Estimated antenna gains at 0.1 m separation for different bandwidths.
1. Introduction
The subject of this chapter is to investigate the effect of bandwidth on a short range indoor UWB channel performance. This research is based on a measurement campaign performed on a wooden desk surface placed in an office room at our faculty building. A vector network analyzer (VNA) is used to sweep the 1 GHz, 2 GHz, 5 GHz and 7.5 GHz bandwidths centered at 6.85 GHz, obtaining the frequency response of the channel. Using the VNA time domain capability the channel impulse response is obtained. This is equivalent to sounding the channel with frequency chirp pulses equal in duration to the inverse of the frequency bandwidth. In other words, a narrower bandwidth results in a wider pulse in time domain.
Although the measurements are performed in same environment for all bandwidths, it is expected that an UWB channel itself would not be equally perceived by different pulse widths. A wider system bandwidth results in shorter pulses, which in turn account for finer temporal and spatial resolution. In this way more multipath components can be resolved as the pulses overlap in a lesser extent.
The fundamental differences between an UWB channel and a narrowband channel arise from the frequency selectivity of the propagation process (Molisch, 2005). As the UWB signal has a wide frequency spectrum which may extend to several gigahertz, the frequency dependence of diffraction/reflection coefficients and dielectric constants can be significant (Di Benedetto et al., 2006). A number of papers report on the effect of carrier frequency on channel parameters. One such investigation (Cassioli et al., 2004) finds a strong dependence between the path loss model exponent and the carrier frequency, yet states that there is no correlation with the bandwidth. (Ghassemzadeh et al., 2005) presented an extensive measurement campaign at two different bandwidths (1.25 GHz and 6 GHz) centered at 5 GHz, reporting mostly minor differences in parameter values between the two bandwidths. Another paper (Choi et al., 2009) models the path loss exponent variation as a function of frequency. A research project (Chang & Tarng, 2007) investigates the effects of bandwidth on observable multipath clustering and Δ
The next section of the chapter explains the measurement procedure and equipment and describes the measurement environment. The subsequent sections present the path loss, shadowing, mean excess delay and RMS delay spread parameter values estimated from the measured power delay profiles, respectively. The obtained results are discussed in the seventh section. The chapter is concluded in the eighth section, briefly summarizing all the key findings of this research.
2. Experimental setup
The measurement campaign took place on an empty desk in one of the offices at our faculty building. The surface of the desk is cleared to ensure the line of sight (LOS) between the transmitter and the receiver. The desk used in measurements is made of wood, chipboard and MDF. Inner walls that separate offices and laboratories are mainly made of plasterboard constructed with a thin wire grid composition, while the external wall and floors are made of reinforced concrete. The floor is entirely covered with wooden parquetry. The doors are made of plywood, while the office furniture is made of various wooden materials and glass. In the vicinity of the desk on which the measurements are conducted is another desk with the measurement equipment and two cabinets in the bottom left corner of the office, as shown on the office plan on Fig. 1. This additional furniture in the office is introduced as an intentional scattering system to approximate a reallife scenario.
A 1.6 m x 0.7 m grid with 0.1 m spacing is drawn on a sheet of paper (Fig. 2) which is fixed on a surface of a plain office desk. This forms a matrix of 17 x 8 measurement points across the desktop. The transmitting antenna (presented by a red disk in Fig. 2) is placed on the top of the 9^{th} column in a fixed position. The receiving antenna (presented by a blue disk in Fig. 2) is moved along the remaining 135 points for each measurement. Both antennas are erected on styrofoam blocks at the same height of 4.5 cm with their azimuth planes being parallel to the desk surface (Θ = 0°).
2.1. Measurement setup
The measurement setup is presented on Fig. 3. The transmitting antenna (Tx) is connected to the port 1 of the HP 8720A vector network analyzer (VNA) by a semirigid coaxial cable. The receiving antenna (Rx) is connected to the port 2 of the VNA by a flexible coaxial cable. The attenuation of the cables is compensated by the calibration procedure. The measured data from the VNA is transferred by the HP interface bus (HPIB) to the data acquisition PC (PCDAQ), where the data is stored for further processing.
2.1.1. Vector network analyzer
A vector network analyzer (VNA) is used to measure the system’s forward complex transmission coefficients
where
Using an inverse Fourier transform ℑ^{1}, the measurement results can be observed in the time delay domain, presenting the power delay profile (PDP) which gives the time distribution of the received signal power from a transmitted pulse:
The measured data acquired from the VNA thus presents the frequency averaged power delay profiles.
The excess delay
By activating the VNA’s averaging function, the last 16 measurements are averaged to reduce the temporal variations in the system. This results in a time averaged power delay profile:
where
Prior to each measurement the VNA is calibrated using a specific calibration standard. The measured data is acquired via HPIB interface to a PC for further analysis and correction. Before the parameter extraction procedure each measured PDP is corrected with respect to the predefined threshold, which was set to 5 dB above the noise level for all measured points. This correction sets all the delay bins with energies below this threshold to zeroenergy bins.
2.1.2. Antennas
A pair of omnidirectional UWB antennas used in this measurement campaign was built according to the design proposed in (Taniguchi et al., 2006). To verify the omnidirectionality of the built antennas, their radiation patterns were measured in a TESEQ 750 GTEM cell at different frequencies generated by a HP 8340A synthesized sweeper. This method is based on measuring the power received by antenna under test (AUT) which is placed inside the cell. The received power was measured using a Rohde&Schwarz NRPZ21 universal power sensor. Electric field in the vicinity of the AUT is measured by an isotropic electric field probe HI4455 positioned next to the AUT. The antenna gain
where
In addition, the VNA was used to measure the VSWR throughout the widest frequency bandwidth of 7.5 GHz, centered at 6.85 GHz. A full port calibration was performed before each measurement. The results, presented on Fig. 5, show that the VSWR is less than 2.5 throughout the entire frequency span.
Finally, the power delay profiles for different bandwidths were measured at a reference distance
where
Applying the Friis equation with
Under the assumption that antenna gains are equal (
The results are shown in Table 1.
Bandwidth  1 GHz  2 GHz  5 GHz  7.5 GHz 
Antenna gain  0.475 dB  0.54 dB  0.035dB  0.26 dB 
The measurement results show that the antenna gain may be approximated to 0 dBi, implying a ±1 dB tolerance.
2.1.3. System resolution and dynamic range
After a preliminary observation of the power delay profiles on various measurement points at different bandwidths, it is noticed that the noise threshold is constant (slightly below 90 dB). Besides that, as the power in the profile decays with delay, there are no observable multipath components above the noise threshold approximately 60 ns after the time instant of transmitting the pulse for each bandwidth. According to this fact the display window from 0 ns to 80 ns is chosen for all considered bandwidths and measurement points. Since this window is sampled at 801 points, the display resolution for all measurements is 0.1 ns.
The temporal resolution of the measurement system is equal to the inverse of the bandwidth and has a value greater than the display resolution. Otherwise, the display window would exceed the maximum observable delay (Hovinen et al., 2002).
Once the VNA was calibrated, the pulse shape was observed it the time delay domain for each different bandwidth (Fig. 6). All pulses have a different peak value at
The VNA provides a windowing feature which is needed because of the abrupt transitions at the start and stop frequencies of the measured frequency span. The band limiting of a frequency domain response causes overshoot and ringing in a time domain response. This limits the usefulness of the time domain measurement in terms of the sidelobes and a pulse wider than the system resolution. The window used in the presented measurements is the normal window, which has a sidelobe level 44 dB below the pulse peak value and extends the pulse width to value of 3.84 times the system resolution for a bandpass mode (HewlettPackard Company, 1989). This has a direct consequence in spreading the pulse spatially, so that the spatial resolution of the system is degraded and the system can no longer resolve the multipath components with the precision expected from the system resolution. Another consequence is that when observing the delays in relation to the time instant of the peak of the transmitted pulse, one can observe a certain amount of energy received in negative delay times. To eliminate this, several power delay profiles where the total energy was not captured are removed from the final set of measured data. Hence, the minimum measurement distance is half of the spatial width of the pulse at a given bandwidth.
Furthermore, the RMS delay spread is estimated for each pulse, as this value is to be subtracted from the measured RMS delay spread of each power delay profile (Section 6). The results are presented in Table 2.
Bandwidth  1 GHz  2 GHz  5 GHz  7.5 GHz 
System resolution  1 ns  0.5 ns  0.2 ns  0.133 ns 
Display resolution  0.1 ns  0.1 ns  0.1 ns  0.1 ns 
Pulse width  3.84 ns  1.92 ns  0.768 ns  0.512 ns 
Spatial width  1.152 m  0.576 m  0.23 m  0.154 m 
Estimated pulse power  1.66 dB  1.66 dB  1.66 dB  1.66 dB 
Pulse RMS delay spread  567 ps  283.4 ps  113.4 ps  75.6 ps 
Fig. 7 shows a detail of normalized power profiles for different bandwidths. All the profiles are measured at 0.78 m antenna separation. It is evident that the measurement system with a narrow bandwidth cannot resolve as much multipath components as the same measurement system with wider bandwidth.
3. Path loss
The total path loss is defined as attenuation in the energy of pulse while propagating from the transmitting antenna to the receiving antenna. This can be expressed as a ratio of total energy of the transmitted pulse vs. total energy of the received power delay profile:
The path loss model (Ghassemzadeh at al., 2003) defines the path loss exponent
where
The total energy received at the Rx antenna is estimated by summing the energy in the entire corrected power delay profile. By subtracting the measured total path loss at a reference distance from this value, we estimate the relative path loss:
where
The relative path loss values estimated from the measured power delay profiles are presented on Fig. 8 at different system bandwidths as a function of antenna separation in a logarithmic scale. In this manner the linear regression lines can be fitted into the measurement data using the least square method. The calculated free space path loss line (7) is added for comparison. Note that the correlation coefficient increases while the path loss exponent slightly decreases with increasing bandwidth.
The path loss exponent values estimated in this measurement campaign are comparable to the values found in literature up to date.
A indoor LOS measurement campaign which took place inside an anechoic chamber on a 1.2 m x 0.8 m rectangular aluminum conductive plate simulating a heavyduty office desk (Suzuki & Kobayashi, 2005) found a path loss exponent value of
Short range indoor LOS measurements have also been performed in a 6 GHz to 8 GHz frequency band, with a transmitterreceiver distance ranging from 1 m to 5 m (Bose, 2006). A path loss analysis yields a path loss exponent of
A 3.6 GHz – 6 GHz measurements in 1 m – 11 m range (Cassioli & Durantini, 2004) report a path loss exponent of
An UWB indoor LOS channel measurement campaign was evaluated in (Cassioli et al, 2001), finding a path loss exponent of
An extensive research based on analysis of over 300000 power delay profiles measured in 4.375 GHz – 5.625 GHz in various indoor LOS scenarios presented in (Ghassemzadeh et al., 2002) finds the average path loss exponent to have a value of
Although the frequency dependence analysis of the channel parameters investigated in (Cassioli et al., 2004) resulted in a wide range of path loss exponent values between 0.6 and 2.2 for indoor LOS scenarios, there is no marked dependence on the system bandwidth.
4. Shadowing
The shadowing term
The shadowing term captures the path loss deviations from its median value.
Fig. 9 presents a comparison of empirical cumulative distribution functions (CDF) of the estimated shadowing term
In a 3.6 GHz – 6 GHz indoor LOS measurements (Cassioli & Durantini, 2004) shadowing is modeled as lognormal, with standard deviation of 1.42 dB for indoor LOS scenarios.
System BW  1 GHz  2 GHz  5 GHz  7.5 GHz  Free space (7) 

30.12 dB  30.24 dB  29.23 dB  28.63 dB  29.16 dB 

2.28  2.24  2.2  2.05  2 

1.87 dB  1.67 dB  1.57 dB  1.39 dB  0 dB 
An UWB indoor LOS channel measurement campaign was evaluated in (Cassioli et al, 2001), finding standard deviation of shadowing term of 5.9 dB.
An extensive research based on analysis of over 300000 power delay profiles measured in 4.375 GHz – 5.625 GHz for indoor LOS scenarios presented in (Ghassemzadeh et al., 2002) reports on the standard deviation of the shadowing term of 1.6 dB.
5. Mean excess delay
Mean excess delay is commonly used to describe the time dispersion characteristics of a transmission channel. It is defined as the first moment of the power delay profile:
The scatter plot of the mean excess delay values, estimated according to (13) from the measured power delay profiles, is presented on Fig. 10 in dependence on the antenna separation for different bandwidths.
Fig. 11 presents a scatter plot of the estimated mean excess delay in function of the estimated total path loss for different bandwidths. We find a significant correlation between these two parameters, which is slightly increasing with the bandwidth.
6. RMS delay spread
RMS delay spread is another parameter that characterizes the time dispersion of the channel. It is defined as a square root of the second central moment of the power delay profile:
As omitting the subtraction of the RMS delay spread of the calibrated pulse would yield overestimated results (Varela & Sánchez, 2001), in order to estimate the RMS delay spread it is necessary to subtract the RMS delay spread of the calibrated pulse from the RMS delay spread of each measured PDPs (Saleh & Valenzuela, 1987):
where
Fig. 13 presents a comparison of empirical cumulative distribution functions (CDF) of the estimated RMS delay spread values and theoretical normal CDFs with respective standard deviations for different bandwidths, shown in Table 4. Note that the mean value of the estimated RMS delay spread values decreases with increasing bandwidth.
System BW  1 GHz  2 GHz  5 GHz  7.5 GHz 
Mean  5.93 ns  4.85 ns  3.85 ns  3.9 ns 
Std. dev.  1.37 ns  1.37 ns  1.25 ns  1.36 ns 
An extensive research based on analysis of over 300000 power delay profiles measured in 4.375 GHz – 5.625 GHz presented in (Ghassemzadeh et al., 2002) reports on normally distributed mean excess delay and RMS delay spread with mean value of 4.7 ns and standard deviation of 2.3 ns for indoor LOS scenarios. This increase is explained to be largely due to paths with longer delays having larger path loss values associated with them.
Fig. 14 presents a scatter plot of estimated RMS delay spread in dependence of the total path loss. It is evident that these two parameters are strongly correlated.
A scatter plot of estimated RMS delay spread in function of the estimated mean excess delay is presented on Fig. 15. A strong correlation is observed for all bandwidths.
7. Discussion
The research presented in this chapter showed a certain dependence of the estimation of channel parameters on the system bandwidth.
First, the estimated path loss exponent slightly decreases with increasing the system bandwidth. An explanation can be found in the frequency selective propagation mechanisms, since different frequency components experience different impact of diffraction, reflection and transmission, where the coefficients that define these processes are dependent on the frequency of the transmitted signal. The estimated path loss exponent is slightly higher than the free space path loss exponent (between 2.05 and 2.28 for different bandwidths). Considering that the path loss exponent estimates for UWB indoor scenarios found in the literature are regularly below 2, the additional loss in the presented measurement campaign might be explained by the influence of the used desk, made of wood, chipboard and MDF.
Second, the estimated standard deviation of the shadowing term evidently decreases with increasing the system bandwidth, which in turn causes the increase in correlation between the estimated total path loss and antenna separation. This can also be explained as a direct consequence of the frequency selective propagation mechanisms. At larger bandwidths, more frequency components will be immune to reflections and diffraction, so that the received signal will be more consistent.
Third, the estimated RMS delay spread decreases with increasing the system bandwidth, because the system with wider spectrum is able to resolve more multipath components.
It is shown that estimated correlations between the various channel parameters investigated in this research are greater as the system bandwidth increases. This is obviously another consequence of better temporal resolution in due to shorter pulses at wider bandwidths.
8. Conclusion
The main scope of this chapter was to explore the validity of well established empirical methods used for UWB channel modeling and to propose the estimated values of the model parameters for a specific propagation environment.
An indoor UWB measurement campaign presented in this chapter was performed on an empty surface of a desk placed in a typical office room. The channel impulse responses were measured on the surface of the desk at different system bandwidths using a VNA method. The channel parameter values were estimated from the measurement results for each different bandwidth. The estimated values are comparable with the ones found in literature.
Analyzing the estimated values we have found that the estimations of the channel parameters show a certain dependence on the system bandwidth. The estimated path loss exponent values, standard deviations of the shadowing term, RMS delay spread mean values and mean excess delay values are all decreasing with increasing bandwidth for this indoor LOS scenario.
The analysis of correlation between the estimated values showed that the correlation coefficients tend to be higher at wider bandwidths.
Acknowledgments
The authors wish to thank Prof. A. Šarolić and Mr. Z. Živković for use of their GTEM cell and assistance in measuring the antennas’ radiation patterns.
References
 1.
Bose R. 2006 Ultra Wideband Channel Modeling for Personal Area Network,  2.
Cassioli D. Durantini A. 2004 A timedomain Propagation Model of the UWB Indoor Channel in the FCCcompliant Band 3.6 6 GHz based on PNsequence Channel Measurements,  3.
Cassioli D. Durantini A. Ciccognani W. 2004 The Role of Path Loss on the Selection of the Operating Bands of UWB Systems,  4.
Cassioli D. Win M. Z. Molisch A. F. 2001 A Statistical Model for the UWB Indoor Channel,  5.
Chang, W.J. & Tarng, J.H. ( 2007 ). Effects of Bandwidth on Observable Multipath Clustering in Outdoor/Indoor Environments for Broadband and Ultrawideband Wireless Systems, IEEE Transactions on Vehicular Technology,00189545 4 56 1913 1923  6.
Choi J. Kang N. G. Sung Y. S. Kang J. S. Kim S. C. 2009 FrequencyDependent UWB Channel Characteristics in Office Environments,  7.
Ghassemzadeh, S.S., Greenstein, L.J., Kavcic, A., Sveinsson, T. & Tarokh, V. ( 2003 ). An Empirical Indoor Path Loss Model For UltraWideband Channels, Journal of communication and networks,12292370 4 5 303 308  8.
Ghassemzadeh S. S. Greenstein L. J. Sveinsson T. Kavcic A. Tarokh V. 2005 UWB Delay Proﬁle Models for Residential and Commercial Indoor Environment,  9.
Ghassemzadeh S. S. Jana R. Tarokh V. Rice C. W. Turin W. 2002 A Statistical Path Loss Model for Inhome UWB Channels,  10.
HewlettPackard Company 1989 Time and Frequency Domain Transforms, In:  11.
Hovinen V. Hämäläinen M. Pätsi T. 2002 Ultra Wideband Indoor Radio Channel Models: Preliminary Results,  12.
Molisch A. F. 2005 Ultrawideband Propagation ChannelsTheory, Measurement and Modeling,  13.
Molisch A. F. 2006 UWB Propagation Channels, In:  14.
Saleh A. A. M. Valenzuela R. A. 1987 A Statistical Model for Indoor Multipath Propagation.  15.
Suzuki, Y. & Kobayashi, T. ( 2005 ). Ultra Wideband Signal Propagation in Desktop Environments, IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences,09168508 9 E88A 2272 2278  16.
Taniguchi T. Maeda A. Kobayashi T. 2006 Development of an Omnidirectional and LowVSWR Ultra Wideband Antenna,  17.
Varela M. S. Sánchez M. G. 2001 RMS Delay and Coherence Bandwidth Measurements in Indoor Radio Channels in the UHF Band,  18.
Živković, Z. & Šarolić, A. ( 2010 ). Measurements of Antenna Parameters in GTEM cell. Journal of Communications Software and Systems,18456421 4 6 125 132