contentscontents guest editorial 1 60 years of department of telecommunications and department of...

ContentsGUEST EDITORIAL 1

60 YEARS OF

DEPARTMENT OF TELECOMMUNICATIONS AND DEPARTMENT OF TELECOMMUNICATIONS AND MEDIA INFORMATICS 2

Boldizsár Bencsáth, Levente Buttyán, István VajdaOn the security of communication network: now and tomorrow 3

Géza Németh, Gábor Olaszy, Klára Vicsi, Tibor FegyóTalking machines?! – Present and future of speech technology in Hungary 8

Márton Csernai, András Gulyás, Zalán Heszberger, Sándor Molnár, Balázs SonkolyCongestion control and network management in Future Internet 14

László Lois, Ákos Sebestyén

Media communications over IP networks – An error correction scheme for IPTV environment 23

Zsolt Parisek, Zoltán Ruzsa, Géza Gordos

Mathematical algorithms of an indoor ultrasonic localisation system 30

CALL FOR PAPERS / GUIDELINES FOR OUR AUTHORS 38

CONTENTS OF THE INFOCOMMUNICATIONS JOURNAL, 2009 40

Infocommunications Journal

ProtectorsGYULA SALLAI – president, Scientific Association for Infocommunications

ÁKOS DETREKÔI – president, National Council of Hungary for Information and Communications Technology

Editorial BoardEEddiittoorr--iinn--CChhiieeff::CSABA A. SZABÓ,

Dept. Telecomm., Budapest Univ. Technology and Economics (BME)

CChhaaiirr ooff tthhee EEddiittoorriiaall BBooaarrdd::LÁSZLÓ ZOMBORY,

Dept. Broadband Communications and Electromagnetic Theory, BME

ISTVÁN BARTOLITS, National Communications Authority

ISTVÁN BÁRSONY, Institute of Technical Physics and Material Science, Hungarian Academy of Sciences (MTA)

LEVENTE BUTTYÁN, Dept. Telecommunications, BME

ERZSÉBET GYÔRI, Dept. Telecommunications and Media Informatics, BME

SÁNDOR IMRE, Dept. Telecommunications, BME

CSABA KÁNTOR, Scientific Association for Infocommunications

LÁSZLÓ LOIS, Dept. Telecommunications, BME

GÉZA NÉMETH, Dept. Telecommunications and Media Informatics, BME

GÉZA PAKSY, Dept. Telecommunications and Media Informatics, BME

GERGÔ PRAZSÁK, National Council for Communications and Information Technology

ISTVÁN TÉTÉNYI, Computer and Automation Research Institute, MTA

GYULA VESZELY, Dept. Broadband Communications and Electromagnetic Theory, BME

LAJOS VONDERVISZT, National Communications Authority

International Advisory CommitteeVOLKMAR BRÜCKNER,

Hochschule für Telekommunikation Leipzig, GermanyMILAN DADO,

University of Zilina, SlovakiaVIRGIL DOBROTA,

Technical University Cluj, RomaniaAURA GANZ,

University Massachusetts at Amherst, USAEROL GELENBE,

Imperial College, London, UKBEZALEL GAVISH,

Southern Methodist University, Dallas, USAENRICO GREGORI,

CNR IIT Pisa, ItalyASHWIN GUMASTE,

IIT Bombay, IndiaLAJOS HANZO,

University of Southampton, UKANDRZEJ JAJSZCZYK,

AGH University of Science and Technology, Krakow, PolandMAJA MATIJASEVIC,

University of Zagreb, CroatiaVACLAV MATYAS,

Masaryk University, Brno, Czech RepublicOSCAR MAYORA,

CREATE-NET, ItalyYORAM OFEK,

University of Trento, ItalyALGIRDAS PAKSTAS,

London Metropolitan University, UKJAN TURAN,

Technical University Kosice, SlovakiaGERGELY ZARUBA,

University of Texas at Arlington, USAHONGGANG ZHANG,

Zhejiang University, Hangzhou, China

This Special Issue is devoted to the 60th anniver-sary of the estalishment of our two predecessordepartments, the Department of Wireline Commu-

nications and the Department of Wireless Communica-tions. Since 1949 not only the organization structure andthe name of departments in this area of our universityhas changed several times but we have been witness-ing several dramatic changes, even paradigm shifts incommunication technologies, telecommunications andbroadcasting. It is sufficient to mention that there wasno Internet at that time (another round figure this year:we are celebrating the 40th anniversary of the birth ofthe Internet as a packet switched wide area computernetwork). To follow these developments, the activitiesand profiles of the two departments underwent corres-ponding changes until today when the convergence ofthese areas has almost fully come true.

The objective of this special issue is to highlight afew topics by invited overview papers that are represen-tative samples from the wide scale of research activi-ties of the two present departments: the Dept. of Telecom-munications (HIT) and the Dept. of Telecommunicationsand Media Informatics (TMIT).

The first paper titled „On the security of communica-tion network: now and tomorrow” by Boldizsár Bencsáth,Levente Buttyán and István Vajda of Crysys Lab (Labora-tory of Cryptography and Systems Security, HIT) discus-ses some security issues in the Internet and sketchesfuture research directions in this field. In particular, theauthors discuss the security issues in wireless network-ed embedded systems through three examples: sensornetworks, vehicular communications, and RFID systems.Finally a brief introduction is given to the field of networkcoding, which is a new and promising research area innetworking.

Speech technology has been an area of intensive re-search worldwide – including Hungary – for several de-cades and the Laboratory of Speech Technology of TMIThas been in the forefront of this research. The paper byGéza Németh, Gábor Olaszy, Klára Vicsi and Tibor Fegyó„ Talking machines?! Present and future of speech tech-nology in Hungary” gives an overview of the challengesand results of the domain and the vision of the develop-ment and the application of the technology will also beintroduced.

The next paper „Congestion control and network ma-nagement in Future Internet” by Márton Csernai, AndrásGulyás, Zalán Heszberger, Sándor Molnár and BalázsSonkoly addresses two key issues in the Future Internetresearch where the general objective is to encourageclean slate designs and thus overcome the barriers ofthe previously prevailing incremental development, pro-posing new visions, architectures and paradigms for thecoming 10-20 years. The first area is congestion controlwhere recent results show that networks operating with-out explicit congestion control (like TCP) may survivewithout congestion collapse if appropriately designedin network resources and if end systems apply appropri-ate erasure coding schemes. The second topic is relatedto the exponential growth of the Internet which makesit hardly impossible to manage the network with traditio-nal centralized approaches (like manager-agent); henceresearch results of complex networks are expected tospread over the Internet with its autonomic behaviors.The authors are pursuing research in these fields withinthe framework of the High Speed Networks Laboratoryof TMIT.

The last paper is titled “Media communications overIP networks – An error correction scheme for IPTV envi-ronment” by László Lois, Ákos Sebestyén, Laboratory ofMultimedia Networks of HIT. Video transmission over IPnetworks has been gaining more and more popularityrecently. One of the crucial problems of video transmis-sion over IP networks through unreliable links is the sus-ceptibility to errors in the transmission path. Packets lostor discarded must be somehow regenerated which canbe accomplished by requesting a retransmission or byrecalculating the packet provided that some redundanc yis introduced in the transmitter side. In addition to a ge-neral overview of issues around the media transmis-sion over IP networks, the paper describes a novel me-thod that can be used for forward error correction in IPTVapplications.

Prof. Dr. Imre SándorHead, Dept. of Telecommunications

Prof. Dr. Sallai GyulaHead, Dept. of Telecommunications and Media Informatics

VOLUME LXIV. • 2009/IV 1

� FOREWORD

Guest [email protected]

[email protected]

In the 40’s, Hungary was considered as one of themost developed countries in the world in the fieldof telecommunications and broadcasting. The te-

lephone line density was one of the highest in Europeby the beginning of WW2. Radio broadcasting started al-ready in 1923, based on the program structure and stu-dio of the so-called telephone news service, inventedby Tivadar Puskás in 1893. Our radio broadcasting in-dustry produced one-third of the radio receivers in themiddle of the 30’s !

To meet the increasing need in electrical engineersspecialized in these new areas, 60 years ago the Buda-pest University of Technology and Economics (then Tech-nical University) decided to establish its Faculty (~School)of Electrical Engineering. It had a new branch called“weak current electrical engineering” in addition to thealready existing – within the Faculty of Mechanical En-gineering – specialization in “heavy current electricalengineering” (~electric power engineering).

This new branch of education started in the acade-mic year of 1949/1950 and two new departments werefounded the same time, the Dept. of Wireline Communi-cations and the Dept. of Wireless Communications. In1971, these two departments were combined in a singleunit called Institute of Communication Electronics whichexisted for 20 years when in 1991 – as a result of a na-tural polarization – it was split again into two units: theDept. of Telecommunications and the Dept. of Telecom-munications and Telematics. The latter was re-named toDept. of Telecommunications and Media Informatics in

2003. These days the two departments cooperate closelyand their scopes mostly complement each other in seve-ral areas, and some healthy competition also exists be-tween them.

The Dept. of Telecommunications was establishedand lead until 2008 by Prof. László Pap, Member of theHungarian Academy of Sciences and is currently leadby Prof. Sándor Imre. It consists of laboratories whichaddress the key research areas of the department andare responsible for the educational activities in theseareas:

• Data and network security• Signal processing and networking algorithms• Computing technologies• Multimedia networks• Acoustics• Networking technologies and network modeling • Network design• Mobile communications and computing

The Dept. of Telecommunications and Telematics wasestablished and lead until 2002 by Prof. Géza Gordos.Since then Prof. Gyula Sallai has been the Head of De-partment. The department focusing on the convergentinformation, communication and media technologies isorganized in the following laboratories representing thekey research and educational areas:

• Infocommunication Networks, Services and Applications

• Infocommunication Management, Strategy and Regulation

• Content Management and Multimedia Systems • Speech and Multimodal Information Systems• Speech Acoustics• Intelligent and Cognitive Media Informatics

In addition, there are two educational laboratories andan accredited and notified Telecommunications Test La-boratory also operates within the department.

For detailed and up-to-date information about the twodepartments we suggest to visit their websites:

www.hit.bme.hu and www.tmit.bme.hu

2 VOLUME LXIV. • 2009/IV

A short summary based on the article by Profs. Géza Gordos and László Pap, published in Hungarian in the Special Issue of “Híradástechnika” devoted to the anniversary of the two departments.

� INVITED PAPER

60 years of Department of Telecommunications and Department

of Telecommunications and Media Informatics

[email protected]

1. Introduction

In this paper, we give a brief overview of Internet secu-rity issues. First, we discuss the security problems ofthe current Internet as we know and use it today. Then,we introduce some future research directions in thefield of Internet security. We continue by considering abroader interpretation of the Internet than it is usuallymeant today, where the network is not limited to PCsand servers, but it also includes various embedded com-puters. This broader interpretation is often referred toas the Internet of Things.

We describe the related security and privacy issuesthrough three examples: wireless sensor networks, ve-hicle communication, and RFID systems. Finally, in thelast part of the paper, we give a short overview on thesecurity of network coding, which is a new promising re-search area that may have impact on the design of fu-ture networks.

Obviously, the general field of security and privacyin communication systems is a large area that cannotbe fully covered in the context of this paper. Our selec-tion of the topics discussed in this paper have beenbiased by our research activities in the Laboratory ofCryptography and Systems Security (CrySyS) of the De-partment of Telecommunications at the Budapest Uni-versity of Technology and Economics. More informationon our research and other activites is available on theweb site of the laboratory at www.crysys.hu.

2. Security of the Internet

The security of the internet has evolved a lot in the lastdecades. Today, nobody can imagine the Internet with-out security mechanisms such as TLS, SSH, PGP, IPsec,or without advanced authentication techniques (smartcards, captchas, two factor authentication tools, etc.)However, lot of the security problems of the Internet re-mained unsolved.

Malware, badware, viruses and worms have beenknown for decades, however, the problem is gettingworse and worse instead of finding proper solution forthe problem. Malware infected hosts are not individualproblematic points anymore, botnets have been created.These botnets might aggregate the resources of milli-ons of computers. Even the estimation of the size of thebotnets is a hard problem.

Using botnets, millions of spam e-mail messages canbe sent out easily and rapidly. Although solutions arecontinuously proposed against spam, spam is still oneof the biggest problems of the Internet. In June 2009,88.9% of the full Internet e-mail traffic was spam.

Most of the Internet servers and services are proneto some kind of Denial-of-Service problems (DoS). Thebasic architecture of the Internet was not designed tobe resilient against such attacks, and therefore the num-ber vulnerable services, servers, protocols is unknownand might very high, therefore, a significant increasemight happen in both the number and the severity of theDoS attacks at any time.

Another unsolved problem on the Internet is crack-ing and defacing web pages and servers. Today, a num-ber of tools and possibilities are available to make ser-vers secure: automatic updates, intrusion detection sys-tems, secure authentication, vulnerability scanners, etc.,but still, the web sites are not secure and cracked fre-quently. There is multiple reasons behind that. Altho-ugh tools are available, they are not used, for differentreasons to make systems secure. Special web contentcannot be updated automatically at all the time. Admi-nistrator do not have right to fulfill necessary operationwithout the owners’ consent.

Grid-computing, cloud-computing also raise new se-curity and reliability concerns. In such cases the legalsituation is even more complicated. The cloud-basedservice might be free or very cheap, but there is no gua-rantee on the reliability, availability and security pro-perties, therefore the end users might have no optionsto solve security problems.


Keywords: information security, privacy, Internet, wireless sensor networks, vehicular communications, RFID systems, network coding

In this paper, we first discuss some security issues in the Internet, and we sketch some future research directions in this field.Then, we discuss the security issues in wireless networked embedded systems through three examples: sensor networks, vehicular communications, and RFID systems. Finally, we give a brief introduction to the field of network coding, which is a new,promising research area in networking.

� INVITED PAPER

On the security of communication network:now and tomorrow

BOLDIZSÁR BENCSÁTH, LEVENTE BUTTYÁN, ISTVÁN VAJDA

Budapest University of Technology and Economic, Department of TelecommunicationsLaboratory of Cryptography and System Security (CrySyS)

{bencsath, buttyan, vajda}@crysys.hu

2.1 Research directionsNot just the problems and solutions have been signi-

ficantly changed during the last years, but even the me-thodology, the point of view on the problems changed a lot.

Future internet: A number of new research projectsstarted to design the next generation of the Internet,including Global Environment for Network Innovations(GENI), a Future Internet Design (FIND), funded by the U.S.National Science Foundation (NFS), or projects in the Se-venth Framework Programme of the European Union(FP7). The main goal of these projects is to redefine thebasic architecture of the Internet. Traditionally the In-ternet provides best-effort services, it is a multi-layerunreliable network. Most of the services are based onTCP/IP where the main task of the routers just to forwardsimple packets. The reason behind many security pro-blems is this approach: because of the architecture andprotocols of the Internet, it is simply impossible or al-most impossible to solve some security problems. Byredesigning the basic blocks and the architecture, newsolutions can be made for the security problems.

The next generation Internet should give more thanbest-effort forwarding of packets. People need servicesinstead of an unreliable transport network, the need qua-lity of service (QoS) agreements. Today, Internet-widesecure authentication is not solved, especially for simplenetwork protocols, and therefore, it is nearly impossi-ble to track back the origin of an attack and punish at-tackers. This list of problems and new tools, featuresmight be continued, but the most important message isthat there is a need and also intention to modify eventhe foundations of the Internet.

Intelligent intrusion-detection: The intrusion detec-tion (IDS, IPS, honeypot, etc.) tools have evolved a lotin the last years. From the basic event detection we ar-rived at complex systems which intelligently distinguishattacks from normal traffic and automatically carry outcountermeasures. This change goes forward to makeglobal intrusion detection systems (possibly with honey-pot systems), to use modern network technologies, likeP2P techniques in intrusion detection, or more intelli-gent reactions to security problems. To this last goal, toprovide more dependable systems by automatic reconfi-guration, an EU FP7 project, DESEREC (www.deserec.eu)has just been finished with the participation of BME’sLaboratory of Cryptography and Systems Security (Cry-SyS).

In the area of trust, reputation and authentication, newmethods give new tools to solve security problems andtherefore there is a very intense research activity in thisfield. The work includes:

• Defining attack and defense incentives and providing new tools based on these properties.

• Using game theoretical methods to make such situations, where there is no use to attack, thus avoiding attacks.

• Providing new ways of authentication possibilitieswhile retaining anonymity and civil rights.

• Dealing with trust on local level and on large scale.

Secure clients and secure, trusted platforms: Lot ofthe problems originate from the fact that the softwareelements and thus the whole client computer cannot betrusted. Trusted computing could provide a situation,where there are no malware programs, or, the can beeasily removed at large scale. However, the concept oftrusted platforms contradicts with the current philoso-phy of the Internet, e.g. the need of the users to installpirated software, downloading music illegally, etc. Thetypical research areas: secure software engineering;formal analysis and proving of protocols, rule sets, oreven formally proven APIs; secure authenticaion in un-trusted environment, etc.

This short introduction cannot provide a full pictureof all the work that is currently in place to provide newsolutions to security problem, we just tried to show themost interesting research areas in this field, that couldhave a great impact to the future of the Internet.

3. Security and privacy in wirelessnetworked embedded systems

3.1 Security in wireless sensor networksIn the last decade, a considerable amount of research

on wireless sensor networks has been carried out allover the world. This new wireless networking techno-logy allows for a number of new and useful applicationsthe monitoring of the parameters of our physical envi-ronment (such as temperature, pressure, humidity, vib-ration, acustic noise, etc.), and the automated collectionand processing of all these monitored data. The poten-tial applications include optimization of agricultural pro-cesses, making ecological observations on large scaleor in environments that are difficult to access physically,forecasting natural disasters such as earthquakes, re-ducing cost in industrial process automation and control,prevention of accidents on the roads, remote monitoringof elderly or chronically ill people, and military tacticalapplications, just to mention a few.

Many of these applications have security require-ments related to the protection of wireless communica-tions on the one hand, and to the increased resistanceof the networking mechanisms againts malicious attackson the other hand. Although, security is a problem andhas been addressed both in wired and in traditional wire-less networks (e.g., in cellular and Wi-Fi networks), thereare new security challanges in wireless sensor networks,and the solutions proposed for wired and traditional wire-less networks can be used only with limited success, ifat all. Such a challenge, for instance, is that the nodes inwireless sensor networks have severe resource con-straints: the nodes are small, battery powered embed-ded computers designed for low energy consumption,and consequently, they have reduced computing, sto-rage, and communication capabilities. Therefore, oneneeds new security mechanisms in wireless sensor net-works that are computationally not so expensive, havesmall code size, and minimize energy consumption.

INFOCOMMUNICATIONS JOURNAL


These requirements are usually not satisfied by the se-curity mechanisms used in traditional networks.

Another distinct security problem in sensor networksis that the nodes are usually physically accessible andthey are not tamper resistant. Thus, they can be relati-vely easily compromised, meaning that an attacker canobtain secrets stored in the node and install rogue soft-ware on the node such that it continues to behave arbit-rari ly. Node compromise may happen in traditional net-works too, however, as the nodes of traditional networksare usually located in locked rooms, the attacker is es-sentially restricted to remote logical attacks. In wirelesssensor networks, node compromise is easier to carry outdue to the easy physical access to the nodes, and wemust always assume that some nodes may have inde-ed been compromised.

In our CrySyS laboratory, we have been working onthe problem of securing wireless sensor networks in thecontext of two projects, UbiSec&Sens (www.ist-ubisec-sens.org) and WSAN4CIP (www.wsan4cip.eu), both fund-ed by the European Commission. In the former project,we participated in the development of a security tool-box for sensor networks inlcuding a random number ge-nerator, new encryption algorithms, new key establish-ment protocols, and security enhancements for routing,clustering, data aggregation, and distributed data sto-rage schemes.

In particular, we developed the Secure-TinyLUNARsecure routing protocol, the RANBAR and CORA resilientdata aggregation algorithms, and the PANEL robust agg-regator node election protocol. In the latter project, weare currently working on dependable networking mecha-nisms for wireless sensors in the context of critical in-frastructure protection applications.

3.2 Security and privacy in vehicle communication systemsUnfortunately, more than 40.000 people die in road

accidents in Europe, and the statistics in the US are si-milar. In addition, another problem is the ever increasingamount of road traffic in large cities that leads to traff icjams and vaste of tremendous amount of time and fuel.In both cases, the situation could be improved if theright information would be available at the right placeat the right time (i.e., if drivers would be informed abo-ut hazardous situations and receive up-to-date informa-tion about the traffic conditions). This could be achievedby letting vehicles to communicate with each other andwith roadside equipments. Such communications mustobviosly be wireless due to the nature of the application.

Most of the large car manufacturers around the worldare seriously considering the idea of vehicle-to-vehicleand vehicle-to-infrastructure communications, and theyinvestigate the related technical problems in nationaland international projects (e.g., NoW, CVIS, Safespot andCoopers projects). Among those technical problems, theyare also looking at security issues. Indeed, it must beclear that vehicle communications can only be adoptedif it cannot be easily misused, or its operation cannot beeasliy disabled. One thing to absolutely avoid is that

s omeone sitting at the roadside injects fabricated mes-sages in the system, and in this way, provokes accidents.A security hole in the system can easily translate intofatalities in this case. Therefore, it is indispensable thatmessages are aunthenticated and their contents are va-lidated, for instance, through consistency checking andcorrelation with other similar messages.

Most safety applications require that the vehicles con-tinuously inform nearby vehicles about their current lo-cation, direction of movement, and speed. For this rea-son, the vehicles send so called heart beat messages,with a rate of several hundred messages per second,which contain these data. This, however, raises privacyproblems, as it becomes rather easy to track the where-abouts of the vehicles by sniffing the wireless channel.Note that, although tracking vehicles is possible by meansof video surveillance technologies, tracking by eaves-dropping is less expensive, more precise, and can becarried out at larger scale. We, and fortunately manyothers, believe that, in modern societies, new techno-logies should be introduced only if they are designedin such a way that they do not make privacy violationseasier than they are today. Therefore, the protection oflocation privacy in vehicle communication systems isan important design requirement.

In our CrySyS laboratory, we have been working onthe problem of securing vehicle communications andlocation privacy enhancing techniques in the contextof the SeVeCom project (www.sevecom.org) that recei-ved funding from the European Commission. In particu-lar, we have investigated various pseudonym schemesand their effectiveness in preventing location trackingin vehicle networks.

3.3 RFID privacySimilarly to the problem of location tracking of ve-

hicles, individuals can be tracked by sniffing the wire-less transmissions generated by various devices thatthey carry with or on them. In particular, it is expected thatin the future many objects will be tagged with RFID tagsthat emit unique identifiers each time they are queriedby a nearby reader. Hence individuals could be identi-fied and tracked by observing the identifiers of their RFIDtags. Unfortunately, RFID tags are even more constrain-ed in terms of resources than the previously describedsensor nodes; hence, it is very challenging to design pro-tocols for them that would prevent this kind of tracking.

In our CrySyS laboratory, we have been working onefficient private identification schemes for low-cost RFIDtags that ensure that external eavesdroppers cannot ob-tain the identifier of the tag from the transactions be-tween the tag and a valid readers.

4. Network Coding

The idea of network coding was born at the beginningof our new millenium, an important development in thetheory and practice of infocommunication. 60 years ago

On the security of communication network


Shannon in his famous publication gave the informationtheoretical foundation for point-to-point communication(channel capacity, existence of optimal channel codes).From the beginning of the 1960’s till the end of 1980’scapacity calculations were extended to small/specialnetworks and channels (e.g. broadcast channels, mul-tiple-access channels, feedback channels).

For practical applications many important resultswere found in the field of random access channels andcode division multiple access channels (CDMA). Thenmore than a decade spent without new ideas in the fieldof information theory of networks when, finally, in theyear of 2000 the network coding was innovated [7].

One the simplest example for the strengh of networkcoding is given in Fig.1. Wireless communication nodesA and B want to exhange their messages a and b, res-pectively Obviously, they can solve this task in four stepsof communication, as shown by version (i). However, theycan also do it by sending only three messages in total:node A and B send their messages to node C, which li-nearly combines the messages by XOR addition (a+b)and broadcasts the sum in one step. It easy to see thatnetwork coding, i.e. allowing linear combination of mes-sages by communication nodes, improves the commu-nication efficiency compared to the classical store-and-forward solution: the store-and-forward solution is thetrivial network coding, when no combination is done atnodes.

The principle of network coding can be extended tonon-binary cases by straighforward generalization. Inretrospect, the network coding in the case of the scena-rio of Fig.1 is trivial. But, how can we find appropriatecombinations for large, general networks? Fortunately,it turns out that nodes can independently choose ran-dom linear combinations to achieve, essentially, optimalcoding.

If a source (or a set of sources) wants to transmit atime flow of messages to destination nodes through thenetwork, then the flow is partitioned into units of fixednumber of messages (called generations) and network

coding combines messages within the same generation.A destination node is able to decode messages of a ge-neration, when it has collected a set of linearly indepen-dent combinations (so called innovative combinations)with set size equals to the size of a generation.

Network coding has important advantages for a di-versity of practical application: for example, improve-ment of network throughput, improved robustness of com-munication in case of link/node failure, decrease of thenumber of communication steps in case of energy criti-cal application (battery powered nodes).

Now, we shortly summarize the possible positioningof network coding in the protocol stack. Network coding

can be placed in each layer, result-ing in different potential applications.If we implement it in the applicationlayer, it has the advantage that rout-ing and MAC layers remain intact,and extension is needed only in thesoftware of the source and destina-tion nodes. A typical application isthe case of overlay communicationtopologies, e.g. P2P file distributionsystems.

When the network coding is im-plemented in the transport layer, adestination node does not send backan ACK for a received packet, butthe node sends back information a-bout possible combinations which

were innovative for it. A source node according to recei-ved needs for innovative information selects combina-tion which is innovative for largest possible set of des-tination nodes.

In the network layer during the discovery of networktopology – which typically some kind of flooding tech-nique – application of network coding is very appropri-ate. The so called opportunistic network coding in wire-less networks is an excellent example for the usage ofnetwork coding in the data link layer. Network nodes areset into promiscuous mode and overhear the wirelesscommunication in their neighborhood, according towhich they can optimize the next network coding stepthey do. Network coding can also be implemented evenin the physical layer, which is called analogic networkcoding.

A serious disadvantage of the network coding is itssensitivity to the so called pollution attack. This meansthat network coding does not check the integrity of the re-ceived packets, and if a – illegally – modified packet (pol-luted packet) is used in a combination it also becomesmodified and will be combined into several further com-binations accross the network, resulting in spreading ofthe pollution, which finally deteriorates the network com-munication. This attack can be stopped by detecting anddropping polluted packets.

For detection cryptographic techniques (MAC, digitalsignature or hash techniques in case of available aut-hentic channels) can be used. This cryptography must fit



Figure 1. An il lustration of the idea of network coding

to network coding by having special algebraic property(homomorphic mapping). However, in case of resourceconstrained environments (e.g. sensor networks) cryp-tographic techniques with high computational comple-xity are excluded.

We can derive the conclusion that the theory andpractice of network coding is developing fast in recentyears. There are many important, potential applications,however, it is an open question yet, when will networkcoding be an ubiquitous option for networking protocols.

5. Conclusions

Communication systems play a fundamental role in to-day’s society, and therefore, their dependability is cru-cial. Dependability includes also security, which meansresistance against intentional attacks. In this paper, wereviewed some of the security issues pertaining in thecurrent Internet, and those expected in future emergingwireless networks and communication systems.

As we could see, there are important challanges a-head of us that must be properly addressed by research-ers and designers of future communication systems,such that we can live in a safer cyber world than we dotoday.

Authors

BOLDIZSÁR BENCSÁTH received his MSc diplomain Computer Science in 2000 from the Budapest Uni-versity of Technology and Economics (BME), andhis MSc diploma in Economics in 2001 the Corvi-nus University of Budapest. He has been workingin BME’s Laboratory of Cryptography and SystemSecurity since 2000, first as a PhD student and thenas a researcher. His research interests are in prac-tical Internet security, protection against spam anddistributed denial-of-service attacks.

LEVENTE BUTTYÁN received the M.Sc. degree inComputer Science from the Budapest University ofTechnology and Economics (BME) in 1995, and ear-ned the Ph.D. degree from the Swiss Federal Insti-tute of Technology – Lausanne (EPFL) in 2002. In2003, he joined the Department of Telecommunica-tions at BME, where he currently holds a positionas an Associate Professor and works in the Labo-ratory of Cryptography and Systems Security (Cry-SyS). His research interests are in the design andanalysis of security protocols and privacy enhanc-ing mechanisms for wireless networked embed-ded systems (including wireless sensor networks,mesh networks, vehicular communications, andRFID systems), and the application of formal met-hods in security engineering.

ISTVÁN VAJDA obtained his MSc degree in Electri-cal Engineering in 1977, and a diploma in Telecom-munications Engineering in 1979 at the BudapestUniversity of Technology and Economics (BME). Hereceived his CSc degree (equivalent to PhD) in 1984,and his Doctor of Sciences (DSc) degree in 1997.Currently, he is a full professor at the Budapest Uni-versity of Technology and Economics, and he is thehead of the Laboratory of Cryptography and Sys-tem Security at the Department of Telecommunica-tions. His research interests include coding theoryand cryptography.

References

[1] MessageLabs Intelligence Reports, Symantec, July 2009. http://www.messagelabs.com/mlireport/MLIReport_2009.07_July_FINAL.pdf

[2] Rajab, M.A., Zarfoss, J., Monrose, F., Terzis, A., “My botnet is bigger than yours (maybe, better thanyours): Why size estimates remain challenging,” Proceedings of 1st Workshop on Hot Topics in Understanding Botnets (HotBots’07), 2007.

[3] Global Environment for Network Innovations (GENI);http://www.geni.net/

[4] NFS NeTS FIND Initiative, http://www.nets-find.net/

[5] Seventh Framework Program,http://cordis-europa.eu/fp7/http://www.future-internet.eu/activities/fp7-projects.html

[6] The Honeynet project, http://www.honeynet.org/

[7] Ahlswede, R. et al, “Network information flow.” IEEE Transactions on Information Theory, July 2000,Vol. 46, No. 4, pp.1204–1216.

On the security of communication network


1. Challenges of speech technology

Speech has been the most natural and most frequentlyused means of human communication. Speech usuallyfulfills the information transmission role between biolo-gical systems (Fig. 1).

The science of speech technology has emerged afew decades ago. Its’ results are used in replacing cer-tain elements of the natural speech communicationchain by artificial solutions (speech recognition, speechsynthesis, human-machine dialogue, diagnosis by speech,speech training, speech-to-speech translation, etc.).

Out of the elements of the natural speech communi-cation chain most practical engineering applicationsrely on the acoustic signal so in the following we shallalso concentrate on this aspect. It should be noted, how-ever, that the language is always behind the acousticform of speech. The linguistic information determinesseveral acoustic components of the spoken message.In order to create successful solutions of language andspeech technology it is not only the processing of theacoustic signal that should be solved. Deep linguisticknowledge should also be coupled with it in order toachieve an artificial system comparable to natural com-munication.

The movie industry has given good visions for thepractical applications of speech technology. One of thekey “actors” is the HAL 9000 speaking computer in 2001:A Space Odyssey that was presented first in 1968 [26].In 1977 in the first episode of Star Wars [1] robots per-ceive, store and present in many ways the multitude ofinformation collected and transmitted through speechcommunication.

These visions of art created the impression for seve-ral people that all these technological breakthroughs canbe reached in a short time. In practice just as interstel-lar spaceships, speaking and thinking robots are still tocome. Because of the gap between huge expectationsand significant but relatively slower technological advan-cements plus short time market success requirementsthere is a certain cyclic nature in the development ofspeech technology.

It is illustrated in Fig. 2 along the dimensions of (tech-nological) maturity and (media) visibility. The figure wascreated by combining Gartner’s 2002 and 2006 key ICT(Information and Communication Technology) and HCI (Hu-man Computer Interaction) forecasts in speech techno-logy related areas. For example natural language searchwas expected to be mature for the market in 2-5 years byanalysts in 2002 (i.e. between 2004-2009). The forecast


Keywords: speech technology, speech synthesis, speech recognition, dialogue systems

Speech technology has been an area of intensive research worldwide – including Hungary – for several decades. This paper will give a short overview of the challenges and results of the domain and the vision of the development and the application of the technology will also be introduced.

� INVITED PAPER

Talking machines?! –Present and future of speech technology in Hungary

GÉZA NÉMETH, GÁBOR OLASZY, KLÁRA VICSI, TIBOR FEGYÓ

Budapest University of Technology and Economics,Department of Telecommunications and Media Informatics

{nemeth, olaszy, vicsi, fegyo}@tmit.bme.hu

Figure 1. The process of speech communication

range changed to 5-10 years in 2006 (i.e. 2011-16). In theevaluation of speech recognition on the desktop similartrends can be observed. It was only speech recognitionin call centers that moved from the 2-5 years categoryin 2002 into the less than two years expectation by 2006.Text-To-Speech (TTS) played the role of emerging techno-logy both in 2002 and 2006 (less than two years until mar-ket penetration). It is important to note that these forecastswere created for the most developed, English speakingUSA market where automation is a frequent businesstarget (e.g. in telephone-based call centers). The real mar-ket situation varies greatly around the world, and eva-luation is a continuous challenge both in Europe as awhole and in our homeland (Hungary) in particular.

In the next section of the paper an overview will begiven about the results of speech technology in an in-ternational and a domestic setting with particular emp-hasis on existing and possible applications. In the 3rdsection a short introduction will be given to the researchand application vision of the area.

2. Domestic and international resultsof speech technology

It is worth looking at both the starting point and the cur-rent situation of R&D in domestic and international set-tings. It should be noted again that successful speechtechnology developments require advances in at leasttwo areas: linguistic analysis and acoustic signal pro-cessing.

Automatic speech generationIn the area of automatic speech generation (often call-

ed speech synthesis) developers achieved as early asin 1984 that an English TTS system became part of theoperating system of Apple computers [5]. This step wastaken in the English versions of Microsoft Windows sys-tems after 2000.

Hungarian research was already in the forefront ofinternational research at the beginning of the 80s whenthe first general purpose, Hungarian TTS system – Hun-garovox – was born in the Institute of Linguistics of theHungarian Academy of Sciences [4]. Since then both lin-guistic analysis and acoustic signal processing algo-rithms have substantially improved. In the latter areathe fourth generation is under study. The first systemsmodeled the human articulatory process by a time-vary-ing filter bank and a simple excitation signal. This so-called formant synthesis solution allowed a coded acous-tic database as small as a few kilobytes. The Hungarovoxsystem was based on this technology, too. The systemhad a strongly robotic voice, with slow speech withoutrhythm and accent but with some level of intelligibility.At the predecessor of the Department of Telecommuni-cations and Media Informatics (BME TMIT) the similarbut improved MultiVox system was implemented in 12languages [12]. The German version of MultiVox was li-censed by an Austrian and a German company.

In co-operation with the developers of the Recogni-ta optical character recognition (OCR) system we coulddemonstrate a Hungarian book-reading system in 1987.The first, commercially availably speech synthesizer for

Talking machines?!


Figure 2. Extended version of the Gartner Hype Cycle [Gartner Hype Cycle 2002, 2006]

the Commodore 64 computer was also developed at thesame BME department [14 – p.269].

In the solutions of the second generation (from the be-ginning of the 90s) waveform segments were cut out fromhuman voice, containing parts of two or three sounds(diphones and triphones, respectively). The acoustic data-base was compiled from these elements. The synthesi-zed waveform was concatenated from these units (c.f.concatenative synthesis). In the following step digitalsignal processing algorithms were applied based on aprosodic model (pitch, timing and intensity). With thissolution a speech quality resembling the given spea-ker could be achieved with a database on 1 to 10.000units, and with storage space in the order of megabytes.In our research the ProfiVox system belongs to this ca-tegory [13].

It was the basis in 1999 for the so-called MailMondó(MailReader) service, which read e-mail messages overthe telephone for subscribers [6]. The same technologyis applied in the SMS reading system for wireline tele-phony subscribers and in the SMSMondó (SMSReader)application for Symbian smartphones [10]. ProfiVox hasalso been integrated in the most widely used screen read-er program for visually impaired people in Hungary (Hun-garian version of Jaws).

The development of the third (corpus-based) techno-logy started in the middle of the 90s. In this case there isno (or maybe some minor) prosodic modification by sig-nal processing. Several hours of (usually read) speechfrom a speaker (or so-called voice actor/actress) arestored. This database is the acoustic database for syn-thesis. In case of good design there is a high probabi-lity that all sound units are available in several proso-dic forms. The storage requirements of these solutionsfall in the gigabyte range.

This technology is applied in the Hungarian name andaddress reading solution of BME TMIT that has alreadyallowed the automation of the reverse directory service(reading out the name and the address of the subscrib-er based on the input phone number) of two mobile ope-rators. In limited domains this technology can approachhuman quality. BME TMIT has prepared solutions for se-veral domains. The weather forecast reader is publiclyavailable (www.metnet.hu), the automatic generation ofauditory version of the price list of devices and servi-ces is applied in the IVR system of a mobile company[11]. The latest demonstration system is a railway time-table information system that “speaks” at the railwaystation of Sárospatak.

The fourth generation of TTS is based on Hidden Mar-kov Models (HMMs). The basic principles are quite diffe-rent from earlier TTS generations. One may say that itgrew out from speech recognition experience. The acous-tic basis in this case is recordings of several hours fromone or more speakers (storage requirements may be inthe terabyte range). These databases are used for train-ing by statistical methods the control parameters of pa-rametric speech coders. It is important that although alarge database is required for training the resulting pa-

rameter database is typically much smaller (even justa few megabytes) which opens up several interestingapplications. The quality of the latest HMM systems app-roach that of the third generation corpus-based systems.There are experiments with so-called hybrid systemswhich provide the data-driven, flexible features of HMMwhile maintaining the high speech quality of corpus-bas-ed systems. Our researchers conduct promising expe-riments in the HMM field, too [16].

It should be remembered, however, that althoughthere are always new solutions the viability of older onesdoes not necessarily cease. They all have certain advan-tages that may be critical in certain applications. For ex-ample it is quite easy to generate whispering voice orspeed up/slow down the synthetic speech with formant(or other parametric) technology which is quite difficultfor corpus-based or HMM approaches.

Automatic speech recognitionThe ASR has been a field of intensive research world-

wide since the middle of the 20th century. From the ini-tial sample-based systems with a vocabulary of a fewwords [15] the technology has advanced to large voca-bulary, continuous, speaker independent technologies.The first Apple operating system containing speech re-cognition was announced in 1993 [5]. The latest ASRsystems of industrial applications are typically basedon HMMs. The basis of the technology was laid down inthe 70s by the researchers of IBM. Nearly forty yearshave passed since then but we still cannot meet “omni-scient machines” that perfectly comprehend our speech.In several narrower domains (e.g. medical dictation)though, there have been applications of regular practi-cal use.

Current ASR systems – beyond standard softwareelements – have basically two language and applica-tion dependent components. Both the acoustic and thelanguage model have to be trained according to the gi-ven application environment.

The acoustic model usually represents the speechsounds as derived from sound samples taken from seve-ral speakers. Even relatively small research databasescontain at least 10 hours of speech of at least 100 spe-akers but there are training databases of up to severalthousand hours. These samples have to be recorded inan environment that is identical (or at least similar) tothe end-user application. For example there are diffe-rent acoustic models for office (wideband) and telephony(narrowband) situations. The general acoustic model canbe adapted to the voice of a particular speaker from arelatively smaller set of training data. The output of pat-tern matching based on just the acoustic models is notaccurate enough. That’s why the language model of ahigher level is required. It is not so surprising if we re-member that human speech perception and understand-ing have several layers, too.

Language models help the recognizer in matchingthe output of the acoustic model (sound sequence) tothe probable linguistic content. In fact, the individual



speech sounds are connected to a complex network ac-cording to the given application environment. In a simplecase, for example command word recognition, the lan-guage model is just a simple vocabulary where the pos-sible commands are listed. In case of the more comp-lex continuous ASR task the linguistic probability ofwords following each other also have to be taken intoaccount. In practice statistical language models train-ed on large text corpora are applied.

In case of agglutinative languages – such as Hun-garian – because of the large number of possible wordforms morpheme-based approaches have also gainedmomentum besides traditional word-based ones. Lang-uage models are always domain specific, the narrowerthe domain the higher recognition accuracy can be ex-pected. In case of isolated command words over 95%accuracy is not rare while in case of the recognition ofspontaneous conversations a result over 60% is regard-ed as quite good on an international scale.

Researchers of BME TMIT have participated in co-ope-ration with industrial partners and with significant statefunding in the creation of several practical applicationsand in the composition of related indispensable data-bases. Their detailed introduction is beyond the scopeof the paper so only a list of major results is given below.

• MKBF 1.0 – ASR engine and development environment:The ASR engine is HMM-based and provides real-timeprocessing in case of moderate size vocabularies(1000-20.000 words). The toolset supports the trainingof both acoustic and language models and allowsN-gram models and speaker adaptation as well.

• Medical report generator:The system allows the direct speech to medical diagnosis transcription [24].

• Classification of prosody and segmentation of speech flow:Prosodic-acoustic processing speech has turnedfrom the interest of speech synthesis research toASR focus as well. An application – based on accentand intonation contour based classification – wasdeveloped for word and phrase boundary detectionfor Hungarian and Finnish. A clause segmentationand a modality detection module was also implemented [21].

• Automatic speech-based emotion recognition [2,17].• Speech databases [22]:

A large amount of labeled and annotated sound material is required for the training of ASR systems.During the preparation of databases statisticallanguage analysis, linguistic and phonetic modeling,corpus design, database qualification and validationtasks have been completed. Diagnostic (oto-rhino-laryngology, radiology, etc.), news (Broadcast News),and audio-visual databases have also been created.

• Multilingual speech corrector (SPECO):In the framework of an EU Copernicus project asystem under the fantasy name of “Magic Box” wasdeveloped. This system provides help for speech

training and speech therapy audio-visually for speech- and hearing-impaired persons in Hungarian,English, German, Slovenian and Swedish. This application will be extended with a prosodic moduleaccording to our latest research results [23,20,18].

• Keyword recognition system: Recognition of a keyword without recognizing previous and following speech segments. Due to integrating both co-articulation and higherlevel pronunciation into the pattern matching process the recognition accuracy may be quitehigh. Because of the lack of the linguistic levelthis approach is not suitable for detecting short keywords. Only one keyword may be found in an announcement. The solution is definitely recommended for recognition of proper names.

• Large vocabulary, speaker independent, real-time application optimized for the transcriptionof broadcast news:This solution takes into account the morphology of the Hungarian language extensively. Consequently the recognition error was nearly halved compared to traditional word-based technologies. In case of speaker adaptation theword error rate was reduced below 20% on a onehour test corpus which is at state-of-the-art levelcompared to similar languages [7].Although automatic speech generation and recog-

nition technologies have improved a lot, “talking ma-chines” (so-called dialogue systems) can be used onlyamong strongly constrained situations. The reason forthis is that modeling such basic phenomena of humancommunication as linguistic, environmental and back-ground knowledge is still at its infancy. In case of natu-ral dialogues we know where and to whom we are talk-ing to and based on our earlier experience we can guessthe topic of the communication, the speaking style ofthe speaker, etc. Most current commercial recognizersdo not convey such basic information as the sex and thespeaking rate of the speaker. Speech synthesizers aretypically not able to change speaking styles, to expressemotions and to adapt to the partner.

Speech based dialogue systemsTaking into account the above mentioned limitations

there are already speech based dialogue systems operat-ing in Hungary. Such systems currently can be success-ful only if the domain of the conversation is sufficientlynarrow and if we inform the human user that the other part-ner is a machine. Such an example is the DrugLine (inHungarian: Gyógyszervonal; www.gyogyszervonal.hu)[9] information system that provides web, wap and tele-phone based interfaces. It ensures the availability of thePatient Information Leaflets of drugs that are approvedby the Hungarian National Institute of Pharmacy throughthree different channels. The speech based dialogue wasimplemented in the telephony version (adapted speech re-cognition and speech recognition subsystems are integ-rated. The phone number of the system is +36-1 8869490.

Talking machines?!


In the USA there is a widespread technology that al-lows the connection of an operator or an appropriate de-partment just by pronouncing the name without keyingin an extension number. A similar technology is avail-able in Hungary as well [13], but is used by smaller orga-nizations yet – such as some local governments – al-though the technology would exhibit its real advantagesin case of large entities (banks, insurance companies,ministries, etc.).

3. Vision for R&D and applications ofspeech technology

In the field of automatic speech generation one of thefocus areas is applying the data-driven, easier to auto-mate HMM technology while preserving the quality ofthe corpus-based approach. Increasing the naturalness,social appropriateness of the generated speech is ofgrowing importance. As a consequence, besides the ge-neral-purpose systems there are increasing numbersof outstanding quality systems in limited domains.

Besides the above mentioned solutions an importantarea is voice telephony access to timetables and ticketordering of public transport systems (railways, local andlong-distance buses) and providing quick access overthe telephone to information of banking, insurance, stateand local government systems quickly, efficiently and7 days/24 hours.

In the area of automatic speech recognition efficientapplications can be implemented in several domainswith currently available technologies. It cannot be reg-arded, however, a market of out-of-the-box solutions.On the contrary, each application needs thorough pre-paration and pre-processing work. In order to achieve abreakthrough for more widespread applications thereshould be advancements in some areas.

• Noise is the hardest limit on recognition accuracy.Noise robust models and noise resistant pre-processing algorithms need even greater attention.This task is language-independent to a large extentso results for a given language may be generalized.

• Another large research area is the recognition ofspontaneous conversational speech. If we look atthe advancement of past decades we can recognizethat technology has proceeded from well definedread speech of both acoustic and linguistic view-point through designed and spontaneous speechto conversational one. The last one is just as “loose”from both acoustic and linguistic viewpoint as thelanguage of Internet forums, for example. Intensiveresearch in this area is of great importance in order to understand natural language communication.

• In case of Hungarian and similar agglutinativelanguages because of the variation of word formsthe size of traditional language models can easilyexceed the limits of several computing platforms.More efficient modeling techniques should be

defined in order to find efficient solutions for theselanguages (e.g. Hungarian, Finnish, Turkish, Arabic, etc.). In Hungarian the relatively free wordorder is another dimension of future research.

Speech technology serving public information accessNearly half of the Hungarian population is not an In-

ternet user. Consequently interactive information ser-vices for all the citizens can only be solved by voice-based telephony. Speech technology is the key to pro-vide automated, cost efficient solutions to this problem.This is the only way to bridge the widening “informa-tion gap”. The idea of “digital public utility” may be worthto be extended to “information public utility” (the accesschannel to information important for the public).

Further information can be obtained from the authorsand from the Hungarian Language and Speech Techno-logy Platform (www.hlt-platform.hu).

Ackowledgements

The authors acknowledge the contribution of the following speech researchers of BME TMIT –

Mátyás Bartalis, András Béres, Tamás Bôhm, Tamás Csapó, Géza Gordos, Krisztián Juhász, Géza Kiss,Laczkó Klára, Péter Mihajlik, György Szaszák, Bálint Tóth,

Zoltán Tüske, Ákos Viktóriusz and Csaba Zainkó – to the results presented in the paper.

Research presented in the paper has been supported byamong others GVOP, NKFP, Jedlik and NTP programs

of the Hungarian Government.

Authors

GÉZA NÉMETH (1959) obtained his MSc in Electri-cal Engineering, major in Telecommunications, atthe Faculty of Electrical Engineering of BME in 1983,his Dr. Univ. degree in 1987 and the PhD degree in1997. Dr. Nemeth is the Head of the Speech Techno-logy Laboratory of BME TMIT. His research areasinclude speech technology, service automation, mul-tilingual speech and multimodal information sys-tems, mobile user interfaces and applications.

GÁBOR OLASZY (1943) is an electrical engineerand phonetician. He graduated from the BME, Fa-culty of Electrical Engineering, Branch of Telecom-munications, in 1967, obtained his Dr. Univ. degree– BME (1985), Ph.D. in linguistics-phonetics (1988),DSc in phonetics (2003). Research areas includeacoustics of speech, development of text-to-speechsystems for different languages, embedding speechsynthesis into applications, research for tools to-wards high quality synthesised speech, combina-tion of stored speech method with synthesised items,modelling of prosody and sound durations for TTS.

TIBOR FEGYÓ (1973) obtained his MSc in TechnicalInformatics at the Faculty of Electrical Engineer-ing and Informatics of in 1997. Research areas in-clude automatic speech recognition, acoustic andlanguage modeling, design and development ofspeech information systems, speech quality mea-surements of telecommunications channels.



KLÁRA VICSI (1948) is a speech acoustic expert.She obtained her M.Sc. degree at Faculty of Scienceof Eötvös Loránd University in 1971, the Dr.Univ.degree in 1982, her Ph.D. degree in 1992 and a DScdegree from the Hungarian Academy of Sciencesin 2005, She obtained a Dr. habil. title from the Bu-dapest Univ. of Technology and Economics in 2007.Research areas inlcude speech acoustics, compu-ter speech recognition, preparation of speech data-bases, psycho-acoustics, project leader in phone-tics and Hungarian speech databases, providing abasis for speech recognition tasks, she was theleader of an international project of the elaborationof multi-modal speech training and developmentprocesses. She was the organizer of numerous in-ternational conferences and summer schools.

References

[1] http://en.wikipedia.org/wiki/Star_Wars_Episode_IV:_A_New_Hope

[2] European COST Action 2102 (Cross-Modal Analysis ofVerbal and Nonverbal Communication),http://www.cost2102.eu/joomla/

[3] Fegyó, T., Mihajlik, P., Szarvas, M., Tatai, P., Tatai, G.,” Voxenter™ – Intelligent Voice Enabled Call Center forHungarian”. In: EUROSPEECH – INTERSPEECH 2003, 8th European Conf. on Speech Communication andTechnology, Geneva, Switzerland, ISCA, pp.1905–1908.

[4] Kiss Gábor, Olaszy Gábor,“Hungarovox – a Hungarian language real-time dialo-gue speech synthetizer system”. Információ Elektronika, 2. (in Hung.),Budapest, 1984, pp.98–111.

[5] MacinTalk, http://en.wikipedia.org/wiki/PlainTalk#The_original_MacInTalk

[6] Németh G., Zainkó Cs., Fekete L.,“Statistical analysis for designing and developing an e-mail reader“, Híradástechnika, Vol. LVI., 2001/1, pp.23–30. (in Hung.)

[7] Mihajlik, P., Tarján, B., Tüske, Z., Fegyó, T.,Investigation of Morph-based Speech Recognition Improvements across Speech Genres, Proc. of Interspeech 2009, Brighton, U.K.

[8] Németh, G., Zainkó, Cs., Kiss, G., Olaszy, G., Fekete, L., Tóth, D.,Replacing a Human Agent by an Automatic Reverse Directory Service; Proc. of 15th Int. Conference on Information System Development, Budapest, Hungary,Springer LNCS, 2006, pp.323–331.

[9] Németh, G., Olaszy, G., Bartalis, M., Kiss, G., Zainkó, Cs., Mihajlik, P.,Speech based Drug Information System for Aged andVisually Impaired Persons, Proc. of Interspeech 2007,Antwerp, Belgium, pp.2533–2536.

[10] Németh, G., Kiss, G., Zainkó Cs., Olaszy G., Tóth, B.,Speech Generation in Mobile Phones, In: D. Gardner-Bonneau and H. Blanchard (Eds.), Human factors and interactive voice response systems,2n edition, Springer, 2008, pp.163–191.

[11] Németh, G., Zainkó, Cs., Bartalis, M., Olaszy, G., Kiss, G.,“Human Voice or Prompt Generation?

Can they Co-exist in an Application?”, Interspeech 2009, Brighton, UK.

[12] G. Olaszy, G. Gordos, G. Németh,The MULTIVOX multi-lingual text-to-speech converter,In: G. Bailly, C. Benoit and T. Sawallis (Eds.): Talking machines: Theories, Models and Applications,Elsevier, 1992, pp.385–411.

[13] Olaszy, G., Németh G., Olaszi, P., Kiss, G., Gordos, G.,“PROFIVOX – A Hungarian Professional TTS Systemfor Telecommunications Applications”, Int. Journal of Speech Technology, Vol. 3, Nr. 3/4.Kluwer Acedemic Publ., December 2000, pp.201–216.

[14] Olaszy Gábor,Electronic speech synthesis. Mûszaki Kiadó, 1989, (in Hung.).

[15] Shoeboxhttp://www-03.ibm.com/ibm/history/exhibits/specialprod1/specialprod1_7.html

[16] Tóth, B., Németh, G.,“Hidden markov chain-based artificial speech generationin Hungarian”, Híradástechnika, 2008, pp.2–6. (in Hung.).

[17] Tóth, Sz.L., Sztahó, D., Vicsi, K.,Speech Emotion Perception by Human and Machine.Proc. of COST Action 2102 International Conference,Patras, Greece, 9-31 October 2007. Revised Papers in Verbal and Nonverbal Features ofHuman-Human and Human-Machine Interaction, Springer, 2007, pp.213–224.

[18] Vicsi, K., Roach, P., Öster, A., Kacic, Z, Barczikay, P., Tantos, A., Csatári, F., Bakcsi, Zs, Sfakianaki, A.,A multimedia, multilingual teaching and training systemfor children with speech disorders. International Journal of Speech Technology, Vol. 3,Kluwer Academic Publisher, 2000, pp.289–300.

[19] Vicsi K, Velkei Sz.,Development of a continuous speech recognition system,3rd Hungarian Computer Linguistics Conference, Szeged, 2005, pp.348–359., (in Hung.).

[20] Vicsi, K.,Computer Assisted Pronunciation Teaching and Training Methods Based on the Dynamic Spectro-Temporal Characteristics of Speech. In: Pierre Divenyi and Georg Meyer (Eds.):“Dynamics of speech production and perception”, IOS Press, Amsterdam, 2006, pp.283–307.

[21] Vicsi K, Szaszák Gy.,Using Prosody for the Improvement of ASR: Sentence Modality Recognition,In: Interspeech 2008, Brisbane, Australia, 2008.

[22] http://alpha.tmit.bme.hu/speech/databases.php[23] http://rcs.hu/sc.htm[24] http://alpha.tmit.bme.hu/speech/research.php[25] http://alpha.tmit.bme.hu/speech/ikta_gastro.php[26] http://en.wikipedia.org/wiki/

2001:_A_Space_Odyssey

Talking machines?!



1. Introduction

The fast changes in the Internet usage and in its tech-nology in the previous years have resulted in a grow-ing expectation of possible paradigm changes in seve-ral mechanisms of the Internet. In this paper we add-ress two challenging areas of these hot topics, namely,the solution for congestion problems and the manage-ment of future Internet.

The congestion is managed in the Internet by a con-gestion control mechanism called the TransmissionControl Protocol (TCP), which is a complex transportprotocol that has gone through several evolution stepssince the beginning of the Internet. This evolution wasdriven by the ever-changing user and application re-quirements and also by the current technological limi-tations. As a result, a number of different TCP versionshave been developed. TCP was originally designed asa reliable connection-oriented end-to-end transport pro-tocol for the fixed wired Internet. However, the situationtoday is rather different concerning the pervasivenessof wireless technologies and also the increasing numberof very high capacity links. It seems that the researchcommunity has arrived to a conclusion that it is veryunlikely that an optimal TCP solution could ever be de-veloped.

On the other hand, a new idea has arisen in the pre-vious years, which basically says that we should de-sign the solution for congestion problems from scratch.The idea challenges the researchers to think over theissue of congestion from a different point of view. Whatif we do not implement any congestion control in thenetwork? How can the congestion be avoided in thatcase? Would it be possible to develop an Internet wherethe packet losses due to congestion events can eff i-

ciently be corrected by erasure coding? The TCP con-cept with its evolution and these exciting questionsare discussed in the first part of the paper.

It is a well-accepted fact by current research con-cerning networks, that the future Internet will be cha-racterized by the tons (in the order of trillions) of parti-cipating communicational entities and the complex andheterogeneous connections between them. Neverthe-less the functional requirements of the next-generationnetworks will be highly diversified, and managing thesenetworks with human interactions will be hardly solvable,thus the need for high-level automatic management isinevitable. Designing such complicated, large-scale sys-tems is a complex task, and we still lack the adequatemethodology for that. The means for describing large-scale networks developed a lot in the past years. Wecan see applications based on these findings, such asthe search in complex networks, which will be discus-sed later in more detail.

Considering the future Internet, as in all complex sys-tems, the analysis of the ongoing processes will requi-re modern tools and methods. The conventional metho-dology, by which the global behavior of the system istreated as the collaborative functioning of different parts,lapses due to the complexity of the system. To handlethese problems, we must come up with a self-organi-zing system model, where the centralized process cont-rol is replaced by distributed functioning and decentra-lized decision making. The monitoring of the networkwill take place on a macro level by analyzing the emer-gent features of the system, and not by independentlyobserving the parts of it. In the second part of the paper,we discuss research topics in the area of large scalesystems and consider self-organizing communicationnetworks.

Keywords: Future Internet, congestion control, complex networks, socio-economics

Future Internet research programs try to ignore and overcome the barriers of incremental development and encourage clean slate designs, propose new visions, architectures and paradigms for the coming 10-20 years. Recent results in congestion control research has shown that networks operating without explicit congestion control (like TCP) may survive without congestioncollapse if appropriately designed in network resources and if end systems apply appropriate erasure coding schemes. The exponential growth of the Internet makes it virtually impossible to manage the network with traditional centralized approaches (like the manager-agent one); hence research results of complex networks are expected to spread over the Internetwith its autonomic behaviors. In this article we give overview and outlook of some interesting research areas connected to the future Internet research.

� INVITED PAPER

Congestion control and network management in Future Internet

MÁRTON CSERNAI, ANDRÁS GULYÁS, ZALÁN HESZBERGER,SÁNDOR MOLNÁR, BALÁZS SONKOLY

Budapest University of Technology and Economics, Department of Telecommunications and Media Informatics

{csernai, gulyas, heszi, molnar, sonkoly}@tmit.bme.hu

2. Congestion Control in Future Internet

Congestion control is a resource and traffic manage-ment mechanism to avoid and/or prevent excessive si-tuations (buffer overflow, insufficient bandwidth) that cancause the network to collapse. It should not be confusedwith flow control, which prevents the sender from over-whelming the receiver. Congestion control has beenaccomplished by the Transmission Control Protocol (TCP)from the very beginning of computer networks and play-ed an important role in the success of Internet.

The original protocol providing a reliable, connec-tion-oriented service on top of IP networks dates backto 1981 (RFC 793). In the mid 1980s, serious incidentswere experienced in the Internet when the network per-formance fell down by several orders of magnitudes.This phenomenon, called congestion collapse, raisedthe urgent need of some more sophisticated controlmechanism in the transport layer. The original solutionfor the congestion collapse was provided in [1] by VanJacobson. An essential part was added to TCP includ-ing the congestion control mechanisms. The congestionmanagement of TCP is composed of two important al-gorithms. The Slow-Start and Congestion Avoidance al-gorithms allow the protocol to increase the sending datarate of sources without overwhelming the network andhelp to avoid congestion collapse. The protocol updatesa variable called congestion window (cwnd, w) that di-rectly affects the sending rate by means of limiting thenumber of unacknowledged packets in the network bas-ed on a sliding window mechanism which involves aself-clocking control. The congestion window variableis adjusted according to various algorithms in differentphases of the connection. The basic mechanism wasincrementally developed and tuned introducing new ad-ditional algorithms, e.g., RTO calculation and delayed

ACK in 1989 (RFC 1122), SACK in 1996 (RFC 2018) andNewReno in 2004 (RFC 3782) just to mention a few. Astandard TCP (TCP Reno) source starts sending accord-ing to the Slow-Start mechanism applying a multiplica-tive increase algorithm. More specifically, the conges-tion window is increased by a constant value for eachacknowledgement received. This yields an exponentialgrowth of the congestion window.

In Congestion Avoidance phase, the congestion win-dow is adjusted by an AIMD (Additive Increase Multipli-cative Decrease) mechanism which results in the clas-sical “sawtooth” trajectory. When no packet loss is expe-rienced, then the window is increased by 1/w per acknow-ledgements (AI) while it is halved as a response to pa-cket loss (MD) as it is shown in the first row of Table 2.The main goal of the TCP’s congestion control mecha-nism is to provide good network utilization, to avoid con-gestion collapse and to share the resources (now thelink capacities) among end-users in a fair way. This isachieved by a distributed, closed-loop feedback mecha-nism. The last requirement regarding the fairness pro-perties of the protocol is an important part of the next-generation transport protocol design.

TCP congestion control had managed successfullythe stability of the Internet in the past decades but it hasreached its limitations in “challenging” network envi-ronments. The new challenges of next-generation net-works (e.g., high speed communication or the communi-cation over different media) generated an urgent needto further develop the congestion control of the currentInternet. In recent years, several new proposals andmodifications of the standard congestion control me-chanism have been developed by different researchgroups all over the world. These new mechanisms andTCP versions address different aspects of future net-works and applications and improve the performance of


Table 1. High speed transport protocols


regular TCP. For example, standard TCP (Reno version)cannot provide acceptable performance in wireless ormobile environments where the propagation delay andthe available bandwidth can suddenly change (e.g., du-ring inter-system handover) which can result in multi-ple back-offs or in extreme cases in disconnection. Inorder to remedy this problem, new TCP versions havebeen dedicated to this environment. The drawbacks ofstandard TCP Reno can be experienced in high speedwide area networks, as well. These networks can be cha-racterized by high bandwidth-delay product (BDP) andTCP cannot efficiently utilize them due to its conserva-tive congestion control scheme. As a response to thisproblem, the research community has proposed severalnew transport protocols recently referred as high speedTCPs or high speed transport protocols.

The huge number of new ideas has resulted in dif-ferent new TCP versions implemented in several envi-ronments. In order to select the “optimal” transport pro-tocol, extensive performance analysis is necessary ina wide range of network environments and applications.In the recent years, many papers were published deep-ening our understanding of these new protocols regard-ing performance characteristics, co-existence issues,and other important properties affecting the possibili-ties of their deployment. In the rest of the paper, a briefoverview is given on some promising TCP versions. Themain properties of the most important variants are pre-sented in Table 1 while a more detailed overview can befound in [2].

In TCP Reno, the congestion event is indicated bypacket losses and the sending rate is reduced whenlosses occur. Protocols considering this type of conges-tion measure are generally referred to as loss-basedprotocols. This one-bit congestion indication does notallow sophisticated congestion control mechanisms.In addition, the permanent oscillation which is an int-rinsic property of this mechanism raises stability issu-es. Therefore, fundamentally different approaches have

also been emerged. In the case of delay-based algo-rithms, the round-trip time (RTT) is regularly calculatedduring the connection and the sending rate is adjustedaccording to the current value of the average delay es-timation. In other words, a “multi-bit” congestion mea-sure (delay) is considered in the control decision. Themost recent TCP versions combine both principles andapply hybrid or combined delay/loss-based mechanisms.Other solutions suggest explicit congestion feedbackfrom the network routers. These mechanisms using ex-plicit congestion indication require the modification ofthe routers, as well.

HighSpeed TCP (HSTCP) [3] is a modification to TCP’scongestion control mechanism for use with TCP connec-tions with large congestion windows. It changes the TCPresponse function to achieve better performance on highcapacity links. HSTCP is based on an AIMD mechanismwhere the increase and decrease parameters (a(w) andb(w)) are functions of the current value of the conges-tion window (see the corresponding row of Table 2) yield-ing an adaptive and more or less scalable algorithm.HSTCP introduces a new relation between the averagecongestion window and the steady-state packet drop(or marking) rate. It is designed to have the standardTCP response in environments with mild to heavy con-gestion (packet loss rates of at most 10-3) and to have adifferent, more aggressive response in environments ofvery low congestion event rate.

Ideas to introduce MIMD mechanisms for TCP havealso been considered. Scalable TCP (STCP) [4] is a goodexample which has been suggested as an efficient trans-port protocol for high speed networks. Here, the mul-t iplicative increase and multiplicative decrease algo-rithm guarantees the scalability of the protocol. Thecongestion window is increased by a constant parame-ter (a) as a response to a received acknowledgement,while it is reduced in a multiplicative manner (by bw)in case of packet losses (see Table 2). A proposed set-ting for the constants are a =0.01 and b =0.125 [4].



Table 2. Details of some TCP vers ions

In order to solve the TCP’s severe RTT (round-trip time)unfairness problems, BIC TCP has been developed [5].BIC TCP combines two schemes called additive incre-ase and binary search. When the BIC TCP source getsa packet loss event, the congestion window is reducedby a multiplicative factor (β); and the maximum windowparameter (Wmax) is set to the value of the congestionwindow just before the reduction while the minimumwindow parameter (Wmin) is set to the current value. Thenthe protocol performs a binary search between theseparameters by jumping to the “midpoint” between thebounds. (More exactly, this jump is based on the B pa-rameter of the protocol.) If packet loss does not occurat the updated window size, that window size becomesthe new minimum; if packet loss occurs, that windowsize becomes the new maximum.

An important restriction is also introduced, the growthcannot be more aggressive than a linear one with a con-stant parameter (Smax). This process continues until thewindow increment is less than a small constant (Smin),when the window is settles down around Wmax (increas-ing slowly on a “plateau”). This mechanism yields an“AIMD-like” behavior where the growing function is mostlikely composed of a linear phase (additive increase)and a logarithmic one (binary search). When the upda-ted window size exceeds the current maximum, then anew equilibrium state has to be found and BIC TCP en-ters into the max probing state. During this phase, thegrowing function is the inverse of the previous ones,more exactly, the window is increased exponentiallyfirst (which is very slow at the beginning) and then li-nearly.

This complex mechanism is also summarized in thecorresponding row of Table 2. The good performance ofthe protocol, including good utilization, linear RTT fair-ness (RTT unfairness is proportional to the RTT ratio asin AIMD), good scalability, and TCP-friendliness, comesfrom the slow increase around Wmax and the aggres-sive linear increase of additive increase and max prob-ing phases. Further research with BIC TCP has beenresulted in CUBIC. CUBIC [6] is an enhanced version ofBIC TCP. It simplifies the BIC window control and impro-ves its TCP-friendliness and RTT-fairness. It is worthnoting that the default TCP protocol of Linux kernelsfrom version 2.6.8 was BIC TCP. From kernel version2.6.19, the default protocol is CUBIC.

The research on the delay-based ideas has result-ed in FAST TCP [7]. FAST TCP has the same equilibriumproperties as TCP Vegas but it can also achieve weight-ed proportional fairness. FAST TCP seeks to restrict thenumber of its packets queued through the network pathbetween an upper (β) and a lower (α ) bound, however,the behavior is usually controlled by a single parame-ter (α ) that can be considered as the targeted backlog(packets in the buffers) along the flow’s path [7]. Undernormal network conditions, FAST TCP periodically up-dates its congestion window based on the comparisonbetween the measured average RTT and the estimatedround-trip propagation delay (when there is no queue-

ing). More exactly, the window is adjusted accordingto the formula presented in Table 2, where γ is the stepsize affecting the responsiveness of the protocol, andbaseRTT is the minimum RTT observed so far which isan estimation of the round-trip propagation delay. Theparameter α controls the equilibrium behaviour, there-fore the appropriate setting of this parameter is crucial.FAST TCP also reacts to packet losses by halving its con-gestion window.

The delay-based control also appears in other pro-posals like TCP-Africa [8]. TCP-Africa is a hybrid proto-col that uses a delay metric to determine whether thebottleneck link is congested or not. In the absence ofcongestion it uses an aggressive, scalable congestionavoidance rule but in the presence of congestion itswitches to the more conservative Reno congestion avoi-dance rule. The combination of the delay-based and theloss-based approaches also appears in TCP-Illinois [9].TCP-Illinois uses loss as a primary congestion signaland delay as a secondary one. The protocol uses anAIMD mechanism but adjusts the increase and decre-ase parameters based on experienced queueing de-lay.

Compound TCP [10] is another important examplewhere a synergy of delay-based and loss-based app-roach has been implemented. It uses a scalable delay-based component in the standard TCP Reno congestionavoidance algorithm. Compound TCP has been develo-ped in the Microsoft Research and it is the default TCPprotocol of Windows Vista and Windows Server 2008.Moreover, it can be installed for other Windows versionsby downloadable hotfixes and in addition, the Linux imp-lementation is also available.

The idea of incorporating accurate bandwidth esti-mations into the TCP congestion control has also open-ed a new path in TCP research. TCP-Westwood [11] is aprominent example where eligible rate estimation me-thods to intelligently set the congestion window andslow-start threshold have been introduced.

Another important group of congestion control pro-tocols is based on explicit congestion notification in-stead of the implicit congestion signals such as packetloss or delay. These congestion control schemes re-quire the assistance of network routers by this meansthe modification of the routers is also necessary. Thisis a serious disadvantage from the aspect of deploy-ment feasibility. One of the main representatives of thisgroup is the eXplicit Control Protocol (XCP) [12] whichgeneralizes the Explicit Congestion Notification (ECN)proposal. Instead of the one bit congestion indicationused by ECN, XCP capable routers inform the sendersabout the degree of the congestion at the bottleneck. Inaddition, XCP decouples the utilization control from fair-ness control.

The history of the research of congestion control pro-tocols revealed that it is difficult to find an optimal pro-tocol that meets all the challenges of the evolving In-ternet. It is very likely that the task to find a universal andoptimal congestion control protocol is impossible. This



view is supported by the fact that the developments ofnew applications show that they use their own conges-tion control mechanisms. These mechanisms in most ofthe cases are not TCP friendly so they cannot work to-gether with TCP efficiently.

An interesting research is proposed in the frameworkof GENI (Global Environment for Network Innovation) ad-vocating a future internet without congestion control. Thebasic idea is that flows do not attempt to relieve the net-work of congestion but rather send as fast as they canwhenever they have data to send. Of course, if all flowsare sending at maximal rates, then the packet loss ratewithin the network is probably high. To overcome thisproblem, flows can use efficient erasure coding.

This solution has several advantages but also rais-es some unsolved problems too. One of the biggest ad-vantage is that we can achieve maximum resource uti-lization. It is because end hosts send packets as fast aspossible and all available network resources betweensource and destination are utilized as much as possi-ble. Links are constantly overdriven so any additionalcapacity is immediately consumed. This solution is themost efficient regarding network resource utilization.Another advantage is that this proposal can use simplerouter architectures. Routers no longer need to bufferpackets to avoid packet loss so no need for expensiveand power-hungry line-card memory. This can also re-sult in significant decrease of the end-to-end packet de-lays for supporting delay sensitive applications. More-over, this solution perfectly fits to a network with all-op-t ical cross-connects.

With all these advantages we can also face a num-ber of problems to solve. The most crucial question isthat what performance can be achieved by implement-ing erasure coding techniques. The promising fact is thatthere is a number of new coding techniques proposedin the last decade with robust characteristics and highperformance like fountain codes [13]. Another problemto solve is how to provide fairness. A mechanism is need-ed in the switches to perform selective packet dropping.As an example the Approximate Fair Dropping (AFD) [14]is a promising candidate to do this task.

Research is in the early phase but researchers canreport some surprising results from their study. It seemsthat the congestion collapse is not as usual in networkswithout congestion control as it was believed [15]. Earlyresults show that efficiency remains higher than 90%for most network topologies as long as maximum sourcerates are less than the link capacity by one or two ordersof magnitude. It is also possible that a simple fair droppolicy enforcing fair sharing at flow level is sufficient toguarantee 100% efficiency in all cases. Of course, thereare several questions unanswered and new challengesto meet because present studies use some assumptionswhich are not fulfilled in practice.

An intensive research is needed to answer the ex-citing question whether we can build the future internetwithout congestion control and forget about TCP and itsall problems.

3. Paradigm shift in managing networks

3.1 Large scale networksOne of the most important processes regarding the

Internet today is the migration from the 20-year-old IPv4protocol to the IPv6. The most demanding issue behindthe process is that we are running out of the availableIPv4 network addresses. The 32 byte address space ofthe IPv4 (~4.3 billion possible addresses) was created,when the computer (mainframe) to people ratio was 1:200.Nowadays this ratio is reaching 1:1 with 1.2 million us-ers, which number can easily double in the near futuredue to the new users of the developing countries (ac-cording to estimates the number of users grow by 150million every year). If we look at the spread of mobiledevices, we can easily calculate that in the near futureone person might even possess 200 network devices.According to assumptions, by 2010 the number of mo-bile devices will reach the number of PCs connected tothe Internet. The typical tendency is that the PDA devi-ces turn into a communicator device, but the real break-through will be the introduction of communications im-plants. In the world of sensors/actuators and intelligentmaterials we will see micro devices integrated into thehuman body, which are capable of wireless connecti-vity and will be used as life supporting or human-com-puter interaction devices. The RFID (Radio FrequencyIDentification) is a really simple sensor network, wherethe active or passive devices can identify themselves,and it is already widely used.

The complexity of the Internet is growing not onlyby the addition of new network devices, but the rise ofnew available online services demands logically con-nected networks apart from the underlying physical one.This tendency is also reflected by the virtual ISPs (In-ternet Service Providers), mobile service providers andthe development of virtual private networks. We can evenmention the widespread use of peer-to-peer communi-cational methods, which all are built on the logicallyconnected overlay networks. This significantly compli-cates the effective manageability of these networks.

For the effective analysis of large-scale complex net-works first we need to develop such theory, which setsaside from the individual characteristics of the nodes,and concentrates on the structure of the connectionsand the character of the network. Moreover, the findingsof this research field can be useful not only in the infor-mation technology. Many real world networks can be de-scribed with complex network models, for instance anorganization, which is a network of people connected toeach other. Also such networks are food webs, the glo-bal economical system or the connections between wordsin a language. We can also mention the diseases, whichspread on the human social network (i.e. STDs). In ge-neral the research on complex networks concentrateson the various characteristics and the dynamic beha-vior of the networks.

Since the ‘50s the complex networks were descri-bed by the Erdôs-Rényi [16] model, which was the only



reasonable and adequately precise approach at thattime. Still researchers presumed that real world net-works are neither completely regular, nor completelyrandom. The widespread use of computers and the In-ternet generated large databases, and these databasesare easily accessible.

By analyzing these topological data, researchersmade two important discoveries in the last two deca-des, one of them is the Watt and Strogatz “small world”effect; the other is the Barabási-Albert scale-free net-work model. The small world effect describes the samephenomena, which was presented in Milgram’s famousexperiment in the ‘60s [17]. He found that even if thereare many billions of people in the world, the shortestroute consists of only several hops between two random-ly chosen individuals in the social network. The scale-free network model highlights another interesting fea-ture of complex networks, namely the complex networkshave scale-free degree distributions, and not Poissondistribution, which is the characteristic of random net-works. The scale-free distribution means that it is highlyprobable for high degree nodes (hubs) to evolve in large-scale network (Fig. 1).

There are three fundamental characteristic featuresof the complex network models, which are worth to em-phasize: average path length, clustering coefficient anddegree distribution [18]. Average path length is the ave-rage of the shortest distances of any two random nodesin the network. This distance represents the effectivesize of the network. It was an interesting discovery thatmost of the real world complex networks have relati-vely short average path length, which feature led to thename “small world”. The examination of the basic para-meters of complex networks was an important step inthis scientific field. Based on these parameters, we canintuitively build up different mathematical models, whichresult in networks with similar statistical characteris-tics.

Another interesting topic in the field of complex net-works is the problem raised by dynamic systems. In-teresting observations were made about the synchro-

nization of routing messages going over the Internet.Although the topology of the network was not specifi-cally built for this purpose, the routers still synchronizetheir message exchanges easily, and when we breakthe synchronicity in one part of the network by introduc-ing some randomness (altering a deterministic proto-col), another part of the network will get in sync [19].We can find more detailed treatment about these kindsof phenomena in the field of self-organizing systems.

3.2 Evolution and self-organization in communicational networks

Today telecommunication systems apply the widelyused global network management paradigm. In this app-roach an external regulating unit controls the system.This unit constantly monitors the state of the systemand the environment. When a problem occurs in thesystem, or the environment changes, the regulator cal-culates the appropriate response, and drives the sys-tem into the respective state (Fig. 2). This solution is onlyfeasible as long as we can find the appropriate respon-se much faster, than the system changes its states. Thiscriterion obviously delimits the permissible comple-xity and dynamism of the system, because the control-ler needs to be much more complicated than the systemitself. For instance we can mention the link-state pro-tocols, which cannot be used on large scale and comp-lex topology.

A natural way to deal with complexity is self-organi-zation, by utilizing the complexity of the system in themanagement plane. In a self-organizing system a largenumber of intricately connected devices achieve a glo-bal function by following simple local rules. It is shownin Fig. 2 that this way the control loop is integrated in-side the system itself, thus the system can organize it-self. Such a system evolves on its own obeying its gi-ven limitations. However, we don’t know certainly whathappens at a given point of the system, we can obser-ve a precisely definable global behavior on the systemlevel. Self-organization is not a feature of the system,but a paradigm, which can be helpful in understanding



Figure 1. Router level models of the Internet: engineering model (left), scale-free model (middle) and the degree distribution (right)

and designing certain real world systems (i.e.: comp-lex telecommunication networks).

Algorithms showing classical signs of self-organi-zation have played an important role from the begin-ning in the evolution and success of the Internet. Wecan mention the TCP protocol, which was previouslydiscussed in the article. This protocol also uses a self-organizing technique while it deals with network con-gestion. It can control traffic flow parameters of a link ina decentralized manner, achieving an emergent result,such as fair resource distribution or high link efficien-c y. During the process every node makes strictly localdecisions based on local information to reach a globalgoal.

Another example is the CSMA/CD algorithm, whichis used in the Ethernet protocol. The rules of CSMA/CDguarantee that the communicating parties are able todetect the simultaneous transmissions and the conse-quent collisions on the common channel, and also pre-vent these collisions without a centralized control me-chanism. If more nodes try to send packets on the sameCSMA/CD channel at the same time, then the parties in-dependently stop transmitting for a random time inter-val, hoping that next time they try to send packets, theirpackets won’t collide with each other. If there is stillcollision on the channel, they increase the wait time in-terval, thus decreasing the probability of another colli-sion. It is easily deductible that following these simplelocal rules the system realizes the fair and efficient re-source distribution of the common transmission chan-nel between the communicating parties.

The self-organizing systems belong to a highly ac-tive interdisciplinary research field, and they play animportant role in many disciplines (biology, physics, so-cial sciences). The systems utilizing these principleshave many advantages, however we can only find fewapplications in the engineering fields. This can be ex-plained by our lack of complete understanding concern-ing the mechanisms of self-organization. Designing suchsystems require an essentially new approach and de-sign methods compared to our traditional ones.

3.3 Search in large-scale networksMany large-scale networks, which are present in na-

ture (human social networks, protein networks, neuralnetworks, etc.), have good searchability as their import-ant feature. Milgram’s experiment [20] showed in 1961that in human social networks not only short routes ex-ist, but people are able to find them very efficiently de-

pending on only the knowledge of a small local part ofthe whole network.

To achieve good searchability, which evolves in aself-organizing manner in nature, is a difficult task in ar-tif icial networks. An important example is any large-scale telecommunication network.

In the early development of the original concept ofthe Internet one of the most important elements wasthe design of a good routing protocol. The base of theframework that is still used consists of designated de-vices called routers, and the network emerging from theconnections between these. This structure has the cha-racteristic that some designated devices must have par-tial or even global knowledge of the whole topology toroute messages. Since to gather the adequate informa-tion about the network topology takes time, the systemmust be quasi-static, and changes in the topology mayhappen with a limited speed. Considering that the nextgeneration Internet will contain nodes by two or threetimes bigger order of magnitude than the present net-work, and these nodes will dynamically change the to-pology of the network, the original concept needs to besignificantly adjusted.

The realization of efficient searchability is conside-red to be an important problem also in peer-to-peer net-works. These networks utilize the Internet as a base struc-ture, and they create an overlay network, which is re-sponsible for the special P2P addressing and contentsearching functions. To cope with the strong networkdynamism, and the high user activity as they connectand disconnect to the network, the system either usesa deterministic, but less scalable “network-flooding”technique, or apply a non-deterministic, more scalablemethod. These applications still highly depend on theunderlying Internet infrastructure. Our research initia-tives aim at developing an overlay network infrastruc-ture (and the corresponding network protocols), whichis able to handle numerous active (connecting, discon-necting or failing) users without a notable performancedecline. The novel idea behind the research is the useof complex network structures instead of the ring struc-tures present in P2P overlay networks.

The ongoing research about future networks moreand more deals with “clean slate” designs. The recom-mended technologies show many similarities with theinfrastructure-free ad-hoc networks. One of these con-cepts is the geographic position based addressing orgeometry addressing and the search algorithms work-ing with them. The communicating terminals can be mark-



Figure 2. The global management (left) and the structure of a self-organizing system (right)

ed by their geographical coordinates, and the routingis based on a simple greedy algorithm: if X is lookingfor Y, X first looks for its neighbor Z, which is closestto Y. In order for this algorithm to work, the network to-pology must meet some given conditions, for instancethe Unit Disk Graph (UDG) is an applicable structure. Insuch cases, when the required connectivity conditionsare not met, and for example it is true that X is not con-nected to Y and every neighbor of X is farther to Y thusthere is no next step, some adjusted procedures needto be used in the algorithm.

The use of virtual coordinates can be the solution inthis case. The task is to assign these virtual coordina-tes to the corresponding terminals, so that the greedycondition is realized, and the routing is always guaran-teed [21]. The condition can be formalized in the follow-ing:

For every pair of nodes X and Y (X≠Y) there existsa node Z that d (Z,Y ) < d (X,Y ), where d (A,B) is the distance between A and B.

The virtual coordinates can be abstracted from the co-ordinates of the Euclidean plane or space, and they canbe chosen from other abstract sets, after we defined acorresponding distance measure. The greedy routingcapable virtual coordinate addressing assignment iscalled greedy embedding. If the greedy embedding isnot possible due to the connection graph’s special cha-racteristics or other circumstances (i.e. dynamic topo-logy change because of moving or failing nodes or links),we need to introduce complementary search procedu-res, for example face routing [22]. Our ongoing researchdeals with analyzing complex network search algorithmsand topology management algorithms, which are efficientat given topology constraints (i.e. max node degree).

The routing techniques presented above have a com-mon feature: they are easily scalable for large-scale net-works, because each node needs to have informationonly about its neighbors. There is no need for large rout-ing address tables, and the decision-making mechanismis really simple. This type of techniques is often calledrouter-free routing. The procedures can be really effec-tive, if they are supplemented with addressing algo-rithms, which can provide network addresses in a self-organizing manner based on local rules. Our researchaims at realizing applications in accordance with ourtheoretical findings concerning these issues.

The management of large-scale, dynamic and struc-ture-free networks is an active research field, and al-though we already have many basic results at will, thegreat challenges of technological applications are stillahead of us.

4. Conclusion

In this paper an overview was given about two rese-arch fields of the future Internet research where para-digm changes are expected. In the first part the issue

of Internet congestion control was discussed. It was pre-sented that Transmission Control Protocol (TCP) has al-ways been serving as a solution for handling and avoid-ing congestion problems in the Internet. The mechanismof TCP was discussed and a short insight was given in-to the TCP versions that were developed during thehistory of the Internet.

A new idea is also discussed for solving the conges-tion problems in the future Internet. It is not based oncontrol but rather on erasure coding. This solution makesthe tempting promise that a future Internet could bedeveloped without any congestion control. The possibi-lity of this solution is a topic of current research. In the se-cond part of the paper the manageability issues of largescale complex networks have been discussed. It is shownthat in the future Internet network management metho-dologies featuring self-organization must play an impor-tant role. As a widely researched area, the special topicof searching in large networks has been presented inmore detail.

Authors

MÁRTON CSERNAI is an undergratuate student on the Budapest Universityof Technology. He is about to finish his MSc degree in next-generation com-munication networks. He is participating in the ongoing research on futureinternet technologies at the Department of Telecommunications and MediaInformatics. His main research fields are large scale complex networks andnext-generation communication networks.

ANDRÁS GULYÁS received M.Sc. and Ph.D. degreein Informatics at Budapest University of Technologyand Economics, Budapest, Hungary in 2002 and 2008respectively. Currently he is a research fellow at theDepartment of Telecommunications and Media Infor-matics. His research interests are complex and self-organizing networks, network calculus and traffic ma-nagement.

ZALÁN HESZBERGER recieved his M.Sc. and Ph.D.degree in electrical engineering at the Budapest Uni-versity of Technology and Economics (BME), Buda-pest, Hungary in 1997 and 2007, respectively. Current-ly he is an assistant professor at the Dept. of Tele-communications and Media Informatics at BME. Hismain research interests are future internet technolo-gies and complex networking.

SÁNDOR MOLNÁR received his M.Sc. and Ph.D. inelectrical engineering from the Budapest Universityof Technology and Economics (BME), Budapest, Hun-gary, in 1991 and 1996, respectively. In 1995 he joinedthe Department of Telecommunications and Media In-formatics, BME. He is now an Associate Professor andthe principal investigator of the teletraffic researchprogram of the High Speed Networks Laboratory. Dr.Molnár has been participating in several EuropeanCOST and CELTIC research projects. He served onnumerous technical program committees of IEEE, ITCand IFIP conferences. He is active as a guest editorof several international journals and is serving in theEditorial Board of the Springer TelecommunicationSystems journal. Dr. Molnár has more than 130 pub-lications in international journals and conferences.His main interests include teletraffic analysis and per-formance evaluation of modern communication net-works.



BALÁZS SONKOLY received his MSc degree in soft-ware engineering from the Budapest University of Tech-nology and Economics in 2002. Now he is a researchengineer at the Department of Telecommunications andMedia Informatics. His interests are in the field of traf-fic modeling and high speed transport protocols. Hehas authored international journal and conference pa-pers in the are of high speed networks and the eval-uation of TCP protocols.

References

[1] V. Jacobson, Congestion avoidance and control, In Proceedings of ACM SIGCOMM ’88, pp.314–329., Stanford, CA, USA, 16-18 August 1988.

[2] S. Molnár, B. Sonkoly, T.A. Trinh, A Comprehensive TCP Fairness Analysis in High Speed Networks, Computer Communications, Elsevier, Vol. 32, Issues 13-14, pp.1460–1484., August 2009.

[3] S. Floyd, Highspeed TCP for large congestion window, IETF RFC 3649, December 2003.

[4] T. Kelly, Scalable TCP: Improving performance in high speedwide area networks, ACM SIGCOMM Computer Communication Review,33(2):83–91, April 2003.

[5] L. Xu, K. Harfoush, I. Rhee, Binary increase congestion control (BIC) for fastlong-distance networks, In Proceedings of IEEE Infocom ‘04, Vol. 4,pp.2514–2524., Hong Kong, China, 7-11 March 2004.

[6] I. Rhee, L. Xu, CUBIC: a new TCP-friendly high-speed TCP variant,In Proceedings of Third International Workshop onProtocols for Fast Long-Distance Networks(PFLDnet 2005), Lyon, France, 3-4 Februar 2005.

[7] D.X. Wei, C. Jin, S.H. Low, S. Hegde, FAST TCP: motivation, architecture, algorithms, performance, IEEE/ACM Transactions on Networking (ToN),14(6):1246–1259, 2006.

[8] R. King, R. Riedi, R. Baraniuk,Evaluating and improving TCP-Africa: an adaptiveand fair rapid increase rule for scalable TCP, In Proceedings of Third International Workshop onProtocols for Fast Long-Distance Networks(PFLDnet 2005), Lyon, France, 3-4 Februar 2005.

[9] S. Liu, T. Basar, R. Srikant, TCP-Illinois: A loss and delay-based congestioncontrol algorithm for high-speed networks, In Proceedings of First International Conference onPerformance Evaluation Methodologies and Tools(VALUETOOLS), Pisa, Italy, 11-13 October 2006.

[10] K. Tan, J. Song, Q. Zhang, M. Sridharan, A compound TCP approach for high-speed and long distance networks, In Proceedings of IEEE Infocom ‘06, Barcelona, Spain,23-29 April 2006.

[11] R. Wang, K. Yamada, M.Y. Sanadidi, M. Gerla, TCP with sender-side intelligence to handle dynamic, large, leaky pipes, IEEE Journal on Selected Areas in Communications23(2):235–248, 2005.

[12] D. Katabi, M. Handley, C. Rohrs, Congestion control for high bandwidth-delay product networks, In Proceedings of ACM SIGCOMM ‘02, Pittsburgh,PA, USA, 19-23 August 2002.

[13] M. Luby, LT-codes, The 43rd Annual IEEE Symposium on the Foundations of Computer Science, pp.271–280., 2002.

[14] R. Pan, L. Breslau, B. Prabhakar, S. Shenker, Approximate fairness through differential dropping,ACM SIGCOMM Computer Communication Review,Vol. 33, Issue 2, April 2003.

[15] T. Bonald, M. Feuillet, A. Proutière, Is the “Law of the Jungle” sustainable for the Internet?,IEEE INFOCOM ‘09, Rio de Janeiro, Brazil, 19-25 April 2009.

[16] P. Erdôs, A. Rényi, “On the evolution of random graphs”, Publ.: Math. Inst. Hung. Acad. Sci., Vol. 5, pp.17–60., 1959.

[17] S. Milgram, “The small-world problem”, Psychology Today, Vol. 2, pp.60–67., 1967.

[18] Xiao Fan Wang, Guanrong Chen, Circuits and Systems Magazine, IEEE, Vol. 3, Issue 1, pp.6–20., 2003.

[19] S. Floyd, V. Jacobson,“The synchronization of periodic routing messages,”IEEE/ACM Trans. Networking, Vol. 2, No. 2, pp.122–136., April 1994.

[20] Milgram, Stanley,“Behavioral Study of Obedience”,Journal of Abnormal and Social Psychology, 67(1963):371–378.

[21] Cedric Westphal, Guanhong Pei, Scalable Routing Via Greedy Embedding, In Proceedings of IEEE INFOCOM’09 Mini-Conference,Rio de Janeiro, Brazil, April 2009.

[22] J. Li, L. Gewali, H. Selvaraj, V. Muthukumar, “Hybrid Greedy/Face Routing for Ad-Hoc Sensor Network,” Euromicro Symposium on Digital System Design(DSD’04), pp.574–578., 2004.



1. Introduction

For seamless playback of digital audio and video con-tact frames must arrive in the decoder at the pace ofthe frame frequency. Moreover, as the misalignment ofaudio and video produces degradation of the perceivedquality, synchronization between audio and video mustbe maintained. In an IP environment, however, audio andvideo frames are transmitted in packets. These pack-ets travel independently within the network thus suffer-ing from timing and alignment problems. It is the deco-der that, through buffering operations, tries to resolvetiming inaccuracies and presents accurately synchroni-sed content to the viewer.

Another important issue is the error prone transmis-sion of information. The IP network provides only verybasic error protection, which, in some cases, is not ade-quate to keep up the quality of the service. Hence, forcertain services more robust forward error correctionand quality control schemes need to be introduced.

The paper is organised around the aforementionedtwo topics and is structured as follows: Section 2 givesan overview on the architecture of the different IP app-lications and reveals their timing requirements. The nextSection deals with the most dynamically evolving app-lication area, IPTV, and focuses mainly on the differenterror correction schemes that can be used in an IPTV en-vironment. Section 4 presents an implemented approachbased on Reed-Solomon encoding and erasure decod-ing [1] for reducing the number of retransmission requ-ests together with some measurement data. Finally, Sec-tion 5 concludes our paper with some possible utiliza-tion of the implemented forward error correction scheme.

2. Media communication applications

2.1 Architecture [2]The quality of the service offered by a media communi-

cation application is determined by the following factors:


Keywords: IPTV, webTV, quality of service, error correction

Video transmission over IP networks has been gaining more and more popularity recently. One of the crucial problems of videotransmission over IP networks through unreliable links is the susceptibility to errors in the transmission path. Packets lost or discarded by the NIC due to CRC errors must be somehow regenerated. Regeneration can be done by requestinga retransmission, or the packet can be recalculated provided that some redundancy is introduced in the transmitter side. After a general description of media transmission over IP links, the paper describes a method that can be used for forward errorcorrection in IPTV applications.

� INVITED PAPER

Media communications over IP networks –An error correction scheme for IPTV environment

LÁSZLÓ LOIS, ÁKOS SEBESTYÉN

Budapest University of Technology and Economics, Department of Telecommunications{lois, sebestyen}@hit.bme.hu

Figure 1. Basic layout of IP appl icat ions

• Quality requirements as set forth by the consumer and the service provider

• User terminal equipments• Devices and media formats used by

the service provider• Quality of service parameters of the network or

the interlinked heterogeneous networksThe basic building blocks of a media communication

network are the encoder and the decoder. The encoderis responsible for converting the video, audio and datacontent to a format that can be transmitted through thenetwork infrastructure, whereas the task of the decoderis to process such video, audio, and data content, restoretheir synchronism, and present them to the viewer.

Depending on the number and layout of the encodersand decoders, several basic network structures are pos-sible (Figure 1). The most simple media communicationapplication consists of a single encoder (server) and asingle decoder (client) which together make up a very ba-sic peer-to-peer video on demand system (Fig. 1, top left).If both peers possess some encoding and decoding ca-pabilities, they both can act simultaneously as a serverand a client. Hence, encoded information can travel inboth directions, and a video telephony system can be setup (Fig. 1, bottom left). The same architecture allows forconferencing services provided that more peers are al-lowed to join the telephony session.

A more sophisticated approach is when the IP infra-structure supports multicast transmission (Fig. 1, righthand side). In a multicast session, a single encoder supp-lies multiple decoders with a common output stream.The packets of this stream are duplicated (multiplied)and routed by special network components thus reduc-ing the burden on the network.

The communication between the server and the cli-ent follows a hierarchical approach, in which each layerhas its own task as depicted in Figure 2.

2.2 Timing accuracy in media communicationMedia communications can also be characterised by

the timing accuracy. The quality of timing accuracy is ba-sically determined by the seamlessness of playbackand the delay introduced by transmission and process-ing. Based on these quality factors three schemes canbe distinguished. The properties of each are summari-sed in Table 1.

Regardless of the scheme used, the primary aim oftransmission is to supply the user with as high an imageand sound quality as can be ensured by the particular net-work in a reasonable time. A key point in that is choos-ing an appropriate buffering strategy which not only de-creases jitter and ensures seamless playback, but al-so allows for either the possibility of retransmission orthe correction of packet by FEC and interleaving.

An offline service is usually implemented by TCP/IPor HTTP protocols and is intended for downloading con-tent to a temporary or final storage. In an offline service,the terminal equipment or a neighbouring network com-ponent has a storage capacity to store the whole con-tent. The primary aim is safe delivery of content, trans-mission delay is only a secondary factor if relevant at all.

An online service such as a video telephony appli-cation tries to minimize the annoying delay of transmis-sion and presentation. As any buffering operations in-crease the delay, the size of the transmitter and recei-ver buffer is usually limited to a couple of frames. As aresult, there is no time to interleave the content in the trans-mitter side and perform FEC encoding (as that would al-so require buffering), or to request the resending of in-formation in the receiver side. Because re-sending isnot possible, unreliable protocols, such as UDP [3], arepreferred.

Most of the current streaming services belong to thenear-line scheme. Due to the lack of interactivity, basicstreaming applications can tolerate longer delay. Buf-



Figure 2. Functional model of the media encoder and decoder

fering for a couple of seconds is common to streamingapplications. The near-line scheme usually uses unre-liable but more complex protocols to transmit informa-tion. In most cases a control channel is used throughwhich control and link state information can be exchan-ged. A typical near-line service using real-time trans-port protocol (RTP [4]) is depicted in Figure 3.

The audio, video, or data sub-streams (depicted bya simple UDP block in Fig. 3) are either treated separa-tely or combined to form a multiplexed stream. While inthe former case separate UDP sessions and control chan-nels are created for each individual sub-stream, in thelater case the sub-streams are multiplexed and only oneaggregate control channel is set up.

3. Streaming services, IPTV

Streaming services usually use UDP/IP or RTP proto-cols. In a streaming environment not only error freetransmission but also timing accuracy is a concern. Adelay of some seconds is tolerable at the start of play-back, but once started, playback must be uninterrupt-ed and seamless. Although a short delay is allowed, ithas to be minimized as much as possible. Since the de-lay introduced by the UDP protocol is minimal, it is con-sidered to be a favourable choice.

Typical streaming services include applications liketelevision through the Internet (webTV) and Internet Pro-tocol Television (IPTV) services wich have been gainingmore and more popularity recently. The difference be-tween the two is while webTV is implemented on thepublic Internet, IPTV uses an IP infrastructure which isa private and closed system maintained by a serviceprovider. It is this service provider who is responsiblefor providing the content, managing user access thro-ugh access control mechanisms, and, in many cases,supplying the users with terminal equipments (so-cal-led set-top-boxes). As the network is closed, the serviceprovider has freedom to prioritise packets carrying te-levision streams over other packets. (This is certainlynot possible in webTV applications.)

IPTV is often bundled with other services like Voiceover IP (VoIP) telephony, and supplementary Internet ac-cess. These three together form a service often referredto as TriplePlay.

The IPTV network is a switched digital video (SDV) ar-chitecture, in which only streams requested by the view-ers are present. Programs that are not being viewed donot appear on the network or on the network segment.This approach, on the one hand, is very economic in thesense that it saves valuable bandwidth for the serviceprovider which can be used to provide value-added ser-

Media communications over IP networks


Table 1.Characterization ofmedia applications by t iming accuracy

Figure 3. A typical near-l ine streaming service based upon the RTP protocol

vices like Video on Demand (VoD). On the other hand, itallows for building an access network with only as muchbandwidth as required by the subscription of each user.Therefore, the bandwidth of the access network for eachuser is determined by the number of programs that canbe simultaneously accessed by them. If a user has ac-cess to only one standard definition (SD), MPEG4-AVC[5] encoded stream at a time, then the data rate can beas low as 2.4 Mbps. (Although 2.4 Mbps is consideredto be a minimal data rate for standard definition AVC,due to technological reasons this is often reduced to 2Mbps.)

The most typical consumer behaviour in an IPTV en-vironment is watching live television streams. Sinceevery program is watched by many users, the streamsare transmitted in multicast mode. In a multicast ses-sion the network and the additional architecture takecare of both packet multiplication and multicast groupmanagement.

If a program change is requested by the user, the useris moved from their previous multicast group corres-ponding to the program they were watching to a newmulticast group corresponding to the newly requestedprogram. Since multicast group switching is a slow pro-cess, and an additional buffering delay is introduced bythe near-line scheme, a common method is to supply theuser with packets of the newly requested stream througha VoD-like unicast connection of higher data rate. A ssoon as the multicast group switching is complete andthe receiver buffer is full, the terminal equipment canswitch to the normal multicast data stream and continuereceiving that.

3.1 Error correction in IPTV networksAs far as the network and the data link layers are

concerned, IP transmission features no forward errorcorrection. The only error resiliency method used in anormal IP environment is the insertion of a CRC codeinto the packet that can be used for error detection. Ifa packet fails the CRC error check in the receiver, thenit is discarded. How the system behaves in the case ofpacket loss is essential from the point of view of the ser-vice.

3.1.1 Retransmission of lost or erroneous packetsPackets can be lost due to congestion in the network,

or they can be discarded by the network layer due to aninvalid CRC code. Either way, missing packets of multi-media services, if not recovered, produce visual arte-facts.

To reduce the quality degradation, missing packetscan be recovered by requesting their retransmission.Retransmission makes sense only if the round trip timeTRTT that consists of the time needed to send the retrans-mission request plus the time the retransmitted packetarrives is less than the time Tbuf the packet spends in thebuffer till it is either presented or used as a reference todecode other frames:

TRTT ≤ Tbuf

To satisfy the above condition the following approa-ches can be followed:

• RTT must be reduced by placing the server asclose to the clients as possible. An IPTV networkusually features one server at a predefined locationand cannot be freely relocated. The problem of relocation, however, can be overcome by installingso-called secondary caching servers. The cachingservers store a well defined portion of the streamsthat pass through them. Upon a retransmission request from a client, they are ready to resend thepacket through a unicast connection.

• The buffering time Tbuf must be increased. Since in an IPTV environment it means that the STBmust be equipped with more memory, it is usuallynot a plausible approach.

3.1.2 Replacement of lost or erroneous packetsMissing packets can be replaced if enough redun-

dancy is introduced in the system, and this redundancycan be used to regenerate the packets that have beenlost. The redundancy information can either be insert-ed in the packets themselves, or it can be transmitted asa separate correction stream. As in the second app-roach the redundancy information can be discarded byreceivers which do not need or do not support them, andthere is no need to recalculate the CRC code of the ori-ginal packets, this is a more plausible method.

The information can be recovered by utilising app-ropriate encoding (like systematic Reed-Solomon enco-ding) provided that the number of erroneous symbolsremains below a well defined upper bound. The methodfor such encoding is the following:

(1) The transmitter appends N-K parity symbols to the K source symbols.

(2) The K source symbols and the appended N-Kparity symbols together form the encoded wordhaving a data length of N symbols.

(3) Out of the N encoded symbols E symbols arecorrupted during transmission and N-E remainintact.

(4) The receiver receives the N encoded symbols,and recovers the E erroneous ones provided that:

2 ⋅E ≤ N–Kif the error locations are unknown (normal errorcorrection), or: E ≤ N–K

if the error locations are known (erasure error correction).Both normal and erasure error correction means a

trade-off between the channel capacity and error correc-tion capability. To be able to correct E symbols out of theN received ones, 2E or E out of the N symbols must beparity information. A well-known systematic FEC encod-ing scheme that satisfies the above conditions and canbe implemented quite easily is the Reed-Solomon encod-ing. In a test bed which is described in the next section,Reed-Solomon encoding and erasure decoding was us-ed to provide a means to regenerate missing data.



4. Error correction as implemented in a real-world IPTV testbed

The method described in the previous section cannot beapplied directly to the Ethernet packets, as that wouldimply an enormous calculation burden on the system.To reduce calculation complexity a so-called interleav-ing approach is followed, which is depicted in Figure 4.

4.1 EncoderThe encoder reserves two memory areas referred to

as Network Data Table and RS Data Table. Each posi-tion within the reserved memory areas can hold one byteof information. The Network Data Table consists of Kcolumns and 1346 rows, and is used to store incomingpackets that need to be Reed-Solomon encoded. TheRS Data Table has N-K columns and holds the parity in-formation, a generated header and some additional in-formation.

The Reed-Solomon encoder works above GF(8), there-fore the upper bound for N is 255. K can be any arbitraryodd number between 1 and N. The difference N-K deter-mines the error correction capability of the erasure Reed-Solomon decoder as seen in Section 3.1.2.

First, a copy of the Ethernet packets arriving at the en-coder is buffered in the Network Data Table in a column-wise direction. If the packet length is less than the ma-ximum packet length of 1340 bytes (corresponding toseven 188-byte transport stream packet plus the RTPheader), the empty positions within the respective co-lumn are zero padded.

Once all K columns in the Network Data Table arecompletely filled, a 6-byte supplementary informationconsisting of the packet length and a CRC code is app-ended to the end of each column. Since this supplemen-tary information can be regenerated in the receiver si-de provided that the packet it corresponds to does ar-rive, this is information is not transmitted.



Figure 4. Network Data Tableand RS Data Tableused for interleaved encoding anddecoding

In the next step, the useful payload of the packets (ex-cluding the header), plus the zero paddings (if any) andthe supplementary information are Reed-Solomon encod-ed row wise. The calculated parity information is storedin the appropriate row of the RS Data Table.

After the parity information is calculated for all rows,a parity header is generated for each column and someadditional information is recorded in the RS Data Table.The header contains the source and destination addres-ses and ports as well as the position of the column with-in the RS Data Table. The additional information includesthe sequence numbers of the RS encoded RTP packets,their positions within the Network Data Table plus pa-rity information calculated from the length and CRC ofthe incoming RTP packets. These data are used in thereceiver side for reconstructing the Network Data Table,spotting any missing packets, and for validating the dataafter reconstruction. Since the sequence numbers andthe positions are crucial for the correct operation of thesystem, they are repeated in every column.

Finally the columns of the RS Data Table are trans-mitted as user packets.

4.2 DecoderThe decoder works in a similar fashion. It first re-

stores the original order of both the normal and the pa-rity packets, and saves the packets in the Network DataTable and RS Data Table respectively. If a packet, eithernormal or parity, is found to be missing, then all positionsin the respective column are marked as erasures. Onceall packets are accounted for (either inserted or markedas erased), the payload of the missing packets can beregenerated by performing erasure RS decoding rowwise. After the payload is restored, the RTP header caneasily be regenerated, thus all the missing informationis regained.

4.3 System architecture and performanceThe architecture of the testbed that uses interleaved

RS encoding for IPTV streams is depicted in Figure 5.Both RS encoding and RS decoding of a predetermined

stream were performed by computers with two NIC cardsworking transparently in a bridged configuration. Theappropriate packets were filtered and passed to a userprogram which then performed RS encoding and RS de-coding as described in Section 4. The packet loss in thenetwork was modelled by a uniform distribution. The ex-pected value of the ratio of lost packets and the parame-ters of the RS forward error correction could be free lychosen. The network traffic after RS decoding was moni-tored by a measurement device.

The measurement results for a packet loss ratio of 10%for different RS parameters are summarized in Table 2.(The packet loss ratio of 10% means that 10 percent ofthe packets after RS encoding is dropped by the error ge-nerator device.)

What is apparent from the figures (apart from the factthat packets with longer parity are more likely to be cor-rected) is that if the increase in bandwidth (i.e. the coderate) is kept constant, then the ratio of lost packets afterRS decoding is decreased with the increase in messagelength. This agrees with our expectations as the longerthe RS codeword, the less probable is that the numberof errors per RS codeword will considerably exceedthe expected value of the distribution, hence it is morelikely that they can be corrected. In case of a codewordlength of around 255 and uniform distribution, all pack-ets could be corrected provided that the increase inbandwidth was twice the packet loss ration after RS en-coding.

5. Conclusions and future work

In the paper we gave a general description of the twomost severe problems of IPTV networks, and present-ed an approach that can be used for error correction inan IPTV environment. Although the method means anincrease in the overall data rate for an IPTV stream (anincrease inversely proportional to the code rate), it canbe effectively used to decrease the unicast traffic thusmaking the caching servers obsolete.



Figure 5. The measurement setup

As far as the future enhancements are concerned,the introduced error correction system can be expand-ed to use low density parity check codes instead of RSencoding.

Authors

LÁSZLÓ LOIS was born in 1971 in Tatabánya. Hereceived his Masters and Ph.D. degree in softwareengineering from the Budapest University of Tech-nology and Economics (BME) in 1995 and 2005, re-spectively. He is currently with the Department ofTelecommunications at BME, and contributes tothe research, development and educational activi-ties. His research interests include source codingapplications, media communications and IP televi-sion.

ÁKOS SEBESTYÉN was born in 1977 in Budapest.He received his Masters Degree in electrical engi-neering at the Budapest University of Technologyand Economics, Department of Telecommunicationsin 2002. Now as a member of the staff of the Depart-ment, he contributes to the educational and re-search and development activities. His research are-as include digital broadcasting (DVB-T/S/C/H/T2)and IP television.

References

[1] Didier F.,“Efficient erasure decoding of Reed-Solomon codes”http://arxiv.org/pdf/0901.1886v1

[2] ITU-T H.264:Advanced video coding for generic audiovisual services,Series H: Audiovisual and multimedia systems,Infrastructure of audiovisual services – Coding of moving video.

[3] RFC 768 – User Datagram Protocol.[4] RFC1889 – RTP:

A Transport Protocol for Real-Time Applications.[5] ISO/IEC 14496-10:2005:

Information technology – Coding of audio-visual objectsPart 10: Advanced Video Coding.



Table 2. Measurement results for a packet loss ration of 10%after RS encoding for different coding parameters

1. Introduction

As part of the AAL (Ambient Assisted Living) program-me, we have developed an indoor ultrasonic localisationsystem at BAY-IKTI (Institute for Applied Telecommuni-cation Technologies of the Bay Zoltán Foundation).

Such a system can be used in many fields, e.g. formonitoring the daily routine of injured or elderly people[1], or for tracking the movement of customers in a su-permarket in order to observe and analyze their shop-ping habits. A similar approach has been applied to theproblem in a number of research projects worldwide,see “The Bat Ultrasonic Location System” developed atCambridge University [2], or the “Ultrasonic LocalisationSystem” developed at HomeLab, Lucerne University [3].

A common shortcoming of such localisation systemsbased on ultrasonic distance measurement is, however,that in order to achieve adequate accuracy (in the 3 to15 centimetres range), one needs to map the exact po-sition of the sensors with much higher precision (in thesub centimetre range). The process for this kind of high-precision positioning, which mustbe performed prior to the first useof the system, can be technicallydemanding (it is usually done ei-ther using a conventional tapemeasure or with a laser range fin-der), hence adding substantiallyto the installation time of the sys-tem.

In contrast with the abovedifficulties, the method we havedeveloped allows the sensors’positions to be determined qua-si-automatically (i.e. without anypreliminary positioning, eithermanual or instrumental) throughthe accurate measurement capa-city of our localisation system.

The rest of the paper is organized as follows. In Sec-tion 2 we introduce our BATSY system, in the next secti-on we discuss how to determine the position of a pointin space with its distance given from a number of knownpoints, in Section 3 we review and correct some of thepositioning errors resulting from possible distortionsin the ultrasonic distance measurement process, andfinally, in Section 4 we propose a method to obtain thesensors’ positions in an automated way.

2. Description of the BATSY system

We have started to build our ultrasonic localisation sys-tem in 2006 to develop a localisation and motion track-ing tool for our AAL (Ambient Assisted Living) laboratory.Since the system relies on distance measurement basedon the speed of an ultrasound signal, it has been namedBATSY (BAT SYstem). As a result of various improvementsmade to the calculation method, we have managed toreach an accuracy of 3 centimetres.

The BATSY ultrasonic locali-sation system consists of threemain components:1. A number (6-8) of sensor

units mounted on walls, forreceiving ultrasound signals.

2. A computer equipped with a radio module, for receivingradio signals and performingcalculations.

3. A mobile node, capable ofemitting radio and ultrasoundsignals simultaneously (Figure 1).The underlying concept of the

system is based on the differencebetween the speed of sound andthat of a radio signal. A radio sig-


Keywords: ultrasound, localisation, motion-tracking, trilateration, self-configuring

Our indoor ultrasonic localization and motion-tracking system (BATSY) is based on a mobile node capable of emitting radio and ultrasonic signals and a number of ultrasound sensors mounted at known positions in a room. The system uses the node’sdistance to the sensors (as derived from the arrival times of the ultrasound signal) for calculating its position. In this paper we discuss the mathematical and measurement problems related to ultrasonic localization, we propose a possiblesolution algorithm, and we present a method for determining the sensors’ position in an automated way.

� FROM OPEN CALL

Mathematical algorithms of an indoor ultrasonic localisation system

ZSOLT PARISEK, ZOLTÁN RUZSA, GÉZA GORDOS

Bay Zoltán Foundation for Applied Research, Institute for Applied Telecommunication Technologies{parisek, ruzsaz}@ikti.hu

Figure 1. The Batsy mobile node

nal emitted by the mobile node arrives to the computeralmost instantly, while the 40 kHz ultrasound wave dis-patched at the same time travels at the speed of sound,thus it reaches the sensors mounted on the walls sig-nificantly later. Therefore, through measuring the diffe-rence between the arrival times of the radio and the ult-rasound signals to some given sensor, one can calcu-late the distance between the mobile node and the sen-sor. Assuming we know the exact location of the sensorsas well as the distances between them and the mobilenode, we can calculate the position of the mobile node.

3. Trilateration: calculating the positionof a point in space with its distancegiven from three other points

Assume we have three known points in space with co-ordinates S1=(x1,y1,z1), S2=(x2,y2,z2), S3=(x3,y3,z3) res-pectively, and we further know their distances, d1,d2,d3from some unknown point (x,y,z). This unknown point islocated at the intersection of three spheres with S1,S2,S3 as their centres and d1,d2,d3 as their radii, respec-tively. This method is based on the measurement of threedistances, so we call it trilateration. Three spheres inter-sect in two points generally, and these two points of in-tersection are symmetrical with respect to the planethrough S1,S2,S3. The two intersection points are obtain-ed as the solution to the following system of equations:

In the course of the actual calculations, differenceslike (x1–x2),(z2–z3),..., appear in the denominator, whichcauses a problem in case one of them is zero. In prac-tice this happens quite often, since the sensors are us-ually mounted on walls and ceilings, thus some of theircoordinates x, y, z are likely to coincide. The easiest wayto avoid this problem is through adding small, indepen-dently chosen random numbers (in the range of a thous-andth millimetre) to the coordinates. This practically doesnot decrease the accuracy of the calculations, while itavoids division by zero with a probability high enough.

In particular adaptations thereis often an obvious opportunity forselecting the correct solution (i.e.the one corresponding to the actuallocation of the mobile node) out ofthe two candidates. For example, incase the three sensors are locatedon the ceiling of a room, the two so-lutions will fall on opposite sides ofthe ceiling, so it is straightforwardto choose the one inside the room.When there is no chance for a solu-tion of this kind, one can use the al-gorithm discussed in Section 4.1.

4. Correcting positioning errors resultingfrom distortions in distance measurement

In practical use, one has to apply different modificationsto the theoretical algorithm described in Section 3.

4.1 Dealing with distortions in distance measurementSometimes, depending on the position of the mobile

node, an obstacle along the straight line connecting themobile node to a sensor may get in the way of the ultra-sound wave. In that case, sound cannot reach the sen-sor along a direct path, hence distance measurementfails. This would not be of any problem in case no resultwas obtained from such occurrences of measurement;but what usually happens is that the sound wave reach-es the sensor along a longer, indirect path. Thus the sen-sor detects a reflected signal, resulting in an estimateddistance greater than the real one.

In order to avoid a possible localisation error causedby corrupted distance values, one should use a great-er number (6-8) of sensors. This way, the estimated lo-cation of the mobile node would be obtained as the in-tersection of more than three spheres. However, therecould still be errors made in the localisation of the cent-res and the measurement of the radii of the spheres(even additional to the previously mentioned ones), inwhich case more than three spheres may have no inter-section at all.

Both problems can be dealt with simultaneously us-ing an “Adaptive Fuzzy Clustering” algorithm [4].

The algorithm consists of two phases. The first phaseinvolves calculating the intersection of all sphere-trip-lets derived from the measured distances, using the tri-lateration algorithm described in Section 3. If we havemeasured n distances, this will give (3

n) possible loca-tions as a set of points in space. Assuming the numberof more or less accurate measurements is sufficientlyhigh, the set of intersection points belonging to sphereswith a correctly measured radius should be concentrat-ed within a relatively small space segment, whereas theintersection points derived from one or more erroneo-usly determined spheres will be dispersed in space es-sentially randomly.

Figure 2. Occurrences of different measured distancesfrom 1000 measures at a distance of 350 cm

Mathematical algorithms...


The second phase consists of finding the point withthe maximum density within the above derived set ofpossible locations. This is done by calculating the vec-tor average of the points, or the “centre of gravity” of theset. Next, we calculate the average distance of all pointsfrom this centre of gravity, and eliminate those whosedistance is greater than average. In this way, we willhave obtained a smaller set with which to repeat the se-cond phase. We do the repetition until the diameter ofthe set falls below a required threshold value.

4.2 Theoretical and practical accuracy of distance measurement

The BATSY system uses 40 kHz ultrasound; its wave-length is about 8.6 mm in room temperature. The sen-sor we used detects the pressure peaks of air waves,so the arrival of an ultrasound packet will presumablybe recorded at a pressure peak. These peaks lie 8.6 mil-limetres apart from each other, which adds an uncer-tainty factor of 8.6 mm to distance measurement.

The sensor’s sampling frequency – in accordancewith the clock pulse of its microcontroller, and taking asgiven the speed of sound – translates into a sub-millimet-ric measurement accuracy.

In Figure 2 we can see the results from 1000 indivi-dual measurements performed at a given distance. Onecan clearly see the occurrence peaks situated 8.6 milli-metres away from each other.

4.3 Distortions related to the directional anglesThe sensors and emitters are directed in space. This

means that the distance between the sensor and the mo-bile node is measured correctly as long as one is po-sitioned facing the other, but when one or both of themare rotated, the measured distance increases with theangle of their axes. This results in a systematic bias re-lated to the sensor’s and the mobile node’s directionalangles (Figure 3).

To offset the bias related to the sensor’s directionalangle, we have to measure it as a function of the angleand distance variables (Figure 4).

In order to eliminate the distortion from the measu-red distance, one needs to know the approximate dis-tance of the mobile node as well as the angle betweenthe sensor-node line and the sensor’s axis.

If the direction of the sensor’s axis is known before-hand (i.e. it was recorded at the moment of the sensorbeing mounted), the bias related to the sensor’s direc-tional angle can be offset using the following two-stepalgorithm.

In the first step we determine the location of the mo-bile node using the method discussed in Section 3. Thislocation will be inaccurate as yet, since it will containa directional bias, but the deviation from the real posi-tion is not significant. So this inaccurate estimated lo-cation is suitable for calculating the angle between thesensor-node line and the sensor’s axis.

In the second step we calculate the range correctionvalue for the given (angle, distance) pair and subtract itfrom the previously measured distance. Do it for all thesensors, and recalculate the location of the mobile nodeusing the method specified in Section 3. This new loca-tion will be free of any directional angle effect.

We have tested the method in our laboratory and foundthat the distance between the positions calculated withand without the sensor’s directional angle correctionis usually less than 20 millimetres. So if such accuracyis not required or the system is low on CPU performance,it can be omitted.

Correcting the bias related to the mobile node’s direc-tional angle is only feasible if the direction of the node’sultrasound emitter can be determined. In this case, onecan use the same method as the one discussed above.In our specific application however, the mobile node’sdirection was not fixed, so we could only measure thelocalisation error related to the sensor’s directionalangle. Our chosen approach then was to place the mo-bile node to a pre-specified location and rotate it aroundwhile simultaneously calculating its indirectly estimat-ed position through the above described algorithm. Wehave found the difference between the real and the cal-culated positions of the node to fall in the 0 to 45 mmrange.

4.4 Distortions resulting from variations in the speed of sound

The speed of sound varies with temperature, humi-dity and air pressure, and so does the outcome of anydistance measurement procedure relying on soundwaves [5]. Whereas the effects of humidity and pres-sure are negligible from our point of view, a 1°C changein air temperature near the 20°C range causes a sub-stantial, 0.176% change in speed. The consequenceof this for our BATSY system is that a sensor will mea-sure an erroneous distance d instead of the real dis-tance d/q, where q is some quotient depending on tem-perature.



Figure 3. The sensor’s (α) and node’s (β) directional angles

Figure 4.Bias in distance measurement as a function of the sensor’sand node’s directional angle, at a distance of 350 cm

Assuming we have distances measured by four sen-sors, the value of q can in theory be obtained by solvingthe following system of equations:

However, we have found that imprecision in the sen-sors’ coordinates and other inaccuracies prevent thisidea from being put into practice. So if our goal is to setup a system operating in room temperature, and unlesswe have temperature data from other sources (e.g. froman internal thermometer), it is better not to use this kindof temperature correction method altogether.

5. Determining the sensor’s coordinatesin an automated way

In order for the positioning system to yield an accura-te result, the sensors’ coordinates have to be measuredwith high precision. In a real-life deployment of the sys-tem, determining the sensors’ exact position is the hard-est and most time-consuming task. Using a traditionaltape-measure or a laser rangefinder takes hours, and itleads to errors in the centimetres range. Is it possible touse the system itself for locating the sensors?

Obviously, it is impossible to have all the coordina-tes determined by the system, since the origin of the co-ordinate-system and the direction of the axes need to bechosen in some arbitrary way. (For convenience, we havechosen the vertical direction as the third axis of the co-ordinate system.) Thus our goal is to come to a self-con-figuring algorithm involving as little technical difficultyas possible and capable of determining the sensors’ co-ordinates in some chosen Cartesian coordinate system.The algorithm we are about to discuss requires the fol-lowing input data (these must be determined manually):

1. Coordinates of an arbitrary sensor S1.2. One of the non-vertical coordinates of some

other sensor S2.3. The side lengths and the orientation of a triangle

arbitrarily drawn on some horizontal plane.

Figure 5. The tetrahedron M1M2M3S1

5.1 Description of the algorithmDenote the mounted sensors by S1,S2,...Sn, and their

coordinates by (x1,y1,z1), (x2,y2,z2), ... (xn,yn,zn), respect-ively. Let M1,M2,M3 be the vertices of our horizontal tri-angle, and let a,b,c denote the distances between them(Figure 5). According to our initial assumption, x1,y1,z1,x2(or y2), a,b,c are known, and our goal is to determinex i,y i,zi (i =1,...n). Without loss of generality, one can im-pose x1=0, y1=0, z1=0.

Step 1:Measurement

Place the mobile unit at location M1, and let the sys-tem measure its distance from the sensors. (Letd11,d12,...d1n denote these n distances.) Repeat the pro-cess with M2 and M3 so to obtain all the distancesdi j (i =1,2,3; j =1,...n).

Step 2:Calculating the distance between S1 and the plane M1M2M3

First, calculate the volume of a tetrahedron with M1,M2,M3,S1 as its vertices. This can be done in two diffe-rent ways. On one hand, according to Tartaglia’s formula[6], we have

On the other hand (using the notation s =(a+b+c)/2),the area of the base triangle is, by Heron’s formula [6],a s follows

Then, we have V =Tm /3, where m is the height of ourtetrahedron as measured from the base triangle M1M2M3. Making use of the equivalence of these two expres-sions for V, one can easily calculate m, which is pre-cisely the distance between S1 and the plane M1M2M3.

Step 3:Calculating the relative positions of M1,M2,M3

Denote by T the foot of the tetrahedron’s altitude lineconnecting S1 to the base triangle M1M2M3 (i.e. T is theorthogonal projection of S1 to the plane M1M2M3). Letr1,r2,r3 be the distances of M1,M2,M3 from T (Figure 6 –on the next page). From the Pythagorean theorem, wehave

Now fix a two-dimensional Cartesian coordinate sys-tem on the plane M1M2M3, with T as its origin and one ofits axes going through M1. As r1 denotes the distanceTM1, point M1 has coordinates (0,r1) in the afore mention-ed system.

Similarly, r2 denoting the distance TM2 and a denot-ing the distance M2M1, coordinates (x,y) of point M2 areobtained as the solution to the following system of equa-tions



The above equation system has two solutions, amongwhich the correct one is to be chosen in accordance withthe orientation of triangle M1M2M3.

The same method can be used for determining thecoordinates of M2. However, in this case the correct oneout of the two solutions is to be selected imposing thefurther restriction that the distance M2M3 must be equalwith c .

The origin of this particular coordinate system is ob-tained through an orthogonal projection of the originalsystem to plane M1M2M3, yet the directions of theiraxes are different. Thus, we have so far determined thepositions of M1,M2,M3 relative to the new coordinate sys-tem, which we further need to rotate around the axisS1T in order to get their coordinates in the original one.

Step 4:Determining the angle of rotation around the axis S1T

Let us return to the three-dimensional space. Thetwo-dimensional relative coordinates of M1,M2,M3 (asdetermined in Step 3) need to be complemented with athird one, which can be expressed as the negative dis-tance between plane M1M2M3 and point S1, that is, (–m).

Starting from these coordinates, and making use ofthe distances d12,d22,d32, the coordinates of S2 in the ro-tated system can be calculated through the trilatera-tion procedure discussed in Section 2. Its two candida-te solutions being symmetrical with respect to plane theM1M2M3, we need to choose the one which lies on thesame side of the plane as where the sensor is actuallylocated.

Denote by (x ’2 ,y ’2 ,z’2) the coordinates of S2 in the ro-tated system. Its coordinates in the original system are(x2 ,y2 ,z2) where only x2 is known for the present. Follow-ing from the properties of the rotated coordinate system,we have z’2= z2 .

If we transliterate the triplet (x ’2 ,y ’2 ,z’2) to a different co-ordinate system derived from the original one through arotation by α around the vertical axis (Fig. 7), the new co-ordinates will be (x ’2cosα –y ’2s inα, x ’2s inα +y ’2cosα, z’2).From the equality of the first coordinates in the two sys-tems, one easily comes to the trigonometrical equation

This equation has two solutions for α, thus yieldingtwo candidates for S2, which lie symmetrically with re-spect to the plane parallel to axes (z,x) and containingS1. Again, we have to select the correct α, the one cor-responding to the particular candidate for S2 which islocated closer to its real position.

Figure 7. Rotation from the temporary coordinate system

to the original one

Step 5:Calculating the coordinates of M1,M2,M3

in the original coordinate systemIn order to come to the absolute coordinates of M1,

M2,M3, their relative coordinates (as calculated in Step 3)need to be multiplied by the matrix

where α is the one determined in Step 4.

Step 6:Obtaining the coordinates of sensors S2,...Sn

From Step 5 we know the positions of M1,M2,M3,along with their distances from S i: d1i ,d2i ,d3i . Thus the c o-ordinates of S2,...Sn can be determined using the trila-teration algorithm discussed in Section 3.

5.2 Practical considerationsSince the system’s overall precision depends high-

ly on the accuracy of the sensors’ coordinates, it is es-sential to reduce any distortions relative to the processas much as possible.

1. The impact of temperature on the speed of soundcan be offset using a reference measure taken at the be-ginning of the process (calibration). Place the mobile



Figure 6. Orthogonal projection to the plane M1M2M3

node to a known benchmark distance from any particu-lar sensor, and let the system measure its distance.Then, through multiplying all measured distances bythe quotient of the afore measured and the real (bench-mark) distances, the distorting effect of temperature iseliminated.

2. The inaccuracy related to the sensors’ directio-nal angle can be reduced through directing the sensorstowards the triangle M1M2M3.

3. The bias related to the mobile node’s directionalangle can be reduced using the method discussed inSection 4, assuming the ultrasound emitters are direct-ed vertically while localising M1,M2,M3.

4. The imprecision relative to the wavelength of the40 kHz ultrasound (as discussed in Section 4.2) can prac-tically be eliminated through taking the average of se-veral measured distances.

5. There might be sensors whose positions cannotbe determined by algorithm 4.1 (e.g. if points M1,M2,M3happen to be out of sight of a particular sensor). In thiscase, the already localised sensors can be used for de-termining the positions of some additional points M4,M5,M6 and then for localising further sensors through Step 6of the algorithm.

5.3 Test resultsThe algorithm described in section 5.1

and further amended in line with the consi-derations discussed in Section 5.2 was test-ed in a room 5x6 metres in area and 2.8 met-res in height. Points M1,M2,M3 were fixedon the surface of a 72 cm high desk, each ofthem situated 1 meter away from any other.We had put all three coordinates of sensorS1 along with coordinate y of sensor S7 in-to the system, and tried to determine the po-sitions of sensors S2,...S7 (Figure 8).

Sensor S3 had no sight of view to pointsM i so the algorithm couldn’t estimate its po-sition. A plant was obstructing the path be-tween S4 and points M i, so we expected in-correct values for its coordinates. The resultsare shown in Table 1. One can see the accu-racy of the method is good enough to deter-mine the sensor’s initial positions to use inthe BATSY localisation system – if there areno obstacles along the ultrasound’s path.

6. Conclusions

In this paper we presented the BATSY ultrasonic localisa-tion system, discussed the positioning algorithm, obser-ved the localisation errors resulting from possible distor-tions in distance measurement, and proposed ways ofreducing them. We have provided an algorithm for con-figuring the system semi-automatically, and tested the re-sults one can expect when operating the system in real-life conditions. The measurements confirmed that our con-figuring algorithm can be used to determine the sensors’position in ideal circumstances, however, when obstac-les blocked the ultrasound’s path, higher errors appeared.

Acknowledgement

We would like to express our gratitude to all those participated in the creation of the BATSY system:

József Bánlaki, Gyula Bakonyi-Kiss, Martin Becker, Bence Csák, Pál Haraszti, Csaba Megyesi, Gábor Ruzsa.



Table 1. Test results

Figure 8. BATSY test configuration

Authors

ZSOLT PARISEK is a researcher at the Bay ZoltánFoundation for Applied Research Institute for App-lied Telecommunication Technologies (BAY-IKTI). Heis doing research in the field of ambient intelligen-ce. He has graduated at the University of Debrecen,Faculty of Informatics as a software engineer. Hisresearch interests include intelligent transportationsystems, algorithms and data mining.

ZOLTÁN RUZSA received his M.Sc in Mathematicsfrom the Eötvös Loránd University of Sciences in2001. He was an assistant professor at the Buda-pest University of Technology and Economics to2007. He is currently employed as a researcher atBAY-IKTI, the Institute for Applied Telecommunica-tion Technologies. His research interests includeoptimisation, graph theory and localisation algo-rithms used in AAL (Ambient Assisted Living) andtraffic analysis.

GÉZA GORDOS received his M.Sc, Ph.D and Dr.Habil. in Telecommunications from the BudapestUniversity of Technology and Economics (BME) in1960, 1966 and 1994, resp. He holds D.Sc. from theHungarian Academy of Sciences. He is full-profes-sor with the BME serving as head of two sectionsin the Department of Telecommunications and Me-dia Informatics since 1976. His previous employ-ments include 3 years at various universities ab-road (UK, USA) and 7 years in industry, among ot-hers as Chairman of the Board of the HungarianTelecommunications Co.(1992-93). His researchactivities have been focusing on the technologyand management of telecommunication systemsand services as well as on speech processing. In2004 he accepted the invitation from the Hungari-an equivalent of NSF to establish and run as Direc-tor the Institute for Applied Telecommunication Tech-nologies (IKTI).

References

[1] Mitilineos A. Stelios, Argyreas D. Nick, Makri T. Effie,Kyriazanos M. Dimitris, Stelios C.A. Thomopoulos,“An indoor localization platform for ambient assisted living using UWB”, International Conference on Mobile Computing andMultimedia, Proceedings of the 6th Int. Conf. on Advances in Mobile Computing and Multimedia, New York, ACM, 2008, pp.178–182.

[2] Andy Ward, Alan Jones, Andy Hopper, “ A New Location Technique for the Active Office”, IEEE Personal Communications, Vol. 4, No.5, October 1997, pp.42–47.

[3] S. Knauth, C. Jost, M. Fercu, A. Klapproth, “Design of an Ultrasonic Localisation System with Fall Detection”, IET Assisted Living 2009 Conference, 24-25 March 2009, London, UK, http://www.ceesar.ch/fileadmin/Dateien/PDF/NewsEvents/IETAL2009_talk.pdf

[4] Jingbin Zhang, Ting Yan, John A. Stankovi, Sang H. Son,“Thunder: towards practical, zero cost acoustic localization for outdoor wireless sensor networks”,ACM SIGMOBILE Mobile Computing and Communications Review, Vol. 11, No.1, 2007, pp.15–28.

[5] O. Cramer, “The Variation of the Specific Heat Ratio and the Speed of Sound in Air with Temperature, Pressure,Humidity, and Concentration”, Journal of the Acoustical Society of America, Vol. 93, No.5, May 1993, pp.2510–2516.

[6] György Hajós, Bevezetés a Geometriába, Budapest, Tankönyvkiadó, 1960, pp.316–319.




Summaries • of the papers published in this special issue

60 years of Department of Telecommunications and Department ofTelecommunications and Media Informatics

A short summary based on the article by Profs. GézaGordos and László Pap, published in Hungarian in theSpecial Issue of “Híradástechnika” devoted to the anni-versary of the two departments.

On the security of communication network: now and tomorrow

Keywords: information security, privacy, Internet,wireless sensor networks, vehicular communications,RFID systems, network coding

In this paper, we first discuss some security issu-es in the Internet, and we sketch some future researchdirections in this field. Then, we discuss the securityissues in wireless networked embedded systems thro-ugh three examples: sensor networks, vehicular com-munications, and RFID systems. Finally, we give a briefintroduction to the field of network coding, which is anew, promising research area in networking.

Talking machines?! –Present and future of speech technology in Hungary

Keywords: speech technology, speech synthesis,speech recognition, dialogue systems

Speech technology has been an area of intensive re-search worldwide – including Hungary – for several de-cades. This paper will give a short overview of the chal-lenges and results of the domain and the vision of thedevelopment and the application of the technology willalso be introduced.


Keywords: Future Internet, congestion control,complex networks, socio-economics

Future Internet research programs try to ignore andovercome the barriers of incremental development andencourage clean slate designs, propose new visions,architectures and paradigms for the coming 10-20 years.Recent results in congestion control research has show-ed that networks operating without explicit congestioncontrol (like TCP) may survive without congestion col-lapse if appropriately designed in network resourcesand if end systems apply appropriate erasure codingschemes. The exponential growth of the Internet makesit virtually impossible to manage the network with tradi-tional centralized approaches (like the manager-agentone); hence research results of complex networks areexpected to spread over the Internet with its autonomicbehaviors. In this article we give overview and outlookof some interesting research areas connected to the fu-ture Internet research.

Media communications over IP networks – An error correction scheme for IPTV environment

Keywords: IPTV, webTV, quality of service, error correction

Video transmission over IP networks has been gain-ing more and more popularity recently. One of the cru-cial problems of video transmission over IP networksthrough unreliable links is the susceptibility to errorsin the transmission path. Packets lost or discarded bythe NIC due to CRC errors must be somehow regener-ated. Regeneration can be done by requesting a retrans-mission, or the packet can be recalculated providedthat some redundancy is introduced in the transmitterside. After a general description of media transmissionover IP links, the paper describes a method that can beused for forward error correction in IPTV applications.

Mathematical algorithms of an indoor ultrasonic localisation system

Keywords: ultrasound, localisation, motion-tracking,trilateration, self-configuring

Our indoor ultrasonic localization and motion-track-ing system (BATSY) is based on a mobile node capableof emitting radio and ultrasonic signals and a numberof ultrasound sensors mounted at known positions in aroom. The system uses the node’s distance to the sen-sors (as derived from the arrival times of the ultrasoundsignal) for calculating its position. In this paper we dis-cuss the mathematical and measurement problems re-lated to ultrasonic localization, we propose a possiblesolution algorithm, and we present a method for deter-mining the sensors’ position in an automated way.

Summaries • of the papers published in this special issue



Call for PapersProspective authors are invited to submit original research papers for publication

in the upcoming issues of our Infocommunications Journal.

Topics of interests include the following areas:

Data and network securityDigital broadcasting

Infocommunication servicesInternet technologies and applications

Media informaticsMultimedia systems

Optical communicationsSociety-related issuesSpace communications

Telecommunication softwareTelecommunications economy and regulation

Testbeds and research infrastructuresWireless and mobile communications

Theoretical and experimentation research results achieved within the framework of European ICT projects are particularly welcome.

From time to time we publish special issues and feature topics so please follow the announcements. Proposals for new special issues and feature topics are welcome.

Our journal is currently published quarterly and the editors try to keep the review and decision process as shortas possible to ensure a timely publication of the paper, if accepted.

As for manuscript preparation and submission, please follow the guidelines published on our website:http://www.hiradastechnika.hu/for_our_authors

Authors are requested to send their manuscripts via electronic mail (preferably)or on a CD by regular mail to the Editor-in-Chief:

Csaba A. SzabóDept. of Telecommunications, Budapest University of Technology and Economics

2 Magyar Tudósok krt., Budapest 1117, HungaryE-mail: [email protected]

Call for Papers


Format of the manuscriptsManuscripts shall be submitted in MS Word. Authors arekindly requested not to use the formatting tools of Wordsuch as automatic title formats etc. The parts of the pa-pers should be numbered in a hierarchical way. Do notuse more than three levels of hierarchy, ideally only two.Use 12 pts font size. Do not use two-column format.

Length of the manuscriptsThe length of papers in printed format is ideally 4-6 jour-nal pages which correspond to 12,000-16,000 charac-ters (incl. spaces). Wherever appropriate, include 1-2 fi-gures or tables per journal page.

Paper structurePapers should follow the standard structure, consistingof Introduction (the part of paper numbered by “1”), andSummary (the last numbered part), and several Sec-tions in between. The Introduction should introduce thetopic, tell why the subject of the paper is important, sum-marize the state of the art with references to existingworks and underline the main innovative results of thepaper. The Introduction should conclude with outliningthe structure of the paper.

Accompanying partsPapers should be accompanied by an Abstract and afew keywords.

AuthorsIn the title of the paper, authors are listed in the ordergiven in the submitted manuscript by their full namesand affiliations (without postal addresses). An e-mail add-ress of the authors (or at least one of them) is also in-cluded in the heading of the paper. No degrees or othertitles of the authors are given.

One of the authors should provide his/her postal add-ress, fax number and telephone number for eventualcorrespondence and communication with the EditorialBoard.

ReferencesReferences should be listed at the end of the paper inthe IEEE format, see below:

a) Last name of author or authors and first nameor initials, or name of organization

b) Title of article in quotation marks c) Title of periodical in full and set in italics d) Volume, number, and, if available, part e) First and last pages of article f) Date of issue

Format of a reference:[11] Boggs, S. A. and Fujimoto, N.,

“Techniques and instrumentation for measurementof transients in gas-insulated switchgear,” IEEE Transactions on Electrical Installation, Vol. ET-19, No. 2, pp.87–92., Apr. 1984.

All references should be referred by the correspondingnumbers in the text.

FiguresFigures should be black-and-white, clear, and drawn bythe authors. Do not use figures or pictures downloadedfrom the Internet. Figures and pictures should be sub-mitted also as separate files. Captions are obligatory.Within the text, references should be made by figurenumbers, e.g. “see Fig. 2.”

When using figures from other printed materials, ex-act references and note on copyright should be includ-ed. Obtaining the copyright is the responsibility of authors.

Contact addressAuthors are requested to

send their manuscripts viaelectronic mail or on a CD

by regular mail to the Editor-in-Chief:

Csaba A. Szabó

Department ofTelecommunications,

Budapest University of Technology and Economics

2 Magyar Tudósok krt., Budapest 1117, Hungary

E-mail: [email protected]

GUIDELINES FOR OUR AUTHORS



2009/I Selected Papers

FOREWORD

Zoltán Czirkos, Gábor HosszúP2P based intrusion detection

Balázs Mezny, Péter Laborczi, Géza GordosAd-hoc adaptive cruise control algorithm

Miklós Máté, Rolland VidaReliable gossiping in inter-vehicle communication

Andrea Farkasvölgyi, Ákos Németh, Lajos NagySimulation and measurement of a MIMO antenna system

Benedek Kovács, Péter FülöpA client-driven mobility frame system – Mobility management from a new point of view

Kristóf Aczél, István VajkNote-based sound source separation of polyphonic recordings

Izabela Krbilová, Vladimír Hottmar, Bohumil AdamecHome access network model specifications

2009/II

IN THIS ISSUE

Levente Buttyán, Gábor Pék, Ta Vinh ThongConsistency verification of stateful firewalls is not harder than the stateless case

László Gyarmati, Tuan Anh TrinhEmpirical evidence and game-theoretical analysis

Szilárd Jaskó, Gyula Simon, Katalin Tarnay, Tibor Dulai, Dániel MuhiCSP-based modelling for self-adaptive applications

Zoltán GálVoIP LAN/MAN traffic analysis for NGN QoS management

András Tóth, Lóránt Vajda, Ferenc VajdaPersonal Electronic Nurse – medical monitoring system

Mihai AvorniculuiUtilization of UML diagrams in designing an events extraction system

Gyula Sallai, Rolland Vida, Csaba A. SzabóThe 60 years’ anniversary of HTE

Lajos HanzóIEEE WCNC 2009: Explores newest advances in wireless communications and cooperative systems

2009/III

IN THIS ISSUE

Balázs Héder, András BertókDetection of sleet attenuation in data series measured on microwave links

Gábor Németh, Gábor Makrai, János TapolcaiMulti-homing in IP networks based on geographical clusters

Oscar Mayora-Ibarra, Denis Gabos, Elizabeth Sucupira Furtado,Roberto Cavaliere, Agostinho Celso Pascalicchio, Rodrigo Filev MaiaA framework for community-oriented interactive digital television

Zoltán Horváth, Dávid VargaChannel allocation technique for eliminating interferencecaused by RLANs on meteorological radars in 5 GHz band

Shahid Mumtaz, Atílio Gamerio, Kazi SaidulEnhanced algorithm for WIMAX: MIESM

Tamás Szálka, Sándor Szabó, Péter FülöpMarkov model based location prediction in wireless cellular networks

Rolland VidaRenewal of Sister Society Agreement between HTE and IEEE ComSoc

2009/IV Special Issue

GUEST EDITORIAL

60 YEARS OF DEPARTMENT OF TELECOMMUNICATIONS AND

DEPARTMENT OF TELECOMMUNICATIONS AND MEDIA INFORMATICS

Boldizsár Bencsáth, Levente Buttyán, István VajdaOn the security of communication network: now and tomorrow

Géza Németh, Gábor Olaszy, Klára Vicsi, Tibor FegyóTalking machines?! – Present and future of speech technology in Hungary

Márton Csernai, András Gulyás, Zalán Heszberger, Sándor Molnár, Balázs SonkolyCongestion control and network management in Future Internet

László Lois, Ákos SebestyénMedia communications over IP networks – An error correction scheme for IPTV environment

Zsolt Parisek, Zoltán Ruzsa, Géza GordosMathematical algorithms of an indoor ultrasonic localisation system

Contents of the Infocommunications Journal, 2009

Editorial Office (Subscription and Advertisements):

Scientific Association for Infocommunications

H-1055 Budapest, Kossuth Lajos tér 6-8, Room: 422

Mail Address: 1372 Budapest Pf. 451. Hungary

Phone: +36 1 353 1027, Fax: +36 1 353 0451


Web: www.hte.hu

Articles can be sent also to the following address:

Budapest University of Technology and Economics

Department of Telecommunications

Tel.: +36 1 463 3261, Fax: +36 1 463 3263


Subscription rates for foreign subscribers:

4 issues 50 USD, single copies 15 USD + postage

Publisher: PÉTER NAGY • Manager: ANDRÁS DANKÓ

HU ISSN 2061-2079 • Layout: MATT DTP Bt. • Printed by: Regiszter Kft.



Our reviewers in 2009

The quality of a research journal depends largely on its reviewing process and, first of all, on the professionalservice of its reviewers. It is my pleasure to publish the list of our reviewers in 2009 and would like to expressmy gratitude to them for their devoted work. Your Editor-in-Chief

Bálint Ary BME*, Hungary

Ferenc Balázs Birdtelecom, Hungary

Jan Bouda Masaryk University, Czech Republic

Levente Buttyán BME, Hungary

Hassan Charaf BME, Hungary

Péter Fazekas BME, Hungary

Enrico Gregori CNR Pisa, Italy

Antonio Grilo INOV, Portugal

Lajos Hanzó University of Southampton, UK

Zalán Heszberger BME, Hungary

Amália Iványi University of Pécs, Hungary

Gábor Jeney BME, Hungary

Csaba Lukovszki BME, Hungary

Oscar Mayora CREATE-NET, Italy

András Micsik Computer and Automation Institute, Hungary

Zoltán Németh BME, Hungary

Roberto Sabatino DANTE, UK

Elio Salvadori CREATE-NET, Italy

Robert Schulcz Mobile Innovation Center, Hungary

Vilmos Simon BME, Hungary

György Takács „Péter Pázmány” Catholic University, Hungary

György Terdik University of Debrecen, Hungary

Attila Török IKTI, Hungary

Gyula Veszely BME, Hungary

László Zömbik Ericsson Research, Hungary

(* BME – Budapest University of Technology and Economics)

contentscontents guest editorial 1 60 years of department of telecommunications and department of...

Documents