A Simple Approach To A Statistical Path Loss Model For Indoor Communications
A Simple Approach To A Statistical Path Loss Model For Indoor Communications
A Simple Approach To A Statistical Path Loss Model For Indoor Communications
d"v (fr)2
r 4d (1)
where PRAD is the radiated power; d, the dstance in meters between transmitting and receiving antennas,
and A the wavelength. n is the exponent of the distance and is assumed a random variable in this model. In
small or local areas a few wavelengths in radius, n varies randomly, its mean and standard deviation being
larger at distances up to about 20A from the transmitting antenna.- At larger distances, the local means tend
to be nearly constant in a given environment under the same propagation conditions. This characteristic in
the behavior of n was observed for various propagation conditions in different buildings: line-of-sight
(LOS), shadowing (NLOS) and total obstruction between the transmitting and receiving antennas by walls
and floors (OBS), and suggests that, in the range of distances encountered in indoor communications, only
one type of statistics can be used to describe path loss.
2. Theory
In (1) the received power, P, can be assumed random and expressed as:
Where Po1 is the free space value of the received power and <ap > is an attenuation (or gain) factor,
dependent on propagation mechanisms which, in general, are not possible to quantify individually, and
include directive gains of antennas, reflection, diffraction, etc. The bracket notation is used to stress the
random nature of the variable. From (1) it can be seen that:
-617-
(aP) =d( 2 (3)
is the excess attenuation referred to free space conditions. Using dB forn, the path loss can now be written
as:
L, =JO(n)log(d) + K (4)
where K is the free space attenuation at 1 m. The value ofn can be obtained directly as:
where, in actual conditions, PRA is known and the sanples of the received power, .PR > are obtained
through measurements in the particular environment. The model given by equation (4) is formally equal to
other models previously reported in the literature [1]-[3], the main difference is that here, the exponent of
distance is treated as a random variable and no further variables are introduced, nor assumptions made
about its statistical behavior.
3. Experimental procedure
Various experiments were made in different buildings of the University of Cantabria, Spain. One of them, of
simple geometry with open areas and straight aisles. The other, of complex geometry, with narrow aisles
occasionally curved, and with numerous obstructions by walls, columns and furiture. Measurement
environments were chosen for well differentiated propagation conditions: line-of-sight (LOS), non line of
sight (NLOS), mainly shadowing, and total obstruction between transmitter and receiver (OBS).
Additionally the effect of depolarization was also measured. All experiments were made with a 10 dBm
unmodulated carter at 1.8 GHz using an REF signal generator as transmitter and a spectrum analyzer as
receiver at distances ranging between 1 and 32 m. Transmitting and receiving antennas were X/4 vertical
monopoles at 2 m and 1.5 m height respectively. A VJ2 rotable dipole was also used for reception, thus the
copolar and crosspolar powers of the received signal could be easily measured at the sane sampling points.
The measurement procedure was mainly manual, and devised in order to make it easily applicable in
practice, bearing in mind that it is essential that signal measurements should be made in a manner that
permits the basic parameters to be extracted with an appropriate level of accuracy [4]. Thus, the sampling
interval, i.e. the distance between individual samples, distance between local areas, and minimum number of
samples per local area, must be properly defined. An empirical approach was followed with this purpose.
Local areas were defined as squares of maximum area of 1 n2, with a minimum separation between them of
20A and several sets of measurements were taken over the same local areas, with sampling intervals of 1,
2.5 and 5 cm. Differences in the mean received power were within ±1.5 dB for I and 2.5 cm intervals, and
in the order of ±5 dB for 2.5 and 5 cm intervals. A maxuium sampling interval of 2.5 cm was considered
adequate for practical purposes.
In order to establish the minimum number of samples required per local area, a similar approach was
followed. Various sets of measurements were made over the sarne local areas in LOS and NLOS conditions,
taking first 500 samples, then 100 and finally several sets of 50 samples. In all cases, the errors in the mean
values of the exponent were less hn 5%/o, and of 7.5% in the standard deviations. As a consequence, a
minimum value of 50 samples per local area was considered sufficient. Such figure is in good agreement
with the criteria suggested by other authors [5], [6], and with those drawn theoretically [7]. The distance
between samples, as well as the number of samples per local area, results in a very simple measurement
procedure to be used by practicing engineers who need to make only a few measurements in order to extract
the basic paraneters to get good idea of path loss behavior, and who do not need to have a deeper
knowledge of channel dynamics. Such knowledge is sufficient in many practical applications where the main
-618-
objective is to dimension the power of a communications system and provides reliable information about the
fading margins required in a practical design.
LOS measurements were performed in corridors and open areas, at distances between 1 and 32 m. NLOS,
between perpendicular corridors, corridors into open areas and open areas partially obstructed by clutter at
distances up to 25 m and, finally OBS measurements were made with the transmitter inside a closed room
and the receiver in separated rooms and corridors at distances up to 18 m from the obstructing wall.
Measurements with transmitter and receiver in different floors were also perfonned. In all cases,
transmission was vertically polarized, and the transmitter kept fixed for each set of measurements while
moving the receiver. Finally, polarization measurements were made taking, at each measurement point, two
samples of the received power, one with the receiving antenna vertical and then, rotating it to an horizontal
position. Measurements were made in 73 local areas in the various environments and for the propagation
conditions already mentioned.
4. Statistics of the exponent
The statistical behavior of either received power, envelope amplitude, or path loss can always be described
by unimodal and asymmetrical probability density functions (pdf), and it is assumed that the statistics of the
exponent of distance should behave in the same way. However, since the relationship between n and the
power is not linear, the function that best describes n will not, necessarily be the same as that of the power,
and it cannot be assumed a prior that n will follow distributions such as Rayleigh, Rice or log-normal.
Fitting of n to these functions was investigated with acceptable results, however, large deviations were
observed in a significant number of cases, particularly at the tails. Such deviations would lead to very
pessimistic predictions of fade margins, unless some corrections are introduced. Several empirical
corrections were attempted to obtain better fittings to actual data, however, no general rule was found for a
correction that worked in all cases. Therefore, fitting to other distributions was investigated, in particular
gamma [8], Weibull [9] and Nakagami [10], with remarkably better results. In most cases, differences in
fitting to these functions were only marginal, with the gamma distribution producing smaller errors and
higher correlation coefficients between the theoretical function and the distribution of the samples. It must
be said that the particular distribution functions used to fit the experimental data were used only as
mathematical tools capable of describing the exponent behavior, and no attempt was made to establish a
relationship between their paraneters and the physical process. Fitting of data to probability density
functions used the mean and standard deviation of the samples as input variables; the resulting cumulative
distributions (cdf) being easily obtained from them. Goodness of fit was measured in terns of the rms error
in pdt's between the theoretical fiuction and experimental data, cdf's, and correlation coefficients in each
case. Fitting errors with gamma distribution were in the order of 0.025 to 0.1 for pdf's, and between 0.01
and 0.068 for cdfts, whereas correlation coefficients were between 0.75 and 0.99 for pdfts and between
0.975 and 0.9995 for cdft's. Similar figures were observed with Weibull and Nakagami distnrbutions.
In all cases investigated, two distinct regions of exponent behavior can be identified. At distances up to
about 20k from the transmitter in LOS and OBS conditions, and from the scattering edge in NLOS, means
and standard deviations are higher and decay, approximately, in exponential form to nearly constant values,
with small fluctuations, at distances greater than, approximately, 20k. Typical values for LOS conditions in
corridors are shown in Table 1.
Table 1
In the near region (d < 204), values of n as high as 8.4 and as low as 0.2 were observed in local areas at 2
m distance, with standard deviations about three times as large as those observed in the distant region (d >
20X), where maximum and minimum values of n were 4.1 and 1.2 respectively. The terns near region and
distant region used before are arbitray, and not necessarily related with the concepts of near (induction)
field and far (Fraunhofer) field and must not be confused with them. No attempt is made here to relate the
observed multipath effects with such concepts. It can be said that the measurements made at distawes
-619-
4
smaller than 20k from the transmitting antenna do not reflect actual propagation conditions in large area,
and from a practical point of view, if values of the exponent in this region are considered, a pessimistic
coverage prediction will result. However, attention must be paid to NLOS (shadowing) cases where, in the
vicinity of diffracting corners, the received power follows a similar behavior than that in the neighborhood
of the transmitting antenna. In practical situations these conditions must be properly evaluated for particular
cases. In Table II, the statistical parameters of the exponent are presented for the various experimental
cases. Such values correspond to samples in the distant region defined before.
Table 2
Mean values of the exponent are in reasonable agreement with those reported by others [11], [12]. Standard
deviations are environment-sensible, being higher where the amount of furiture, columns and objects is
larger. This effect has direct consequences on the fade margin as can be seen in figure 1, where the gamma
cumulative distribution for two LOS cases in comrdors in different buildings is shown. The mean values of
the exponent are almost equal (1.96 and 1.97), however the standard deviation is larger in building 1. From
the figure it seems clear that the second building offers better LOS propagation conditions. For the cases
shown in figure 1, 99% of values of the exponent will be below 3.75 for building 1, and below 2.6 for
building 2 and, for a distance of, say, 100 m, the excess path loss with respect to free space conditions will
be 22.8 dB for the first building and 7.8 dB for the second. The differences in the expected path losses in
worst conditions (99% of values of the exponent in this example), are due only to the different standard
deviations, since mean values are approximately equal.
Figure 1
Similar plots can be produced for other environments and propagation conditions. As mentioned before, the
gamma distribution function fits very well in general, to experimental data, and equation (4) can be easily
used to describe the statistical behavior. Since the mean value of the exponent is assumed constant in the
range of distances of interest, there is no need to specifically describe the large area behavior with this
model. Furthermore, all sets of samples under the same propagation conditions in a given environment can
be grouped in a larger set whose parameters adequately describe the path loss behavior. A remark must be
made for the case of total obstruction (OBS) between antennas: a total obstacle introduces a fixed
attenuation which, in general, is not known and causes that the exponent after the obstacle be distance-
dependent. Such dependence has not been found significant in the range of distances of the experiments,
except at short distances from the obstacle. However, in this situation, the following expression can be used
for the exponent:
fl(d) - LOBS
n)= no +-JlOlog(d) 6
(6)
where n(d) is the mean value of the exponent after the obstacle, nO is its value in the region before the obstacle, LOBS
is the mean of the attenuation introduced by the obstacle, and d is the distance from the obstacle to the measurement
point. Loss can be easily obtained with the above expression, through measurements in the regions before and after
the obstacle.
Polarization effects [13]. In the indoor propagation environment, electromagnetic waves suffer
depolarization as a consequence of multiple scattering. The amount of depolarization was measured, as
described in Section 3, in most of the experiments performed in building 2, and the statistical behavior of
the exponent for copolar and crosspolar components analyzed. The results are resumed in Table 3 in terms
of the exponent values, as well as path loss difference between copolar and crosspolar components.
Table 3
-620-
Values in Table 3 suggest that the amount of depolariztion depends strongly on clutter in the environment,
being greater when LOS and OBS conditions prevail. Crosspolar components can also be characterized
with the gamma distribution function, and it is interesting to notice that, in the experiments performed,
observed standard deviations are very similar for copolar and crosspolar components. In the worst
conditions observed, the received crosspolar component was higher than the copolar in about 35% of the
samples, which strongly suggests the convenience of polarization diversity in such in such circumstances.
6. Conclusions
A very simple model in which the exponent of distance is considered a random variable has been
investigated to statistically characterize path loss at 1.8 GHz in indoor environments. Such model can be
used by working engineers who, through a few measurements, can obtain the basic infornation needed to
estimate the power budget of communications systems in a particular environment, and who do not require a
deeper knowledge of channel dynamics. The exponent can be descnrbed with a gamma distribution function
through which, the necessary information about fading can be extracted. Since the behavior of the exponent
is fairly constant in large areas, only one distribution function is necessary to characterize it. Polarization
effects were also measured and strong depolarization was observed in cases of shadowing (NLOS), and
total obstruction in the signal path, suggesting the convenience of polarization diversity in such cases.
References
1. Motley A.J. & Keenan, J.M.P. "Personal communication radio coverage in buildings at 900 MH[z and
1700 MHz". Electron Lett. Vol 24, No 12, pp. 763-764, June 1988.
2. Radiowave propagation model document. COST23 1 TD(93)/13. Barcelona, January 19-22, 1993.
3. Rappaport, T.S. - Wireless Communications. Principles and Practice. IEEE Press - Prentice Hall PTR.
1996.
4. "Measurements and modeling recommendations". IEEE Trans. on Vehic. Techn. Vol. VT- 37, No 1, pp.
4546, Feb. 1988. Special Issue on Mobile Communications.
5. Lee, W.C. "Estimate of local average power of a mobile radio signal". IEEE Trans. on Vehic. Techn.
Vol. VT-34, No 1, pp. 22-27. Feb. 1985.
6. Parsons, J.D. The Mobile Radio Propagation Channel. Pentech Press. London. 1992.
7. Kendall, M. & Stuart, A. The Advanced Theory of Statistics. Vol 1. Distribution Theory. 4th. Ed.
Charles Griffin & Co. Ltd. 1977.
8. CCIR Report 1007-1.
9. Ireson, W.G. Reliability Handbook. McGraw Hill Book Co. New York, 1966.
10.Nakagami, M. "Te m-distribution-A general formula of i ty distribution of rapid fading".
Pergamon Press. Symposium Publications Division. London, 1960.
1 I.Alexander, S.E. "Characterising buildings for propagation at 900 MHz". Electron. Ien. Vol. 19, No 20.
p. 860. Sept. 1983.
12.Bach Andersen J., Rappaport, T. & Yoshida, S. "Propagation measurenents and models for wireless
conununications channcls". IEEE Comm. Mag. Jan 1995. pp. 4249.
13.Per=-Vega C. and Garcia Garcia J.L. "Polarisation behaviour in the indoor propagation channel".
Elcctro. Leo Vol 33, N 10. pp. 898-899, May 1997.
-621-
Table 1. Typical values of the exponent (n) and its standard deviation (a.,) for LOS conditions in corridors.
distance 2 4 8 16 32
n 2.49 2.21 1.94 1.96 2.05
an 1.26 0.81 0.47 0.46 0.38
Table 3. Statistical parameters of exponent and path loss difference for copolar and crosspolar components.
-622-
Figure 1. Gamma cumulative distributions for LOS conditions in corridors in two different
buildings.
cdf(n)
-623-