IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. VT-21, NO.

1, FEBRUARY 1972 27

A Power-Spectral Theory of Propagation

in the Mobile-Radio Environment

Abstract-The statistical properties of radio propagation between a ANTENNA

BEARING
DIRECTION OF A T Y P I C A L
COMPONENT WAVE
mobile unit and a base-station terminal arederived. The power spec-
trum of the transmission coefficient in the multipath medium is used
to determine probability distributions of amplitude and phase, correla-
tions of fields versus time and space at mobile and base stations, level-
crossing rates and durations of fades, and random frequency modula-
tion. Duality between the power spectrum and density of time delays
*:jl-'i---- VEHICLE
DIRECTION OF V
SPEED,

is shown. The correlations versus frequency and the coherence band-

width then follow from the density of time delays. The performance of 'MOBILEANTENNA
standard diversity systems is then predicted.
There is a review of results previously presented by Clarke [ 11 , how-
ever, thederivations given herein utilizeexpressionsof thepower TRANSMITTED. cos w l : & [ e J Y ' ]
spectrum rather than expressions of the component waves. The power-
RECEIVED.~[T(f,t)eJW']
spectral approach, used throughout, allows directapplication of pre-
vious statistical analyses, particularly thoseof Rice [2]. f=f,cosa W H E R E f m : Kx

I. INTRODUCTION

I N A TYPICALmobile-radiosituationthepropagationbe-
tween base station and mobile stationis not only by a direct
line-of-sight route, but via many paths, largely by way of scat-
tering by reflections from or diffraction around buildings and
terrain. Thus the signal received by the mobile at any point
would consist of a large number of generally horizontally trav-
elling uniform plane waves whose amplitudes, phases, andangle
will concentrate on the statisticalnatureoftheshort-term
of arrival relative to the direction of vehicle motion (Fig. 1)
are random. These plane waves interfere and produce a vary- variationsandthetechniquesofcommunication in their
ing field strength pattern with minima and maxima spaced on presence.
the order of a quarter-wavelength. With the short wavelengths The majority of Sections I1 and I11 is a review of results pre-
encountered at UHF and microwave frequencies, the mobile- sented previously by Clarke [ I ] . Themethodofderivation
receiversignal fades rapidly and deeply as the mobile station applied herein, however, uses Rice's [ 2 ] formulas for the first-
moves throughtheinterferencepattern. By reciprocity,the and second-order statistics in terms of moments of the power
base-station receiver experiences the same rapid fading as the spectrumratherthanreturningtotheexpressionofthe re-
mobile transmitter moves. ceived process as asumofincidentplane waves. Forex-
As the mobile station moves from one location to another, ample, the cross correlations of (39), (40), and (41) [(41) is a
different scatterers and terrain change the plane waves incident new result], were obtained by noting that the signals received
on the mobile receiver. Thus, superimposed on the rapid fad- by different antennas result from passing the same power spec-
ing, are slow variations in the average field strength of the in- trumthroughdifferentfilters so thatspectral analysis for
terference
pattern.Theselong-termvariations,
sometimes multiple-terminal networks [3] applies.
Correlation versus
referred t o as shadow effects, are usually handled in system
frequencyfollowsdirectlyfromcorrelation versus timeby
recognizing thedualitybetweenpower-spectrumanddelay
design by increasingmarginsintransmitterpower andco-
distribution. Section 11-C presents new results for correlation
channel interference to cope with worst-case conditions.How-
betweenbase-stationantennas,andtheexpressionfor level-
ever, the deep rapid fades associated with the short-term varia-
crossing ratesfordirectionalantennas (53) andthepower
tions radicallydegrade communicationqualityatUHFand
spectrum of random FM are extensions of Clarke's results.
microwave frequencies, and are best minimized by using fre-
quency modulation and diversity techniques. In this paper we 11. TRANSMISSION
COEFFICIENT
A convenientwayofcharacterizingamobile-radiocom-
Manuscript received April 9, 197 1.
Theauthor is with the Crawford Hill Laboratory, Bell Telephone municationchannel (MRCC) is intermsofitstransmission
Laboratories, Inc., Holmdel, N. J. 07733. coefficient. The transmission coefficient represents the ampli-
28 IEEE TRANSACTIONS FEBRUARY
TECHNOLOGY,
ON VEHICULAR 1972

tude and phase of the received signal when a unit amplitude spread around the transmitted frequency. This received signal
continuous wave (CW) signal is transmitted. In complex conforms to Rice's model of narrow-band Gaussian noise [ 2 ] .
notation, Rice has computedthestatistical behaviorof narrow-band
transmitted: Gaussian noise in terms of its power spectrum.
The power spectrum of T(f,,t)/& (the output of a spec-
cos (of)
= Re [e'"'] trum analyzer, shifted so that f, coincides with 0 Hz) can be
received: expressed in terms of the strength of the signal received with
eachDoppler shift.From Fig. 1, theDoppler shift f i n fre-
Re [ T ( f ,t ) e j w f ] quency from the carrier frequency f,, for a signal arriving at
where an angle a relative to vehicle motion, is

f w / 2 7 ~frequency
, of transmitted CW signal, Hz f = f" cos a (1)
t time, s where
T ( f ,t) complex transmission coefficient.
f
V
= -
In an MRCC, the transmission coefficient is a function of time
andfrequency;that is, the MRCC is time varying and
" x
dispersive. is the maximum Doppler shift at the vehicle speed Vand car-
The rapid fading, already described, as the mobile unit moves rier wavelength A. We denote by p(a)da the total power of
through the interference pattern is represented by decreases in the plane waves arriving within da of the angle a (that would
the magnitude of T(f,t) as time isvaried. Variations in the be received by an isotropic antenna with the correct polariza-
phase of T ( f , t )as time is varied is often termed random FM. tion) and by the unit polarization vector of these plane waves
Variations in amplitude and phase of T(f,t) as frequency is ep(a) (assume da small enough that ep(a)does not vary within
varied are called frequency selectivefading and phase distor- da). We denotethe power gain of the mobile antenna by
tion of the channel, respectively. G(a - ,$)(gain relative to an isotropic antenna) and the polar-
In statisticaltermsT(f,t) is acomplexstochastic process ization of theantennapattern by eg(a- E), where ,$ is the
w h c h is quasi-stationarywith respect toand t. By quasi- azimuthal bearingof theantennabeam.Notethat we have
stationary, we mean the joint probability density of {Ti(fi, ti)}, assumed all waves are travelling horizontallyandthatthe
i = 1, * . * ,N , remains unchanged if the set of {fi}are all trans- antenna is pointing in ahorizontaldirection. Thisis suf-
lated by the same frequency shift so long as all {&} remain ficientgeneralization tohandlethemajority ofcases. Thus
withn the stationary range AF,and if the set of { t i }are trans- the power contributed to the received signal by plane waves
lated by the same time shift so long as all {ti} remain within arriving withinthe angle da is the power arriving in that
thestationary range AT, A F is determinedbythe range of angularinterval that would be received by an
isotropic
frequency over which the reflective and diffractive properties antenna of the same polarization times the antenna gain times
of the terrain and buildings remain approximately unchanged. the square of the parallel fraction of polarization:
A T is determined by how long the effectsof terrain and build-
ings and the vehicle velocity remain approximately unchanged . eg(a-
p(a)da[ep(a> $11 c(a- O . (3)
(e.g., the time it takes a vehcle to travel from the center of a From (l), we have
block to an intersection in a city).
The quasi-stationary propertiesof T( f, t ) allow us t o describe
its statistics in simple terms. In turn, the statistical behavior
[
dcu = - fm df: (4)

of T ( f ,t ) determines the performance of various diversity and Thus, inserting (4) into (3) and combining the two angles (&a)
modulation systems that might be proposed for mobile-radio which give Dopplershift f , the power-spectral density of
communication. TU,, t)iv'T is

A . Power Spectrum of T( f,t) where

Fora given transmittedfrequency,say, f,, the received
signal T( f,, t ) as mentioned before is a result of many plane
waves, each shifted in frequency by theDopplershiftap-
propriate to the vehicle motion relative to the direction of the For example, assume the vertically polarized parts of the in-
plane wave. Thus, in a typical case, the received signal consists cident waves are uniformly distributed with respect to azimuth
of a large number of sinusoids of comparative amplitude and angle a. Thenthe following vertically polarized antennas re-
random phase, whose frequencies are confined to the Doppler sult in the spectra shown in F i g . 2 and 3:
GANS: ENVIRONMENT
IN MOBILE-RADIO 29

3 ) vertical loop in plane of vehicle motion, G(a)= 312 cos’ cy,

4) beam antenna directed perpendicular to vehicle motion,

G(a) = Go, if ja I < /3/2; C(a) = 0, otherwise,

otherwise
5 ) beam antenna directed along vehicle motion,

I1 d(f) ,, otherwise.
G(al=+cos‘a
(1 0)
One may thmk of thespectraobtainedwithdifferentan-
tennas as different filter shapes through which the same white
noise is passed. Following Rice, we may now compute much
( C)
of the first- and second-order (with respect to the time vari-
Fig. 2. Typicalpowerspectra.Assume: p ( a ) [ e p ( a ). eg(a - $)I = 1.
(a) Verticalmonopole. (b) Verticalloop in planeperpendicular to able) statistics of T ( J t) in terms of the moments of the power
vehicle motion. (c) Vertical loop in plane of vehicle motion. spectrum.

B. Probability Density of T(J t)

ANTENNA
BEARING We may divide the transmission coefficient into its in-phase
‘Q$\ Go IF I o - F I < B / Z
0 OTHERWISE
T, and quadrature parts T,:
V TC(0 = Re [TWl Ts(t) = Im [T(t)l (1 1)
(a) where we have dropped fargument
from
theto simplify nota-
tion. Rice has shown that for a narrow band of sinusoids, as

pm/ ;
t

-fm

-fmslngi
’ 0 1

I
,<

f,sln;

-
1
,

[=go”
*
f
J(f1.

I fm
GO

0 OTHERWISE
IF

I fml
<sinB
experienced in mobile-radio reception, the in-phase and quad-
rature
components
anyat given time t are independent Gaus-
sian random variablesprobability
the
with density

f ( x )= x [- 21
1
exp (1 2)

fl [
2GO
IF I > L > l-cos- where
ti_
..__ d(f
$if)=
f,J= fm
i‘\ __a,

0 OTHERWISE
x = Tc(t) or X = Ts(t)
f
0
-fm
ifm
fm(l-cosT)

E.0’
-
(b)
bo = lfm
fm
S ( f ) df= mean received power (1 3 )

so that the probability that x lies between x and x + dw is

Fig. 3. Powerspectraforbeamantenna.(a) Beam antennapattern.
(b) spectra.
Power f ( x )dx. n e cumulative
probability
distribution
(probability
that x is less than X)is therefore

1) vertical monopole, G(a) = 3 / 2 ,

3
(’ = f m \ / I - ( f / f m 12 where the error function is defined by
2 ) vertical loop in planeperpendicular
G(a) = 312 sin’ a,
2
to vehicle motion,
erf(y) = 1
6
ly (- exp $)du.

(7)
The Gaussian distribution is plotted in Fig. 4.
30 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, FEBRUARY 1972

Fig. 4. Probability density and cumulative distribution of in-phase and quadratureparts of transmission coefficient.

These probability densities are plotted in Fig. 5 . Two impor-

tant properties of any probability density f ( x ) are its mean m
and variance u2 defined by

The means andvariances of the preceding random variables are

given below.
~~~ ~~~~

Variable
Random Square
MeanMean Variance

R (2 - ;)bo
8:arctan(+)
e 0
bl.[T(t)exp(jwt)]=R(t)coo[wtt8(1)]

(b)
The deep fades encountered in mobile radio are a property
Fig. 5. (a) Rayleighprobabilitydensity of amplitude.(b)Uniform
of the Rayleighdistribution. If oneaskedwhat level one
probability density of phase.
could expect to exceed 99 percent of the time, the amplitude
distribution (18) predicts that R = O.l&; i.e., 1 percent of
Rice has also shown that the amplitude
the time the amplitudecan be expected to be more than 20 dB
R(t) = I T ( t )I = dT: (t)+ T,'(t) (1 6 ) below its rms value.
If there is a plane-wave component much stronger than the
and phase
average component, the received signal will consist of a sine
e(t) = arctan ($) wave plus a narrow band of Gaussian noise. For example, the
mobile unit may be in direct line of sight with the base station,
so that the sinusoid corresponds to the direct-path contribu-
are independent random variables. The probability density of tion and the narrow-band Gaussiannoise corresponds to the
the amplitude is Rayleigh, contribution of nearby scatterers. The probability density of
the amplitude then becomes the Rice distribution

Io, otherwise

F ( R ) = 1 - exp (- 5) for a sine-wave component Q cos 2nfst plus scattered waves

withmeanpower bo, where f , can be anyDoppler-shifted
frequency and Io is the modified Bessel function of the first
while the phase is uniformly distributed from -'IT to +n, kind with index zero.

for -II < 0 < II C. Correlation

(19) Thecorrelationof the transmissioncoefficientversustime
0, otherwise.
used
distance
and
be
may t o determine
performance
of di-
GANS: PROPAGATION IN MOBILE-RADIO ENVIRONMENT 31

where

The autocorrelation of the amplitude squared is

W ( r ) R 2 ( t + T ) ) = 4bi(l + p z ) . (3 1)
Since the mobile unit is moving, a time shift of 7 implies a
translation { = VT. Thustheprecedingcorrelationsmay be
considered as correlations versus spatial separation by setting

As an example, we consider the foregoing case of uniformly

scattered waves and the vertical monopole (denoted E,, since
it responds to the vertical electric field component), the loop
in theplaneperpendicular to vehicle motion(denoted H,,
since it responds to the x component of magnetic field), and

d7)= Lfm fm
S(f) cos 2nf7 df
the loop in the plane of vehicle motion (denoted H y , since it
responds to the y componentofmagneticfield).Fromthe
preceding formulas, we have (k = 2n/h)

h(7) = Lfm fm
S(f) sin 2nf7 df (24)
vertical monopole:

Notethatthecorrelationsfollowdirectlyfromthe power
spectrum S ( f ) by the Fourier transforms of(24).Thefore-
going impliesthe following moments (7 = 0):
( T Z ) = ( T i ) = b o ,( T c T s ) =0 perpendicular loop:
(T,T,')=-(T;T,)=bl,
(T,T;)=(T,T,')=O 1
- hH,(7) = 0
( T i z ) = ( T i z )= b z , ( T;
T;) = o (25) bo
where

b, = ( 2 4 " i_, s ( n f n
fm
df. (26) =Jo(W +Jz(W (34)
parallel loop:
We note that in the preceding, when S ( f ) is symmetrical about
its center frequency, h(7) and b,, with n an odd integer, are
zero.
If we define

then the autocorrelation of the amplitude becomes [4] where J o and Jz are the Bessel functions of the first kind of
order 0 and2,respectively.Thesecorrelationfunctionsare
plottedin Fig. 6. As seen fromthefigure,abroad-beamed
antennainauniformlyscattered field becomesrapidlyde-
where E is the complete elliptic integral of the second kind. corre!ated, dropping to a small correlation within a distance of
The autocorrelation of the phase is determined from the fol- a half-wavelength.
lowing integral: By using power-spectral analysis for multiple-terminal linear
systems, one may compute thecross-power spectrumSTA T , ( f )
between two antennas, A and B, from their individual power
spectra [ 3 ] . Thecross-power spectrumoftwostochastic
processes is defined as theFouriertransform of their cross
+P(n- cos-1 correlation. Therefore, the cross correlation of the transmis-
dx (29)
(1 - P 2 ) 3 ' 2 sioncoefficient can be obtainedfromthe cross-powerspec-
32 TRANSACTIONS
IEEE ON VEHICULAR TECHNOLOGY, FEBRUARY 1972

BEARING OF ANTENNA B
BEARING OF ANTENNA A f

t k = -a
21

10 OTHERWISE

0
- fm t
I
OR f /A
2

Fig. 6 . Normahzed correlation of real part of transmission coefficient

Fig. 8. Two overlapping beam antennas.

for vertical monopole E,, perpendicularloop
loop H y .
H,, and parallel At 7 = 0, or the same point in space, { = 0, we see that E,, H, ,
and H, are all uncorrelated (and hence independent,since they
are complex Gaussian random variables). Thustheybecome
desirable components to useas separate branches in a three-
branch diversity system.
Consider twobeamantennas w h c h overlap from 0 1 ~to
01, as in Fig. 8. The voltage on antenna B relative to that on
antenna A due to a component with Doppler shft f is zero
unless 01 = cos-' ( f / f m ) lies within CYAand aB, where it is given
by the square root of the ratio of the antenna gains

-fmt OR %/x (41) otherwise,

Fig. 7. Normalized cross correlation of E, and H y .
trum by its inverse transformation times two. As an example,
we will compute the cross correlation of H y and E, :
S E ,=HsyE(*f )E , ( n X * ( f )
where SE,E,(~)is the power spectrum of E, given in (6) and
K(nis the ratio of Hy to E, for a given Doppler shift f (i.e.,
a plane wave at an angle 01 = cos-' ( f / f m ) )
1 I f
X(f)=;cosa= --
77 f m
where 77 is the free-space plane-wave impedance. Thus
from the Fourier transform inversion (plotted in Fig. 7)
671
(E,*(t)Hy(t+ 7)) = - j J 1 (2nfm 7) =
77
Similarly, we can show that
(E,*(t)H,(t + 7)) = 0 = (H,*(t)Hy(r t 7 ) ) .
Thus the cross correlation ( T j ( t )TB(t + 7)) is the same as
theautocorrelation of abeamantennawithangularwidth
O~B - CUAand gain d-.
D. Correlation at the Base Station
(36) By reciprocity, the transmission coefficient for transmission
from the base station and reception at the mobile unit is the
same as for transmission from the mobile unit and reception
by the base station. Thus the power spectrumandauto-
correlation versus time are identical for signals received at the
mobile unit and at the base station. However, our method of
relatingspatialcorrelation to timecorrelationassumed the
(37)
base station f k e d and the mobile unit moving with velicity V
in thedirection of the spatialseparationinquestionatthe
mobile unit.Therefore, t o determinecorrelation versus dis-
tance at the base station, we mustconsider the mobileunit
fixed and the base station moving as in Fig. 9.
The main difference between the mobile and base stations is
that the scatterers are mainly in the vicinity of the mobile unit
because the base station is mounted up andawayfrom
scatterers. As aresult,theplane wavesarrivingat the base
7) station from the mobile are restricted to a narrow azimuthal
angular spread. If oneassumesan omnidirectionalantenna
(vertical monopole) on the mobile unit and a uniform ring of
(40) scatterers surrounding the mobile unit at a radius a, then the
 -
GANS:PROPAGATION IN MOBILE-RADIOENVIRONMENT 33

MOBILEUNIT

y,

RING OF S C A T T E R E R S OF RADIUS a
?-
d?

\\ \ d = DISTANCEFROM B A S E TO MOBILE

Fig. 9. Base-station
geometry. Fig. 10. Power
spectrum
for
fixed
mobile
unit
and
moving
base
station.

power arriving at the base station in an angular widthdol is where

a
v = - (sin g) { $ = k{ cos t .
where, from the geometry shownin Fig. 9, d
Thus adistancemovedatthebasestation gives the same
(43) decorrelation as a distance a/d (sin [){ moved at the mobile.
Typically, a = 100 ft’ and d = 10 000 ft, so that one must
where move ontheorderof 100 timesfartheratthe base station
thanthemobileinorder to obtainanindependentlyfading
K constantdependent on power level oftransmitter, dis- signal. Distances moved at the base station along the line be-
tance, and antenna gain tween base andmobileresultinpractically no decorrelation
d distance from base station to mobile unit according t o ourapproximationinneglectinghigherorder
( bearing of mobile unit relative to base-station motion. terms in a/d.
Substituting the foregoing into ( 5 ) gives the power spectrum
(plotted in Fig. 10) E. Level-Oossing Rates and Average Durationof Fades
To evaluate receiver performance, it is of interest to know
the rate at which the input signal falls below a given level and
how long it remains below that level on the average. The rate
where ft = f,, cos t . We have neglected higher order terms of at which a stochastic process x ( t ) decreases below a level X i s
a/d. given by Rice [ 2 ] as
By comparing the spectrum of (44) and that of a vertical 0
monopole at the mobile unit, given in ( 6 ) >we see that they N(X)=-j=x’f(X,x’)dx’ (47)
are identical for the following substitutions:
a - where f ( x , x ‘ ) is the joint probability density of x and x’. If
f,, +. 7dfA -f? f -+f - fc. (45) we are discussing the amplitude of the transmission coefficient,
then
Thus the correlations of the base station (BS) are related to
bile
the atthose vertical monopole (VM) by f(R7 R ’ ) = f ( R >f(W (4 8)

Recentmeasurements(e.g., [ 6 ] and[SI ), asdiscussedinSection

111-A, indicate a 1/4-gs spread in time delays in suburban areas, placing
the ring of scatterers at a radius of about 125 ft. In urban areas, as dis-
cussed in Section 111-A, YoungandLacy’sdata [ 5 ] indicatea5-ks
(46) spread in time delays, giving a radius of about 2500 ft.
34 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, FEBRUARY 1972

X
'

-- V
,q i
n
f(R,R')=d7
R Bb
2n
exp
[- BYBT;R'a

dB RELATIVE TO MEAN SQUARE L E V E L -

d B RELATIVE TO MEAN Fig. 12, Average duration of fade.
SQUARE LEVEL

(b)
Fig. 11. Fadingrate.(a)Fadingprocess. (b) Level-crossing rate of Equation (54) is plotted in Fig. 12 for the vertical monopole
vertical monopole. case.The average duration of a 20-dBfade at 30 mi/hand
900 MHz is about 1 ms.
where f ( R ) is given in (1 8) and F. Random FM
As the transmission coefficient varies in time its phase varia-
tions 0' appear as a random FM in an FM demodulator.
Amplitude fading can be alleviated in an FM systembyin-
where creasing the transmitter power until only a negligible number
B = bob2 - b: (50) of fades fall below the FM threshold. However, increasing the
transmitter power does not alleviate random FM. Fortunately,
and bo, b and bz are defined in (25). Then (47) gives random FM is small at UHF frequencies and lower, but it be-
comes moreimportantat X band. Rice has expressed the
probability density of the random FM in terms of the spectral
moments
For a vertical monopole, ( 5 1) becomes

Although the rms value of 0' diverges due t o ( 5 5 ) , a practical

NvM(R) is plotted in Fig. 11. demodulator is limited in amplitude and bandwidth and would
For an endfire beam antenna of beamwidth /3, (51) gives give negligible output except when the maximum Doppler fre-
quency is ontheorder of 1 kHz. Rice's expression forthe
autocorrelation of 0' may be transformed to obtain its power
spectrum, plotted in Fig. 13. Note that the spectrum decreases
02 as l/f for f / f m S 1 and is quite small for frequencies greater
than 5 fm .

Thus the directional antenna has a reduced fading rate, but its
relative fading rateatdifferent levels is the same as for the
omnidirectional vertical monopole. Thefollowingis anex-
ample of the fading rate in a typicalcase: at 30 mi/hand
900 MHz, 20-dB fades occur at a rate of 111s for the vertical
monopole case.
The average duration of fade is found by dividing the fading
rate into the cumulative probability distribution
111. STATISTICAL DEPENDENCE
ON FREQUENCY
The variation of the transmission coefficient with time is due
to the fact that it is composed of a sum of sinusoids (in corn-
plex notation), the ith onebeing given by
exp [ P ( f , +A>( f - A 4 ) l
where f;: is the Doppler shift and A t i the time delay of the ith
wave. As time increases, each wave changes relatively in phase
in proportion t o its Doppler shift, so that the statistics of the
time variations are determined by the weighting of power at
GANS: PROPAGATION I N MOBILE-RADIO ENVIRONMENT 35

00

IO

SPECTRUM
OF e‘
I

\
2
(Af
1
0
-I
2na %-Aq -
T A -
2 7h

Fig. 13. Power spectrum of random FM. o=EXPERIMENTAL POINTS FROMHOFFMAN’S DATA OF ENVELOPES
OF T W O D I F F E R E N T TRANSMISSION FREQUENCIES
ASSLlMlNG A = 1 / 4 p s
the various Doppler frequencies S(n. As one varies the trans- (b1
mitted frequency, the phase of each wave changes relatively in Fig. 14. Effect of time delays. (a) Several delay distributions. (b) Auto-
phase in proportion to its delay At,so that the statistics of the correlation versus frequency separation.
frequency variations are determined by the weighting of power
at the various delays W(At). Thus we may consider a duality where 6 is the Diracdelta function. Thesedensitydistribu-
existing between frequency and time variations, with the de- tions are displayed in Fig. 14.
lay distribution W(At) acting as the dualof the power spec-
trum S(f). A . Correlation Versus Frequency
FromYoungand Lacy’smeasurements [5] ofimpulsere- Since W(At) is the dual of S(f),to find the autocorrelation
sponse at 450 MHz in New YorkCity,Clarke [ l ] hasap- of thetransmissioncoefficient at twodifferentfrequencies,
proximatedthe averageover their delay distributions as we needonlytaketheFouriertransformationof W(At) as
(Maxwell distribution) follows:

(T*(f,,t) T(f,- Ai t ) )= 2 W(At)exp (tj2nAfAt) d A t

I

i 0,
if A t > o
if A t < 0
so that
(59)

where K is the total power received by the antenna, A t is the

delayrelative to adirectpathfrombasestation to mobile
unit, and A is the standard deviation of the delays. Although
in any one location we must use the actual delay distribution
W(At) rather than the average over all locations ( W(At)), ( 5 6 )
could serveas a typical estimate ofW(At). A simpler model
is the exponential distribution
The normalizedcorrelationcoefficientcalculatedfromthe
preceding example delay distributions W1, Wz, and W 3 , given
in (56), ( 5 7 ) ,and (58), respectively, are
lo, if A t < 0.
At the other extreme, we may approximate the delay distribu-
tion as being concentrated at two values
K
W , ( A t ) = - [&(At)t &(At - 2A)] (5 8)
2
36 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, FEBRUARY 1912

where ANTENNAS

2nA fA
2nAf A
b= d m
and ,F, is the confluent hypergeometric function. These cor-
relationfunctionsareplottedinFig. 14 alongwithexperi-
M
1
mentalpointscomputed [7] from Hoffman’s [6] measure- BRANCH COMBINER
DIVERSITY OUTPUT
mentsofenvelopes of signals withdifferenttransmission
frequencies in New Providence, N. J., assuming a standard de-
viation in time delays of 1/4 ps. Young and Lacy’s data [5]
indicateda5-psspread in timedelaysin New York City at \
450 MHz. This factor of 20 difference in time delay spread is
AMPLIFIERSWITH
considered to bemainlydue to the differenceintheurban CONTROLLED
COMPLEXGAIN
andsuburbanenvironmentsandonlypartlyduetothedif-
ference in carrier frequencies.

B. Coherence Bandwidth
Thecoherencebandwidth of achannel is themaximum
width of the band in which the statistical properties of the
transmissioncoefficients of two CW signalsarestronglycor-
related. If onetransmitsasignalwhosebandwidthexceeds
the coherence bandwidth, it is not possible to correct for the
variations in amplitudeand phase of thetransmissioncoef- and
ficient with frequency by means of a single complex correc-
tion factor applied over the complete bandwidth. Also, if one W(At)=J:p(u,
At)G({ - a)da. (63)
attempts to correct for the amplitude and phase change on a
signal due to the transmission coefficient by detecting a pilot
that was transmitted with the signal, the separation frequency It seemsphysicallyreasonable thatAt is dependenton cy,
betweenpilotand signalshouldbe less thanthecoherence being less for a in the direction of thebase station.
bandwidth.
IV. DIVERSITY
For the sake of definiteness, we will define coherence band-
width as that frequency separation for which the magnitude In order to minimize the rapid fluctuations in signal strength,
of the normalized complex correlation coefficient first drops it is possible to combine several antennas that are fading in-
below 3/4. (As will beshown later,thiscorresponds to a dependently. This technique is referred t o as space diversity.
IO-dB drop in the1-percentprobability level of amaximal The linear diversity combining may be divided into three main
ratio combiner due to error from using a pilot that has correla- types, depicted in Fig. 15:
tion to the signalof p = 3/4 instead of 1 .) From the curves maximal ratio:
of p ( A n given in Fig. 14, we see that the coherence bandwidth q=g? (64)
is only slightly dependent on the shape of the delay distribu-
tion, ranging fromaboutl/lOAtoabout1/6Afortheex- equal gain:
amples given. For our purposes, we can approximate the coher-
encebandwidth by l / 8 A . Thus, in New York City,
at
450 MHz, where A E 5 ps, the coherence bandwidth is about
25 kHz; while in the suburbs, at 860 MHz,where A 1/4 ps, selection:
the coherence bandwidth may extend to 500 kHz. The relative
phasechange of T(f,,t) versustransmissionfrequencywas
alsomeasured in thesuburbsat860 MHz [8]. Using these
measurements and the dual of (29), one arrives at a spread in where gi is the signal phasor ontheithbranch(excluding
timedelays of 1/2 p s , in approximateagreementwiththe modulation), cyi is thecomplex weighting factoroftheith
1/4-psspreadobtainedfromtheamplitudecorrelationmea- branch, and we assume equal thermal noise for each branch.
surements [ 6 ] . The improvement in average signal-to-noise ratio by using M
If one wishes todeterminecorrelationsforshifts in both independent branches of each kind ofdiversity is shown in
frequency and time, it is necessary to know the joint distribu- Fig. 16.Theeffectontheprobabilitydistributionofthe
tion p(cy, At), wherep ( a ) (which gives S(f) by (5)) is given by signal-to-noise ratio for four branches is shown in Fig. 17. The
effect of equal-gain diversity on level-crossingrates is shown
p ( a ) = j -0p ( c y , A ? ) d A t (6 2) in Fig. 18. The foregoing results of diversity effects are com-
puted by applying the complex Gaussian statistics, described
GANS:PROPAGATIONINMOBILE-RADIO ENVIRONMENT 37

DIVERSITY

NUMBER OF B R A N C H E S
Fig. 16. Change in average level by diversity.

Fig. 18. Effect of equal-gain diversity on level-crossing rates,

2 B R A N C H S E L E C T I O ND I V E R S I T Y

“I
-15 1 ab!
o w
K

PERCENT OF TIME L E V E L EXCEEDED d

z
v)
Fig. 17. Change in probability distribution by four-branch diversity.

d d
previously, of the transmission coefficients of each branch to P R O B A B I L I T Y l N S T 4 N T A N E O U S SNR
IS L C S S T H A NT H EO R D I N A T E
the diversity systems (see, for example, [9]).
The preceding statistics can also be used to determine the Fig. 19. Effect of correlation on diversity.
effects of correlation between branches on the diversity per-
formance. For example, the probability distribution of two-
branch selection diversity changes as a function of the magni-
tude of the normalized complexcorrelationcoefficient p as ASSUMING:

\
shown in Fig. 19 [9] . The correlationbetween branches % p S E CS P R E A D
could be due tothe finitespacing betweenantennas (see I NT I M ED E L A Y
Fig. 6, for example). 60 M P H
Assume the weighting factors are determinedfroma pilot 900 M H z
transmitted along with signal and that the pilot is separated in
time or frequency from the signal. Then, due to the time or ;-IO[- 1
. 4 BRANCHES

frequencyseparation,theamountthatthe pilot correlation

with the signal p decreases from unity may be computed by z -I5
the preceding techniques. The effect of this error in weighting
factor,due to decorrelation of thepilot,on
combining is shown in Fig. 20 [IO] .
maximal ratio
5
Y
YI
4
;-20
u
Ly
0
‘ 0.75 0.5
. _I- ._
0.25
- 0 o* C O R R E L A T I O N

1 I
V. SUMMARY 0
I
0.805 1.17 1.48 I . ~ * R A D p R M SP H A S E
ERROR
The basic statistical properties of the short-term variations -._c I ,
0 184 31 8 550 kHz Af FREQUENCY
in the transmission coefficient of a mobile-radio communica- SEPARATION
tions channel have been derived as a function of time, space, L.
1.27 2.25 10.3 -rns ADtE LTAI MY E
and frequency. The application of thesestatistical properties
to standard diversity systems has been indicated. Fig. 20. Combiner degradation due to error in pilot.
38 IEEE TRANSACTIONS
VEHICULAR
ON
TECHNOLOGY, VOL. NO.
VT-21, 1 , FEBRUARY 1972

ACKNOWLEDGMENT [31 A . Papoulis, Probability, Random Variables, and Stochastic Pro-

It is a pleasure to acknowledgethehelpful suggestionsof
W. C. Jakes, Jr., in reviewing this paper.
REFERENCES
[ l ] R. H. Clarke, “ A statisticaltheory of mobile-radio reception,”
Bell Syst. Tech. J., vol. 47, pp. 957-1000, July-Aug. 1968.
[2] S . 0. Rice,,“Mathematical analysis of random noise,” Bell Syst.
Tech. J., vol. 23,pp. 282-332, July1944; also ibid., vol. 24,
pp. 46-156, Jan. 1945.
cesses. New York: McGraw-Hill, 1965,pp. 352-353.
( 4 1 W. B. Davenport, Jr., and W. L. Root, Introduction to Random
Signals and Noise. New York: McGraw-Hill, 1958, p. 170.
[5 1 W. R. Young, Jr., and L. Y. Lacy, “Echos in transmission at 450
megacycles fromland-to-carradiounits,” Proc. IRE, vol. 38,
pp. 255-258, M a . 1950.
H. H. Hoffman,unpublished memo.,Dec. 15, 1961.
J. F. Ossanna, Jr., unpublished memo., Ost. 15, 1962.
M. J . Gansand S . W. Halpern,unpublishedmemo., Mar. 31,
1969.
M. Schwartz, W. R. Bennett, and S . Stein,Communication Sys-
t e r n and Techniques. New York: McGraw-Hill, 1966, ch. 10.
M. J. Gans, unpublished memo., Mar. 10, 1969.

The Accuracy of Vehicle Location by

Trilateration in a Dense
Urban Environment
HAROLD S T A R A S , SENIOR MEMBER, IEEE, A N D STEPHEN N . HONICKMAN

Abstract-A set of linear formulas are derived for estimating the mostof the distance between the vehicle and fixed sensors placed
probable location of a vehicle in a metropolitan area by trilateration throughoutthecity (in hyperboliclocationsystems,differ-
techniques, given that multipath introduces errors in rangemeasure-
ences in distance are measured rather than distance directly).
ments. These formulas are applied to a phase-ranging system (although
they are also applicable to pulse-ranging systems), and the results ob-In a pulse-ranging system, the time of arrival of the leading
tained suggest that the probability of large location errors may be tooedge of a narrow pulse is determined, while in a phase-ranging
high for many applications of interest. system, the phase of a low-frequency sinusoid (approximately
3 kHz) is measured, the low-frequency signal being modulated
ontoa high-frequencycarrier.Multipath, as wellas equip-
INTRODUCTION
I. ment characteristics, introduce errors in the measurement of
distance. It is the purpose of this paper to relate the errors in
A UTOMATICvehicle monitoring (AVM) in metropolitan
areas is becoming important for various urban functions. themeasurement ofdistance to the errorinlocatingaran-
Themostimmediateneedappears to beforpolicedepart- domly moving vehicle and to determine the statistics of loca-
ments that would like to know the disposition and status of tion errors.
their vehicles continuously,andinurban mass transit(bus) The accuracy of a pulse-ranging system does not appear to
systems that need real-time bus location information for sched- be too seriously degraded by multipath in a dense urban en-
uling purposes. In addition, there are under consideration new vironmentsinceonly the leadingedgeof the directpulse is
system
conceptswhich also
require
vehicle-location
in- required in the measurement of distance; however, the band-
formation. width requirement of such a system can be very large. Typ-
There are, of course, several physical principles on which a ically, 10 MHz maybe
necessary for reasonable location
vehicle locationsystem canbeused.Oneof themoreob- accuracy. If there areonly several thousand vehicles in the
vious ones is a pulse-ranging or phase-ranging trilateration sys- system, this much bandwidth is considered by many to be very
tem. These trilaterationsystemsdependuponmeasurement expensive in the use of valuable spectrum space. A phase-
ranging system, on the other hand,can be far more economical
Manuscript received March 1, 1971;revised November 10, 197 1. in the use of the spectrum. For example, 50 kHz appears to
The authors are with the RCA Laboratories, Princeton, N.J. 08540. besufficientfordeterminingthelocationof 2000 vehicles/