IEEE TRANSACTIONS ON INSTRUMENTATION A N D MEASUREMENT. VOL. 39. NO. 5 .
OCTOBER 1990 75 I
Measurement and Analysis of the Indoor Radio
Channel in the Frequency Domain
STEVEN J. HOWARD AND KAVEH PAHLAVAN,
Abstract-Using a network analyzer, several experiments for the fre-
quency-domain characterization of the indoor radio channel in the 0.9-
1.1-GHz band are performed. In the experiments, the frequency re-
sponse measurements are taken at spatially distributed locations
throughout the test area by fixing the receiver in a central location and
moving the transmitter to different locations. The experiments were
performed in a high-rise office building and a three-story building with
offices and laboratories. For each experiment, the exponent of the
power-distance relationship and the statistics of the 3-dB width of the
frequency correlation function are determined from the frequency-do-
main data. The approximation to the impulse response of the channel
is obtained from the inverse Fourier transform of the frequency re-
sponse. An empirical exponential relationship between the 3-dB width
of the frequency correlation function and the inverse of the rms delay
spread of the impulse response is derived.
I. INTRODUCTION
R ECENTLY, indoor radio propagation has been stud-
ied extensively for the evaluation of different trans-
mission systems for application in universal portable
phones and wireless local area networks. Radio propa-
gation studies can be performed either in the time domain
or in the frequency domain. The reported wide-band mea-
surements and modeling for the indoor radio channels
have been performed almost exclusively in the time do-
main [ l]-[5]. These measurements determine the channel
impulse response by sending a narrow pulse and by ob-
serving the effect of the channel on the received signal.
The limited measurement in [4] reports the changes ob-
served in the channel frequency response when an object
moves close to the transmitter or the receiver. This paper
presents the results of the frequency-domain measure-
ments in two different buildings, compares the statistical
behavior of the two, and relates the results to the time-
domain statistics.
Coherent wide-band frequency-domain measurements,
presented in this paper, provide magnitude and phase of
the frequency response of the channel. As a result, the
exact time-domain response is also obtained by taking the
inverse Fourier transform of the measured data. In appli-
cations such as channel modeling or performance calcu-
lation, the phase information is necessary. The wide-band
Manuscript received December 13. 1989; revised May 18, 1990. This
work was supported in part by the National Science Foundation under Grant
NCR-8703435 and a Raytheon Company Fellowship.
S . J . Howard is with Raytheon Company. Marlboro. MA 01752.
K. Pahlavan is with the Department of Electrical Engineering. Worces-
ter Polytechnic Institute, Worcester, MA 01609.
IEEE Log Number 9037397.
0018-9456/90/1000-0751$01.OO
SENIOR MEMBER, IEEE
time-domain measurements in [ 11-[5] provide only the
magnitude of the time-domain response. The transmitted
power in the frequency-domain measurement has a con-
stant envelope, as opposed to the time-domain measure-
ment where the ratio of the peak to average transmitted
power is large. This fact allows a larger area to be mea-
sured and reduces the effects of nonlinearities. Also, the
set up for the frequency measurements is easier and the
measurement time is shorter when it is compared with the
time-domain measurements explained in [ 11 and [2].
11. MEASUREMENT
SYSTEM
The block diagram of the measurement system used for
frequency-domain characterization of the indoor radio
channel is shown in Fig. 1. The main component of the
measurement system is a network analyzer that outputs a
swept frequency signal and analyzes the received signal.
The time to sweep the frequency band is 400 ms. The
signal generated by the network analyzer is used as the
input to a 45-dB transmitter RF amplifier. The output of
the RF power amplifier is propagated by a dipole antenna.
The signal from the receiver dipole antenna is passed
through an attenuator and a series of amplifiers with a gain
of 60 dB. The output of the amplifiers is returned to the
network analyzer to determine the frequency and time re-
sponse of the channel. The measured data is then read and
stored by the PC controller for further analysis.
The choice of the 200-MHz band centered at 1 GHz is
for a variety of reasons. The 200-MHz bandwidth in the
frequency domain gives an equivalent resolution in the
time domain of 5 ns, which is what many of the time-
domain systems are capable of producing. We want to
cover the 900-MHz band because of allocation of this band
for indoor transmission, and the many other reported time-
domain measurements are for this band. The components
used in the construction of the measurement system have
a flat response in the 1-GHz range.
Fig. 2 shows a plot of the magnitude and phase of a
typical frequency response H ( f , x ) measured at a location
x , and the corresponding magnitude of the time-domain
response 1 h (7,x ) l obtained from the inverse Fourier
transform. The magnitude of the frequency response in
decibels, the phase of the frequency response in degrees,
and the magnitude of the time response on a linear scale
are shown. The received signal level is adjusted by means
of the attenuators to maintain an approximately 0-dB re-
0 1990 IEEE
 ~

152 I E E E T R A N S A C T I O N S ON I N S T R U M E N T A T I O N A N D M E A S U R E M E N T . VOL. 39. N O . 5. O C T O B E R 1990

RECEIVE
ANTENNA
TRANSMIT

v
I 3WBm Max
ATTENUATOR
1-6440 4MB 3WBm

-n
U
100

ki
pERsoNALc7xmnER 5 0
WrmGplB
rJTEwPLE j
e
-100

900 950 1000 1050 1100

Frequency (YHx)
ceived signal power relative to the transmitted signal (b)
power. The attenuator settings and receiver-transmitter
separation are recorded during the measurement process.
The frequency response consists of 801 complex samples
at a frequency spacing of 0.25 MHz for a frequency span
of 200 MHz, which is centered at 1 GHz. From this fre-
quency response, a time response of 4000-ns duration is
derived. The time response is truncated to show only that
portion with significant energy. The frequency selective
nature of the channel is seen to result in deep nulls at
certain frequencies. The phase is linear for most of the
frequency band, except for those frequencies at deep nulls
where a phase jump is observed. The time-domain re-
sponse illustrates the multipath propagation which causes (C)

the frequency selectivity.
111. DESCRIPTION OF THE MEASUREMENTS
In this paper, the frequency response measurements are
spatially distributed throughout the test area, such as the
floor of a building, by fixing the receiver in a central lo-
cation and moving the transmitter to different locations.
The locations are selected based on the existence or likely
existence of communication equipment for wireless local
area networks. As shown in [6], the maximum rate of
variations in the channel characteristics is below 10 Hz
(time variations of the order of 100 ms) which are caused
by movements close to the transmitter or the receiver; the
acquisition time in our measurement system is 400 ms.
To avoid the variations of the channel during the acqui-
sition time, the surrounding environment is kept station-
ary by preventing movements close to the transmitter and
the receiver. The objective of experiments is to determine
the effect of location and path obstructions on radio-wave
propagation. The results from three experiments are re-
ported in this paper.
Fig. 2. An example of a measurement made with the network analyzer.
(a) The magnitude of the frequency response in decibels. (b) The phase
of the frequency response in degrees. (c) The magnitude of the inverse
Fourier transform of the frequency response.
The measurements were obtained from 128 different lo-
cations in two buildings. The area covered in each build-
ing is on the order of a picocell (less than 50-m radius)
where picocell is the smallest cell size considered for the
future digital cellular portable radio systems. The first set
of measurements were obtained from an office, located at
the 16th floor of a 32-story building in downtown Worces-
ter, MA. The office consists of a central open area sur-
rounded by small offices. A total of 70 frequency re-
sponses were collected. The receiver was placed in a
central location and the transmitter was moved to different
locations for each frequency response measurement. The
measurements are divided into two groups, 46 mobile
(GI) and 24 fixed (G2). Fixed measurements were taken
from locations where data transmission devices such as a
computer terminal existed. Mobile measurements were
 ~

HOWARD AND PAHLAVAN: INDOOR RADIO C H A N N E L I N FREQUENCY D O M A I N 153

taken from locations where data transmission by portable - 1

phone would occur, such as the middle of a room or areas

close to desks.
The third group of measurements (G3) was performed
on the second floor of the three-story Atwater Kent Lab-
oratories at the Worcester Polytechnic Institute. The re-
ceiver was placed in the central computer terminal room.
Measurements from 58 locations close to computer ter-
minals in the same room, the adjacent laboratories, the
power systems laboratory across a hallway, and offices
across another hallway were taken.

IV. MEASUREMENT
ANALYSIS
The main objective of the indoor radio wave propaga-
tion measurements is to determine the radio coverage and
data rate limitations in various buildings [ 1]-[5], [71-[8]. Fig. 3. A scatter piot of the power (dB) versus the distance on a log scale
for experiment G I . Also shown is the line with the minimum mean square
The radio coverage is related to the distance-power rela- error fit to the data.
tionship in the area, and the data rate is limited by the
frequency selective fading multipath characteristics of the
channel. In the time-domain measurements, the data rate TABLE I
limitations are studied by examining the statistics of the VALUESOF 0 1 , d (dB), r , A N D A (dB) DERIVEDFROML I N E AREGRESSION
R
OF POWER (dB) ON 10 log,, OF DISTANCE FOR EACHOF THE THREE
rms delay spread [ 11-[5]. In this section, the statistics of GLOBAL EXPERIMENTS
the 3-dB width of the autocorrelation function of the fre-
quency response are derived and an empirical relation be- EXPERIMENT Q d(dB) r A(&)
G1 2.599 2.979 -0.93 4.29
tween the 3-dB width of the frequency correlation func-
G2 2.463 2.595 -0.91 -5.6
tion and the rms delay spread of the channel is developed.
G3 2.465 4.563 -0.81 -6.75

A. Received Power versus Distance

For a fixed transmitter power (P,), the received power
(P,) decreases with distance ( d ), as
P,(d) = Ad-"
where a is the exponent of the power-distance relation-
ship and A is a constant set by the transmitted power and
the measurement system gain. When the logarithm of (1)
is taken, the linear relationship
dB. The correlation coefficient, r = -0.93, indicates that
decreasing received power is highly correlated with in-
(1) creasing distance. Table 1 gives the value of a , d (dB), r,
and A (dB) determined for the three global experiments.
The a values are all very close to 2.5 with a high degree
of correlation between the average received power and the
distance.

10 log,o[Pr(d)] = 10 loglop] - l o a log,o[d] (2) B. Correlation in Frequency Domain

between power in decibels and 10 logio of the distance The complex autocorrelation function of the frequency
results. For free space, a = 2; values of a obtained from response
time-domain measurements are given in [ 1 1 , [2]. The val- W

ues of a reported for office environments are normally

between 2 and 3, but lower and higher values have been
R(Af, ). = j -w
H ( f , ). H * ( f + A f , ). df (3)

reported for other indoor environments.
In this paper, the average power for each measurement
is calculated by averaging the power over all sample points
of the measured frequency response. Using linear regres-
sion analysis [9], the minimum mean square error
(MMSE) line is calculated for the dependence of average
power (dB) on 10 loglo of the distance for each global
experiment. The slope of the regression line gives the ex-
perimental value of - a.
Fig. 3 shows a scatter plot of received power (dB) ver-
sus distance on a log scale for G 1 and the MMSE line
fitted to the data. The linear regression analysis gives a
= 2.599, and the standard deviation of the average re-
ceived powers from the regression estimates is d = 2.979
is computed for all frequency responses. The 3-dB width
of 1 R(Af, x ) l is a measure of the similarity or coherence
of the channel in the frequency domain, which is in-
versely proportional to the delay spread of the channel.
Fig. 4 shows the cumulative distribution function (CDF)
of the 3-dB widths B,. for each of the three global exper-
iments. The CDF shows the sample probability that B,. is
greater than the value given on the abscissa. The results
for the high rise building, G 2 and G 1, are very close. The
experiment at WPI (G3) has slightly smaller 3-dB widths.
Although the external structure of the buildings are im-
mensely different and the floor plans are also quite differ-
ent, the CDF's and their medians are very similar. Table
I1 gives the statistics of each experiment.
154 IEEE TRANSACTIONS ON INSTRUMENTATION A N D MEASUREMENT. VOL 39. NO. 5. OCTOBER 1990

0 5 10 15 20 25
3dB Width (Wiz)
Fig. 4 . The cumulative distribution function of the 3-dB widths for each
of the three global experiments.

7 10 20 30 50
Delay Spread (ns)
TABLE 11
STATISTICS OF 3-dB WIDTH A N D RMS DELAYSPREAD FOR EACHEXPERIMENTFig. 5 . A scatter plot of the 3-dB widths versus the rms Delay spreads for
each global experiment in a log-log scale. Also shown are the lines with

EXl'ERIMENT I zL1 1 1 I 1
NO.

LIRE-
MENIS
3dB

(MHz)
3dB 3dB
WIDTH WIDTH WIDTH DELAY
MEDIAN MEAN S.D.

(MHz)
RMS RMS
DELAY
SPREAD SPREAD

(MHz)
the MMSE fit to the data.

(ns)
'
.D.
TABLE I11
GI 47 5.75 8.04 7.21 24.16 5.91
VALUES OF 0. d (LOG). r , A N D c DERIVEDFROM LINEAR
REGRESSION
OF
G2 24 5.75 7.04 4.44 24.3 4.5 THE LOG 3-dB WIDTHON T H E LOG R M S DELAYSPREAD
G3 58 5.25 7.41 7.47 28.5 6.17
EXPERIMENT I B I d(log) I I 1 C
G1 I 1.803 1 0.089 I -0.93 I 1974.15
C. Relation Between 3-dB Width and RMS Delay G2 I 2.046 I 0.091 I 4.89 I 4113.86
Spread G3 I 1.935 I 0.070 I -0.96 I 3680.7
For the time responses derived from frequency-domain
measurements, the rms delay spread rrmais computed by
nm cients for each experiments. For the experiments, the 0's
are around 1.9, which is similar to results for the outdoor
mobile radio channels [ 101. The correlation coefficients
for these experiments are more than 89 % .

V . SUMMARY A N D CONCLUSIONS

where Indoor radio propagation measurements in the 0.9-1.1

m GHz frequency band using a network analyzer were re-

-
j T l h ( r ,.)I2
-m
dr
r=
jm( h ( r ,. ) I 2
-m dr
Analysis and simulation shows that the maximum data rate
without diversity or equalization is few percent of; ;7
[7], [8]. Using the inverse relationship between width in
the frequency domain and duration in the time domain, a
relationship of the form B, = Cr;: between the 3-dB
width (MHz) of the frequency correlation function and the
rms delay spread (ns) of the channel in time domain is
determined from a linear regression of the logarithms. Fig.
5 shows the scatter plot of 3-dB width of the frequency
correlation function versus the corresponding rms delay
spread of all three global experiments on a log-log scale
and the lines with the MMSE fit to the logarithm of the
data. For G1, the empirical relationship is determined as
B, = 1974~;:~. Table I11 gives the values of 0,C, the
standard deviation of the logarithm of the 3-dB widths
from the regression estimates, and the correlation coeffi-
ported. The objective of the experiments were to deter-
mine the effect of transmitter location on the received sig-
(5) nal in a central station. These experiments were performed
in an office in the 16th floor of a 32-story building, and
the 2nd floor of the three-story Atwater Kent laboratory
building at the Worcester Polytechnic Institute, Worces-
ter, MA. The measurements in the high rise building were
further broken down to distinguish locations where fixed
equipment such as PC's might reside, and mobile loca-
tions where portable equipment such as cordless phones
might be used. From the frequency-domain data base, the
exponent of the power-distance relationship was found to
be around 2.5 for the all three global experiments. The 3-
dB width of the frequency correlation function was al-
ways less than SO MHz. The empirical relationship be-
tween the 3-dB width of the frequency correlation func-
tion was shown to be exponentially related to the inverse
of the rms delay spread with the exponent ranging from
1.8 to 2.0. Within a picocell in the buildings with vast
difference in exterior, the characteristics of the channel
did not vary significantly when the interiors were similar.
HOWARD A N D PAHLAVAN: INDOOR RADIO C H A N N E L IN FREQUENCY DOMAIN
ACKNOWLEDGMENT
The authors would like to thank R . Ganesh for his help
during the measurements.
REFERENCES
[ l ] K. Pahlavan, R. Ganesh. and T. Hotaling, Multipath propagation
measurements on manufacturing floors at 910 MHz, Electron. Lett.,
vol. 3, pp. 225-227, Feb. 1989.
[2] A. M. Saleh and R. A. Valenzuela, A statistical model for indoor
multipath propagation, IEEE J . Select. Areas Cotnmun.. vol. SAC-
5 , pp. 128-137, Feb. 1987.
[3] D. M. J . Devasirvatham, Time delay spread measurements of wide-
band radio systems within a building, Electron. Lett., pp. 949-950.
Nov. 1984.
[4] A. A. M. Saleh, A. J. Rustako, Jr., and R. S . Roman, Distributed
antennas for indoor radio communications. I Trans. Comrnu-
nications. vol. COM-35, pp. 1245-1251. Dec. 1987.
755
151 R. Ganesh and K. Pahlavan. On the modeling of fading multipath
indoor radio channels. in Proc. I GLOBECOM 89. Dallas. TX,
NOV.30, 1989, pp. 1346-1350.
[6] S . J. Howard and K. Pahlavan, Doppler spread measurements of the
indoor radio channels, Electron. Lett., vol. 26, pp. 107-109. Jan.
1990.
[7] T. A. Sexton and K. Pahlavan, Channel modeling and adaptive
equalization of indoor radio channels. lEEE J . Selecr. Areas Com-
mun., vol. SAC-5, pp. 128-137, Feb. 1987.
[8] S . J . Howard and K . Pahlavan. Performance of a DFE modem eval-
uated from measured indoor radio multipath profiles, in Proc.
ICC90. Atlanta, GA. Apr. 16-19, 1990.
191 G. W. Snedecor and W. G . Cochran. Statisrical Methods. Ames, IA:
Iowa State University Press, 1976.
[IO] A. S . Bajwa and J . D. Parsons, Large area characterisation of urban
UHF multipath propagation and its relevance to the performance
bounds of mobile radio systems, Inst. Elerr. Eng. Proc., vol. 132.
pt. F. no. 2, pp. 99-106. Apr. 1985.