Int. J. Communications, Network and System Sciences, 2010, 3, 870-876
doi:10.4236/ijcns.2010.311118 Published O nline Novem ber 2010 (http://
Copyright © 2010 SciRes. IJCNS
UWB Channel Modeling Improvement in Indoor
Line-of-Sight (LOS) Environments
Vahid Tabataba Vakili, Dariush Abbasi-Moghadam, Mostafa Yahyaabadi
School of Electrical Engineering, Department of Telecommunications, Iran University of Science & Technology (IUST),
Narmak, Iran
Received August 1, 2010; revised September 19, 2010; accepted October 21, 2010
Channel measurement and modeling are important issues when designing ultra wideband (UWB) communi-
cation systems. A Precise model of the channel response is inevitable for designing a UWB telecommunica-
tion system. In this article signal propagation in indoor environment and LOS condition is evaluated and the
appropriate model of this scenario is presented. Parameters such as the power delay profile, mean excess de-
lay, delay spread, “NP10dB” are analyzed and simulated. Based on the analysis results, the proposed model is
presented. This model is based on Two-cluster approach but its average power delay profile is described with
power function and cluster time of the arrival is modeled by the modified exponential distribution. Finally
UWB channel parameters of the proposed model, Saleh and Valenzuela (S-V) and Two-cluster models are
compared. Measurement and simulation results show that considerable improvement for mean excess delay,
delay spread and “NP10dB” of proposed model comparing with S-V and Two-cluster models, this means the
channel is better described, which mean the channel is described more precisely.
Keywords: Channel Modeling, S-V Model, Power Delay Profile, Ultra Wideband
1. Introduction
UWB technology has been employed for several decades
in military and commercial communications applications
like high-speed mobile local area networks, imaging and
surveillance systems, ground penetratio n radars, automo-
tive sensors, medical monitors and recently wireless
personal area networks. FCC has allocated band width
from 3.1 GHz to 10.6 GHz to ultra wideband systems [1].
In recent years, these systems have gained more attention
because of their advantages over narrowband systems.
This system’s RF signal consists of ultra short pulses
with low power spectral density. Low transmission
power (–41.3 dBm) and large bandwidth together render
the power spectral density of the transmitted signal ex-
tremely low, which allows the frequency-overlay of a
UWB system with other existing radio systems such as
GPS, IEEE802.11x and WLNA [2]. Spread-spectrum
communication systems using ultra-short impulses have
seen a renewed interest because of its fine resolution in
delay to the order of several nanoseconds though at the
cost of an ultra wide frequency band.
Channel transmission is a propagation environment
through which the signal passes from transmitter to re-
ceiver. The propagation channel influences design as-
pects such as con struction of the matched filter, choice of
the Rake receiver structure, and search algorithms for
geolocation of transceivers. As propagation environment
of ultra wideband is usually indoor and crowded, the
transmitted signal passes through different paths to re-
ceiver like other wireless channels, so the received signal
is a combination of multi-path components which has a
destructive form over the transmitted signal. These dif-
ferent multipath components are realized by different
delays, various phases and amplitudes, therefore these
three parameters should be included in the channel
model. Precise modeling of channel is essential for de-
signing UWB systems. S-V model which has been con-
sidered as the accepted standard by IEEE802.15.3a
committee is the most well known model for modeling
ultra wideband channels [3]. But S-V model is a standard
model for wireless propagation in NLOS and is not pre-
cise for LOS. Also one of the other problems of this
model is differentiation of clusters and statistical model-
ing of them [4]. For solving these issues, another model
has been offered which is called Two-cluster [5]. This
model is based on S-V model for ultra wideband channel
including several stochastic clusters while in this model
only two deterministic clusters are considered. Unfortu-
nately, since in Two-cluster model, the second cluster is
larger that S-V model clusters, we cannot define the av-
erage power delay profile with exponential function. A
model has been proposed in this paper in which the av-
erage power delay profile is defined by power function.
The paper is organized as follows: in Section 2, meas-
urement conditions are explained, in Section 3, Two-
cluster and S-V models are introduced, Section 4 pre-
sents the proposed model, while in Section 5 new model
is simulated and finally concluding remarks are pre-
sented in Section 6.
2. Measurement
Propagation measurements have been made to character-
ize the UWB signal propagation channel. Using short
pulse with sub-nanosecond width, the impulse response
of the channel can be observed. Figure 1 is the block
diagram of the experimental setup. Transmitter consists
of a pulser, a periodic pulse generator, and a transmitting
antenna. The periodic pulse gen erator generates a regular
frame clock signal with period of 1 microsecond, which
triggers the pulser. The periodic pulse generator is also
connected to a DSO (Digitized Sampling Scope) by co-
axial cable to provide a trigger signal for a measurement
of the receiving antenna output. The pulser generates
Gaussian-shaped pulses with sub-nanosecond duration.
Figure 2 shows the output signals of a typical pulser.
Since the antenna system differentiates and filters the
pulser’s output, a more complex waveform is detected by
the DSO. Incoming signal is differentiated at receiving
antenna and observed at DSO. The DSO takes samples
over many periods of the transmission to construct one
received waveform, and averages several such waveform
measurements. Figure 2(b) shows a typical signal meas-
ured by the DSO in an indoor setting. The LOS path
component of the signal is shown in the first two or three
nanoseconds of the response, and is followed by a num-
ber of multipath components.
Experiments were carried out in a laboratory with size
Figure 1. A block diagram of the measurement apparatus [6].
of 2.7 m (height), 13.5 m (length) and 8.5 m (wide). The
transmitter antenna has been located 165 cm far from the
floor near the southern wall. LOS receiver, which is
named F1, has been located 9.5 meter far from the
transmitter, near the western wall. The distance from
receiver antenna to floor is 120 cm. In Figure 3, the plan
of laboratory has been shown from above. In receiver
location, a square matrix of 7 × 7 = 49 sensors with 15
cm spacing has been provided.
CLEAN algorithm was used to extract the CIR from
measurements data of the received waveforms [8,9]. The
power delay profile [10,11], defined by (1) is averaged
over 49 measured channels and is plotted in Figure 4.
49 )(
Experimental data have been used as criteria for exami-
nation of model’s accuracy. Therefore, each model im-
pulse response is a random process which cloud be ade-
quately presented by three random variables (parameters):
mean excess delay, RMS delay spread NP10dB. CDF and
mean of each parameters are computed by simulation
using at least 1000 runs. It is obvious, CDF and mean of
a model which is closer to measured CDF and mean will
be selected as a more precise channel model as described
in the following section.
Figure 2. (a) Pulser output; (b) Received signal.
Copyright © 2010 SciRes. IJCNS
Figure 3. Plan of laboratory where the LOS propagation
measured experiment was performed [7].
Figure 4. Average power delay profile versus excess delay
in a semi logarithmic scale for the 49 LOS locations in the
3. S-V and Two-Cluster Model
It can be seen from Figure 4, due to multipath compo-
nents in UWB systems, each cluster consists of several
rays. This definition of ultra wideband systems were in-
troduced by “Saleh and Valenzuela” for the first time [3,
11-13]. The following impulse response has been pro-
posed for UWB channel [14]:
 L
lkllk TtXth
00 ,, )()(
where L denotes the number of clusters, l is the cluster
index, k is ray index, L is total number of clusters, K is
the total number of rays in each cluster. X Indicates shad-
owing factor and lk , is ray gain for kth ray of lth cluster.
These two parameters are modeled by lognormal distri-
bution. Also l is lth cluster time arrival, lk ,
is time
arrival of kth rays of lth cluster. Assuming time of arrival
distribution as Poisson, time interval distribution should
be considered as expon ential.
S-V model has several problems. First, this model is
not accurate for LOS and indoor applications. Second,
finding the accurate values of parameters like cluster
arrival rate and inter cluster exponential decay constant
is difficult for modeling cluster behaviors based on ex-
perimental data .This would require some specific algo-
rithms of distinguishing clusters from one another. In
order to compensate for this disadvantage, an UWB
channel model with two deterministic clusters and sto-
chastic arriving rays within each cluster were proposed
[5].This model is based on S-V model, but the only dif-
ference is that instead of several clusters with random
time of arrival, only two clusters are considered with
deterministic time of arrival (Figure 4). Also, instead of
determining the gain of first ray of each cluster stochas-
tically in S-V model, the gain of first ray in first and
second clusters are computed deterministically. The other
model components, i.e., gain and time of arrival of next
rays in each two cluster are modeled statistically. The
channel has the following response:
00,0,)( )( )( N
kk Tttth
where lk , is multipath gain coefficient, Tm is the time
interval between two clusters, k,0
is the delay of the
kth multipath component relative to the first cluster arri-
val time, lk ,
is the time delay of the kth multipath
component relative to the second cluster arrival time (Tm),
M is the number of paths in the first cluster, and N is the
number of paths in the second cluster. The parameters of
this model can be divided into deterministic and stochas-
tic parts. In order to calculate the deterministic parame-
ters, it is enough to have length, width and height of the
room, electromagnetic properties of reflecting surface,
polarization and bandwidth of transmitted and received
signal. The statistical parameters are modeled like S-V
4. Proposed Model
The proposed model is based on Two-cluster model. The
difference is that in this model, the average power delay
profile is defined by power function instead of exponen-
tial function and the time interv al between times of rays’
arrival is modeled by modified exponential function in-
stead of exponential model. The reasons for above modi-
fications are explained in following sections.
Copyright © 2010 SciRes. IJCNS
4.1. Deterministic Part of Model
Deterministic part of the model includes the path gain of
the first ray of two clusters and their time delay. In this
model, LOS ray is considered as the first ray of the first
cluster. The arrival time of this ray is assu med to be zero
and its gain is determined based on path loss characteris-
tic of fr ee space:
0,0 (4)
where c is light speed, m
is the geometrical mean of the
upper and lower signal frequency limits, d is the distance
between the transmitter and receivers’ antenna. The first
ray of the second cluster is a ray which is reflected once
from six reflecting surfaces (four walls, ceiling and floor
of laboratory). The arrival time of this ray is calculated
as follows:
min{ }, 1,2,3,4,5,6
where 0
is the length of direct path and i
is the length
of reflected rays. For computing the gain of this ray ex-
cept path loss of free space, we should consider the loss
resulting from reflection [15]. Therefore:
0,1 0
min{ }
 
With , and
k for ver-
tical polarization and
k for horizontal
polarization and 120a/r
. Here, r
relative dielectric constant and the conductivity of re-
flecting surface [16] respectively. By the abovemen-
tioned definitions, the deterministic parameters are cal-
culated and presented in Table 1.
4.2. Statistical Part of Model
The gain and arrival time of the next rays in both clusters
are modeled statistically. The gain of these rays is mod-
eled as follows, but with slight difference with S-V
1/0,1/0,1/0,  kP kkk
where 1/0,k will be with equal probability. 1/0,k
is lognormal fading of rays with standard deviation of δ:
1 ),,( )( log202
1/0,1/0,  kNormal kk
Determination of depends on the fun
ty of average power delay profile. In Two-cluste
r m
el, the average podelay profile is defined by
exponential function:
21/0,0 ke
As Figure 5 shows, the average power
does not follow the exponential function accurately
delay profile
ntinuous line shows the exponential function). The
reason why is the fact that the second large cluster con-
sists of several small clusters are not considered in mod-
eling of the average PDP as exponential function of time.
In order to achieve more accuracy, we define the av-
erage power delay profile as a power function of time:
21/0, )( (1
 m
where is ray gain for multi-path,
lk ,
lk ,
is arrival
time rays of lth cluster and T
own in Figure 5. The dashed line shows
of kth m
arrival t.
The adjustment of power delay profile with power
function is sh
is the second cluster
wer function. As Figure 5 shows, the average power
Table 1. Values of deterministic parameters of model.
Path gain
tion rizontal
Excess Delay (ns)
0.009 0.009 0 LOS
0.0069 0.0058 5. Rd 44eflecte
Figure 5. PDP modeled with Power function and Exponen-
tial function.
Copyright © 2010 SciRes. IJCNS
delay profile based on power function, compared with
exponential, is more fitted to measured PDP data. The
mean and variance of error between averaged PDP based
on power function and measured PDP and also error be-
tween averaged PDP based on exponential function and
measured PDP are presented in Table 2. As this table
shows the difference between variances of these two
function models is 2.0166 dB respectively. Hence we
should compute lk ,
based on new definition of aver-
age power delay profile.
Since 1/0,k
is modeled with lognormal distribution
(Equation 8), we can have:
mal (11)
From (12) and (13) we have:
),,( b
bb ezEezm
),(),( 222
kkmNormalmNormalk (13)
)(22 1/0, ,][
eE Varm
 Varm k (14)
By equalizing (10) and (14) we conclude that:
mk T
 m
T (15)
Therefore, in the proposed model is the
no k,0/1
ion log-
rmal fading term with standard deviat
and lk,
which was defined in (Equation 15). Arrival time of ray
in Two-cluster model described with exponential func-
tion, but in proposed model, arrival time of rays de-
scribed with modified exponential function (Equation
16). Comparison of arrival time of ray of these two mod-
els with measured data shows that the prop osed model is
improved. So in the proposed model, the time interval
between times of rays’ arrival is modeled with modified
exponential distribution:
,0/1 1,0/1
(| )(1)exp(())
 
11 ,0/11,0/1
exp( ())
k k
  
where b is close to 1. The values of b,,,
n in Table 3
esult of Two-cluster model,
-V model and the proposed model have been compared.
ariances of error.
wer function
were used in simulation have been show.
5. Simulation Results
n this section simulation r
This comparison has been done regarding mean and cu-
mulative distribution function of three key parameters,
i.e.,mean excess delay, RMS delay spread and “NP10dB”.
First, several channels have been simulated for both pre-
vious models and proposed model by MATLAB 7 Then
the parameters of each simulated channel are computed
and recorded and the mean and cumulative distribution
function for each parameter was obtained from these
recorded values. In order to compare and evaluate these
models, the obtained values of mean and cumulative dis-
tribution function in measured data have been considered
as reference. In Table 4 the result of mean value of pa-
rameters is compared. The average value of mean excess
delay in the proposed model, Two-cluster model and S-V
model are 14.68 ns, 15.28 ns and 16.04 ns, respectively.
Also, the mean value for RMS delay spread in the pro-
posed model, Two-cluster model and S-V model are
17.54 ns 17.62 ns and 18.76 ns, respectively. The mean
value of “NP10dB” in the proposed model, Two-cluster
model and S-V model are 7.85, 8.91 and 8.93, respec-
tively and “NP10dB” in measured data is 7.51. Compare
these values with obtained values from measurements
show that the average relative error proposed model pa-
rameters, mean excess delay and delay spread and
“NP10dB” in comparison with the Two-cluster model
were 4.2 percent, 0.45 percent, 14.11 percent, compared
S-V model, respectively, 9.5 percent, 6.88 percent and
14.38 percent decrease.
Table 2. V
PoExponential function
22.2731 24.2897
error variance
Tabaramete proposed modimu-
le 3. Value of prs used inel s
(1 /ns)
2(1 /ns)
0.b 095
Table 4. Average value of key parameter.
14.68 15.28 16.04 14.25
Mean Excess
17.54 17.62 18.76 17.52
RMS Delay
7.85 8.91 8.93 7.51
pread (ns)
Copyright © 2010 SciRes. IJCNS
Figure 6. CDF of mean excess delay.
Figure 7. CDF of RMS delay spread.
Figure 8. CDF of number of significant multipaths.
In Figures 6-8 the CDF of mean excess delay, RMS
delay spread and “NP10dB” have been drawn for three
models. As it can be seen, CDF of mean excess delay
and RMS delay spread in the propos ed model is closer to
the measured values than Two-cluster model and S-V
model. Also CDF of “NP10dB” is better for the proposed
model comparing to the previous models. The ray arrival
time in the proposed model, which is described with ex-
ponential distribution using 1
, 2
, is more precise
compared to S-V and Two-clodels. Furthermore,
the ray arrival time and the ray gain of the proposed
model are closer to the experimental results.
6. Conclusion
In this paper, a model has been presented for indoor LOS
t o
he modified
n. Results show that the cumulative
ons of three parameters of proposed
odel has been recovered comparing with S-V model
uster m
B channel. The proposed model for indoor L
WB channel doesn't need the description of clusters
and knowing the mean parameters of arrival rate of clus-
ter and their variance comparing to S-V model. Also,
instead of approximating the gain of first ray with a
mean value, it is determined accurately and sub stitu ted in
the model. Moreover instead of simulating the time arri-
val of several clusters with statistical distribution, we
should compute deterministically an d use the time arrival
of two clusters. As in Two-cluser mdel, several clusters
of S-V model are considered as the second cluster. In
proposed model, the average power delay profile was
approximated with power function. Furthermore in the
oposed model, times were modeled by tpr
exponential functio
distribution functi
and Two-cluster model. The mean value for mean excess
delay and RMS delay spread and “NP10dB” in the pro-
posed model have been recovered 0.6 ns, 0.08 ns and
1.06 comparing with Two-cluster model and 1.36 ns,
1.22 ns and 1.08 comparing with S-V model. Therefore
the proposed model fits better for data and has less vari-
ance, hence it can model the ch annel better than S-V and
Two-cluster model in LOS environment.
7. References
[1] M. Z. Win and R. A. Scholtz, “Characterization of Ul-
tra-Wide Band Width Wireless Indoor Communication
Channel: A Communication Theoretical View,” IEEE
JSAC, Vol. 20, No. 9, December 2002, pp. 1613-1627.
[2] Multiband OFDM Physical Layer Proposal for IEEE
802.15 Task Group 3a.
[3] A. Saleh and R. Valenzuela, “A Statistical Model for
Indoor Multipath Propagation,” IEEE Journal on Selected
Areas in Communications, Vol
5, No. 2, February 1987,
pp. 128-137.
Copyright © 2010 SciRes. IJCNS
Copyright © 2010 SciRes. IJCNS
Systems Engi-
lectronics, Vol. 18, No. 4, December 2007,
tromagnetic Compatibility, Vol. 46, No. 4,
IEEE 19th International
of the Power Delay Profile,” Elsevier Interna-
3a, 2002.
[4] S. Venkatesh and J. Ibrahim, “A New 2-Cluster Model for
Indoor UWB Channel Measurement,” IEEE in Antenna
and Propagation Society Symposium, Vol. 1, June 2004,
pp. 946-949.
[5] Y. Wang, N. T. Zhang, Q. Y. Zhang and Z. Z. Zhang,
“Characterizing Ultra-Wide Band Indoor Line-of-Sight
Wireless Channel,” Elsevier Journal of
neering and E
pp. 673-678.
[7] D. Cassioli, M. Z. Win and A. F. Molisch, “The Ul-
tra-Wide Bandwidth indoor Channel: From Statistical
Model to Simulations,” IEEE Journal on Selected Areas
in Communications, Vol. 20, No. 6, August 2002, pp.
[8] Y. Wang and N. T. Zhang, “A New Multi-Template
CLEAN Algorithm for UWB Channel Impulse Response
Characterization,” IEEE Conference on Communication
Technology, Guilin, November 2006, pp. 1-4.
[9] C. Buccella, M. Feliziani and G. Manzi, “Detection and
Localization of Defects in Shielded Cables by
Time-Domain Measurement with UWB Pulse Injection
and Clean Algorithm Post Processing,” IEEE Transac-
tions on Elec
of Tr
November 2004, pp. 597-605.
[10] J. H. Kim and Y. K. Yoon, “The Multipath Delay Spread
Model for the LOS Case,” Antennas and Propagation
Society International Symposium, San Diego, July 2008,
pp. 1-4.
[11] R. G. Vaughan and N. L. Scott, “Super-Resolution of
Pulsed Multipath Channels for Delay Spread Characteri-
zation,” IEEE Transactions on Communications, March
1995, Vol. 47, No. 3, pp. 343-347.
[12] Y. F. Chen, “Statistical Properties of IEEE UWB Channel
Models and Their Application,
Symposium on Indoor and Mobile Radio Communica-
tions, Cannes, September 2008, pp. 1-5.
[13] J. Wout, D. Jeffrey, V. Leen and M. Luc. “Statistical
tional Journal of Communications, In Press, Available
online, 7 July 2009.
[14] J. Foerster, “Channel Modeling Sub-Committee Report
Final,” IEEE P802.15-02/368r5-SG
[15] V. Marshall and G. Skite k, “Electromagnetic Concep t and
Applications,” Prentice-Hall International, Inc., 1990.
[16] P. R. Barnes and F. M. Tesche, “On the Direct Calculation
ansient Plane Wave Reflected from a Finitely Con-
ducting Half Space,” IEEE Transaction on Electromag-
netic Compatibility, Vol. 33, No. 2, May 1991, pp. 90-96.