Progress In Electromagnetics Research B, Vol.

49, 155176, 2013

ON THE STUDY OF EMPIRICAL PATH LOSS MODELS
FOR ACCURATE PREDICTION OF TV SIGNAL FOR
SECONDARY USERS
Nasir Faruk
1, 2, *
, Adeseko A. Ayeni
1
and Yunusa A. Adediran
2
1
Department of Telecommunication Science, University of Ilorin,
P.M.B. 1515 Ilorin, Kwara State, Nigeria
2
Department of Electrical and Electronics Engineering, University of
Ilorin, P.M.B. 1515 Ilorin, Kwara State, Nigeria
AbstractDemand for wireless communication technologies and
systems keep increasing and has reached the peak where the capacity
can only be achieved by improving spectrum utilization. The spectrum
allocated to TV broadcast systems can be shared by wireless data
services through exploiting spatial reuse opportunities (Spatial TV
white space). Path loss models are used extensively in signal
prediction, coverage optimization and interference analysis. Recently
it is being used in estimating distances for safe operation of secondary
users in TV white space. Peculiarities of these models give rise to high
prediction errors when deployed in a dierent environment other than
the one initially built for. It is however not very clear which model
gives the best t and what the penalties are for using the models
outside the intended coverage area. In this paper, we assess the tness
of nine empirical widely used path loss models using ve novel metrics
to gauge their performance. In order to achieve this, eld strength
measurements were conducted in the VHF and UHF regions along six
dierent routes that spanned through the urban, suburban and rural
areas of Kwara State, Nigeria. A program was developed in VB 6.0
language to compute the path losses for the empirical models. The
measurement results were converted to path losses and are compared
with the models prediction. The results show that no single model
provides a good t consistently. However, Hata and Davidson models
provide good tness along some selected routes with measured RMSE
values of less than 10 dB. ITU-R P.1546-4, Walsch Ikegami (WI),
Received 13 January 2013, Accepted 16 February 2013, Scheduled 19 February 2013
* Corresponding author: Nasir Faruk ([email protected]).
156 Faruk, Ayeni, and Adediran
Egli, CCIR and FSPL perform woefully, with higher RMSE and SC-
RMSE (Spread Corrected RMSE) values. Further analysis on the
error spread as a function of distance along 60 km route revealed that
Hata and Davidson models show symmetry up to about 30 km with
slight divergence between 24 km and 30 km after which Davidson model
gives lower prediction error along the route. The prediction errors
for Davidson model distributes nearly symmetrically around the mean
error of 2.15 dB. It is noteworthy that the Gaussian error distribution
within the window of 5 dB dominates the frequency counts. However,
the error counts for CCIR model closely follow normal distribution
with a mean error of 6.37 dB but Hata, FSPL, Walsch Ikegami and
ITU-R P. 529-3 models do not follow normal distribution curve.
1. INTRODUCTION
Demand for wireless communication technologies and systems keeps
increasing and has reached the peak where the capacity can only
be achieved by improving spectrum utilization. Spectrum allocated
to TV operators can potentially be shared by wireless data services,
either at the times when the primary service is switched o or by
exploiting spatial reuse opportunities. For a couple of years researchers
have focused on how to evaluate/quantify TV white space. The
TV white space can be temporal (i.e., times/periods the primary
service is o) or spatial (i.e., where TV signals cannot be successfully
received),technically when the reception level is less than 116 dBm for
digital TV (DTV) and 94 dBm for analogue TV [1]. The temporal
white spaces, have not been subjected to extensive research because the
idea is that the DTV (digital television) will be operating for 24 hours
daily. In order to recover the spatial TV white space, signal prediction
techniques are required to make a decision whether the location is
white space or not, and the decision is based on threshold (i.e., if the
received signal level at the position is greater than a certain value as
described in the IEEE 802.22 draft. The incumbent systems currently
operating in the TV bands are analogue TV with sensitivity value
of 94 dBm, digital TV with sensitivity of 116 dBm and wireless
microphone with 107 dBm. In this regard, Federal Communications
Commission (FCC) in the United States announced 114 dBm as the
criteria of the empty spaces for TV white space [1]. The whole idea of
these is to free more spectrums (white spaces) for secondary access so
that low power, low-range wireless devices in a strictly localized manner
(Keep-out distance) can utilize the white space without interfering with
the TV transmission. Figure 1 shows the spatial deployment scenario
Progress In Electromagnetics Research B, Vol. 49, 2013 157
Figure 1. Deployment scenarios for co-channel and adjacent channel
TV band devices in TV white space [2].
of TV band devices (TVBDs) also known as white space devices

, in
TV white space (TVWS).
Figure 1 shows potential application scenario of primary user (TV
broadcast service) and secondary user (TVBD) in TVWS. Several
applications have been proposed to be deployed in TVWS as reported
in [31] these include UMTS and LTE extension, Wi-Fi-2, Wimax and
public safety and emergency networks. Public safety has been receiving
attention globally due to global deserter; in this regard, [23] focused on
deploying TETRA (terrestrial trunk radio access) in TVWS. TETRA
systems is currently used by government agencies, emergency services,
(police forces, re departments, ambulance) for public safety networks,
rail transportation sta for train radios, transport services and the
military. Figure 2 shows typical deployment scenario of TETRA
network in TVWS.
1.1. Predicting TV Coverage Using Path Loss Models
Today, propagation models are used extensively in coverage planning
and optimization and signal prediction, and is found very useful
for interference analysis. Path loss models are applied in cellular
environments, xed wireless access systems and TV broadcast systems.
They are to be used here for the prediction of TV coverage. The

White space device is an FCC-certied wireless device that can be used in the RF
spectrum below 700 MHz. The devices are divided into two categories: xed and
personal/portable.
158 Faruk, Ayeni, and Adediran
Figure 2. TETRA system deployments in TV white space [23].
success and peaceful coexistence between the primary users and the
secondary users (white space devices) depend on the propagation
characteristics of the channel. Received signal prediction models
would play an important role in the coverage optimization and maybe
eciently used based on FCCs rule to predict locations for safe
operation for secondary users. The existing path-loss models have
been classied into theoretical and empirical models. The theoretical
models predict transmission losses by mathematical analysis of the
path geometry of the terrain between the transmitter and the
receiver and the refractivity of the troposphere [3]. Empirical models
add environmental-dependent loss variables to the free-space loss to
compute the net path loss in the corresponding environment. These
models require measurements and so considered more accurate in view
of its environmental compatibility. Path loss models will help in
the design of transmission strategy such as the transmit power and
frequency. These models can dier in their properties with locations
due to dierent terrain environment.
Most existing TVWS studies employ the use of propagation curves
such as the ITU Radio communications Sector (ITU-R) P.1546-2,
Egli, Okumura and Hata models for predicting the TV coverage.
These models are built based on measurements conducted in regions
that are dierent from Nigeria; suitability in terms of usage may
therefore vary due to environmental factors and terrain prole. In
addition, peculiarities of these models gives rise to high prediction
errors when deployed in a dierent environment other than the one
initially built for. These errors may consequently aect secondary
operations. This raises the question of whether to adopt or modify
the existing prediction models or to build a new model that will
minimize the errors and protect the primary users from excessive
Progress In Electromagnetics Research B, Vol. 49, 2013 159
interference from secondary users. Interference is not the only case;
the error could also have eect on the amount of white space recovery
and could have signicant impacts on the deployment of secondary
networks. For instance Camp et al. [4] show that wireless mesh network
planned with a given path loss model can massively under or over
provision as a results of small change in model parameters. This is
a big issue as over-provision would add cost during roll out phase
while under provision would aect the QoS of the network. Anang et
al. [12] show that cellular systems information capacity changes due
to propagation loss and system parameters, including the path loss
exponent. It was concluded that decrease in path loss exponent causes
severe interference. It is however not very clear, which models give the
best t and what the penalties are for using the models outside the
intended area. Therefore, it is necessary to have accurate assessment of
the propagation models in order to modify a model or choose a better
model to achieve high accuracy thereby minimizing errors and thus,
increasing exibility in local spectrum usage.
In this paper, we assess the tness of nine widely used empirical
path loss models using ve novel metrics to gauge their performance.
The focus in this paper is the ecacy of these models at predicting
path loss values for safe operation of secondary users in the chosen
environment. In order to achieve these, eld strength measurements
were conducted in the VHF and UHF frequencies along six dierent
routes that spanned through the urban, suburban and rural areas. A
program was developed in VB 6.0 language to compute the path loss
for the empirical models. The measurement results were converted
to path losses and are compared with the models prediction. The
chosen models are Hata [5], COST 231 [6], Walsch [7] and Ikegami [8],
Egli [9], ITU-R P.529-3 [10], ITU-R P.1546-4 [11], CCIR [13],
Davidson [14] and FSPL [15]. This paper is organised as follows
Section: Section 1 provides introduction; Section 2 presents the related
work; method of data collection is presented in Section 3; Section 4
provides the metrics used; Section 5 presents the results and, nally,
Section 6 concludes the paper.
2. RELATED WORK
There are lots of published research that worked on analyzing the
ecacy of path loss models. In such cases, the authors often
collect measurement data in an environment of interest and make an
assessment of whether the models t in. [16] and [17] provide practical
lower bounds on the prediction accuracy of path loss models. In the
works 30 propagation models that had been published in the last 70
160 Faruk, Ayeni, and Adediran
years were considered. A large scale measurement was conducted in
the diverse set of rural and urban environments. In the end, it was
concluded that no single path loss model was able to predict path
loss consistently. In [18], a comparative assessment of ve models was
presented with respect to the data collected in the urban and suburban
environments at 910 MHz. However, the paper does not provide a
conclusion about which model gives the best results. [32] provides a
comparative analysis using four empirical models for WCDMA and
GSM systems based on drive test data collected from Kano city
and Abuja city, which are all urban areas in Nigeria. In all of the
measurements taken, it was found that COST 231 and Hata give fairer
results for Kano and Abuja environment. The work is considered as
an extension of the one presented in [3] where COST 231 was found
more suitable for use in the GSM 1800 band for Kano environment.
[28] provides a comparison of empirical propagation path loss models
for xed wireless access systems based on measurement conducted in
Cambridge, UK. It was found that, among the contenders, the ECC-33
model, the Stanford University Interim (SUI) model, and the COST-
231 model show the most promise and that the SUI model shows quite
a large mean path loss prediction error. [19] presents similar results
to that in [28]. Also [20] conducted a mobile propagation path loss
studies at VHF/UHF bands in Southern India. In the work, eld
strength was measured at 200, 400 and 450 MHz and their result
shows that Hatas prediction method gave better agreement in all
cases. This work is similar to that presented by [21]. Achtzeh et
al. [22] analysed the accuracy of three widely used path loss models in
predicting TV signal strength using data carried out in a medium-sized
central European city. In the work, spatial statistics based technique
was employed for estimating the coverage. Also in [23], three empirical
path loss models, i.e., ITU-R P.1546-3, Hata and ETRI (Electronic
Telecommunication Research Institute) models were used to calculate
propagation distances for safe operation of TETRA system on DTV
white space. In [33] ITU-R model is used to address spectrum sharing
issues between IMT-advanced, xed wireless systems (FWS) and TV
broadcast service. In the work, the path loss model was used to
investigate inter system interference between Wimax, FWS and mutual
coexistence between them.
The amount of white space that would be free for the TVBD in
accordance with the regulatory guidance on interference has been a
subject of extensive studies. The amount of white space has been
acquired in several places across the globe, for the United States
Harrison et al. [24] and for central Europe by Van de Beek et al. [25].
Very few studies exist for outside the United States, such as [26], which
Progress In Electromagnetics Research B, Vol. 49, 2013 161
attempts to quantify TVWS capacity in the United Kingdom in a
limited area, and [27] for southern Europe using ITU-R model. Of
recent, [2] presented experimental and simulation results for the use of
TVBD in TVWS. In the work, keep-out distance which is the minimum
separation distance of TVBD from the DTV protected contour was
obtained using the Okumura, ITU-R 1546, FCC and measurements of
UHF signals in Korea. The schematic of the deployment scenario is
shown in Figure 1. So far, there has not been any published report
from Africa indicating such studies. U.S and related studies cannot be
directly extrapolated to Nigeria case due to dierences in deployment
scenario, activities of the primary users, regulatory aspects and terrain
prole. The work presented in this paper is the rst of its kind
in Nigeria that carries out an extensive analysis of large number of
propagation models using large amount of data set produced from real-
time measurements.
3. MEASUREMENT CAMPAIGN
This section provides the steps followed during data collection and
it gives the description of the equipment used. The propagation
measurements were conducted in Ilorin (Long 4

36

25

E, Lat
Table 1. Description of measurement routes.
1
2
3
4
5
6
Suburban area. It has regular building structure
with dual carriage way.
S/N Route Description
No.
Points
Route
(km)
STD
DEVIATION
PL (dB)
ASADAM
UNILORIN
GAMBARI
MURTALA
OLD
JEBBA
BODE-
Sa'adu
12,712
24,310
26,634
12,004
18,418
18,0274
10
7
11
8
13
60
10.33
6.71
11.87
13.95
13.93
15.28
Urban area. It has very complex terrain; some
areas are very high whereas, some parts are very
low. Within the University, heavy trees cover the
road. Along the route, the road is very narrow with
averagely two-storey buildings.
Dense urban area. It is historical area with very
old buildings around. It is very busy commercial
area.
Urban area. It has regular building structure with
arrange of three-storey buildings with dual
carriage road.
Urban area. It has regular building structure with
arrange of two-storey buildings with teo-lane road
Route spans from the urban to rural areas. It has
regular building structure with average of two-
storey buildings within the city, then outside the
city hotspots villages at an average distance of 15
km interval.
162 Faruk, Ayeni, and Adediran
8

25

55

N) and its environs within Kwara State, Nigeria. Ilorin is

a large city characterized by a complex terrain due to the presence
of hills and valleys within the metropolis. Outside the metropolis,
the routes are covered with thick vegetation. The altitude within
the transmitters coordinates is 403.7 m; this can be as low as 150 m
when travelling within and outside the metropolis. (see Figure 10 for
path prole for Bode-Saadu route) Six routes were covered during the
measurement campaign. The routes are Olorunshogo via ASADAM,
University of Ilorin (UNILORIN) via Pipeline, GAMBARI via Agaka,
MURTALA Mohd way, Old Jebba Road and BODE-Saadu.
Table 1 shows details of the measurement routes. Figure 3 shows
the screens hot of the measurement routes and Figure 4(b) shows an
aerial view of Ilorin metropolis and outside Ilorin. NTA Ilorin and
Kwara TV transmitters were utilized. NTA transmits on channel 5 at
203.25 MHz while Kwara TV transmits on channel 35 at 583.25 MHz.
While the transmission is taking place, a dedicated Agilent spectrum
analyser was placed inside a vehicle and driven at an average speed of
40 km/h along these routes. Details of the transmitter and the analyser
can be found in Table 2. Field strength was measured continuously and
stored in an external drive for subsequent analysis. Total route length
and number of points were 109 km and 286,870 respectively.
Figure 3. Measurement routes in Ilorin.
Progress In Electromagnetics Research B, Vol. 49, 2013 163
(a) (b)
Figure 4. (a) Mountain and vegetation cover outside Ilorin (BODE-
Saadu route). (b) Aerial view of Ilorin metropolis.
4. PERFORMANCE METRICS
The performance of the models is analysed using ve metrics;
Prediction error, root mean square error (RMSE), spread-corrected
root mean square error (SC-RMSE), normalized error probability
density function and rank correlation. The Prediction error, , is
the dierence between the measured path loss (P
i
) at distance i, and
models predicted path loss (p
m,i
) and is evaluated using Equation (1).

i
= P
i
p
m,i
(1)
Other sub metrics are the maximum and mean prediction error of
sample (n
j
)
MaxError = max
i
(
i
) (2)
MeanError =
1
n
j
j

i
(3)
RMSE also known as Root Average Squared Predication Error
(RASPE) and it is the most apparent metric for analysing error of
predictive models. We compute the prediction error values using
Equation (1) for each model as a function of distance from the
transmitter. The overall RMSE for a given model m, for a given data
164 Faruk, Ayeni, and Adediran
Table 2. Measurement equipment and conguration.
Spectrum Analyzer N9342C Agilent, 100 Hz7 GHz
Displayed average noise level (DANL) 164 dBm/Hz
Preamplier 20 dB
Resolution bandwidth (RBW) 10 kHz
Center frequency (NTA)
Center frequency (KWARA TV)
203.25 MHz
583.25 MHz
Impedance 50 Ohms
Receiver Antenna: Diamond RH799 RH 795
Frequency range 70 MHz1 GHz
Form Omni directional
Height 1.5 m
Gain 2.51 dBi
NTA Ilorin Transmitter
Power 2.4 kW
Frequency 203.25 MHz
Antenna height above the ground 185 m
Cable Type RFS HEL FEX 512
Impedance 50 Ohms
Coordinates 4

36

25

E, 8

25

55

N
set n is dened as;
RMSE =

_
1
n
n

i=1
|
m,i
|
2
(4)
A RMSE value closer to 0 indicates a better t. However, the
acceptable RMSE for a model is about 67 dB for urban areas [30] and
1015 dB for suburban and rural areas [29]. Another important metric
is the SC-RMSE, which helps to extract the impact of dispersion from
the overall error. This has the eect of reducing the error associated
with a noisy link. Computing SC-RMSE is similar to that of RMSE;
the only dierence is that the error is obtained by subtracting the
standard deviation from the absolute value of the error.

m,i
= |
m,i
|
i
(5)
SC RMSE =

_
1
n
n

i=1

m,i

2
(6)
Progress In Electromagnetics Research B, Vol. 49, 2013 165
The fourth metric is error distribution, i.e., the probability density
function of a Gaussian (Normal) random variable. Firstly, the model
has to follow normal distribution curve. Secondly, the error counts
from 0 to 10 dB should dominate the frequency counts since 010 dB
RMSE is the chosen performance criteria in this work.
The fth metric is the Spearmans rank correlation coecient ().
Which is a nonparametric measure of statistical dependence between
the measured and predicted path losses across the links. It assesses
how well the relationship between two variables can be described using
a monotonic function. A perfect Spearman correlation of +1 or 1
occurs when each of the variables is a perfect monotone function of the
other.
=

i
_
P
i
P
i
_ _
P
m,i
P
m,i
_
_

i
_
P
i
P
m,i
_
2

i
_
P
i
P
m,i
_
2
(7)
where, P
i
and P
m,i
are the mean measured path loss at distance i, and
mean models predicted path loss respectively.
5. RESULTS AND DISCUSSION
Figures 5 through 9 provide the graphical depictions of measured and
prediction path losses along the ve predened routes. Figure 5 shows
the comparison of the measured path loss with the predicted path
loss as a function of distance for ASADAM route. Within the rst
2 km along the route, CCIR model agrees with the measured path loss;
thereafter, CCIR over estimates the path loss. Walsch Ikegami, ITU-
R P.1546-4 and Egli models underestimate the path loss throughout
the range of interest. ITU-R P.529-3, Hata, Davidson and COST 231
model give better results. For the overall route, COST 231 model
provides the best result with RMSE value of 0.4 dB which is a fantastic
result. ITU-R P.529-3 and Davidson models are actually derivatives of
Hata and the results are, thus, expected to be the same for distance
of less than 20 km. Hata model turns out to give RMSE value of
10.7 dB and SC-RMSE value of about 4 dB. However, ITU-R P.1546-4,
Walsch, Ikegami, Egli, CCIR and FSPL perform woefully with higher
RMSE and SC-RMSE values. The corresponding error statistics in
terms of the RMSE and SC-RMSE are shown in Table 3.
Figure 6 depicts the result of UNILORIN route. The path
loss obtained using the empirical model resembles that presented in
Figure 5. For all the measurement routes studied, ITU-R P.529-3,
Hata and Davidson models give the best results followed by COST 231
model, except for MURTALA route in Figure 8 where CCIR model
performs better with RMSE of 7.9 dB. Refer to Tables 3 and 4 for
166 Faruk, Ayeni, and Adediran
.
60
65
70
75
80
85
90
95
100
105
110
115
120
125
130
135
140
Measured Path Loss
WI Egli
Hata
0 1 2 3 4 5 6 7 8 9 10 11 12
Distance (km)
P
a
t
h

L
o
s
s

(
d
B
)
ITU-R P.5293 COST 231
Davidson
CCIR ITU-R P.1546-4
Figure 5. Comparison of empir-
ical models with measured path
loss for ASADAM route.
60
65
70
75
80
85
90
95
100
105
110
115
120
125
130
135
140
Measured Path Loss
WI Egli
Hata
0 1 2 3 4 5 6 7
Distance (km)
P
a
t
h

L
o
s
s

(
d
B
)
ITU-R P.5293
COST 231
Davidson
CCIR ITU-R P.1546-4
Figure 6. Comparison of empir-
ical models with measured path
loss for UNILORIN route.
60
70
80
90
100
110
120
130
140
150
Measured Path Loss
WI Egli
Hata
0 1 2 3 4 5 6 7 8 9 10 11 12
Distance (km)
P
a
t
h

L
o
s
s

(
d
B
)
ITU-R P.5293
COST 231
Davidson
CCIR ITU-R P.1546-4
13 14
Figure 7. Comparison of empir-
ical models with measured path
loss for OLD JEBBA route.
60
70
80
90
100
110
120
130
140
150
Measured Path Loss WI
Egli
Hata
0 1 2 3 4 5 6 7 8 9
Distance (km)
P
a
t
h

L
o
s
s

(
d
B
)
COST 231
Davidson CCIR ITU-R P.1546-4
Figure 8. Comparison of empir-
ical models with measured path
loss for MURTALA route.
statistical RMSE and SC-RMSE values for all the models across all the
measurement routes. Along ASADAM, UNILORIN and GAMBARI,
COST 231, Hata, ITU-R P.529-3 and Davidson models give better ts.
But, for MURTALA and OLD JEBBA routes, all the models perform
badly except for CCIR with RMSE of about 7.9 dB along OLD JEBBA
route. The reason for the high values of errors along MURTALA route
is attributed to the fact that the route is the busiest in the city. There
were present of scatters, moving vehicle and girder. Also the route
Progress In Electromagnetics Research B, Vol. 49, 2013 167
Table 3. Root mean square error for six measurement routes.
ROUTES
Hata
(dB)
COST
231
(dB)
WI
(dB)
ITU-R
P.529-3
(dB)
EGLI
(dB)
ASADAM 10.786 0.404 84.807 10.786 67.613
UNILORIN 0.866 9.468 68.297 0.866 68.431
GAMBARI 3.893 15.084 99.487 3.893 82.293
MURTALA 27.493 37.330 108.092 27.493 100.551
OLD JEBBA 18.972 29.266 106.906 18.972 91.090
BODE-Saadu 50.117 68.598 253.326 56.954 124.907
ROUTES
CCIR
(dB)
DAVID
SON
(dB)
ITU-R
P.1546-4
(dB)
FSPL
(dB)
-
ASADAM 51.132 10.786 52.440 113.918 -
UNILORIN 30.147 0.866 17.643 89.355 -
GAMBARI 36.452 3.893 24.310 128.598 -
MURTALA 7.971 27.493 43.716 132.933 -
OLD JEBBA 18.139 18.972 37.132 133.684 -
BODE-Saadu 16.510 14.408 33.344 311.294 -
has building structure with average of three-storey buildings with dual
carriage road.
Figure 9 provides the graphical depiction of measured and
prediction path losses for Bode-Saadu route, up to 20 km, ITU-R
P.529-3, Hata and Davidson models give the same RMSE values,
after which Davidson model provides better results over Hata. The
recoded RMSE values for Hata and Davidson are 50 dB and 14.4 dB
respectively. This result obviously favors Davison model for the
fact that six correction factors were included into Hata model.This
extended the range from 20 km to 300 km. Figure 4(a) shows mountain
and vegetation cover just about 10 km from the city the terrain prole
for this route is shown in Figure 10.
Along this route, there were high prediction errors for most of
the models, because of the complex nature of the terrain. At the
transmitters location, 400 m altitude was recorded; this can be as
low as 275 m just 5 km away from the transmitter and 150 m at a
168 Faruk, Ayeni, and Adediran
70
80
90
100
110
120
130
140
150
160
170
Measured Path Loss
Egli
Hata
0 10 20 30 40 50 60
Distance (km)
P
a
t
h

L
o
s
s

(
d
B
)
ITU-R P.5293
COST 231
Davidson
CCIR
Figure 9. Comparison of empir-
ical models with measured path
loss for Bode-Saadu route.
A
l
t
i
t
u
d
e

(
m
)
0 5 10 15 20 25 30 35 40 45 50 55 60
Distance (km)
NTA Transmitter's altitude
150
200
250
300
350
400
450
Figure 10. Path prole of Bode-
Saadu route.
Table 4. Spread corrected root mean square error (SC-RMSE).
ROUTES
HATA
(dB)
COST
231
(dB)
WI
(dB)
ITU-R
P.529-3
(dB)
EGLI
(dB)
ASADAM 3.920 6.774 49.629 3.920 32.512
UNILRIN 4.968 4.018 37.650 4.968 37.784
GAMBAR 16.616 14.534 56.429 16.612 39.235
MURTALA 20.435 12.146 58.615 20.435 51.073
OLD JEBBA 25.344 19.149 52.585 25.344 36.770
BODE-Saad 55.158 43.467 137.13 48.221 9.479
ROUTES
CCIR
(dB)
DAVID
SON
(dB)
ITU-R
P.1546-4
(dB)
FSPL
(dB)
-
ASADAM 18.947 3.920 17.261 78.740 -
UNILRIN 3.132 4.968 13.393 58.708 -
GAMBAR 0.804 16.616 24.061 85.540 -
MURTALA 28.962 20.435 20.420 83.456 -
OLD JEBBA 28.284 25.344 17.877 79.363 -
BODE-Saad 68.965 72.884 21.947 195.104 -
Progress In Electromagnetics Research B, Vol. 49, 2013 169
distance 60 km from the transmitter. This would obviously aect the
signal reception and thus; contribute to the error. In addition, except
for ITU-R P.529-3, ITU-R P.1546-4, CCIR and Davidson models, the
validity of the transmission distance for other models is less than 20 km.
Prediction error is expected to be higher when used to predict path
loss for distances greater than 20 km. Furthermore, we studied the
prediction and spread corrected errors as a function of distance for
Hata and Davidson models for Bode-Saadu route. Figures 11 and 12
show the results.
In Figures 11 and 12, it is worth noting that Hata and Davidson
prediction models show symmetry up to about 30 km with slight
divergence between 24 km and 30 km window for both metrics, after
which Hata model under predicts the path loss. Davidson model gives
the best result along this route. This indicates that, for wider range
path loss prediction, Davidson model would perform better than the
widely used Hata model. Interestingly, the error spreading for both
models follows the terrain prole (See Figures 10 & 11).
Figure 13 depicts the distribution histograms of the predication
error for the eight empirical models considered along the BODE-Saadu
route. The solid line indicates the probability density function (PDF)
of a Gaussian (Normal) random variable. In this scenario, Davidson
and CCIR models show similar shapes of their PDFs. The error were
normalized to t in to the Gaussian normal distribution.
In Figure 13(a), the prediction errors are nearly distributed
symmetrically around the mean error of 2.15 dB. It can be observed
that the error distribution within the 5 dB window dominates the
frequency counts. This indicates good tness of the model in terms of
predicting path loss in the region. However for CCIR model, which is
the second that performs better along this route the prediction error
15 20 25 30 35 40 45 50 55 60 65
HATA
Davidson
Distance (km)
0 5 10
20
15
10
5
0
-5
-10
-15
-20
-25
P
r
e
d
i
c
t
i
o
n

E
r
r
o
r

(
d
B
)
Figure 11. Prediction error
along BODE-Saadu route.
15 20 25 30 35 40 45 50 55 60 65
HATA
Davidson
Distance (km)
0 5 10
20
15
10
5
0
-5
-10
-15
-20
-25
S
p
r
e
a
d

C
o
r
r
e
c
t
e
d

E
r
r
o
r

(
d
B
)
Figure 12. Spread corrected
error along BODE-Saadu.
170 Faruk, Ayeni, and Adediran
closely follows normal distribution with a mean error of 6.37 dB as
shown in Figure 13(b). From all indications, the model underestimates
the path loss since the model is found to have negative skew. The error
counts are quite high and spread along the distribution.Hata and ITU-
R P.529-3 models do not follow the normal distribution curve, despite
the fact that the mean errors were found to be 2.2 dB comparable
with that of Davidson model that gives a better spread in the error
distribution. Egli model gives fair distribution with mean error of
15 dB and slightly follows the normal curve as shown in Figure 13(h).
FSPL and Walsch-Ikegami models provide worst results under this
0
5
10
15
1
5
20
40
60
80
95
99
0
5
1
5
20
40
60
80
95
99
0
5
10
15
20
1
5
20
40
60
80
95
99
0
5
10
15
1
5
20
40
60
80
95
99
-25 -20 -15 -10 -5 0 5 10 15 20 25
Normalized Prediction Error (dB)
0.5
D
i
s
t
r
i
b
u
t
i
o
n
C
u
m
u
l
a
t
i
v
e

D
i
s
t
r
i
b
u
t
i
o
n
-25 -20 -15 -10 -5 0 5 10 15 20 25
-20 -15 -10 -5 0 5 10 15 20
Normalized Prediction Error (dB)
-20 -15 -10 -5 0 5 10 15 20
D
i
s
t
r
i
b
u
t
i
o
n
C
u
m
u
l
a
t
i
v
e

D
i
s
t
r
i
b
u
t
i
o
n
-10 -5 0 5 10 15 20 25
Normalized Prediction Error (dB)
30
-10 -5 0 5 10 15 20 25 30
D
i
s
t
r
i
b
u
t
i
o
n
C
u
m
u
l
a
t
i
v
e

D
i
s
t
r
i
b
u
t
i
o
n
D
i
s
t
r
i
b
u
t
i
o
n
C
u
m
u
l
a
t
i
v
e

D
i
s
t
r
i
b
u
t
i
o
n
0.5
-5 0 5 10 15 20
Normalized Prediction Error (dB)
-5 0 5 10 15 20
(a) (b)
(c) (d)
Progress In Electromagnetics Research B, Vol. 49, 2013 171
0
5
10
15
5
20
40
60
80
95
0
5
10
15
1
5
20
40
60
80
95
99
0
5
10
10
20
40
60
80
0
5
10
15
20
1
5
20
40
60
80
95
99
0 10 20
Normalized Prediction Error (dB)
0.5
D
i
s
t
r
i
b
u
t
i
o
n
C
u
m
u
l
a
t
i
v
e

D
i
s
t
r
i
b
u
t
i
o
n
(e) (f)
(g) (h)
25 30 35 40 45 50 55 60
Normalized Prediction Error (dB)
25 30 35 40 45 50 55 60
-10 -5 0 5 10 15 20 25
Normalized Prediction Error (dB)
30
-10 -5 0 5 10 15 20 25 30
D
i
s
t
r
i
b
u
t
i
o
n
C
u
m
u
l
a
t
i
v
e

D
i
s
t
r
i
b
u
t
i
o
n
25 30 35 40 45 50
Normalized Prediction Error (dB)
15 20
25 30 35 40 45 50 15 20
30 40
0 10 20 30 40
D
i
s
t
r
i
b
u
t
i
o
n
C
u
m
u
l
a
t
i
v
e

D
i
s
t
r
i
b
u
t
i
o
n
D
i
s
t
r
i
b
u
t
i
o
n
C
u
m
u
l
a
t
i
v
e

D
i
s
t
r
i
b
u
t
i
o
n
Figure 13. (a) Davidson model. (b) CCIR model. (c) Hata model.
(d) ITU-R P. 529-3 model. (e) FSPL. (F) COST 231. (g) Walcsh-
Ikegami. (h) EGLI.
172 Faruk, Ayeni, and Adediran
Figure 14. Spearmans correlation between measured and models
prediction.
metric; the least frequency counts was found to be 30 dB and 55 dB for
WI and FSPL models respectively and the error spread do not follow
the normal distribution curve.
Figure 14 gives the rank correlation coecient between measured
and models prediction. In terms of this metric all the models perform
better along all the routes except for UNILORIN where the correlation
value is less than 0.5 at p < 0.001. The correlation coeecient of 0.81
(p < 0.001) was calculated along BODE-Saadu route for HATA model
this indicates very strong correlation. However, this metric does not
provide consensus on which model performs best at rank ordering.
Above all, the results show strong correlation between the measured
and prediction path losses for all the routes.
6. CONCLUSIONS
In this paper, we have assessed the tness of nine widely used empirical
path loss models using ve novel metrics to gauge their performance
based on eld strength measurements along six routes that spanned
through the urban, suburban and rural areas of Kwara State, Nigeria.
The performance criteria were based on prediction error, RMSE,
SC-RMSE, normalized error probability density function and rank
correlation. The results show that no single model provides a good t
consistently. The measure of tness is when the RMSE value is 010 dB
in the urban scenario and 1015 dB in the rural scenario. However,
Hata and Davidson models provide good tness along some selected
measurement routes with measured RMSE values of less than 10 dB.
Progress In Electromagnetics Research B, Vol. 49, 2013 173
ITU-R P.1546-4, Walsch, Ikegami, Egli, CCIR and FSPL models
perform woefully, with higher RMSE and SC-RMSE values. Further
results on the error spread as a function of distance along a wider
route revealed that Davidson model gives a better t over Hata this
is perhaps expected since Hata model is only valid for a maximum
transmission distance of 20 km.
The novel feature of this study is that, the work provides detailed
error analysis of the path loss models of which is the rst of its
kind in Nigeria to carry out an extensive analysis of large number
of propagation models using large amount of data set produced from
real time measurements. Other important contributions of this work
are:
1. The provision of the error bounds for the models studied and
introduction of a new matric, i.e., normalized error probability
density function. Other metrics are commonly used when
assessing models performance, but this new metric helped in
studying the error distribution counts for each model across the
link and aid towards judging which model t into Gaussian normal
distribution.
2. Error spread as a function of distance, to examine the impact of
the terrain prole on error as a function of distance. This metric
helps to demarcate the point of divergence between Hata and
Davison models despite they had the same RMSE and SC-RMSE
values for considerable distance, i.e., d < 20 km, and consequently,
this gives an insight of their performance in terms of distance.
3. These bounds will provide guidelines for researchers and practising
engineers in choosing appropriate path loss model(s) for coverage
optimization and interference analysis for wireless devices
operating in the TV band in our environment and also, to predict
TV coverage and keep-out distances for potential secondary users
operation in the TV white space.
However tuning of Davidson model is necessary to minimize the
RMSE values within the acceptable ranges that would cover all the
routes. Minimizing these errors would of course, have signicant
impact on the amount of white space recovery and thus increase
exibility in local spectrum usage and deployments of the secondary
networks in the TVWS.
REFERENCES
1. FCC, Second report and order and memorandum opinion and
order, ET Docket No. 08-260, Nov. 2008.
174 Faruk, Ayeni, and Adediran
2. Kang, K. M., J. C. Park, S. I. Chio, and B. J. Jeong,
Deployment and coverage of cognitive radio networks in TV
white space, IEEE Communications Magazine, Vol. 50, No. 12,
8894, Dec. 2012.
3. Ayeni, A. A., N. Faruk, L. Olawoyin, M. Y. Muham-
mad, and M. I. Gumel, Comparative assessments of some
selected existing radio propagation models: A study of
Kano City, Nigeria, European Journal of Scientic Research,
Vol. 70, No. 1, 120127, Euro Journals Publishing, Inc., 2012.
http://www.europeanjournalofscienticresearch.com/ISSUES/E-
JSR 70 1 12.pdf.
4. Camp, J., J. Robinson, C. Steger, and E. Knightly, Measurement
driven deployment of a two tier urban mesh access networks,
Proceedings of the 4th International Conference on Mobile
Systems, Applications and Services (Mobisys06), 96109, ACM,
New York, USA, 2006.
5. Hata, M., Empirical formula for propagation loss in land mobile
radio services, IEEE Trans. Vehicular Technology, Vol. 29, No. 3,
317325, Aug. 1980.
6. COST 231, Urban transmission loss models for mobile radio in
the 900 and 1800 MHz bands (revision 2), COST 231 TD(90)
119 Rev. 2, The Hague, the Netherlands, Sep. 1991.
7. Walsch, J. and H. L. Bertoni, A theoretical model of UHF
propagation in urban environment, IEEE Trans. Antennas
Propagation, Vol. 36, 17881796, 1988.
8. Ikegmi, F., S. Yoshida, T. Tacheuchi, and M. Umehira,
Propagation factors controlling mean eld strength on urbn
street, IEEE Trans. Antennas Propagation, Vol. 32, No. 8, 822
829, 1980.
9. Egli, J. J., Radio propagation above 40 MHz over irregular
terrain, Proc. IRE, Vol. 45, No. 10, 13811391, 1957.
10. ITU-R P.529-3, VHF/UHF propagation data and prediction
methods required for the terrestrial land mobile services, 1994.
11. ITU-R P.1546-4, Method for point-to-area predictions for ter-
restrial services in the frequency range 30 MHz to 3000 MHz,
2009, available on http://www.itu.int/rec/R-REC-P.1546-4-
200910-I/en [Accessed on 31/01/2013].
12. Anang, K. A., P. B. Rapajic, R. Wu, L. Bello, and
T. I. Eneh, Cellular system information capacity change at higher
frequencies due to propagation loss and system parameters,
Progress In Electromagnetics Research B, Vol. 44, 191221, 2012.
Progress In Electromagnetics Research B, Vol. 49, 2013 175
13. CCIR: http://w3.antd.nist.gov/wctg/manet/calcmodels r1.pdf
[Accessed on 13/11/2012].
14. Davidson, A report on technology independent methodology for
the modeling, simulation and empirical verication of wireless
communications system performance in noise and interference
limited systems operating on frequencies between 30 and
1500 MHz, IEEE Vehicular Technology Society Propagation
Committee, TIA TR8 Working Group, May 1997.
15. Feher, K., Wireless Digital Communications Modulation & Spread
Spectrum Applications, Prentice Hall, Inc., USA, 1995.
16. Phillips, C., D. Sicker, and D. Grunwald, Bounding the practical
error of path loss models, International Journal of Antennas and
Propagation, Vol. 2012, 121, 2012, doi:10.1155/2012/754158.
17. Phillips, C., D. Sicker, and D. Grunwald, Bounding the error of
path loss models, New Frontiers in Dynamic Spectrum Access
Networks (DySPAN), IEEE Symposium on Dynamic Spectrum
Access, 7182, May 2011.
18. Delisle, G. Y., J. P. Lefevre, M. Lecours, and J. Y. Chouinard,
Propagation loss pre diction: A comparative study W with
application to the mobile radio channel, IEEE Trans. Vehicular
Technology, Vol. 34, No. 2, 8696, 1985.
19. Sharma, P. K. and R. Singh, Comparative analysis of
propagation path loss models with eld measured databases,
International Journal of Engineering Science and Technology,
Vol. 2, 20082013, 2010.
20. Rao, T. R., M. S. V. Bhaskara, M. V. S. N. Prasad, M. Sain,
A. Iqbal, and D. R. Lakshmi, Mobile radio propagation path
loss studies at VHF/UHF bands in Southern India, IEEE
Transactions on Broadcasting, Vol. 46, No. 2, 158164, Jun. 2000.
21. Prasad, M. V. S. N. and I. Ahmed, Comparison of some path
loss prediction methods with VHF/UHF measurements, IEEE
Transactions on Broadcasting, Vol. 43, No. 4, 459486, 1997.
22. Achtzehn, A., J. Riihijarvi, G. Martinez, M. Petrova, and
P. Mahonen, Improving coverage prediction for primary multi-
transmitter networks operating in the TV white spaces, 9th
Annual IEEE Communication Society Conference on Sensor,
Mesh and Adhoc Communication and Networks (SECON), 547
555, 2012.
23. Kim, H., H. Sunahara, and A. Kato, Study on environmental
improvement for DTV white space utilization with narrow band
system, International Journal of Network and Communications,
Vol. 2, No. 4, 3846, 2012.
176 Faruk, Ayeni, and Adediran
24. Harrison, K., S. Mishra, and A. Sahai, How much white-
space capacity is there?, Proceedings of IEEE DySPAN, 110,
Apr. 2010.
25. Van de Beek, J., J. Riihijarvi, A. Achtzehn, and P. Mahonen, TV
white space in Europe, IEEE Transactions on Communication
Mobile Computing, Vol. 11, No. 2, 178188, Feb. 2012.
26. Nekovee, M., Quantifying the availability of TV white spaces
for cognitive radio operation in the UK, Proc. IEEE ICC
2009, Workshop, 15, Cognitive Wireless Networks and Systems,
Dresden, Germany, Jun. 2009.
27. Makris, D., G. Gardikis, and A. Kourtis, Quantifying TV
white space capacity a geolocation-based approach, IEEE
Communication Magazine, Vol. 50, No. 9, 145152, Sep. 2012.
28. Abhayawardhana, V. S., I. J. Wassell, D. Crosbsy, M. P. Sellars,
and M. G. Brown, Comparison of empirical propagation path
loss models for xed wireless access systems, IEEE Vehicular
Technology Conference, Vol. 1, 7377, Spring, 2005.
29. Blaunstein, N., D. Censor, D. Katz, A. Freedman, and
I. Matityahu, Radio propagation in rural residential areas with
vegetation, Progress In Electromagnetics Research, Vol. 40, 131
153, 2003.
30. Parson, J. D., Mobile Radio Propagation Channel, Wiley,
Chichester, West Sussex, England, 1992.
31. COGEU, Cognitive radio systems for ecient shar-
ing of TV white spaces in European context, 73
101, Jan. 2010Dec. 2012, available on http://www.ict-
cogeu.eu/pdf/COGEU D7.1%20(ICT 248560).pdf [Accessed
on 10/04/2012].
32. Surajudeen-Bakinde, N. T., N. Faruk, A. A. Ayeni, M. Y. Muham-
mad, and M. I. Gumel, Comparison of propagation models for
GSM 1800 and WCDMA systems in selected urban areas of
Nigeria, International Journal of Applied Information Systems
(IJAIS), Vol. 2, No. 1, May 2012.
33. Shamsan, Z. A., T. A. Rahman, and A. M. Al-Hetar, Point-
point xed wireless and broadcasting services coexistence with
IMT-advanced system, Progress In Electromagnetics Research,
Vol. 122, 537555, 2012.