D E P A R T M E N T O F E L E C T R I C A L A N D I N F O R M A T I O N E N G I N E E R I N G D E G R E E P R O G R A M M E I N E L E C T R I C A L E N G I N E E R I N G

ULTRA WIDEBAND INDOOR RADIO CHANNEL MEASUREMENT AND MODELLING

Author _________________________________ Lassi Hentilä Supervisor _________________________________ Seppo Karhu Accepted _______________ Grade _______________

Hentilä L. (2004) Ultra Wideband Indoor Radio Channel Measurement and Modelling. University of Oulu, Department of Electrical and Information Engineer-ing. Master’s Thesis, 79 p.

ABSTRACT

The topic of this thesis project is ultra wideband (UWB) indoor radio channel measurement and modelling. The objective is to construct a statistical model of an UWB radio channel for UWB radio system performance analysis and simu-lations. A baseline of the study is to develop a relatively simple channel model that characterises the measured channel in which UWB devices are expected to work in the future. The channel model must be easy to generate for different environments in order to reduce complexity in performance simulations. How-ever, a simple channel model which well characterises the measured channel is very difficult to implement.

Frequency domain measurements were carried out with a vector network analyser in a 3.1–8.0 GHz range using omnidirectional vertically polarised an-tennas. As a result, a complex transfer function of the radio channel was ob-tained. The measurement environments included office rooms and lecture halls at the University of Oulu. During the measurements, the transmission antenna was in a fixed position and the receiving antenna was moved along a stepped track. The measurement results were recorded on a computer and analysed with Matlab. The transfer function of the channel was transformed to a time domain impulse response using an inverse Fourier transform. The impulse re-sponse was used to construct a power delay profile to extract the final channel characteristics. According to the channel characteristics, a modified version of the IEEE 802.15.3a channel model was created. Path loss analysis was per-formed directly from the transfer function of the channel.

The results show that an UWB indoor radio channel is a very multipath-rich and frequency-selective channel. When the three 100 MHz sub-bands of the measured transfer function are compared with each other, carrier frequency dependence on the statistical properties of the channel is found. The phenome-non is easily seen by observing the amplitude distributions and delay spread of the power delay profiles. The amplitude distributions for an UWB seem to con-form better to a lognormal distribution rather than to a Rayleigh or Rice distri-bution. A free space loss model can be utilised to approximate the slope of the path loss of the indoor UWB signal. Multipath propagated signals can be easily distinguished in the channel impulse response, since its delay resolution is high. A high delay resolution makes it possible to position the shortest signal compo-nent in the impulse response with an accuracy of about one hundred picosec-onds in the measured band. It corresponds to a few centimetres in distance.

Key words: radio channel sounding, impulse radio, multipath propagation, power delay profile, modified IEEE 802.15.3a channel model, vector network analyser.

Hentilä L. (2004) Sisätilan ultralaajakaistaisen radiokanavan mittaus ja mallin-nus. Oulun yliopisto, sähkö- ja tietotekniikan osasto. Diplomityö, 79 s.

TIIVISTELMÄ

Työn aiheena on sisätilan ultralaajakaistaisen radiokanavan (UWB, Ultra wi-deband) mittaus ja mallinnus. Tavoitteena on muodostaa UWB-radiokanavasta tilastollinen malli, jota voidaan hyödyntää UWB-radiojärjestelmän suoritusky-kytutkimuksissa ja simuloinneissa. Lähtökohtana on muodostaa mahdollisim-man yksinkertainen malli, joka luonnehtii hyvin mitattua radiokanavaa, jossa UWB-laitteet tulevaisuudessa toimivat. Kanavamallin on oltava helposti gene-roitavissa eri ympäristöille siten, ettei suorituskykysimuloinneista tule liian kompleksisia. Yksinkertainen ja mitattua kanavaa hyvin kuvaava malli on kui-tenkin vaikea toteuttaa.

Taajuustason radiokanavamittaukset tehtiin vektoripiirianalysaattorilla 3,1–8,0 GHz:n taajuuskaistalla käyttäen ympärisäteileviä pystypolarisaatioantenne-ja. Tuloksena saatiin radiokanavan kompleksinen siirtofunktio. Mittauspaikka-na oli Oulun yliopiston toimistohuoneita ja luentosaleja. Mittausten ajan lähe-tinantenni oli paikallaan ja vastaanotinantennia liikutettiin askelrataa pitkin. Mittaustulokset tallennettiin tietokoneelle ja analysoitiin Matlabilla. Kanavan siirtofunktiosta laskettiin käänteinen Fourier-muunnos, jolloin saatiin kanavan impulssivaste. Impulssivasteesta muodostettiin tehoviiveprofiili, josta laskettiin kanavan ominaisuuksia kuvaavat parametrit. Lopputuloksena muodostettiin IEEE 802.15.3a -kanavamallista modifioitu versio. Etenemisvaimennusanalyysi tehtiin suoraan kanavan siirtofunktioista.

Mittaustulokset osoittavat UWB-radiokanavan olevan hyvin moniteinen ja taajuusselektiivinen. Kun mitatun siirtofunktion kolmea 100 MHz:n alakaistaa verrataan toisiinsa, havaitaan kanavan tilastollisten ominaisuuksien riippuvan myös signaalin kantotaajuudesta. Tämä ilmiö näkyy selvästi tehoviiveprofiilien amplitudijakaumien ja viivehajeen tarkastelussa. Amplitudijakauma UWB-signaalille näyttää noudattavan lognormaalijakaumaa paremmin kuin perintei-siä Rayleigh- tai Rice-jakaumia. UWB signaalin etenemisvaimennusta sisätilas-sa voidaan approksimoida vapaan tilan vaimennuksen mallilla. Monitie-edenneet signaalikomponentit voidaan helposti erottaa kanavan impulssivas-teesta, koska sen viiveresoluutio on hyvin suuri. Suuri viiveresoluutio mahdollis-taa suoraan edenneen signaalikomponentin paikannuksen impulssivasteesta noin sadan pikosekunnin tarkkuudella käytetyllä mittauskaistalla. Se vastaa muutamaa senttimetriä välimatkassa mitattuna.

Avainsanat: radiokanavan luotaus, impulssiradio, monitie-eteneminen, tehovii-veprofiili, modifioitu IEEE 802.15.3a -kanavamalli, vektoripiirianalysaattori.

TABLE OF CONTENTS

ABSTRACT TIIVISTELMÄ TABLE OF CONTENTS PREFACE LIST OF SYMBOLS AND ABBREVIATIONS 1. INTRODUCTION...............................................................................................10 2. OVERVIEW OF ULTRA WIDEBAND TECHNOLOGY ................................12

2.1. Principles of UWB Technology.................................................................12 2.2. UWB Applications.....................................................................................15

3. MULTIPATH RADIO CHANNEL....................................................................16 3.1. Characterisation of Time-Variant Linear Channels...................................16 3.2. Multipath Propagation ...............................................................................18

3.2.1. Effects of Multipath Propagation...................................................19 3.2.2. Power Delay Profile .......................................................................21 3.2.3. Doppler Effect................................................................................23

3.3. Path Loss....................................................................................................24 3.4. Noise and Interference Sources in a Radio Channel..................................25 3.5. Link Budget ...............................................................................................25

4. UWB RADIO CHANNEL MEASUREMENT SYSTEM..................................26 4.1. Basic Techniques for Channel Sounding...................................................26

4.1.1. Narrowband Sounding ...................................................................26 4.1.2. Wideband Sounding.......................................................................28

4.2. UWB Channel Sounding ...........................................................................28 4.2.1. Previous Indoor UWB Measurement Campaigns..........................29 4.2.2. Measurement Setup........................................................................29 4.2.3. Calibration of the System...............................................................33

4.3. Measurement Environment........................................................................33 5. DEVELOPMENT OF UWB CHANNEL MODELS .........................................35

5.1. Data Post-Processing .................................................................................35 5.1.1. Signal Analysis Using an IFFT......................................................36 5.1.2. Positioning of the Shortest Signal Path..........................................38 5.1.3. Effect of Windowing......................................................................39

5.2. Multipath Amplitude Fading .....................................................................40 5.3. Proposed UWB Channel Models...............................................................44

5.3.1. Path Loss Model.............................................................................44 5.3.2. Multipath Model ............................................................................48

5.4. Resulting UWB Channel Model ................................................................52 5.4.1. Modified IEEE 802.15.3a Channel Model.....................................52 5.4.2. RMS Delay Spread and Mean Excess Delay .................................56 5.4.3. Number of Paths.............................................................................58

6. DISCUSSION .....................................................................................................59 7. SUMMARY ........................................................................................................61 8. REFERENCES....................................................................................................63 9. APPENDICES.....................................................................................................68

PREFACE

This Master’s thesis was completed as a part of the CUBS (Concepts for Ultra Wide-band Radio Systems) project at the University of Oulu, Centre for Wireless Commu-nications (CWC). CUBS is a continuation of the first general ultra wideband FUBS (Future Ultra Wideband Radio Systems) project carried out at CWC in 1999–2003. The sponsors of CUBS are the National Technology Agency of Finland (Tekes), Elektrobit Ltd., the Finnish Defence Forces and CWC.

The supervisor of this thesis was Prof. Seppo Karhu and the instructor was M.Sc. Veikko Hovinen, both at the University of Oulu, Telecommunication Laboratory. I would like to thank them for their role in the development and completion of this work. Extended discussions with them assisted me tremendously and prepared me for future work. Special thanks to the project manager, Lic.Tech. Matti Hämäläinen. He gave valuable instructions and diligently reviewed the manuscript. I would also like to thank Prof. Jari Iinatti for being as a second examiner and for reviewing the manuscript.

My thanks go also to the many people who have contributed to my work. I am very grateful to my colleagues Niina Laine, Raffaello Tesi and Tommi Jämsä for ra-dio channel-related discussions. I want to acknowledge Pekka Nissinaho and Juha Pihlaja for their technical assistance and Marko Sonkki for being as a helper during the measurements.

I want to express my warmest gratitude to my wife Pauliina for her constant sup-port and love during this work. I would also like to thank my little son for being a source of inspiration at the home front in this work, and for adding perspective to my view of what is important and what is less important in life. Finally, I wish to thank my parents, sisters and brothers for being interested in my work and studies.

Oulu, May 31, 2004 Lassi Hentilä

LIST OF SYMBOLS AND ABBREVIATIONS B Bandwidth Bc Coherence bandwidth Bf Fractional bandwidth c Speed of light in a vacuum ( 2.998108 m/s) d Distance between the transmitter and the receiver d0 Reference distance, which is the shortest far field distance of the

antenna Dm Test quantity of the Kolmogorov-Smirnov test

mD Critical values of the Kolmogorov-Smirnov test

f Frequency F Fourier transform fc Carrier frequency fD Doppler shift fH Highest frequency limit fL Lowest frequency limit fm Maximum Doppler shift G(f, ) Output Doppler spread function h(t, ) Input delay spread function of the channel h(tk) Impulse response of the channel at time k I0() Modified Bessel function of the first kind and zeroth order IRnorm Normalised impulse response K Rician parameter Kr Number of clusters in the channel model L Number of delay samples Lc Number of multipath components inside the cluster m Number of samples in the dataset M Number of frequency points per swept band n Refractive index of air ( = 1.0003 ) N Number of pulses used to transmit one symbol NP10 dB Number of paths within 10 dB of the peak multipath component NP85 % Number of paths capturing 85 % of the channel energy PL Path loss in decibels PLFS(d) Free space loss in decibels as a function of distance d PL(d) Path loss in decibels as a function of distance d pk,l Variable which is either +1 or –1 pr(r) Probability density function Pr Received power Pt Transmitted power P(x) Cumulative distribution of the hypothesis function r Signal amplitude rs Amplitude of the dominant signal component rs

2 Power of the dominant signal component t Time

Shh(t, ) Delay cross power spectral density Sm(x) Cumulative distribution of the dataset S() Delay power spectral density T(f, t) Time-variant transfer function of the channel Tc Coherence time Tf Length of the time hopping frame

ilT Arrival time of the first path of the lth cluster related to the ith chan-

nel realisation Tp Pulse width tk Time index U(, ) Delay-Doppler-spread function v Relative velocity between the transmitter and receiver w(t) Channel output in the time domain Xσ Zero-mean Gaussian random variable with standard deviation Xi Lognormal shadowing term z(t) Complex envelope of the input signal Angle between an incoming horizontal wave and the direction of

motion i

lk , Gain coefficient of the kth multipath of the lth cluster related to the

ith channel realisation Level of significance k,l Corresponds to the fading of the kth ray of the lth cluster Path loss exponent r Ray decay factor Cluster decay factor h Height difference between two points on the surface t Time shift i Angle of incidence λ Wavelength λr Ray arrival rate, i.e., the arrival rate of the path within each cluster Cluster arrival rate Mean value l Reflects the fading associated with the lth cluster 1 Standard deviation of the cluster lognormal fading term in dB 2 Standard deviation of the ray lognormal fading term in dB L Standard deviation of a lognormally distributed signal x Standard deviation of lognormal shadowing term for total multi-

path realisation in dB

2 Mean power r

2 Mean power of the randomly scattered component Delay A Absolute delay (delay of the first component in the PDP) e Total excess delay

ilk ,τ Delay of the kth multipath component relative to the lth cluster arri-

val time ilT

m Mean excess delay max Upper bound of the detectable delay of the channel in the time do main RMS RMS delay spread Doppler shift variable in the frequency domain

AET Acoustic-to-Electrical Transformer ARA Antenna Research Associates Incorporated BER Bit Error Rate BPSK Binary Phase Shift Keying CDF Cumulative Distribution Function CIR Channel Impulse Response CM Channel Model COST Cooperation in Scientific and Technical Research CW Continuous Wave CWC Centre for Wireless Communications DC Direct Current DS Direct Sequence DSL Digital Subscriber Line DSO Digital Sampling Oscilloscope DS-SS Direct Sequence Spread Spectrum DUT Device Under Test DVD Digital Versatile Disc EAT Electrical-to-Acoustic Transformer EIRP Effective Isotropic Radiated Power ETSI European Telecommunications Standards Institute FCC Federal Communications Commission FFT Fast Fourier Transform FIR Finite Impulse Response GPIB General Purpose Interface Bus GSM Global System for Mobile Communications H0 Null hypothesis H1 Alternative hypothesis IEE Institution of Electrical Engineers IEEE Institute of Electrical & Electronics Engineers IF Intermediate Frequency IFFT Inverse Fast Fourier Transform IR Impulse Radio ISI Inter-Symbol Interference LNA Low Noise Amplifier LOS Line-of-Sight LPD Low Probability of Detection LPI Low Probability of Interception

NLOS Non-Line-of-Sight OLOS Obstructed Line-of-Sight PA Power Amplifier PDA Personal Digital Assistant PDF Probability Density Function PDP Power Delay Profile PG Processing Gain PN Pseudo-Random Noise PPM Pulse Position Modulation PRF Pulse Repetition Frequency RF Radio Frequency RMS Root Mean Square RX Receiver SIMO Single Input, Multiple Output SS Spread Spectrum SV Saleh-Valenzuela Channel Model TDC Time Domain Corporation TH Time Hopping TH-SS Time Hopping Spread Spectrum TM-UWB Time Modulated Ultra Wideband TX Transmitter UHF Ultra High Frequency US Uncorrelated Scattering UWB Ultra Wideband VHF Very High Frequency VCO Voltage-Controlled Oscillator VNA Vector Network Analyser WCDMA Wideband Code Division Multiple Access WLAN Wireless Local Area Network WSS Wide Sense Stationary WSSUS Wide Sense Stationary and Uncorrelated Scattering

1. INTRODUCTION

Development of wireless communication systems is currently receiving a great deal of global attention due to the increasing demand for personal communication ser-vices. The consumer electronics market is expanding fast and the need for higher and higher data rates is evident. All the existing technologies are band limited, which is a limiting factor when striving for higher capacities. A new technology referred to as Ultra Wideband (UWB) may be one solution high data rate and short-range applica-tions.

The background of UWB is in impulse radio technology. The only radio technolo-gies used in the early history of wireless radio around the turn of the 20th century were baseband or impulse radios. Impulse radio has also been used in military com-munications since the sixties, and it has been further developed for non-communication systems. Later on in the 1990s, when impulse radio was studied more, high performance communications systems became the focus of new devel-opment. The old technology, which affected the birth of the wireless radio, is now being revitalised for modern communication systems. [1]

The Federal Communications Commission (FCC) in the United States has speci-fied a spectral mask for unlicensed UWB communications. The allocated spectrum consists of a 7.5 GHz bandwidth from 3.1 to 10.6 GHz, with a maximum allowed effective isotropic radiated power (EIRP) level of –41.3 dBm/MHz. [2]

Regulatory work in Europe is still ongoing, but the framework will very likely fol-low the mask defined by the FCC. Co-existence studies are under consideration, since there are some contradictory results and no jointly accepted interference mod-els and definitions. The European Telecommunications Standards Institute (ETSI) has specified a draft mask, which is presented in Figure 1 together with the FCC mask. [3]

Figure 1. FCC and ETSI (draft) UWB spectrum masks.

ETSI Indoor Limit FCC Indoor Limit Part 15 Limit

- 51.3+87 log(f/3.1)

ETSI Indoor Limit FCC Indoor Limit Part 15 Limit

- 51.3+87 log(f/3.1)

ETSI Indoor Limit FCC Indoor Limit Part 15 Limit

- 51.3+87 log(f/3.1)

3.1 10.6

11

In addition to the regulatory work being done in Europe, two technical proposals for an Institute of Electrical & Electronics Engineers (IEEE) UWB communication standard are still under consideration. The focus is on classical single band and newer multi band methods. The classical single band approach is the case considered in this thesis. The new method uses a multi band approach in which information is divided in multiple > 500 MHz radio frequency (RF) sub-bands at the same time. The latter approach has more proponents, which now include the majority of the UWB device companies [4].

The goal of radio channel modelling is to provide a relatively simple path loss and multipath model of the environments in which the desired UWB devices are ex-pected to operate. Irrespective of its simplicity, the model should correspond to measurements which have been made in order to characterise the channel. Since the environments usually may differ considerably, it is difficult to generate one model that characterises well all the environments. Generation of a path loss model from the measured data is quite straightforward, but multipath modelling can be carried out using many different approaches. A power delay profile (PDP), which shows the relative received power as a function of delay, is utilised to extract the multipath channel model parameters.

The purpose of this thesis is to study and measure the characteristics of an indoor UWB radio channel with laboratory equipment, and to create a reasonable channel model for further performance simulations. This is done by first studying the meas-urement system and programming the control of the antenna movement carriage. The channel is then measured in defined positions at the University campus and the data is stored for post-processing analysis. The post-processing consists of several steps, which lead to the final channel model.

The structure of this thesis is as follows. Chapter 2 presents an overview of UWB technology. Chapter 3 deepens into the theory of the radio channel. Chapter 4 dis-cusses channel measurement technologies and the setup used in this thesis project. Chapter 5 concentrates on the channel model analysis and the results. Chapter 6 draws conclusions and describes further development of the topic, and finally Chap-ter 7 summarises the work.

12

2. OVERVIEW OF ULTRA WIDEBAND TECHNOLOGY

The principles of UWB technology and some of its commercial and military applica-tions are introduced in this chapter. The definition of UWB, its advantages in con-ventional spread spectrum technology and basic pulse shapes are also presented.

2.1. Principles of UWB Technology

In general, an UWB radio system is a type of spread spectrum (SS) wireless commu-nication, and it is defined by the FCC [2] as follows: ‘UWB device is any device where the fractional bandwidth is greater than 0.2 or it occupies 500 MHz or more of spectrum.’ The fractional bandwidth Bf is defined as

( )

LH

LHf

2ffff

B+−

= , (1)

where fH is the highest –10 dB and fL is the lowest –10 dB frequency limit.

An UWB radio system does not use conventional SS technology, although it is re-alised using either direct sequence SS (DS-SS) or time hopping SS (TH-SS) tech-niques. The main differences between these two techniques are their duty cycle and peak-to-average power ratio. Duty cycle is determined as the ratio of the pulse dura-tion to the pulse repetition period. In the DS technique, chips, which are used to spread the information signal over a wide bandwidth, are sent continuously without any time break between them. The TH technique utilises very short impulses and discontinuous transmission, which means there is a silent time interval between the transmitted pulses. This is the reason why the TH technique is also called an impulse radio (IR). For an UWB system using pulse position modulation (PPM) as a modula-tion method, the TH technique is an inevitable choice. When antipodal signalling such as binary phase shift keying (BPSK) is used, either the DS or TH technique can be applied [5].

The basic pulse shapes for UWB transmission are based on different derivatives of short Gaussian pulses. In [6] an UWB transmitter was developed which generates ultra-short Gaussian pulses. The first derivative of the Gaussian pulse in the time domain and in the corresponding frequency domain representations are shown in Figure 2 as functions of pulse width Tp. The technology in [6] is based on a time-modulated ultra wideband (TM-UWB) scheme, which uses PPM as one of the modu-lation methods. The system works in the baseband, so no carrier is needed in the transmission of the signal. Transmission without a carrier is possible, since a single band UWB signal does not contain direct current (DC). It is evident from Figures 2a and b that short pulses have a wider spectrum, and vice versa. The duration of the pulse is typically less than one nanosecond [7] and thus, the –10 dB bandwidth is re-

13

spectively more than one gigahertz. In addition, the pulse repetition frequency (PRF) and pulse waveform directly influence the shape of the frequency spectrum. The main effects are changes in the centre frequency and –10 dB bandwidth. Depending on the modulation method an additional effect can also be observed in the frequency domain of the spectrum. For instance, PPM smoothes the spectrum [6].

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1

−1

−0.8

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

1

Time [ns]

Am

plitu

de

Gaussian Monocycle in Time Domain

Tp = 0.25 ns

Tp = 0.5 ns

Tp = 0.75 ns

Tp = 1.0 ns

Tp = 2.0 ns

0 1 2 3 4 5 6 7 8 9 10 11−70

−60

−50

−40

−30

−20

−10

0

Frequency [GHz]

Pow

er [d

B]

Gaussian Monocycle in Frequency Domain

Tp = 0.25 ns

Tp = 0.5 ns

Tp = 0.75 ns

Tp = 1.0 ns

Tp = 2.0 ns

a) b) Figure 2. Normalised Gaussian monocycle (a) in the time domain and (b) in the fre-quency domain.

The advantages of TM-UWB communication compared to conventional DS-SS are very high bandwidth, which makes high processing gain possible, low chip rate and duty cycle, and thus, low signal processing complexity. The processing gain is achieved by spreading one symbol over many pulses using a pseudo-random noise (PN) code. Low transmission power together with PN-coded pulse-based signal leads to a noise-like ultra wideband signal. The processing gain due to pulse repetition coding is defined as [8]

( )NPG 101 log10= , (2)

where N is the number of pulses used to transmit one symbol.

Because of the short pulse duration, the signal energy is spread over a wide fre-

quency band. The low duty cycle induces very low spectral power density, as one can see from Figure 3. Therefore, UWB communication is not very vulnerable to jamming, and the probability of interception (eavesdropping) and detection becomes very low.

Figure 3 illustrates a rough comparison of the power spectral densities of Global System for Mobile Communications (GSM), Wideband Code Division Multiple Ac-cess (WCDMA) and UWB transmission.

14

PowerSpectralDensity

Frequency

UWB, B > 500 MHz

Wideband B = 5 MHz (WCDMA)

Narrowband, B = 200 kHz (GSM)

Figure 3. Power spectral densities of UWB, wideband and narrowband transmission. B denotes bandwidth.

Pulses of duration Tp are sent at random intervals inside time-hopping frames of length Tf. If the multipath delay of the channel is smaller than the time hopping frame, there is no, in practice, inter-symbol interference (ISI) between adjacent sym-bols. The time hopping technique is illustrated in Figure 4.

PulseAmplitude

Time

Tp Tf

Figure 4. Frame structure for time hopping technique.

Since the pulse repetition interval (frame length) is relatively long compared to the

pulse width, the tail of the transmitted pulse is attenuated before the next pulse is sent. The processing gain due to the low duty cycle is

=

p

f102 log10

TT

PG . (3)

The total processing gain is calculated as [8]

21tot PGPGPG += . (4)

One can calculate a reasonable value for the processing gain in an UWB system. For example, if the duty cycle is given the value of 1 % (Tp = 500 ps and Tf = 50 ns) and the number of pulses used to transmit one symbol is 10, the total processing gain be-comes 30 dB. Compared to conventional spread spectrum systems, such as WCDMA with a maximum processing gain of 25 dB for speech, UWB exceeds this value eas-

15

ily (10log10(3.84 Mcps / 12.2 kbps) = 25 dB, where 3.84 Mcps is the chip rate for WCDMA and 12.2 kbps is the bit rate for speech).

Now, chips for UWB applications offer data rates up to 100 Mbps, which is good enough, e.g., for streaming video and other high data rate applications [9]. The frame length is relative to the data rate, so the longer the frame length is, the lower is the data rate.

2.2. UWB Applications

UWB applications can be divided into two main categories: military and commercial applications. Other categories could be short-range high data rate communication, wide-range radars and positioning systems [10]. The low probability of interception (LPI) and detection (LPD) features provided by TM-UWB are the most important benefits for military use. As stated earlier, the UWB signal is also robust against jamming, which together with its LPI / LPD features makes it an inevitable choice for military use. Some of the military applications include tactical handheld radios, tags, different type of radars, precision location systems and wireless intercom sys-tems [10].

Nevertheless, UWB technology offers the most interesting applications for com-mercial use, such as high speed data wireless local area networks (WLAN), intelli-gent tags for transportation systems, collision avoidance radars and precision location systems [10]. High-speed wireless communication links between home appliances, such as televisions, camcorders, computers and stereo systems would be an interest-ing commercial application in the future. Possible UWB applications for a wireless home network are illustrated in Figure 5 [11]. A review of UWB applications coming in the near future is reported also in [12].

Figure 5. Possible UWB applications for a wireless home network. DVD is a Digital Versatile Disc, PDA is a Personal Digital Assistant and xDSL is a Digital Subscriber Line (of any type).

16

3. MULTIPATH RADIO CHANNEL

The signal path from a radio transmitter (TX) output to a receiver (RX) input is called a radio channel. Generally, antennas are included in the radio channel, but it is more appropriate to consider the channel without antennas as the propagation chan-nel or propagation medium. However, the channel model and the measurements in this thesis deal with the signal environment seen by a receiver, thus the effects of the antennas are included. Figure 6 describes a radio communication system [13].

RADIO CHANNEL

PROPAGATION CHANNEL

TRANSMITTER RECEIVER

RADIO CHANNEL

PROPAGATION CHANNEL

TRANSMITTER RECEIVER

Figure 6. Radio communication system. In theory, the behaviour of a radio channel could be predicted by solving Max-

well’s equations. However, exact solutions of real-life radio channels cannot be found, because the computation would become too complex. Therefore, approximate techniques and measurements are required in order to obtain knowledge about radio channel characteristics.

3.1. Characterisation of Time-Variant Linear Channels

Bello [14] has proved that a radio channel may be regarded as a linear channel, whose statistical characteristics are randomly time-variant. He developed system functions, generally known as Bello functions, which characterise a time-variant lin-ear channel modelled by a linear filter. When z(t) is the complex envelope of the in-put signal, the channel output w(t) is given by the convolutional relationship

∞

∞−

−= τττ dtzthtw )(),()( , (5)

where

t is time, is delay and h(t, ) is the time-variant input delay spread function.

The input delay spread function is commonly called the impulse response of the

channel. This is the time domain representation of the radio channel. Another way is

17

to consider the radio channel in the frequency domain as a time-variant transfer func-tion T(f, t) as

∞

∞−

−== τττ τπ deththFtfT fj 2),(),(),( , (6)

where

f is frequency and F depicts a Fourier transform. If the impulse response h(t, ) is Fourier transformed with respect to time t, the

Doppler spectrum of each of the delay taps is obtained. The result is a delay-Doppler-spread function U(, ), where depicts the Doppler shift variable in the frequency domain. To determine another appropriate Bello function, which shows the Doppler spectrum associated with each frequency component, either U(, ) or T(f, t) is Fourier transformed with respect to or t, respectively. The resulting func-tion is the output Doppler-spread function G(f, ). The relationships between the Bello functions are presented in Figure 7. The Doppler effect is further discussed in section 3.2.3. From the channel modelling point of view, the functions h(t, ) and T(f, t) are inevitable, and they can be determined by time domain or frequency domain measurements, respectively.

The autocorrelation function of the input delay spread function h(t, ) is the delay cross power spectral density Shh(t, ). When t = 0, Shh(t, ) expresses the delay power spectral density S(), also known as a power delay profile (Figure 7).

INPUT DELAY SPDEAD

FUNCTION

OUTPUT DOPPLER-SPREAD FUNCTION

DELAY-DOPPLER-SPREAD FUNCTION

TIME-VARIANT TRANSFER FUNCTION

),( tfT

),( fG

),(τU

),(th

F fF 1− tF

1−F

F tF fF 1−

1−F

DELAY CROSS POWER SPECTRAL DENSITY

),( tS hh ∆

auto-

correlation

DELAY POWER SPECTRAL DENSITY

0=∆ t

( IMPULSE RESPONSE )

S()

Figure 7. Fourier transform relationships between the Bello functions for time-variant linear channels and the transition from the impulse response (input delay spread function) to the delay power spectral density via the delay cross power spec-tral density.

In practice, a radio channel is typically a multipath channel where the transmitted signal arrives at the receiver via multiple propagation paths with different delays, amplitudes and phases [15]. In order to simplify the analysis of multipath radio

18

channel characteristics, the Bello functions can be assumed to be wide sense station-ary, both in time and frequency. The channel is then called a WSSUS channel. In the case of WSS (Wide-Sense Stationary) the channel is wide sense stationary in time, and in the case of US (Uncorrelated Scattering) the channel is wide sense stationary in frequency. A stochastic process is said to be a WSS process if it has a constant mean value and the autocorrelation function depends only on the time difference. US assumption means the autocorrelation function is independent of frequency [14]. In terms of a multipath radio channel, the contributions from scatterers with different time delays are uncorrelated. In practice, no radio channels are stationary, but charac-terisation of the channels without WSSUS assumption becomes extremely difficult [16].

3.2. Multipath Propagation

Signal propagation between the TX and the RX in a multipath radio channel passes along a direct line-of-sight (LOS), a non-line-of sight (NLOS) and an obstructed line-of-sight (OLOS) path. NLOS paths are caused by reflection, scattering, refraction and diffraction. The effects of multipath propagation can be described in both the frequency and time domains, as it is pointed out in section 3.2.1. The propagation mechanisms can be divided into four categories, which are discussed next. Figure 8 illustrates multipath propagation [17].

Reflection occurs when an electromagnetic wave encounters a smooth, large sur-face compared to the wavelength of the signal. Surface roughness is defined by the Rayleigh criterion [17] as

icos8<∆h , (7)

where

h is the height difference between two points on the surface, is the wavelength of the propagating signal and i is the angle of incidence. When the surface is shading into rough in comparison to the wavelength, the sig-

nal scatters from it and the reflected component becomes weaker. Thus the scattering happens when a radio wave encounters a particle whose dimensions are comparable to or smaller than the wavelength. In such a case, the wave spreads out in all direc-tions and only a fraction of the signal continues directly towards the receiver [17].

When an electromagnetic wave passes from one transparent medium to another, it changes velocity and bends, i.e., changes direction. The amount of bending depends on the refractive indices of the mediums and the angle between the propagation di-rection of the wave and the normal of the surface separating the two mediums. This phenomenon is called refraction.

19

Diffraction is a phenomenon in which the wave front of the propagating wave bends in the neighbourhood of an obstacle, into the shadow. As a result, the bent wave makes transmission possible even if no LOS connection exists between the TX and the RX. Huygen’s principle explains an important feature of this phenomenon: Each point on a primary wave front serves as the source of spherical secondary wavelets such that the primary wave front later is the envelope of these wavelets [17].

TX

RX(1)

RX(2)

NLOS(1)

LOS

NLOS(2)

OLOS

LOS = line-of-sight, direct pathNLOS(1) = non-LOS, diffracted pathNLOS(2) = non-LOS, reflected pathOLOS = obstructed LOS, refracted path

Figure 8. Multipath propagation.

3.2.1. Effects of Multipath Propagation

A short pulse is sent though a multipath radio channel. Multipath propagated signals arrive at the receiver with different delays, amplitudes and phases. The received su-perimposed signal is spread over time and is called the impulse response of the chan-nel.

The maximum detectable delay of the band-limited signal above the noise level at the receiver is given as max. Bandwidth B of the receiver multiplied by max plays a significant role when the properties of the measured radio channel are analysed. If Bmax << 1, signals collected by the receiver merge into one peak. This kind of chan-nel is called a narrowband radio channel, which can be modelled by a one-ray chan-nel model. If Bmax 1, the channel is still narrowband, but a few different signal paths can be identified. If Bmax >> 1, the impulse response broadens and the delay resolution increases, since the sent pulse is wideband. Due to the delay resolution of the impulse response, many individual paths of the signal can be identified. The higher the bandwidth of the signal, the more details of the propagating environment can be identified [18]. When the signal is ultra wideband, the delay resolution be-comes very good, which leads to reasonable accuracy for indoor positioning systems [19]. An example of a measured indoor ultra wideband radio channel impulse re-sponse is depicted in Figure 9a. The frequency band used in the figures is from 230 to 390 MHz. The delay resolution in Figure 9a is 6.25 ns.

20

The effect of multipath propagation is seen in the frequency domain as the fre-quency selectivity of the channel. Different signal paths either strengthen or weaken the signal, depending on the delay difference of the arriving signals, as shown in Figure 9b. In order to have a frequency-selective radio channel, the relative time de-lay of the paths must be significantly different. This condition is fulfilled when the relative time delay is comparable to or shorter than the duration of the transmitted symbols of the system [17].

0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2−90

−80

−70

−60

−50

−40

−30

Rec

eive

d P

ower

[dB

m]

Propagation Delay [µs] 240 260 280 300 320 340 360 380−100

−95

−90

−85

−80

−75

−70

−65

−60

−55

−50

−45

Rec

eive

d P

ower

[dB

m]

Frequency [MHz] (a) (b) Figure 9. Example of an indoor ultra wideband radio channel (a) impulse response and (b) transfer function measured in an office environment. Multipath Fading Short-term fluctuation in the strength of the received signal is called fast fading, and long-term fluctuation is known as slow fading or shadowing. Fast fading is caused by the destructive or constructive addition of multipath signal components when the mobile terminal moves or when there is some other movement in the surroundings. The distance between two successive fast fading minima is typically about half a wavelength.

Two probability distributions, called Rayleigh and Rice are widely used to de-scribe the statistics of this phenomenon [15]. Rayleigh distribution is usually de-scribed as a good approximation for the measured fading amplitude statistics for mo-bile fading channels in NLOS situations. The theoretical probability density function (PDF) of Rayleigh fading is given by [15]

)

2(

2r2

2

)(

r

er

rp−

= , (8)

where r is signal amplitude,

2 is the mean power of the received signal and r2/2 is the short-term signal power.

21

In practice, there is occasionally a dominant signal path, which is either a LOS component or a strong reflected component. In this case, the best fitting theoretical distribution usually is a Rician distribution. The PDF of the Rician function is given by [15]

)

(

)( 202

2r

2

2s

2

s

rrrr

Ier

rp

+−

= , (9)

where rs is the dominant signal component and I0() is the modified Bessel function of the first kind and zeroth order.

The Rician PDF is often expressed in terms of the Rician parameter K, defined as

[15]

2r

2s

2

rK = , (10)

where

rs2/2 is power of the dominant signal component and

r2 is the mean power of the randomly scattered component.

Shadowing is caused by dynamic evolution of the propagation paths, wherein new

paths arise and old paths disappear as the mobile terminal moves. The statistics of the mean received amplitude typically follow a lognormal distribution [15]

)

2

) )(ln((

2L

r2

2

2

1)(

−−

=r

er

rp , (11)

where is the mean value of signal strength in decibels and L is the standard deviation in decibels.

The standard deviation is typically in the order of 4 to 10 dB [15].

3.2.2. Power Delay Profile

The power delay profile of a radio channel is obtained by calculating an autocorrela-tion of the channel impulse response and observing it at time zero, as discussed in section 3.1. A sketch of the PDP and its parameters is shown in Figure 10, which is modified from [17]. The difference between the first and last signal components in the PDP of the radio channel is defined as total excess delay, usually called just ex-cess delay e. The delay of the first component in the PDP expresses the absolute de-

22

lay A of the channel. The distance between the TX and the RX can be calculated us-ing A, if the direct path exists. A represents the arrival time of the first signal path in the PDP. Therefore, it is the most important delay time considered in positioning sys-tems.

Power

Delay

RMS

A

m

e

Figure 10. Power delay profile of a radio channel. The following parameters, which can be extracted from the channel measurements,

are represented as discrete time. Discrete time representation considers a finite num-ber of samples from the measured data. Sampling converts a continuous signal into a discrete time signal.

Mean excess delay m corresponds to the centre of gravity of the measured delay profile defined by [20, 34]

=

=−

=L

kk

L

kkk

th

tht

1

2

1

22A

m

)(

)()(

τ , (12)

where h(tk) is the sampled impulse response of the channel, tk is delay at time k and L is the number of delay samples.

A widely used channel modelling parameter for describing the delay spread of a channel is root mean square (RMS) delay spread RMS, defined by [20, 34]

=

=

−−=

L

kk

L

kkk

th

tht

1

2

1

22Am

RMS

)(

)()( . (13)

23

RMS delay spread characterises the multipath channel well, since it takes into ac-count the relative power of the delay taps. The excess delay, which presents the amount of total delay spread is inversely proportional to coherence bandwidth of the channel as [21]

ec

1∝B . (14)

Thus, large excess delay corresponds to a narrow coherence bandwidth. The rela-

tionship between the coherence bandwidth and the bandwidth of the transmitted sig-nal expresses whether the radio channel is frequency selective or frequency non-selective, as depicted in section 3.2.1. The radio channel is frequency selective if the bandwidth of the transmitted signal is large compared to the coherence bandwidth.

3.2.3. Doppler Effect

The Doppler effect, which is one of the main causes of channel variability, is directly proportional to both the carrier frequency and the relative velocity between the TX and the RX. Hence, higher carrier frequencies and mobility induce higher Doppler shifts, and thereby faster channel fluctuations. The Doppler shift is given by [17]

cosmD ff = , (15) where fm is the maximum Doppler shift given by

cv

ff cm = , (16)

where v is the relative velocity between the TX and the RX, c is the speed of light,

is the angle between an incoming horizontal wave and the direction of motion and

fc is the carrier frequency. Doppler spread is seen in the frequency domain as a spreading of the received sig-

nal bandwidth relative to the transmitted bandwidth. Thus, the received signal band-width is not the same as the transmitted one. The variability of the channel is seen in the amplitude distribution of the multipath propagated components. If the amplitude distribution is Rayleigh-like in a certain path, the channel causes a Doppler spread. The more the amplitude distribution is Rice-like, the less Doppler spread is found. The rate of change of the channel can be defined with the Doppler spread [22]. The

24

time domain duplicate of the Doppler spread is coherence time, which is inversely proportional to the Doppler spread of the channel, as [17]

mc

1f

T ∝ . (17)

3.3. Path Loss

A signal propagated through the radio channel undergoes attenuation, which is called path loss. Path loss is defined in decibels as [23]

=

r

t10log10

PP

PL , (18)

where Pt is transmitted power and Pr is received power.

For isotropic antennas, free space loss is defined in decibels as

=

4log20)( 10FS

ddPL , (19)

where d is the distance between the TX and the RX.

Mean path loss is described as a function of the distance between the TX and the

RX. Since path loss is generally defined as the median value, it is necessary to pro-vide its variations. In UWB measurement campaigns, for example in [24], the total normalised mean energy was found to be lognormally distributed. Path loss is usually presented in decibels as [23]

0100FS )/(log10)()( XdddPLdPL ++= , (20)

where d0 is a reference distance, which is a far field distance of the antenna, is a path loss exponent and

Xσ is a zero-mean Gaussian random variable with standard deviation σ, both given in decibels.

In [25] quite comprehensive UWB path loss measurements were done in different

indoor environments. Also in this thesis, one task is to determine the path loss model of a typical indoor office environment. The basis of link budget estimation is given in (18)–(20). The link budget is described in more detail in section 3.5.

25

3.4. Noise and Interference Sources in a Radio Channel

Noise and interference in a multipath radio channel modify the transmitted signal, affecting the complexity of signal recovery in the receiver. There are two types of noise sources in a wireless radio channel, as shown in Figure 11 [17]. Multiplicative noise is a result of, for example, directional characteristics of antennas, reflections from buildings and different surfaces, absorptions of walls, trees and atmosphere, scattering from rough surfaces, diffractions from edges and refractions from atmos-pheric layers. The receiver itself, atmospheric effects, cosmic radiation, interference and some external devices cause additive noise. In channel modelling, noise is in-cluded in the channel characteristics, since it is one part of the measured channel. Therefore, channel parameters which have been extracted from the measured data give good approximations of the behaviour of a real radio channel in that specific measured environment.

x +

MULTIPLICATIVE

NOISE

ADDITIVE

NOISE

x +

MULTIPLICATIVE

NOISE

ADDITIVE

NOISE

Figure 11. Two types of noise sources in a wireless radio channel.

3.5. Link Budget

Link budget planning is part of the radio network planning process, which helps to estimate the required coverage, capacity and quality of service requirements in the network. The link budget determines the power requirements for the required bit er-ror rate (BER) of the system. Link budget calculation involves addition and subtrac-tion of decibel gains and losses within the radio link. When the gains and losses of various system components are determined and summed, the result is an estimation of the maximum acceptable path loss for the system. Then, using (20), the coverage of the system can be calculated. To reach an accurate answer, factors such as trans-mission powers, amplifier gains, cable and matching losses, noise figures, transmit and receive antenna gains, path losses, required signal-to-noise ratio, fade margin and interference levels must be taken into account [17]. Some of the link budget fac-tors that have been used in the measurements for this master’s thesis are listed in Table 1 in section 4.2.2.

26

4. UWB RADIO CHANNEL MEASUREMENT SYSTEM

There are many types of radio channel measurement techniques, as discussed in [15]. A narrowband technique is most widely used to excite the channel. There are also many types of wideband channel sounding techniques based on continuous wave (CW) and pulse techniques [26]. The best way to observe multipath channel effects is to transmit short pulses through it. However, generation of short pulses is difficult, and the receiver needs to be able to quickly process a very high bandwidth signal. Channel sounding can be performed either in the time domain or the frequency do-main. Since the impulse response and the transfer function are Fourier transform pairs, it is sufficient to measure only one of them in order to gain knowledge of the channel characteristics. In general, the choice of the sounding method is made be-tween narrowband, wideband or ultra wideband and between the time or frequency domain. For example, one criterion can be the fractional bandwidth Bf. In addition, the choice of a particular technique depends on the set of desired channel model pa-rameters and the application. The main study in this thesis is concentrated on UWB measurement techniques and measurement campaigns.

4.1. Basic Techniques for Channel Sounding

4.1.1. Narrowband Sounding

A signal is said to be narrowband if the inverse of its bandwidth is much greater than the delay differences of the different propagation paths. Narrowband sounding tech-niques are usually based on unmodulated RF carrier (single tone) transmission [16]. The amplitude and phase of the received signal fluctuate strongly due to multipath propagated signal components, which arrive at the receiver in random phases and amplitudes.

Narrowband sounding techniques are also used to achieve the wideband or even ultra wideband frequency response of the channel. In this technique the tone is swept over the required band by linearly stepping through a band of frequencies. The method is usually considered to be wideband, even though it is based on narrowband sounding. The vector network analyser (VNA) which was used in the UWB channel measurements in this thesis project utilises the frequency sweeping technique. The transfer function of the channel is obtained by measuring a scattering parameter S21 with the VNA, where the radio channel is the device under test (DUT). A generalised block diagram of the VNA is presented in Figure 12 [27]. To gain the knowledge of the DUT, the VNA must provide a source for stimulus, signal-separation devices, receivers for signal detection, and display / processing circuitry for reviewing the re-sults. The source is usually a built-in phase-locked voltage-controlled oscillator (VCO) [27]. The measurement system based on the frequency sweeping technique is discussed in more detail in section 4.2. The technique is quite affordable to realise (compared to pulse generators), but there are some drawbacks in it. The sweeping consumes some time to go through a wide frequency band. Thus, it is not possible to use the technique in wideband mobile measurements, where the terminals or the en-

27

vironment is moving during the sounding. Measurements of Doppler shift or angle of arrival are also not possible [16].

Figure 12. Generalised network analyser block diagram. A modified channel sounder based on the VNA has been developed in [27] and

presented in Figure 13. The traditional VNA-based sounder is confined to use within a distances of some tens of metres, since the cables of both the TX and RX antennas are connected to the VNA. The modified sounder eliminates this problem by employ-ing an external transmitter and receiver. The triggering pulse for the synchronisation is sent via an external radio link, and thus arbitrary radio link distances can be meas-ured.

PC withLabView

Network Analyzer

TX-port RX-port

50 Ω

GPIB

Sweeping signalgenerator

external triggerpulse 5/10 MHz

external triggeringpulses

LNAPA

external frequency reference

external frequency reference5/10 MHz

Radio

AET

AFGU

Radio

EAT

EAT = Electrical-to-Acoustic Transformer

AET = Acoustic-to-Electrical Transformer

Synchronization/triggering

Radio channel probing signal

(optional)

Transmitter

LNA = Low Noise A m plif ierAFG U = Function GeneratorPA = Pow er Am plif ierG PIB = General Purpose Interface Bus

Figure 13. Modified frequency domain UWB channel sounding system.

28

4.1.2. Wideband Sounding

In order to acquire deep knowledge of the effects of multipath propagation on a radio channel, the sounding technique needs to utilise a wideband probing signal. Relevant channel characteristics, such as average delay and delay spread, can be extracted from the wideband measurement data.

The most frequently used technique for wideband channel sounding uses DS-SS transmission with an IQ demodulator ( I is the in-phase component and Q represents the quadrature component) and a sliding correlator at the receiver. This technique was previously called swept time-delay cross correlation [16] and was used the first time in a wideband sounder by Cox [29]. In this technique a reference PN sequence at the receiver has a slightly slower chip rate than at the transmitter. The received signal is cross-correlated with the reference PN sequence. The difference between the chip rates cause noise in the correlation, which decreases the dynamic range of the sounder and therefore also its performance [30].

In addition, instead of using a sliding correlator, a stepping correlator can be used at the receiver [31]. In this technique a locally generated reference PN sequence is time-shifted relative to the received signal in discrete steps. Sampling rates increase year by year, and hence the stepping correlator is more competitive than a sliding correlator.

A filter matched to the waveform of the sounding pulse is one possible alternative to the correlator in the receiver. The technique that utilises a matched filter is known as the convolution matched filter technique. The receiver is simpler than the correla-tor receiver, since no continuous local PN sequence generation is needed. Therefore, the technique is asynchronous. [16]

A single pulse-based method for channel sounding is periodic pulse sounding [16]. Short pulses are used to excite the channel. As presented in the theory of linear sys-tems in section 3.1, the received signal represents the convolution of the sounding pulse with the impulse response of the channel. In a pulse-based measurement sys-tem a digital sampling oscilloscope (DSO) is typically used in the receiver to detect the multipath components. Periodic and relatively high-rate pulse transmission is ob-viously needed in order to observe the time-variant behaviour of the channel. The pulse repetition interval must also be relatively long compared to the pulse width, so that the tail of the received pulse is attenuated sufficiently before the next pulse is received.

One of the advantages of the DS-SS technique over the pulse techniques is that less peak power is needed, since transmission is continuous.

4.2. UWB Channel Sounding

During the past few years, ultra wideband channel measurement campaigns have been carried out in many laboratories around the world [32–43]. The focus of the measurements has been to deepen knowledge of the behaviour of the radio channel when the width of the frequency band gets as wide as an ultra wideband. Previous

29

wideband channel models have usually been based on measurements of up to several hundred megahertz bandwidths at the maximum.

4.2.1. Previous Indoor UWB Measurement Campaigns

Most of the UWB channel measurements have been carried out with a VNA utilising the frequency sweeping technique [33–37, 40–42]. In addition, a few time domain pulse-based techniques have been used to excite the channel [32, 38, 43]. Research-ers from the University of Southern California and Time Domain Systems Inc. (TDC) performed the first indoor UWB channel sounding campaign in 1997 [32]. Their measurement system consisted of a periodic pulse generator that transmits an approximately one nanosecond pulse every 500 ns and a DSO as a receiver. A few years after the first campaign the frequency sweeping technique utilising a VNA came into focus [33–37, 40–42]. New analyser models allowed quite fast frequency sweeping using a several gigahertz bandwidth and enough sample points for rea-sonably accurate data post-processing. An interim goal for many researchers was to measure the indoor UWB radio channel in response to the IEEE 802.15.3a Call for Contributions on Ultra-wideband Channel Models (IEEE P802.15-02/279r0-SG3a) [44]. In the IEEE conference on Ultra Wideband Systems and Technologies in 2002, five papers concerning VNA-based channel measurements were presented [33–37].

TDC [38] and AT&T Labs [24] have published measurement results using pulse-based techniques. TDC has developed a scanning receiver based on a TM-UWB sig-nal. The transmitter emits a pulse train, a tracking correlator synchronises with the received pulse train, providing coherent transmission, and the scanning correlator scans the received signal in the same manner as a DSO. The measurement system employed by AT&T Labs sounds the channel using nanosecond pulses and records the response using a DSO.

After the IEEE 802.15.3a Channel Modelling Sub-committee Report [39], re-searchers from CEA-LETI [40], the University of Aalborg [41] and France Telecom R&D [42] published their measurement results and UWB channel models based on VNA measurements at the First International Workshop on Ultra Wideband Systems (IWUWBS) in 2003 in Oulu, Finland.

RADIOLABS at the University of Rome [43] has carried out an indoor measure-ment campaign based on a different measurement system. The sounding was done in the FCC-compliant 3.6–6.0 GHz band using a carrier modulated by a train of pulses shaped by a PN sequence. This technique has some advantages over the other tech-niques, such as the higher transmitter power level and accurate phase recovery.

4.2.2. Measurement Setup

The UWB radio channel measurement system used in this thesis project consists of an Agilent 8720ES vector network analyser [45], an Agilent 83017A wideband am-plifier [46], an Antenna Research Associates (ARA) CMA-118/A wideband conical monopole antenna pair [47], 8 and 15 m coaxial cables, a stepped track for antenna

30

movement, a control computer with LabVIEWTM 6i software and a Leica precision laser meter [48]. The measurement setup and analysis diagrams are shown in Figure 14 [33].

Channel impulse response

Network Analyzer Agilent 8720ES

PC withLabVIEW 6i software

GPIB-bus

TX-port

RX-port

Computer network

AmplifierAgilent 83017A

TX (ARA CMA-118/A) RX (ARA CMA-118/A)

Analysis softwarein Matlab

Channel characterization

Inverse Fourier /Chirp-ZTransformation

Channel models

Channel frequency response

Statistical analysis

Figure 14. Measurement setup and analysis diagrams.

The network analyser was operated in a transfer function measurement mode,

where port 1 was the TX port and port 2 was the RX port. An external amplifier was connected to port 1 to increase the transmission power level. The transfer function and impulse response of the amplifier are shown in Appendix 1. The antennas are vertically polarised conical monopole antennas having typically omnidirectional ra-diation patterns and constant phase centres. Appendices 2–8 show the radiation pat-terns of the antennas. The sweep time of the VNA depends on the frequency points within the sweep band, being automatically adjusted by the VNA. Table 1 lists the main parameters of the measurements. The antenna gain given by the manufacturer was not verified by measurements, but the radiation patterns shown in Appendi-ces 3–5 give the gain values of the antennas that were used.

The selected frequency band falls within the FCC spectrum mask from 3.1 GHz to 10.6 GHz for UWB transmission. The frequency band used in the measurements is from 3.1 GHz to 8.0 GHz, giving a bandwidth of 4.9 GHz. In addition, 5.0 GHz band from 1.99 to 6.99 GHz was measured in some cases. The loss of the cables can be seen from Appendix 9. The maximum EIRP of the system can be calculated from the values in Table 1 by adding the antenna and amplifier gains to the transmission power and subtracting the cable losses. The result is 26.2 dBm, which corresponds to 420 mW. The maximum number of frequency points per sweep is 1601.

The values above can be used to calculate the maximum detectable delay of the channel as [45]

31

,)1(

max BM −=τ (21)

where M is number of points in the swept frequency band.

Table 1. Measurement setup and link budget parameters Parameter Value Frequency band (a) 3.1 to 8.0 GHz Frequency band (b) 1.99 to 6.99 GHz Bandwidths (a) / (b) 4.9 GHz / 5.0 GHz IF bandwidth of the VNA 3.0 kHz Number of points over the band 1601 Maximum detectable delay (a) / (b) 326.5 ns / 320 ns Sweep time 800 ms Dynamic range 90 dB Average noise floor –120 dBm Transmission power +5 dBm Amplifier gain (min/mean/max) 35.3 / 35.9 / 36.9 dB Amplifier output power (max) +30 dBm Amplifier delay 0.60 ns Antenna gain (typical/max @ 25° elevation) 0 / 4.2 dBi Antenna polarisation Vertical TX cable loss (min/max) 3.5 / 8.0 dB RX cable loss (min/max) 0.6 / 1.0 dB EIRP (max) 26.2 dBm = 420 mW

Using (21) and the parameters defined in Table 1, the maximum detectable delay

is 326.5 ns, which corresponds to about 98 m in distance. It is reasonably long for the indoor environments where the measurements were carried out.

Because the VNA is used to transmit and receive the sounding signal simultane-ously, the frequency of the received signal is shifted compared to the transmitted one due to propagation delay. The frequency shift f is calculated by [33]

,sw

tr

=∆

tB

tf (22)

where ttr is the propagation time of the signal and tsw is sweep time. Using (21) and the values from Table 1 so that the propagation time is equal to the

maximum detectable delay max, the frequency shift f was found to be 2 kHz, which

32

is less than the 3 kHz intermediate frequency (IF) bandwidth of the VNA. If f were larger than the IF bandwidth, the received signal would be distorted by the IF filter.

The measured transfer function data was transferred via General Purpose Interface Bus (GPIB) to the control computer, where the collected data was stored for later analysis. The stored data was later converted into suitable format for MatlabTM post-processing.

The dynamic range of the VNA reported by Agilent [45] is 100 dB, but the cable losses decrease it to approximately 90 dB. The transfer functions of the cables are shown in Appendix 9. Using the free space loss formula (19), the range of the meas-urement system can be approximated to be 94 m when using 8.0 GHz and 243 m with 3.1 GHz. The highest frequency is the limiting factor of the system.

The measurement setup and analysis diagrams are shown in Figure 14 [33]. The analysis part is described in detail in the post-processing part in section 5.1. The measurement system was placed on a trolley as shown in Figure 15a, making the sys-tem easier to move. In order to enhance the antenna positioning accuracy, a stepped track (antenna carriage) was used at the RX end. The stepped track with the antenna is shown in Figure 15b. During the measurements, the RX antenna was moved along the track in 1.0 cm steps. The length of the track is 2.35 m, giving 235 different an-tenna positions. The TX antenna was in a fixed position during the measurement campaign. The antenna heights at both ends were 1.34 m, measured from the radia-tion centres of the antennas.

The stepped track at the RX end also made it possible to consider the measurement data as if there had been a single input, multiple output (SIMO) transmitter and re-ceiver, respectively.

(a) (b)

Figure 15. Measurement setup: (a) trolley and (b) antenna with carriage.

33

4.2.3. Calibration of the System

Before each measurement session, the VNA was calibrated using full 2-port calibra-tion. The calibration procedure required the amplifier and the antennas to be discon-nected from the measurement setup. After calibration, a zero delay value refers to the connectors of the antenna elements. The antenna and the amplifier bring some excess delay, which is independent of the environment. Appendix 1 presents the delay of the amplifier. The effect of the amplifier was taken into account during the post-processing of the data. The frequency domain’s raw measurement data was divided by the transfer function of the amplifier before the inverse Fourier transform (IFFT) to compensate the delay and the gain. Since the transfer function of the antenna is not available from the manufacturer and now it is not possible to measure the antenna patterns, the effects of the antennas have not been removed from the measurement data. Therefore, the antennas are included in the resulting radio channel model.

4.3. Measurement Environment

The measurements were carried out at spatially distributed locations, mainly in the Tietotalo building at the University of Oulu. In addition, some measurement data was collected from lecture halls on the main university campus. In the Tietotalo building the transmitting antenna was placed in room TS440. The track of the receiving an-tenna was located in two rooms adjacent to TS440, which were TS441 and TS472. A 1.5 m wide corridor borders all three rooms, so all the measured links included two walls. The wall material of rooms TS440 and TS441 is plasterboard and of room TS472, concrete.

This construction contains different measurement distances from 4 m to 10 m and two kinds of environments for analysis: NLOS1 and NLOS2. NLOS1 contains non-line-of-sight with two plasterboard walls between the antennas. NLOS2 contains plasterboard and concrete walls.

Three track positions in TS441 and two in TS472 give five positions. Given that, one track has 235 positions, the total number of measurement positions is 705 in the NLOS1 case and 470 in the NLOS2 case. Figure 16 illustrates the measurement envi-ronment in the Tietotalo building.

The LOS and part of the NLOS1 measurements were performed in the university lecture halls SÄ118, L5 and L6. The construction blueprints of the measurement sites are presented in Appendices 10–11. The TX antenna was located in a fixed position 2 to 4 metres from the wall. The RX end was moved along a straight line about 1 m from the wall both inside and outside the room. The wall between the TX and RX was a single layer brick wall in the case of SÄ118 and L5 and a solid double brick wall in the L6 case.

The LOS measurements were performed in the frequency band from 1.99 to 6.99 GHz. In the cases of NLOS1 and NLOS2, the frequency band from 3.1 to 8.0 GHz was used.

34

1,51

m

2,98

m

4,10 m

4,10 m

8,72

m1,59 m

4,18 m

TX

TS440

TS441

TS472

RX

stepping rail

Corridor

5,0

m

3,0

m

1,0

m

RX

RX

RX

RX

stepping rail

NLOS2

NLOS1

1,51

m

2,98

m

4,10 m

4,10 m

8,72

m1,59 m

4,18 m

TX

TS440

TS441

TS472

RX

stepping rail

Corridor

5,0

m

3,0

m

1,0

m

RX

RX

RX

RX

stepping rail

NLOS2

NLOS1

Figure 16. Floor plan of the office building where the channel measurements were carried out.

stepped track

35

5. DEVELOPMENT OF UWB CHANNEL MODELS

In the literature, various different channel models describe and parameterise the be-haviour of an UWB multipath channel. The time domain model is most frequently used [24, 32–35, 38–40, 42–43]. The linearity approach, as presented in section 3.1, is used in these cases to obtain the channel impulse response function. All the pa-rameters of the impulse response function are random numbers or processes [14]. In order to model these parameters, the channel measurement campaign has to be per-formed empirically to get the random processes.

The frequency domain model of the channel is also used in some cases [49]. In those cases, a frequency transfer function is used as a representation of the linear time-variant channels.

This chapter discusses the channel modelling. Data processing steps to used obtain the channel model, IFFT analysis, positioning of the paths, effects of windowing, multipath amplitude fading, proposed UWB channel models, path loss and multipath modelling, RMS delay spread and number of paths are considered.

5.1. Data Post-Processing

Before the channel model was extracted from the raw measurement data, various data processing and signal analysis stages were performed. All the signals detected in different rooms and positions during the measurement campaign were used to obtain the PDPs of each position. The effects of small-scale and large-scale statistics were analysed separately. Small-scale statistics were extracted from one track position containing 235 antenna positions, which correspond to a distance of 2.35 m. Large-scale statistics were achieved by merging all the track positions into the same pool and averaging the PDPs spatially.

In the previous channel measurement campaigns presented in section 4.2.1, the statistical properties of the channel were studied using the full measured frequency band. This approach assumes that the frequency does not affect the channel statistics. In this thesis the analysis was based on a 100 MHz sub-band (SB) starting from the lowest frequency (lower sub-band), another 100 MHz sub-band ending at the highest frequency (higher sub-band) and a median sub-band, as shown in Figure 17. The re-sults of the different sub-bands were then compared with each other. 100 MHz was chosen so that the results could also be compared to the previous wideband channel measurement campaigns, e.g., [50]. The small-scale transfer functions of the meas-ured full band and the sub-bands in TS441 are presented in Figure 18a and b.

Lower SB

4.9 GHz3.1 GHz 8.0 GHz

100 MHz100 MHz 100 MHz

Median SB Higher SB

Figure 17. Illustration of the sub-band division.

36

(a) (b)

Figure 18. Transfer functions of the measured bands: (a) full band and (b) sub-bands. Windowing was used to obtain the arrival time of the first path in the PDP, but the

channel model parameters were extracted without windowing. Windowing sharpens the edge of the PDP, making positioning of the arrival time easier. In addition, win-dowing distorts the frequency spectrum and underestimates the delay spread. Win-dowing consists of multiplying the signal in the frequency domain by a window function that is zero outside some defined range. In this thesis a Hamming window was used, since it blurs in frequency but produces much less leakage than, e.g., a rec-tangular window. The effect of windowing is discussed additionally in section 5.1.3.

For all impulse responses, normalisation was performed by setting the channel en-ergy at each position to unity so that the area under each PDP is equal to one. The normalised impulse response IRnorm is obtained by

=

=L

kkth

thIR

1

2norm

)(

)(. (23)

The PDP is then a squared value of the IRnorm. Normalisation makes it possible to

compare the statistics of PDPs that were measured at different positions.

5.1.1. Signal Analysis Using an IFFT

The signal measured using the VNA is the transfer function of the channel. An IFFT was used to transform the measured frequency domain data to the time domain. The IFFT is usually taken directly from the measured raw data vector (typical method). This processing is possible since the receiver has a down-conversion stage with a mixer device. This method is referred to as a complex baseband IFFT, and it is suffi-cient for modelling narrowband and wideband systems. Since single-band UWB sig-nals are carrierless and pulse-based, no mixer device is needed at the receiver. Thus, signal processing in the UWB system becomes easier.

37

There are two common techniques for converting the signal to the time domain, which both lead to approximately the same results. The first approach [40] is based on Hermitean signal processing, which gives a better pulse shape. The second ap-proach [51] has been found to be an easier and more efficient way to obtain the same pulse shape accuracy. These two approaches are introduced next.

Hermitean Signal Processing Using the Hermitean processing, the passband signal is obtained with zero padding from the lowest frequency down to DC, taking the conjugate of the signal and re-flecting it to the negative frequencies. The result is then transformed to the time do-main using an IFFT. This Hermitean method is shown in Figure 19 [52]. The signal spectrum is now symmetric around DC. The resulting double-sided spectrum corre-sponds to a real signal. The time resolution of the received signal is more than twice that achieved using the baseband approach. This improvement in accuracy is impor-tant, since one purpose of UWB channel modelling is to accurately separate the dif-ferent signal paths.

*

Figure 19. Idea of the Hermitean approach. Conjugate Approach The conjugate method involves taking the conjugate reflection of the passband signal without zero padding. Using only the left side of the spectrum, the signal is con-verted using an IFFT with the same window size as in the Hermitean method. The technique is presented in Figure 20. The result using the conjugate method is practi-cally the same as that stated in the Hermitean case. The conjugate method is more efficient in data processing, since the matrix calculations in the post-processing stage become easier. Matlab adds the zeros itself during the IFFT, and therefore the mem-ory requirement is decreased. In this method, 3 dB of power needs to added in order to maintain the channel energy, since the measured spectrum is one-sided. The im-

38

pulse responses of these two methods and the baseband method are shown in Figure 21.

Figure 20. Idea of the conjugate approach.

22 23 24 25 26 27 28 29 30

−0.06

−0.04

−0.02

0

0.02

0.04

0.06

Propagation delay [ns]

Rec

eive

d am

plitu

de

Different IFFT MethodsTypical methodHermitean approachConjugate approach

Figure 21. Impulse responses of the different IFFT methods.

From Figure 21, it can be seen that for all the methods, the main reflection posi-tions are the same and the amplitudes are very close. Thus, the approach based on the left side conjugate produces an adequate pulse shape with lower processing complex-ity.

5.1.2. Positioning of the Shortest Signal Path

Each of the PDPs has an initial delay before the first and other multipath components arrive. The initial delay is the time that is spent by the radio wave propagating through the shortest path from the TX to the RX. Typically, the shortest path is not the LOS path, but the arrival time of the path is usually taken as the arrival time of the LOS component. After the initial delay is defined from the measurement data, a distance between TX and RX can be calculated. Respectively, if the distance be-tween the antennas is accurately measured, the arrival time of the shortest path is eas-

39

ily extracted from the PDP. The arrival time of the first signal component in the pro-file can be estimated as [24]

ncd/A =τ , (24)

where n is the refractive index of air ( = 1.0003).

In order to accurately measure the distance between the antennas, a Leica Disto

precision laser meter was used. Leica’s standard deviation of the measuring accuracy is ± 3 mm for a 100 m distance [48]. However, the error is less than one millimetre for the distances of antenna separation in the channel measurement campaign. The accuracy of the Leica was proved by the test measurements. Figure 22 illustrates the positioning of the direct path performed with two methods. The difference in metres can be calculated from the sample interval. The sample interval is 62.5 ps, which cor-responds to 1.9 cm in distance. Thus, the maximum error in positioning is less than 4 cm. The sample number in Figure 22 depicts only the vector index. The UWB sig-nal provides very high accuracy for direct path positioning from the measurement data.

0 50 100 150 200 250256

258

260

262

264

266

268

270

Sam

ple

num

ber

Position of the antenna on the rail

Positioning of the direct path in NLOS1 case

Propagated direct pathMeasured direct pathSample interval = 62.5 ps

Figure 22. Positioning of the first component of arrival of the PDP.

5.1.3. Effect of Windowing

Let us assume the situation where the channel transfer function is unity over the whole 4.9 GHz bandwidth. This corresponds to an ideal band limited channel. By transforming this response to the time domain using an IFFT, the impulse response of the transmitted signal was found and is shown in Figure 23a. The red signal in the figure depicts a sinc function, which is the impulse response of an ideal low pass fil-ter, which cuts off at half the sampling rate. Since the bandwidth is limited, it is nec-essary to use windowing in the frequency domain in order to get less oscillation at the passband close to the cut-off frequencies. It must be noted that windowing modi-

40

fies the signal by distorting the spectrum and therefore, the channel model that was generated in this thesis project was considered without windowing. In addition, win-dowing underestimates the RMS delay spread. Figure 23a illustrates the effect of windowing on the ideal response and Figure 23b on the measured signal. The meas-ured result shows strongly the effect of windowing on the received signal shape. The noise level decreases, making it easier to position the first arrival component.

0 0.5 1 1.5 2 2.5 3 3.5 4

−0.5

0

0.5

1

Time [ns]

Nor

mal

ized

Am

plitu

de

Effect of Windowing (Hamming)No windowingWindowing(Hamming)

0 50 100 150−80

−70

−60

−50

−40

−30

−20

−10

0Effect of Windowing in Frequency Domain

Propagation delay [ns]

Rec

eive

d P

ower

[dB

]

No windowingHamming window

(a) (b)

Figure 23. Effect of Hamming windowing on the signal shape: (a) an ideal impulse response and (b) the measured impulse response.

5.2. Multipath Amplitude Fading

Amplitude fading in a multipath radio channel may follow different distributions de-pending on the measurement environment. Rayleigh and lognormal distributions are the best candidates in a NLOS channel and Rice in a LOS channel [20]. The analysis here was divided into a small-scale and a large-scale statistics. The small-scale area was chosen to be 43 wavelengths, calculated according to the 5.5 GHz centre fre-quency. That contains 235 positions on the track and 1880 sweeps altogether. Ampli-tude fading distributions were calculated to show the variability of the amplitude through the small-scale PDPs. A comparison was done by fitting the measured am-plitudes to lognormal, Rayleigh and Rice distributions. The Kolmogorov-Smirnov test [53] was used to show the reliability of the fit. The test is standard for testing data for a cumulative distribution function (CDF). It determines if the dataset and the hypothesis differ significantly by comparing their CDFs. The test has two hypothe-ses, H0: the distributions of the datasets do not differ and H1: the distributions differ. The test quantity Dm is [53]

)()(max xSxPD mxm −= , (25)

where P(x) is the cumulative distribution of the hypothesis function, Sm(x) is the cumulative distribution of the dataset and m is the number of samples in the dataset.

41

Dm is compared to the values in a table of critical values

mD and the level of sig-nificance is then obtained. The level of significance is the probability of a false re-jection of the null hypothesis in the test. If Dm <

mD , the assumed CDF is an accept-

able fit to the dataset for a significance of . Otherwise, the hypothesis H0 must be rejected.

For the measured data, a significance of 1 % is used to evaluate the reliability of the fit. Traditionally, 1 % and 5 % are the most commonly used values. The tables below show how the different CDFs fit to the data. Table 2 depicts all the measure-ment environments from LOS to NLOS2 using the full measured band. Table 3 com-pares the fit in the case of LOS with different sub-bands and the NLOS1 case is com-pared in Table 4. The sub-bands of the NLOS2 case are presented in Table 5.The pass rate stands for the passing percentage of the test.

Table 2. Comparison of pass rates of multipath fading distributions using a full band FULL BAND

Pass rate of Lognormal

(%)

Pass rate of Rayleigh

(%)

Pass rate of Rice (%)

LOS (lecture hall)* 99.5 52.4 45.6 NLOS1 (office)** 83.3 13.6 0.8 NLOS2 (office)** 78.1 7.6 1.3

* Large-scale ** Small-scale Table 3. Comparison of pass rates of multipath fading distributions in SÄ118 Large-scale SÄ118 – LOS

Pass rate of Lognormal

(%)

Pass rate of Rayleigh

(%)

Pass rate of Rice (%)

Lower sub-band (100 MHz) 96.8 71.0 41.9 Median sub-band (100 MHz) 96.8 58.1 64.5 Upper sub-band (100 MHz) 96.8 67.7 32.3 Full band (5.0 GHz) 99.5 52.4 45.6

Table 4. Comparison of pass rates of multipath fading distributions in TS 441 Small-scale TS 441 – NLOS1

Pass rate of Lognormal

(%)

Pass rate of Rayleigh

(%)

Pass rate of Rice (%)

Lower sub-band (100 MHz) 87.1 3.2 3.2 Median sub-band (100 MHz) 83.9 12.9 9.7 Upper sub-band (100 MHz) 93.6 0 16.1 Full band (4.9 GHz) 83.3 13.6 0.8

42

Table 5. Comparison of pass rates of multipath fading distributions in TS 472 Small-scale TS 472 – NLOS2

Pass rate of Lognormal

(%)

Pass rate of Rayleigh

(%)

Pass rate of Rice (%)

Lower sub-band (100 MHz) 60.0 23.3 0 Median sub-band (100 MHz) 83.3 16.7 3.3 Upper sub-band (100 MHz) 86.7 3.3 6.7 Full band (4.9 GHz) 78.1 7.6 1.3

It is evident from the tables that the lognormal distribution fits best to the data in

all the cases. The same result was obtained in some previous UWB measurement campaigns, such as [24] and [40]. Figure 24 shows an example of the CDF fitting result.

Figure 24. Cumulative distribution functions of the measured data and the fitting. When observing the large-scale sub-band approaches, it can be seen from Table 3

that the Rayleigh distribution improves the percentage value in the LOS case in both sub-bands. This results from the fact that when the bandwidth is decreased, more signal components merge into one path in the PDP and thus the statistical process of the given path amplitudes becomes more and more random, which is Rayleigh-like. Proof of the Rayleigh-like nature is also provided in various previous wideband radio channel measurements, which are discussed in detail in [20]. The figures below show the change in the delay resolution in the PDP when the bandwidth is decreased. Figure 25 illustrates the LOS channel, Figure 26 the NLOS1 channel and Figure 27 the NLOS2 channel.

43

0 20 40 60 80 100 120 140 160 180 200 220−130

−120

−110

−100

−90

−80

−70

−60

−50

Rel

ativ

e R

ecei

ver P

ower

[dB

]

Delay [ns]

Full Band vs. 100 MHz Sub−Bands in LOS

Full band 2.99 − 7.99 GHzLower sub−band 2.99 − 3.09 GHzMedian sub−band 5.44 − 5.54 GHzHigher sub−band 7.89 − 7.99 GHz

Figure 25. PDPs of the channel obtained with sub-band division. Note the used fre-quency band, which is different from the other cases.

0 20 40 60 80 100 120 140 160 180 200 220−130

−120

−110

−100

−90

−80

−70

−60

Rel

ativ

e R

ecei

ver P

ower

[dB

]

Delay [ns]

Full Band vs. 100 MHz Sub−Bands in NLOS1

Full band 3.1 − 8.0 GHzLower sub−band 3.1 − 3.2 GHzMedian sub−band 5.5 − 5.6 GHzHigher sub−band 7.9 − 8.0 GHz

Figure 26. PDPs of the channel obtained with sub-band division.

44

0 20 40 60 80 100 120 140 160 180 200 220−130

−120

−110

−100

−90

−80

−70

Rel

ativ

e R

ecei

ver P

ower

[dB

]

Delay [ns]

Full Band vs. 100 MHz Sub−Bands in NLOS2

Full band 3.1 − 8.0 GHzLower sub−band 3.1 − 3.2 GHzMedian sub−band 5.5 − 5.6 GHzHigher sub−band 7.9 − 8.0 GHz

Figure 27. PDPs of the channel obtained with sub-band division.

Figures 25–27 show that when the bandwidth of the signal is decreased, many ad-

jacent multipath components merge into one path. That makes it very difficult to separate and position a certain path from the PDP. The figures also show that a signal in the higher sub-band decays more rapidly compared to the median and lower sub-bands. This effect is evident from (19) whereby the free space loss can be calculated.

5.3. Proposed UWB Channel Models

Papers [24–25, 33, 35–36, 38–40, 42–43] consider UWB channel measurements and modelling. Several different statistical channel models have been generated. The va-riety of models consist of a path loss model [25, 36, 38, 42], multipath models such as a tapped-delay-line model commonly called a finite impulse response (FIR) model [24, 43], a dual slope model [33], a single cluster model [35] and a multi-cluster model, i.e., a modified Saleh-Valenzuela (SV) model [39–40].

5.3.1. Path Loss Model

The path loss model is based on a study of propagation properties of the UWB chan-nel. Path loss is defined as the power reduction from the transmitter to the receiver location, as discussed in section 3.3. A general path loss formula (20) contains a variable Xσ, which gives the statistical variability of the path loss values. In order to obtain reliable statistics of the model by measurements, a large amount of measure-

45

ment positions in different environments must be included in the study. Quite com-prehensive measurements and the path loss model are presented in [25].

In this thesis project, path loss was studied in all the measured environments. The path losses are calculated by averaging the transfer functions over the frequency band as a function of distance, as [25]

−= =

1601

1

2i10 ),(

16011

log10)(i

fdHdPL , (26)

where H(fi) is the channel transfer function.

Averaging over frequencies can be explained by the fact that path loss is relatively

insensitive to frequency, as depicted in Figure 28, where the red colour corresponds to higher path loss and blue corresponds to weaker path loss. In addition, the focus was to investigate the path loss relative to the total received power over the measured band. Figure 29, Figure 30 and Figure 31 depict the path losses in the different meas-urement cases. The path loss exponent is calculated from the slope of the linear re-gression line, which is shown in the path loss figures.

Figure 28. Example of path loss as a function of frequency and distance. The path losses presented in Figure 30 and Figure 31 prove that an indoor UWB

LOS radio channel can have a path loss exponent below the case of free space loss, i.e., two. This can be explained by the fact that an UWB indoor radio channel is very rich in signals reflected from the walls. The path loss exponents reported in the litera-ture [25, 35, 38–40, 42, 43] concerning an UWB channel modelling are presented in Table 6 where the campaign numbers depict the reference numbers listed in Chap-ter 8. The table show that the path loss exponent in LOS channel is typically slightly

46

below two and in NLOS channel between two and four. The figures below prove the same by showing the equal values.

5 6 7 8 9 1070

72

74

76

78

80

82

84

86

88

NLOS1

NLOS2

TX−RX Separation [m]

Pat

h Lo

ss [d

B]

Path Loss vs. TX−RX Separation

Path loss exponent = 2.03Path loss exponent = 1.45

Figure 29. Path loss in TS441 (NLOS1) and TS472 (NLOS2).

3 4 5 6 7 8 9 10 1530

40

50

60

70

80

90

TX−RX Separation [m]

Pat

h Lo

ss [d

B]

Path Loss vs. TX−RX Separation

Measured in L5Path loss exponent = 1.90Measured thru L5L6 wallPath loss exponent = 3.00

Figure 30. Path loss in L5 and through the two-sided brick wall between L5 and L6.

In Figure 29 the path loss exponent, particularly for NLOS2, is slightly suspicious.

The exponent in NLOS indoor cases is typically around three, as shown in Figure 30 (measured through the L5/L6 wall). This was also proved in [25]. As noticed in Figure 29, the amount of different TX-RX separations is low. This might be the rea-son for the low path loss exponents in those cases.

47

3 4 5 6 750

55

60

65

TX−RX Separation [m]

Pat

h Lo

ss [d

B]

Path Loss vs. TX−RX Separation

Measured in SÄ118Path loss exponent = 1.70Measured thru SÄ118 wallPath loss exponent = 2.16

Figure 31. Path loss in SÄ118 (LOS) and through a normal brick wall.

Table 6. Comparison of path loss exponents in different measurement campaigns

Measurement campaign [25] [35] [38] [39] [40] [42] [43] (LOS) 2.07 1.58 - 2 1.6 1.5 1.92 (NLOS) 2.95 1.96 2.1 - 3.7 2.5 3.67

3 4 5 6 7 8 9 10 1515

20

25

30

35

40

45

50

55

60

TX−RX Separation [m]

Pat

h Lo

ss [d

B]

Path Loss vs. TX−RX SeparationFull bandPath loss exponent = 3.01Lower sub−bandPath loss exponent = 3.04Median sub−bandPath loss exponent = 2.92Higher sub−bandPath loss exponent = 4.10

Figure 32. Excess path loss of the different sub-bands in the case of L5/L6 wall.

Figure 32 show the excess path loss of the different sub-bands. It evidently proves,

as expected, smaller path loss for the lower sub-band and vice versa. It is shown in Figure 33 that the entire path loss is larger than free space loss since

the signal is propagated through a wall. It is also seen that the entire path loss follows the outline of the free space loss plus some additional attenuation.

48

Figure 33. Path loss in the case of the L5/L6 wall (red at the top), free space loss (in the middle), free space loss +35 dB (black) and the excess loss (blue at the lowest).

The excess loss could be modelled separately utilising, e.g., a COST231 (Coopera-

tion in Scientific and Technical Research) multi-wall model [54]. However, it is not sufficient for UWB because of the possibility of adapting the free space loss model.

5.3.2. Multipath Model

The multipath model is obtained by investigating the multipath propagated signals in the PDP. The basic tapped-delay-line is extracted from the PDP, giving relative power values to the taps with a given delay. The number of taps is directly propor-tional to the complexity of the model. The UWB signal results in a PDP with very high accuracy, and therefore the number of paths in the model should be a large value. This is one reason why the tapped-delay-line model was not generated in this thesis project. Another and more reasonable motive is that the measured PDPs have distinct clusters. The proposed model for the channel having the cluster phenomenon is a modified IEEE 802.15.3a (SV [55]) model defined in [39]. The model presented in [55] is modified in order to fit the measured UWB channel data to the model. As presented in section 5.2, the amplitude seems to be lognormally distributed rather than Rayleigh distributed. In addition, each cluster and the rays inside the cluster are assumed to have independent fading. Figure 34 shows the idea of the modified IEEE 802.15.3a channel model. It is a compound from [39] and [55].

49

Delay

Amplitude

Overallenvelope

Clusterenvelope

Cluster 0

T1T0

e -T/

e -/

Cluster 1

Arrivals

. . .

Figure 34. Modified IEEE 802.15.3a channel model: An illustration of exponential decay of mean cluster power within clusters.

The following theory for the modified IEEE 802.15.3a model was originally pre-sented in [39]. The model consists of the discrete time impulse response defined as

−

=

−

=

−−=1

0

1

0

c r

)(L

l

K

k

ik,l

il

ik,lii )T(tXth , (27)

where

ik,l is the gain coefficient of the kth multipath of the lth cluster related to

the ith channel realisation, i

lT is the arrival time if the first path of the lth cluster related to the ith

channel realisation in nanoseconds, i

k,l is the delay of the kth multipath component relative to the lth cluster

arrival time ilT ,

iX is a lognormal shadowing term, Lc is the number of clusters and Kr is the number of paths (rays) inside the cluster. In the model, for the first cluster, T0 = 0 and for the first ray within the lth cluster,

00, =l . The number of channel realisations is freely selectable that must be large

enough in order to fulfil the statistical requirements. According to the model, lT and

k,l are described by the independent interarrival exponential probability density functions, as the Poisson processes as

50

( ) ( )[ ] 0exp 11 >−−= −− l,TT|TTp llll (28)

( ) ( )[ ] 0exp 1rr1 >−−= −− k,|p )l(kk,l)l(kk,l , (29)

where Λ is the cluster arrival rate in 1/ns and λr is the ray arrival rate in 1/ns, i.e., the arrival rate of the path within

each cluster. The Poisson process is one of the most commonly used in the queuing theory to

model, e.g., an arrival process of calls. The Poisson process is a good model when the population is large and, the arrivals are independent. Thus the process describes well the cluster and path arrival distributions.

The channel coefficients are defined as

, k,llk,lk,l p= (30) where

k,lp is a variable which takes into account the signal inversion due to re-flections and is either +1 or –1,

l reflects fading associated with the lth cluster and

k,l corresponds to fading associated with the kth ray of the lth cluster.

r//0

2 k,ll eeE T

k,ll−−=

, (31)

where E[·] is defined as the expectation value, 0Ω is the mean energy of the first path of the first cluster,

Γ is cluster decay factor in nanoseconds and γr is ray decay factor in nanoseconds.

The cluster and the ray decay factors describe the slopes of the attenuation of the

clusters and the rays, respectively, as shown in Figure 34. The values depend, e.g., on the reflective properties of the walls and on the room sizes [55].

Since the lognormal shadowing of the total multipath energy is captured by the term iX , the total energy contained in the terms i

k,l is normalised to unity for each realisation, as mentioned earlier. This shadowing term is characterised as

)N()(X i

2x10 ,0log20 ∝ , (32)

where x is standard deviation of the lognormal shadowing term for total multipath realisation in dB.

51

Two additional parameters describe the model:

1 is standard deviation of the cluster lognormal fading term in dB and

2 is standard deviation of the ray lognormal fading term in dB.

All the model parameters can be found from the measured data by matching the main characteristics of the channel. Mean excess delay, RMS delay spread and the number of multipath components within 10 dB of the peak (NP10 dB) are reasonable characteristics, which can be used to determine the model parameters [39]. The char-acteristics and the parameters of the model in [39] are listed in Table 7. In addition, the number of paths capturing 85 % of the energy (NP85 %) and the total channel en-ergy were calculated from each of the channel realisations. The 85 % threshold was chosen to minimise the modelling error over the entire data set [39]. The channel models CM 1–CM 4 are:

CM 1: LOS model from 0 to 4 m, CM 2: NLOS model from 0 to 4 m, CM 3: NLOS model from 4 to 10 m and CM 4: NLOS model from 4 to 10 m (represents an extreme NLOS).

Table 7. Modified IEEE 802.15.3a model parameters and characteristics Model Parameters CM 1 CM 2 CM 3 CM 4 Λ [1/ns] 0.0233 0.4 0.0667 0.0667 λ [1/ns] 2.5 0.5 2.1 2.1 Γ [ns] 7.1 5.5 14.00 24.00 γr [ns] 4.3 6.7 7.9 12 1 [dB] 3.3941 3.3941 3.3941 3.3941 2 [dB] 3.3941 3.3941 3.3941 3.3941 x [dB] 3 3 3 3 Model Characteristics Mean excess delay, m [ns] 5.0 9.9 15.9 30.1 RMS delay, RMS [ns] 5 8 15 25 NP10 dB 12 15 24 41 NP85 % 20 33 64 123 Channel energy mean [dB] -0.4 -0.5 0.0 0.3 Channel energy std [dB] 2.9 3.1 3.1 2.7

It must be noted that a complex tap model was not adopted here, but the result is

real valued. Thereby, the phase information is not included in the channel model.

52

5.4. Resulting UWB Channel Model

5.4.1. Modified IEEE 802.15.3a Channel Model

The parameters for the modified IEEE 802.15.3a channel model were found by searching reasonable values for the , , , r, 1, 2 and x which fit to the measured data. One cycle in the process is the following:

1. Give the initial values for , , , r, 1, 2 and x.

2. Determine the cluster arrivals according to (28). 3. Determine the ray arrivals for each cluster according to (29). 4. Calculate the amplitudes for each of the rays according to (30). 5. Impose the lognormal shadowing term on utilising (32). 6. Construct the impulse response realisations according to (27). 7. Calculate the mean excess delay m, RMS delay spread RMS and the number of

paths within 10 dB of the peak NP10 dB from the channel realisations. 8. Compare m, RMS and NP10 dB with the values which are extracted from the

measurement data. 9. Change some of the initial parameter values and return to step two.

The iteration process lasts until the results match with the measurement data. Table 8 lists the final Modified IEEE 802.15.3a -model parameters, which are based on the measurements in the LOS, NLOS1 and NLOS2 cases.

Table 8. Modified IEEE 802.15.3a model parameters and characteristics

Model Parameters LOS NLOS1 NLOS2 [1/ns] 0.05 0.1 0.05 [1/ns] 16 6 9 [ns] 16 19 24 r [ns] 1.03 2 5 1, 2 [dB] 3.4 3.4 3.4 x [dB] 2 2 2 Model Characteristics m [ns] 8.8 15.0 18.9 RMS [ns] 14 18 21 NP10 dB 9 16 27 NP85 % 15 39 65 Channel energy mean [dB] -0.6 -0.5 0.1 Channel energy std [dB] 2.1 2.3 2.4

Using the parameters from the table above, one hundred channel realisations were

constructed with the simulator, which was attached in [39]. The simulator works as it was presented at the beginning of section 5.4.1. The following figures illustrate the channel realisations in the LOS, NLOS1 and NLOS2 cases. The figures clearly show

53

the clustering phenomenon. In Figures 35, 37 and 39, different colours depict differ-ent clusters. Figures 36, 38 and 40 show that the LOS case is much clearer and less chaotic than the NLOS1 and NLOS2 cases.

Figure 35. One hundred impulse response realisations in LOS.

Figure 36. One hundred impulse response realisations in LOS (absolute values).

54

Figure 37. One hundred impulse response realisations in NLOS1.

Figure 38. One hundred impulse response realisations in NLOS1 (absolute values).

55

Figure 39. One hundred impulse response realisations in NLOS2.

Figure 40. One hundred impulse response realisations in NLOS2 (absolute values).

56

5.4.2. RMS Delay Spread and Mean Excess Delay

RMS delay spread is a time domain parameter which is typically used to give an idea of the channel’s character. It is calculated from the PDP, as presented in (13). All the time domain parameters are obtained from the PDPs by taking into account the thresholds presented in Figure 41. The initial delay is removed first. The most accu-rate way to remove the initial delay is to compare the exact distance measured with a Disto laser meter and the position of the corresponding path from the measured data. This method, which gives the optimal value for initial delay, was presented in section 5.1.2. The average noise level is typically around –60 dB above the maximum multi-path component in the normalised PDP, but it is estimated separately for all the cases by averaging the measured data before the first multipath component arrives [24]. RMS delay spread and mean excess delay are then calculated from the data, which is 15 dB above the noise level. A dynamic range of approximately 45 dB is then ob-tained for the final channel modelling.

RMS delay seems to be typically between 14 ns and 21 ns in indoor environments. A LOS channel provides 14 ns as an average, NLOS1 provides 18 ns and NLOS2, which can be referred to as an extreme NLOS, provides 21 ns as an average value. The values for indoors, as presented in the model in [39], are 5.28 ns, 14.28 ns and 25 ns for the same type of channels, respectively. The values in [39] are proposed for the same distances that were measured in this thesis project, but their environmental parameters differ.

Figure 41. Typical power delay profile in an indoor LOS channel. As depicted in Figure 42, variance of the RMS delay spread is quite large and it

varies in different channels. The difference between the measured and simulated re-sult can be explained by environmental properties. In Figure 42a, one room was in-cluded in the investigation and in Figure 42b, many rooms were considered.

57

0 50 100 150 200 25013

14

15

16

17

18

19

20

21

RMS Delay Spread in NLOS1

RM

S D

elay

Spr

ead

[ns]

Channel Number0 50 100 150 200 250

5

10

15

20

25

30

35

RMS Delay Spread in NLOS1

RM

S D

elay

Spr

ead

[ns]

Channel Number a) b)

Figure 42. RMS delay spread comparison through 235 channels in NLOS1: a) has been calculated from the measured PDP and b) depicts a simulation result.

The mean excess delay is 8.8 ns for LOS, 15 ns for NLOS1 and 18.9 ns for NLOS2. The values in the model in [39] are 5.0 ns, 14.18 ns and undefined, respectively. Pre-vious measurements provide in the order of 30 ns for the extreme NLOS case [39]. One reason for the difference between the measured value and the value in the model in [39] might be the way of positioning the time of arrival of the first path in the PDP. If the positioning is based only on the choice of the first path after the noise level is cut, the mean excess delay provides a bigger value. The real position of the arrival time of the first path in the PDP is later than the noise-cut position.

The delay spread parameters of the separate sub-bands are presented in Table 9, where the difference from the full band is evident.

Table 9. Comparison of delay spread values for separate sub-bands

LOS, L5 / L6 m [ns] RMS [ns] Lower sub-band (100 MHz) 10.9 13 Median sub-band (100 MHz) 9.9 15 Upper sub-band (100 MHz) 15.5 16 Full band (4.9 GHz) 8.4 14 NLOS1, TS 441 Lower sub-band (100 MHz) 12.4 18 Median sub-band (100 MHz) 13.3 17 Upper sub-band (100 MHz) 9.4 16 Full band (4.9 GHz) 10.0 16 NLOS2, TS 472 Lower sub-band (100 MHz) 29.8 21 Median sub-band (100 MHz) 26.7 26 Upper sub-band (100 MHz) 29.6 43 Full band (4.9 GHz) 15.3 22

58

The values in Table 9 can also be compared to the results of the previous wideband measurement campaigns. For example, in [50] with a 100 MHz bandwidth, the RMS delay spread is 33 ns and 48 ns for 90 and 60 MHz centre frequencies, respectively. With the lower centre frequency, signal attenuation is reduced, which makes the RMS delay spread larger. However, this is not the case with the UWB measurements and its sub-band observations, as depicted in Table 9.

5.4.3. Number of Paths

The number of paths within 10 dB of the peak counted in the PDP is a significant parameter when discussing the channel models. It has a direct relationship to the complexity of the channel simulator and the whole communication system planning. If the number of paths is a high value, system simulations become long. In addition, the devices in the communication system need much more hardware in order to be able to pick up all the desired multipath arrivals.

The number of paths within 10 dB of the maximum multipath component in the measured data is about 10 to 30 in the cases from LOS to NLOS2. The number of paths containing 85 % of the energy is from 15 to 65, including the environments from LOS to NLOS2, respectively. Both parameters were listed in Table 8.

59

6. DISCUSSION

The aim of this thesis project was to measure and model an indoor ultra wideband radio channel. Planning and carrying out the measurement campaign was interesting, since it allowed contributing many key aspects to make the measurements successful. Therefore, it was an important and vital part of this project. However, the largest and most difficult parts were the post-processing and analysis of the measured data. The measurement campaign succeeded well and the goals of generating radio channel models of the required environments were fulfilled.

One question that was brought up is the WSSUS assumptions for UWB communi-cations and channel modelling. How far do the WSSUS assumptions hold, has also been the question among a few other researchers in this sector. None of the UWB radio channels fulfils the assumptions absolutely, but in order to significantly sim-plify channel analysis, the assumptions are supposed to be valid [16].

When analysing the three 100 MHz sub-bands of the measured UWB channel, some interesting findings were attained. The amplitude fading distributions seem to vary in the different sub-bands, which can be seen in the pass rates of the Kolmo-gorov-Smirnov test. In the test, Rayleigh, Rice and lognormal distributions were con-sidered. According to the test, the most probable distribution for the UWB was the lognormal distribution, also in the sub-bands, even though the Rayleigh and Rice dis-tributions increased the percentage value in the sub-bands. The Rayleigh distribution is typically used to model the amplitude fading process in wideband NLOS channels and the Rice distribution in LOS channels [20]. One can ask if the amplitude distri-bution is different from the UWB. Of course, the distribution includes the effect of the operational environment, and thus the comparison must be done in similar cases.

One significant difference between wideband and UWB channel modelling is that in UWB modelling a very high delay accuracy of the received signal is achieved and thus a sharp channel impulse response is obtained. This is one of the advantages compared to narrowband and wideband channel modelling. UWB channels make it possible to generate very multipath-rich models that enable some extra improve-ments in the analysis, such as multipath separation accuracy.

It is important to consider what differences there are in the delay spread values of the separate sub-bands. Physically, signals with a lower centre frequency at the lower sub-band should have a larger RMS delay spread compared to the median or higher sub-bands. Signals in the lower sub-band have a larger wavelength, which makes them more resistant against the attenuation of the walls and furniture. One of the most interesting cases is the extreme non-line-of-sight (NLOS2) case where the RMS delay spread increases as the centre frequency increases. This finding diverges from the previous physical interpretation. Again, the phenomenon could be explained in such a way that a signal with a higher centre frequency reflects more sensitively from the walls inside the room and thus increases the RMS delay spread.

The UWB signal propagation phenomena seem to need much more and deeper in-vestigation, since some assumptions that have been proved in the narrowband and wideband signal analysis are no longer necessarily valid. One of these might be the US assumption, which says that the statistical characteristics of the radio channel are not dependent on the frequency. However, in carefully analysing the sub-bands a

60

slight frequency dependency was found. Therefore, it appears that some interesting findings may be found with UWB channel measurements that typical wideband sounders cannot detect from the same channel.

Path loss analysis is also an interesting subject of additional discussion. Generally, path loss is defined at the carrier frequency. That is why path loss should be defined as a function of frequency. In UWB transmission, it is calculated by averaging over the frequencies in order to obtain the path loss from the overall energy captured by the given link. The averaging is also supported by the fact that path loss is quite in-sensitive to the frequency in the indoor UWB case. The effects of the antennas are included in the resulting path loss. In addition, the effects of the antennas distort the final channel model results. This is one source of error in the channel model analysis, if only the ringing effect of the antennas is similar to the effect of the multipath propagation. If a real UWB system worked with the same antennas as in the meas-urements, there would not be any problem since all the impacts are taken into ac-count in the channel model. In addition to the antenna effect, the cables add some extra source of error when they are turned and moved from place to another. The er-ror cause by the cables is, however, insignificant.

If the path loss analysis was chosen to be large and more comprehensive, the amount of measurements should be much larger in order to get reliable results. Then the analysis would be extended, for example, to excess loss and compared to the dif-ferent types of indoor models such as the COST231 Multi-wall model [54].

The resulting channel model is different from the models that were constructed in the previous channel measurement and modelling campaigns. This model describes best the environment where the measurements were carried out, but it is possible to adapt it to similar types of environments. In order to construct a model for the “typi-cal” office environment, more comprehensive measurements should be performed in different types of office rooms. Care must be taken in applying the model outside the measurement environment, which makes radio channel measurement as an everlast-ing project.

The value of the measurement results and the constructed channel model is quite considerable, since the model can be utilised in a channel simulator. A simulator makes it possible to analyse and observe the performance of the UWB system in an indoor channel. The previous measurement campaigns provided channel models only for their “typical” environments, and therefore the modelling of our channels is vital. Because of the large number of sweeps, the statistical reliability of the measurements is quite high.

When considering further development, some things would be done a bit differ-ently. The sub-band analysis that was briefly analysed in this thesis project would be the focus in the future. The measurements would be done separately for all the sub-bands, which would retain the delay resolution in the power delay profile. In addi-tion, 500 MHz sub-bands, which is the case in a multiband UWB technology, would be another topic of investigation. After all, the study of the UWB radio channel will continue for many years.

61

7. SUMMARY

In this thesis project some typical UWB indoor radio channels were measured and modelled. The objective was to construct a statistical channel model of the environ-ments in which UWB devices are expected to operate in the future. The focus was on the multipath model, where the IEEE 802.15.3a model was modified and applied. In addition, path loss was investigated and modelled.

Three different environments were studied: line-of-sight (LOS), non-line-of-sight (NLOS1) and extreme non-line-of-sight (NLOS2). Several radio links were measured in each NLOS environment, utilising a stepped track at the receiver end. One track location including 235 antenna positions was selected as the boundary of the small-scale and large-scale consideration. In the LOS case, the measured environment was a lecture hall where each antenna position was included in the small-scale considera-tion and all the positions together in the large-scale consideration. The effects of the LOS and NLOS situations were investigated separately and added to the final chan-nel model analysis.

The thesis begins with an overview of UWB technology. The principles of an UWB and its commercial and military applications were presented. It was pointed out that UWB technology might be an inevitable and unbeatable choice for the high data rate and short-range applications in the future.

A theoretical discussion of the UWB radio channel was presented. The characteri-sation of time-variant linear multipath channels, divided into the effects of multipath propagation, power delay profile, Doppler effect and path loss, was studied in detail. Noise and interference sources were also discussed.

Various measurement techniques employed for radio channel sounding were pre-sented. The focus was on the setup that was used in the measurements of this thesis project. The basic techniques of channel sounding include narrowband techniques such as the single tone sounder and the vector network analyser (VNA) based sounder. Wideband and UWB sounding techniques and the measurement campaigns were presented in more detail. The sounder used in this thesis project is based on the VNA and is therefore referred to as a narrowband sounder, although the idea was to measure an UWB radio channel. The measurement environments were also presented in detail.

The UWB channel model analysis was divided into four main sections. Data post-processing steps included different IFFT methods such as Hermitean and conjugate approaches, positioning of the shortest signal path and a discussion of the effect of windowing. The conjugate approach, which was found to improve signal processing, was utilised to transform the measured transfer function to the time domain. Position-ing of the shortest arrival signal in the power delay profile was carried out by using a method where the real laser-measured distance between the antennas and the signal path extracted from the data were compounded. This method gives the best position-ing accuracy of the first path. Windowing that was performed in the frequency do-main caused distortion in the spectrum. Underestimation of the delay spread and deg-radation of the positioning accuracy of the shortest path in the PDP was also found in the time domain observations. Small-scale and large-scale amplitude fading were in-vestigated from the standpoint of the fading distributions. Rayleigh, Rician and log-

62

normal distributions were tested to fit the data to the cumulative distributions using the Kolmogorov-Smirnov test. Evidently, the lognormal distribution fits the best in all the cases, with a significance level of 1 %.

The free space loss model seems to be a rather good outline of the path loss in the LOS indoor UWB channels. The path loss exponent increases slightly above two when the LOS changes into a typical NLOS. The most distinct difference compared to free space loss is the excess attenuation caused by the furniture, walls and other obstacles. The multipath channel model was obtained from the data by investigating the aver-age PDPs of different environments. A modified IEEE 802.15.3a channel model based on the Saleh-Valenzuela channel model was constructed. The signal compo-nents in the PDPs arrive in distinct clusters, making the IEEE 802.15.3a model a good candidate. A couple of key parameters, including the cluster and ray arrival rates, the cluster and ray decay factors and the standard deviations of the fading and shadowing terms, define the model. The cluster arrival rates were 0.05 GHz for the LOS channels, 0.1 GHz for the NLOS1 channels and 0.05 GHz for the NLOS2 chan-nels. The corresponding ray arrival rates were 16, 6 and 9 GHz, respectively. The cluster decay factors were 16, 19 and 24 ns and the ray decay factors were 1.03, 2 and 5 ns for the LOS, NLOS1 and NLOS2 channels, respectively. The parameters were recovered by iteration and by matching the channel characteristics to the meas-ured RMS delay spread, mean excess delay and the number of multipath components within 10 dB of the peak. The measured values for RMS delay spread were 14, 18 and 21 ns for LOS, NLOS1 and NLOS2, respectively. The corresponding values for the mean excess delays were 8.8, 15.0 and 18.9 ns. The numbers of paths within 10 dB of the peak in the power delay profile were 9, 16 and 27 for LOS, NLOS1 and NLOS2, respectively. The resulting channel model describes best the environment where the measurements were carried out, but it can be adapted to offices or residen-tial rooms with plasterboard wall materials and typical room sizes.

The main differences between the IEEE 802.15.3a channel model and the modified version generated in this thesis project are the values of RMS delay and mean excess delay spread in the LOS channels. The unmodified model proposes 5 ns for both RMS and mean excess delay as opposed to the modified model’s proposal of 14 and 8.8 ns, respectively. The margin is explained by the environmental properties.

It appears that with UWB channel measurements, some interesting findings can be found, which typical wideband sounders cannot detect in the same channel. One sig-nificant finding is that a very high accuracy of the received signal is achieved, which makes it possible to separate the distinct paths from the power delay profile.

The sub-band analysis that was briefly analysed in this thesis would be the focus in the future. A division into 500 MHz sub-bands would also be included, which is the case in multiband UWB technology.

63

8. REFERENCES

[1] Aiello G.R. & Rogerson G.D. (2003) Ultra-Wideband Wireless Systems. IEEE Microwave Magazine 4, p. 36–47.

[2] Federal Communications Commission (2002) Revision of Part 15 of the

Commission’s Rules Regarding Ultra-Wideband Transmission Systems, First Report and Order. ET Docket 98–153, FCC 02–48, 117 p.

[3] Presentation of the European UWB Cluster to the Radio Spectrum Committee

(12.4.2004) 7th Meeting of the Radio Spectrum Committee, Brussels, March 3, 2004, URL: http://europa.eu.int/information_society/topics/radio_spectrum/ docs/ppt/19.

[4] UWB Multi-Band Coalition (12.4.2004) URL: http://www.uwbmultiband.org. [5] Somayazulu V.S. (2002) Multiple Access Performance in UWB Systems Us-

ing Time Hopping vs. Direct Sequence Spreading. In: Wireless Communica-tions and Networking Conference, March 18–21, Orlando, USA, Vol. 2, No. 3, p. 522–525.

[6] Time Domain Corporation (1998) Comments of Time Domain Corporation,

In the Matter of Revision of Part 15 of the FCC’s Rules Regarding Ultra Wideband Transmission Systems. ET Docket 98–153. 46 p. + 7 app.

[7] Time Domain Corporation (22.1.2004) PulsON® Technology Overview.

URL: http://www.timedomain.com/Files/downloads/techpapers/PulsONOver view7_01.pdf.

[8] Hämäläinen M., Hovinen V., Tesi R., Iinatti J. & Latva-aho M. (2002) On the

UWB System Coexistence with GSM900, UMTS/WCDMA, and GPS. IEEE Journal on Selected Areas in Communications 20, p. 1712–1721.

[9] XtremeSpectrum Inc. (6.4.2004) TRINITY for Media-Rich Wireless Applica-

tions, Product Brief. URL: http://www.xtremespectrum.com/PDF/xsi_trinity _brief.pdf.

[10] Multispectral Solutions Inc. (6.4.2004) Recent Applications of Ultra Wide-

band Radar and Communications Systems. URL: http://www.multispectral. com/pdf/AppsVGs.pdf.

[11] Daniele N. (2003) Introduction to UWB Impulse Radio. In: Ultra Wide Band

Summit 2003, Dec. 2–5, Paris, France, PowerPoint presentation, 59 p.

64

[12] Hentilä L. & Oppermann I. (2002) Wireless Location Companies and Patents. Center for Wireless Communications (CWC), Internal document, University of Oulu, 69 p.

[13] Talvitie J. (1997) Wideband Radio Channel Measurement, Characterisation

and Modelling for Wireless Local Loop Applications. Acta Universitatis Ou-luensis, Series C, Technica 99. Oulun yliopisto, Oulu, 94 p.

[14] Bello P.A. (1963) Characterization of Randomly Time-Variant Linear Chan-

nels. IEEE Transactions on Communication Systems 11, p. 360–393. [15] Parsons J.D. (1992) The Mobile Radio Propagation Channel. Pentech Press,

London, 316 p. [16] Parsons J.D., Demery D.A. & Turkmani A.M.D. (1991) Sounding Techniques

for Wideband Mobile Radio Channels: A Review. IEE Proceedings I 138, p. 437–446.

[17] Saunders S.R. (1999) Antennas and Propagation for Wireless Communication

Systems. John Wiley & Sons Ltd., New York, USA, 409 p. [18] Abouraddy A.F. & Elnoubi S.M. (2000) Statistical Modeling of the Indoor

Radio Channel at 10 GHz Through Propagation Measurements - Part I: Nar-row-Band Measurements and Modeling. IEEE Transactions on Vehicular Technology 49, p. 1491–1507.

[19] Christ R. & Lavigne R. (2000) Radio Frequency-Based Personnel Location

Systems. In: IEEE 34th International Carnahan Conference, Oct. 23–25, Ot-tawa, Canada, p. 141–150.

[20] Hashemi H. (1993) The Indoor Radio Propagation Channel. Proceedings of

the IEEE 81, p. 943–968. [21] Proakis J.G. (2001) Digital Communications. Fouth Edition. McGraw-Hill

Inc., Boston, 1002 p. [22] Hämäläinen M. (2002) Kaksitieradiokanavan vaikutus ilmailun hajaspektri-

tiedonsiirtoon VHF- ja UHF-alueilla. Lisensiaatintyö. Oulun yliopisto, sähkö-tekniikan osasto, Oulu, 78 p.

[23] Rappaport T.S. (1996) Wireless Communications: Principles and Practice.

Prentice Hall, New Jersey, 641 p. [24] Cassioli D., Win M.Z. & Molisch A.F. (2002) The Ultra-Wide Bandwidth In-

door Channel: From Statistical Model to Simulations. IEEE Journal on Se-lected Areas in Communications 20, p. 1247–1257.

65

[25] Ghassemzadeh S.S., Greenstein L.J., Kavi A., Sveinsson T. & Tarokh V. (2003) An Empirical Indoor Path Loss Model for Ultra-Wideband Channels. KICS Journal of Communications and Networks 5, p. 303–308.

[26] Hewitt A. & Vilar E. (1988) Selective Fading on LOS Microwave Links:

Classical and Spread-Spectrum Measurement Techniques. IEEE Transactions on Communications 36, p. 789–796.

[27] Agilent Technologies Inc. (2000) Agilent AN 1287-2, Exploring the Architec-

tures of Network Analyzers, Application Note. USA, 12 p. [28] Hämäläinen M., Hentilä L., Pihlaja J. & Nissinaho P. (2003) Modified Fre-

quency Domain Radio Channel Measurement System for Ultra Wideband Studies. In: The Fourth Finnish Wireless Communications Workshop, Oct. 16–17, Oulu, Finland, p. 132–135.

[29] Cox D.C. (1972) Delay Doppler Characteristics of Multipath Propagation at

910 MHz in a Suburban Mobile Radio Environment. IEEE Transactions on Antennas and Propagation 20, p. 625–635.

[30] Talvitie J. & Poutanen T. (1994) Self-Noise as a Factor Limiting The Dy-

namic Range in Impulse Response Measurements Using Sliding Correlation. In: IEEE Third International Symposium on Spread Spectrum Techniques and Applications, July 4–6, Oulu, Finland, Vol. 2, p. 619–623.

[31] Hämäläinen M., Nykänen A., Hovinen V. & Leppänen P. (1997) Digital

Stepping Correlator in a Wideband Radio Channel Measurement System. In: IEEE 8th International Symposium on Personal, Indoor and Mobile Commu-nications, Sept. 1–4, Helsinki, Finland, p. 1120–1124.

[32] Win M.Z., Scholtz R.A. & Barnes M.A. (1997) Ultra-Wide Bandwidth Signal

Propagation for Indoor Wireless Communications. In: IEEE International Conference on Communications, June 8–12, Montreal, Canada, Vol. 1, p. 56–60.

[33] Hovinen V., Hämäläinen M. & Pätsi T. (2002) Ultra Wideband Indoor Radio

Channel Models: Preliminary Results. In: IEEE Conference on Ultra Wide-band Systems and Technologies, May 21–23, Baltimore, USA, p. 75–79.

[34] Keignart J. & Daniele N. (2002) Subnanosecond UWB Channel Sounding in

Frequency and Temporal Domain. In: IEEE Conference on Ultra Wideband Systems and Technologies, May 21–23, Baltimore, USA, p. 25–30.

[35] Kunisch J. & Pamp J. (2002) Measurement Results and Modeling Aspects for

the UWB Radio Channel. In: IEEE Conference on Ultra Wideband Systems and Technologies, May 21–23, Baltimore, USA, p. 19–23.

66

[36] Ghassemzadeh S.S., Jana R., Rice C.W., Turin W. & Tarokh, V. (2002) A Statistical Path Loss Model for In-Home UWB Channels. In: IEEE Confer-ence on Ultra Wideband Systems and Technologies, May 21–23, Baltimore, USA, p. 59–64.

[37] Prettie C., Cheung D., Rusch L. & Ho M. (2002) Spatial Correlation of UWB

Signals in a Home Environment. In: IEEE Conference on Ultra Wideband Systems and Technologies, May 21–23, Baltimore, USA, p. 65–69.

[38] Yano S.M. (2002) Investigating the Ultra-Wideband Indoor Wireless Chan-

nel. In: IEEE 55th Vehicular Technology Conference, May 6–9, Birmingham, USA, Vol. 3, p. 1200–1204.

[39] Foerster J. (2003) Channel Modeling Sub-committee Report Final. Submitted

to IEEE P802.15 Working Group for Wireless Personal Area Networks (WPANs), IEEE P802.15-02/490r1-SG3a, Feb. 7, 52 p.

[40] Keignart J. & Daniele N. (2003) Channel Sounding and Modelling for Indoor

UWB Communications. In: The First International Workshop on Ultra Wide-band Systems, June 2–5, Oulu, Finland, Proceedings on CD, 5 p.

[41] Kovács, I. & Eggers P.C.F. (2003) Short-Range UWB Radio Propagation In-

vestigations Using Small Terminal Antennas. In: The First International Workshop on Ultra Wideband Systems, June 2–5, Oulu, Finland, Proceedings on CD, 5 p.

[42] Pagani P., Pajusco P. & Voinot S. (2003) A Study of the Ultra-Wide Band

Indoor Channel: Propagation Experiment and Measurement Results. In: The First International Workshop on Ultra Wideband Systems, June 2–5, Oulu, Finland, Proceedings on CD, 5 p.

[43] Cassioli D. & Durantini A. (2004) A Time-Domain Propagation Model of the

UWB Indoor Channel in the FCC-Compliant Band 3.6 – 6 GHz based on PN-Sequence Channel Measurements. In: IEEE Vehicular Technology Confer-ence, May 17–19, Milan, Italy, 5 p.

[44] Responses to IEEE 802.15.3a Call for Contributions on Ultra-wideband chan-

nel models (7.4.2004). URL: http://grouper.ieee.org/groups/802/15/pub/2002/ Jul02.

[45] Agilent Technologies Inc. (2000) Reference Guide 8719ET / 20ET / 22ET

8719ES / 20ES / 22ES Network Analyzers. USA, 289 p. [46] Agilent Technologies Inc. (2000) 83000A Series Microwave System Amplifi-

ers, Product Overview. USA, 15 p.

67

[47] Antenna Research Associates Inc. (1994). Technical Manual for Conical Monopole Antenna CMA-118/A 1.0-18.0 GHz. Beltsville, USA, 29 p.

[48] Leica Geosystems (2001) Disto Pro / Pro a, Käyttäjän käsikirja, Versio 1.0,

Suomi. 60 p. [49] Witrisal K., Yong-Ho K. & Prasad R. (1998) Frequency Domain Simulation

and Analysis of the Frequency-Selective Ricean Fading Radio Channel. In: The Ninth IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, Sept. 8–11, Boston, USA, Vol. 3, No. 3, p. 1131–1135.

[50] Krishnamurthy P., Beneat J., Marku M. & Pahlavan K. (1999) Modeling of

the Wideband Indoor Radio Channel for Geolocation Applications in Resi-dential Areas. In: IEEE 49th Vehicular Technology Conference, Houston, USA, Vol. 1, p. 175–179.

[51] Conversation with Veikko Hovinen, Telecommunication Laboratory, Univer-

sity of Oulu, Finland, May 2003. [52] Velera G., Lobeira M., Torres R.P. & Garcia J.L. (2002) UWB Channel

Model Contribution from University of Cantabria and ACORDE. In: The 21st Meeting of the IEEE 802.15 SG3a, Nov. 11–15, Koloa, USA.

[53] Vaughan R. & Andersen J.B. (2003) Channels, Propagation and Antennas for

Mobile Communications. The IEE Electromagnetic Waves Series 50, IEE, London, UK, 753 p.

[54] European Commission / COST Telecommunications (1999) Digital Mobile

Radio Towards Future Generation Systems, COST231 final report, Brussels, Belgium, 474 p.

[55] Saleh A.A.M. & Valenzuela R.A. (1987) A Statistical Model for Indoor Mul-

tipath Propagation. IEEE Journal on Selected Areas of Communications 5, p. 128–137.

68

9. APPENDICES

Appendix 1 Frequency and impulse responses of Agilent 83017A amplifier

Appendix 2 Construction and close-up figures of the ARA antenna

Appendix 3 Radiation pattern of the ARA antenna (4.0 GHz, azimuth)

Appendix 4 Radiation pattern of the ARA antenna (6.0 GHz, azimuth)

Appendix 5 Radiation pattern of the ARA antenna (8.0 GHz, azimuth)

Appendix 6 Radiation pattern of the ARA antenna (4.0 GHz, elevation)

Appendix 7 Radiation pattern of the ARA antenna (6.0 GHz, elevation)

Appendix 8 Radiation pattern of the ARA antenna (8.0 GHz, elevation)

Appendix 9 Frequency responses of the TX and RX antenna cables

Appendix 10 Construction blueprint of the measurements in SÄ118

Appendix 11 Construction blueprint of the measurements in L5 / L6

Appendix 1 Frequency and impulse responses of Agilent 83017A amplifier 69

3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 835.2

35.4

35.6

35.8

36

36.2

36.4

36.6

36.8

37

Gai

n [d

B]

Frequency [GHz]

Agilent 83017A amplifier; frequency response

Average = 35.87 dB

0 1 2 3 4 5 6 7 8

15

20

25

30

35

[dB

]

Propagation Delay [ns]

Agilent 83017A Amplifier; Impulse Response

Delay = 0.6 ns

Appendix 2 Construction and close-up figures of the ARA antenna 70

Appendix 3 Radiation pattern of the ARA antenna (4.0 GHz, azimuth) 71

Appendix 4 Radiation pattern of the ARA antenna (6.0 GHz, azimuth) 72

Appendix 5 Radiation pattern of the ARA antenna (8.0 GHz, azimuth) 73

Appendix 6 Radiation pattern of the ARA antenna (4.0 GHz, elevation) 74

Appendix 7 Radiation pattern of the ARA antenna (6.0 GHz, elevation) 75

Appendix 8 Radiation pattern of the ARA antenna (8.0 GHz, elevation) 76

Appendix 9 Frequency responses of the TX and RX antenna cables 77

Appendix 10 Construction blueprint of the measurements in SÄ118 78

Door

steelcolumn8,5 cm width

SÄ118 - through the wall m easurem ent - 021018RX in corridorNum ber of sweeps 32Freq range 1.99 - 6.99 G HzTX height 1,10 mRX height 1 ,10 m

1,47 m

5,94 m

,64

m

,45 m

10,34 m

3,68

m

Solid brick wallthickness 0,13 m

concretepillar

Solid brick w

allSol

id b

rick

wal

lw

ith c

halk

boar

d

8,08 m

2,65

m

1,00

m

Door

steelcolumn8,5 cm width

SÄ118 - through the wall m easurem ent - 021018RX in corridorNum ber of sweeps 32Freq range 1.99 - 6 .99 G HzTX height 1,10 mRX height 1 ,10 m

1,47 m

5,94 m

,64

m

,45 m

10,34 m

3,68

m

1,00

m3,

65 m

Solid brick wallthickness 0,13 m

concretepillar

Solid brick w

allSol

id b

rick

wal

lw

ith c

halk

boar

d

8,08 m

Appendix 11 Construction blueprint of the measurements in L5 / L6 79

L5 - through the wall measurement - 021111RX and TX in the lecture hall L5Number of sweeps 32Freq range 1.99 - 6.99 GHzTX height 1,10 mRX height 1,10 m.

Door

Solid double brickthickness 0,21 m

15,66 m

,45 m

5,70 m5,57 m

1,83 m

1,00

m

15,07 m

3,00

m

hollowmetalwall

concretepillar

Solid brick w

allthickness 0,13 m

L5 - through the wall measurement - 021112RX in L6 and TX in L5Number of sweeps 32Freq range 1.99 - 6.99 GHzTX height 1,10 mRX height 1,10 m.

Door

Solid double brickthickness 0,21 m

15,66 m

5,57 m

1,83 m

1,00

m

13,23 m

1,79

m

hollowmetalwall

concretepillar

Solid brick w

allthickness 0,13 m

5,70 m

,45 m