IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. COM-33, NO.

7 , JULY-1985 665

Analysis and Simulation of a Digital Mobile Channel

Using Orthogonal Frequency Division Multiplexing
LEONARD J. CIMINI, JR., MEMBER, IEEE

Abstract-This paper discussestheanalysis and simulation ofa available spectrum, it is desirable to look for channel designs
technique forcombatingtheeffects of multipathpropagation and whichprovidegoodperformance-forbothspeechanddata
cochannelinterference on a narrow-band digitalmobilechannel.This transmission,andwhicharealsobandwidthefficient.The
systemusesthediscreteFouriertransform to orthogonallyfrequency channeldesignspresentedinthispapercouldaccommodate
multiplex many narrow subchannels, each signaling at a very low rate, speech or data transmission. For the narrow channel assumed,
into one high-rate channel. When this technique is used with pilot-based a low-bit-rate speech coder would be required. For example, a
correction, the effects of flatRayleighfadingcan be reduced signifi- 7.5 kHz channel using t h e system proposed in this paper can
cantly. An improvement in signal-to-interference ratio of 6 dB can be support 8.6 kbits/s.Inwhatfollows, the channel will b e as-
obtainedoverthe bursty Rayleighchannel. In additim, with each sumed to be transmitting data symbols.
subchannelsignalingat a low rate,thistechniquecanprovideadded In a conventional serial data system, the symbols are trans-
protection against delay spread. To enhance the behavior ofthe technique mitted sequentially, with the frequency spectrum of each data
are used. symbol allowed to occupy the entire available bandwidth. Due
in a heavily frequency-selective environment, interpolated pilots
A frequency offset reference schemeis employed for the pilots to improveto the bursty nature of the Rayleigh channel, several adjacent
protection against cochannel interference. symbols may be completely destroyed during a fade. To illu-
strate the severity of the problem, consider the following ex-
ample.Assumethatthere isa cochannelinterfererwithan
I. INTRODUCTION averagepower level 17 dB below that of the desired signal.
EVERE multipath propagation, arising from multiple scat- This condition occurs approximately 10 percent of the time in
S tering by buildings and other structures in the vicinity of a a cellular mobile system. A fade 17 dB below theaverage level
mobileunit,makesthedesign of amobilecommunication will bury the desired signal in the interference. At a carrier fre-
channel very challenging [ 1 1 . This scattering produces rapid quency of 850 MHz and a vehicle speed of 60 mph, the aver-
random amplitude and phase variations in the received signal age fade duration fora fade 1 7 dB below the local meanof t h e
as t h e vehicle moves in the multipath field. In addition, the desired signal is 0.75 ms [ 11. For a data rate of 10 kbits/s,7 o r
vehicle motionintroducesaDopplershift,whichcausesa 8 adjacent bits would be destroyed during such a fade.
broadening of t h e signal spectrum.Measurementsconfirm In a serial system, higher data rates can be achieved, at the
that the short4erm statistics of the resultant signal envelope expense of a degradation in performance, by using higher order
approximate a Rayleigh distribution. modulationsor,attheexpenseofincreasedchannelband-
Multipath fading may also be frequency selective, that is, width,bydecreasingthesymbolinterval. However, delay
t h e complex fading envelope of the received signal at one fre- spread imposes a waiting period that determines when the next
quency may be only partially correlated with the received en- pulse can be transmitted. This waiting period requires that the
velope at a different frequency. This decorrelation is d u e t o signaling be reduced t o a rate much less than the reciprocal of
the difference in propagation time delays associated with the t h e delayspread t o preventintersymbolinterference. De-
various scattered waves making up the total signal. The spread creasing the symbol interval makes the system more susceptible
inarrivaltimes,knownasdelayspread,causestransmitted t o delay spread impairments.
data pulses t o overlap,resultinginintersymbolinterference. Aparallel ormultiplexeddatasystemofferspossibilities
Inatypicalurbanenvironment,aspread of severalmicro- for alleviating many of the problems encountered with serial
seconds and greater can be occasionally expected. systems. Aparallelsystemis one in which several sequential
There is an additional impairment in a cellular mobile sys- streams of data are transmitted simultaneously, so that at any
tem. The available radio channels are reused at different loca- instant many data elements are being transmitted. In such a
tions within the overall cellular service area in order to use the system, the spectrum of an individual data element normally
assigned spectrum more efficiently. Thus, mobilessimultane- occupiesonlyasmallpartoftheavailablebandwidth.Ina
ously using the same channel in different locations interfere classical parallel data system, the totalsignal frequency band is
with each other. This is termed cochannel interference and is dividedinto N nonoverlapping frequencysubchannels.Each
often the dominant impairment. subchannel is modulated with a separate symbol and, then, the
In addition, there is a long-term variation of t h e local mean N subchannels are frequency multiplexed. A more efficient use
of the received signal, called shadow fading. Shadow fading in of bandwidth can be obtained with a parallel system if t h e
a mobile radio environment is caused by large obstacles block- spectra of the individual subchannels are permitted to overlap,
ing t h e transmission path. This impairment is alleviated in cel- with specificorthogonalityconstraintsimposed t o facilitate
lularsystemsbyusingtransmittedandreceivedbase-station separation of the subchannels at the receiver.
signals at two different geographical locations [ 11, and will Aparallel approach has the advantage of spreading out a
not be discussed in this paper. fade over many symbols. This effectively randomizes the burst
Given t h e harshmobileenvironmentandthe,scarcityof errors caused by the Rayleigh fading, so that instead of several
adjacent symbols being completely destroyed, many symbols
are only slightly distorted. This allows precise reconstruction
PaperapprovedbytheEditor for ,hdio Communication of the IEEE of a majority of them. A parallel approach has the additional
CommunicationsSociety for publicationwithoutoralpresentation.Manu- advantage of spreading out the total signaling interval, thereby
script received June 18, 1984; revised January 14, 1985. reducing the sensitivity of the systemto delay spread.
The author is with AT&T Bell Laboratories, Holmdel, NJ 07733. Several systems have previously used orthogonal frequency

0090-6778/85/0700-0665$01.00 0 1985 IEEE

 666 IEEE'TRANSACTIONS ON COMMUNICATIONS, VOL. COM-33, NO. 7 , JULY 1985

cos W O t

SERIAL
DATASTREAM d(n)=a(n)+jb(n) D (t)
b ENCODER L MuLnux,
4
f* =- CHANNEL
At
(SPEECH OR DATA)

CHANNEL

sinw,,,,,t (b)
Fig. 1. BasicOFDMsystem.(a)Transmitter. (b) Receiver.

divisionmultiplexing(OFDM) [3]-[8]. Inparticular,inthe tral shape is chosen so that interchannel interference does not
early 1960's, this technique was used in several high-frequency occur; that is, the spectraof the individual subchannels are zero
militarysystems(forexample,KINEPLEX [ 9 ] , ANDEFT at the other subcarrier frequencies. The N serial data elements
[ 101, KATHRYN [ 1 11, [ 121 ), where fast fading was not a (spaced by At = l/f, where fs is the symbol rate) modulate
problem.Similarmodemshavefoundapplicationsinvoice N subcarrierfrequencies,whicharethenfrequency Chvision
bandwidth data communications (for example, [ 131 ) t o allevi- multiplexed.Thesignalinginterval T hasbeenincreased to
ate the degradations caused by an impulsive noise environment. N A t , which makes the system less susceptible t o delay spread
In this paper, a parallel system which uses the OFDM tech- impairments.In
addition,
the
subcarrier
frequencies are
nique is described. In Section 11 an analysis and simulation of separated by multiples of 1/T so that, with no signal dislortion
the'basic system, using pilot-based correction, is presented. In in transmission, the coherent detection of a signal element in
Section 111 apractical 7.5 kHzchanneldesign is presented, any one subchannel of the parallel system gives no output for
along with a discussion of several of the problems encountered areceivedelementinany othersubchannel. Using a twodi-
in reliablyretrievingthepilotsused in thedatacorrection mensional digital modulation format, the data symbols d(n)
process. Several solutions t o these problems are also presented. can, be represented as a ( n ) ' + j b ( n ) (where a ( n ) and b ( n ) are
This investigation is simplified by the assumption that the real sequences representing the in-phase and quadratur,e com-
sole source of additive signal degradation is cochannel interfer- ponents,respectively)andthetransmittedwaveformcanbe
ence-thermal noise is assumed negligible. Man-made environ- represented as
mental noise, such as that caused by automotive ignitions or N-1
neonlights, isalsoignored.However,theseimpairmentsare
basically impulsive and their effect should be greatly reduced
by this technique.
11. BASIC PRINCIPLES O F OPERATION +
where f n = fo n A f and A f = 1 / N A t . This expression ;tnd the
followinganalysescanbeeasilyextendedtoincludepulse
A , Orthogonal Frequency Division Multiplexing (OFDM) shaping other than the assumed rectangular shape.
When an efficient use of bandwidth is not required, the most Theoretically,
M-ary digital modulation scheme:: using
effective parallel system uses conventional frequency division OFDM can achieve a bandwidth efficiency, defined as 'bit rate
multiplexing where the spectra of the different subchannels do per unit bandwidth, of logz M bits/s/Hz. This is easily shown
not overlap. In such a system, there is sufficient guard space as follows. Given that the symbol rate of the serial data stream
between adjacent subchannels to isolate them at the receiver is l/At, thebitrateforacorrespondingM-arysyljtem is
using conventional filters. A much more efficient use of band- log2 M / A t . Eachsubchannel,however,transmitsat z! much
width can be obtained with a parallel system if t h e spectra of lower rate, log2 M / ( N A t > . The total bandwidth of t h e 'OFDM
the individual subchannels are permitted to overlap. With the system is
addition of coherent detection and the use of subcarrier tones
separated b y t h e reciprocal of t h e signaling element duration
(orthogonal tones), independent separation of the multiplexed
tones is possible.' where fn is the nth subcarrier and6 is the one-sided bandwidth
Consider the system shown in Fig. 1. The transmitted spec- of the subchannel (where the bandwidth is considerec: as the
CIMINI: ANALYSIS AND SIMULATIONOF DIGITAL MOBILE CHANNEL 667
COCHANNEL
INTERFERER

PILOT

/STREAM\
/ I
R (m)
' PHASE GAIN 6h1)
*CORRECTION CORRECTIONFFT-,

Fig. 2.
- -
hL

OFDM systemimplementedwith
PARALLEL

. t
TO SERIAL
CONVERTER

an FFT.
a(n),
7-
DECODER

distance t o t h efirst null). The subcarriers are uniformly spaced Furtherreductions in complexityarepossiblebyusingthe
so that f N - - fo = (N - 1)Af. Since Af = l/NAt due to the fastFouriertransform(FFT)algorithm to implementthe
orthogonalityconstraint, f N - 1 - fo = (1 - ( l / N ) ) ( l / & ) . DFT whenN is large.
Therefore, the bandwidth efficiency P becomes
C. Pilot-Based Correction
If t h e transmissionchannel is distortionless, the orthogo-
(3)
nality of the subcarriers allows the transmitted signals t o b e
received without error at the receiver. Consider t h e system in
Fig. 2 with t h e block of data represented by the sequence of
Fororthogonalfrequencyspacingandstrictlyband-limited N complex numbers {d(O), d ( l ) , .-, d ( N - 1)). These complex
4
spectra (bandwidth Af) with 6 = Af = 1/2NAt, = log2 M numbers are generated byethe data encoder from a binary data
bits/s/Hz. In reality, however, the spectra overflow this min- sequence. A DFT is performed on this block of data, giving
imum bandwidth by some factor CY so that 6 = ( I +
a ) ( l / the transmitted symbols2
2NA[) and the efficienc,y (3) becomes N- 1

p=- log' < log2 M.

CY
1 +- Notice that this is a sampled version of ( 5 ) where the complex
N notation has been retained. All future analyses will be done in
To obtain the highest bandwidth efficiency in an OFDM sys- the complex domain. Under the assumption of a distortionless
tem,N must belarge and CY must be small. channel, the received data sequence (the output of the inverse
DFT) will beexactlythetransmittedsequenceduetothe
B. Implementation of OFDM Using the Discrete Fourier orthogonality of the subcarrier tones (exponentials).
Transform If the transmission channel distorts the signal, this orthog-
The principal objections to the use of parallel systems are onality is impaired. In a flat Rayleigh fading environment (i.e.,
the complexity of the equipment required to implement the the environment is not frequency selective), the effects of the
system, and the possibility of severe mutual interference amongRayleigh channel can be represented as a multiplicative noise
subchannelswhenthetransmissionmediumdistortsthe sig- process on the transmitted signal. This multiplicative process
nal. The equipment complexity (filters, modulators, etc.) can ischaracterized b y acomplexfadingenvelopewithsamples
be greatly reduced by eliminating any pulse shaping, and by Z ( m ) = A ( m ) e l e ( m )where t h e A ( m ) are samples from a Ray-
using the discrete Fourier transform (DFT) t o implement the leigh distribution and the B(m) aresamplesfromauniform
modulation processes, as shown in [71, 181 . There it is shown distribution [ 11. These samples multiply the sequence of (6)
that a multitone data signal iseffectively the Fourier trans- t o give
form of the original data stream, and that a bank of coherent R ( m ) = Z(m)D(m). (7)
demodulators is effectively an inverse Fourier transform. This
can be seen by writing (1)as The output data sequencei ( k ) is the inverse DFT of (7),

n =0

Letting t = mAt, the resulting sampled sequence D ( m ) is seen

as t h e real part of the DFT of the sequence d(n).l The act of
truncating the signal t o t h e interval (0, NAt) imposes a sinx/x
frequency response on each subchannel with zeros at multiples
of 1/T. This spectral shape has large sidelobes, andgives rise t o
significant interchannel interference in t h e presence of multi- N- 1
path. This point will b e discussed in more detail in Section111. = d(n)z(k - n)
n =o
' It is convenient in this paper to think of d(n) as being in the frequency
domainand D(m) as being in the time domain,contrarytothe, usual * Throughout this paper, all indexes will be assumed to belong to the set (0,
engineering interpretation [8]. 1, 2, ..., N - 1 ) .
668 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. COM-33, NO. 7 , JULY 1985

where z ( n ) is t h e inverse DFT of Z ( m ) . It can be seen from If unlimited gain and phase correction is used [i.e., Z p ( , m )=
(8) that there is a complex-weighted averaging of the samples Z d ( m ) ], Z(k - n ) = 6 k n , there is no intersymbol interference,
of thecomplexfadingenvelope. If Z ( m ) = 1 for all m (the and ( 12) becomes
distortionless channtl), z ( k - n ) is simply the Kronecker delta
function 6 k n and d ( k ) = d(k).Inthe presenceoffading,
z ( k - n ) # 6 k n and

N-1
Theonlydistortion is causedby thecochannelinte~ferer.
n=O However,. since Zi(rn) and Z d ( m ) are statistically independent,
n#k the desired signal may be in a fade when the interferer is not,
and unlimited gain correction may boost the interferer alrerage
The second term on the right represents the interchannel (in- energy above that of the desired signal.
tersymbol) interference caused by the loss of orthogonality. One alternative t o unlimited gain and phase correctio!l is t o
Without correction for the fading, the output sequence is cor- have a limit on the gain correction, so as not to follow t:he de-
rupted by intersymbol interference even if there is no cochan- sired signal into deep fades [ l ] . This is done at the expenseof
ne1 interferer. increasedintersymbolinterference,duetoimperfectcorrec-
Pilot-basedcorrectionprovidesanamplitudeandphase tion of the desired signal. In this situation, the correction sig-
reference which can be used to counteract the unwanted ef- nal is of the form
fects of multipathpropagation.Similarconsiderationshave
been analyzed for single-sideband mobile-radio systems [ 141,
[ 151.Coherentdetection,bydefinition,requiresaphase
reference; however, gain correction is also needed in an OFDM
system in a fading environment to remove intersymbol inter-
ference. If phase and gain correctionis employed in t h e absence where E is the gain limit and is defined relative t o t h e average
pf cochannelinterference, it is easilyshown,in (9), that value of the local field strength. Therefore, in (1 2),z ( k -- n ) #
d ( k )= d(k). 6,,, resulting in intersymbol interference. Consequently, there
Inacellularmobilesystem, thedominanttransmission is a tradeoff between increasing the intersymbol interference
impairmentoftencomesfromotherusersusingthesame and boosting the cochannel interference energy.
carrierfrequency.It is assumedherethatthedesiredsignal Another alternative is t o develop an optimum gain coset-
anda single undesiredcochannelinterfererarereceived si- tion factor which takes both distortion effects into account.
multaneously, and that both are digital signals modulated by An optimum gain correction factor F ( m ) has been deri’rfd b y
differentdatasequenceswithidenticalsignalingrates.It is minimizing the mean-square distortion betweenD(m) andD(m).
alsoassumed that they are subject to mutually independent The derivationof F ( m ) has beenomitted for the sake of brevity.
Rayleigh fading. The correction sequence then becomes
When a cochannel interferer is present in the received sig-
nal,it is notadvantageoustodounlimitedgaincorrection, Zc(m = z,( m) F ( m )
due to the possibilityof enhancing the energy of the interferer
during deep fades of the desired signal. The detrimental effects
of unlimited gain correctioninthepresence
interferercanbeseenasfollows.Let
of a cochannel
D(m) bethe desired
=Z,(m) [1+ .(3)>’].
transmitted signal sequence and let I ( m ) be the correspondin
cochannel interferer sequence. With Z d ( m ) = A d ( m ) e j e d ( m y This correction procedure would be more difficult to imple-
and Z i ( m ) = A j ( m ) e l e l ( m ) the desiredandinterferercom- ment than the gain limiting procedure described above...*
plexfadingsequences,respectively,thesequencepresentat In addition to the impairments caused by intersymbol and
t h e receiver can be represented as cochannel interference, frequency-selective fading may also be
present.Thisphenomenoncausesadecorrelation of:there-
R ( m ) = z d ( n z > D ( m >+ f i z i ( W ( m > (10) ceived signal envelopes at different frequencies, lessening the
effectiveness of thepilot-correctionprocedure,sinceadata
where y is theinterference-to-signalpowerratio (SIR-l). point which is being corrected may be decorrelated from the
R ( m ) is corrected by a complex correction sequence Zc(rn) = corresponding pilot complex fading envelope.
Z,(m), the complex pilot fading envelope, giving Finally, one of t h e major advantages of the OFDM tech-
nique is its ability to “average” out impairments, making ,the
bursty Rayleigh channel appear much less bursty. The extent
t o whichthisaveragingapproachesaGaussianchannelde-
pendsonthecorrelationbetween samplesof thecomplex
fading envelope. It can be seen that as N increases, more inde-
Taking t h e inverse DFT of (1 l), the received data sequence pendentfadesareaveraged.Thisenablesbursterrors to be
becomes randomizedandtherebyaidsinbiterrorcorrectiorl.This
property will be more evident in the simulation results, which
N-l
N--l 1 z j ( m )
=
n=o
+
d(n)z(k - n ) fi X - -
m=o N Z p ( m )
indicate that the curves for the bit error rate fall betwcen the
linear Rayleigh channel curves and the exponential Glmssian
i channel curves. For large N and high vehicle speeds, the bit
error curve approaches that for a Gaussian channel.

D.Distortion Analyses
Several mechanisms contribute to t h e overall distortion of
t h e desired signal. In this section, emphasis is on the contribu-
tionsdue to gainlimiting,evident in increasedinter!;ymbol
interference,andduetocochannelinterference.Thedistor-
CIMINI: ANALYSIS AND SIMULATION OF DIGITAL MOBILE CHANNEL 669
tion resulting from decorrelation of the pilot due to frequency-
selective fading or due to interference on the pilot is considered
in Section 11-F.
First,considerthecaseofgain-limitedcorrection,where
the amplitude correction is bounded to follow fades only as
deep as E. Assume that the random processes which produce
therandomsequencesareergodic,therebypermittingthe
equivalence of time and ensemble averages. The pilot complex c
fadingenvelopeataparticularinstant
z d ( m ) and the correction sequence is
in time is Z,(m) =
a
s!
8
Z,(m> = max ( A d ( m ) ,e ) e i e d ( m ) . (16) -
10 dB
The corrected output samples become
=I5dB
=iodB

I I I I I
-30 -20 -10 0 10
€(dB)
Fig. 3. Signal-to-distortionratiofor a flat Rayleigh fadingenvironment
when gain-limited correction is used.

The signal-to-distortion ratio (SDR) can be defined as in [ 141,

where x denotes a time average of X . Assuming I D ( m ) Iz =

]l(m)
='1 , the denominator in( 1 8) reduces t o
I b ( m ) - D(m>l 2

I I I
0 10 20 30 40
SIR (dB)
Fig. 4. Signal-to-distortion ratio for a flat Rayleigh fadingenvironment
when the optimum gain correction factor is used.

Assuming time averagescanbereplaced by expected values nal. Notice, there is a definite maximum which is fairly flat.
and assumingdd(m) andAi(m)are statistically independent and Although SDR as defined here is an analog transmission qual-
Rayleigh distributed, ( 1 8) becomes, after some manipulations, ity measure, it does indicate the degree to which intersymbol
interference, caused by imperfect gain correction, and cochan-
ne1 interference are problems. These factors are critically im-
S portantin digital transmission.The SDR alsoclearlyshows
the tradeoffs which must be made when choosing the appro-
priate gain limit. This, in turn, directly affects the bit error
rate (BER), as shown in the next section.
Similar results can be derived for optimum gain correction,
+ SIRT 1 as in (IS), and the SDR can be shown to be

where E,(x) = -[@ +

In (x) + (-l)"x"/nn!)J and \k
is Euler's constant (=0.57721566 .-). The SDRin ( 2 0 ) is plot-
ted in Fig. 3 for several values of SIR. Obviously, if SIR =
(no cochannel interference), the results reduce to that in [ 141
and no gain limit should be used. However, for SIR 00 t h e<
- SDR = [(&) [ & In (SIR) -

which is plotted versus SIR in Fig. 4. This curve indicates t h e

I]-'
1 (21)

curves dearly indicate the tradeoff between intersymbol in- best performance for a given SIR. Notice, by comparing Figs.
terference,causedbygainlimiting,andboostingoftheco- 3 and 4, that using gain-limited correction does not sacrifice
channel interference average energy, caused by unlimited gain much if the gain limit is in the vicinity of the maximum. Both
correction. If unlimited gain correction is used, SDR = -00, in- oftheseresultscouldbeusedasan aid in determining the
dicating that the interferer completely distorts the desired sig- appropriate level for gain limiting for agiven SIR.
 670 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. COM-33, NO. 7 , JULr‘1985

1 1 1 1 1 1 1 1 1 1 1 1
E. Simulation Results
Initial simulations were performed assumingflatRayleigh
10 0
- N= 512 v.60 MPH, OPTIMUM GAIN
CORRECTION
fading.Inthesesimulations,it wasalsoassumed
pilots could be recovered perfectly; that
thatthe
is, there is no inter-
--- N=512, v 560 MPH, GAIN-LIM~TEDCORRECTION
***.. N . 5 1 2 , ~ =30MPH, GAIN-LIMITED CORRECTIIW
ference or distortion on the pilots. A symbol rate and chan- .-e- N.128, v = ~ O MPH, GAIN-LIMITED CORRECTION
nel bandwidth of 7.5 kHz3 were used and the BER was deter-
mined for several values of SIR (100 000 bits were used in t h e lo-’
simulationtoprovidestatisticalsignificance). Based o n this
choice for the bandwidth, the maximum bit rateis 7.5 log2 M
kbits/s. Several parameters were vaned in this initial investiga-
tion, the most important being N , the number of subchannels,
and u, the vehicle speed. Both quantities are very important
factors in determining the ability of this system to effectively 40-2
randomizethebursterrorscreated by t h e Rayleighfading. E
The fading rate is directly proportional to the vehicle speed. W
m
In particular, at a carrier frequency of 850 MHz, independent
fades are about 7 in apart, giving a fade every 6.6 ms at 60
mph. Therefore, for a given value of N , higher vehicle speeds
10-3
shouldresultinbetterperformancebecausemorefadesare
included in t h e averagingprocess.Similarly, for a given ve-
hiclespeed, if N islarge, the total signalinginterval is also
large and more fades are again used in the averaging process.
The results shown in Fig. 5 , where quadrature phase shift
keying (QPSK) has been employed, indicate the improvement I 0-4
possible if OFDM is used with gain correction under the as-
sumptions of ideal pilot recovery and a flat Rayleigh fading
environment. Results for both optimum gain correction, as in
(1 5), and gain-limited correction, as in (14), are given. These
resultsclearlyindicate theeffects of vehiclespeedandthe
number of subchannels. Ata carrier frequency of 850 MHz 10-5
and with a vehicle speed of 60 mph, with gain-limited correc- 0 2 4 6 8 10 1 2 44 16 18 2 0 22
24 26
tion,improvementsinSIRof 6-7 dB4 havebeenobtained
SIR (dB)
using 5 12 subchannels ( T = 6 8 ms). This is in comparison to a
flat Rayleigh channel using coherent detection ( N = 1) with Fig. 5 . Simulation results assuming perfect pilot recovery in a flat kayleigh
QPSK. A reduction in speed t o 30 mphresults in aloss in fadingenvironment (QPSK, = 7.5 Hz).
performance of less than 1 dB. A reduction of N to 128 sub-
channels-(T = 17 ms) results in an additional 2 dB loss, be-
cause fewer independent fades are included in the averaging
process. For the cases where gain-limitedcorrection is used,
the BER curves shown are for the “best” absolute gain limit.
If the optimum gain correction factor can be determined, an
additional improvement of 1 dB can be obtained. The sensi-
tivity of the BER on the gain limit, shown in Fig. 6, indicates
that adaptive gain limiting, or some “intelligent” guess at the
gain limit based on the distortioncurves, may be required.

F. Effects ofFrequency-Selective Fading

I
-
SIR=IbdB

When good correlation exists between the fading statistics

of the pilot tone and those of the fading information signal,
almo‘st total suppression of the unwanted amplitude and phase 10’~
fluctuations is possible. The simulation results given in Section SIR
11-E were obtained under the assumption that the fading on
t h e pilot and the desired signal were totally correlated. This is
a valid consideration when there is no interference on the pilot /
/
and when the fadingis not frequency selective.
In general, however, the mobile environment is frequency I d4
selective, due to the existence of a spread in arrival times of
t h e various multipath components. In this case, the correlation
in phase and amplitude between two pilots separated in fre-
quency is high for small frequency separation, and falls essen-
tially to zero as the separation substantially exceeds the correla-
tion bandwidth [ 1 ] . The gain correction process, aswill be seen,
requires a high degree of correlation between the phase and
1 6 ~ I
-24 -22 -20-18 -16 -14-12 -10 - 0 -6 - 4 - 2

Such a channel allows a factor of 4 improvement in spectral efficiency

over the current 30 kHz cellular mobile telephone service channel. Fig. 6. Sensitivity of BER tovariations in gainlimit ( N = 512, u = 60
All comparisons in this paper will be made at a BER level of mph,f, = 7.5 kHz).
ULATION
AND
ANALYSIS
CIMINI: CHANNEL
MOBILE
OF DIGITAL 67 1
amplitude variations of the pilot and that of the phase and

r
amplitude variations imposed on the data.
A simple'way to estimate the effects of a delay spread en-
vironment is to compute the equivalent decorrelation between loo
t h e pilot and the data caused by the frequency-selective fading -NO DELAYSPREAD
(forexample, see [ 121).Thesecalculationsdependonthe ---DELAY SPREAD (A'5OpdI 500 HZ SEPARATIOI
model used t o describe t h e dispersive channel. In the simula- ,o-~ -.-.
DELAYSPREAD ( A e 5 0 p ) , j kHz SEPARATION
tions, t h e delay spread channel is simply modeled as a two-im-
pulse channel response with equal-amplitude signal and echo
separated by some time A. As in [2, sect. 9.81 , an approxima-
tion model will be assumed for the delay distribution. In this
model, the probability density function of the delay is repre-
sented as t w o equal-amplitude,equallylikelyimpulses sepq- io-2
rated by some delay A. This is the model which was employed
in the simulation andis a sufficient approximation fors n/2. <
The rms delay spread in this case is simply A/2. The corre- a
w
sponding complex correlation coefficient can be shown to be m

lom3

where s is the frequency separation in rad/s,T is the separation

in time between two samples, and wm is the maximum Dop-
pler shift in rad/s. LettingT = 0 (without loss of generality) \
IO-^
1
p(s, a) = - [l + cos (sa)]1 / 2 GAUSSIAN
&- CHANNEL
The coherence bandwidth, defined as the frequency separation
when the envelope correlation ( p 2 ) is 0.5, is easily computed 6
to be l/A. According to publisheddatafor New YorkCity SIR (dB)
[ 161, t h e rms delay spread is usually less than 3 ps (A = 6 ps).
However, othermeasurements in Newark [ 171 suggest that Fig. 7. Simulationresultsin a frequency-selectiveenvironment (QPSK, f,
= 7.5 kHz, N = 512, u = 60 mph, gain-limited correction).
values of A as high as 5 0 ps are possible. Values of A = 25 p s
were found to occur often enough and with sufficient repeat-
ability in Newark t o b e regarded seriously. The usefulness of (a particularly bad case measured in Newark). This value has
the OFDM technique has been illustrated for a flat Rayleigh been chosen to determine what degradations occur in an ap-
fading environment. It remains to show how much the system proximately worst-case
dispersiveenvironment. Frequency
degrades in a delay spread environment where the pilots and separations between the pilot and the data of 500 Hz and 1
the data become decorrelated. kHz have been used. The results are shown in Fig. 7 and indi-
In the simulation, a bound on the effects of delay spread cate that, for severe frequency-selective fading, a loss of 5 or
hasbeenobtainedbyappropriatelydecorrelatingthepilot 6 dB in BER is obtained over the flat fading case for 1 kHz
fading sequence and the fading data sequence. At a specific separation, and less than 2 dB for 500 Hz separation. Similar
instant in time, the correlation hetween the complex fading degradationsoccurwhenthere is interferenceonthepilot,
envelope of the pilot, Z,(rn), and the complex fading envelope either from the data spreading into the pilot due to the Dop-
of the data, Z,(m), is some valuep < 1. It is easily shown that pler shift induced by the motion of the mobile, or by a co-
t h i s will be true if the pilot complex fading envelopeis chosen channel pilot. The interference from a cochannel pilot decor-
as relates t h e received pilot and data signal -envelopes. Thus, co-
channelpilotinterferenceinducescorrectiondistortionina
mannerverysimilar to that induced by operating in a high
delay'spreadenvironment.Obviqusly,inapracticaljmple-
where Zx(rn) is statistically independent from Z d ( m ) . In the mentation of this modem, some techniques must be devised
previous section, p was assumed to'be unity. Obviously, this toovercomethedetrimentaleffects of delayspreadand
process causes additional distortion. Using limited gain correc- pilot interference. These will be treated
tion, it can be shown that in a frequency-selective environment . in. the next section.
the SDR is III. PRACTICAL C~NSIDERATIONS
A . Problems in Pilot Retrieval
Inapractical OFDM system,amethodforaccurately
correcting the data is required, whichin turn requires a method

-
P+ cos @
E
erf ( E ) +
(
1 - p2 +-
S:R) I-'
E1(e2)

(25)
for reliably retrieving the pilot signals. In addition, the com-
plex fading envelope of the pilot signal' must be highly cor-
related with the complex fading envelope,of the data. A pilot
signal, located somewhere within t h e transmission band, must
be sent with the'data. This pilot will be distorted in transmis-
sionbycochannelinterference,byadjacentdatasymbols
where @ = arctan [-sa]. This reduces to (20) for p = 1 and spreading into the pilot due to the motion of the vehicle, b y
reduces to results in [ 141 when SIR = m and @ = 0. filtering processes, and b y t h e decorrelqtion between the pi-
Simulation results have been obtained assuming A = 5 0 ps lot fading envelope and the signal fading envelope caused by
672 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL.NO.
COM-33, 7 , JULY 1 9 8 5

afrequency-selectiveenvironment.Techniques
sented which alleviate these distortions.

B. Techniques for Reliable Pilot Retrieval

will bepre-

When there is adequate separation between the pilot tone 0

n
\
\\
\
8.

1-1
~
BLOCK # I

N-&
I
,‘
I

N-IO
I m\
\
1

‘8.

1-1
BLOCK # 2

N-P bl-1
20’
.g
I

,
and its neighboring information components, it is possible t o
separate the spread pilot from the spread signal components,

I
SHAPEANDMODULO
without fear of overlap, by suitable filtering in the receiver. EXTENDBOTHBLOCKS
Assuming a t o p vehicle speed of 60 mph and transmission at
850 MHz, each subchannel spreads +80 Hz (due to the Dop- -
plershift).Therefore,a200 Hz spacingbetween
frequency and the nearest data subcarrier provides more than
enough protection against overlap. However, due to the use of
the DFT, each subchannel possesses a sin xfx spectrum which
has. fairly large sidelobes, and which may cause problems if the
orthogonality of the subcarriers is impaired. The sidelobes can
be reduced either by filtering the individual subchannels be-
the pilot

0
BLOCK # I

N-lo
N -1
p\
I

I
I
I/
8

,’
‘,I‘
.2
/

\\

1-lo
89

4
I

N- I
I
I
BLOCK # 2
7 N-1

fore transmission (for example, see [ 6 ] ) ,or by extending the \ v J

frame of dataintimeandbyrequiringgradualratherthan BLOCK IS NOW N+k?
LONG
abrupt rolloffs of the transmitted waveform [ 81. In this paper, Fig. 8. Modulo extension and shaping of data block to reduce interchannel
thelattertechnique is employed,to avoid thecomplexity interference caused by impairment of orthogonality.
created by prefiltering. In particular, the data block is modulo
extended in time, and then the extended sections are shaped 3
with a raised cosine and added as shown in Fig. 8. This reduces
the sidelobes but widens the main lobe. The length of the ex-
tension determines the width of t h e main lobe beyond that of
the sin x/x spectrum.
When the effects of the data on the pilot signal have been
sufficientlyattenuated,cochannelpilotinterferencecanbe
attacked. A frequencyoffsetreferencetransmissionscheme 2 @
j!
works well t o minimize t h e effects of cochannel pilot interfer- /
ence. In this scheme, the slot for pilots in the total bandis
further divided into several slots for cochannel interferer pilots.
-
0

9
3
I
I
/

I
Specifically, in this paper, three adjacent slots are made avail- I
able,increasingthetotalpilotallocationto1000Hz.This I
techniqueeffectivelyincreasesfrequencyreusebyafactor 1 i
of 3. i
I
In a frequency-selective environment, a high correlation be- I
tween the complex fading envelopes of the pilot and the data
must be ensured. Based on the results in Section 11-F,this de-
pendsonthe severity of the delayspreadand onthe fre- / PILOT 9 2 ;
quency separation between the pilot and the data subcarrier.
Assuming a harsh delay spread environment (for example, A = 7 c 2 0 kHz
50 ps), thepilotanddatamustbeverycloseinfrequency f
(<500 Hz) to constrain the loss due to decorrelation to 2 dB. Fig. 9. Possible amplitude response for two-impulse channel ni,odel for
If a 7.5 kHz channel is used, placing t h e pilot in the center of delay spread, A = 50 ps, A(w) = 2 lcos wA/2(.
the band results in data symbols at the edges that are 3.75 kHz
away from the pilot. This large separation causes a large degra-
dation in BER. With smaller amounts of-delay spread (for ex-
I yrt I
I I I I??[
I I

ample, 5 ps) this is not a problem. This large degradation can 250 Hz I kHz 1 kHz 3 kHz I kHz 1 kHz 250 Hz
be avoided if two pilots, separated in frequency, are used and
the appropriate complex correction signal is obtained by inter-
polation in frequency between these two pilots.
In general, separate amplitude ( A ) a d phase ( 8 ) interpola-
1 PILOTS FOR DESIRED SIGNAL
4
tion is notappropriate.Forthetwo-impulsechannelmodel COCHANNEL PILOTS
for delay spread which is being used here, the amplitude re-
sponse is as shown in Fig. 9. It is possible f0.r both pilot fading Fig. 10. Datasetup used in simulations.
envelopes t o be at peaks, due to the randomness of the initial
position of the 7.5 kHz data window. The resulting amplitude include measures t o limit adjacent.channe1 interference. This is
and phase interpolation will be inadequate to follow the null accomplished by leaving 250 Hz gaps at each end of th.e band.
infrequencycausedbythefading.Interpolationseparately The final data setup used in the simulations is shown, in Fig.
t h e real (in-phase, I ) and imaginary (quadrature, Q ) parts of 10. The total bandwidth used by the data has now been re-
the complex fading envelopes of the pilots will enable such a duced to 5 kHz.
correction to occur. Simulation results, presented in Fig. 12,
indicate that I-Q interpolation gives 8 dB improvement over C. Basic System Setupand Pertinent Parameters for
A-8 interpolation for 50 ps delay spread. Similar results will Sirnuletion
be obtained for other models for the dispersive channel. The Simulationswereperformedincorporating all of :thede-
technique depends only on nulls occurring in the channel fre- sign features presented in the previous section. The pertinent
quency response. system parameters were carefully selected, based on a:n initial
Finally, the design of a mobile telephone system must also assumption of a7.5 kHz channel. T , the total signaling interval,
CIMINI: ANALYSIS AND SIMULATION OF DIGITAL MOBILE CHANNEL 673

IO0
SPREAD,A =50 p ~ 4-ARY8
........DELAY
N = l , NO DELAY SPREAD
, A-8

._._._DELAY SPREAD, h = 5 0 ~ 84-ARY.

. 1-0
(--I2
4-ARY
(Q PSK) dmin= 4.41

I o-2
E
W
m
8-ARY
(7PSK-01
1o -~

1o -~

16-ARY
(QA M 1 ___+____ dmin = 0.63
I o-~
x x x x 2 4 6 8 I O 12 14 16 18 20 22 24
x x S I R (dB)
Fig. 1 1 . Modulation formats used insimulations(averageenergy = 1). Fig. 12. Simulation results (N = 128, u = 30 mph,f, = 7.5 kHz).

should be chosen to be much greater than the average dura- wereobtainedfor

binary
(BPSK),4-ary
(QPSK), 8-ary
tionbetweenfades, so thatmanyfadesareaveragedinthe (7PSK-0), and 16-ary (QAM) modulation schemes. The 8-ary
receiver. In t h e previous simulation, T was 68 ms and good constellationusedaseven-pointphase shiftkeying pattern
results were obtained. However, based o n subjective opinions, with an additional data point at the origin. This requires a two-
a signaling interval of 68 ms was shown t o result in a notice- stepdetectionprocedure,but gives 1.7dBimprovementin
ableroundtripdelayforspeechtransmissionwhenbuffer- performanceoverconventional8-PSK,basedoncomparison
ingandprocessingtimesareincluded.Also,bymaking T
of theminimumdistancesineachconstellation.Thesepat-
too large, the subchannels are closer together and the inter- terns could form the basis of a multimode system.
channel interference due to spreading may become unaccept- Simulation results for a flatRayleighfadingenvironment
able.An interval on the order of 20ms ( N = 128) is more are shown in Fig. 12. Gain-limited correction has been used in
satisfactory.Thisdecreasein T resultsinadegradation of
the data correction process. These results are computed for a
about 2 dB in performance compared t o T = 68 ms. vehiclespeed of 30mph.Fora.vehiclespeed of 60mph,
Note that the average interval between fades tends to in-
there is an improvement of 1 dB. Each pilot was also run 10
finity as t h e speed of the mobile approaches zero. Therefore, dBhigherinenergythananyindividualdatapoint,forad-
T must be very large at low speeds to obtain any averaging.
ditional protection of the pilots. By using the frequency off-
A solution to this problem is to make the vehicles always ap- setreferencetechniquetocombatcochannelpilotinterfer-
pear t o move at some substantial speed, such as 30 mph. One ence, the total pilot-to-distortion ratio is 25 dB, a reasonable
way thiscanbedone is byforcinganantenna to oscillate
level of distortion. The pilot filters were implemented, in the
when the vehicle is moving at low speeds, as in[ 181. simulation,aslow-passFIRfilterswith100 Hzpassbands.
Assuming an initial sample rate of 7.5 kHz and letting N = These filters added no significant distortion to the pilots. The
128 gives T = 17.06 ms and A f = 58.59 Hz. Assuming a total results in Fig. 12 indicate the improvements which are attain-
interval of 20 ms, these parameters allow 2.93 ms of extension. able with this technique. Gains of 2-3 dB were obtained for
Sacrificing two bands of 1000 Hz each for pilot protection and OFDM-QPSKover coherent QPSK.An additional3 dB is
250 HZ at either end for adjacent channel interference protec- gained over differential QPSK (DQPSK).
tion leavesspace for86datachannelsandabandwidth ef-
Fig. 12 also indicates the BER for QPSK when there is sig-
ficiency of 0 = 0.57 log2 M bits/s/Hz. For QPSK, this corre- nificant delay spread. Delay spread was implemented as a sta-
sponds to a maximum data rate o f 8.6 kbits/s. tistically independent delayed echo added t o t h e desired sig-
nal. At a BER = l o p 2 , there is a loss of only 2 dB in SIR for
D.Simulation Results A = 50 ps. For A = 10 ps, the loss is about 0.5 dB. The losses
In this simulation, several different two-dimensional modu- due to delay spread have been minimized by interpolating in
lation formats wereconsidered.TheseareshowninFig. 11, frequency tenappropriatelyspacedpilots,whicharethen
alongwiththecorrespondingminimumdistances.Results each used to correct the data sequence. The appropriate sym-
 674 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. COM-33, NO. 7 , JULY 1985

PILOT # I PILOT #2
n=n4 n=n2
COCHANNELINTERFERER

--., CHANNEL

T
1

DELAY
CHANNEL

'?,(m)
R (m)
+
INTERPOLATE
' IN
.. EXTRACT

FREQUENCY
. 4PPROPRIATE
SYMBOLS

t n=6,16,
40,50,60,
fiz (m). 70,80,90,
113,121
P12?(m)
INTERPOLATED PILOT
LOCATIONS
. .
- 1
Fig. 13. Delay spreadimplementation and pilot correctionprocedure.

bols are then chosen at the output. The correction procedure TABLE I
is indicated in Fig. 13 and the symbols corrected by each pilot DATA SYMBOLS CORRECTED BY EACH INTERPOLATED PILC~T
are given in Table'I.
Variations suFhaschanging thepilotlocations,changing INTERPOLATED DATA SYMBO.LS
the number and locationsof the interpolated pilots, and chang- PILOT POSITION CORRECTED
ing the length of the time extension have a large effect on the 8
BER. Based upop numerous trials, it has been determined that 16
for ten interpolated pilot's, t h e system simulation used in this 40 38.44
section is close to an optimum. 50 45,54
60 55.64
IV. CONCLUSIONS A N D FURTHERINVESTIGATIONS 70 65.74
Inthispaper,acellularmobileradiosystembasedon 80 75,84
orthogonallyfrequencydivisionmultiplexingmanylow-rate 90 85,91
subchannelsintoonehigherratechannel was analyzedand 113 109.11 7
simulated. This technique, when used with pilot-based correc- 121 I 18,123
tion, was shown to provide large improvements in BER per-
formance in a f l a t Rayleigh fading environment. Degradations
due to severe delay spread were kept t o a minimum by fre-
quency interpolation of two pilots: Using a frequency offset SomealternativecodingschemesfortheOFDM;;ystem
scheme for the pilots limited the effects of cochannel pilot have been briefly considered. Two possibilities for coding in
interference.Theseconsiderationsprovidedvery'goodBER thissystemareanalogandUngerboeckcodes.Analogcodes
performance; at the expense of a decrease in the overall band- have recently gained quite a bit of interest (for example, see
width efficiency. The averaging ability of the OFDM system, [ 191). One approach is t o use fixed transmitted symbols to
whichmakes' the
bursty
Rayleigh channelappear nearly correct symbols in error. In initial trials for the OFDM system,
Gaussian,providesthislargeBERimprovement.Additional this technique was not very effective, because the largl: noise
improvementsarepossible if largersignalingintervalsare term on eachsymbolcaused by cochannel interference was
permissible:' enough to offset the effectiveness of the algorithm. Further
Other factors which also impair the orthogonality of the study in this area may prove fruitful. Ungerboeck codf's [20]
subcarriers, such as phase and gain hits, phase jitter, and fre- map data symbols into a higher order constellation and, in a
quency offset, have not been considered. In [ 131, these im- Gaussian, white noise environment, have been shown'to pro-
pairments were considered for a voice-bandwidth data modem. vide as much as 3 dB. coding gain without sacrificing data rate
There it is shown that some form of frequency domain equali- or requiring more bandwidth. These codes, although untested
zation can be used to alleviate the effects of these impairments. in a fading environment, present a possibility for our system.
TION
AND
ANALYSIS
CIMINI: CHANNEL
MOBILE
OF DIGITAL 675
ACKNOWLEDGMENT multiplexing,” Ph.D. dissertation, Univ. Illinois, Urbana-Champaign,
1982.
The author would like to acknowledge gratefully N. Sollen- [14] K. W. Lelandand N. R.Sollenberger,“Impairmentmechanismsfor
berger and K. Leland for their helpful suggestions throughout SSB mobile communications at UHF with pilot-based Doppledfading
the course of this project. correction,” Bell Syst. Tech. J., vol. 59, pp. 1923-1942, Dec. 1980.
[I51 J . P. McGeehanandA. J. Bateman,“Theoreticalandexperimental
REFERENCES investigation
of
feedforward
signalregeneration
as
means
a of
combatingmultipathpropagationeffectsinpilot-basedSSBmobile
W. C. Jakes, Jr., Ed., Microwave Mobile Communications. New radiosystems,” IEEE Trans. Veh. Technol., vol.VT-32,pp.106-
York:Wiley,1974. 120,Feb.1983.
W. C. Y. Lee, Mobile Communications Engineering. NewYork: [16]D. C. CoxandR.P.Leck,“Distributionsofmultipathdelayspread
McGraw-Hill,1982. andaverageexcessdelayfor910MHzurbanmobileradiopath,”
R.W.Chang,“Synthesisofband-limitedorthogonalsignalsfor IEEE Trans. Antennas Propagat., vol. AP-23, pp. 206-213, Mar.
multichanneldatatransmission,” Bell Syst. Tech. J., vol. 45, pp. 1975.
1775-1796,Dec.1966. [17] K. W.LelandandN.R.Sollenberger,privatecommunication.
-, “Orthogonalfrequencydivisionmultiplexing,”U.S.Patent [18] W. C. Wong, R. Steele, B. Glance, and D. Horn, “Time diversity with
3 488 445, tiled Nov. 14, 1966, issued Jan. 6, 1970. adaptive error detection to combat Rayleigh fading in digital mobile
B. R.Saltzberg,“Performanceofanefficientdatatransmission radio,” IEEE Trans. Commun., vol.COM-31,pp.378-387,Mar.
system,” IEEE Trans. Commun. Technol., vol. COM-15, pp. 805- ‘1983.
813,Dec.1967. [19] J. K. Wolf, “Redundancy, the discrete Fourier transform, and impulse
R. W. Chang and R. A. Gibbey, “A theoretical study of performance noise cancellation,’’IEEE Trans. Commun., vol. COM-31, pp. 458-
ofanorthogonalmultiplexingdatatransmissionscheme,” ZEEE 461, Mar. 1983.
Trans. Commun. Technol., vol. COM-16, pp. 529-540, Aug. 1968. [20]G.Ungerboeck,“Channelcodingwithmultilevel/phasesignals,”
[71 J. Salzand S. B. Weinstein,“Fouriertransformcommunication IEEE Trans. Inform. Theory, vol. IT-28, pp. 55-67, Jan. 1982.
system,”in Proc. ACM Symp. Probl. Optimiz. DataCommun.
Syst., Pine Mountain, GA, October 13-16, 1969, pp. 99-128.
S. B. Weinstein and P. M. Ebert, “Data transmission by frequency-
divisionmultiplexingusingthediscreteFouriertransform,” IEEE
*
Trans. Commun. Technol., vol.COM-19,pp.628-634,Oct.1971.
M.L.
Doelz.E. T. Heald, and
D.
L.
Martin,“Binarydata Leonard J. Cimini, Jr. (S’77-M’82) was born in
transmission techniques for linear systems,” Proc. IRE, vol. 45, pp. Philadelphia, PA, onApril19,1956.Hereceived
656-661,May1957. the B.S.E., M.S.E., and Ph.D. degrees in electrical
G.C.Porter,“Errordistributionanddiversityperformanceofa engineering from the University of Pennsylvania,
frequency-differential PSKHFmodem,” IEEETrans.Commun. Philadelphia,in1978,1979,and1982,respec-
Technol., vol. COM-16, pp. 567-575, Aug. 1968. tively. During his graduate work he was supported
M. S. Zimmerman and A. L. Kirsch, “The AN/GSC-IO (KATHRYN) by a National Science Foundation Fellowship.
variableratedatamodemforHFradio,” IEEETrans. Comrzun. He has been employed by AT&T Bell Laborato-
Technol., vol.COM-15,pp.197-205,Apr.1967. ries, Holmdel, NJ, since 1982, working in the area
P. A. Bello, “Selective fading limitations of the Kathryn modem and , , of mobile radio systems. His main research interests
some system design considerations,” IEEE Trans. Commun. Tech- are in the generalareas of signalprocessingand
nol., vol.COM-13,pp.320-333,Sept.1965. communications systems. He also teaches courses in communication systems
W. E. Keasler,Jr.,“Reliabiledatacommunicationsoverthevoice at Monmouth College, West Long Branch, NJ.
bandwidthtelephonechannelusingorthogonalfrequencydivision Dr. Cimini is a member of Tau Beta Pi and Eta Kappa Nu.