can be seen that the

modified Huang and Chang’s 8-bit priority resolver presented in this paper that corresponds to active-low logic features the least power among its counterparts, accounting for just 16% of the total power dissipation of Huang and Chang’s original 8-bit priority resolver and 34% of the power dissipated by active-high logic based modified Huang and Chang’s 8-bit priority resolver. Overall, the modified active-low Huang and Chang’s priority resolver circuit enables approximately 4× average reduction in power dissipation compared to its active-high version and the original Huang and Chang’s priority resolver macro.

Although two power optimized active-high and active-low 8-bit priority resolver designs were presented in this work, it is to be noted that these cannot be used as such for realizing higher order priority resolvers, neither can the original Huang and Chang’s design, by utilizing the parallel priority lookahead architecture of Kun et al. [13]. This is due to the reason that the primary outputs of these 8-bit macros are initially driven to the ‘set’ state, contrary to ‘resetting’ in case of Kun et al.’s and Mohanraj et al.’s [14] 8-bit priority encoder designs. Hence, proposition of a new cascading structure which would suit the original Huang and Chang’s or modified Huang and Chang’s 8-bit priority resolver designs as discussed in this paper is deemed necessary, and this suggests scope for further work.

REFERENCES

[1] E.D. Adamides, P. Lliades, I. Argyrakis, P. Tsalides, A. Thanailakis, “Cellular logic bus arbitration,” IEE Proc. Computers and Digital Techniques, vol. 140, no. 6, November 1993, pp. 289-296.

[2] S. Murugesan, “Use priority encoders for fast data comparison,” Electronic Engineering, vol. 42, pp. 24, July 1989.

[3] H.-M. Lam, C.-Y. Tsui, “A MUX-based high-performance single-cycle CMOS comparator,” IEEE Trans. on Circuits and Systems, Part II – Express Briefs, vol. 54, no. 7, July 2007, pp. 591-595.

[4] J.L. Hennessy, D.A. Patterson, Computer Architecture – A Quantitative Approach, 3rd edition, Morgan Kaufmann Publishers, NY, 2002.

[5] R. Hashemian, “Highly parallel increment/decrement using CMOS technology,” Proc. 33rd IEEE International Midwest Symposium on Circuits and Systems, vol. 2, 1991, pp. 866-869.

[6] C.-H. Huang, J.-S. Wang, Y.-C. Huang, “Design of high-performance CMOS priority encoders and incrementer/decrementers using multilevel lookahead and multilevel folding techniques,” IEEE Jour. of Solid-State Circuits, vol. 37, no. 1, January 2002, pp. 63-76.

[7] J.G. Delgado-Frias, J. Nyathi, D.H. Summerville, “A programmable dynamic interconnection router with hidden refresh,” IEEE Trans. on Circuits and Systems, Part I, vol. 45, November 1998, pp. 1182-1190.

[8] D.H. Summerville, J.G. Delgado-Frias, S. Vassiliadis, “A flexible bit-pattern associative router for interconnection networks,” IEEE Trans. on Parallel and Distributed Systems, vol. 7, May 1996, pp. 477-485.

[9] H. Kadota, J. Miyake, Y. Nishimichi, H. Kudoh, K. Kagawa, “An 8-kbit content-addressable and reentrant memory,” IEEE Jour. of Solid-State Circuits, vol. SC-20, 1985, pp. 951-957.

[10] N. Mohan, W. Fung, M. Sachdev, “Low-power priority encoder and multiple match detection circuit for ternary content addressable memory,” Proc. IEEE International SOC Conference, 2006, pp. 253-256.

[11] J.G. Delgado-Frias, J. Nyathi, “A VLSI high-performance encoder with priority lookahead,” Proc. 8th Great Lakes Symposium on VLSI, 1998, pp. 59-64.

[12] J.-S. Wang, C.-S. Huang, “A high-speed single-phase-clocked CMOS priority encoder,” Proc. IEEE International Symposium on Circuits and Systems, 2000, pp. V-537-V540.

[13] C. Kun, S. Quan, A.G. Mason, “A power-optimized 64-bit priority encoder utilizing parallel priority look-ahead,” Proc. IEEE International Symposium on Circuits and Systems, 2004, pp. II-753-II-756.

[14] J. Mohanraj, P. Balasubramanian, K. Prasad, “Power, delay and area optimized 8-bit CMOS priority encoder for embedded applications,” Proc. 10th International Conference on Embedded Systems and Applications, 2012, pp. 111-113, Nevada, USA.

[15] S.-W. Huang, Y.-J. Chang, “A full parallel priority encoder design used in comparator,” Proc. 53rd IEEE International Midwest Symposium on Circuits and Systems, 2010, pp. 877-880.


85

APPENDIX:

TRUTH TABLE OF ACTIVE-HIGH 8-BIT PRIORITY RESOLVER

Primary inputs Primary outputs

IP1 IP2 IP3 IP4 IP5 IP6 IP7 IP8 OP1 OP2 OP3 OP4 OP5 OP6 OP7 OP8

1 d d d d d d d 1 0 0 0 0 0 0 0

0 1 d d d d d d 0 1 0 0 0 0 0 0

0 0 1 d d d d d 0 0 1 0 0 0 0 0

0 0 0 1 d d d d 0 0 0 1 0 0 0 0

0 0 0 0 1 d d d 0 0 0 0 1 0 0 0

0 0 0 0 0 1 d d 0 0 0 0 0 1 0 0

0 0 0 0 0 0 1 d 0 0 0 0 0 0 1 0

0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1

d – Binary don’t care state

During the falling-edge of clock, pMOS transistors ‘p1’ to ‘p8’ turn-on and the primary outputs are driven ‘high’.

During the rising-edge of clock, pMOS transistors ‘p1’ to ‘p8’ are turned-off. When the look-ahead input (LAin) of the priority resolver is ‘1’,

based on ‘high’ input requests and the priority assignment, an appropriate ‘high’ primary output is produced – this is shown above.

TRUTH TABLE OF ACTIVE-LOW 8-BIT PRIORITY RESOLVER

Primary inputs Primary outputs

IP1 IP2 IP3 IP4 IP5 IP6 IP7 IP8 OP1 OP2 OP3 OP4 OP5 OP6 OP7 OP8

0 d d d d d d d 1 0 0 0 0 0 0 0

1 0 d d d d d d 0 1 0 0 0 0 0 0

1 1 0 d d d d d 0 0 1 0 0 0 0 0

1 1 1 0 d d d d 0 0 0 1 0 0 0 0

1 1 1 1 0 d d d 0 0 0 0 1 0 0 0

1 1 1 1 1 0 d d 0 0 0 0 0 1 0 0

1 1 1 1 1 1 0 d 0 0 0 0 0 0 1 0

1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 1

d – Binary don’t care state

During the falling-edge of clock, pMOS transistors ‘p1’ to ‘p8’ turn-on and the primary outputs are driven ‘high’.

During the rising-edge of clock, pMOS transistors ‘p1’ to ‘p8’ are turned-off. When the look-ahead input (LAin) of the priority resolver is ‘1’,

based on ‘low’ input requests and the priority assignment, an appropriate ‘high’ primary output is produced – this is shown in the Table.


86

Study of the Cell Towers Radiation Levels in Residential Areas

Sabah Hawar Saeid

College of Engineering-University of Kirkuk-IRAQ

Abstract: The rapid development of cellular communication systems all over the world has caused the appearance of many hundreds of mobile telephone base stations in every city. Installation of base station antennas has produced concerns about health and in some cases has resulted in litigation in court. Independent researches and measurements on electromagnetic fields in areas close to base stations was discussed in many countries, as well as a comparison of the level of exposure of local populations and current exposure limits. Continuous exposure to microwave radiation from cell phone towers cause serious health problems over the years. Two of the most important factors in these measurements are the distance and the direct line of sight to the antenna site. In this study, measurements have been carried out at various places near the cell towers inside residential areas in Kirkuk-Iraq. It has been found that the radiation levels were above the recommended values. The results of this study show that the amount of power density is more than ten times greater than the recommended safety power density. Keywords- Cell towers radiation levels, GSM base stations, Safety power density. I. INTRODUCTION Worldwide, the use of mobile telephony has increased considerably with the introduction of the digital GSM 900 systems in the 1990s [1-3]. With increase in cell phone communication, number of cell towers getting installed is increasing every day. This increased use of mobile phones has led to an important deployment of base stations. The number of base stations in any country depends on several factors as the number of network providers, the number of users and the topography [4]. In Kirkuk, currently there are hundreds of cell phone towers, and to meet the communication demand, the number will increase rapidly. Such base stations are often situated close to dwellings or houses and have become the reason for concerns of parts of the population in the recent years. Some of the base stations are planted right in a home of residence. The concerned population often wants to know the level of exposure due to the base stations, if these levels of exposure might be health relevant and if the levels comply with national and international

standards, guidelines and regulations. To answer these questions, local and national authorities network providers and private persons often contract qualified institutions to evaluate the exposure level in restricted areas. The cell tower transmits in the frequency range of 869 - 894 MHz (CDMA), 935 - 960 MHz (GSM900) and 1805 - 1880 MHz (GSM1800) [5]. A base station and its transmitting power are designed in such a way that mobile phone should be able to transmit and receive enough signal for proper communication up to a few kilometers. These cell towers transmit radiation 24x7, so people living within 10’s of meters from the tower will receive 10,000 times stronger signal than required for mobile communication [6]. Majority of these towers are mounted near the residential and office buildings to provide good mobile phone coverage to the users. In cities, millions of people reside within these high radiation zones. The cellular base stations are transmitting continuously even when nobody is using the phone. We know from a variety of scientific studies, that significant biological effects result from the non-thermal effects of extremely periodic -pulsed - HF-radiation as are utilized in the most common modern digital cellular and cordless phone systems round the world. Not all standards and guidelines throughout the world have recommended the same limits for exposure. For example, some published exposure limits in Russia and some eastern European countries have been generally more restrictive than existing or proposed recommendations for exposure developed in North America and other parts of Europe. Very limited information is available on the exposure to cellular base station radiation in residential areas at different distances and directions to antenna sites [7, 8]. II. RADIATION FROM THE CELL TOWER A GSM900 base station antenna transmits in the frequency range of 935 - 960 MHz. This frequency band of 25 MHz is divided into twenty sub-bands of 1.2 MHz, which are allocated to various operators [9]. There may be several carrier frequencies (1 to 5) allotted to one operator with upper limit of 6.2 MHz bandwidth. Each carrier frequency may transmit 10 to 20W of power. So, one operator may transmit 50 to 100W of power and there may be 3-4 operators on the same roof top or tower, thereby total transmitted


87

power may be 200 to 400W [10]. In addition, directional antennas are used, which typically may have a gain of around 17 dB (numeric value is 50), so effectively, several KW of power may be transmitted in the main beam direction. Radiated power density can be calculated for N number of base stations at distances Rn using [11]:

(1)

where, Ptn = Transmitter power in Watts from nth station. Gtn = Gain of transmitting antenna of nth station. Rn = Distance from the antenna nth station in meters. The simplest case is the one when a human is exposed to a single base station antenna (N=1), as shown in the Fig. (1), [12].

Fig. (1) Human exposed to a single base station

antenna. Following the above computational steps, the power density for the frequency 945 MHz, using equations 1 were computed. Table (1) gives the power density for Pt = 20 W, Gt = 17 dB (Gt = 50.12), for various distances from the transmitting tower. The power density values given in Table-1 are for a single carrier and a single operator, (N=1). If multiple carriers are being used and multiple operators are present on the same roof top or tower, then the values will increase many times. However, radiation density will be much lower in the direction away from the main beam. One should know actual radiation pattern of the antenna (which unfortunately is not made public) to calculate exact radiation density at a point.

Table (1): Power density and received power at various distances from the transmitting tower

R in m Pd in mW/m2 Pd in dBm 1 79766.43 49.018 2 19941.60 42.998 3 8862.94 39.476

4 4985.40 36.977 5 3190.66 35.039 6 2215.73 33.455 7 1627.89 32.116 8 1246.35 30.956 9 984.77 29.933

10 797.66 29.018 20 199.42 22.998 30 88.63 19.476 40 49.85 16.977 50 31.91 15.039

100 7.977 9.018 200 1.994 2.997 300 0.886 -0.526 400 0.499 -3.019 500 0.319 -4.962

Fig. (2): shows the graph of power density versus distance for a typical base station.

Fig. (2): power density versus distance. III. MEASUREMENTS PROCEDURES: The base station antennas in the selected sites were the more common panel antennas, which divide the area around the base station into three sectors. With this arrangement of the three antennas the entire region around the base stations were covered. The signals radiated are for digital mobile telephone systems that operate with GSM frequency band of 900 MHz. All measurements were made with a 3 Axis Radio Frequency Electromagnetic Field Tester (Model: EMF-839) [13]. This equipment is specially developed for measuring or monitoring electromagnetic field, for example: cell-phone stations. It is used for broadband devices of monitoring the wide range radio frequency electromagnetic field value, which allows each received radio signal, in the range of 100 KHz to 3 GHz. IV. RADIATION MEASUREMENTS

0 50 100 150 200 250 300 350 400 450 500-10

0

10

20

30

40

50

Distance in meters

Pow

er d

encs

ity in

dB

m


88

For the purpose of this research, one major service provider in Kirkuk is selected, that provided GSM coverage to all the regions under consideration. The major concern was the radiation emitted by base station antennas. Considering stations that are sited within 50m from residential buildings, and densely populated areas. Measurements of the base station signals conducted from 9:00 AM to 2:00 PM local time in Kirkuk. The power radiated by base stations are highly dependent on the number of subscribers making calls at the same time. Thus, the measured radiated power depends on time, place, direction, distance of measurement and season. For hot countries or hot seasons in countries, the acceptable maximum radiated power density should be much lower. The number of measurements taken in this research work were 147 measures, at distances about 50 meters around the base stations in Kirkuk. The average power density was 71.226 mW/m2, which is more than twice of theoretically calculated power density, (as shown in table (1)), using equation (1). Fig. (3): shows the comparison between measured power density in different countries [14- 21] with safety code 6 given by equation (2) [22].

(2) where f is the frequency (MHz), in the range of (300-1500 MHz). So the calculated maximum safety power density is equal to (6.3 mW/m2).

Fig. (3) Comparison between measured power

densities in different countries.

V. CONCLUSION In this research work, measurements have been carried out at various places at distances about 50m away from the cell towers inside residential areas in Kirkuk-Iraq. 147 measures have been taken at different directions. The results of this study show that the average power density in more than 90% of the measures was 71.226 mW/m2. This amount of power density is more than ten times greater than the recommended safety power density which is equal to or less than 6.3 mW/m2. Comparison between the power density in many countries shows that the minimum measured power density was in France then Germany and India, and the worst is in Iraq. Therefore, many comprehensive studies are necessary to be done in this country, to protect peoples from the risk of the exposure to this high power density of the radiation of cells phone tower especially in residential areas. REFERENCES [1] Mann, S. M.; Cooper, T. G.; Allen, S. G.;

Blackwell, R. P.; Lowe, A. J.; "Exposure to Radio Wave Near Mobile Phone Base Stations, NRPB-R321, 2000.

[2] Neubauer, G.; "Exposure next to Base Stations in Austria", BEMS Proceeding Book, Munich, 2000.

[3] Pinho, P.; Casaleiro, J.; "Influence of the Human Head in the Radiation of a Mobile Antenna", PIERS Proceedings, Moscow, Russia, August 18-21, pp.666-669, 2009.

[4] Shalangwa, D. A.; "Measurement of Exposure of Radio Frequency Field Radiation from GSM Masts", Journal of Electrical and Electronics Engineering Research, Vol. 2, No. 3, pp. 75-84, May 2010.

[5] Ayeni, A. A.; Braimoh, K. T.; Ayeni, O. B.; "Effect of GSM Phone Radiation on Human Pulse Rate", Journal of Emerging Trends in Computing and Information Sciences", Vol. 2, No. 11 pp.

[6] Kumar, G.; "Cell Tower Radiation", Electrical Engineering Department, IIT Bombay, December, 2010.

[7] Saeid, H. S.; Najat, M. M.; "Study of Health Hazards due to Radiation of Electromagnetic Power from Mobile Base Stations Using MATLAB", American Journal of Scientific Research, Issue 41, December, pp. 16-25, 2011.

[8] Saeid, H. S.; "Calculation of the Mobile Phone Radiation Level in Respect to the Exposure Standards", Journal of Science & Technology, University of Science and Technology, Yemen, Vol. 13, No. 1, 2008.

[9] Parsons, J. D.; "The Mobile Propagation Channel", Second Edition, John Wiley & Sons Ltd, 2000.


89

[10] Kaushal, M.; Singh, T.; Kumar, A.; "Effects of Mobile Tower Radiation & Case Studies from Different Countries Pertaining the Issue", International Journal of Applied Engineering Research, Vo. 7, No. 11, 2012.

[11] Saeid, H. S.; "Human Exposure Assessment in the Near-Field of Antennas Used by Mobile Research", Journal of Asian Scientific Research, Vol. 2, No. 4, 2012

[12] Saeid, H. S.; "Theoretical Estimation of Power Density Levels around Mobile Telephone Base Stations", Journal of Science & Technology, University of Science and Technology, Yemen, Vol. 13, No. 2, 2008.

[13] Lutron Electronic; 3Axis Radio Frequency Electromagnetic Field Meter" Model: EMF-839.

[14] Haumann, T.; Munzenberg, U.; Maes, W.; Sierck, P.; "HF-Radiation Level in Residential Areas", Report on HF-Radiation of GSM Cellular Phone Towers, pp.327-333.

[15] Arnelli, C.; Roggia, G.; Trinchero, D.; "Low Cost Measuring Methods Applied to an Electromagnetic Site Survey of a Complex Environment", Department of Electronics, Torino Polytechnics, Italy.

[16] Nedhif, S.; "Health Safety and Field Strength Exposure in ICS Telecom", White Paper, May 2008.

[17] Pllana, M. I.; Ahma, L.; Hamiti, E.; "Human Exposure Assessment in the Vicinity of 900 MHz GSM Base Station Antenna", International Journal of Communications", Vol. 1, Issue 2, 2007.

[18] Abdlati, M.; "Electromagnetic Radiation from Mobile Phone Base Stations at Gaza", Journal of the Islamic University of Gaza (Natural Sciences Series), Vol. 13, No. 2, pp. 129-146, 2005.

[19] Hoong, K. N.; "Non-Ionizing Radiations-Sources, Biological Effects, Emissions and Exposures", Proceedings of the International Conference on Non-Ionizing Radiation at UNITEN, 20th – 22nd October, 2003.

[20] Kumar, N.; Kumar, G.; "Biological Effects of Cell Tower Radiation on Human Body", ISMOT, New Delhi, India, December 16-19, 2009.

[21] Neubauer, G.; Giczi, W.; Schmid, G.; "Analysis of Exposure Levels Next to GSM Base Stations", ARC Seibersdorf Research GmbH, Austria.

[22] Thansandote, A.; Lecuyer, D. W., Gajda, G. B.; McNamee, J. P.; "Limits of Human Exposure to Radio Frequency Electromagnetic Fields in the Frequency Range from 3kHz to 300 GHz", Safety Code 6, Environmental Health Directorate, Health Protection Branch, Minister of Health, Canada.


90

A Closed-form Delay Estimation Model for

Current-mode High Speed VLSI Interconnects

M.Kavicharan

Dept. of ECE


Warangal, India

[email protected]

N.S.Murthy

Dept. of ECE


Warangal, India

[email protected]

N. Bheema Rao

Dept. of ECE


Warangal, India

[email protected]

Abstract—Closed-form model for the delay estimation of

current-mode Resistance Inductance Capacitance (RLC)

interconnects in VLSI circuits is presented. The existing Eudes

model for interconnect transfer function approximation is

extended and applied for further accurate estimation of delay.

With the equivalent lossy interconnect transfer function, finite

ramp responses are obtained and line delay is estimated for

various line lengths, per unit length inductances and load

capacitances. The estimated delay values of extended Eudes

model are compared with the existing Eudes model against

HSPICE W-element model. The obtained delay values of Eudes

model worst-case error percentage is 14.3% whereas our

extended Eudes model is in good agreement with those of

HSPICE results within 2% for the line lengths of 1mm to 10mm.

Keywords— Delay, ramp input, Eudes model, RLC

interconnects, MacLaurin series, transient analysis, transfer

function, SOC.

I. INTRODUCTION

Advancement in technology leads to chip size

reduction and increased on-chip interconnection complexity.

The performance improvement by using traditional modeling

methods [1]-[4] in future technologies is not beneficial. To

reach the speed requirement of VLSI circuits at future

technology nodes demands for alternative signaling technique

that provides an attractive solution to some of the disputes

caused by aggressive interconnect scaling. As per the paper

[5], the conventional voltage-mode signaling is not able to

meet the speed requirements of future technology generations.

So, current-mode signaling has been investigated as a

substitute for high speed data transmission over interconnects.

Signaling in global lines is a major bottleneck in high

performance VLSI systems due to the dominant limitation of

signal propagation delays as compared to circuit delays. So,

accurate and efficient delay estimation models are required in

current-mode signaling. Accurate estimation of propagation

delay in global interconnects plays a predominant role in the

early design stages of VLSI systems as compared to local

interconnect delays because, global wires supports main

functions like clock, signal distribution between the functional

blocks and provides power/ground to all functions on a chip.

Starting from lumped RC model to distributed RLC

model, various techniques [6]-[8] based on analytical closed-

form formulations have been proposed to model delay in

voltage-mode interconnects. Similarly for current-mode RC

interconnects, closed-form delay analysis model was presented

in [9] and the analysis does not include the fast edge input

response. In [10] closed-form delay model for distributed

current-mode RC line is presented, which included the

practical fast edge input response issue but this model could

not include the inductance effect in current-mode interconnect.

A delay estimation model [11], which is derived using the

concept of absorbing inductance effect into equivalent RC

model, then modified nodal analysis (MNA) was used.

Various closed-form delay models [9]-[13] for on-

chip interconnects in current-mode signaling have more

inaccuracy in terms of delay estimation. This paper presents a

closed-form delay estimation model for current-mode on-chip

RLC interconnects by using the existing Eudes model [14],

which was basically for modeling of PCB RLCG

interconnects. But, in this paper we have implemented Eudes

model and extended Eudes model for current-mode RLC

interconnects and are compared against standard HSPICE tool.

At deep sub micron VLSI designs Conductance effect can be

ignored so, we implemented Eudes model and its extension for

RLC interconnects. These mathematical models are

implemented in MATLAB for easy of analysis. These models

are validated under different ranges of parameters and

compared with respect to HSPICE tool.

According to signaling point of view, both voltage and

current-mode driver circuits can be approximated by a voltage

source and a linear resistance. As compared to voltage-mode

receiver circuits, current-mode receiver circuits provide a low-

impedance path, so a resistive and parallel low capacitive path

is needed at the receiver. Therefore, for sensing a current-

mode signal, low input impedance receiver termination is

required, which provides current path to ground.


91

The remainder of the paper is organized as follows. Section II briefly describes the Eudes model and extension of Eudes model for the analysis of current-mode RLC interconnects. Section III reports the validation of the presented models and the results are compared with standard HSPICE. Conclusions and future scope appear at the end.

II. ANALYSIS OF CURRENT-MODE RLC INTERCONNECT

USING EUDES MODEL AND ITS EXTENSION

The analysis of on-chip current-mode RLC interconnects

begins with Telegraphers equation in frequency domain. All

the closed-form RLC interconnects models assume quasi-

TEM mode of signal propagation. The Telegrapher's equations

are a pair of linear partial differential equations which

illustrate the voltage and current on a transmission line with

distance and time as transmission line variables.

The solution of interconnects are described by telegrapher’s

equations as

szIsLRszVx

,,

szsCVszIx

,,

(1)

where ‘s’ is the Laplace-transform variable, z is a

variable which represents position; szV , and szI , stand

for the voltage and current vectors of the transmission line,

respectively, in the frequency domain; and R, L and C are the

per unit length (p.u.l.) resistance, inductance, and capacitance

matrices, respectively.

The solution of (1) can be written as an exponential

matrix function as

sI

sVe

sdI

sdVd

,0

,0

,

,

(2)

where

0

0

Y

Z

In (2) ‘d’ is the length of the transmission line, with Z=R+sL

and Y=sC. The exponential matrix of (2) can be written in

terms of cosh and sinh functions as

YZdYZdY

YZdYZYdd

e coshsinh

sinhcosh

0

1

0 (3)

where 1

0 )( YZYY

A. Eudes model [14]

This model was developed for the characterization of

RF/digital PCB RLCG interconnections for the prediction of

the high-speed signal transient responses. This model

considers the second order MacLaurin series approximation of

the interconnection RLCG-model transfer matrix and the

transient responses obtained from the resulting interconnection

system transfer function. We applied this model to current-

mode RLC interconnects and obtained the delay estimation.

The direct time domain representation of (2) is not

possible, so it is difficult to analytically estimate the delay of

interconnects. This equation can be realized by applying the

second order MacLaurin series for each element of the transfer

matrix. The resultant is derived as

H(s) =

2

.1

2

.1

YZY

ZYZ

(4)

A single current-mode RLC line is shown in Fig. 1. The

line is driven by a step input and 1-V finite ramp with rise

time of 0.1 ns. Current-mode interconnect driver is modeled

similar to voltage-mode driver equivalent series resistance Rs

but the main difference is with current-mode interconnect

receiver, which is modeled as a resistor RL in parallel with

capacitive load CL for low impedance path which enhances

high speed data transmission over longer lines

Fig. 1.Equivalent circuit model of the current-mode single line RLC

interconnect.

The transfer function matrix in (4) denotes the

approximation model of Transmission line without source and

load impedances. As per Fig. 1, the source and load

impedances matrices need to be cascaded with second order

interconnect transfer function to find the complete circuit

transfer function

H1(s) =

10

1 sR*

2

.1

2

.1

YZY

ZYZ

*

11

01

lZ

(5)

where

1. LL

L

lsCR

RZ

The Zl is an equivalent load impedance of line and is

represented as parallel combination of load resistance RL and

CL.


92

The frequency domain transfer function at the far end is

H111(s)=

211

21

1

ZYRZ

ZYR

ZYs

ls

(6)

Based on the ABCD parameter model matrix and their

expansion leads to the voltage transfer function (6) in

frequency domain which can be inverse Laplace transformed

to get the time domain response for the estimation of delay.

The equation (6) is the result of Eudes model which

approximates current-mode RLC interconnect transfer

function model using second order MacLaurin series. But this

model has less accuracy so, we extended his work with fourth

order MacLaurin series for better accuracy and mathematical

analysis is implemented in MATLAB.

B. Extended Eudes model

We have extended the existing Eudes model and obtained

an efficient model for current-mode RLC interconnects to get

the accurate delay estimation. This model is compared with

existing Eudes model and HSPICE and found that our model

is having good accuracy at cost of slight increase in

computation time

The complicated form (2) is realized by applying

Extended Eudes model and is given as

H(s) =

2421

6

62421

23

32

ZYZYYY

ZZ

ZYZY

(7)

The complete transfer function matrix obtained using

cascade of source, above interconnect matrix (7) and load

impedances matrices is written as H2(s)

H2(s)=

10

1 sR

2421

6

62421

23

32

ZYZYYY

ZZ

ZYZY

11

01

lZ

(8)

After multiplication of above matrices leads to

frequency domain voltage transfer function H112(s) as first

term, written as

2421

6

1

6421

1

2332ZYZY

RZ

ZZ

YYR

ZYZYs

l

s

(9)

These transfer function models H111(s) and H112(s) of

Eudes and extended Eudes model are inverse Laplace

transformed using MATLAB to get time domain responses

and estimation of delay.

III. SIMULATION RESULTS

The single current-mode RLC line is presented in this

section to demonstrate the validity and efficiency of the

proposed method. The results were obtained using MATLAB

R2010a operating on HP 64-bit Intel i5 processor with clock

speed of 2.53 GHz and are also compared with HSPICE using

the W-element method.

The typical interconnect parameters [11] considered for

simulation of single current-mode RLC interconnect are given

in table-I. The Eudes model and its extended model are

implemented in MATLAB for the same set of input

parameters.

Table 1. Interconnects parameter for model validation [11]

Unit-length

Inductance

Unit-length

Capacitance

Unit-length

Resistance RL CL

400nH/m-

100nH/m

50pF/m-

100pF/m

20k Ω/m-

40k Ω/m

10 Ω-

10kΩ

10fF-

600fF

We considered the rise time of the ramp input signal

is 0.1ns, so the maximum frequency necessitated for the

presented models responses have been approximated of about

3.5GHz from (10). In order to demonstrate the adaptableness

of the proposed modeling method, we varied line lengths, line

inductance and load capacitance and observed the delay

statistics.

rtf

35.0max (10)

Fig 2. 50% delay versus Line length for various models

As per the data in Fig. 2, it is apparent that as the line

length increases from 1mm to 10mm range of SOC global line

applications, the delay increases and various models (HSPICE,

Eudes and extended Eudes model) delay plots are cascaded.

As per calculations of delay, the Eudes model has maximum

error percentage of 14.3% whereas extended Eudes model


93

matches very well with HSPICE and the maximum error

percentage is 1.26% for the line length of 10mm. However, for

shorter lines (0.3cm) and lower, all the models matches to that

of HSPICE with acceptable error percentage within 1%, but

for longer lengths the extended Eudes model plays a

prominent role in estimating the accurate delay of current-

mode interconnect.

Fig 3. 50% delay versus line inductance for 0.2cm long line

In Fig. 3 there is a clear trend of increasing delay for

various per unit length Line inductances of the range 100nH/m

to 800nH/m. As observed that, both the presented models give

acceptable error percentage of within 1% for the line length of

0.2cm.

Fig 4. 50% delay versus load capacitance for 0.2cm long line

Fig. 4 depicts the 50% delay values for various

interconnect load capacitances in the range of 10fF to 800fF. It

can be observed from Fig. 4 that, the delay values of Eudes

model deviates with HSPICE but, the extended Eudes model

nicely matches with HSPICE and gives acceptable error

percentage with in 0.2% for the line length of 0.2cm.

To find the longer line response we increased the

length of the line from 0.2cm to 1cm along with variations in

line inductances and load capacitances. Figures 5 and 6 are the

plots of 50% delay values for line inductance and load

capacitance variations at longer length of 1cm.

Fig 5. 50% delay versus line inductance for 1cm long line

As shown in Fig. 5 the Eudes model has maximum

error percentage of 14.6% whereas the extended Eudes model

maximum error percentage is of 5%. It can be noticed that,

the extended Eudes model matches with HSPICE until

600nH/m with an error percentage of 2% for the line length of

1cm.

Fig 6. 50% delay versus load capacitance for 1cm long line

Further analysis in Fig. 6 shows that, for 1cm long

line with various capacitive loads, the existing Eudes model

fall short of to estimate the delay accurately with maximum

error percentage of 14.4% but, the extended Eudes model has

maximum error percentage of 2.3% and matches with

HSPICE, independent of interconnect load capacitance

variations.

To find the interaction between the line length values

and delay, we varied the values of the per unit length

inductance in the Figures 3 & 5 for line lengths of 0.2cm and

1cm. Similarly to find the effect of line length on delay for

various models, we varied the value of the capacitive load in

the Figures 4 & 6 for 0.2cm and 1cm lines. For both of these

cases transient responses were found using HSPICE, Eudes

model and its extension model to find the delay of line under

various interconnect parametric variations. The extended

Eudes model has good accuracy at the cost of slightly higher

computation time. These models produce stable results under

various interconnect parameters.


94

CONCLUSION

This paper presents Eudes model and an extension of Eudes

model based closed form models for delay estimation of current-

mode high speed VLSI interconnects. The purpose of the current

study is to estimate the delay of current-mode VLSI interconnects

and to find the interaction between delay for various lengths, line

inductances and load capacitances using existing Eudes model and its

extension against HSPICE tool. This research has shown that, the

existing Eudes model fall short of to give good accuracy for longer

lines (>0.3cm) but, our extended model is better at longer lines. A

single line interconnect has been used for validating the presented

models (Eudes model and extended Eudes model) by comparing with

the HSPICE. In SOC (system on chip) applications, for global lines

of lengths 0.3 cm and below both methods matches nicely with

HSPICE but, above 0.3cm length the extended Eudes model is found

to be more accurate than existing method. The present study confirms

previous findings and contributes additional extension that suggests,

for longer lines existing Eudes model underestimates the delay with

large error percentage but, extended Eudes model is having better

accuracy. These methods can be used to estimate the signal integrity

characteristics of Carbon nano tubes.

REFERENCES

[1] H. B. Bakoglu, Circuits, Interconnections, and Packaging for VLSI,

Addison-Wesley Company, Reading , MA 1990.

[2] Y.I. Ismail, E.G. Friedman, Optimum repeater insertion based on a CMOS delay model for on-chip RLC interconnect, Proc. IEEE ASIC(1998), pp. 369–373.

[3] Adler and E. G. Friedman, “Repeater design to reduce delay and power in resistive interconnect,” IEEE Trans. Circuits Syst. I, vol. 45,pp. 607–616, May 1998.

[4] K. Banerjee, A. Mehrotra, A power-optimal repeater insertion methodology for global interconnects in nanometer designs, IEEE Trans. Electron. Devices 49 (11) (2002).

[5] A. Maheshwari and W. Burleson, "Differential current-sensing for on-chip interconnects," IEEE Tran. on Very Large Scale Integration Systems, vol. 12, no. 12, pp. 1321-1329, December 2004.

[6] W. C. Elmore, “The transient response of damped linear networks with particular regard to wideband amplifiers,”J. Appl. Phys., vol. 19, no.1, pp. 55–63, Jan. 1948.

[7] T.Sakurai, “Closed-form expressions for interconnection delay, coupling, and crosstalk in VLSI’s,”IEEE Trans. Electron Devices, vol. 40, no. 1, pp. 118–124, Jan. 1993.

[8] Kahng A B, Muddu S, “An analytical delay for RLC interconnects”, IEEE Trans. Computer-Aided Design of Integrated Circuits and Systems, 1997, 16(12): 1507−1514.

[9] E. Seevinck, P. van Beers, H. Ontrop, “Current-Mohe Techniques for High-speed VLSI Circuits with Application to Current Sense Amplifier for CMOS SRAM’s,” IEEE J. of Solid-state Circuits, Vol 26, No. 4, April1991.

[10] Bashirullah, R.; Wentai Liu; Cavin, R., III, "Delay and power model for Current-mode signaling in deep submicron global interconnects," Custom Integrated Circuits Conference, 2002. Proceedings of the IEEE 2002, vol., no., pp.513,516, 2002.

[11] Mingcui Zhou; Wentai Liu; Sivaprakasam, M., "A closed-form delay formula for on-chip RLC interconnects in Current-mode signaling," Circuits and Systems, 2005. ISCAS 2005. IEEE International Symposium on , vol., no., pp.1082,1085 Vol. 2, 23-26 May 2005.

[12] Jadav, S.; Khanna, G.; Kumar, A.; Saini, G., "Low power high throughput Current-mode signalling technique for global VLSI interconnect," Computer and Communication Technology (ICCCT), 2010 International Conference on , vol., no., pp.290,295, 17-19 Sept. 2010.

[13] R. Bashirullah, W. Liu, and R. Cavin, “Current-mode signaling in deep submicrometer global interconnects,”IEEE Trans. Very Large Scale Integr. (VLSI) Syst., vol. 11, no. 3, pp. 406–417, Jun. 2003.

[14] T.Eudes, B. Ravelo, and A. Louis,“Transient response characterization of the high-speed interconnection RLCG-model for the signal integrity analysis,” PIER Journal, Vol. 112, pp. 183–197, 2011.


95

AC-DC & DC-DC Converters for DC Motor Drives Review of basic topologies

G.Ch.Ioannidis*, C.S.Psomopoulos

*, S.D.Kaminaris

*,

P.Pachos*, H.Villiotis

*, S.Tsiolis

*, P.Malatestas

*,

G.A.Vokas**

*Department of Electrical Engineering

**Department of Electronic Engineering

Technological Education Institute of Piraeus

Piraeus, Greece

S.N. Manias

School of Electrical & Computer Engineering

National Technical University of Athens

Zografou, Greece

Abstract—This paper deals with a comprehensive survey on

the topic of AC/DC & DC/DC converters for DC Motor Drives. A

substantial number of different AC/DC and DC/DC topologies

appropriate for DC motor drives are presented. This critical

literature review brings out merits, demerits, and limitations

besides giving the basic operating principles of various

topologies.

Keywords—controlled-rectifiers, choppers; dc motor drives;

hard switching; soft-switching;

I. INTRODUCTION

DC motors have been available for nearly 100 years. In fact the first electric motors were designed and built for operation using direct current power. Although AC motors are mainly used in industry for high speed operation (over 2500 rpm) because they are smaller, lighter, less expensive, require virtually no maintenance comparing to their DC counterparts, the latter are still used. The reasons for this are that they exhibit wide speed range, good speed regulation, starting and accelerating torques in excess of 400% of rated, less complex control and usually less expensive drive. Today, DC motors are still used in several applications as in industrial production and processing of paper pulp, textile industries, in electric vehicle (EV) propulsion and in public transport such as TRAM (trolley) and METRO. The control of these motors is usually made of power electronics devices, such as controlled rectifier-fed (thyristor-fed) DC drives or chopper-fed DC drives and because of their simplicity, ease of application, reliability and favorable cost have been a backbone of industrial applications.

Fig. 1. Quadrants of operation of a DC motor

DC motor drives can be categorized according to the way they manage the energy generated during braking of the DC

motor ([1]-[4]). In this perspective, there are non-regenerative and regenerative DC drives in industry. Non-regenerative DC drives are the most conventional type in common usage. They are able to control motor speed and torque in one direction (first-quadrant, Fig.1). With the addition of an electro-mechanical (magnetic) armature reversing contactor or manual switch (units rated 2 HP or less) the controller output polarity is reversed and the same is true for the direction of rotation of the motor armature (third-quadrant, Fig.1). In both cases torque and rotational direction are the same. Regenerative DC drives are also known as four-quadrant drives and they are capable of controlling not only the speed and direction of motor rotation, but also the direction of motor torque. The term regenerative describes the ability of the drive under braking conditions to convert the mechanical energy of the motor and connected load into electrical energy which is returned (or regenerated) to the AC power source. When the drive is operating in the first and third quadrants, both motor rotation and torque are in the same direction and it functions as a conventional non-regenerative unit. The unique characteristics of a regenerative drive are apparent only in the second and fourth quadrants. In these quadrants (Fig.1), the motor torque opposes the direction of motor rotation which provides a controlled braking or retarding force. A high performance regenerative drive is able to switch rapidly from motoring to braking modes while simultaneously controlling the direction of motor rotation. A regenerative DC drive is essentially two coordinated DC drives integrated within a common package. One drive operates in the first and fourth quadrants and the other operates in the second and third quadrants.

Another way to classify DC motor drives is according to the type of the converter which is utilized in order to control the speed and the torque of the DC motor ([2]-[4]). When a controlled rectifier circuit (one or three phase) is used the respective category is called: Controlled Rectifier-Fed (Thyristor-Fed) DC Motor Drive. In case that a DC to DC converter is used the respective category is called: Chopper-Fed DC Motor Drive. Both of these categories can further subdivided into non-generative and generative drives based on what was mentioned earlier.

In the literature many different converters have been presented and analyzed whose operation is based on controlled rectifier circuits, single-phase or three-phase thus, AC-DC


96

converters ([2]-[9]). These converters can operate at one, two or even four quadrants and are used in various applications depending on the particular requirements. The manufacture of efficient semiconductor switches led to the development of converters which receive as input DC voltage, operating at relatively high frequency (several tens kHz), exhibiting high response speed and are used to control DC motors ([2]-[5], [11]-[26]). In order to increase further the switching frequency of these DC-DC converters and diminish the ripple current, especially in case of low-inductance DC motors, but also to reduce more the size, weight and volume of the overall drive, soft-switching DC-DC converters have been developed ([11]-[26]). These converters initially have been proposed for switched-mode power supplies ([11]-[15]), but they cannot be directly applied to dc motors, especially for electric vehicle propulsion. Apart from suffering excessive voltage and current stresses ([13], [15]), they cannot handle backward power flow during regenerative braking ([16]). In recent years soft-switching DC-DC converters suitable for DC motor drives have been presented in the literature ([17]-[26]).

This paper deals with a comprehensive survey on the topic of AC-DC & DC-DC converters for DC Motor Drives. A lot of publications are reviewed and classified into two major categories. Some of them are further classified into several subcategories.

II. CONTROLLED RECTIFIER-FED DC DRIVES

The thyristor DC drive remains an important speed-controlled industrial drive, especially where the higher maintenance cost associated with the DC motor brushes is tolerable. The controlled (thyristor) rectifier provides a low-impedance adjustable DC voltage for the motor armature, thereby providing speed control. For motors up to a few kilowatts the armature converter can be supplied from either single-phase or three-phase mains, but for larger motors three-phase is always used. A separate thyristor or diode rectifier is used to supply the field of the motor: the power is much less than the armature power, so the supply is often single-phase, as shown in Fig.2.

Fig. 2. General closed-loop controlled rectifier -fed DC motor drive

The main power circuit usually consists of a one or two or

four or six-thyristor circuit, which rectifies the incoming AC supply to produce a DC supply to the motor armature. The assembly of thyristors, mounted on a heatsink, is usually referred to as the 'stack'. By altering the firing angle of the thyristor/s the mean value of the rectified voltage can be varied,

thereby allowing the motor speed to be controlled. The controlled rectifier produces a DC bus with a pronounced ripple in the output voltage. This ripple component gives rise to pulsating currents and fluxes in the motor, and in order to avoid excessive eddy-current losses and commutation problems, the poles and frame should be of laminated construction. It is accepted practice for motors supplied for use with thyristor drives to have laminated construction, but older motors often have solid poles and/or frames, and these will not always work satisfactorily with a rectifier supply. It is also the norm for drive motors to be supplied with an attached “blower” motor as standard. This provides continuous through ventilation and allows the motor to operate continuously at full torque even down to the lowest speeds without overheating.

sins m tν ν ω=

s Ti i=

oV

iα

Rα

aL

fVbe ,f fr L

FD

fdi

T

Fig. 3. Single phase half wave converter drive

Low power control circuits are used to monitor the principal variables of interest (usually motor current and speed), and to generate appropriate firing pulses so that the motor maintains constant speed despite variations in the load. The speed reference (Fig.2) is typically an analogue voltage varying from 0 to 10 V, and obtained from a manual speed-setting potentiometer or from elsewhere in the plant. The combination of power, control, and protective circuits constitutes the converter. Standard modular converters are available as off-the-shelf items in sizes from 0.5 kW up to several hundred kW, while larger drives will be tailored to individual requirements. Individual converters may be mounted in enclosures with isolators, fuses etc., or groups of converters may be mounted together to form a multi-motor drive.

sins m tν ν ω=

11T

1Ti

12T

11D

12D

oV

iα

Rα

aL

fVbe ,f fr L

Fig. 4. Single phase half-controlled asymetrical converter drive

A separately excited DC motor fed through single phase half wave converter ([2]-[6]) is shown in Fig. 3. Single phase half wave converter feeding a DC motor offers only one quadrant drive. Such type of drives are used up to about 0.5 kW DC motor. For this converter the average output voltage can be calculated versus the firing angle α as:


97

( )1 cos , for 02

mo

VV a α π

π= + < < (1)

where mV is the maximum value of the applied line voltage

A separately excited DC motor fed through a single-phase asymmetrical semiconverter is shown in Fig. 4. The armature voltage cannot be at any instant negative because the diodes cannot have a positive potential difference in their terminals. This means that this converter topology cannot regenerate. Its operation is therefore confined to the first quadrant of the va-ia (or torque-speed) diagram (motoring mode) and is used up to 15 kW DC drives. The diodes which offer the freewheeling path for the main power-circuit (armature current) should be ultra-high speed diodes, in order to protect the circuit from undesirable overvoltages. For one-quadrant operation, half-controlled converter exhibits better characteristics than the fully controlled one, such as, less harmonics distortion to the input current, increased mean value of the output voltage for the same firing angle, power factor improvement and cheaper control circuits. For this converter type, considering continuous-mode operation the average output voltage can be calculated as ([2]-[6]):

( ) παπ

<<+= 0for ,cos1 aV

V mo

(2)

sins m tν ν ω=

11T

11Ti

13T

14T 12T

oV

iα

Rα

aL

fVbe ,f fr L

Fig. 5. Single phase fully controlled converter drive

sins m tν ν ω=

11T

11Ti

13T

14T 12T

oV

iα

Rα

aL

be,f fr L

22T

23T21T

24T

sins m tν ν ω=

fV

Fig. 6. Single phase dual converter drive

In Fig.5 the armature voltage is varied by single phase full wave converter. It is a two quadrant drive and is limited to applications up to 15kW. The armature voltage can be varied between π/2 mV+ and π/2 mV− and this possibility allows

operation in the first and fourth quadrant of the ao iv − diagram

or, in other words, in the first and second quadrant of the

torque-speed diagram. The reversal of the armature or field voltage allows operation in the second and third quadrant. For continuous-mode operation the output voltage is:

παπ

<<= 0for ,cos2

aV

V mo

(3)

In Fig. 6, there are two single phase full wave converters (in back-to-back connection), either converter 1 operates to supply a positive armature voltage Vo, or converter 2 operates to supply negative armature voltage –Vo. Converter 1 provides operation in first and fourth quadrants, and converter 2 provides operation in second and third quadrants. It is four-quadrant drive and provides four modes of operation: forward motoring, forward braking (regeneration), reverse motoring, and reverse breaking (regeneration).

If converter 1 operates at a firing angle of α1 then the armature voltage is:

( ) παπ

<<+= 0for , cos1 1aV

V mo

(4)

And similarly, if converter 2 operates at a firing angle of α2 then the armature voltage is

( )2cos1 aV

V mo +=

π (5)

11T

13T

14T

12T

oV

iα

Rα

aL

fVbe ,f f

r L

15T

16T

ai

bi

ci

Fig. 7. Three-phase fully controlled converter

It is should be noted that in this case, inverting operation occurs by reversing the current flow through the motor armature rather by reversing the motor counter electromotive force (CEMF), which require a field-reversal. Thus, the field reversing is not required and much more rapid motor reversal is possible. In addition, the DC motor drives provide the fastest dynamic response to changes in torque or speed commands. The continuity of armature current is a desirable feature for the satisfactory operation of the control system. Continuous-current operation can be obtained by including additional inductance in series with the motor armature circuit, but even a large inductor cannot ensure continuous-current operation under all conditions of load and speed.

Other converter configurations, as the three-phase AC to DC converters can be used to reduce the size of the necessary inductor considerably, even though they do not eliminate it completely.


98

In Fig. 7 a three-phase fully controlled converter ([2]-[6]) is shown. It is the most popular and frequently used AC to DC converter for large power motor-control applications (up to 140kW). A half-controlled configuration, with three thyristors replaced by diodes is possible, but it can operate only in the

first quadrant of the ao iv − (or torque-speed) diagram. In

addition it has the disadvantage of introducing even harmonics into the line-current waveforms and is therefore unsuitable for large power applications. For continuous-mode operation the output voltage is:

παπ

<<= 0for ,cos3 ,

aV

VLm

o (6)

where LmV , is the maximum value of the line voltage.

11T13T

16T 12T

oV

iα

Rα

aL

be,f fr L

22T

25T 23T

26T

fV

15T

14T

24T

21T

Fig. 8. Dual three-phase fully controlled converter

As in the case of the single-phase fully controlled converter, the three-phase fully controlled converter provides the possibility for two-quadrant operation (motoring and regeneration or braking modes respectively). If four-quadrant operation is required, the dual three-phase fully controlled converter topology can be used (Fig. 8).

III. CHOPPER-FED DC MOTOR DRIVES

A chopper is a static power electronic device that converts fixed DC input voltage to a variable DC output voltage. It could be considered as DC equivalent of an AC transformer since it behaves in an identical manner. Choppers are now being used all over the world for rapid transit systems. They are also used in trolley cars, marine hoist, forklift trucks and mine haulers. Chopper systems are characterized by high efficiency, fast response and regeneration operation capability. The power semiconductor devices which are employed in these circuits can be force commutated thyristor, power BJT, MOSFET and IGBT. GTO and MCT based choppers are also used. All these devices actually operate like a switch. When the switch is off, no current can flow. Current flows through the load when switch is “on”. The power semiconductor devices have on-state voltage drop in the range of 0.5V to 2.5V across them and together with their switching characteristics, their power losses are determined.

Depending on the way that the transition from one switching state to another is implemented, the converters are divided in two categories: a) hard-switching converters and b) soft-switching converters.

A. Hard-Switching Converters For DC Drives

I

VOff

On

Soft-switching

Hard-switching

Safe Operating Area

snubbered

Fig. 9. Hard and soft switching characteristics

Since 1970, conventional PWM power converters have been operated in a switched mode operation. Power switches have to cut off the load current within the turn-on and turn-off time intervals under the hard-switching conditions. Hard-switching refers to the stressful switching behavior of the power electronic devices. The switching trajectory of a hard-switched power device is shown in Fig.9. During the turn-on and turn-off processes, the power device has to withstand high voltage and current simultaneously, resulting in high switching losses and stress.

Dissipative passive snubbers are usually added to the power circuits so that the dv/dt and di/dt of the power devices could be reduced, and the switching loss and stress are diverted to the passive snubber circuits. However, the switching loss is proportional to the switching frequency, thus limiting the maximum switching frequency of the power converters. Typical converter switching frequency was limited to a few tens of kHz (typically 20 kHz to 50 kHz) in early 1980’s. The stray inductive and capacitive components in the power circuits and power devices still cause considerable transient effects, which in turn give rise to electromagnetic interference (EMI) problems.

Rα

aL

be

inV D

SW

oV

Fig. 10. 1-quadrant hard switching DC motor drive

A single-switch chopper using a Thyristor, BJT, MOSFET or IGBT is presented in Fig.10. It can only supply positive voltage and current to a DC motor, and is therefore restricted to the first quadrant of (motoring) operation. For continuous conduction mode of operation the following expressions are valid ([5]):

ino VDV ⋅= (7)

( )1r inV V D D= − (8)


99

in bsw p p

a

V eI D I

R T

τ−

−= − (9)

1D p p b

a

DI I e

T R

τ−

−= − (10)

(1 )

1 1 / 1DT D T T

inp p

a

VI e e e

Rτ τ τ

−− − −

−

= − − −

(11)

/a aL Rτ = (12)

Where , , , , , , , sw Do b r p pV e V I I I T D− are the average output

voltage, back EMF voltage, output AC ripple voltage, average switch current, average diode current, peak to peak output ripple current, switching period and duty cycle respectively.

In Fig.11a a two-quadrant hard-switching DC motor drive is presented ([5]). Devices SW1 and D1 form a first-quadrant chopper and energy is delivered from the DC source Vin to the DC motor (motoring mode). Devices SW2 and D2 form a fourth-quadrant chopper and energy is delivered from the DC motor (regenerating mode) to the DC source Vin. For continuous conduction mode of operation (7), (8), (11) and (12) are still valid. The average switches and diodes currents are more complicated to be estimated because they depend on a) the polarity of the mean output current and b) the polarity of the max and min peak output current. In case that converter works in the first

t quadrant, (9) and (10) give the average

currents of SW1 and D1 respectively while are zero for SW2 and D2.

Rα aLbe

1SW2D

inV

oV

2SW1D

Rα aL

be

1SW2D

inV

2SW 1D

3SW4D

4SW 3D

oV

a) b)

Fig. 11. a) 2-quadrant and b) 4-quadrant hard-switching DC motor drives

A four-quadrant (H-bridge) DC chopper is shown in Fig. 11b ([5]). The four switches of the bridge result into a number of different control methods which can be used to produce four quadrant output voltage and current (bidirectional voltage and current). All methods should employ complementary device switching in each leg (either T1 or T4 on but not both and either T2 or T3 on, but not both) so as to minimize distortion by ensuring current continuity around zero current output.

One control method involves controlling the H-bridge as two virtually independent two-quadrant choppers, with the restriction that no two switches in the same leg conduct simultaneously. One chopper is formed with T1 and T4 together with D1 and D4, which gives positive output current but bidirectional voltage ±Vo (Q1 and Q2 operation). The

second chopper is formed by grouping T2 and T3 with D2 and D3, which gives negative output current but bi-direction voltage ±Vo (Q3 and Q4 operation). The second control method is to unify the operation of all four switches.

With both control methods, the chopper output voltage can be either multilevel or bipolar, depending on whether zero output voltage loops are employed or not. Bipolar output states increase the ripple current magnitude, but facilitate faster current reversal, without crossover distortion. Operation is independent of the direction of the output current. Two generalized unified H-bridge control approaches are considered: bipolar and three-level output.

For bipolar output voltage the average output voltage and AC ripple voltage are given below ([5]):

( )2 1o inV D V= − (12)

( )2 1r inV V D D= − (13)

The peak to peak output ripple current is twice the value given by (11).

For three-level output and for D≤0,5

( )2 1o inV D V= − (14)

( )2 1 2r inV V D D= − (15)

and for D≥0,5

( )2 1o inV D V= − (16)

( )2 (2 1) 1r inV V D D= − − (17)

B. Soft-Switching Converters For DC Drives

In all hard-switching DC to DC converter topologies the controllable switches operate in such mode that the entire load current is turned on and off. In this mode of operation the semiconductors are subjected to high switching stresses which in turn increase linearly their losses with the increase of switching frequency. Moreover, the problem of EMI becomes intense due to large di/dt and dv/dt. These shortcomings of switched mode converters are exacerbated when the frequency is increased in order to reduce the converter size and weight and hence to increase power density.

The above mention shortcomings are minimized if each switch in a converter, changes its state when the voltage across it or current through it, is zero at switching instant. This is succeeded by using a simple LC resonant circuit which shapes the current or voltage waveform such that the power device switches at zero-current (ZC) or zero-voltage (ZV) condition. Such topologies are termed ‘resonant soft-switching’ converters. Some soft-switching DC-DC topologies have been specially developed for dc motor drive ([16]-[26]), having the capability of bidirectional power flow for both motoring and


100

regenerative braking. Their operation is generally dictated by resonant elements and the characteristic impedance Z and angular frequency ω are defined as:

rrrr CLCL /1 ,/ ==Ζ ω (18)

SD

1SW

SC

D

2SWDC

RL

Rα

aL

be

inV oV

Fig. 12. Zero-voltage multi-resonant converter fed DC motor drive

A two-quadrant (2Q) Zero-Voltage Multi-Resonant (ZVMR) is presented in Fig.12 ([18]). This converter is created by adding a resonant inductor and two resonant capacitors to a conventional 2Q-PWM DC drive. It has been developed for bidirectional power flow for both motoring and regenerative braking DC motor drives. This soft switching converter not only possesses the advantages of achieving high switching frequencies ( >100 kHz) maintaining low current ripple of DC motor, with practically zero switching losses due to Zero-Voltage-Switching (ZVS) for all switches, but also provides full ranges of voltage conversion and load variation.

The ZVMR converter uses all built-in diodes of the power switches and absorbs all major parasitic. It should be noted that the ZVT technology is highly desirable for power MOSFET based power conversion. It is due to the fact that the power MOSFET device generally suffers from severe capacitive voltage turn-on losses. The 2Q–ZV–MR converter can handle both no-load up to short-circuit condition without any additional measures since it behaves as a constant current source after reaching the maximum output current.

The power rating of the semiconductors (MOSFET) associated with the MR cell are higher as compared with the conventional 2Q-PWM DC drive, due to the circulating energy and the conduction losses.

Rα

aL

be

inV

oV

Fig. 13. Zero-voltage-transition converter fed DC motor drive

A two-quadrant (2Q) Zero-Voltage-Transition (ZVT) converter ([19]) is presented in Fig.13. This converter has been also developed for bidirectional power flow for both motoring and regenerative braking DC motor drives. Compared with the conventional 2Q-PWM DC drive, it needs additional components: a resonant inductor, a resonant capacitor and two auxiliary switches.

The 2Q-ZVT converter exhibits some important advantages such as: Zero Voltage Switching (ZVS) for all main switches and diodes, unity device voltage and current stress during both the motoring and regenerative modes of operation, simple circuit topology, same resonant tank for both forward and backward power flows, full utilization of all built-in diodes of the power switches. These characteristics lead to the achievement of high switching frequency (>100 kHz), high power density and high efficiency.

Also, the operation of this converter requires the use of a DSP system in order to generate the appropriate control signals of the semiconductor switches.

Rα

aL

be

inVoV

Fig. 14. Zero-current–transition converter fed DC motor drive

A two-quadrant (2Q) Zero-Current-Transition (ZCT) converter ([21]) is presented in Fig.14. This converter also provides bidirectional power flow for both motoring and regenerative braking DC motor drives. The 2Q-ZCT converter, compared with its conventional PWM counterpart, needs additional components: a resonant inductor, a resonant capacitor and two auxiliary switches.

The main advantages of this converter are: ZCS for all main and auxiliary switches and diodes, minimum voltage and current stress, low cost simple circuit topology, same resonant tank for both forward and backward power flows, and full utilization of all built-in diodes of the power switches. All the previous mentioned characteristics lead to the achievement of switching frequency in the range of 50 kHz, high power density and high efficiency.

Rα Lbe

inV

oV

SαDα

′Sα

′Dα4S

4D

1S

1D

/ 2aC

/ 2aC3S

3D

2S

2DbS

bD

′bS

′bD

/ 2bC

/ 2bC

aL bL

Fig. 15. 4Q-ZVT converter fed DC motor drive

This converter is particularly useful for medium-power DC motor applications (a few kW), employing insulated-gate bipolar transistor (IGBT) as power devices, which generally suffer from diode reverse recovery during turn-on and severe inductive turn-off switching losses. Finally, a DSP system is also required for the proper operation of this converter.

A four-quadrant (4Q) Zero-Voltage-Transition (ZVT) converter ([21]-[22]) is presented in Fig.15. This converter has been developed mainly for MOSFETs like its two-quadrant


101

ZVT ancestor, and for motoring and regenerative braking in both forward and reversible operations for DC motor drives. The main advantages of this converter are: ZVS for all main and auxiliary switches and diodes, unity voltage and current stress, low circulating energy and simple circuit topology. Furthermore, the same resonant tank for both forward and backward power flows is used, full utilization of all built-in diodes of the power switches is achieved thus minimizing the overall hardware count and cost. All these lead to the achievement of high power density (up to 5kW) and high efficiency. To achieve ZVS operation, two sets of resonant tanks are utilized: inductor La, resonant capacitors Ca/2 with auxiliary switches Sa and Sa' for soft switching S1 and S4 and inductor Lb, resonant capacitors Cb /2 with auxiliary switches Sb and Sb' for soft switching S2 and S3. The DC motor can be considered to be simultaneously fed by two 2Q-ZVT converters as shown in Fig. 15.

Rα Lbe

bLbCaL

aCinV

oV

′Sα

′Dα

Sα

Dα

4S

4D

1S

1D

3S

3D

2S

2D′bS

′bD

bS

bD

Fig. 16. 4Q-ZCT converter fed DC motor drive

A four-quadrant (4Q) Zero-Current-Transition (ZCT) converter ([22]) is presented in Fig.16. This converter has been developed mainly for IGBTs, like its two-quadrant ZCT ancestor, and for motoring and regenerative braking in both forward and reversible operations for DC motor drives.

To achieve ZCS operation, two sets of resonant tanks are required, inductor La, resonant capacitor Ca, with auxiliary switches Sa and Sa' for soft switching of S1 and S4 and inductor Lb, capacitor Cb, with auxiliary switches Sb and Sb' for soft switching of S2 and S3. The DC motor can be considered to be simultaneously fed by two 2Q-ZCT converters. Therefore this four-quadrant ZCT converter has the same characteristics as its two-quadrant ZCT ancestor. This topology can be used for high frequency dc motor drive applications up to about 5 kW.

IV. COMPARATIVE EVALUATION OF DC MOTOR DRIVES

Controlled rectifier-fed DC drives remain common in industries, such as metals, cranes, mining and printing. For motors up to a few kilowatts the DC motor can be supplied from either single-phase or three-phase mains, but for larger motors (>15-20kW) three-phase is always used. Standard modular converters are available as off-the-shelf items in sizes from 0.5 kW up to several hundred kW. There are different controlled rectifier circuits available depending on the application. Single-phase controlled rectifiers are classified into one-quadrant, two-quadrant and four-quadrant operation topologies.

For one-quadrant operation half wave converters and asymmetrical semi-converters are available. The first converters are used for applications up to about 0.5 kW DC motors while the latter up to 15 kW. For one-quadrant operation, half-controlled converter exhibits better

characteristics than the fully controlled one, such as: less harmonics distortion to the input current, increased mean value of the output voltage for the same firing angle, power factor improvement and cheaper control circuits. For two-quadrant operation the full wave converter is the appropriate one for applications up to 15kW. Also, if a four-quadrant operation is needed the single-phase dual converter is the right choice for up to 15kW.

Three-phase controlled rectifiers are also classified into one-quadrant, two-quadrant and four-quadrant operation topologies. Although half-controlled 3Φ converters are available, fully controlled converter is the most popular and frequently used AC to DC converter for large power motor control applications (up to 140kW). A half-controlled configuration, with three thyristors replaced by diodes has the disadvantage of introducing even harmonics into the line current waveforms and is therefore unsuitable for large power applications. Furthermore, half-controlled converter is one-quadrant converter while the fully controlled one is a two-quadrant converter. When a four-quadrant operation is required function a dual three-phase fully controlled converter is also available.

Choppers are now being used all over the world for rapid transit systems. They have replaced conventional controlled-rectifier converters in many DC motor applications due to their high efficiency, fast response and regeneration operation capability. Due to the high switching frequency the armature ripple current is decreased and therefore motor losses and torque ripple are also decreased.

As it was mentioned in the previous section, choppers are divided into hard-switching and soft-switching converters. Both of these converters are further classified as: one, two or four-quadrant converters. Hard-switching converters utilize thyristor, GTO, MCT, BJT, MOSFET and IGBT depending on the DC motor power and required frequency. Thyristors and GTOs are utilized for high power applications (up to several hundred kW) and for low switching frequencies (up to several hundred Hz). Power BJTs and mainly IGBTs and MOSFETs are used when high frequency (in the range of 20-50 kHz) and for low to medium power is required. All these choppers usually employ PWM control techniques.

When even higher switching frequencies are required, i.e. for low-inductance DC motors, soft-switching choppers can be employed in order to reduce switching losses on one hand and to limit Electromagnetic Interference (EMI) on the other. Furthermore, soft-switching techniques can reduce converter losses and therefore increase converters’ efficiency.

More specifically, a two-quadrant (2Q) Zero-Voltage Multi-Resonant (ZVMR) converter not only possesses the advantages of achieving high switching frequencies (>100 kHz) maintaining low ripple current, with practically zero switching losses due to Zero-Voltage-Switching (ZVS) for all switches, but also provides full ranges of voltage conversion and load variation. On the other hand, the power rating of the MOSFET associated with the MR cell are higher as compared to the conventional 2Q-PWM DC drive (hard switching), due to the circulating energy and the conduction losses.


102

A 2Q-ZVT converter is proposed as a better solution in place of ZVMR when MOSEFTS are employed or a 2Q-ZCT one when IGBTs are used. The main reasons are that this converter achieves ZVS for all main switches and diodes, unity device voltage and current stress during both the motoring and regenerative modes of operation, simple circuit topology, same resonant tank for both forward and backward power flows, full utilization of all built-in diodes of the power switches. In this way high switching frequency (>100 kHz), high power density and high efficiency are achieved. The 2Q-ZCT converter is intended to be used for medium-power DC motor applications (a few kW) and for switching frequency in the range of 50 kHz, employing insulated-gate bipolar transistor (IGBT) as power devices. Both of the last two soft-switching converters need a DSP system for the proper driving of theirs semiconductor switches.

Finally, 4Q-ZVT and 4Q-ZCT converters are available for MOSFETs and IGBTs respectively. These converters can be used for high frequency DC motor drives and for applications up to about 5 kW. A DSP system is also needed for the proper operation of these converters.

V. CONCLUSIONS

In this paper a review of the basic topologies for DC motor drives is attempted. DC motors are still used in several applications as in industrial production and processing of paper pulp, textile industries and in public transport such as TRAM (trolley) and METRO.

DC motor drives are categorized according to the converter it is utilized. The two main categories controlled rectifier (thyristor) and chopper-fed DC drives are presented in order to outline their basic characteristics. Moreover, the basic equations, key advantages and application fields were attempted to be presented. Both of the previous mentioned categories can further subdivided into non-generative and generative drives and this operation characteristic has been exploited for their comparative evaluation.

ACKNOWLEDGMENT

Research co-funded by the E.U. (European Social Fund) and national funds, action “Archimedes III–Funding of research groups in T.E.I.”, under the Operational Programme “Education and Lifelong Learning 2007-2013”.

REFERENCES

[1] S. J. Chapman, Electric Machinery Fundamentals, New York: WCB/McGraw-Hill,1998

[2] N.Mohan, T.M. Undeland, Power Electronics, Converters, Applications and Design, John Wiley & Sons, 1995

[3] S.N.Manias, Power Electronics, Simeon Publications, 2012 (in Greek)

[4] P. Malatestas, Motor Drives, Tziolas Publications, 2013 (in Greek)

[5] B.W. Williams, Power Electronics: Devices, Drivers, Applications, and Passive Components, McGraw-Hill , 1992

[6] M. Nedeljkovic and Z. Stojiljkovic, “Fast current control for thyristor rectifiers,” IEE Proceedings- Electr. Power Appl., vol. 150, no. 6, pp. 636- 638, November 2003.

[7] Alfio Consoli, Mario Cacciato, Antonio Testa and Francesco Gennaro, “Single chip integration for motor drive converters with power factor Capability,” IEEE Transactions on Power Electronics, vol. 19, no. 6, pp. 1372-1379, November 2004

[8] Manoj Daigavane, Hiralal Suryawanshi and Jawed Khan, “A Novel Three Phase Series-Parallel Resonant Converter Fed DC-Drive System,” Journal of Power Electronics, vol. 7, no. 3, pp. 222-232, July 2007.

[9] R.Gupta, R.Lamba, S.Padhee, “Thyristor based speed control techniques of DC motor: A comparative analysis,” International Journal of Scientific and Research Publications, vol. 2, no. 6, June 2012

[10] Chau, K. T., Y. S. Lee, and A. Ioinovici, “Computer-aided modeling of quasi-resonant converters in the presence of parasitic losses by using MISSCO concept,” IEEE Transactions on Industrial Electronics, vol. 38, no. 6, pp.454-461, 1991.

[11] Chan, C. C., and K. T. Chau, “A new zero-voltage-switching dc/dc boost converter,” IEEE Transactions on Aerospace and Electronic Systems, vol. 29, no. 1, pp.125-134, 1993.

[12] Chau, K. T., “New Constant-frequency multi-resonant boost convertor,” IEE Electronics Letters, vol. 30, no. 2, pp.101-102, 1994.

[13] Chong, C. C., Chan, C.Y., and Foo, C. F., “ A quasi-resonant converter-fed dc drive system,” Proceedings of the European Power Electronics Conference, pp. 372- 377, 1993

[14] Hua, G., and F. C. Lee, “Soft-switching Techniques in PWM Converters,” IEEE Transactions on Industrial Electronics, vol. 42, no. 6, pp.595-603, 1995.

[15] Luo, F. L., and L. Jin, “Two-quadrant DC/DC Soft-switching Converter,” Proceedings of IEEE Power Electronics Specialists Conference, vol. 1, pp.173-178, 2000.

[16] Uma, G., and C. Chellamuthu, “Modeling and Design of Fuzzy Speed Controller for Constant Frequency Zero Current Switched Converter Fed DC Servo Motor for Battery Operated Vehicles,” Proceedings of International Conference on Power System Technology, vol. 1, pp. 211-215, 2000.

[17] T.W.Ching, “Soft-switching converters for electric vehicle propulsion,” Journal of Asian Electric Vehicles,vol.5, no.2, pp.1019-1026, December 2007.

[18] Chau, K. T., T. W. Ching, and C. C. Chan, “Constant frequency multi-resonant converter-fed dc motor drives," in Proc. IEEE Industrial Electronics, Control, and Instrumentation Conf., pp. 78-83, 1996

[19] Chau, K. T., and T. W. Ching, “A new two-quadrant zero voltage transition converter for dc motor drives,” International Journal of Electronics, vol. 86, no. 2, pp. 217-231, 1999,

[20] Ching, T. W., and K. T. Chau, “A new two-quadrant zero current transition converter for dc motor drives,”International Journal of Electronics, vol. 88, no. 6,, pp. 719-735, 2001

[21] Ching, T. W., "Four-quadrant zero-voltage-transition converter-fed dc motor drives for electric propulsion," Journal of Asian Electric Vehicles, vol. 3, no. 1, pp. 651-656, 2005

[22] Ching, T. W., "Four-quadrant zero-current-transition converter-fed dc motor drives for Electric Propulsion," Journal of Asian Electric Vehicles, vol. 4, no. 2, pp. 911-918, 2006

[23] Ching, T. W., “Review of Soft-switching Technologies for High-frequency Switched-mode Power Conversion,” International Journal of Electrical Engineering Education, paper No. 4042, 2008.

[24] Felix Joseph, X., Pushpa Kumar, “Design and Simulation of a Soft Switched Dc Boost Converter for Switched Reluctance Motor”, American Journal of Applied Sciences 9 (3): 440-445, 2012

[25] S. Waffler and J.W. Kolar, “Comparative Evaluation of Soft-Switching Concepts for Bi-directional Buck+Boost Dc-Dc Converters,” International Power Electronics Conference (IPEC), 21-24 June 2010.

[26] Premananda Pany, R.K. Singh, R.K. Tripathi, “Bidirectional DC-DC converter fed drive for electric vehicle system”, International Journal of Engineering, Science and Technology, Vol. 3, No. 3, 2011, pp. 101-110, 2011.


103

Secure Communication for Cognitive Networks Based on MIMO and Spreading LDPC Codes

Yang Xiao, Kaiyao Wang Institute of Information Science, Beijing Jiaotong University

Beijing 100044, P.R.China

Emails: [email protected], [email protected]

Abstract—The existing cognitive radio (CR) systems lack the ability to deal with the attack of a pretended primary user (PPU), which can palsy cognitive network. To solve the problem, in this paper, a secure cognitive radio (CR) system is proposed. The proposed systems combines MIMO and spread spectrum techniques as well as low density parity check (LDPC) codes to cancel the electronic interference of PPU. The simulation results in Rayleigh flat-fading channel show that, comparing with MIMO cognitive system without the proposed method, the proposed approach can effectively cancel electronic interference from enemy and channel noises interference, and obtain about 9dB gain.

Index Terms—secure communication, MIMO cognitive systems,

electronic interference cancellation, protograph low density

parity check (LDPC) codes, spread spectrum

I. INTRODUCTION

Nowadays, radio spectrum has become a very precious resource in wireless communication. There is little radio frequency spectrum left over. The cognitive radio (CR) technology can increase radio frequency spectrum usage efficiency by spectrum sharing. As a promising solution for this problem, the CR technology is being studied extensively [1-8]. Two kinds of users are served in cognitive systems, primary users (PUs) and cognitive users (CUs). However, in the existing schemes [1-8], CUs and PUs share the same frequency spectrum in a time division. If an attacker pretends a primary user, i.e. a PPU, it is difficult to guarantee the multiple CUs communicate over the frequency spectrum using by PPU.

More recently, multi-terminal communication with confidential messages has been studied intensively [9-16]. Most of existing secure communication schemes are focus on the study of secrecy capacity. They interested in protecting the common message against eavesdropping. For the secure transmission of information over an ergodic fading channel in the presence of an attacker, and the attacker can pretend a empowered user. The secrecy capacity of such a system with ergodic fading channel is characterized under the assumption of asymptotically long coherence intervals. [9, 10] investigated the rate-equivocation region and secrecy capacity region for ergodic fading channel. The impact of fading on secure communication also was studied in [11-16], and some of them

exploited the feedback and cooperation to improve the secrecy for CR networks. However, the above existing secure communication schemes did not consider that the attacker (PPU) can send electronic interference signal to CR client receiver. Generally, when detecting the channel is occupied by PU, CUs had to withdraw and keep off. Thus, it is difficult to guarantee reliable communications under the condition of strong electronic interference from a PPU by the approaches of [9-16]. To solve the problems, in this paper, a secure MIMO cognitive system is proposed, in which the multiple CUs communicate over Rayleigh flat-fading channels and a PPU sends an electronic interference signal to CUs. We are interested in protecting the secure communication between multiple CUs against electronic interference signal from PPU.

Different from [9-16], the proposed secure MIMO cognitive system introduces spread spectrum technique and LDPC codes for the electronic interference cancellation of PPU. This scheme can ensure multiple CUs secure communication under the condition of strong electronic interference from PPU. Protograph LDPC codes [17-19] and spread spectrum are applied to the proposed secure MIMO cognitive system, the information transmission of CR network can be of the anti-interference ability and error correcting capability through channel coding and spread spectrum mapping, which can cancel electronic interference of PPU and channel noises interference. In our scheme, spread spectrum mapping does only require DSP (Digital Signal Processor) programming to be achieved, but does not require the circuit and spread spectrum synchronization. Protograph LDPC codes have the advantage of high-speed decoding, low error floor and low iterative decoding threshold [20-23]. The protograph LDPC codes used in this method can be of fast encoding. Thus, it can simplify the hardware complexity of communication equipment.

In the proposed secure MIMO cognitive system, the diversity combining is used to increase information transmission reliability. The different antennas of source send the same signals to obtain diversity gain. The electronic interference of PPU and channel noise are looked as a new noise in the receiver. The expected information can be obtained through despreading mapping and channel decoding of the proposed secure MIMO cognitive system. The simulation results in Rayleigh flat-fading channel show that, comparing with MIMO cognitive system without the proposed method, the proposed secure MIMO cognitive system can effectively

___________________ This work was supported by the Beijing Natural Science Foundation of China: (No. 4102050).


104

cancel electronic interference from PU and channel noises interference, and obtain about 9dB gain.

II. ELECTRONIC INTERFERENCE CANCELLATION OF PPU

In this section, we proposed a method that electronic

interference cancellation based on spread spectrum LDPC codes in MIMO cognitive system. This method can make multiple CUs secure communication under the condition of strong electronic interference from PU. In this paper, we assume the channel parameters in a transmission cycle of signals frame are constant. In the method, the diversity combining is used to increase information transmission reliability. For the receiver, the received signals were combined by using equal gain combining. The strong electronic interference signals from enemy and channel noise looked as a new noise. It minimizes new noise interference by despreading mapping and decoding.

In this paper, the MIMO cognitive system block diagram showed in Figure 1, including an interference source, a base station and k mobile terminals. In Figure 1, the interference source PPU and the base station both have 2 antennas, and each mobile terminal also has 2 antennas. In this paper, it is assumed that the fading is quasistatic and the channel state information (CSI) of the PPU is not available at the source, while CR mobiles and CR base-station know the CSI of them, which provides the chance to cancel the attack of PPU for the proposed system.

Fig.1 MIMO cognitive system block diagram From Fig. 1, we know that PPU can attack the

communication between CR mobiles and CR base-station in uplink or in downlink by sending an electronic interference signal in the bandwidth of CR network, if PPU occupies the bandwidth of CR network. To solve the problem, we propose a method that electronic interference cancellation based on spread spectrum LDPC codes in MIMO cognitive system under two conditions (uplink communication system and downlink communication system). Refs. [9]-[16] only consider how receivers avoid their information not to be

obtained by eavesdropper. Their approaches are not available to our application in Fig.1.

A. Uplink communication system

In the uplink, the mobile terminals transmit data, and the base station receives data. The proposed MIMO cognitive system uplink communication system model is shown in Figure 2.

information source 1

LDPC encoder

1m 1x

1m

1c

y

1c

spread spectrum mapping

1r

txtT

RFmodulation

1T

km kxkc kT

kmkc

information source k

LDPC encoder

RFmodulation

Interference source

spread spectrum mapping

RFmodulation

RFdemodulation

despreadingmapping

despreadingmapping

LDPC decoder

LDPC decoder

information sink 1

information sink k

tT

1T

kT

2r

equal gain combining

1y2y

Fig. 2 MIMO cognitive system uplink communication

system model The information bit from mobile terminal is denoted

1 (1 )Mi i k×∈ ≤ ≤Zm , M is the length of information bit.

Protograph LDPC codes encode the information bit im , and the

encoded information is denoted 1 (1 )Ni i k×∈ ≤ ≤Zc , N is the

length of codes. It can be written as = ic m G , where G is the generate matrix of protograph LDPC codes.

The spread spectrum signal is denoted 1 ( ) (1 )N K

i i k× ⋅∈ ≤ ≤Zx by spread spectrum mapped encoded

information ic . The mapping relation of baseband signals and

codes as following: 1 ,0→ →−d d , where 1 (1 )K

i i k×∈ ≤ ≤Zd is the spreading codes sequences, and K

is the length of spreading codes sequences. In the proposed scheme, spread spectrum mapping does

only require DSP (Digital Signal Processor) programming to achieve, and not require the circuit and spread spectrum synchronization. Thus, it can simplify the hardware complexity of communication equipment.

Spread spectrum signal [ ]T(1) (2) ( )i i i i N=x x x x

is stored in a N K× two-dimensional matrix. Baseband spread spectrum signal was transmitted successive through the transmitting antenna. The spread spectrum signal block

diagram is shown in Figure 3, where ( )jix n denotes the

thj chip of the thn encoded information bit.

The baseband signal from mobile terminals (1 )i i k≤ ≤ can

be written as ( ) ( )i i in c n= ⋅x d (1)


105

where, ( )ic n is the thn encoded information bit, 1( ) K

i n ×∈Zx is the thn baseband spread spectrum signal

sequence. Baseband signal ix after the RF modulating obtain send

signals (1 )i i k≤ ≤T . Interference source signals tx after the

RF modulating obtain send signals tT .

1(1)ix 2 (1)ix 3(1)ix (1)Kix

1(2)ix 2 (2)ix 3(2)ix

1(3)ix 2 (3)ix 3(3)ix

1( )ix N 2 ( )ix N 3( )ix N

(2)Kix

(3)Kix

( )Kix N

Fig. 3 Spread spectrum signal block diagram In BPSK (Binary Phase Shift Keying) system, the sending

signal from mobile terminal i can be written as 0

1( ) cos(2 ) ( ) ( )i i c i i cn

T t A f t c n d p t nTπ= ⋅ × ⋅ ⋅ −∑ (2)

where, 0iA is the transmission amplitude, cf is the carrier

frequency, 1p is impulse response of pulse shaping filter.

The receive signal in the receiver can be written as

, , 1 ,( ) cos(2 ) ( ) ( )

( ) ( 1,2)

j i j c i j i i c i ji n

r t A f t c n d p t nT

v t j

π ϕ τ= ⋅ + × ⋅ ⋅ − −

+ =

∑ ∑ (3)

where, ,i jA and ,i jϕ are the carrier amplitude and phase

position, ,i jτ is the transmission delay, v is channel white

Gaussian noise. The baseband uplink channel model of MIMO cognitive

system is shown in Figure 4. The channel transfer matrix from mobile terminals

(1 )i i k≤ ≤ to base station is denoted 2 2i

×∈QH , and

,1 ,2

,3 ,4

i ii

i i

h h

h h

⎡ ⎤= ⎢ ⎥⎣ ⎦

H , where the entries of (1 )i i k≤ ≤H are

independent and identically distributed (i.i.d.). The channel transfer matrix from interference source to base station is

denoted 2 2t

×∈QH , and ,1 ,2

,3 ,4

t tt

t t

h h

h h

⎡ ⎤= ⎢ ⎥⎣ ⎦

H . ix is transmission

signal from mobile terminals (1 )i i k≤ ≤ . 1 ( )N Kt

× ⋅∈Zx is

interference signal from PPU interference source. 1 ( )

1 2, N K× ⋅∈Qv v are channel white Gaussian noise.

1 ( )1 2, N K× ⋅∈Qy y are the received signals in base station

receiver.

1x

1y

2x

tx

1H

1x

2x

kx

kx

2y

tx

2H

kH

tH

Fig.4 The baseband uplink channel model of MIMO

cognitive system In the base station receiver, the received signals 1 2,y y

after equal gain combining, the combining signals 1 ( )N K× ⋅∈Qy

can be obtained. The electronic interference of PPU and channel noise are looked as a new noise. The expected information can be obtained through dispreading mapping and channel decoding of the proposed system.

In a transmission cycle of signals frame, the baseband signals model in base station receiver is as follows.

1 1

12 2

( ) ( )( ) ( )

( ) ( )( ) ( )

ki t

i ti i t

n nn n

n nn n=

⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤= ⋅ + ⋅ +⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥

⎣ ⎦ ⎣ ⎦⎣ ⎦ ⎣ ⎦∑

x xy vH H

x xy v (4)

where 11 2( ), ( ) Kn n ×∈Qy y are the thn received signals

sequences. After spread spectrum mapping, 1( ) Kt n ×∈Zx is

the thn interference signals sequences. 11 2( ), ( ) Kn n ×∈Qv v are

the thn channel white Gaussian noise sequences. Equation (4) can be written as

1 ,1 ,2 ,1 ,2 11

2 ,3 ,4 ,3 ,4 21

( ) ( ) ( ) ( ) ( ) ( )

( ) ( ) ( ) ( ) ( ) ( )

k

i i i t t ti

k

i i i t t ti

n h h n h h n n

n h h n h h n n

=

=

= + ⋅ + + ⋅ +

= + ⋅ + + ⋅ +

∑

∑y

y x x v

x x v

(5)

According to (1), (5) can be written as

1 ,1 ,2 ,1 ,21

1

2 ,3 ,4 ,3 ,41

2

( ) ( ) ( ) ( ) ( )

( )

( ) ( ) ( ) ( ) ( )

( )

k

i i i i t t ti

k

i i i i t t ti

n h h c n h h n

n

n h h c n h h n

n

=

=

= + ⋅ ⋅ + + ⋅

+

= + ⋅ ⋅ + + ⋅

+

∑

∑y

y d x

v

d x

v

(6)

where id is the spreading codes sequences, and

0( )Ti j i j⋅ = ≠d d , A( )T

i j i j⋅ = =d d , A is a positive integer.

In this paper, we define


106

4

,1

(1 )i i jj

h i kα=

= ≤ ≤∑ and4

,1

t t jj

hα=

= ∑ .

In the base station receiver, the received signals 1y and 2y

after equal gain combining, (6) can be written as 2

1 21 1

( ) ( ) ( ) ( ) ( ) ( )K

i i i t t ii i

n n n c n n nα α= =

= + = ⋅ ⋅ + ⋅ +∑ ∑yy y d x v (7)

The base station receiver extracts mobile terminal 1 signal,

the combining signals y after dispreading mapping, (7) can be written as

21 1

11 1

2

1 1 11 1

21

1 1 11

1 1

( ) ( ) ( ( ) ( ) ( ))A A

1( ( ) ( ) ( ) )

A

( ) 1( ) ( )

A A

( ) ( ) '( )

T TK

i i i t t ii i

KT T T

i i i t t ii i

TTt

t ii

t t

c n n c n n n

c n n n

nc n n

c n x n v n

α α

α α

α α

α α

= =

= =

=

= ⋅ = ⋅ ⋅ + ⋅ + ⋅

= ⋅ ⋅ ⋅ ⋅ + ⋅ ⋅ + ⋅

⋅= ⋅ + ⋅ + ⋅ ⋅

′= ⋅ + ⋅ +

∑ ∑

∑ ∑

∑

d dy d x v

d d x d v d

x dv d

(8)

It can also be written as

1 1 1 't tα α ′= ⋅ + ⋅ +c c x v (9)

where, t′x is the interference signal after dispreading

mapping, ′v is channel white Gaussian noise after dispreading mapping.

From the above equation, after despreading mapping, the interference from other mobile terminals has been eliminated, and the received signals only are of the interference from interference source and channel noise.

According to (9), we can see that signal 1c includes co-

channel interference t tα ′⋅ x from interference source signal t′x .

Let interference of PPU be C t tα ′= ⋅I x . Because interference

CI is not controllable, the despreading mapping in (9) cannot

completely cancel the interference CI . In order to solve this

problem, we use protograph LDPC codes to cancel the interference CI . The design of protograph LDPC will be

provided in next section. The interference source signal and channel noise

interference in equation (9) look as a new noise, it can be written as

C t tα ′ ′= ⋅ +v x v (10)

According to (10), signal 1c can be written as

1 1 1 Cα= ⋅ +c c v (11)

Then, decoding the received signals 1c by protograph

LDPC codes, we can obtain the information 1m from

transmitter. It can be written as

1 1 1 1dec( ) dec( )Cα= = ⋅ +m c c v (12)

where dec( )⋅ denotes the protograph LDPC decoding.

Now, we analyze the channel coefficient influence the BER performance of cognitive base station receiver.

After normalization processing, (9) can be written as

1 11 1

1't

t

αα α

′ ′= + ⋅ + ⋅y c x v (13)

Generally, the base station receiver extracts signal of

mobile terminal i , (13) can be written as 1

' (1 )ti i t

i i

i kαα α

′ ′= + ⋅ + ⋅ ≤ ≤y c x v (14)

From (14), in this paper we define

R ti

i

αα

= (15)

as channel matrix ratio. In the uplink communication system receiver, the value of

iα are different, and the value of tα is fixed. This means that

the greater value of iα , the smaller value of R i , and the

smaller interference of tt

i

αα

′⋅ x and1

'iα⋅ v to received signals.

The BER performance depends on the value of R i in the

uplink communication system. We will see it in the following simulations.

B. Downlink communication system

In the downlink, the base station transmits data, and the

mobile terminals receive data. The downlink communication system model of MIMO cognitive system is shown in Figure 5.

1m 1x

1m

1c

1y1c

km kxkc

kc

txtT

tT

T

T

x

1,1r

1,2r

kmky,1kr

,2kr

1,1y

1,2y

,1ky

,2ky Fig.5 MIMO cognitive system downlink communication

system model The MIMO cognitive system baseband downlink channel

model is shown in Figure 6. The downlink channel model of MIMO cognitive system

baseband The baseband signals from base station can be written as

1 1

( ) ( ) ( )K K

i i ii i

n n c n= =

= = ⋅∑ ∑x x d (16)


107

The channel transfer matrix from interference source to mobile terminal (1 )i i k≤ ≤ is denoted tiH ,

and ,1 ,2

,3 ,4

ti titi

ti ti

h h

h h

⎡ ⎤= ⎢ ⎥⎣ ⎦

H .

Interference signal Base Station data transmission

Base Station

x

x

Mobile Terminal k

Mobile Terminal 1

Mobile Terminal 2

1,1y

1,2y

2,1y

2,2y

,1ky

,2kytxInterference

Source

tx

1H

2H

kH

1tH

2tH

tkH

In a transmission cycle of signals frame, the baseband

signals model in mobile terminal 1 receiver is as follows.

1,1 11 1

1,2 2

( ) ( ) ( )( )

( ) ( ) ( )( )t

tt

n n nn

n n nn

⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎡ ⎤= ⋅ + ⋅ +⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥

⎣ ⎦ ⎣ ⎦⎣ ⎦⎣ ⎦

y x vxH H

y x vx (17)

Equation (17) can be written as

1 1,1 1,2 1,1 1,2 1

2 1,3 1,4 1,3 1,4 2

( ) ( ) ( ) ( ) ( ) ( )

( ) ( ) ( ) ( ) ( ) ( )t t t

t t t

n h h n h h n n

n h h n h h n n

= + ⋅ + + ⋅ +

= + ⋅ + + ⋅ +y

y x x v

x x v (18)

According to (16), (18) can be written as

1,1 1,1 1,2 1,1 1,21

1

1,2 1,3 1,4 1,3 1,41

2

( ) ( ) ( ) ( ) ( )

( )

( ) ( ) ( ) ( ) ( )

( )

K

i i t t ti

K

i i t t ti

n h h c n h h n

n

n h h c n h h n

n

=

=

= + ⋅ ⋅ + + ⋅

+

= + ⋅ ⋅ + + ⋅

+

∑

∑

y d x

v

y d x

v

(19)

In this paper, we define4

,1

ti ti jj

hα=

= ∑ .

In the mobile terminal 1 receiver, the received signals 1,1y

and 1,2y after equal gain combining, (19) can be written as

1 1,1 1,22

1 11 1

( ) ( ) ( )

( ) ( ) ( )K

i i t t ii i

n n n

c n n nα α= =

= +

= ⋅ ⋅ + ⋅ +∑ ∑yy y

d x v (20)

The mobile terminal 1 receiver extracts the signal from base station, the combining signals 1y after dispreading

mapping, (20) can be written as

21 1

1 1 1 11 1

2

1 1 1 1 11 1

21

1 1 1 11

1 1 1

( ) ( ) ( ( ) ( ) ( ))A A

1( ( ) ( ) ( ) )

A

( ) 1( ) ( )

A A

( ) ( ) '( )

T TK

i i t t ii i

KT T T

i i t t ii i

TTt

t ii

t t

c n n c n n n

c n n n

nc n n

c n x n v n

α α

α α

α α

α α

= =

= =

=

= ⋅ = ⋅ ⋅ + ⋅ + ⋅

= ⋅ ⋅ ⋅ ⋅ + ⋅ ⋅ + ⋅

⋅= ⋅ + ⋅ + ⋅ ⋅

′= ⋅ + ⋅ +

∑ ∑

∑ ∑

∑

d dy d x v

d d x d v d

x dv d

(21)

It can also be written as

1 1 1 1 't tα α ′= ⋅ + ⋅ +c c x v (22)

where, t′x is the interference signal after dispreading

mapping, ′v is channel white Gaussian noise after dispreading mapping.

From the above equation, after despreading mapping, the interference from other mobile terminals has been eliminated, and the received signals only are of the interference from interference source and channel noise.

The PPU interference source signal and channel noise interference in equation (22) look as a new noise, it can be written as

1C t tα ′ ′= ⋅ +v x v (23)

According to (23), signal 1c can be written as

1 1 1 Cα= ⋅ +c c v (24)

Then, decoding the received signals 1c by protograph

LDPC codes, we can obtain the information 1m from

transmitter. It can be written as

1 1 1 1dec( ) dec( )Cα= = ⋅ +m c c v (25)

where dec( )⋅ denotes the protograph LDPC decoding.

III. PROTOGRAPH LDPC CODES

In this section, we give the construction of protograph LDPC codes with fast encoding, which is used to cancel the electronic interference and channel noises interference.

The proposed rate 1/2 protograph LPDC ( , )N M codes are

short codes with good BER performance, simple structure and low encoding complexity. The size of parity check matrix

[ ]a b=H H H of protograph LDPC codes is M N× , where

matrix aH and bH consists of 8 8× sub matrices.

The matrix bH is shown as below,

8 8

b

×

⎡ ⎤⎢ ⎥⎢ ⎥=⎢ ⎥⎢ ⎥⎣ ⎦

I 0

I IH

0 I I

(26)

where I and 0 are 16 16N N× identity and zero matrices,

respectively. We give the matrix aH . It can be written as


108

1 2 6 7

3 4 5 8 9a

⊕ ⊕⎡ ⎤= ⎢ ⎥⊕ ⊕ ⊕⎣ ⎦

L L L LH

L L L L L (27)

where matrix 1L ~ 8L are 4 4N N× permutation matrices.

The permutation matrices we used in this paper are similar to [26].

Permutation matrix kL has non-zero entries in row i and

column ( )πk i for 0, , 4 1i N∈ − and

( ) ( )( ) ( )( )π 16 mod4 16 mod16 16k k k

N Ni i N i N iθ φ= + ⋅ + ⋅ +⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦ (28)

The permutation matrix can be divided into 4 4× unit circulate sub-matrices, the matrix aH consist of 8 8× sub

matrices. The matrix aH also can be written as

1,1 1,2 1,8

2,1 2,2 2,8

8,1 8,2 8,8

a a a

a a a

a

a a a

⎡ ⎤⎢ ⎥⎢ ⎥= ⎢ ⎥⎢ ⎥⎢ ⎥⎣ ⎦

I I I

I I IH

I I I

(29)

where the matrix ,i jaI is identity matrix right shift ,i ja bit.

Now, we give the fast encoding algorithm of protograph LDPC codes.

Suppose, codes vector is denoted [ ]=c S P , where

[ ]1 2 8=S S S S is the information vector,

and [ ]1 2 8=P P P P is the check vector.

According to the channel coding theory, we know that T× =H c 0 , it can be written as

[ ] [ ]Ta b ⋅ =H H S P 0 (30)

1,1 1,2 1,8

2,1 2,2 2,8

8,1 8,2 8,8

1 2 8 1

1 2 8 1 2

1 2 8 7 8

a a aT T T T

a a aT T T T T

a a aT T T T T

× ⊕ × ⊕ × ⊕ =× ⊕ × ⊕ × ⊕ ⊕ =

× ⊕ × ⊕ × ⊕ ⊕ =

I S I S I S P 0I S I S I S P P 0

I S I S I S P P 0

(31)

Then we can obtain 1 2 8T T T, , ,P P P .

1,

2 ,

8,

8

11

8

2 11

8

8 71

j

j

j

aT Tj

j=

aT T Tj

j=

aT T Tj

j=

⎧ = ×⎪⎪⎪

= ⊕ ×⎪⎨⎪⎪⎪

= ⊕ ×⎪⎩

∑

∑

∑

P I S

P P I S

P P I S

(32)

Equation (32) is the proposed fast encoding algorithm. If the information codes vector S and check matrix H are

known, applying (32), we can obtain the code vector

[ ]=c S P .

The sub-matrices of the parity check matrix H are unit circulate matrices, if the first row of matrix is known, the other rows can be obtained by first row of right shift. Thus we only need to store the first row of each sub-matrix in the encoding process. This can reduce the computation and storage space.

The proposed fast encoding algorithm (32) simplifies the hardware complexity of LDPC encoder.

IV. EVALUATIONS AND SIMULATION RESULTS

In this section, we evaluate performances of the proposed method that electronic interference cancellation based on spread spectrum LDPC codes in MIMO system. According to the design approach of protograph LDPC codes in Section III, let 1024N = , we can obtain the parity check matrix H of rate 1/2 protograph LDPC ( 1024, 512)N M= = codes by (27) and

(28), where the functions kθ and ( )k jφ are defined in Table 1.

Table 1 Description of kθ and ( )k jφ

k kθ (0)kφ (1)kφ (2)kφ (3)kφ

1 0 39 40 31 9

2 2 20 18 53 11

3 0 12 30 1 38

4 2 29 59 8 43

5 3 24 23 7 2

6 0 41 17 18 47

7 2 10 15 14 3

8 1 45 6 39 41

9 3 51 5 49 26

According to [27], we can obtain the check matrix H of rate 1/2 Tanner LDPC codes ( 1024, 512)n k= = , where

5, 2, 128a b m= = = . Theorem 1[28]: If and only if the elements of THH are 0

or 1 except in diagonal line, the LDPC codes has no girth-4, where H is the parity check matrix.

Now, we test girth_4 of protograph LDPC codes and Tanner LDPC codes by applying theorem 1, as shown in Fig.6.

The girth_4 test of protograph LDPC codes and Tanner

LDPC codes According to Fig.6, we observe that the elements of

THH are 0 or 1 except in diagonal line, and we found that there is no girth-4 in both protograph LDPC codes and Tanner LDPC codes.

In the simulation results, method 1: electronic interference cancellation based on spread spectrum protograph LDPC codes; method 2: electronic interference cancellation based on spread


109

spectrum Tanner LDPC codes; method 3: electronic interference cancellation based on spread spectrum codes.

Conditions of simulation experiment: Rayleigh fading channel, the decoders of two codes (protograph LDPC codes and Tanner LDPC codes) use the same BP algorithm [28], the same code rate 1/2, 4000 data frames per SNR point, the spreading codes sequences of method 1 and 2 use 128-order hadamard matrix, the spreading codes sequences of method 3 use 256-order hadamard matrix. The code rates of three methods are 1/256.

Take a MIMO system with 16 mobile terminals, a base station and an interference source for example. We evaluate the BER performances of the proposed interference cancellation based on spread spectrum protograph LDPC codes under two conditions (uplink communication system and downlink communication system).

Uplink communication system The channel transfer matrix from mobile terminal 1 to base

station is 1

0.4503 0.3714

0.3359 0.4208

⎡ ⎤= ⎢ ⎥⎣ ⎦

H ; The channel transfer matrix

from mobile terminal 2 to base station is

2

0.3106 0.2874

0.2417 0.3462

⎡ ⎤= ⎢ ⎥⎣ ⎦

H ; The channel transfer matrix from

interference source to base station is 0.9201 0.3746

0.3314 0.8947t

⎡ ⎤= ⎢ ⎥⎣ ⎦

H . 1 1.5784α = , 2 1.1859α = and

2.5208tα = .

11

2.5208R 1.5971

1.5784tα

α= = = , 2

2

2.5208R 2.1256

1.1859tα

α= = = ,

1 2R R< .

In the base station receiver, we use spread spectrum protograph LDPC codes to extract mobile terminal 1 signal under three conditions (equal gain combining signals, the first antenna signals, the second antenna signals). The BER performance comparison of three conditions in uplink communication system is show in Fig.8.

compare BER performance under three conditions (equal

gain combining signals, the first antenna signals, the second antenna signals) in uplink communication system

According to BER performance in Fig.8, we observe that equal gain combining signals obtained about 3dB and 4 dB

gains comparing with the first antenna signals and the second antenna signals in Rayleigh flat-fading channel, respectively. It verifies that the equal gain combining can increase information transmission reliability.

The BER performance comparison of the method 1, 2and 3 in uplink communication system is show in Fig.9.

(a) Base station extracts mobile terminal 1 signal

(b) Base station extracts mobile terminal 2 signal compare BER performance of method 1, 2and 3 in uplink

communication system According to BER performance in Fig.9 (a), we observe

that method 1 obtained about 4dB and 9dB gain comparing with method 2 and 3 in Rayleigh flat-fading channel, respectively. According to BER performance in Fig.9 (b), we observe that method 1 obtained about 4dB and 10dB coding gain comparing with method 2 and 3 in Rayleigh flat-fading channel, respectively. It verifies the validity of the proposed approach; it can effectively cancel electronic interference from enemy and channel noises interference.

Compare Fig.9 (a) and (b), we observe that method 1,2 and 3 in Fig.9 (a) obtained about 1dB, 1dB and 2dB gain comparing with method 1,2 and 3 in Fig.9 (b) , respectively.

And the value of 1R in Fig.9 (a) is smaller than 2R in Fig.9 (b) .We can see that, the smaller value of R , the better BER performance in uplink communication system.


110

downlink communication system The channel transfer matrix from base station to mobile

terminal 1 is 1

0.3503 0.2114

0.2459 0.3208

⎡ ⎤= ⎢ ⎥⎣ ⎦

H ; The channel transfer

matrix from base station to mobile terminal 2 is

2

0.3306 0.2274

0.2523 0.3463

⎡ ⎤= ⎢ ⎥⎣ ⎦


interference source to mobile terminal 1 is

1

0.9201 0.3746

0.3314 0.8947t

⎡ ⎤= ⎢ ⎥⎣ ⎦


interference source to mobile terminal 2 is

1

0.9501 0.4046

0.3714 0.9347t

⎡ ⎤= ⎢ ⎥⎣ ⎦

H .

The BER performance comparison of the method 1, 2and 3 in downlink communication system is show in Fig.10.

(a) Mobile terminal 1 extracts cognitive base station signal

(b) Mobile terminal 2 extracts cognitive base station signal compare BER performance of method 1, 2and 3 in

downlink communication system According to BER performance in Fig.10 (a), we observe

that method 1 obtained about 4dB and 10dB coding gain comparing with method 2 and 3 in Rayleigh flat-fading channel, respectively. According to BER performance in Fig.10 (b), we observe that method 1 obtained about 3dB and 8dB coding gain comparing with method 2 and 3 in Rayleigh flat-fading channel, respectively. And the method 2 have error floor

between -13dB and -11dB. It verifies the validity of the proposed approach; it can effectively cancel electronic interference from enemy and channel noises interference.

V. CONCLUSIONS

In this paper, we propose a method that electronic interference cancellation based on spread spectrum LDPC codes in MIMO cognitive system. Protograph LDPC codes and spread spectrum are applied to MIMO cognitive system in the proposed scheme, the information transmission can be of the anti-interference ability and error correcting capability through channel coding and spread spectrum mapping, which can cancel electronic interference and channel noises interference. This method can make multiple CUs secure communication under the condition of strong electronic interference from PU. The simulation results in Rayleigh flat-fading channel show that, comparing with MIMO cognitive system without the proposed method, the proposed approach can effectively cancel electronic interference from enemy and channel noises interference and obtain about 9dB gain.

REFERENCES

[1] S. Haykin. Cognitive radio: brain-empowered wireless

communications [J]. IEEE J. Sel. Areas Commun, 2005,

23(2):201-220.

[2] Hou Y. Thomas, Shi Yi, Sherali Hanif D. Spectrum sharing for

multi-hop networking with cognitive radios [J]. IEEE J. Sel.

Areas Commun, 2008, 26(1):146-155.

[3] H. Islam, Y. Liang, A. Hoang. Joint power control and

beamforming for cognitive radio networks [J]. IEEE Trans. on

Wireless Communications, 2008, 7(7):2415-2419.

[4] Ma Jun Li, Geoffrey Ye, Juang Biing Hwang. Signals

processing in cognitive radio[J]. Proceedings of the IEEE, 2009,

97(5):805-823.

[5] Mitola, Joseph. Cognitive Radio Architecture Evolution[J].

Proceedings of the IEEE, 2009, 97(4):626-641.

[6] Yang Xiao, Kiseon Kim, Guangzhi Qu. A Cognitive Spatial

Multiplexing Scheme for MIMO-CDMA Networks[C]. Proc. of

Conference on Wireless, Mobile and Multimedia Networks,

Beijing, China, 2010: 147-150.

[7] Yang Xiao,Yingkang Zhang, Guangzhi Qu, et al. Spatial

Multiplexing Algorithms of Cognitive Base-Stations[C]. Proc.

of Conference on Wireless, Mobile and Multimedia Networks,

Beijing, China, 2010: 221-224.

[8] Gao, Cunhao, Shi, Yi, Hou, Y. Thomas, et al. On the

Throughput of MIMO-Empowered Multihop Cognitive Radio

Networks [J]. IEEE Trans. on mobile computing, 2011,

10(11):1505-1519. [9] Y. Liang, H. V. Poor. Multiple access channels with confidential

messages[J]. IEEE Trans. Inf. Theory, 2008,54(3): 976–1002. [10] R. Liu, I. Maric, R. Yates, et al. The discrete memoryless

multiple access channel with confidential messages [C]. Proc. of


111

IEEE Int. Symp. Information Theory, Seattle, USA, 2006: 957–961.

[11] E. Tekin, A. Yener. The Gaussian multiple access wire-tap channel [J]. IEEE Trans. Inf. Theory, 2008, 54(12): 5747–5755.

[12] E. Ekrem, S. Ulukus. Effects of cooperation on the secrecy of multiple access channels with generalized feedback [C]. Proc. of Conference on Information Sciences and Systems, Princeton, USA, 2008:791–796.

[13] R. Bassily, S. Ulukus. A new achievable ergodic secrecy rate region for the fading multiple access wiretap channel [C]. Proc. of Conference on Communications, Control and Computing, Monticello, USA, 2009: 819–826.

[14] O. Simeone, A. Yener. The cognitive multiple access wire-tap channel [C]. Proc. of Conference on Information Sciences and Systems, Baltimore, USA, 2009:158–16.

[15] Y. Liang, H. V. Poor, S. Shamai. Secure communication over fading channels [J]. IEEE Trans. Inf. Theory, 2008, 54(6): 2470–2492.

[16] P. Gopala, L. Lai, H. El Gamal. On the secrecy capacity of fading channels [J]. IEEE Trans. Inf. Theory, 2008, 54(10): 4687–4698.

[17] X. Tang, R. Liu, P. Spasojevic, H. V. Poor. On the throughput of secure hybrid-ARQ protocols for Gaussian block-fading channels [J]. IEEE Trans. Inf. Theory, 2009, 55(4): 1575–1591.

[18] Ruoheng Liu, Yingbin Liang, H.V. Poor. Fading Cognitive Multiple-Access Channels With Confidential Messages [J]. IEEE Trans. Inf. Theory, 2011, 57(8): 4992–5005.

[19] X. He, A. Yener. Cooperation with an untrusted relay: A secrecy perspective [J]. IEEE Trans. Inf. Theory, 2010, 56(8): 3807–3827.

[20] E. Ekrem, S. Ulukus. Secrecy in cooperative relay broadcast channels [J]. IEEE Trans. Inf. Theory, 2011, 57(1): 137–155.

[21] J. Thorpe. Low density parity check (LDPC) codes constructed

from protographs [C]. JPL Workshop on Interplanetary Network,

USA, 2003:42-154.

[22] A. Abbasfar, D. Divsalar, K. Yao. Accumulate-repeat-

accumulate codes [J]. IEEE Trans. on Communications, 2007,

55(4): 692-702.

[23] Y. Xiao, K. Kim. Good encodable irregular quasi-cyclic LDPC

codes [C]. Proc. of Conference on Communication Systems,

Singapore, 2008, pp.1291-1296.

[24] Kaiyao Wang, Shaohai Hu, Yang Xiao, et al. Construction of

protograph LDPC codes based on Jacket matrices [C]. Proc. of

Conference on Signals Processing, Beijing, China, 2010:1604-

1607.

[25] Kaiyao Wang, Yang Xiao, Kiseon Kim. Construction of

protograph LDPC codes with circular generator matrices [J].

Journal of Systems Engineering and Electronics, 2011, 22(5):

840-847.

[26] CCSDS131.1-O-2. Low density parity check codes for use in

near-Earth and deep space applications [S].USA: CCSDS, 2007.

[27] R. M. Tanner, D. Sridhara, A. Sridharan, et al. LDPC block and

convolutional codes based on circulant matrices [J]. IEEE Trans.

on Inforation Theory, 2004,50(12):2966-2984.

[28] Y. Xiao, M. H. Lee. Low complexity MIMO-LDPC CDMA

systems over multipath channels [J]. IEICE Trans. on

Communications, 2006, E89-B(5) :1713- 1717.


112

Linear Precoding for the Downlink of Cognitive Radio Network

Pengpeng Lan, Yang Xiao, Jinfeng Kou

Institute of Information Science Beijing Jiaotong University Beijing 100044,P.R.China

Emails: [email protected], [email protected], [email protected]

Abstract—In this paper, a multiple-input–single-output (MISO) overlay cognitive radio (CR) network is considered in which a primary base station (PBS) and a cognitive base station (CBS) share the same spectrum to transmit the information. We consider a novel linear precoding technique to cannel the interferences which is bring about by the CBS’s transmission. Three theorems about the design of the precoding and decoding matrices are put forward. With the linear precoding and decoding, the BER performance of the PBS and CBS are improved significantly. At the same time, the capacity of the PBS and CBS channels are analyzed. Keywords—multiple-input–single-output(MISO);overlay; Cognitive radio (CR); precodin; decoding

I. INTRODUCTION The use of cognitive radio (CR) has been recognized as a

promising solution to improve the utilization efficiency of the current severely underutilized spectrum [1-3]. By allowing unlicensed secondary users (SUs) to coexist either opportunistically or concurrently with the licensed primary users (PUs), the shortage of wireless spectrum resources is eased.

In the last decade, research on CR has focused on three main spectrum sharing paradigms: interweave underlay, and overlay [4]. Each of them requires a different level of cognition about the surrounding environment and a different level of sophistication which leads to different challenges. In interweave CR, CR uses advanced spectrum-sensing techniques to detect spectrum holes (opportunities) in the PU’s licensed band and transmits its own signals in the detected free bands. It must stop transmission as soon as the PU returns to the used bands so as not to cause any interference to the PU. In underlay CR, the CR operates under the constrain that the interference to the primary receiver must not exceed a given level [5]. On the other hand, the overlay CR can learn the primary message by overhearing primary ARQ transmissions and utilize knowledge of the message to perform spectrum overlay during retransmissions of the primary system[6]-[9]. As an alternative, [10] and [11] consider explicit cooperation of the primary and secondary systems.

In this paper, we consider a MISO overlay CR network where the PBS and CBS simultaneously transmit in the same

spectrum to 2 PUs and SUs. We consider the communication scenario where PBS and CBS transmissions cause interference to each other. Such a scenario is justified in situations when the CBS is transmitting in the service area range of the PBS, thereby causing interference to each other [12]-[18].

As traditional in the overlay CR, it is assumed that the CBS has noncausal knowledge of the primary data vector [16-18]. This assumption is well documented in literature for the overlay CR model [6]–[8]. In this paper, we develop a scheme that the PBS and the CBS use linear precoding to cancel the interference to PUs and SUs to make PBS and the CBS sharing a same channel. At the same time, we have proposed three theorems about the construct of the precoding and decoding matrices. BER performance and capacity of the channels are analyzed.

The rest of the paper is organized as follows. In Section II, we describe the system model of the overlay MISO CR system; give the three theorems and relevant demonstrations. The condition of simulation and result are given in Sections III. Finally, conclusion is drawn in Section VI.

The following notations are used in this paper: Matrices and vectors are denoted using boldface upper and lower-case letters, respectively. Superscripts 1( )−i , ( )Ti ,and ( )Hi stand for the inverse operation, the transpose, and the conjugate transpose, respectively. I and 0 represent the identity matrix and zero matrix, respectively. x y× denotes the space of ( )x y× matrix with complex entries. ( )20,σCN denotes

the complex normal distribution with mean μ and variance 2σ .

II. SYSTEM MODEL AND PROBLEM FORMULATION

The system model we are interested in is shown in Fig. 1. The primary network has 3 terminals :one PBS uses 2 antennas and 2 PUs are equipped with a single antenna each. The secondary network consists of one CBS uses 2 antennas, 2 CUs with a single antenna each. The response of the primary downlink channel, the secondary downlink channel and interference channels between the PBS to CUs and the CBS to PUs are denoted by ppH , ccH , pcH , and cpH

respectively, where 2 2, , ,pp cc pc cp×∈H H H H . The CSI of all ___________________

This work was supported by the Beijing Natural Science Foundation of China: (No. 4102050).


113

links is assumed to be perfectly known at the CBS, whereas the PBS only knows its own downlink channel. Such assumption can be justified as follows: knowledge of ppH is obtained by extracting the

CSI feedback from the PUs to the PBS [12]; the CSI of ccH is estimated at the CUs by decoding the CBS’s pilot signals and feeding back it to the CBS; the CSI of pcH is obtained by the CUs by

synchronizing to primary’s pilot signals in every frame and then feeding back this information to the CBS [15]. cpH is obtained by

listening to PU transmission to the PBS and assuming channel reciprocity due to time-division duplex mode.

The information symbol vector corresponding to PUs can be denoted as ,1 ,2[ ]T

p p cu u=u and CUs as ,1 ,2[ ]Tc c cu u=u .

ps and cs represent the transmitted symbol vectors from the PBS and

CBS after precoding. pP and cP are the total transmitted powers for

primary and cognitive signals, respectively. For simplicity, we assume equal power allocation for PUs and

CUs. ,1 ,2pp p pdiag P P=P and ,1 ,2cc c cdiag P P=P

denote the power loadings for the PUs and CUs signals, respectively.

p pp p=u P u and c cc c=u P u [14]are the primary and secondary

symbol vectors after power loading, respectively.

Fig.1. the system model of the CR network

In this paper, we consider linear precoding at the PBS and CBS.

pT and cT are precoding matrices at PBS and CBS. β is the

average transmitted power normalization factor. Thus, the transmitted precoded symbol vectors at the PBS and CBS can be expressed as:

p pp β=

T us (1)

c cc β=

T us (2)

The received symbol vectors at the PUs and CUs is given by: p pp p cp c p= + +y H s H s n (3)

c pc p cc c c= + +y H s H s n (4)

Where 2(0, )CNp pσn I∼ ∼ and 2(0, )CNc cσn I∼ ∼ are

the additive white Gaussian noise vectors at the PUs and SUs,. To simplify the analysis, we let:

, , ,pp cp p p

p cpc cc c c

⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤= = = =⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦⎣ ⎦

H H n yH H n y

H H n y (5)

Thus, (3) and (4) can be rewritten as:

p p c c

p p p c c c

β β

= + +

= + +

y H s H s n

H T u H T un

(6)

To detect pu and cu from (6), we need to design decoding

matrices pU and cU as well as the precoding matrices pT and cT ,

these matrices need satisfy the following theorem. Theorem 1: The received symbol vectors py and cy at the PUs and

SUs are formulated in (3) and (4), where the transmitted precoded symbol vectors ps and cs at the PBS and CBS are given by (1) and

(2).We can separate pu and cu from (6) if and only if

p c c c p p= =U H T U H T 0 (7)

Proof: Necessity is obvious, we only prove the sufficiency. If the precoding matrices pT and cT as well as decoding

matrices pU and cU satisfy (7), multiplying pU to (6), we have

p p p p p c c cp p

p pp

β β

β

= + +

= +

U H T u U H T uU y U n

V uU n

(8)

where

p p p p=V U H T (9)

Similarly, multiplying cU to (6), we have

c p p p c c c cc c

c cc

β β

β

= + +

= +

U H T u U H T uU y U n

V uU n

(10)

where

c c c c=V U H T (11)

Further, by multiplying 1p−V and 1

c−V to (8) and (10), respectively,

we get

1 1pp p p pβ− −= +

IuV U y V U n (12)

1 1cc c c cβ− −= +

IuV U y V U n (13)

From Theorem 1, we can obtain some algorithms to construct the precoding matrices pT and cT as well as decoding

matrices pU and cU . One of the algorithms is as following theorem.

Interference channel

ppH p pT u

Transmitting Base-stations Receiving users

PBS

CBS c cT u

pcH

cpH

ccH

PU1

Communication channel

py

cy

PU2

CU1

CU2


114

Theorem 2: If p cc cp⎡ ⎤= −⎣ ⎦U H H , c pc pp⎡ ⎤= −⎣ ⎦U H H , 1( )p p p

−=T U H , 1( )c c c−=T U H , and

cc cp cp cc=H H H H (14)

pc pp pp pc=H H H H (15)

then

pp pβ

= +u

U y U n (16)

cc cβ

= +u

U y U n (17)

Proof: If p cc cp⎡ ⎤= −⎣ ⎦U H H , c pc pp⎡ ⎤= −⎣ ⎦U H H , due to (7),

we have cp

p c cc cp cc cp cp cc

cc

⎡ ⎤⎡ ⎤= − = −⎢ ⎥⎣ ⎦

⎣ ⎦

HU H H H H H H H

H (18)

pp

c p pc pp pc pp pp pcpc

⎡ ⎤⎡ ⎤= − = −⎢ ⎥⎣ ⎦⎢ ⎥⎣ ⎦

HU H H H H H H H

H (19)

Due to (14) and (15), (18) and (19) can be simplified into

cp

p c cc cp

cc

⎡ ⎤⎡ ⎤= − =⎢ ⎥⎣ ⎦

⎣ ⎦

HU H H H 0

H (20)

pp

c p pc pppc

⎡ ⎤⎡ ⎤= − =⎢ ⎥⎣ ⎦ ⎢ ⎥⎣ ⎦

HU H H H 0

H (21)

Then, (8) and (10) can be changed into:

p p p pp pβ

= +U H T u

U y U n (22)

c p p pc cβ

= +U H T u

U y U n (23)

Substitute 1( )p p p−=T U H and 1( )c c c

−=T U H into (22) and

(23), we will have the results in (16) and (17). In applications, the channel conditions (14) and (15) of Theorem

2 are very strict. PBS and CBS need to select the PUs and CUs to satisfy the channel conditions to ensure the detections of pu and

cu from (6). To solve the problem, we have following theorem to

construct pT , cT , pU and cU , which can remove the limitations of

(14) and (15) of Theorem 2.

Theorem 3: If 1 1p cp cc

− −⎡ ⎤= −⎣ ⎦U H H , 1 1c pp pc

− −⎡ ⎤= −⎣ ⎦U H H , 1( )p p p

−=T U H , 1( )c c c−=T U H , then pu and cu can be

obtained from (16) and (17).

Proof: If 1 1p cp cc

− −⎡ ⎤= −⎣ ⎦U H H , 1 1c pp pc

− −⎡ ⎤= −⎣ ⎦U H H , due to

(7), we have

1 1 1 1cp

p c cp cc cp cp cc cc

cc

− − − −⎡ ⎤⎡ ⎤= − = −⎢ ⎥⎣ ⎦

⎣ ⎦= − =

HU H H H H H H H

H

I I 0

(24)

1 1 1 1pp

c p pp pc pp pp pc pcpc

− − − −⎡ ⎤

⎡ ⎤= − = −⎢ ⎥⎣ ⎦⎢ ⎥⎣ ⎦= − =

HU H H H H H H H

H

I I 0

(25)

From Theorem 1, (8) and (10) are changed into (22) and (23).

Substitute 1 1p cp cc

− −⎡ ⎤= −⎣ ⎦U H H and 1 1c pp pc

− −⎡ ⎤= −⎣ ⎦U H H into 1( )p p p

−=T U H and 1( )c c c−=T U H , we can obtain pT and cT ,

then from (22) and (23), we will have the results in (16) and (17).

III. SIMULATION RESULTS From Theorem 3, we can get the decoding matrices

1 1p cp cc

− −⎡ ⎤= −⎣ ⎦U H H and 1 1c pp pc

− −⎡ ⎤= −⎣ ⎦U H H , precoding

matrices 1( )p p p−=T U H and 1( )c c c

−=T U H .

The decoding results are obtained from (16) and (17),

pp pβ

= +u

U y U n ,

cc cβ

= +u

U y U n ,

where β is the average transmitted power normalization factor. According to theorem 3, we design the the precoding and decoding

matrices in the simulations. Then, we analysize the BER performance of PBS and CBS in the system model, capacity of the primary user and the cognitive user channel are given too.

A. BER simulation

Conditions of simulation experiment: BPSK modulation, Rayleigh flat-fading channel, code rate 0.75, 10000 data frames per SNR point.

Assume the cognitive downlink channel ccH is

0.95 0.210.37 0.85cc

⎡ ⎤= ⎢ ⎥⎣ ⎦

H ,it is estimated at the CUs by decoding the

CBS’s pilot signals and feeding back the estimate of CSI to the CBS. The knowledge of pcH is obtained at the CUs by synchronizing to primary’s pilot signals in every frame and then feeding back this

information to the CBS. And let 0.4 0.380.21 0.54pc

⎡ ⎤= ⎢ ⎥⎣ ⎦

H . cpH It is

obtained by listening to PU transmission to the PBS and assuming channel reciprocity due to time-division duplex mode. Knowledge of

ppH is obtained by extracting the CSI feedback from PUs to PBS.

Each of them are 0.62 0.310.24 0.53cp

⎡ ⎤= ⎢ ⎥⎣ ⎦

H and0.9 0.30.4 0.73pp

⎡ ⎤= ⎢ ⎥⎣ ⎦

H .


115

Fig.3 the BER performance of the PBS and CBS with linear

precoding and without

From the Fig.3, we can see that with the increase of the SNR, the BER performance of the PBS and CBS with linear precoding is decrease significantly.

B. Channel capacity

We compute the capacity by utilizing a pre-whitened channel matrix in order to make the interference-plus-noise appear white. Apply the proposed scheme, due to Theorem 2, there will be no interference for the sharing channel of the PBS and CBS. Thus, the capacity for PBS and CBS can be computed by ppH and ccH ,

2log (det( ))Tpp pp ppC I SNR= + ⋅ H H (26)

2log (det( ))Tcc cc ccC I SNR= + ⋅ H H (27)

The following results of Fig. 4 and Fig.5 are based on the calculations of (26) and (27).

Conditions of simulation experiment: Using the Monte Carlo method, the simulation results are the average of 10,000 channels.

Fig.4 the capacity of the primary user channel

Though the BER performance of the primary user and the cognitive user with linear precoding is different. The capacity of the primary user channel ppH and the cognitive channel ccH are very close.

Fig.5 the capacity of the cognitive user channel

IV.CONCLUSIONS

In this paper, we have considered spectrum sharing in a CR MISO downlink network, where the PBS and CBS transmit to a number of PUs and SUs, respectively. We have proposed the linear precoding and decoding to exploit the interferences the CBS to the PBS and given the method to design the linear precoding and decoding matrices. The simulation result shows the efficiency of the proposed scheme.

REFERENCES [1] Federal Communications Commission, “Spectrum policy task force,”

Rep. ET Docket No. 02-135, Nov. 2002. [2] Y. Xiao, Y. K. Zhang, Qu Guangzhi, Kim Kiseon, “Spatial multiplexing

algorithms of cognitive base-stations”, IET 3rd International Conference on Wireless, Mobile and Multimedia Networks (ICWMNN 2010), 2010, pp.221-224.

[3] Y. L. An, Y. Xiao, “Multi-users cooperative transmitting algorithm in cognitive WLAN”, Journal of Networks, 2012, 7(8):1164-1169.

[4] A. Goldsmith, S.A. Jafar, I. Maric, and S. Srinivasa, “Breaking spectrum gridlock with cognitive radios: An information theoretic perspective,” Proc.IEEE, vol. 97, no. 5, pp. 894–914, May 2009.

[5] R. Zhang, Y.-C. Liang, and S. Cui, “Dynamic resource allocation in cognitive radio networks,” IEEE Signal Processing Magazine, vol. 27,no. 3, pp. 102–114, May 2010.

[6] N. Devroye, P. Mitran, and V. Tarokh, “Achievable rates in cognitiveradio channels,” IEEE Trans. Inf. Theory, vol. 52, no. 5, pp. 1813–1827,May 2006.

[7] A. Jovicic and P. Viswanath, “Cognitive radio: An information theoretic perspective,” IEEE Trans. Inf. Theory, vol. 55, no. 9, pp. 3945–3958,Sep. 2009.

[8] J. Jiang and Y. Xin, “On the achievable rate regions for interference channels with degraded message sets,” IEEE Trans. Inf. Theory, vol. 54,no. 10, pp. 4707–4712, Oct. 2008.

[9] S. H. Seyedmehdi, Y. Xin, and Y. Lian, “An achievable rate region for the causal cognitive radio,” in Proc. 45th Annu. Allerton Conf. Commun.,Control Comput., Monticello, IL, Sep. 2007, pp. 783–790.

[10] Y. Han, A. Pandharipande, and S. Ting, “Cooperative decode-and-forward relaying for secondary spectrum access,” IEEE Transactions on Wireless Communications, vol. 8, no. 10, pp. 4945–4950, Oct. 2009.

[11] Y. Li, P. Wang, and D. Niyato, “Optimal power allocation for secondary users in cognitive relay networks,” in Proc. IEEE Wireless Communications and Networking Conf. (WCNC), Mar. 2011, pp. 862–867.

[12] C. Masouros and T. Ratnarajah, “Utilization of primary-secondary crossinterference via adaptive precoding in cognitive relay assisted


116

MIMO wireless systems,” in Proc. IEEE ICC, Kyoto, Japan, Jun. 5–9, 2011,pp. 1–5.

[13] C. Masouros and T. Ratnarajah, “Interference as a source of green signal power in cognitive relay assisted co-existing MIMO wireless transmissions,”IEEE Trans.Commun., vol. 60, no. 2, pp. 525–536, Feb. 2012.

[14] A. Jovicic and P. Viswanath, “Cognitive radio: An information theoretic perspective,” IEEE Trans. Inf. Theory, vol. 55, no. 9, pp. 3945–3958,Sep. 2009.

[15] F. Negro, I. Ghauri, and D. T. M. Slock, “Transmission techniques and channel estimation for spatial interweave TDD cognitive radio systems,”in Proc. 43rd Asilomar Conf. Signals Syst. Comput., Pacific Grove, CA,Nov. 1–4, 2009, pp. 523–527.

[16] Y. L. An, Y. Xiao, G. Qu, “Multi-source cooperative MIMO cognitive network based on distributed interference alignment”, 4th IET International Conference on Wireless, Mobile & Multimedia Networks (ICWMMN 2011), 2011, pp.160-164.

[17] Y. F. Li, Y. Xiao, M. H. Lee, “MIMO interference alignment for cognitive radio network, 4th IET International Conference on Wireless”, Mobile & Multimedia Networks (ICWMMN 2011), 2011, pp.127-130.

[18] Y. Y. Li, Y. Xiao, M. H. Lee, “Decoding algorithm and BER performance of interference alignment and cancellation system in MIMO networks”, 4th IET International Conference on Wireless”, Mobile & Multimedia Networks (ICWMMN 2011), 2011, pp.131-136.


117

Cooperative Spatial Multiplexing for CR Users Sharing a Common Channel with Primary Users*

Yang Xiao, Jinfeng Kou

Institute of Information Science Beijing Jiaotong University Beijing 100044,P.R.China

Emails: [email protected], [email protected]

Abstract—To enable mobile stations (MSs) of secondary users (SUs) in cognitive radio (CR) networks sharing a common channel with MSs of primary users (PUs) is an important but challenging issue. In this paper, a cooperative spatial multiplexing (CSM) scheme is proposed, where base station (BS) in CR network has K antennas and each SU or PU has one antenna only, and BS supports PUs and SUs by K spatial channels. To ensure communication quality of the network, the paper applies LDPC as the channel coding for BS, PUs and SUs. Simulations verify the proposed scheme with good spatial multiplexing capacity and BER performance.

Index Terms—wireless communication, cognitive MIMO, spatial multiplexing, encoding and decoding algorithms

I. INTRODUCTION The demand for high data rates in wireless networks

significantly increased over the last few years and stimulated the interest in efficient utilization of so called multiple-input multiple-output systems (MIMO) [1-19]. By utilizing multiple antenna architectures at both BS and MSs of CR network, significantly higher channel capacities close to that of wired networks are possible [1, 2, 16-19]. While, there are two problems need to be solved: first one is co-channel interference for SUs’ and PUs’ in a network, since SUs’ and PUs’ channels are not sure to be orthogonal each other; second one is MSs implementation complexity [16-19]. In the uplink, sophisticated multiuser detection (MUD) can be implemented at the BSs to counter the co-channel interference, however in the downlink, the MS receivers are usually required to be simple. In particular, in broadcast scenarios (downlink situation), where no joint processing of the received signals is possible, MIMO precoding schemes in [4, 5] are attractive since they can be viewed as the counterpart to successive cancellation which is applicable in uplink scenarios only.

Different from classical MIMO, the existing cooperative spatial multiplexing (CSM) schemes for MIMO network [10-15] need several base stations (BSs) and mobile stations (MSs) to support one data transmitting. However, there are four vital problems for this scheme to hinder CSM’s

application. The first problem is the exact synchronization among the MSs and BSs; the second one is the spatial channels’ estimation for BSs and MSs [20]. The third one is space-time coding and decoding for the BSs and MSs, every one of the MSs and BSs needs to know the spatial channels’ parameters [16, 20]; and the fourth one is the complexity of the implementation systems and MIMO network. That is to say, we can not follow the existing CSM schemes [10-15], and had to propose new CSM solutions. The proposed CSM systems including the MIMO base stations (BS) and MIMO MS are designed, while we can not apply the existing CSM schemes [10-15] simply because of the above problems.

The proposed cooperative spatial multiplexing systems in this paper are different from the existing CSM schemes [10-15]. First, in our CSM, one CR network has one BS and many MSs of (PUs and SUs), BS has K antennas, while each MS only has one antenna, MSs have not used MIMO technique: no space-time coding and decoding. Second, in the proposed scheme, the CSM provides the BS of K antennas to support K MSs by one frequency channel only, the existing CSM schemes have not considered our approach. Third, the proposed CSM scheme solves the co-channel interference in CR network by LDPC [21-25], but the existing CSM schemes have no ability to solve the problem. Fourth, there is the synchronization problem among the MSs and BSs of the existing CSM schemes, one BS can support K MSs with independent spatial channels. Fifth, contrary to the existing CSM, in the proposed CSM, MSs need not to estimate the spatial channels’ parameters, which make their implementation simplified greatly.

The following contents are organized as following. The proposed CSM downlink and CSM uplink schemes based on MIMO-CR network are given in Section II and Section III, respectively. In Section IV, we give the simulation results of a MIMO-CR network. Finally, Section V concludes the paper.

II. THE PROPOSED CSM DOWNLINK OF MIMO-CR NETWORK Consider the downlink channel of MIMO-CR network

shown in Fig. 1. In the proposed MIMO CR systems, the BS

_____________________________________________This work was supported by the Beijing Natural Science Foundation of China: (No. 4102050).


118

in CR network uses K antennas to transmit and receive the MSs’ data stream of real symbols. From Fig.1, we see that the K downlink paralleled sequences for K MSs are transmitted from K antennas of BS transmitter. Since BS has K antennas, BS can divide K MSs into SU group and PU group who share the same frequency bandwidth in CR network.

In Fig.1, to keep the MIMO channels be independent, the distance of two MSs is lager than 10 radio signal wavelengths, and the distance of two receiving antennas of BS is lager than 4 radio signal wavelengths, else the channel matrix may be singular.

The proposed CSM scheme in Fig. 1 can provide K spatial channels to support K MSs of PUs and SUs at one frequency band-width, while there is one PU only for existing CR network only at one frequency band-width, and all SUs had to back-off.

AntK

MS1

Ant1

Base station Transmitter

( )1ks

( )kKs

Ant1

AntK

( )kKr

( )1

kr

h11

h1K

hK1

hKK

MSK

Fig. 1 The downlink channel of MIMO-CR network

We consider using MIMO precoding at the transmitter MIMO-CR BS, the purpose is to simplify the receiver implementation of MSs of PUs and SUs. The implementation complexity of proposed MIMO-CR systems is moved to BSs: downlink MIMO ZF precoding, so there is no space-time coding and decoding for MSs of PUs and SUs.

The proposed BS transmitter structure supports spatial multiplexing scheme for downlink transmission, and it is shown as in Fig. 2. In the proposed MIMO-CR network, MSs need not estimate the MIMO channel parameters, and they only need to provide BS an uplink pilot for k th time-slot.

AntK

( )1

ks

( )kKs

Ant1

( )1

ku

( )kKu

Precoding

by 1

( )k−H

LDPC ENC

LDPC ENC

Channel Estimation

( )1

kb

( )kKb

Fig. 2 MIMO precoding transmitter of BS for MSs

Fig. 2 only provides base-band processing blocks of the BS transmitter. Considering K MSs, the BS transmitter include K processing blocks as that in Fig.2. BS can determine the spatial multiplexing number by the number of PUs and MUs.

At downlink, firstly, BS’s LDPC codec encodes the MSs data streams ( )

1ku , …, ( )k

Ku of k th time-slot into K

streams )(1

kb ,…, ( )kKb for MS1 ,…, MSK.

Secondly, the LDPC streams )(1

kb , …, ( )kKb of MSs

are precoded by inversed channel matrix as a linear combination of the K sub-streams ( )

1ks ,…, ( )k

Ks ,

( )

( )

( ) ( )

( ) ( )

( )

( )

1

1 11 1 1

1

k k k kK

k k k kK K KK K

s h h b

s h h b

−⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎢ ⎥ ⎢ ⎥ ⎢ ⎥

=⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦ ⎣ ⎦

…

…

(1)

where ( ) , , 1,..., kijh i j K∈ denote the channel coefficient

that the path from the transmitter antenna i to the MS’s antenna j experiences independent Rayleigh fading, shown in Fig.1, and we let

( ) ( )

( ) ( )

11 1

( )

1

k kK

k

k kK KK

h h

h h

⎡ ⎤⎢ ⎥

= ⎢ ⎥⎢ ⎥⎣ ⎦

H

…

…

(2)

In the proposed BS system, the channel parameters of 1

( )k−H are provided by the channel estimation model of BS,

shown in Fig.2. BS in the proposed scheme accesses the MSs according

to the spatial channels to be independent for the K MSs, which requires ( )kH should be full rank. This is always

possible since the number of active MSs at different positions in CR network is always larger than K. The BS only does not select the MSs to be near each other [16].

In the proposed scheme, we need not require the MSs to estimate the channel parameters, since it will greatly increase the implementation complexity of MSs. The channel estimation is arranged on BS utilizing the approach of [9].

The received signals at the antennas of MSs are given by

( )

( )

( ) ( )

( ) ( )

( )

( )

( )

( )

1 11 1 1 1

1

k k k k kK

k k k k kK K KK K K

r h h s n

r h h s n

⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥

= +⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦

…

…

(3)


that the path from the transmitter antenna i of BS to the receiver antenna of MS j experiences independent Rayleigh fading. The receiver noises ( ) , 1,..., k

jn j K∈ of the


119

antenna j are normalized to be )1,0(CN (zero-mean unit-variance complex Gaussian).

Since the precoding at BS is adopted, the received signals ( ) , 1,...,k

ir i K= , from the MSs’ antennas, are independent, and can be further written as

( )

( )

( ) ( )

( ) ( )

( ) ( )

( ) ( )

( )

( )

( )

( )

( )

( )

( )

( )

1

1 11 1 11 1 1 1

1 1

1 1

k k k k k k kK K

k k k k k k kK K KK K KK K K

k k

k kK K

r h h h h b n

r h h h h b n

b n

b n

−⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥

= +⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦

⎡ ⎤ ⎡ ⎤⎢ ⎥ ⎢ ⎥

= +⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦

… …

… …

(4) From (4), the proposed MS receiver can be very simple, whose base-band circuit is shown in Fig.3. The soft information ( )k

ir needs to be decoded by LDPC decoder in Fig.3 [21-25].

Anti

( )kiu

( )kir

LDPC DEC

Fig. 3 The MS receivers of MIMO CR network

The proposed CSM downlink scheme of MIMO CR

system, the MSs need not estimate the spatial channel parameters, the base station of CR network provides K spatial channels for K MSs. This is one of main contributions of the proposed CSM scheme. However, in the downlink of wireless network, the existing CSM receivers need to joint-recover the transmitted data streams from BS by channel matrices, and MSs had to estimate Rayleigh channel parameters by the pilot signals of BS [10].

III. THE PROPOSED CSM UPLINK OF MIMO-CR NETWORK Considering the uplink of the CSM MIMO-CR network

shown as in Fig.4, where there are K antennas at BS but there is one antenna of at each MS, assuming there are K active MSs in the network. The MSs transmit the K streams through its single transmitting antenna through the flat fading channels in the MIMO-CR network. The proposed BS assigns spatial channels to the MSs of PUs and SUs according to their spatial positions, and MSs should not near each other.

The proposed MS transmitter of MIMO-CR network is shown in Fig.5. At the uplink, firstly, MS’s LDPC codec encodes its data streams ( ) , 1,...,k

iu i K= into LDPC

streams ( ) , 1,...,kib i K= [21-25]. Then, the LDPC

streams are transmitted different MSs, and they are received through Rayleigh fading channel by BS.

AntK

MS1

Ant1

Base station Transmitter

( )1

kr

( )kKr

Ant1

AntK

( )kKb

( )1

kb

h11

h1K

hK1

hKK

MSK

Fig. 4 the uplink of the CSM MIMO network

Anti

( )kiu

LDPC ENC

( )kib

Fig.5 The MS transmitter of MIMO-CDMA network Different from the existing CSM schemes for MSs, from

Fig.5, we see that the proposed MS transmitter has no space-time coding. The CSM gain is obtained by the proposed BS MIMO receiver.

The proposed MIMO-CR BS receiver K MSs is shown in Fig. 6. In the proposed CSM scheme, each MS provides an uplink pilot sequence to BS. BS obtains the spatial channel parameters by the MS’s pilot [9], this is different from the other existing CSM scheme.

AntK

( )1

ks

( )kKs

Ant1

( )1

ku

( )kKu

Decoding

by 1

( )k−H

LDPC DEC

LDPC DEC

Channel Estimation

( )1

kb

( )kKb

Fig. 6 MIMO receiver of MIMO-CD BS The received signals at K antennas of BS are given by ( )

( )

( ) ( )

( ) ( )

( )

( )

( )

( )

1 11 1 1 1

1

k k k k kK

k k k k kK K KK K K

r h h b n

r h h b n

⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥

= +⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦

…

…

(5)


that the path from the MS i to the receiver antenna j of BS experiences independent Rayleigh fading. The receiver noise


120

vectors ( ) , 1,...,kjn j K= of the antenna j are normalized to

be )1,0(CN (zero-mean unit-variance complex Gaussian).

From ( ) , 1,...,kir i K= , the MIMO decoding model

recovers ( ) , 1,...,kiz i K= , the LDPC coded streams

with channel noise by using 1( )k−H ,

( )

( )

( ) ( )

( ) ( )

( ) ( )

( ) ( )

( )

( )

( )

( )

( )

( )

( ) ( )

( ) ( )

( )

( )

1

1 11 1 11 1 1 1

1 1

1

1 11 1 1

1

k k k k k k kK K

k k k k k k kK K KK K KK K K

k k k kK

k k k kK K KK K

z h h h h b n

z h h h h b n

b h h n

b h h n

−

−

⎛ ⎞⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎜ ⎟⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥

= +⎜ ⎟⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎜ ⎟⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥

⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦⎝ ⎠

⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎢ ⎥ ⎢ ⎥ ⎢ ⎥

= +⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦ ⎣ ⎦

… …

… …

…

… (6) where ( ) , 1,...,k

iz i K= need to be decoded into ( ) , 1,...,kiu i K= by LDPC decoder of the BS [21-25],

which is shown in Fig.6.

IV. SIMULATIONS We assume that in a MIMO-CR network, there are one BS

and four MSs (one is PU, three are SUs), BS has 4 antennas, and each MS has one antenna only, BS obtains the perfect channel estimation by the approach of [9], and the transmitted signals are modulated by BPSK.

To verify the LDPC role [21-25] in the proposed CSM system of MIMO-CR network to be valid, we simulate the downlink and uplink of MIMO-CR network with LDPC in MIMO-CR network. In following two experiments, we use the same QC LDPC codes in [21], in our system simulation,

127p = , 2a = and 7b = . Thus, the QC code has the code rate 1/2, and code length to be 1270. Experiment 1: BER simulation of the proposed downlink MIMO-CR system with LDPC. BS had to use the proposed CSM to let one frequency band-width to support one PU and three SUs.

In the downlink network’s simulation, the two systems (one with LDPC, the other no) transmit 200 data frames under same SNR and channel conditions, where

0.9 0.31 0.29 0.130.19 0.78 0.4 0.210.24 0.3 0.71 0.330.11 0.17 0.4 0.83

⎡ ⎤⎢ ⎥⎢ ⎥=⎢ ⎥⎢ ⎥⎣ ⎦

H.

The obtained BER results are shown in Fig. 7. Observing simulation results in Fig.7, we find that the PU and SUs have similar BER performance at MIMO-CR downlink, if BS has perfect channel estimation. When SNR is larger than 6, the BER of PU and SUs is smaller than 510− , which is expected by us. From (4), we also know that the precoding algorithm

can complete the spatial separation for K MSs’ data screams. For the MSs, the received data is independent each other, and the multiuser interference has been removed due to (4).

0 2 4 6 8 10 12 14 16

10-5

10-4

10-3

10-2

10-1

100

SNR(dB)

BER

PU

SU1

SU2

SU3

Fig. 7 The bit error rates (BER) of MIMO-CR downlink

Experiment 2: BER simulation of the proposed uplink

MIMO-CR system with LDPC, the simulation parameters are same with Experiment 1. The obtained BER results for PU and SUs are shown in Fig. 8.

Observing simulation results in Fig.8, we find that the PU MS has best BER performance, which in fact is determined by the channel matrix H . When SNR is larger than 9dB, the BER of PU is smaller than 510− , while BER of SU2 is smaller than 510− , SNR needs to be larger than 15dB. The problem comes the second term of (6), where 1−H may enlarge the

noise term

( )

( )

1k

kK

n

n

⎡ ⎤⎢ ⎥⎢ ⎥⎢ ⎥⎣ ⎦

in (6), which is different from (4).

0 2 4 6 8 10 12 14 16

10-5

10-4

10-3

10-2

10-1

100

SNR(dB)

BER

PU

SU1

SU2

SU3

Fig.8 The bit error rates (BER) of MIMO CR uplink

V. CONCLUSIONS In this paper, we proposed a CSM system with inversed

channel matrices precoding and decoding algorithms for MIMO-CR network to enable MSs of PUs and SUs sharing a common channel. The CSM framework and algorithms of downlink and uplink are provided. It is possible to make PUs and SUs work at independent spatial channels in the


121

MIMO-CR downlink and uplink. The simulation results showed that the proposed MIMO-CR network can achieve the good error rate performance in Rayleigh channels. Because MS has no complicated matrix decoding and channel estimation, the implementation of MSs becomes easy and can be accepted in future MIMO-CR network.

REFERENCES [1] I. E. Telatar, “Capacity of multi-antenna Gaussian channels,”

Euro.Trans. Telecom., vol. 10, pp. 585–595, Nov. 1999. [2] G. J. Foschini and M. J. Gans, “On limits of wireless

communications in a fading environment when using multiple antennas,” Wireless Personal Commun., vol. 6, pp. 311–335, 1998.

[3] G. G. Raleigh and J. M. Cioffi, “Spatio-temporal coding for wireless communications,” IEEE Trans. Commun., vol. 46, pp. 357–366, Mar. 1998.

[4] R. F. H Fischer, C. A. Windpassinger, “Improved MIMO precoding for decentralized receivers resembling concepts from lattice reduction”, 2003 IEEE Global Telecommunications Conference, (GLOBECOM '03), 1-5 Dec. 2003, Volume 4:1852 –1856.

[5] Y. Xiao, Y. Zhao and M. H. Lee, “MIMO Precoding of CDMA Systems”, Proc. 8th International Conference on Signal Processing, Beijing, 2006, Vol. 1: pp.397-401.

[6] Y. Xiao and M. H. Lee, “MIMO Multiuser Detection for CDMA Systems”, Proc. 8th International Conference on Signal Processing, Beijing, 2006, Vol. 1:pp.381-384.

[7] Y. Hou, Y. Xiao, “Cancelling Co-Channel Interference for MIMO-OFDM Systems Based on DCT and LDPC”, 2008 IET International Conference on Wireless Mobile & Multimedia Networks, 2008, pp.268-271.

[8] Y. Xiao and M –H Lee, “Low Complexity MIMO-LDPC CDMA Systems over Multipath Channels”, IEICE TRANS. COMMUN., VOL. E89-B, No. 5 May 2006, pp.1713-1717.

[9] Y. Xiao, J. L. Liu, H. J. Yin, K. Kim, MIMO Spatial Multiplexing Systems with Uplink Pilot and LDPC Codec, 5th International Conference on Wireless Communications, Networking and Mobile Computing, 2009. WiCom '09. 24-26 Sept. 2009, pp.1-5.

[10] F. Verde, D. Darsena and A. Scaglione, “Cooperative Randomized MIMO-OFDM Downlink for Multicell Networks: Design and Analysis”, IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 58, NO 1, pp.384-402, 2010.

[11] H. Skjevling, D. Gesbert, “Precoded Distributed Space-Time Block Codes in Cooperative Diversity-Based Downlink”, IEEE Transactions on Wireless Communications, Vol. 6, No. 12, pp. 4209-4212, 2007.

[12] A. Özgür, O. Lévêque and D. N. C. Tse, “Hierarchical Cooperation Achieves Optimal Capacity Scaling in Ad Hoc Networks”, IEEE Transactions on Information Theory, Vol. 53, No. 10, pp.3549-3572, 2007.

[13] O. Simeone, O. Somekh, H. V. Poor, S. Shamai, Distributed MIMO in Multi-Cell Wireless Systems via Finite-Capacity Links, ISCCSM 2008, Malta, 12-14 March 2008, pp.203-206.

[14] Hilde Skjevling, David Gesbert, Are Hjorungnes, “Receiver-Enhanced Cooperative Spatial Multiplexing with Hybrid Channel Knowledge”, Proc. IEEE Conference on Acoustics, Speech, and Signal Processing, ICASSP 2006, (Toulouse, France), IEEE, 14-19 May 2006. IV:65-68

[15] S. H. Lee and S. Y. Chung, “Degrees of Freedom of Cooperative MIMO in Cellular Networks”, Proceedings of IEEE International Conference on Communications (IEEE ICC 2009), 2009, pp.1-5.

[16] Y. Xiao, MIMO Multi-antennas Wireless Communication Systems, Press of Posts and Telecommunications, Beijing, 2009.

[17] Hochwald, T. L. Marzetta, C. B. Papadias, A Transmitter Diversity Scheme for Wideband CDMA Systems Based on Space-Time Spreading, IEEE Journal on Selected Areas In Communications, 2001, 19(1): 48-60.

[18] Y. Xiao, K. S. Kim, G. Z. Qu. A Cognitive Spatial Multiplexing Scheme for MIMO-CDMA Networks. Proc. of Conference on

Wireless, Mobile and Multimedia Networks, Beijing, China, 2010: 147-150.

[19] Y. Xiao, Y. K. Zhang and G. Z. Qu, et al. Spatial Multiplexing Algorithms of Cognitive Base-Stations. Proc. of Conference on Wireless, Mobile and Multimedia Networks, Beijing, China, 2010: 221-224.

[20] W. –J. Chen, Y. Xiao and Y. Zhao, The algorithm implement of WCDMA channel estimation, 7th International Conference on Signal Processing Proceedings, 2004, v3: 1894-1897.

[21] Y. Xiao and M. H. Lee, Construction of good quasi-cyclic LDPC codes, Proceedings of IET International Conference on Wireless Mobile and Multimedia Networks (ICWMMN '06), Hangzhou, China, Nov. 2006, pp.172-176.

[22] Y. Xiao, K. Kim, Alternative good LDPC codes for DVB-S2, Proc. 9th International Conference on Signal Processing (ICSP 2008), Beijing, Oct. 26-29, 2008, pp.1959–1962.

[23] Y. Xiao, K. Kim, Good encodable irregular quasi-cyclic LDPC codes, 11th IEEE Singapore International Conference on Communication Systems (ICCS 2008), 19-21 Nov. 2008, pp.1291-1296.

[24] Y. Xiao, Evaluations of good LDPC codes based on generator matrices, Proc. of 2006 8th International Conference on Signal Processing, Beijing, China, Nov.16-20, 2006. pp. 2187-2190.

[25] Y. Xiao, Turbo and LDPC Codecs and their Applications, Press of Posts and Telecommunications, Beijing, 2010.


122

A Spatial Coding Approach for MIMO Cognitive

Radio Networks’ Bandwidth Sharing

Yanru Qiao, Yang Xiao Institute of Information Science, Beijing Jiaotong University

Beijing 100044, P.R.China Emails: [email protected], [email protected]

Abstract—The existing cognitive network can’t work together with licensed (primary) users’ network at the same frequency-time domain, and secondary users (SUs) of cognitive network only wait the frequency band occupied by primary users (PUs) to be free. To solve the problem, this paper proposed a spatial coding approach for MIMO cognitive network, where a MIMO base-station with six antennas provides three different spatial codes for three users such as one PU and two SUs, then the SUs can share the bandwidth of PUs. The spatial codes’ design for encoding and decoding vectors is provided. Simulation results verify the proposed approach.

Keywords-cognitive radio;MIMO system; spatial coding; bandwidth sharing.

I. INTRODUCTION To deal with the conflicts between spectrum

congestion and under-utilization, cognitive radio (CR) has been recently proposed as a smart and agile technology which allows non-legitimate users to utilize licensed bands opportunistically [1-8]. CR should have the abilities to sense the unemployed spectrum, select the frequency dynamically and control transmitted power. However, the interference caused by sharing the same radio channel becomes an obstacle that limits the whole system performance, such as the system throughput [6]. Thus, another aim of this system should be to maximize the output of the cognitive network without affecting the primary user.

In practical application, cognitive network and primary users’ network exist in a same area. Since the bandwidth is originally licensed to the PU, it can be allocated to the SUs when PUs don’t occupy the bandwidth, which leads to the usage efficiency of the bandwidth to be very lower. However, if PUs and SUs operating at the same frequency simultaneously, two kinds of interference would be generated inevitably [7,8]. One is the interference from SUs to PU, the other is inversely related. Since the PUs always have a higher priority compared with the SUs in bandwidth usage, the main challenge is to protect PUs from the interference of SUs when SUs operating at the same frequency simultaneously, whereas few solutions about the problem can be found.

The conventional CR scheme had to shut the

sub-carrier of the SUs so that the SUs’ operation would not interfere with the PUs at the expense of using this victim band. A spatial coding approach for MIMO CR networks is proposed by us to ensure PUs and SUs work at same band-width, and SUs need not to back-off. Applying the proposed approach, the SUs could utilize the victim band for its data transmission, with the condition that the SUs and PUs do not interfere with each other’s operation. The proposed spatial coding approach for CR networks introduces IA technique [10-13] so that both PU and SUs can work well in the same frequency simultaneously. The proposed spatial coding technique is only based on linear pre-coding at the transmitters and zero-forcing at the receivers, which requires the channel knowledge to be known exactly [14]. Simulations results verify the proposed approach to be valid.

II. MIMO CR DOWNLINK BASE-BAND SYSTEM

In this section, without loss of generality, a MIMO CR downlink base-band system with one base-station (BS) and three mobile stations (MSs) is considered as shown as Fig.1. We assume that the MIMO channel is assumed to be time-varying and the channel coefficients can be acquired by the approach of [14], thus the channel matrix is perfectly known as well.

In Fig.1, the BS transmitter are equipped with 6 antennas and MS receivers are equipped with 2 antennas. The BS transmitter provides three spatial coding flows

, 1, 2,3i is i =v from is three groups of antennas, each

group has two antennas. is represents the signal from

antenna group i and , 1, 2,3i i =v represents the pre-coding vector at antenna group i , which is a 2 1× vector.

The thi (1 3)i≤ ≤ receiver intended signal along with interference from other BS transmitter antenna group

( )j j i≠ . The base-band system model in Fig. 1 at a certain time instance can be expressed by

3

1,i ii i i ij j j i

j j i

s s= ≠

= + +∑r H v H v n (1)

where ijH is a 2 2× channel matrix of the link between *This project is supported by the Beijing Natural Science Foundation of China: (No. 4102050).


123

MS receiver i and antenna group j , and ijH is the

interfering link between antenna group j and unexpected

receiver i . in accounts for the random noise generated in the radio frequency front-end of the receiver i , and

in denotes 2 1× additive noise vector, which is assumed

to be i.i.d. Gaussian distributed with zero mean and 2σ variance.

Fig.1 MIMO CR downlink base-band system The proposed pre-coding vector is a kind of spatial

codes, which let multi-users’ interference to be cancelled at the receivers.

In other words, BS in Fig. 1 constructs pre-coding vector , 1, 2,3i i =v to encode the transmitting signals

, 1, 2,3is i = and decoding vector , 1, 2,3i i =u for MSs to decode the signal at MS receivers.

The intended pre-coding vector , 1, 2,3i i =v and

decoding vector , 1, 2,3i i =u should satisfy the following constraints:

=0 , Hi ij i i j∀ ≠u H v (2)

rank( )=1Hi ii iu H v (3)

Then, the MS receiver i can acquire its expected signal from BS by multiplying its decode vector H

iu . The proposed decoder procedure is simple, the

decoded signal iy at receiver i can be obtain by Hi iu r

3

1,

H H H Hi i i i ii i i i ij j j i i

j j i

Hi i i i

y s s

c s

= ≠

= = + +

= +

∑u r u H v u H v u n

u n

(4) where H

i i ii ic = u H v , which can determine the BER performance of the receiver i .

Now, we construct the pre-coding vector , 1, 2,3i i =v , which need to satisfy the conditions (2)

and (3). Due to (2), we have

112 2 13 3 2 2 12 13 3

121 1 23 3 2 3 23 21 1

131 1 32 2 2 1 31 32 2

−

−

−

⎧− = =⎫⎪⎪− = ⇒ =⎬ ⎨

⎪ ⎪− = =⎭ ⎩

H v H v 0 v H H v

H v H v 0 v H H v

H v H v 0 v H H v

(5)

Then the 1v will be given through 1 1 1

1 31 32 12 13 23 21 1− − −=v H H H H H H v (6)

From (6), 1v is the eigenvector of 1 1 1

31 32 12 13 23 21− − −H H H H H H . BS can solve 1v from (6).

Substitute 1v into (5), we can obtain 3v 1

3 23 21 1−=v H H v (7)

and from (5) further obtain 2v

12 12 13 3

−=v H H v (8)

Then substitute , 1, 2,3i i =v into (2) so as to solve the

decoding vector , 1, 2,3i i =u , which is a 2 1× vector.

1 12 2 1 13 3

2 21 1 2 23 3

3 31 1 3 32 2

0

0

0

H H

H H

H H

⎫= =⎪

= = ⎬⎪= = ⎭

u H v u H v

u H v u H v

u H v u H v

(9)

In the proposed approach, BS with six antennas uses three pre-coding vectors , 1, 2,3i i =v to provide three independent spatial channels for three MSs (PU and SUs), which means that one PU MS can share its bandwidth with two SU MSs. Here, the pre-coding vectors

, 1, 2,3i i =v and decoding vectors , 1, 2,3i i =u are the proposed spatial codes. Now, we study whether the pre-coding vectors

, 1, 2,3i i =v and decoding vectors , 1, 2,3i i =u can be used in MIMO CR uplink base-band system.

III. MIMO CR UPLINK BASE-BAND SYSTEM Similar to the MIMO CR downlink base-band system

in Fig.1, a MIMO CR uplink base-band system with one base-station (BS) and three mobile stations (MSs) is considered as shown as Fig.2.

The MS transmitters provide three spatial coding flows , 1, 2,3i is i =v from their antennas, is represents

the signal from MS i and , 1,2,3i i =v represents the

pre-coding vector at MS i , which is a 2 1× vector. BS divides its six antennas into three groups: group thi (1 3)i≤ ≤ , each group has two antennas.

The received signal of the BS group thi (1 3)i≤ ≤ has the interference from other

MS ( )j j i≠ . The base-band system model in Fig. 2 at a certain time instance is expressed by

3

1,i ii i i ij j j i

j j i

s s= ≠

= + +∑r G v G v n (10)

where ijG is a 2 2× channel matrix of the link between

BS

MS1

MS2

MS3

1 1sv

2 2sv

3 3sv

3r

2r

1r 11H

21H 31H

12H

22H

13H

33H23H


124

antenna group i and the transmitter of MS j , and ijG is

the interfering link between antenna group i and unexpected transmitter j . in accounts for the random noise generated in the radio frequency front-end of the BS receiver of antenna group i , and in denotes 2 1× additive noise vector, which is assumed to be i.i.d. Gaussian distributed with zero mean and 2σ variance.

Fig.2 MIMO CR uplink base-band system

If the band-width of downlink and uplink of the CR network are same, the channel matrix ij ij≈G H , MSs

can use the pre-coding vectors , 1, 2,3i i =v in downlink system, else needs to repeat the algorithms in Section II to re-construct , 1, 2,3i i =v .

MSs in Fig. 2 construct pre-coding vector , 1, 2,3i i =v to encode the transmitting signals

, 1, 2,3is i = and decoding vector , 1, 2,3i i =u for BS to decode the signal at BS receivers.

The intended MS pre-coding vector , 1, 2,3i i =v

and BS decoding vector , 1, 2,3i i =u should satisfy the following constraints:

=0 , Hi ij i i j∀ ≠u G v (11)

rank( )=1Hi ii iu G v (12)

Then, the BS receiver antenna group i can acquire its expected signal from BS by multiplying its decode vector H

iu . The proposed BS decoder procedure is simple, the

decoded signal iy at BS receiver i can be obtain by Hi iu r ,

3

1,

H H H Hi i i i ii i i i ij j j i i

j j i

Hi i i i

y s s

d s

= ≠

= = + +

= +

∑u r u G v u G v u n

u n

(13) where H

i i ii id = u G v .

The 1v will be given through

1 1 11 31 32 12 13 23 21 1

− − −=v G G G G G G v (14)

From (14), 1v is the eigenvector of 1 1 1

31 32 12 13 23 21− − −G G G G G G . 1v can be solved from (14).

Similar to (7) and (8), MSs can obtain 3v and 2v 1

3 23 21 1−=v G G v (15)

12 12 13 3

−=v G G v (16) Similar to (9), MSs obtain the decoding vector

, 1, 2,3i i =u , which is a 2 1× vector.

1 12 2 1 13 3

2 21 1 2 23 3

3 31 1 3 32 2

0

0

0

H H

H H

H H

⎫= =⎪

= = ⎬⎪= = ⎭

u G v u G v

u G v u G v

u G v u G v

(17)

In practical application, it is difficult for MSs with lower complexity to complete the construction of spatial codes , 1,2,3i i =v and , 1,2,3i i =u , such as the calculation of (14) and obtaining the parameters of the matrices in (14). We can resort to the approach in [14] to move the computation complexity of the spatial codes to BS, BS estimates the parameters of the matrices in (14), calculate (14) and all spatial codes , 1,2,3i i =v and , 1,2,3i i =u .

IV. RESULTS OF SIMULATION

In this section, our attention will focus on the MIMO CR downlink base-band system with three MSs, namely one primary user (MS1 in Fig.1) and two secondary users SU1(MS2 in Fig.1) and SU2(MS3 in Fig.1). A specific example will illustrate the performance of our spatial coding approach. We check whether the proposed approach can ensure PUs and SUs to share the band-width, and SUs need not back-off. In Fig.1, the three original signal sequences at each group of antennas are transmitted to their corresponding receivers.

The parameters of channel matrices in (1) and Fig. 1 are assumed to be generated randomly,

11

0.9554 0.15480.1332 0.8677

H⎡ ⎤

= ⎢ ⎥⎣ ⎦

,12

0.2501 0.06860.3277 0.2994

H⎡ ⎤

= ⎢ ⎥⎣ ⎦

,

13

0.5916 0.23590.2033 0.3984

H⎡ ⎤

= ⎢ ⎥⎣ ⎦

,21

0.5017 0.79600.6508 0.2334

H⎡ ⎤

= ⎢ ⎥⎣ ⎦

,

22

0.6008 0.51580.1125 0.8378

H⎡ ⎤

= ⎢ ⎥⎣ ⎦

, 23

0.2208 0.27760.4982 0.6525

H⎡ ⎤

= ⎢ ⎥⎣ ⎦

,

31

0.3173 0.27420.5098 0.1973

H⎡ ⎤

= ⎢ ⎥⎣ ⎦

, 32

0.1112 0.39640.2974 0.4208

H⎡ ⎤

= ⎢ ⎥⎣ ⎦

,

33

0.6115 0.09190.6938 0.7021

H⎡ ⎤

= ⎢ ⎥⎣ ⎦

.

With (6)-(8), BS calculate the encoding vector iv , and the encoding matrix is as follows,

BS

MS1

MS2

MS3

1 1sv

2 2sv

3 3sv

3r

2r

1r

11G21G

31G

12G

22G

13G

33G23G


125

1

0.95190.3065

v⎡ ⎤

= ⎢ ⎥−⎣ ⎦,

2

0.83000.1905

v⎡ ⎤

= ⎢ ⎥⎣ ⎦

, 3

0.05480.7979

v⎡ ⎤

= ⎢ ⎥⎣ ⎦

.

Similarly, the decoding vector iu can be solved by using

(9), 1

0.32900.2207

u⎡ ⎤

= ⎢ ⎥−⎣ ⎦, 2

0.54800.2336

u⎡ ⎤

= ⎢ ⎥−⎣ ⎦,

3

0.42480.2180

u−⎡ ⎤

= ⎢ ⎥⎣ ⎦

.

0 5 10 15 20 25 30 35 4010

−8

10−7

10−6

10−5

10−4

10−3

10−2

10−1

100

SINR(dB)

BE

R

PUSU1SU2

Fig.3 the BER curve of each user Then BS sends the decoding vector to its

corresponding receiver through control channel so that each MS receiver just uses (13) to separate the desired signal from interferers.

Fig.3 shows the BER curve of each MS user. In the simulation, we have

1 1 11 1 0.3143Hc = =u H v , 2 2 22 2 0.2680Hc = =u H v ,

and 3 3 33 3 0.0850Hc = =u H v , 1 2 3c c c> > . From Fig.3 we can see that the BER of PU is best due

to 1 2 3c c c> > , when SNR=12dB, the PU’s BER <10-7. SU1 and SU2 can share the band-width with PU, their

BERs are smaller than 10-7, when SNRs are larger than 18dB and 26dB, respectively, which are determined by

2c and 3c .

V. CONCLUSIONS This paper develops a spatial coding approach for

MIMO cognitive radio network. Limited the paper size, we only consider the approach to be applied in 2 2× MIMO perfect channel. Because the network has cognitive competence, we assume the spatial channels to be known perfectly. Compared with the conventional MIMO CR system, this proposed system keeps PUs and SUs to share the same band-width at the same time, which reduces the implementation complexity of MS receivers. This paper provides the algorithms to construct the

encoding and decoding vectors from downlink channel and uplink channels. Different from the IA technique [10-13], the paper simplified the spatial coding process MIMO base-station and PU and SU receivers, and there is no requirement for PU and SU MSs to be linked by Internet and joint decoding. PU and SU receivers can independently use simple spatial codes to separate the desired signal from interferers other than collaborative decoding, which makes the proposed spatial coding process practical.

REFERENCES [1] Federal Communications Commission, “Spectrum Policy Task

Force," Rep. ET Docket no. 02-135, Nov. 2002. [2] J. Mitola and G. Q. Maguire, “Cognitive radio: making software

radios more personal," IEEE Personal Commun., vol. 6, pp. 13-18, Aug. 1999.

[3] T. A. Weiss and F. K. Jondral, “Spectrum pooling: an innovative strategy for the enhancement of spectrum efficiency," IEEE Commun. Mag., vol.42, pp. S8-14, Mar. 2004.

[4] Y. Xiao, K. S. Kim, G. Z. Qu. A Cognitive Spatial Multiplexing Scheme for MIMO-CDMA Networks[C]. Proc. of Conference on Wireless, Mobile and Multimedia Networks, Beijing, China, 2010: 147-150.

[5] Y. Xiao,Y. K. Zhang, G. Z. Qu, et al. Spatial Multiplexing Algorithms of Cognitive Base-Stations[C]. Proc. of Conference on Wireless, Mobile and Multimedia Networks, Beijing, China, 2010: 221-224.

[6] Hamdi, K.; Wei Zhang; Letaief, K.; , "Opportunistic spectrum sharing in cognitive MIMO wireless networks," IEEE Transactions on Wireless Communications, vol.8, no.8, pp.4098-4109, August 2009.

[7] S. Haykin, “Cognitive radio: brain-empowered wireless communications” IEEE Trans. Selected Areas in Communications, Vol.23, No. 2, pp. 201-220, 2005.

[8] H. Zamiri-Jafarian, M.A. Jannat-Abad, "Cooperative Beamforming and Power Allocation in the Downlink of MIMO Cognitive Radio Systems," Vehicular Technology Conference Fall (VTC 2010-Fall), 2010 IEEE 72nd , vol., no., pp.1-5, 6-9 Sept. 2010.

[9] M. Maddah-Ali, A. Motahari, and A. Khandani, “Signaling over MIMO multi-base systems: Combination of multi-access and broadcast schemes,” in Proc. IEEE Int. Symp. information Theory (ISIT), Seattle, WA, USA, July 2006, pp. 2104– 2108.

[10] R.Tresch, M.Guillaud, E. Riegler, "On the achievability of interference alignment in the K-user constant MIMO interference channel," IEEE/SP 15th Workshop on Statistical Signal Processing,(SSP '09), pp.277-280, 2009.

[11] O. El Ayach; S. W. Peters, R. W. Heath, "Real world feasibility of interference alignment using MIMO-OFDM channel measurements," Military Communications Conference, 2009. MILCOM 2009. IEEE , vol., no., pp.1-6, 18-21 Oct. 2009.

[12] Tiangao Gou and Syed A. Jafar, “Degrees of Freedom of the K-User M x N MIMO interference Channel,” 42nd Asilomar Conference on Signals, Systems and Computers, 2008, pp.126-130

[13] Krishna Gomadam, Viveck R. Cadambe, and Syed A. Jafar, “Approaching the Capacity of Wireless Networks through Distributed Interference Alignment,” IEEE Global Telecommunications Conference, 2008, pp.1-6.

[14] Y. Xiao, J. L. Liu, H. J. Yin, Kim Kiseon, MIMO Spatial Multiplexing Systems with Uplink Pilot and LDPC Codec, 5th International Conference on Wireless Communications, Networking and Mobile Computing, pp 1272-1276, Beijing, 2009.


126

Optimizing Power Allocation in a Cellular

DS/FFH-CDMA System under Rayleigh Fading

P.Varzakas Technological Educational Institute of Lamia

Department of Electronics

GR-35 100, Lamia, Greece

e-mail: [email protected]

Abstract—The optimization between the average received power and the theoretically achievable average channel capacity per user (in the Shannon sense) of a hybrid direct-sequence/fast frequency hopping code-division multiple-access (DS/FFH-CDMA) cellular system, when operating in a Rayleigh fading environment, is examined. The theoretical analysis leads to a novel-closed form expression for the optimal average received power value based on the maximization of the achieved spectral efficiency, estimated in terms of the available average channel capacity per user. Finally, respective numerical results are presented, which will be useful for the initial practical design of a DS/FFH-CDMA cellular system when operating in a Rayleigh fading environment.

Keywords: Hybrid CDMA systems, Power allocation, Cellular systems, Rayleigh fading.

I. INTRODUCTION Hybrid systems are attractive because they can combine

the advantages of both Direct Sequence (DS) and Frequency Hopping (FH) systems while avoiding some of their advantages. A hybrid DS/FFH-CDMA system can combine the anti-multipath effectiveness of DS system with the good antipartial-band-jamming and the good antinear-far problem features of FH system, [1]. Although, in general, in cellular networking, the hybrid DS/FFH-CDMA technique is not the standard technique remains an interesting technique for a number of reasons. Then, in a non-cellular or wireless local environment, where we have point-to-point communication without using a base station, this is a more flexible and cheaper method than the cellular approach. Because of the absence of base stations power control, playing a key role in reducing the near-far effect is no longer possible. FFH is therefore used to beat the near-far effect. On the other hand there is still DS with its well-known advantages: CDMA, jamming rejection, fading rejection and security.

Following the method and the hybrid system described firstly in [1], here, the spectral efficiency of a DS/FFH-CDMA cellular system is evaluated in terms of each user’s achievable average channel capacity. The channel capacity expression, establishes an upper bound limit for reliable information transmission over a bandlimited additive white Gaussian noise (AWGN) environment. When the channel side information (CSI) is not available at the transmitter, the source data is transmitted at a constant rate. Since no CSI is available at the transmitter, data transmission takes place over all fading states including deep fades where the data is lost and hence the effective channel capacity is significantly reduced. In cellular mobile radio, where signal fading is a

considerable capacity degradation factor, channel capacity can be estimated in an average sense and used as a figure of merit for system’s operation. This average channel capacity formula would indeed provide the true channel capacity, if channel side information were available at the receiver, [2]. It must be noticed, that the following analysis does not solve the problem of the capacity region, i.e., the set of information rates at which simultaneously reliable communication of the messages of each user is possible.

The final equation, theoretically derived, to the author’s best knowledge, is the first time such expression has been exposed, thus avoiding complex algorithms or lengthy simulations. A comparison of the theoretical results derived here, with alternative methodologies described previously in literature, is not possible because these methodologies assume specific conditions for the system’s operation and no one of these considers the case of the maximization of the available average channel capacity per user, [3-7]. Hence, a novel-closed form expression for the optimal average received power, in a Rayleigh fading environment, with respect to the maximization of the achieved spectral efficiency, estimated in terms of the available average channel capacity per user, is derived and respective numerical results are presented. The final expression can be very useful for the practical design of a DS/FFH-CDMA cellular system, specifically in the power control algorithm applied and for an initial quantitative analysis.

The remainder of this paper is organized as follows. Section II describes the system’s model applied in the analysis and the operation of the considered hybrid DS/FFH-CDMA cellular system in an ideal non-fading AWGN and a Rayleigh fading environment. Numerical results and graphs are presented and discussed in section III. Final conclusions are outlined in the last section.

II. SYSTEM’S MODEL AND OPTIMAL RECEIVED

POWER IN A RAYLEIGH FADING ENVIRONMENT We consider a cellular DS/FFH-CDMA system where it

is assumed that accommodates K of users per cell, in contrast to a variable allocation of users, which is governed by a birth–death process, implying a dynamic channel capacity model, and the users within each cell can be approximately orthogonalized, [8]. This is accomplished by choosing hopping sequences (shifts of the same basic sequence) that are orthogonal within the cells. In addition, hopping sequences need be assigned for minimum inter-cell correlation, meaning that any two users in adjacent co-channel cells interfere only at one hop during the period of

the hopping sequence, [9]. However, the theoretical results


127

derived in the paper can be applied directly in a DS/SFH-CDMA system, with available number of users per cell, considering that the number of users per cell K represents the mean value of users per each cell.

Then, the original transmitted signal is only corrupted by AWGN and co-channel interference (CCI) power. During each frequency hop, a DS signal is transmitted in the form

of a spread signal with bandwidth Wds=Gp⋅Ws, where Gp is the processing gain applied and Ws is the signal bandwidth. The totally allocated system’s bandwidth Wt is equal to:

Wt=M⋅Wds=M⋅Gp⋅Ws (1)

where M (M>1) is the number of hops per transmitted bit. In addition, we consider the CCI power resulting only from the first tier of the DS/FFH-CDMA cellular system assuming a cell cluster size equal to twelve and where all base stations' and mobile units' antennas are assumed omnidirectional. Thus, the channel capacity required for error-less transmission of a signal of bandwidth Wds will be given by the Shannon-Hartley theorem, [10]:

)S(1logWC DS/FFH,2dsDS/FFH, ii +⋅= (2)

where Si,,DS/FFH, i=[1,..,12K], is the average signal-to-interference plus noise ratio (SINR) received at the i-th user as it reaches the boundary of a cell (i.e. the worst case of operation). Assuming that in the downlink all mobile units of a certain cell will receive equal average signal power from their cell site when appropriate power control scheme is applied, [11], then, for a fourth power law path loss, the average received signal power Pr at the distance r by the i-th user, i=[1,..,12K], will be:

Pr=4rα −⋅ (3)

where α is a constant factor, [11]. Therefore, the SINR received by the mobile unit as it reaches the boundary of a

cell, DS/FFHSi,

, can readily be determined by considering the

average CCI power resulting from the totally 11K co-channel interfering users: 2K, 3K and 6K, respectively, located at distances R, 2R and 2.633R, as follows, [11,12], i.e.:

[ ]

M

PK)(2.3123PWN

M

P

M

1(2.633R)α6K(2R)α3KRα2KPWN

P

,S

rhds0

r

4-4-4-

hds0

r

DS/FFH

⋅⋅⋅+=

=⋅⋅⋅+⋅⋅+⋅⋅⋅+

′=

=

⋅

⋅

i

(4)

since, for a FFH transmission scheme, the CCI power, as seen by a desired signal, originates, on the average, from 1/M of the co-channel users, [13], and P

/r is the average

received signal power by the mobile unit as it reaches the boundary of a cell and N0 is the noise power spectral density of the AWGN. However, it must be noted that (4) does not take into account the voice activity cycle, sector-reuse parameterization, lognormal variations and a random location model for the users’ positions, as required in describing real commercial systems. Furthermore, CCI is considered as Gaussian distributed interference even for small values of the number of system’s users, [14,15]. In

eq.(4),rP′ is the user’s average received signal power, in each

of the M frequencies, being equal to:

rP′ =

M

Pr (5)

assuming that, in the FFH case, the totally transmitted signal power is equally shared, by hopping, among the M different

carrier frequencies. In eq.(4), hP is the probability of hit, for

the FFH case, approximated by:

M

1Ph ≅ (6)

Thus, eq.(4) can be rewritten in the form:

SK)(2.3123M

1MG

SS

p

DS/FFH

⋅⋅⋅+=

⋅

i,

(7)

where S=(Pr/N) is the average received signal-to-noise ratio (SNR) over signal bandwidth Ws and N=N0·Ws is the AWGN power over signal bandwidth Ws.

Assuming that the physical channel of bandwidth Wds is greater than the coherence bandwidth Wcoh of the Rayleigh fading channel, the maximum number Mds of uncorrelated resolvable paths is approximated by, [16],:

1)WG(1∆][WM spdsds +∆⋅⋅≈+⋅= (8)

where ∆ is the maximum delay spread of the fading channel and [.] returns the largest integer less than, or equal to, its argument. Compared to DS transmission, where it can be said that the diversity effect is gained in parallel, the diversity in FFH transmission is achieved sequentially. Hence, a M hops per transmitted bit FFH system, can be seen as equivalent to an M-branch maximal-ratio combining (MRC) space diversity system, [17]. Therefore, the average

channel capacity per user ⟨Ci⟩DS/FFH,Rayleigh, normalized over the total system’s bandwidth Wt, is given by:

( ) ( )( ) ( ) dγ

S

γexp

S ! 1M

γγ+1log

WM

C

W

C

DS/FFH,0

M

DS/FFH,

1M

2

ds

RayleighDS/FFH,

t

RayleighDS/FFH,

⋅

−⋅

−=

=⋅

=

∫∞ −

⋅

ii

ii

(9)

where ⟨.⟩ indicates average value and Si,DS/FFH=⟨ γ ⟩, given

by eq.(7), is the average received SINR in each of the M frequencies where the DS signal is transmitted. Furthermore, if path-diversity reception, provided by a MRC RAKE receiver, is also applied to the DS/FFH-CDMA system, then additional diversity will be achieved. Hence, assuming that the multipath intensity profile (MIP) has equal path strengths on the average, the SINR after path-

diversity applied, in each of the M frequencies,DS/FFH,pt,Si

will be given by, [18],:

( ) ( ) SK2.3123M

1

W

W

S M

SK2.3123M

1MG

S M

SMS

s

t

ds

p

ds

DS/FFH,dsDS/FFHpt,,

⋅⋅⋅+⋅=

⋅⋅⋅+⋅⋅=

=⋅= ii

(10)

where DS/FFH,Si

=⟨ γ ⟩ is the average received SINR in each

of the M frequencies in a Rayleigh fading channel (the


128

suffice 'pt' refers to the path-diversity reception applied).

Applying directly eq.(10) to eq.(9), ⟨Ci⟩DS/FFH,Rayleigh, normalized over the total system’s bandwidth Wt, is rewritten as:

( ) ( )( ) ( ) dγ

S

γexp

S! 1M

γ γ+1log

WM

C

W

C

DS/FFHpt,,

M

DS/FFHpt,,

1M

0

2

ds

RayleighDS/FFH,

t

RayleighDS/FFH,

⋅

−⋅

⋅−⋅=

=⋅

=

−∞

∫ii

ii

(11)

The problem of maximization of the normalized average channel capacity per user, can be stated as follows:

( ) ( )( ) ( ) dγ

S

γexp

S! 1M

γ γ+1logmax

DS/FFHpt,,

M

DS/FFHpt,,

1M

0

2 ⋅

−⋅

⋅−⋅

−∞

∫ii

(12)

The combined average spread SINR after diversity

reception i.e. ⋅DS/FFH,pt,SiMds, that maximizes eq.(12), equals

to 6 dB, [19], i.e.,

( )6.0

s

t

2

ds

DS/FFH,

2

dsdsDS/FFHpt,,

10

SK2.3123M

1

W

W

S M

SMMS

=⋅⋅⋅+

⋅=

=⋅=⋅ii

(13)

applying directly eq.(10). The eq.(13) can be rewritten as:

( )0.6

s0

r

p

s0

r

2

ds 10

WN

PK2.3123

M

1MG

WN

P

M =

⋅⋅⋅⋅+⋅

⋅⋅

(14)

Then, the optimal received power Pr,op (the new suffice 'op' refers to the optimal value) can be found directly from eq.(14), as following i.e.:

K9.2MM

GWNM3.9P

2

ds

ps0

2

opr, ⋅−⋅

⋅⋅⋅⋅= (15)

III. NUMERICAL RESULTS The optimal received power Pr,op (expressed in Watt

(W)), given by eq.(15), is plotted in Figure 1 as function of the number of users per cell K, where the following values are assumed: (i) totally constant allocated system’s bandwidth: Wt=10MHz, (ii) signal bandwidth: Ws=30KHz, (iii) number of hops per transmitted bit: M=5, (iv) total multipath spread of the urban Rayleigh fading channel: ∆=3µsec.

2 4 6 8 10

12

14

16

18

20

22

K (users/cell)

Figure 1. Optimal received power Pr,op in a DS/FFH-CDMA cellular system versus the number K of users per cell in a Rayleigh fading environment.

As it can be seen directly from Figure 1, the required value of optimal received power Pr,op is increased as the number of users per cell K is increased, indicating that when the number of users per cell increases, and consequently the CCI power increases respectively, the required optimal received power Pr,op must be increased respectively in order to minimize the impact of the CCI power on the system’s performance.

2 4 6 8 10

0

1

2

3

4

5

M (hops/bit)

Figure 2. Optimal received power Pr,op in a DS/FFH-CDMA cellular system versus the number M of hops per transmitted bit in a Rayleigh fading environment.

In addition in Figure 2, the optimal received power Pr,op,

(expressed in Watt (W)), is plotted as function of the number M of hops per transmitted bit, for: S=30dB, Wt=10MHz, Ws=30KHz, ∆=3µsec and K=10 users per cell as an indicative value (in real cellular systems the actual number K of users per cell is of the order of 50). As it can be seen directly from Figure 3, the required value of optimal received power Pr,op is increased as the number M of hops per transmitted bit increased, indicating that although an increased value of number M of hops per transmitted bit provides increased inherent diversity potential, the CCI power is still sufficient and then, an increased value of the

Pr,op (W)

Pr,op (W)


129

received power Pr,op is needed finally, in order to minimize the effect of the CCI in the system’s performance.

IV. CONCLUSIONS In this paper, we estimate the optimal average received

power of a cellular DS/FFH-CDMA system when operating in a Rayleigh fading environment, which maximizes the achieved spectral efficiency. Spectral efficiency is expressed in terms of the average channel capacity available to each user, when additional path-diversity reception is applied. It is derived, without applying complex theoretical algorithms or lengthy simulations, a general expression, which relates the optimal average received power with all system’s parameters. However, it must be noticed that, similar expression has been recently derived by the author in [20], but concerning the optimization of the processing gain applied. The final expression can be useful for the initial practical design of a DS/FFH-CDMA cellular system in a Rayleigh fading environment and for the power control scheme applied but a simulation process or an experimental comparison with other methods must be described analytically, in a future paper, in order to compare with the theoretical results derived in this paper.

REFERENCES [1] P.Varzakas and G.S.Tombras, "Spectral efficiency for a hybrid DS/FH

CDMA system in cellular mobile radio", IEEE Trans. on Veh.

Technology, vol.50, no.6, pp.1321-1327, Nov.2001.

[2] A.J.Goldsmith and P.Varaiya, "Capacity of fading channels with channel side information", IEEE Trans. on Inform. Theory, vol.43, no.6, pp.1986-1992, Nov.1997.

[3] S.Songsong and W.A.Krzymien, "Hybrid ARQ and optimal signal-to-

interference ratio assignment for high-quality data transmission in DS-CDMA", European Trans. on Telecommunications, vol.12, no.1, pp.19–29, 2001.

[4] S.Kandukuri and S.Boyd, "Optimal power control in interference-limited fading wireless channels with outage probability

specifications", IEEE Trans. on Wireless Communications, vol.1, no.1, pp.46–55, 2002.

[5] M.Xiao, N.Shroff and E.Ching, "A utility-based power-control scheme in wireless cellular systems", IEEE/ACM Trans. on Networking,

vol.11, pp.210–221, 2003.

[6] H.Boche and S.Stanczak, "Convexity of some feasible QoS regions and asymptotic behavior of the minimum total power in CDMA

systems", IEEE Trans. on Communications, vol.52, no.12, pp.2190–2197, 2004.

[7] T.Alpcan, T.Basar, R.Sricant and E.Altman, "CDMA uplink power

control as a noncooperative game", Wireless Networks, vol.8, pp.659–669, 2002.

[8] N.Livneh, R.Meidan, M.Ritz and G.Silbershatz, "Frequency hopping CDMA for cellular radio", Proc. of the International Commsphere

Symposium, Herzilya, Israel, pp.10.5.1–10.5.6, Dec. 1991.

[9] A.Lempel and H.Greenberger, "Families of sequences with optimal

hamming correlation properties", IEEE Trans. Inform. Theory, vol. IT-1, pp.90–94, Jan.1974.

[10] C.E.Shannon, "Communication in the presence of noise", Proc. of

IRE, vol.37, pp.10-21, Jan.1949.

[11] M.K.Simon and M.S.Alouini, ‘Digital Communication over Fading

Channels: A Unified Approach to Performance Analysis’, Wiley: New York, 2000.

[12] W.C.Y.Lee, ‘Mobile Communications Design Fundamentals’, 2nd ed.

New York: Wiley, 1993.

[13] C.D’Amours and A.Yongacoglu, "Hybrid DS/FH-CDMA system employing MT-FSK modulation for mobile radio", in Proc.

PIMRC’95, Toronto, Canada, pp.164–168, Sept.1995.

[14] K.S.Gilhousen, I.M.Jacobs, R.Padovani, A.Viterbi, L.A.Weaver, and

C.E.Wheatley, "On the capacity of a cellular CDMA system", IEEE Trans. on Veh. Technol., vol. 40, pp.303–312, May 1991.

[15] E.A.Geraniotis, "Coherent hybrid DS-SFH spread-spectrum multiple

access communications", IEEE J. Select. Areas Commun., vol. SAC-3, Sept.1985.

[16] J.G.Proakis, and M.Salehi, ‘Digital Communications’, 5rd Edit.,

McGraw-Hill, 2008.

[17] R.Kohno, R.Meidan and L.B.Milstein, "Spread spectrum access methods for wireless communications", IEEE Commun. Mag., pp.58–67, Jan.1995.

[18] Y.Murata, R.Esmailzadeh, K.Takakusaki, E.Sourour and M.Nakagawa, "Path diversity for FFH/PSK spread-spectrum

communication systems", IEEE J. Select. Areas Commun., vol.12, no.5, pp.970-975, June 1994.

[19] F.Lazarakis, G.S.Tombras and K.Dangakis, "Average Channel Capacity in a Mobile Radio Environment with Rician Statistics", IEICE Trans. on Commun., vol. E-77B, no.7, pp.971-977, July 1994.

[20] P.Varzakas, "On the optimum processing gain for a hybrid DS/FFH-CDMA cellular system over Rayleigh fading channels", International

Journal of Communication Systems, vol. 24, no.7, pp.902–910, July 2011.


130

Statistics of the Channel Capacity in a Cellular

DS/FFH-CDMA Rayleigh Fading System

P.Varzakas Technological Educational Institute of Lamia

Department of Electronics

GR-35 100, Lamia, Greece

e-mail: [email protected]

Abstract—In this paper, we derive analytically the probability density function (pdf) of the theoretically achievable average channel capacity per user (in the Shannon sense) for a constant total system’s allocated bandwidth hybrid direct-sequence/fast frequency hopping code-division multiple-access (DS/FFH-CDMA) cellular system, when operating in a Rayleigh fading environment. Then, the analysis leads to a theoretical novel-closed form expression for the pdf of the channel capacity, which relates the pdf of the channel capacity per user and all system’s parameters. In addition, the probability that the channel capacity per user does not exceed the available average channel capacity per user, in a Rayleigh fading environment, is calculated and respective numerical results are presented to investigate the sensitivity of this probability value from various system’s parameters. Finally, the derived expression can be useful for the initial practical design of a DS/FFH-CDMA cellular system and for a quantitative analysis.

Keywords: Channel capacity, Hybrid DS/FFH-CDMA

systems, Cellular systems, Rayleigh fading.

I. INTRODUCTION

The channel capacity expression, [1], establishes an upper bound limit for reliable information transmission over a bandlimited additive white Gaussian noise (AWGN) environment. When the channel side information (CSI) is not available at the transmitter, the source data is transmitted at a constant rate. Since no CSI is available at the transmitter, data transmission takes place over all fading states including deep fades where the data is lost and hence the effective channel capacity is significantly reduced. In cellular mobile radio, where signal fading is a considerable capacity degradation factor, channel capacity can be estimated in an average sense and used as a figure of merit for system operation, [2]. This average channel capacity formula would indeed provide the true channel capacity, if channel side information were available at the receiver, [3].

The hybrid spread-spectrum systems have recently received considerable interest in commercial, smart grid, and military communication systems because they accommodate high data rates with high link integrity, even in the presence of multipath effects and interfering signals. Then, in this paper, we consider the statistics of the average channel capacity available on the downlink of a DS/FFH-CDMA cellular system, when operating in a Rayleigh fading environment. Following the method and the hybrid system described firstly in [4], and described again here, only for presentation reasons, the achievable average channel capacity of each user’s of a cellular DS/FFH-

CDMA system is evaluated, representing an optimistic upper bound, in an average sense, useful in evaluating practical modulation and coding schemes. It must be noticed, that the following analysis does not solve the problem of the capacity region, i.e., the set of information rates at which simultaneously reliable communication of the messages of each user is possible. Hence, the pdf of the channel capacity (in the Shannon sense) per user for the DS/FFH-CDMA cellular system, under consideration, is analytically derived and it is related with transmission/reception’s parameters as the average received signal-to-noise ratio (SNR) value, the number of hops per transmitted bit, the number of users per cell, the bandwidth of the DS transmission and the signal’s bandwidth. The theoretical final equation, to the author’s best knowledge, is the first time such expression has been exposed, thus avoiding complex algorithms or lengthy simulations. In addition, the probability that the channel capacity per user does not exceed the estimated average channel capacity per user, is calculated, and respective numerical results are presented. However, a simulation process must be described analytically, in order to compare with the theoretical results of this paper and previous published research works, [5]. We are still working on this, for a future paper, but results are not yet derived due to complicated system’s parameters. Then, the analytical description of a respective simulation process remains, this time, due to complicated system’s parameters, an open research problem.

The remainder of this paper is organized as follows. Section II describes the system’s model applied in the analysis. In section III, the channel capacity statistics in the considered hybrid DS/FFH-CDMA cellular system, in a Rayleigh fading environment is examined, while numerical results and graphs are presented and discussed in section IV. Final conclusions are outlined in the last section.

II. SYSTEM’S MODEL A number of basic assumptions, of the cellular DS/FFH-

CDMA system, are set in this section. At first, we consider the twelve co-channel cells, in the first tier, of a cellular DS/FFH-CDMA system, as shown in Figure 1. The users within each cell, can be approximately orthogonalized so that they do not interfere with one another, while in a typical cellular DS-CDMA system multiple-access interference (MAI) power is the dominant source of interference. This is accomplished by choosing appropriate hopping sequences that are orthogonal within the cells. In addition, hopping sequences need be assigned for minimum inter-cell


131

correlation, meaning that any two users in adjacent co-channel cells interfere only at one hop during the period of the hopping sequence, [6]. Under these assumptions, in a cellular DS/FFH-CDMA system, the original transmitted signal is only corrupted by AWGN and co-channel interference (CCI) power.

Figure 1. A DS/FFH-CDMA cellular system and its interference. In addition to the above assumptions, we consider that

the cellular hybrid DS/FFH-CDMA system accommodates K of users per cell. During each frequency hop, a DS signal is transmitted in the form of a spread signal with bandwidth

Wds=Gp⋅Ws, where Gp is the processing gain and Ws is the signal bandwidth. Hopping from one frequency to another is determined by a pseudo-random sequence, while, in parallel, bandwidth spreading over bandwidth Wds is accomplished by multiplying the information data by a pseudo-random sequence. Respectively, at the receiver, the signal is de-hopped by a frequency synthesizer controlled by an identical pseudo-random sequence. After de-hopping, the signal is de-spread using a synchronized and identical pseudo-random sequence to that used in the transmitter. Then, the totally allocated system’s bandwidth Wt is equal to:

Wt=M⋅Wds=M⋅Gp⋅Ws (1) where M (M>1) is the number of hops per transmitted bit and assuming no guard band between adjacent channels with bandwidth Wds. However, the analytical description of a frequency hopping sequence assignment is beyond the scope of this work.

Finally, the analysis covers the base-to-mobile link, i.e., the downlink transmission, while a fixed number of simultaneously transmitting users per each cell is assumed. Although a dynamic user population is a reasonable assumption for DS/FFH-CDMA practical cellular systems, the theoretical results derived in the paper can be applied directly in a DS/FFH-CDMA system, with a variable number of users per cell, considering that the number of users per cell K, represents the mean value of users per cell in a birth-death model describing the variable allocation of users, [7].

III. STATISTICS OF THE CHANNEL CAPACITY IN

A RAYLEIGH FADING ENVIRONMENT We consider the CCI power resulting only from the first

tier of the DS/FFH-CDMA cellular system assuming a cell cluster size equal to twelve and where all base stations' and

mobile units' antennas are assumed omnidirectional. The channel capacity available to all 12K users is limited only by CCI power, since, as already mentioned in the previous section, the K users of each cell are assumed mutually orthogonal. Clearly, transmission of each user signal (assumed Gaussian at the system input) with arbitrarily small BER depends on CCI level. Thus, the channel capacity Ci ,DS/FFH required for error-less transmission of a spread signal of bandwidth Wds will be given by the Shannon-Hartley theorem when arbitrarily complex coding and delay is applied, [1]:

)S(1logWC DS/FFH,2dsDS/FFH, ii +⋅= (2)

where Si,,DS/FFH, i=[1,..,12K], is the average signal-to-interference plus noise ratio (SINR) received at the i-th user as it reaches the boundary of a cell. In order to simplify the mathematical solution, we approximate all hexagon cells of the considered system by circular regions of radius R with the same area. Assuming that in the downlink all mobile units of a certain cell will receive equal average signal power from their cell site, when appropriate power control scheme is applied, then, for a fourth power law path loss, the average received signal power Pav at the distance r by the i-th user, i=[1,..,12K], will clearly be:

Pav=α·r-4 (3)

where α is a constant factor, [8]. Therefore, for the DS/FFH-CDMA system, the SINR received at the mobile unit as it reaches the boundary of a cell (worst case

scenario), DS/FFHSi, , can readily be determined by

considering the average CCI power resulting from the eleven co-channel cells of the first dominant tier of interfering cells, i.e., from 11K interfering users, and neglecting all inter-cell interference, i.e.:

[ ]

M

PK)(2.3123PWN

M

P

M

1(2.633R)α6K(2R)α3KRα2KPWN

P

,S

avhds0

av

4-4-4-

hds0

av

DS/FFH

⋅⋅⋅+=

=⋅⋅⋅+⋅⋅+⋅⋅⋅+

′=

=

⋅

⋅

i

(4)

since, for a FFH transmission scheme, the CCI power, as seen by a desired signal, originates, on the average, from 1/M of the co-channel users, [9], and N0 is the noise power

spectral density of the AWGN. In addition, avP′ is the user’s

average received signal power, in each of the M frequencies, being equal to:

avP ′ =

M

Pav (5)

assuming that, in the FFH case, the totally transmitted signal power is equally shared, by hopping, among the M different carrier frequencies. In the presented work, it is assumed that the average CCI power resulting from the eleven co-channel cells of the first dominant tier of interfering cells and consequently the total average CCI power results from all the simultaneously transmitting users being located within this cluster (only from first tier) i.e. at distances R, 2R, 2.633R as it is presented in eq.(4). However, similar results can be derived assuming a different cluster size as seven. Furthermore, CCI is considered as Gaussian distributed interference even for small values of the number of system’s


132

users, [10]. In eq.(4), Ph is the probability of hit, for the FFH case, approximated by, [11],:

M

1Ph ≅ (6)

Thus, eq.(4) can be rewritten in the form:

SK)(2.3123M

1MG

SS

p

DS/FFH

⋅⋅⋅+=

⋅

i,

(7)

where S=(Pav/N) is the average received signal-to-noise ratio (SNR) over signal bandwidth Ws and N=N0Ws is the AWGN power over signal bandwidth Ws. Following eq.(2), the channel capacity (in the Shannon sense) for the twelve cells of the cellular DS/FFH-CDMA system under consideration, that is, the total channel capacity available to all 12K users, will be given by the sum of the individual rates:

∑∑==

+⋅==12K

1

DS/FFH,2ds

12K

1

DS/FFH,DS/FFH )S(1logWCCi

i

i

i (8)

where Si,DS/FFH is given by eq.(7). Since, in practice, Si ,DS/FFH, i=[1,..,12K], is well below unity (in linear scale), [8], eq.(8) can be approximated by:

)SK121(logWC DS/FFH,2dsDS/FFH i⋅⋅+⋅≅ (9)

However, it must be notice that eq.(9) is an

approximation of eq.(8) only if 1SK12 DS/FFH, <<⋅⋅i

.

We consider now the previously described cellular DS/FFH-CDMA system operating in a Rayleigh fading environment. We assume that the physical channel of bandwidth Wds is greater than the coherence bandwidth Wcoh of the Rayleigh fading channel. The radio channel is modeled as a slowly fading, time-invariant and discrete multipath channel and, thus, it appears to be frequency-selective to the transmitted DS signals. In order to simplify the followed mathematical analysis, the maximum number Mds of uncorrelated resolvable paths is approximated by, [12],:

1)WG(1∆][WM spdsds +∆⋅⋅≈+⋅= (10)

where ∆ is the maximum delay spread or total multipath spread of the fading channel (assumed known or measurable and much less than the bit interval in order to avoid inter-symbol interference (ISI)), and [.] returns the largest integer less than, or equal to, its argument. Although the number of resolvable paths Mds may be a random number, it is approximated by eq.(10) in order to simplify the followed mathematical presentation.

The FFH transmission calls for a kind of diversity reception since each "chip" of the same information bit is transmitted using M different carrier frequencies. Since, the bandwidth Wds is assumed greater than the coherence bandwidth Wcoh of the Rayleigh fading channel, fading will independently affect each of these M frequencies, and then frequency diversity will be obtained. Compared to DS transmission, where it can be said that the diversity effect is gained in parallel, in FFH transmission the diversity is achieved sequentially. Hence, an M hops per transmitted bit FFH system can be seen as equivalent to an M-branch maximal-ratio combining (MRC) space diversity system, [13]. Therefore, the average channel capacity per user

⟨Ci⟩DS/FFH,Rayleigh, is given by:

( )( ) ( )

( )( ) ( )

( ) ( ) dγγpγ1logWG

dγS

γexp

S ! 1M

γγ+1logWG

dγS

γexp

S ! 1M

γγ+1logWC

M

0

2sp

DS/FFH,0

M

DS/FFH,

1M

2sp

DS/FFH,0

M

DS/FFH,

1M

2dsRayleighDS/FFH,

⋅+=

=

−

−=

=

−

−=

∫

∫

∫

∞

∞ −

∞ −

ii

ii

i

(11)

where ⟨.⟩ indicates average value and Si,DS/FFH=⟨ γ ⟩, given

by eq.(7), is the average received SINR in each of the M frequencies where the DS signal is transmitted and no correlation between the M fading patterns is assumed and pM(γ) is the pdf of the combined instantaneous SINR Si,DS/FFH=γ of the spread signal over the bandwidth Wds, with no correlation among the Mds branches, given by:

( )( ) ( )

−⋅

−=

−

DS/FFH,

M

DS/FFHi,

1M

MS

γexp

S ! 1M

γγp

i

(12)

As shown from eq.(11), the channel capacity per user Ci,DS/FFH is a random variable in a fading environment, [2], since the SINR is also a random variable. Using the pdf of the SINR, given by eq.(12), the pdf pCi,DS/FFH(Ci,DS/FFH) of the channel capacity Ci ,DS/FFH is derived as following:

( ) ( )

( )

( ) ( )

( ) ( ) ( )[ ]

( )

( )( )[ ] ( ) ( )[ ]bCaCaexpbexp1Caexpb

1M

a

S

1CaCaexp

S

1exp

1aCexpS

1

MΓW

ln2

W

γ)(1ln2

S

γexpγ

S

1

! 1M

1

W

γ)(1ln2γp

dC

dγγpCp

DS/FFH,DS/FFH,

1M

DS/FFH,

M

DS/FFH,

DS/FFH,DS/FFH

DS/FFH,

1M

DS/FFH,M

DS/FFH,ds

dsDS/FFH,

1M

M

DS/FFH,

ds

M

DS/FFH

MDS/FFH,C DS/FFH,

+⋅+⋅⋅−⋅−⋅⋅⋅−

=

=

+⋅+⋅⋅−⋅

⋅−⋅⋅⋅

=

=+⋅

⋅

−⋅⋅⋅

−=

=+⋅

⋅=

=⋅=

−

−

−

iii

i

ii,

i

i

i

ii

i,

ii

(13)

where in eq.(14) b=1/Si,DS/FFH, a=ln2/Wds, Γ(Μ)=(M-1)! is the Gamma function, [14], and Si ,DS/FFH is given by eq.(7).

We consider now the problem to finding the probability pnon-exc<Ci>DS/FFH,Rayleigh that the available channel capacity per user Ci,DS/FFH does not exceeds the available average channel

capacity per user ⟨Ci⟩DS/FFH,Rayleigh given by eq.(11). Then, using directly eq.(13), pnon-exc<Ci>DS/FFH,Rayleigh is calculated from:

( )

( )( )[ ] ( ) ( )[ ] DS/FFH,

C

0

DS/FFH,DS/FFH,

1M

DS/FFH,

M

DS/FFH,

C

0

DS/FFH,C exc-non

dCbaCaCbexpexp1aCexpb !1M

a

dC Cpp

RayleighDS/FFH,

RayleighDS/FFH,

DS/FFH,RayleighDS/FFH,

iiii

iiC

i

i

ii

∫

∫><

−

><

><

++−−−

=

==(14)

Then, eq.(14) relates directly, the probability

pnon-exc<Ci>DS/FFH,Rayleigh that the available channel capacity per user Ci,DS/FFH does not exceeds the average channel capacity

per user ⟨Ci⟩DS/FFH,Rayleigh in a Rayleigh fading environment, with all the system’s parameters, since the factors a and b, in eq.(14), are dependent of the number of users per cell K, the processing gain applied Gp, the average received SNR S over the signal bandwidth Ws, the number of hops per bit M and the transmitted bandwidth Wds.


133

IV. NUMERICAL RESULTS The pdf pCi ,DS/FFH(Ci ,DS/FFH) of the channel capacity

Ci ,DS/FFH, given by eq.(13), is plotted in Figure 2 as function of the channel capacity per user Ci ,DS/FFH for K=10 users per cell as an indicative value (in real cellular systems the actual number K of users per cell is of the order of 50). In addition, the following values for system’s parameters are assumed: (i) totally constant allocated system’s bandwidth: Wt=10MHz, (ii) signal bandwidth: Ws=30KHz, (iii) number of hops per transmitted bit: M=8, (iv) signal bandwidth of DS transmission: Wds=1.25MHz, (v) processing gain: Gp=41.6 and (vi) average received SNR over the signal bandwidth Ws: S=20dB.

0 2 4 6 8 10

0

5. µ 10-42

1. µ 10-41

1.5 µ 10-41

Figure 2. Probability density function pCi,DS/FFH(Ci,DS/FFH) of channel

capacity Ci,DS/FFH for a hybrid DS/FFH-CDMA cellular system versus the channel capacity Ci,DS/FFH (expressed in bits/sec) when operating in a Rayleigh fading environment for: K=10, Wt=10MHz, Wds=1.25MHz, Ws=30KHz, M=8 and Gp=41.6.

The integral in eq.(14) is calculated numerically as it can

not be expressed in closed form. Then, in Figure 3, the pnon-exc<Ci>DS/FFH,Rayleigh is plotted as function of the average

channel capacity per user ⟨Ci⟩DS/FFH,Rayleigh (expressed in bits/sec) assuming the following values for system’s parameters: (i) number of users per cell: K=10, (ii) totally constant allocated system’s bandwidth: Wt=10MHz, (iii) signal bandwidth: Ws=30KHz, (iv) signal bandwidth of DS transmission: Wds=Wt/M (v) processing gain: Gp=Wds/Ws, (vi) average received SNR over the signal bandwidth Ws: S=20dB and for (a): M=5, (b): M=8 and (c): M=10.

0 5 10 15 2010-32

10-30

10-28

10-26

10-24

Figure 3. Probability pnon-exc<Ci>DS/FFH,Rayleigh that the channel capacity

per user Ci,DS/FFH does not exceeds the average channel capacity per user

⟨Ci⟩DS/FFH,Rayleigh, versus the average channel capacity per user

⟨Ci⟩DS/FFH,Rayleigh (expressed in bits/sec) for a hybrid DS/FFH-CDMA cellular system when operating in a Rayleigh fading environment for: K=10, S=20dB, Wt=10MHz, Ws=30KHz, Wds=Wt/M, Gp=Wds/Ws and for (a): M=5, (b): M=8 and (c): M=10.

Finally, in Figure 4, the pnon-exc<Ci>DS/FFH,Rayleigh is plotted as function of the average channel capacity per user

⟨Ci⟩DS/FFH,Rayleigh (expressed in bits/sec) assuming the following values for system’s parameters: (i) number of users per bit: M=8, (ii) totally constant allocated system’s bandwidth: Wt=10MHz, (iii) signal bandwidth: Ws=30KHz, (iv) signal bandwidth of DS transmission: Wds=1.25MHz (v) processing gain: Gp=41.6, (vi) average received SNR over the signal bandwidth Ws: S=20dB and for (a): K=10, (b): K=50 and (c): K=100.

0 5 10 15 20

10-49

10-46

10-43

10-40

Figure 4. Probability pnon-exc<Ci>DS/FFH,Rayleigh that the channel capacity

per user Ci,DS/FFH does not exceeds the average channel capacity per user

⟨Ci⟩DS/FFH,Rayleigh, versus the average channel capacity per user

⟨Ci⟩DS/FFH,Rayleigh (expressed in bits/sec) for a hybrid DS/FFH-CDMA cellular system when operating in a Rayleigh fading environment for: S=20dB, Wt=10MHz, Ws=30KHz, Wds=1.25MHz, Gp=41.6 and for (a): K=10, (b): K=50 and (c): K=100.

As it can be seen directly from Figures 3 and 4, the probability pnon-exc<Ci>DS/FFH,Rayleigh that the channel capacity per user Ci ,DS/FFH does not exceeds the average channel

capacity per user ⟨Ci⟩DS/FFH,Rayleigh, is increased as the number of hops per bit M is increased or the number of users per cell K is increased, indicating that, in both cases, the available average channel capacity per user is seriously limited by the total CCI power that appears in general in a DS/FFH-CDMA cellular system and finally, the probability that the instantaneous channel capacity per user is smaller than the average channel capacity per user, has a significant value and therefore a CCI power cancellation scheme is needed to mitigate the total CCI power that appears.

V. CONCLUSIONS

In this paper, the pdf of the available average channel capacity per user (in the Shannon sense) for a hybrid DS/FFH-CDMA cellular system when operating in a Rayleigh fading environment is analytically is examined. Then, it is derived theoretically without applying a lengthy simulation process or complex theoretical algorithms, a novel mathematical general expression which relates the pdf of the average channel capacity with system’s parameters. In addition, the probability that the channel capacity per user does not exceed the available average channel capacity per user, is derived. The final expressions can be very useful for the system’s engineers and for an initial quantitative analysis of a DS/FFH-CDMA cellular system, when operating in a Rayleigh fading environment.

REFERENCES

[1] T.M.Cover and J.A.Thomas, ‘Elements of Information Theory’, 1st

Edit., Hoboken, NJ: John Wiley, 2006.

pCi,DS/FFH(Ci,DS/FFH)

Ci,DS/FFH (bits/sec)

⟨⟨⟨⟨Ci⟩⟩⟩⟩DS/FFH,Rayleigh (bits/sec)

pnon-exc<Ci>DS/FFH,Rayleigh

(a)

(b)

(c)

⟨⟨⟨⟨Ci⟩⟩⟩⟩DS/FFH,Rayleigh (bits/sec)

pnon-exc<Ci>DS/FFH,Rayleigh

(a)

(b)

(c)


134

[2] W.C.Y.Lee, "Estimate of Channel Capacity in Rayleigh Fading Environment", IEEE Trans. on Veh. Technology, vol. 39, no.3, pp.187-189, Aug. 1990.

[3] A.J.Goldsmith and P.Varaiya, "Capacity of fading channels with channel side information", IEEE Trans. on Inform. Theory, vol. 43, no.6, pp. 1986-1992, Nov. 1997.

[4] P.Varzakas and G.S.Tombras, "Spectral efficiency for a hybrid DS/FH CDMA system in cellular mobile radio", IEEE Trans. on Veh. Technology, vol. 50, no.6, pp. 1321-1327, Nov. 2001.

[5] M.Olama, S.Smith, T.Kuruganti, and M.Xiao, "Performance study of hybrid DS/FFH spread-spectrum systems in the presence of frequency-selective fading and multiple-access interference," 2012 IEEE International Workshop Technical Committee on Communications Quality and Reliability (CQR), art. no. 6267102, pp. 1-5, 15-17 May 2012.

[6] A.Lempel and H.Greenberger, "Families of sequences with optimal Hamming correlation properties", IEEE Trans. on Info. Theory, vol. 1, pp. 90-94, Jan. 1974.

[7] M.S.Alencar and I.F.Blake, "The capacity for a discrete-state Code Division Multiple Access Channel", IEEE J. Select. Areas Commun., vol. 12, no.5, pp. 925-937, June 1994.

[8] W.C.Y.Lee, Mobile Communications Design Fundamentals, John Wiley, 1993.

[9] C.D’Amours and A.Yongacoglu, "Hybrid DS/FH-CDMA System Employing MT-FSK Modulation for Mobile Radio", IEEE 6th

International Symposium on Personal, Indoor and Mobile Radio

Communications, (PIMRC 1995), pp. 164-168, Toronto, Canada, 27-29 Sept. 1995.

[10] K.S.Gilhousen, I.M.Jacobs, R.Padovani, A.Viterbi, L.A.Weaver and C.E.Wheatley, "On the capacity of a cellular CDMA system", IEEE Trans. on Veh. Technology, vol. 40, pp. 303-312, May 1991.

[11] E.A.Geraniotis, "Coherent hybrid DS-SFH spread-spectrum multiple-access communications", IEEE J. Sel. Areas in Commun., vol. SAC-3, no.5, pp. 695-705, Sept. 1985.

[12] J.G.Proakis, ‘Digital Communications’, 5th Edit., McGraw-Hill, 2008.

[13] R.Kohno, R.Meidan, and L.B.Milstein, "Spread spectrum access methods for wireless communications", IEEE Commun. Mag., pp.58–67, Jan. 1995.

[14] L.S.Gradshteyn and I.M.Ryzhic, ‘Table of Integrals, Series and Products’, New York: Academic, 1980.


135

Modeling the value chain with object-valued Petri nets

Jaroslav Zacek Faculty of Science, Department of Computer Science

University of Ostrava Ostrava, Czech Republic [email protected]

Frantisek Hunka Faculty of Science, Department of Computer Science

University of Ostrava Ostrava, Czech Republic [email protected]

Zdenek Melis Faculty of Science, Department of Computer Science

University of Ostrava Ostrava, Czech Republic

[email protected]

A substantial part of the economic theories is based on conversion and exchange process. This process can be described as a value chain, which can be considered as a cyclic model with complex attributes. There is a serious problem how to express resources and their conversions in a complex cyclic model during the simulation and how to identify these converted resources in every step of the simulation. This paper introduces the Object-valued Petri (OV-PN) modification as a new formalism to create a cyclic model of the value chain. According to the modification we had to define a new path and pass of the OV-PN. We also had to determine new properties. Properties are based on the OV-PN and reflect needs of model requirements. A new formalism is verified on a common enterprise value chain.

Value chain; Object-valued Petri nets; cycle Petri nets; simulation; model validation

I. INTRODUCTION The value chain is a modelling technique to formalise and

monitor the competitiveness of the business. It focuses on the flow of resources between internal business processes that are interconnected to each other. A product increases its value when it passes through a stream of production chain. That is the fundamental notion in value chain analysis [3]. The REA value chain is a network of business processes. The purpose of the network is to directly or indirectly contribute to the creation of the desired features of the final product or service, and to exchange it with other economic agents for a resource that has a greater value for the enterprise [10]. The value chain definition implies that it is important to find a suitable formalism for the simulation of the model run for practical realization. Existing theories such as state machines, Petri nets [5], or neural networks were considered while searching for a correct formalism. Value chains have specific requirements for descriptive formalism allowing their validation and simulation (according to [17]). Specific type of the value chain is supply chain [12]. State machines are not expressive enough to solve this problem. Despite the fact that neural networks are

expressive enough for describing processes in the value chain there is significant complexity in simulation. Two independent neural networks must be created for the simulation of the value chain. The first network is able to validate the model and the second implements simulation steps. In both cases, the neural network must learn these properties. Therefore the process becomes time and implementation consuming. On the other hand the neural network approach is very flexible and can be used to solve multilevel problems (for example multilevel SPAM control [11]).

The Petri Net theory matches the description of the model states more closely, but its expressivity, especially for P/T Petri nets, is very limited [7]. General token is not able to capture such a complex structure, for example an object representing the resource. Therefore this article suggests using the object-valued Petri nets (OV-PN) to ensure the simulation and the validation of value chains. It also discusses some specific properties of the OV-PN and defines new properties for the value chain domain. Main advantage of using the Petri net theory is possibility to create an automatic deterministic process of the code generation [13].

THE VALUE CHAIN AND ITS SIMULATION PROCESS The value chain consists of two main parts: processes and

links that form a chain with other processes (similar to supply chain described in [15]). These parts create the interconnected network of processes increasing the value of the resource. The value chain creates a cyclic bond that means all processes have their inputs and outputs connected together and form a full closed chain. Each process can have more than one input and more than one output. Multiple types of resources can form the input and the output. In figure 1 there are two significant examples of resource distribution. Resources Plan and Money, needed for purchasing the Material, enter to the Purchase process. Sales process produces output Money that enters into two another processes - Purchase process and Acquisition process.


136

Fig. 1. Enterprise Value Chain

The simulation of the value chain process is used to monitor the competitiveness of the business. Model elements show processes that increasing the value of corporate resources. Each step of the process increases the value of the company output, and therefore it can be understood as a value chain [1]. The value chain is the set of mutually interdependent activities that are interrelated together by their inputs and outputs.

Enterprise value system shows the flow of resources between participants within the enterprise and it can be obtained by analysis of the company value chains. The value chain shows the flow of resources across business processes [14].

Figure 1 shows an example of the value chain depicting the flow of resources between the enterprise and business partners. The model example consists of 5 processes:

• Purchase process expresses the purchase of the raw material from vendors.

• Production process is internal conversion process creating the product.

• Sales process illustrates the sales of the created product to the customer.

• Acquisition process arranges labour for Planning process.

• Planning process prepares purchase plane.

For purposes of the planning and detailed analysis of business value chains it is necessary to record the flow of resources and their changes in time. For these cases, it is possible to use the simulation of the flow of resources across business processes. In every step of the simulation an exchange process performs exchanging of resources and the conversion process creates a new product or modifies characteristics of an existing product. In case of complex value chain the simulation can determine which links are inefficient and where is the place for the subsequent optimization. The simulation can be also used for analysis of the economic situation of the enterprise, such as stores status, financial estimates, an efficiency of the production, or logistics. You can also simulate the way of a product from an initial purchase of raw materials, through its production and finishing by sale to determine the total financial and time costs for one product.

II. OBJECT-VALUED PETRI NET Object-valued Petri net is an extension of P/T Petri net.

This extension has been introduced in [8]. Object-valued Petri nets are used as formalism for validation and synchronization of complex object models.

Definition 3.1: Object-valued Petri net

Petri net is extended to a 6-tuple (P, T, F, V, R, C), where:

• 𝑃 is a finite nonempty set of places,

• 𝑁 is a finite nonempty set of transitions,

• 𝑇 ∩ 𝑃 = ∅ (P and T are disjoint),

• 𝐹 ⊆ 𝑃×𝑇 ∪ 𝑇×𝑃 is a finite set of arcs (flow relation),

• V is a finite set of object data types,

• R is a finite set of transforming functions 𝑅:𝑃 ∪ 𝑇 →𝜓(𝑉), where 𝜓(𝑉) is the power of the set of object data types.

• C is a set of capacity function. 𝐶:𝑃 → 𝑁 ∪ 𝜔,𝑁 ⊆ ℕ and 𝜔 denotes infinite.

• 𝑀!:𝑃 → 𝑉!" is the initial marking of the token. ∀𝑝 ∈𝑃:𝑀!(𝑝) ∈ 𝑅(𝑝)!", where 𝑅(𝑝)!" is the multiple set of the object data type tokens in p.

The main idea of the Object-valued Petri net is an object-valued token that provides adequate expressivity to describe resources represented by complex object structures. The token carries basic information to identify the specific object instance. Initial marking consists of the multiple set of object data types deployed across the net. Firing of each token means change in marking of the net and also change of the token type. However token identification remains and therefore we can identify the token in every step of the simulation process. If the model is partly linked with the Object-valued Petri net theory we have to define the path of tokens. Formalism itself defines necessary basis to create the model, unfortunatelly that does not ensures the sequence of movements into desirable result. Object-valued Petri net realizes transition as soon as the transition is feasible. Nevertheless the real model can require other conditions to realize the transition (for instance lazy constructions). Therefore we have to state the new definition of the path and pass of the model.

Definition 3.2: Path of the OV-PN

Let OV-PN = (P, T, F, V, R, C) be an Object-valued Petri net with initial marking M0. The path from the place 𝑢! ∈ 𝑃 ∪𝑇 to following place 𝑢! ∈ 𝑃 ∪ 𝑇 is the sequence (𝑢!, 𝑢!,… , 𝑢!), where (𝑢! , 𝑢!!!) for 1 ≤fi ≤ n.


137

Definition 3.3: Pass of the OV-PN

Object-valued Petri net OV-PN = (P, T, F, V, R, C) with initial marking M0 is feasible when:

1. Must exist an initial place where .

2. Must exist exactly one final place where .

3. Every place lies on the pass between initial place i and final place o.

In this context the first condition is understood as a marking of the input of the model that can be represented by more than one input parameter. If this condition is set to be strict to the value of input marks the model cannot realize calling of the method with more than one input parameter. The output of the model is usually one because of the standard method construction in object-oriented paradigm [2]. Third condition expresses the fact that every place and every transition exists on the path between the initial place and the final place. Therefore the Object-valued Petri net should not have blind paths and every call in the model should be reachable form initial place by passing finite number of transitions representing a flow relation F (similarly to [16]). Similarly to the initial place every place in the model exists in the flow relation F is able to reach the final place of the model.

Boundedness and safeness

The ordinary Petri net (P/T Petri net) defines the boudedness mechanism to limit the tokens in all reachable markings. A place in the Petri net is called k-bounded if it does not contain more then k tokens in every making in the net, including the initial marking. Moreover the Petri net is bounded if and only if its reachability graph is finite. The special case of the boundedness is safeness attribute. If the net is 1-bounded it is called safe.

Object model synchronized by Petri net mechanism can be bounded at the places level as in ordinary Petri net. Every method can produce more than one output during the simulation. Places may store these outputs as Object-valued tokens (similar to colour evaluation in [6]). By applying safeness rule the places in model stores only one object-valued token and the model becomes less complex.

Conservation

Created model cannot be strictly conservative. In the first step in figure 1 the Purchase process consumes two inputs and produces one output. The Sales process consumes only one input and produces two outputs - two object-valued tokens parameterized as Money. Moreover the Purchase process requires a synchronization mechanism. The model cannot have a constant token count for every marking from set of the reachability set

Liveness and deadlock

All methods in object-oriented paradigm can be executed more than once [4]. However by executing some method an

internal state of the object can be altered. That means if we need to apply liveness property to whole model, every method must be considered as an atomistic operation.

Generally the transition 𝑡 ∈ 𝑇 is alive if:

∀𝑝 ∈• 𝑡:𝑀 𝑝 ≠ ∅ and 𝑝 ∈ 𝑡• ∶ 𝑀 𝑝 = ∅.

It means that transition becomes active, if there are tokens on all transition’s entrances and the place that follows the transition is empty. The net is alive if there is at least one live transition in every step of simulation process otherwise a deadlock occurs. Deadlock is solved on a higher abstraction level and requires user intervention.

III. OBJECT-VALUED PETRI NET EXTENSIONS The main condition of the value chain is cyclicality.

However the Object-valued Petri net has two definitions that limit the path of the net (definition 3.2) and pass of the net (definition 3.3). First definition says that O-V Petri net with a specific marking M0 has a specific sequence from one place to another. The value chain has also specific sequence that defines the path of the chain. Moreover the cyclic chain consists of many single paths connected to each other. To express a general value chain principle with the Petri net theory we have to define a cyclic Petri net:

Definition 4.1: Cyclic Object-valued Petri net

A marked Petri net (OV-PN, M0) is cyclic Petri net if from every reachable marking M it is possible to return into M0 (i.e.

).

According to [9] we must also define the inverse of an ordinary Petri net:

Definition 4.2: The inverse of an Object-valued Petri net

For a Petri net OV-PN, its inverse 𝑂𝑉 − 𝑃𝑁 = (𝑃,𝑇,𝐹) is given by:

• 𝑇 = 𝑡|𝑡 ∈ 𝑇 and

• 𝐹 𝑡, 𝑝 = 𝐹 𝑝, 𝑡 and 𝐹 𝑝, 𝑡 = 𝐹 𝑡, 𝑝 for every 𝑝 ∈ 𝑃 and 𝑡 ∈ 𝑇

The definition of the inversion of the Object-valued Petri net is presented for completeness only. In the real model of the value chain there is usually no backward path. For instance the company cannot convert the product to the raw material. On the other hand this definition gives the robust tool to verify cyclicality of the net. Algorithms to verify cyclicality could be simplified to perform the token verification. Every token in the Object-valued Petri net have the unique instantiation number. The inner value of the Object-valued token is changed during the pass of the net, however instantiation number stays unchanged despite the value transformation. The modelling tool can set up the initial marking and make finite steps of the firing. If the net is cyclic the specific instantiation returns to the initial marking.

According to the facts above we can redefine the original Object-valued Petri net tuple for modelling the value chain:

Pi∈ ∅=• i

Po∈∅=•o

TPu ∪∈

∑ ∑∈ ∈≠ℜ

S S iii i

pMpMMρ ρ

)()(:)( 00

),(,( 00 MPNOVMMPNOVM −ℜ∈⇒−ℜ∈


138

Definition 4.3: Petri net for a value chain simulation

Petri net for value chain simulation is a 5-tuple (P, T, F, S, R), where:

• 𝑃 is a finite nonempty set of places,

• 𝑇 is a finite nonempty set of transitions,

• 𝑇 ∩ 𝑃 = ∅ (P and T are disjoint),

• 𝐹 ⊆ 𝑃×𝑇 ∪ 𝑇×𝑃 is a finite set of arcs (flow relation),

• S is a finite set of resources,

• R is a finite set of transforming functions 𝑅:𝑃 ∪ 𝑇 →𝜓(𝑆), where 𝜓(𝑆) is the power of the set of resources.

Moreover we must claim boundedness of elements:

• ∀𝑝 ∈ 𝑃∃𝑡 ∈ 𝑇: 𝐹 𝑝, 𝑡 ∈ 𝐹,

• ∀𝑝 ∈ 𝑃∃𝑡 ∈ 𝑇: 𝐹(𝑡, 𝑝) ∈ 𝐹,

• ∀𝑡 ∈ 𝑇∃𝑝 ∈ 𝑃: 𝐹 𝑝, 𝑡 ∈ 𝐹,

• ∀𝑡 ∈ 𝑇∃𝑝 ∈ 𝑃: 𝐹 𝑡, 𝑝 ∈ 𝐹

and their connection to cyclic model:

• ∀𝑝 ∈ 𝑃:𝑀 𝑝 ∈ ℜ 𝑀! ,𝑀! ∈ ℜ 𝑀 𝑝 ,

• every 𝑡 ∈ 𝑇 is reachable from any 𝑝 ∈ 𝑃 in limited count of steps.

Naturally we also must specify the properties of the new definition:

Boundedness and safeness

In the value chain the one resource can be transferred into the more than one process (i.e. money to buy a new material and money to fund innovations).

The safeness of the net is the matter of discussion. In fact there are two possibilities. The net can be safe and that means one place stores only one object-valued token. That property can be convenient to verify the whole conversion process and user can focus to one resource and transformation process. This simulation is similar to redefining business processes in the company. By applying the safeness property the whole model became simple to understand and verification of the process flow is much easier.

The second view on the value chain simulation is to get statistic data and optimize workflow parameters. The model must simulate the conversion process with more than one Object-valued token. A typical example is a production process creating the specific product. At the beginning of the simulation the company needs to know how many products must be created to cover money for a product development. In the short term the first view can set the margin of the seller and express the production process. The simulation of the second view takes longer and works with multiple tokens. The price of the product decreases with time and by the long term simulation the company can reveal if the price model has been set correctly. Therefore the second view cannot be safe form a Petri net point of view.

Liveness and deadlock

Object-valued Petri net must fulfill liveness property because of the object-oriented paradigm construction. Cyclic Petri nets are based on general OV-PN, but it differs on boudedness and safeness property. Therefore liveness property must be changed.

Transition 𝑡 ∈ 𝑇 is alive if:

• ∀𝑝 ∈• 𝑡:𝑀 𝑝 ≠ ∅

• 𝑝 ∈ 𝑡• ∶ 𝑀 𝑝 + 𝑈(𝑡) ≤ 𝐾(𝑝), where 𝑈(𝑡) is a number of resources produced by transition t

The net is live if there is at least one live transition in every step of the simulation process.

The value chain consists of processes and links. Process itself consists of atomic operations that can be repeated infinitely with the same result. For example: a production process is defined by precise methodology how to produce a product on the serial assembly. The parameters of the process are set at the beginning of the serial assembly (i.e. speed of the line) and usually remains unchanged during production. Therefore we naturally apply the liveness property to the Object-Valued Petri net model of the value chain. All processes remain the same despite the fact that the Object-valued token flow through the process.

The process itself can have more than one input link. In figure 1 the Purchasing process requires the Money and the Plan. Plan process takes more time to create a specific purchasing plan and sales process delivers the Money immediately after the product has been sold. In this specific step of the model simulation the deadlock occurs. That means the execution of the Purchasing process is delayed until both inputs provided with links are available. These cases can be problematic and generally can be solved on a higher abstraction level, i.e. modelling tool. If the model is validated a deadlock cannot occur because of the cyclicality property of the value chain. Moreover that implies that every process must be reachable.

Conservation

The conservation property means that one object-valued token cannot be duplicated when it passes the transition and the transition has the same number of inputs and outputs. In other words the count of the Object-valued tokens is same in every step of the simulation. The basic models of the value chain can be conservative. However most models in the real world are more complex and there is big challenge to apply the conservation rule to express a chain of resources. For example money in the real world is an input to more than one process - production process, planning, development, etc.. From a Petri net point of view we must duplicate tokens with specific inner attributes and sends them to other transitions. Therefore the model does not have the constant token count for every marking from the reachability set.

IV. VALUE CHAIN SIMULATION EXAMPLE During the transformation the value chain elements are

mapped into the modified cyclic Object-Valued Petri net


139

elements. Similar transformation process can be found in [18]. Processes that exist in the value chain will be represented as transitions and all properties mentioned above will be applied. Links that exist in the value chain will be composed of two arcs and one place. The Arc indicates the direction of an Object-valued token and the Place carries the Object-valued token(s) that represents information.

For example we used the value chain from figure 1. The transformed result is shown in Figure 2. The Petri Net consists of five transitions and six places. The simulation starts with the Token set on the place Product and it takes five cycles before repeating:

1. The company has product to sell. The Product enters into the Sales process. That generates Money for the Purchase process and the Acquisition process (token is divided into two places).

2. Money enters into the Acquisition process and creates the Computer work. The Purchase process is not executed because of insufficient Plan input.

3. The Computer work enters into the Planning process and generates the Plan.

4. The Purchase process transition has all needed inputs to perform firing. Money and Plan tokens are exchanged for Material.

5. In the last step the Material enters into the Production process and creates a new Product.

Fig. 2. Transformed value chain

The model is cyclic. That means the cycle 1 does not have to be the first and all steps are realized infinitely. The order is only that matters. The key in the value chain simulation process, except the synchronization primitives, is the Object-valued token. In the beginning of the simulation the Object-valued token is parameterized as a Product. In the first step the token is transformed into the Money and split into the two tokens with specific ratio used to determine the inner value. The association of the tokens to the value chain can be identified through the instantiation ID, and inner values can be changed as needed. Moreover in the second cycle, one of tokens enters to Acquisition process and transforms (parameterizes) into the Computer work. That means the token has different inner values and even a data structure. Analogical transformation changes the Computer work into the Plan

structure in the third cycle. In the fourth cycle two tokens are merged together by Purchase process and the result is the Object-valued token parameterized as a Material. The merge process can be performed because of the same token instantiation ID. In the last step the token is transformed to the Product by the Production process and the chain is closed.

There is only one token in figure 2 and all links are limited to 1. We can simulate the whole conversion process with more than one token and we can establish a capacity function on every place in the net. All splits and merges of tokens are identified by instantiation ID and therefore they are distinguishable. That means we can recognize the specific token as a part of the cyclic chain and the base for optimization of processes in the model.

V. CONCLUSION The paper introduced a new formalism based on the Object-

Valued Petri net to create, synchronize and manage cyclic models of the value chain. The paper described a basic theory of economic models based on conversion and exchange processes and introduced a value chain term in the first part. The paper also described why current formalisms such as neural network and state machines are not suitable to build a value chain model. Paragraph 3 shows an Object-value Petri net theory focused on the path and pass of the net. This theory is suitable to build a value chain model because the Object-valued token can be used to express resources of the value chain and their transformations. However Object-valued Petri net are not cyclic and have strictly defined pass and path of the net. The definition of an extended Object-valued Petri net formalism - definition 4.3 - solves this problem and adds the cyclicality. All basic properties are discussed and redefined for the new cyclic object-oriented model. An extended Object-Valued Petri net formalism solves all problems mentioned in the second paragraph and can be applied to any cyclic model. The proposed formalism has been verified on the ordinary value chain and basic steps of the simulation are described in paragraph 4.

ACKNOWLEDGMENT The paper was supported by the grant reference no.

SGS08/PRF/2013 provided by Ministry of Education, Youth and Sports.

REFERENCES

[1] Fiala, J., Ministr, J.: Pruvodce analyzou a modelovanim procesu, VB-TU Ostrava, 2003, ISBN 20-248- 0500-6.

[2] Shilling, J.: Three Steps to Views: Extending the Object-Oriented Paradigm, OOPSLA 89 Proceedings, 1989.

[3] Hunka, F., Zacek, J., Melis, Z., Sevcik, J.: REA Value Chain versus Supply Chain, Scientific Papers of the University of Pardubice, 2011, s. 68-77, ISSN 1211-555X.

[4] Zacek J., Hunka F.: CEM: Class executing modeling, World Conference on Information Technology, 2010, page 1597-1601, ISSN 1877-0509.

[5] Sklenar, J., Caruana, E.: Using Timed Petri Nets in Discrete Simulation, Proceedings of the Industrial Simulation Conference, 2004 (ISC-2004), Malaga, 2004, page 7-11.

[6] Zhao, X., Wei, C., Lin, M., Feng, X., Lan, W: Petri Nets Hierarchical Modelling Framework of Active Products Community, Advances in Petri


140

Net Theory and Applications, 2010, page 153-174, ISBN 978-953- 307-108-4.

[7] Martinik, I.: Methodology of object-oriented programmatic system development using theory of Object Petri nets, dissertation thesis, VB-TU Ostrava, 1999.

[8] Zacek, J., Hunka, F.: Object model synchronization based on Petri net, 17th International Conference on Soft Computing MENDEL 2011, Brno University of Technology, Faculty of Mechanical Engineering, 2011, s. 523-527, ISBN 978-80-214-4302-0.

[9] Bouziane, Z., Finkel, A.: Cyclic Petri Net Reachability Sets are Semi-linear Effectively Constructible, CONCUR’11 Proceedings of the 22nd international conference on Concurrency theory, Springer-Verlag, Berlin, ISBN: 978-3-642-23216-9.

[10] Geerts, G. L., McCarthy, W. E.: Using Object Oriented Templates from the REA Accounting Model to Engineer Business Process and Tasks, Paper presented at European Accounting Congress, Gratz, Austria, 1997.

[11] Sochor T., Davidova A.: Potential of Multilevel SPAM Protection in the Light of Current SPAM Trends, 10th International Conference on Networking, Sensing and Control, Paris-Evry University, France, 2013.

[12] Kersten W., Blecker T., Ringle M. Ch.: Managing the Future of Supply Chain, Eul Verlag, ISBN 978-3-8441-0180-5, 2012.

[13] Ding Z., Liu J., Wang J.: A Petri Net Based Automatic Executable Code Generation Method for Web Service Composition, Proceedings of the 2012 International Conference on Information Technology and Software Engineering, pp 39-48, ISBN 978-3-642-34530-2, 2013.

[14] van Hee M. K., Sidorova N., van der Werf M. J.: Business Process Modeling Using Petri Nets, Transactions on Petri Nets and Other Models of Concurrency VII, pp 116-161, ISBN 978-3-642-38142-3, 2013

[15] Wang W. J., Ip H. W., Muddada R. R., Huang L. J., Zhang J. W.: On Petri net implementation of proactive resilient holistic supply chain networks, The International Journal of Advanced Manufacturing Technology, ISSN 0268-3768, Springer-Verlag, 2013.

[16] Darondeau P., Demri S., Meyer R., Morvan Ch.: Petri Net Reachability Graphs: Decidability Status of First Order Properties, Logical Methods in Computer Science, Vol. 8(4:9), pp. 1–28, 2012.

[17] Bocewicz G., Wójcik R., Banaszak Z.: Cyclic Scheduling for Supply Chain Network, Trends in Practical Applications of Agents and Multiagent Systems, Springer Berlin Heidelberg, pp 39-47, 2012.

[18] Li J., Zhou M., Xianzhong D.: Reduction and Refinement by Algebraic Operations for Petri Net Transformation, Systems, Man and Cybernetics, Part A: Systems and Humans, IEEE Transactions on (Volume:42 , Issue: 5), pp 1244 - 1255, ISSN 1083-4427, 2012.


141

Generalized Net Model for Telecommunication Processes in Telecare Services

Mikhail Matveev, Velin Andonov, Krassimir Atanassov (IEEE Member)

Institute of Biophysics and Biomedical Engineering Bulgarian Academy of Sciences, Sofia, Bulgaria

[email protected], [email protected], [email protected]

Maria Milanova Multi-profile Hospital for Active Medical Treatment

and Emergency Medicine "N.I.Pirogov" Sofia, Bulgaria

[email protected]

Abstract—In a series of papers, Generalized Net (GN) models of processes, related to tracking changes in the health status of adult patients, have been presented. The contemporary state-of-the-art of the telecommunications and navigation technologies allow these models to be further extended to the case of active and mobile patients. This requires the inclusion of patient’s current location as a new and significant variable of the model. Various opportunities are considered for the retrieval of this information, with a specific focus on the optimal ones, and a refined GN model is herewith proposed.

Index Terms—Generalized nets, Modelling, Telecare, Telecare services.

I. INTRODUCTION In a series of papers, Generalized Net (GN, see the

Appendix) models of processes, related to tracking the changes in health status of adult patients have been presented (see, e.g., [1, 2]). They are a continuation of the ideas for description of processes, taking place in hospital units, using the apparatus of GNs (see, e.g., [3–5]). The so constructed nets give the possibility to model the logical conditions for the realization of the processes, to simulate these processes, as well as search ways for their optimization.

In [1], the processes for generation of signals from sensors around adult patients and their transmission through different telecommunication tools to the respective hospital units, have been described. In [2], a GN-model of the telecommunication processes between the adult patients and hospital units, has been discussed. Below, a GN-model of the processes for signal classification and the reaction of the medical staff of the hospital units, is constructed. In the Appendix, short remarks of GNs are given.

II. THE GENERALIZED NET MODEL The GN model (see Fig. 1) consists of. • six transitions Z1, Z2, Z3, Z4, Z5 and Z6. • sixteen places l1, l2, …, l16. • four different types of tokens representing the patients,

the dispatchers that monitor the signals from the sen-sors, the medical doctors who examine the patients and the medical specialists.

The tokens π1, π2, ..., πk which represent the patients enter the net in place l4 with initial characteristic “patient; name of the patient; current health status”.

The tokens δ1, δ2, …, δl which represent dispatchers enter the net in place l8 with initial characteristic: “dispatcher; name of the dispatcher; information about all received signals”.

The tokens μ1, μ2, ..., μm which represent the medical doctors who examine the patients enter the net in place l9 with initial characteristic: “medical doctor; name of the medical doctor; specialty”.

The tokens σ1, σ2, ..., σm which represent the medical specialists who examine the patients enter the net in place l9 with initial characteristic: “medical doctor; name of the medical doctor; specialty”.

The six transitions will be described in details below. The first transition Z1 has the form:

Z1 = ⟨l4, l10, l15, l1, l2, l3, l4, R1⟩,

where

,

15

10

4,43,42,44

43211

truefalsefalsefalselfalsefalsefalsetruelWWWfalsel

llllR =

and the predicates in the index matrix R1 have the meanings: • W4,2 = “the sensor detected a change in patient’s condition”, • W4,3 = “the patient should be transported to hospital” • W4,4 = ¬ W4,2 ,

where ¬ P is the negation of the predicate P. When the truth-value of the predicate W4,2 is true, the token

πi enters place l2 with characteristic “signal of the sensor about the current patient”.

When the truth-value of the predicate W4,3 is true, the token πi enters place l3 with characteristic “name of the patient; current status”.

In place l4, the π-tokens receive the characteristic “current status of the patient”.


142

Fig. 1. GN model of the telecommunication processes,

taking place in telecare services

The second transition Z2 has the following form:

Z2 = ⟨l2, l8, l5, l6, l7, l8, R2⟩, where

,

8

7,26,25,22

87652

truefalsefalsefalselfalseWWWlllll

R =

and the predicates in the index matrix R2 have the meanings: • W2,5 = “medical doctor should be sent to perform examin-

ation of the patient at home”, • W2,6 = “no action is necessary”, • W2,7 = “the patient should be transported to a medical center”.

When the truth-value of the predicate W2,5 is true, the token πi enters place l5 with characteristic “a decision to visit the patient has been taken”.

When the truth-value of the predicate W2,6 is true, the token πi enters place l6 with characteristic “a decision to ignore the signal has been taken”.

When the truth-value of the predicate W2,7 is true, the token πi enters place l7 with characteristic “a decision to transport the patient to a medical center has been taken”.

The third transition Z3 has the form:

Z3 = ⟨l1, l5, l9, l9, l10, R3⟩, where

,

10,99,99

5

1

1093

WWlfalsetruelfalsetruelll

R =

and the predicates in the index matrix R3 have the meanings • W9,10 = “a medical doctor should be sent to examine the

patient”, • W9,9 = ¬ W9,10.

In place l9, the μ-tokens do not obtain any new charact-eristics.

When the truth-value of the predicate W9,10 is true, the corresponding μi token representing the medical doctor enters place l10 with characteristic “name of the medical doctor who will visit the patient”.

The forth transition Z4 has the following form:

Z4 = ⟨l7, l12, l13, l11, l12, R4⟩, where

,

13

12,1211,1212

7

12114

truefalselWWltruefalselll

R =

and the predicates in the index matrix R4 have the meanings • W12,11 = “specialists should be sent to bring the patient to

the hospital”; • W12,12 = ¬ W12,11.

In place l11 the current token σi receives the characteristic “names of the specialists who will bring the patient to the hospital”.

In place l12 the tokens receive the characteristic “names of the staff on duty”.

The fourth transition Z5 has the form:

Z5 = ⟨l3, l11, l13, l14, R5⟩, where

.

11

3

14135

falsetrueltruefalselll

R =

In place l13, the tokens receive the characteristic “time for completing the transportation of the patient”.

In place l14, the tokens receive the characteristic “condition of the patient upon arrival at the hospital”.

The sixth, final, transition Z6 has the following form:

Z6 = ⟨l14, l16, l15, l16, R6⟩, where

,

16,1615,1616

14

16156

WWltruefalselll

R =

and the predicates in the index matrix R6 have the meanings


143

• W16,15 = “all medical procedures are completed”; • W16,16 = ¬ W16,15.

In place l15, the current πi token receives the characteristic “condition of the patient upon discharge from hospital”.

In place l16, the current πi token receives the characteristic “condition of the patient during the procedures”.

Finally, we mention that place l8 represents the processes, described by the GN from [3], while places l14, l15 and l16 correspond to the processes, modeled by the GN from [4]. On the other hand, the present GN model elaborates into further details the basic idea presented in [5].

III. CONCLUSIONS The so constructed GN model traces the logical stages of the

final part of the process of communication between the sensors connecting mobile adult patients and the staff of the respective hospital unit. The developed model can be used for simulation of the processes of decision making of the appropriate specialists, who must either visit the respective adult patient or to transport him/her to the hospital unit. The model permits simulation of different scenarios e.g. the situation, in which many patients simultaneously require medical assistance.

ACKNOWLEDGMENT This work was partly funded by the project FP7-PEOPLE-

2009-IRSES-247541-MATSIQEL.

REFERENCES [1] Andonov, V., et al. Generalized Net Model for Telecare

Services, IEEE Conf. “Intelligent Systems”, Sofia, Bulgaria, 6-8 Sept. 2012, 221-224.

[2] Andonov, V., T. Stojanov, K. Atanassov, P. Kovachev, General-ized Net Model for Telecommunication Processes in Telecare Services, 1st International Conference on Telecommunications and Remote Sensing, Sofia, Bulgaria, 29-30 Aug. 2012, 158-162.

[3] Chakarov, V., et al. Generalized net model for some basic clinical administrative decision making. 1st European Conf-erence on Health Care Modelling and Computation Craiova, Aug. 31 - Sept., 2, 2005, 72-78.

[4] Matveev M., et al. Dynamic model of intensive care unit work-flow based on generalized nets. International Electronic Journal “Bioautomation”, Vol. 2, 2005, 85-92.

[5] Shannon, A., et al. The generalized net modelling of information healthcare system. Int. Conf. "Automatics and Informatics'06", Sofia, 3-6 Oct. 2006, 119-122.

[6] Atanassov, K., Generalized Nets, World Scientific, Singapore, 1991.

[7] Atanassov, K., On Generalized Nets Theory, “Prof. Marin Dri-nov” Publishing House of the Bulgarian Academy of Sciences, 2007.

APPENDIX: SHORT REMARKS ON GENERALIZED NETS Generalized Nets (GNs, see [6, 7] are extensions of the

apparatus of mathematical modelling of Petri Nets and other modifications of theirs. GNs are a tool intended for the detailed modelling of parallel and concurrent processes.

A GN is a collection of transitions and places ordered according to some rules (see Fig. 2). The places are marked by circles. The set of places to the left of the vertical line (the transition) are called input places, and those to the right are called output places. For each transition, there is an index matrix with elements called predicates. Some GN-places contain tokens – dynamic elements entering the net with initial characteristics and getting new ones while moving within the net. Tokens proceed from an input to an output place of the transition if the predicate corresponding to this pair of places in the index matrix is evaluated as “true”. Every token has its own identifier and collects its own history that could influence the development of the whole process modelled by the GNs.

Two time-moments are specified for the GNs: for the beginning and the end of functioning, respectively.

A GN can have only a part of its components. In this case, it is called reduced GN. Here, we shall give the formal definition of a reduced GN without temporal components, place and arc capacities, and token, place and transition priorities.

Formally, every transition in the used below reduced GN is described by a triple: Z = ⟨L′, L″, r⟩, where:

l'1

l'm

Z

l"1

. . .. . .

l'i

. . .. . .

l"n

l"j

Fig. 2. A GN transition

(a) L′ and L″ are finite, non-empty sets of places (the tran-sition’s input and output places, respectively); for the transition these are

L′ = m21 'l,...,'l,'l and L″ = n21 "l,...,"l,"l ;

(b) r is the transition’s condition determining which tokens will pass (or transfer) from the transition’s inputs to its outputs; it has the form of an Index Matrix (IM):


144

)1,1(

)(

'...'

...'

"..."..."

,

,

1

1

njmi

predicater

r

l

l

llll

r

ji

ji

m

i

nj

≤≤≤≤

−

=

where ri,j is the predicate that corresponds to the i-th input and j-th output place. When its truth value is “true”, a token from the i-th input place transfers to the j-th output place; otherwise, this is not possible.

The ordered four-tuple

E = ⟨A, K, X, Φ⟩

is called a reduced Generalized Net if: (a) A is the set of transitions; (b) K is the set of the GN’s tokens; (c) X is the set of all initial characteristics which the tokens

can obtain on entering the net; (d) Φ is the characteristic function that assigns new

characteristics to every token when it makes the transfer from an input to an output place of a given transition.

Many mathematical operations (e.g., union, intersection and others), relations (e.g., inclusion, coincidence and others) and operators are defined over the GNs. Operators, being of six types (global, local, hierarchical, reducing, extending and dynamic operators) change the structure of the GN, the strategies of token transfer, etc.


145

Design of a Dynamic CMOS Incrementer/Decrementer

and a Parallel Cascading Architecture

B. Archanadevi, V. Anbumani, T. Malathy,

P. Balasubramanian

Department of Electronics and Communication Engineering

S. A. Engineering College, Chennai 600 077, India

[email protected]

N. E. Mastorakis

Division of Electrical Engineering and Computer Science

Military Institutions of University Education

Hellenic Naval Academy, Piraeus 18539, Greece

[email protected]

Abstract—Dynamic CMOS based transistor level designs of

incrementer/decrementer circuit is presented in this work. The

design of a new 8-bit decision module is first described. This is

followed by elucidation of an original cascading architecture to

realize larger size incrementer/decrementer circuits. From

SPICE simulations corresponding to a 0.25µµµµm CMOS process

technology, it is inferred that an 8-bit incrementer/decrementer

embedding the new decision module macro dissipates 48% less

power for incrementing and 30% less power for decrementing

than the one incorporating a conventional macro. Further, 16-

bit and 32-bit incrementers/decrementers constructed using the

proposed cascade consume 21% and 23% reduced average

power for increment and decrement operations respectively than

their conventional counterparts.

Keywords—Incrementer/decrementer; Dynamic CMOS logic;

Low power; Full-custom design; Digital integrated circuit

I. INTRODUCTION

Microprocessors, microcontrollers and application-specific processors [1] usually require a program counter to perform operations such as incrementing and/or decrementing with respect to address calculations for data access, branching and storage functions. In this context, an incrementer/decrementer circuit plays an important and crucial role [2] [3]. Generally, incrementer/decrementer circuits can be optimally designed in a full-custom manner at the transistor level [2] – [4] rather than a semi-custom synthesis using standard cells [5]. In this regard, references [3] and [4] deal with the design of an 8-bit incrementer/decrementer circuit based on the principle of priority encoding [4] [6] [7]. Based on the survey of existing literature [2] – [5] and to the best of our knowledge, no other better circuit level design exists compared to reference [3]. The basic circuit module can eventually be cascaded to implement larger size increment/decrement functionality at the physical level, which can be subsequently made available either on-chip or off-chip. Moreover, circuit level designs can be better optimized for area/delay/power parameters than gate level designs. Reference [3] has also put forward a scheme to realize higher order incrementer/decrementer circuits through multi-level lookahead and folding techniques.

The rest of this paper is structured as follows. Section 2 describes the least significant one bit principle underlying the design of the decision module block. Section 3 details the new 8-bit dynamic CMOS based decision module macro design, which forms the ‘heart’ of the incrementer/decrementer circuit. In Section 4, a novel cascading architecture proposed to build larger size incrementer/decrementer circuits is presented, followed by the documentation of the simulation results and the concluding remarks in Section 5.

II. INCREMENT/DECREMENT – OPERATING PRINCIPLE

The basic template for performing increment/decrement operation is illustrated through the circuit shown in Figure 1.

Fig. 1. Priority encoding based 8-bit incrementer/decrementer module [3] [4]


146

The 8-bit incrementer/decrementer circuit depicted in Figure 1 contains three sub-modules viz. data-in selector, decision module, and data-out selector. I0 to I7 represent the primary inputs, D0 to D7 denote the data inputs to the 8-bit decision module, and Y0 to Y7 signify the primary (incremented/decremented) outputs. Here, I0 (and hence D0) and Y0 are the least significant bits, and assume the highest priority among the input and output bits. The data-in selector consists of an Inc/Dec’ signal whose value indicates the type of operation to be performed; increment if Inc/Dec’ = 1, and decrement otherwise. Clk refers to the clock signal which is used to synchronize the circuit operation. During the falling-edge of clock, the 8-bit decision module is pre-charged as all the nMOS transistors (n1 to n8) turn-on and eventually the decision module outputs are driven to 1 (set). However, during the rising-edge of the clock, nMOS transistors n1 to n8 will turn-off and pMOS transistors p1 to p8 turn-on. Based on the priority, the least significant output bit corresponding to a least significant input bit, and the less significant output bits are retained as 1, while the more significant output bits are driven from 1 to 0 during the clock evaluation phase. Due to the high switching activity encountered with initial ‘setting-up’ of all the outputs and subsequent high-to-low transitions of select output bits, and with widespread usage of pMOS transistors, the average power dissipation of the incrementer/decrementer circuit shown in Figure 1 is likely to be very high. LA_in, LA_int, and LA_out are the respective input, internal and output lookahead signals, which are used to build larger size incrementers/decrementers via cascading.

To explain the increment/decrement circuit operation, let us consider an example primary input as 11001011. To enable the decrement function, Inc/Dec’ signal is set to 0 in Figure 1. Hence, the output of the data-in selector is the same as its input, i.e. D7 to D0 = I7 to I0 = 11001011. According to the ‘least significant one bit’ (LSOB) principle [3], the output of the 8-bit decision module is equal to 00000001, since a 1 is present in the least significant position of the data input. As a result, only the least significant but highest priority output bit is driven to 1, while the other more significant but low priority output bits are driven to 0. By EXORing the data input and the decision module output in the data-out selector module, we obtain the final result as 11001010, which is indeed the desired decrement value. Assuming the same primary input, to perform increment operation, Inc/Dec’ signal is set to 1 in the data-in selector module. This complements the primary input, and hence D7 to D0 becomes 00110100. With this as the input supplied to the decision module, as per the LSOB principle, the decision module will output 00000111. EXORing this with the primary input (11001011); the data-out selector produces 11001100, which is the required increment value.

III. PROPOSED 8-BIT DECISION MODULE DESIGN

A new 8-bit decision module has been designed on the basis of the LSOB principle, as shown in Figure 2, utilizing the dynamic CMOS logic style. This decision module can in fact replace the one shown in Figure 1 to facilitate binary increment or decrement operations on demand. It can be seen in Figure 2 that there is no separate lookahead output, but only a lookahead input provision. Nevertheless for cascading, we resort to a parallel schema which is discussed in Section 4.

The decision module macro shown in Figure 2 implements the following output equations.

Y0 = D0 + D1 + D2 + D3 + D4 + D5 + D6 + D7 (1)

Y1 = D0’ (D1 + D2 + D3 + D4 + D5 + D6 + D7) (2)

Y2 = D0’ D1’ (D2 + D3 + D4 + D5 + D6 + D7) (3)

Y3 = D0’ D1’ D2’ (D3 + D4 + D5 + D6 + D7) (4)

Y4 = D0’ D1’ D2’ D3’ (D4 + D5 + D6 + D7) (5)

Y5 = D0’ D1’ D2’ D3’ D4’ (D5 + D6 + D7) (6)

Y6 = D0’ D1’ D2’ D3’ D4’ D5’ (D6 + D7) (7)

Y7 = D0’ D1’ D2’ D3’ D4’ D5’ D6’ D7 (8)

With reference to Figure 2, during the falling-edge of the clock (Clk), pMOS transistors pc1 to pc8 turn-on, and outputs Y0 to Y7 are reset. During the leading clock edge, the pMOS transistors turn-off, and the circuit processes data inputs and produces the desired outputs. For example, given LA_in is 1, and for a rising clock edge, if D1 is 1, and all the other inputs are 0, nMOS transistors ev1, ev2 and ev3 would turn-on thus resulting in outputs Y0 and Y1 becoming set, while the remainder of the outputs continue to remain reset. Since the outputs are initially reset, and only the requisite outputs are set during the evaluation phase, the average switching activity would be potentially less for this decision module compared to that shown in Figure 1, highlighting the likelihood of a low power design.

D0

D1

D2

D3

D4

D5

D6

D7

Y7

Y6

Y5

Y4

Y3

Y2

Y1

Y0

LA_in

Clk

pc8

pc7

pc6

pc5

pc4

pc3

pc2

pc1

ev1ev3

ev2

Fig. 2. Proposed nMOS based 8-bit decision module macro

Comparing the 8-bit decision modules of Figures 1 and 2, it is evident that pMOS transistors are predominant in the conventional design, while nMOS transistors are widely prevalent in the proposed design – this could indeed translate


147

into substantial savings in terms of power dissipation with an aspect ratio of two governing the width of pMOS versus nMOS transistors, except for those nMOS transistors which lie in the critical path whose widths could be made comparable to or higher than pMOS transistors to spruce signal propagation.

IV. CASCADING ARCHITECTURES

A. Existing Architecture

In reference [3], a linear cascade of increment/decrement blocks (granularity of the basic block being 8) is used to realize higher order incrementer/decrementer circuits. The output lookahead of a fundamental increment/decrement module serves as the input lookahead for the next module in the cascade. The maximum delay encountered with lookahead signal propagation would be governed by a logarithmic order: O[log2 2

(N/8)-1], where N signifies the overall size of the

incrementer/decrementer circuit. Figure 3 portrays a 16-bit incrementer/decrementer as a sample, featuring a series connection of two 8-bit increment/decrement building blocks. The 8-bit Inc/Dec Block 1 in the cascade of Figure 3 assumes higher priority than the succeeding module (Inc/Dec Block 2).

Fig. 3. 16-bit incrementer/decrementer realized using the cascaded structure of [3], utilizing 8-bit incrementer/decrementer block(s) as shown in Fig. 1

To reiterate the significance of primary inputs and outputs, inputs and outputs having the highest subscript are referred to as ‘most significant’, while those inputs and outputs featuring lesser subscripts are identified as ‘less significant’. In terms of priority assignment however, the 8-bit increment/decrement block comprising least significant primary inputs and outputs is accorded the ‘highest’ priority, while the successive blocks are associated with descending order of priority. To explain the operation of priority based 16-bit incrementer/decrementer shown above, we revert back to Figure 1 for assistance. The equations for internal and output lookahead signals with respect to Figure 1 are given by,

LA_int = (D3 + D2 + D1 + D0 + LA_in)’ (9)

LA_out = D7 + D6 + D5 + D4 + D3 + D2 + D1 + D0 + LA_in

(10)

Considering the 8-bit increment/decrement module shown in Figure 1, and with respect to the above equations, if LA_in is 1 during the rising clock edge, pMOS transistors p1 to p8 and op1 to op8 will turn-on, which would drive all the outputs of the decision module to 0. Consequently, outputs Y0 to Y7 will reflect the values of inputs I0 to I7. If any or all of the data

inputs of the decision module are equal to 1 simultaneously, then more pMOS transistors will turn-on to produce a similar output, nevertheless at the expense of more switching activity. Supposing LA_in is not 1, but if the lower orders nibble (D3 to D0) is equal to 1, then the internal lookahead signal LA_int becomes equal to 0, and turns-on pMOS transistors op5 to op8. This eventually drives the higher order output nibble of the decision module to 0.

Referring back to Figure 3, it can be seen that LA_in1 is connected to ‘ground’. Under this condition, if any or all of the primary inputs I7 to I0 are 1, then as per the internal circuit diagram of the 8-bit Inc/Dec Block 1 depicted by Figure 1, LA_out1 and consequently LA_in2 will become 1. This implies that irrespective of the value of Inc/Dec’ signal, outputs Y15 to Y8 will equal I15 to I8, which means the most significant byte has not been incremented or decremented, and only the least significant byte will be subject to an increment or decrement operation depending on the value of Inc/Dec’ control signal. Again, referring back to Figure 1, we find that during the falling-edge of the global clock, the pMOS transistor marked as ‘pla’ will turn-on and hence LA_out will become equal to 0. Relating this with Figure 3, it may be noted that if none of the inputs I7 to I0 are 1, then LA_out1 and subsequently LA_in2 are equal to 0. Under this scenario, if any or all of the more significant inputs I15 to I8 equals 1, then during the rising edge of the clock, an incremented or decremented value will be visible only in the output byte Y15 to Y8.

B. Proposed Architecture

If any primary input byte contains a 1, then increment or decrement operation is to be invoked for the respective primary input group and the less significant primary input groups, while no incrementing or decrementing needs to be performed for the more significant groups of primary inputs. In other words, 8-bit incrementer/decrementer modules corresponding to more significant primary input groups can be disabled – disabling can be done purely on the basis of primary data inputs, and hence a sequential transfer of lookahead signals from one stage to another is not necessary. This observation has led us to the idea of parallel reset of unwanted increment/decrement building blocks, and the novel cascading architecture that satisfies this requirement is shown in Figure 4 for the case of a 16-bit incrementer/decrementer.

Fig. 4. 16-bit incrementer/decrementer circuit implemented using the novel cascading architecture


148

Let us now briefly consider the operation of the 16-bit circuit shown in Figure 4. The 8-bit Inc/Dec Blocks shown herein are based on the structure portrayed by Figure 1, but featuring the 8-bit decision module macro that was shown in Figure 2. In Figure 4, LA_in1 is connected to the ‘supply’, hence depending on inputs I7 to I0; the corresponding outputs could either reflect the inputs or incremented or decremented value. If any of I7 to I0 is a 1, the output of the NOR gate ‘nor8’ becomes 0, and it disables the 8-bit Inc/Dec Block 2. In such an instance, outputs Y15 to Y8 would just reflect the values of primary inputs I15 to I8, and no incrementing/decrementing action is performed. On the other hand, if I7 to I0 is equal to 0, the lookahead input LA_in2 is driven to 1 as ‘nor8’ equals 1, and therefore based on any of I15 to I8 being equal to 1, incremented or decremented value is output in Y15 to Y8.

Comparing the 8-bit decision module of Figure 2 with the one shown in Figure 1, it is clear that the separate logic associated with lookahead signaling has been eliminated in the former – this enables a reduction in device count and tends to facilitate a less complex and more elegant physical design. However, comparing the cascade topologies of Figures 3 and 4, it is seen that an extra NOR gate (8-inputs) is required. In fact, for constructing larger sized incrementers/decrementers, both NOR gates and AND gates would be required, as shown in Figure 5 for a generic N-bit incrementer/decrementer – these extra gates tend to slightly offset the reduction achieved in the number of transistors and power dissipation. But the advantage is that parallel reset of select increment/decrement blocks is feasible and static implementations can be used for NOR and AND gates. In general, (N/8) basic increment-cum-decrement blocks are required to realize an N-bit incrementer/decrementer, and the basic equation for a K

th

stage input lookahead is given by (11), with K ranging from 2 to N/8. The symbol ‘•’ signifies logical conjunction in the equation given below.

LA_inK = I8(K-1)-1 + … + I8(K-1)-8’ • I8(K-2)-1 + … + I8(K-2)-8’ •

… • I8(K-6)-1 + … + I8(K-6)-8’ • I7 + … + I0’ (11)

V. RESULTS AND CONCLUSION

Firstly, 8-bit decision module circuit macros which form the heart of the incrementer/decrementer were constructed based on existing and proposed methods at the transistor level. Secondly, 8-bit incrementer/decrementer basic building blocks, corresponding to both existing and proposed designs, were realized physically on the basis of the structure shown in Figure 1. Finally, 16-bit and 32-bit incrementer/decrementer circuits were designed based on both existing and proposed cascading architectures by utilizing 8-bit increment/decrement modules and extra elements. All the circuits designed using

Tanner tools [8] pertain to a 0.25µm bulk CMOS process technology. A range of input patterns was used to verify the functionality of the circuits designed, and also to estimate the average power dissipation for a nominal clock frequency of 100MHz using Tanner SPICE – the power values are given in Table 1.

TABLE I. AVERAGE POWER DISSIPATION VALUES OF DIFFERENT

INCREMENTER/DECREMENTER CIRCUITS, ESTIMATED USING TSPICE

Design

Style

Incrementer/Decrementer Circuit and Total Power

Incrementer/decrementer

size and operating mode

Power

(mW)

%age

decrease

Existing

[3]

8-bits Increment 263.8 -

Decrement 288.7 -

16-bits Increment 1334 -

Decrement 1888 -

32-bits Increment 2951 -

Decrement 4142 -

Proposed

8-bits Increment 136.9 48.1%

Decrement 203.5 29.5%

16-bits Increment 948 28.9%

Decrement 1258 33.4%

32-bits Increment 2567 13%

Decrement 3612 12.8%

The respective savings in power dissipation obtained for the proposed approach over the existing method for various incrementers/decrementers in mentioned in the last column of the above Table. From the simulation results, it is seen that the overall power savings garnered by the proposed approach for incrementing operation is 30%, while it is around 25% for decrementing operation, compared to the existing approach. However, the margin of average power savings for the former in comparison with the latter tends to narrow down with increase in the size of the incrementer/decrementer – this is due to the addition of extra logic gates in the former to feed the lookahead input signal of constituent 8-bit Inc/Dec Blocks. Nonetheless, the overall difference in device count between the two approaches for the above circuits is just about 3%.

REFERENCES

[1] S. Furber, ARM System-on-Chip Architecture, 2nd edition, Pearson Education Limited, 2000.

[2] R. Hashemian, “Highly parallel increment/decrement using CMOS technology,” Proc. 33rd IEEE International Midwest Symposium on Circuits and Systems, vol. 2, 1991, pp. 866-869.

[3] C.-H. Huang, J.-S.F. Wang, Y.-C. Huang, “Design of high-performance CMOS priority encoders and incrementers/decrementers using multilevel lookahead and multilevel folding techniques,” IEEE Journal of Solid-State Circuits, vol. 37, no. 1, January 2002, pp. 63-76.

[4] C.-H. Huang, J.-S.F. Wang, Y.-C. Huang, “A high-speed CMOS incrementer/decrementer,” Proc. IEEE International Symposium on Circuits and Systems, vol. 4, 2001, pp. 88-91.

[5] S. Veeramachaneni, L. Avinash, M.K. Krishna, P.S. Reddy, M.B. Srinivas, “A novel high-speed binary and gray incrementer/decrementer for an address generation unit,” Proc. International Conference on Industrial and Information Systems, 2007, pp. 427-430.

[6] J. Mohanraj, P. Balasubramanian, K. Prasad, “Power, delay and area optimized 8-bit CMOS priority encoder for embedded applications,” Proc. 10th International Conference on Embedded Systems and Applications, 2012, pp. 111-113, Nevada, USA.

[7] S.-W. Huang, Y.-J. Chang, “A full parallel priority encoder design used in comparator,” Proc. 53rd IEEE International Midwest Symposium on Circuits and Systems, 2010, pp. 877-880.

[8] Tanner EDA. Available: http://www.tannereda.com


149

Fig. 5. Proposed cascade architecture for realizing N-bit incrementer/decrementer


150

Authors Index

Alic, K. 43 Minea, M. 50 Anastasijevic, A. 73 Mohorcic, M. 43 Anbumani, V. 146 Mohottige, N. 73 Andonov, V. 142 Murthy, N. S. 77, 91 Archanadevi, B. 146 Pachos, P. 96 Bădescu, I. 50 Panchal, P. 82 Balasubramanian, P. 82, 146 Pertovt, E. 43 Bozic, M. 73 Psomopoulos, C. S. 96 Budimir, D. 73 Qiao, Y. 123 Darbandi, M. 56, 65 Rabbi, K. 73 Hunka, F. 136 Rao, N. B. 77, 91 Ioannidis, G. Ch. 96 Saeid, S. H. 87 Jazyah, Y. H. 25, 36 Srivastava, R. 82 Kaminaris, S. D. 96 Švigelj, A. 43 Kavicharan, M. 77, 91 Tsiolis, S. 96 Kou, J. 113, 118 Varzakas, P. 127, 131 Lan, P. 113 Villiotis, H. 96 Malatestas, P. 96 Vinitha, C. 82 Malathy, T. 146 Vokas, G. A. 96 Manias, S. N. 96 Wang, K. 104 Mastorakis, N. E. 82, 146 Xiao, Y. 104, 113 Matveev, M. 142 Xiao, Y. 118, 123 Melis, Z. 136 Zacek, J. 136 Milanova, M. 142


151

recent advances in electronics and ...recent advances in electronics and communication systems...

Documents