Y

2908

IEEE TRANSACTIONS ON COMMUNICWIONS, V a . 42, NO. 10, OCIDBER 1994

A Technique for Orthogonal Frequency Division Multiplexing Frequency Offset Correcti*on

Paul H. MQose, Member, IEEE
is being suggested as an efficient modulation for applications ranging from modems [4], to digital audio broadcast [5]. One of the principal advantages of OFDM is its utility for transmission at very nearly optimum performance in unequalized channels and in multipath channels. As described in [31451, intersymbol interference (ISI) and intercarrim interference (ICI) can be enthly eliminated by the of inserting between symbols a small time in a guard interval. The length of the guard

I. INTRODUCTION HE TECHNIQUE described in this paper has been developed to correct frequency offset errors in digital communications systems employing orthogonal frequency division multiplexing (OFDM) as the method of modulation. The aim of the paper is twofold; to show the effect offset errors have on the signal-to-noise ratio of the OFDM carriers and to present an algorithm to estimate offset so that it may be removed prior to demodulhhn. OFDM is a bandwidth efficient signalling scheme for digital communications that w s first proposed by Chang [l]. a The main difference bepeen frequency division multiplexing (FDM) and OFDM, is that in OFDM the spectrum of the individual carriers mutually overlap, giving therefore an optimun spectrum efficiency (asymptotically Q b/Hz for 2Q-ary modulation of each carrier). Nevertheless, the OFDM carriers exhibit orthogonality on a symbol interval if synthesized such t a they are spaced in frequency exactly at the reciprocal ht of the symbol interval. Fortunately, this synthesis can be accomplished perfectly, in principle, utilizing the discrete Fourier ttansform (dft) as first described by Darlington [2] and later, for data modems, by Weinstein and Ebert [3]. With the recent evolution of integrated circuit digital signal processing (dsp) chips, OFDM has become practical to implement and

Paper approved by M. J. Joindot, the Editor for Radio Communications of the I F Comtnunications Society. Manuscript received November 27,1991; EE
revised June 29,1992. The author is with the Electrical and Computer Engineering Department, Naval FWgraduate School. Monteaey, CA 93943. USA. EEE Log Number 9401947.

S symbols do not overlap because of the guard interval, I 1 is eliminated, too. One of the principal disadvantagesof O M is sensitivity to frequency offset in the channel. For example, the coded OFDM system developed by CCETT (Centre Commun. d'Etudes de Telediffusion et Telecommuications) for digital sound broadcasting to mobile receivers incorporates an AFC (automatic frequency control) loop in the receiver to reduce frequency offset caused by tuning oscillator inaccuracies and doppler shift [6]. There are two deleterious effects caused by frequency offset; one is the reduction of signal amplitude in the output of the filters matched to each of the carriers and the second is introduction of IC1 from the other carriers which are now no longer orthogonal to the filter. Because, in OFDM, the carriers are inherently closely spaced in frequency compared to the channel bandwidth, the tolerable frequency offset becomes a very small fraction of the channel bandwidth. Maintaining sufficient open loop frequency accuracy can become difficult in links, such as satellite links with multiple frequency translations or, as mentioned previously, in mobile digital radio links that may also introduce significant Doppler shift. The effects of frequency offset are presented in Section II. In Section 1 1 we present an algorithm to estimate fresuency 1, offset from the demodulated data signals in the receiver. The algorithm extends to OFDM,with important differences, a method described by Simon and Divsalar [7] for single carrier MPSK. The technique involves repetition of a data symbol and comparison of the phases of each of the carriers between the successive symbols. Since the modulation phase values are not changed, the phase shift of each of the carriers between successive repeated symbols is due to the frequency offset. The frequency offset is estimated using a maximum likelihood estimate (MLE) algorithm. Performance of the algorithm as

0090-6778/94$04.00 0 1994 IEEE

MOOSE: OFDM FREQUENCY OFFSET CORRECTION

2909

a function EJN, (individual OFDM carrier energy to one- synchronization at this stage of the receiver) leading to a sided. spectral density of additive white Gaussian noise) and received sequence as given by (4). It is assumed that the frequency offset is @lu&d in Section III. impulse response of the channel does not change (much) In the event that the frequency offset exceeds f1/2 the during the symbol plus guaTd interval (this corresponds to i n t e d e r spacing, the maximum lmt of the algorithm, a slow-fading in a radio frequency channel). iis strategy is requkd for initial acquisition. One such strategy The insertion of guard intervals renders the received carriers is described in Section IV. orthogonal on the N point symbol interval. However, the demodulation process, which is implemented with a dft (the dft is equivalent to matched filter reception in the absence of H. OFDM SLATION I A N frequency offset) is affected by frequency offset. That is, CHANNEL WITH FREQUENCY OFFSET

An OFDM transmission symbol is given by the N point complex modulation sequence

N-I

n=O

k=-K

n = 0 , 1 , 2 ,... N - 1 ;

N L 2 K + 1 . (1)

the kth element of the dft sequence, consists of three components;

It consists of 2K + 1 complex sinusoids which have been m d l t d with 2K 1 complex modulation values { X k } . ouae We note that the iadividual sinusoids are orthogonal on the

The first component is the modulation value Xk modified by the channel transfer function. This component experiences N-1 an amplitude reduction and phase shift due to the frequency = (1/N))Xk)2akl (2) offset. Since N is always much greater than TE, N sin ( m / N ) nO = may be replaced by m. The second term is the IC1 caused by the frequency offset where Z n k = ( 1 / N ) X k e 2 * j n k / . and is given by We also note that the N point discrete Fourier transform (dft) of (1) is ,the N point sequence K (XIH~){(sin7FE)/(Nsin(n(l-k E ) / N ) ) ) Ik = D m N (xn} I=-K symbol interval, that is

I f k

.,jm(N-1)/Ne-j*(1-k)/N

(8)

= {XO, X I , . X K , 0, 0 , . -. 0, x-K 0,

. ..x--2,X - l }
(3)

of modulation values. Equation (1) is the inverse discrete Fourier transform (IDFT) of (3) and defines a practical modulation-carrier synthesis technique for generating OFDM with perfect orthogonality. Atr passing through a bandpass channel, the complex fe envelope of the received q e n c e can be expressed as

In order to evaluate the statistical properties of the ICI, some further assumptions are necessary. Specifically, it will be aSSUmed that E[X&] 0 and E[X&X;]= IxI2bl&, is, = that the modulation values have zero mean and are uncorrelated. With this provision E[&] = 0, and
K
I=-K
l#k

-{sin
n = 0 , 1 , 2 ,..., N - 1
(4)

/ { N sin (.(a
K-k

- IC -I-C ) / N ) } ~ .

(9)

The average channel gain, E { l H l I 2 } = [HI2,is constant so it can be separated from the sum and (9) becomes
E[IIk12] =

where

Hk

is the transfer function of the channel at the

fresueacy of the kth carrier, E is the relative frequency offset

o the cbilmel (the ratio of the actual frequency offset to f t h e i n k er sjmcing), and UIn is the complex envelope of additive white Gaussian noise (AWGN). Let the actual symbol transmitted be the N + Ng point sequence
{XN--Ng,..-rxN-2, Z N - 1 , 20, Z l , . . - , ~ N - l }

IX121H12{sinnE}2

.
p=-K-k

1/{Nsin(a(p+e)/N))2. (10)
P#O

The sum in (10) can be bounded for E = 0. It consists of 2K positive terms. The interval of the sum is contained within the longer interval -2K 5 p 5 2K, its looation dependent on k. with Ng greater than or equal to the time spread of the Recall that 2K 5 N- 1.Also note the following; the argument channel. The Ng point p m o r signal allows the received of the sum is periodic with period N, it is an even function of symbol sequence to reach steady state by n = 0 (we assume p , and it is even about p = N / 2 . Thus the 2K terms of the sum

(5)

2910

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 42, NO. 10, OCIOBER 1994

m. m Q U E N C Y OFFSET ESTIMATION
Expcrimnt: ++++++ Thwry (Lowrr BoMd): sdidlinca

If an OFDM transmission symbol is repeated, one receives, ih the absence of noise, the 2N point sequence
r n = (1/N)

IK
2N-1

XkHkeznjnk+e)lN i
= 0, 1,...,2 N - 1. (16)

k=-K

The kth element of the N point dft of the first N points of (16) is
N-1
5~

am

0.1

ais

a2

as

a3

0.35

0.4

0.4s

0.5

Rlk = x r n e - z z j n k / N ; k = 0, 1, 2,. ..,N - 1, (17)

n=O

Fig. 1. SNR versus relative frequency offset.

are a subset of the N terms in the intervals -N/2 5 p 5 -1 and 1 5 p 5 N/2 for every k. Consequently,

and the kth element of the dft of the second half of the sequence is

R~~=
n=N N -1

rne-2xjnklN
rn+Ne-2jnkIN; k = 0, 1,
n=O

I.

.,N - 1.(18)
(19)

Observe that (sin?rp/N) 2 (2p/N)2 for lpl 5 N/2. Therefore,

N f2

But from (16),

rn+N

= rneznje--f R2k = R1keznj.

2 C 1 / ( N sin (rP/2
p=l
N/2

Including the AWGN one obtains

Observe that between the first and second DWs, both the IC1 and the signal are altered in exactly the same way, by upper bounds the sum for small E. Numerically, we have a phase shift proportional to frequency offset. Therefore, if determined that the sum in (10) is bounded by 0.5947 for offset E is estimated using observations (20) it is possible to obtain accurate estimates even when the offset is too large for E < 0.5 so that satisfactory data demodulation. E[11k12] 5 0.59471X121H12(~in~~)2; 5 0.5 161 (13) It is shown in the Appendix that the maximum likelihood estimate (MLE) of E is given by upper bounds the variance of the IC1 for values of frequency offset up to plus or minus one half the carrier spacing. ; (1/27r)tan-l = Equation (13) may be used to give a lower bound for the S N R at the output of the dft for the OFDM carriers in a channel with AWGN and frequency offset. Thus, This is an intuitively satisfying result since, in the absence of SNR 2 1x121H12{sin?rE/(TE)}2/ E noise, the angle of Y2kYik is ~ A for each k. Fig. 2 shows {0.5947)X121H12(sinx)2 E[Iwk12]). (14) simulation results for the estimate of E obtained using (21) for It is easily established that ~ x ~ 2 ~ ~ ~ 2 = ~ [ ~ ~ versus2E ] values of E,/No corresponding to 17 and 5 dB. / &/No k ~ where E, is the averaged received energy of the individual carriers and N0/2 is the power spectral density of the AWGN A. Statistical Properties of the Estimate in the bandpass transmission channel. Therefore, (14) may be The conditional mean and variance of i given E and {&} more conveniently expressed as can be approximated as follows. Consider the complex products Y2kYck from which we estimate E. For a given E, subtract L {~c/~o){s~~E/(~E)}2/ the correspondingphase, ~ K Efrom each product to obtain the , (1 0.5947(E,/N,)(~inm)~} (15) tangent of the phase error K with equality at E = 0. Quation (15) is plotted in Fig. 1 versus fresuency offset E for values of E,/N, equal to 11, 17, 23, tan [2?r(i- E)]= and 29 dB. Simulation results for N = 256, K = 96, and 8 PSK modulation in a flat channel are included for validation. It can be seen that the bound of (15) is quite tight for small values of E but is about 3 dB too low at E = 0.5. p=l
Zp=1

< 2c1/(2p) < - c l / p 2

loo

= 2 / 1 2 = 0.882

(12)

MOOSE O?DM FREQUENCY OFFSET CORRECTION

2911

0.5

' 0 o.os ai 0.15 a2 0.25 03 a35 a4 0.6 a5 . F g 2. MLE relative frequency offset estimate versus relative frequency i.
Offset

Fig. 3. MLE relative frequency o f e estimate error standard deviation fst versus EsINo.

E = 0 and 0.45. The theoretical standard deviation from (25) is plotted for comparison. We conclude that (21) will give very accurate estimates of the relative frequency offset E. Under i - E "N (1/27r) normal conditions for communicationssignalling, the accuracy is sufficient to correct E to well within tolerances (see (14) and Fig. 1) for negligible losses due any residual error in the offset estimate. The limits of accurate estimation by (21) are 161 10.5, that is, f112 the intercarrier spacing. As E 40.5, 2 may, due to noise and the discontinuity of the arctangent, jump to -0.5. (23) When this happens the estimate is no longer unbiased and, At high signal-to-noiseratios, a condition compatible with suc- in practice, it becomes useless. Thus, for frequency offsets cessful communikations signaling, (23) may be approximated exceeding one half the carrier spacing, an initial acquisition strategy must be prescribed. One such strategy is discussed in bY I Section IV. I 2 - E M (1/27$

For 1 - E ( << 1/2n, ?he tangent can be approximated by its 2 argument so that

B. Frequency W s e t Estimution in a Multipath Channel

It is evident from (25) and (26) that the mean and variance of the offset estimate do not depend on the actual received frequency coefficients {Rk}. Furthermore, if the symbol pair has been received through an unknown multipath channel, and as described in Section I it has been preceded by a periodic from which we find that precursor of length Ng 2 N , the time spread of the channel t E[; - I , {Rk}] = 0. (25) then the carriers remain a their steady state values throughout the duration of both symbols because the modulation values Therefore, for small errors, the estimate is conditionally un- are repeated. Thus, as no guard interval is required between biased. the symbol pair, the algorithm of (21) can be used without Tk conditional variance of the estimate is easily determined modification. for (24). Fig. 4 shows six amplitude responses of a channel with five multipaths whose arrival times have been uniformly distributed vu[; 1 % {Rk}] 1/{(2r)2(Es/No)} = (26) over an interval T, = T/16. The paths have equal weight and random phases so that the overall channel exhibits frequency where selective Rayleigh fading as is evident b m the figure. Fig. 5 N-I shows estimates of E f o (21) with Ng = N/16 for the rm same conditions as Fig. 2. It can be seen that the estimate is n=O unaffected by the multipath. is the t t l symbol energy. Since the total energy is the sum oa o the energies o the 2K 1 carriers, the error variance f f IV. ACQUISITION of the offset estimate wl in practice be very low. Fig. 3 il shows the sample standard deviation of the error in the relative In the event that the frequency offset is greater than f1/2 f"cy offset estimate for 100 simulation trials of (21) the carrier spacing, a strategy for initial acquisition to bring fst versus E,/N, for 2K 1 =193 carriers and for o f e s of the offset within the limits of the algorithm must be developed.
/ K
\ \

2912

IEEE TRANSACTIONS ON C O M M U " S , VOL. 42, NO. 10, OCTOBEX 1994

I
10

I
50 100

4 0

-100

-50

150

Fig.4. Multipathf

ion channel.

'r/
035

05,

a3

au
ai

repeated symbols by a repetition of the fist full length data symboi or by the use of an AFC loop as shown in [6]. To illustrate, consider the following example for a digital audio broadcasting service. Assume an intercarrier spacing for the data stream of 1 kHz, and a frequency offset uncertainty in the system dominated by the long term accuracy of the oscillators in the receiver that heterodyne the received signal to IF and quadrature demodulate to obtain the complex envelope. Assume VHF radio transmission at 150 MHz and overall oscillator uncertainty (long term stability) of 1 part in lo5. Thus,,S , =1500 Hz and Afinitial must be greater than 3000 Hz.At regular intervals in the data stream of 1 ms (plus guard interval) symbols insert a short symbol of length 250 ps (4 kHz carrier spacing) and repeat it once. From this repeated shortened symbol estimate E. Assume Ec/No is 11 dB for the data symbols and that there are 200 carriers, so that Es/No is 34 d33. The shortened symbols have 114th energy so Es/No is only 28 dB for the initial estimate of E . From (25) we find that Uainitial = 0.0063 SO that Uedata = 0.025. From (15) we See that residual offset errors of this magnitude cause IC1 resulting in a signal-to-interferenceratio of 24.3 dB in the data symbols for only 0.2 dB loss in overall S N R at &/No of 11 dB. This example illustrates a situation for which the initial offset acquisition estimate with shortened data symbols is of sufficient accuracy that refinement of the estimate with longer symbols is not necessary.
V. DISCUSSION CONCLUSIONS AND

Fi 5. MLE relative frequency offset estimate versus relative frequency o& i a multipatb channel. n

We envision that, if continuous, the OFDM symbol stream will be punctuated at appropriate intervals with repeated symbols. A continuous symbol stream occurs in applications such as d g t l audio broadcast [53. A second possibility is that OFDM iia modulation is used in session oriented digital data or voice communications such as digital radio [8]. Here, we envision that the session initiation interval will include one or more lepemzd symbols. The basic strategy for initial fresuency offset acquisition, in &her case, is to shorten the dft's and use larger carrier spacings such that the phase shift does not exceed f ? r . The fkqtwncy offset in Hz is S = c/T = EA where A f is the f intercarrier spacing and T is the symbol interval. Let us assume that the initial frequency offset is no greater than *Smm. Then

determines the minimum initial carrier spacing, and corresponding dft lengths. If the average power of the shortened symbols is kept the same, the variance of the estimate of initial w l ' e larger than for the longer data symbols since there ilb is less symbol energy. Also, the offset estimate error for the shortened symbols, since it estimates the fraction of carrier spacing, corresponds to a proportionately larger fractional offwt for the longer data symbols. However, the MLE estimate i s accurate that in practice the i i i l estimate still may s a nta be adequate. If not, it is refined by following the shortened

We have seen that, as expected, frequency offset in OFDM causes serious loss of S N R of the dft outputs due primarily to ICI. A lower bound for S N R has been derived and simulation results show that the bound is quite accurate for small offsets, but about 3 dB too pessimistic as the offset approaches 1/2 the carrier spacing. An algorithm for maximum l i k e l i h d estimate of frequency offset using the dft values of a repeated symbol has been presented. It has been shown that for small error in the estimate, the estimate is conditionally unbiased and is consistent in the sense that the variance is inversely proportional to the number of carriers in the OFDM signal. Furthermore, both the signal values and the IC1 contribute coherently to the estimate so that it is possible to obtain very ht accurate estimates even when the offset is too great, t a is there is too much ICI, to demodulate the data values. Since the estimation error depends only on total symbol energy, the algorithm works equally well in multipath spread channels. However, it is required that the frequency offset as well as the channel impulse response be constant for a period of two symbols. The accuracy required of frequency offset correction depends on how much residual offset can be tolerated. Offset induced IC1 can be treated quite satisfactorily as additional AWGN since its source is the multiplicity of other OFDM carriers that are zero mean and uncorrelated random processes. 1 Note W the SNR defined in Section 1 [see (14) and (15)] is just &/No in the absence of offset. Thus we may interpret (15) as the effective &/No of the carriers, or if divided

T :

2913

by tbe number of bits encoded in each of the carriers, their effective Eb/No. Required &,/No course depends qxm the of maddation constellation, the fading statistics of the chamel the forward error control coding, if any, employed in the OPDM system and t e desired BER (see, for example, [5, h Figs. 11-13]). Tbe acquisition range of the algorithm presented here is f / the intercarrier spacing of the repeated symbol. It is 12 independeat of the modulation constellations chosen for the carriers and whether the symbols are coherently or differentially encoded. The AFC loop shown in [6] does not require a repeated symbol. However its acquisition range is only f 1 / 2 m of the intercarrier spacing for m-ary PSK.The initial frequency t h offset a te time of the initiation of the communicationsession may be greater than 1/2 the intercarrier spacing and thus even outside the range of the MLE algorithm. In this event, a strategy is required for initial acquisition. We propose to use a pair of shortened data symbols whose carrier spacing is sufticiently large to insure that the algorithm will operate within its range. Due t the low variance of the initial estimate, o fiuther refinement will normally not be required. It may be advantageous to use shortened repeated symbols for tracking offset variations too, instead of an AFC loop, because this reduces the time during which the channel must be stable.
APPENDIX
MAXIMUM LIKELIHOOD ESTIMATE OF DIFFERENTIAL

To find the conditional density function in (A.8), note that

Y = (Y1- W l ) H ( @ )+ W 2 2
so that
Y2

(A.9)

= Y i H ( 8 ) W2 - W i H ( 8 ) .

(A.10)

If W1 and W2 are Gaussian, zero mean white random vectors with variance u2,then the conditional density function in (A.6) is multivariate Gaussian with mean value vector Y l H ( 8 ) and 2M x 2M covariance matrix K = E [ ( W 2 - W l H ( Q ) ) t ( W 2- w , H ( e ) ) ] = 2 2 1 . (A. 11) We note that K is independent of 8, therefore,

6 = megx[f(y2 I 8, = m $ [ ~ ( e ) J Y,)]
with

(A.12)

J ( @ ) = (Y2 - Y i H ( @ ) ) ( Y 2- Y I H ( B ) ) ~ . (A.13)
Using the fact that

H(8)[dH(8)/d8]t [dH(8)/d8]Ht(8=0 + )
we can find that

(A.14)

PHASE

Let M complex values {Zk}be represented by a length 2M row vector

dJ(B)/dQ = -Y2[dH(Q)/d8ltYt -Yl[dH(O)/dQ]Yd. (A.15) Using (A.4), it follows directly that (A.15) is identically zero when 8 = Q such that
sin (
~ ) F ~ R Y ~ YR I Y ~ , , ] COS ( 6 ) [ ~ 2 r y 4 R Y~RY,~,I. ~ = -

Z = [ZIR Z ~ R Z M R Zir .
*

221

* *

.Z M I ]
(A.1)

= [ Z R ZZ].

(A. 16) Therefore,

Consider the random vectors

Y1 =R1

+Wl

Y2 = R l H ( 6 )

W2

(A.2) (A.3) (A.17)

where

W) =

[ .
4

c=COS(Q)I

& S = sin(8)I

is the maximum likelihood estimate (MLE) of 0.

04.4) ACKNOWLEDGMENT is a 2M x 2M rotation matrix. The maximum likelihood estimate of the parameter 8, given the observations Y 1 and The author would like to thank Dr. S. Pupolin of University Y2 (see, example, Sage and Melsa, [9, p. 1961) is the value of Padova for his helpful suggestions to the original manufor of 8 that maximizes the conditional joint density function of script. The author is also indebted to the anonymous reviewers the observations. That is whose constructive remarks have improved the quality of this Paper. 6 = m8a [ f ( Y i ,Y2 I Q)] (A.5) which can be written as
REFERENCES

e = m e [ f ( Y 2I Q, YI)f(YlI 8 1 1.
But 9 gives no information about Y 1 ,that is
f W 1

(A.6)

[I] R. W. Chang, Synthesis of band-limited orthogonal signals for multichannel data transmission, Bell Syst. Tech. J., vol. 45, pp. 1775-1796.

Dec. 1%.

I Q) = W

l >

(A.7)

[2] S. Darling, Ondigital single-sidebandmodulators,IEEE Truns. Circuit Theoty, V O ~ .CT-17, W.409-414, Aug. 1970. [3] S. B. Weinstein and P. M. Ebert, Data transmission by frequency-

so that
Q

division multiplexingusing the discrcte Fourier transform, IEEE Trans. C o m n . Technol., vol. COM-19, pp. 62-34. Oct. 1971. [4] J. A. C. Bingham, Multicarrier modulation for data transmission: An idea whose time has come, ZEEE Commun. Mug., vol. 28, pp. 17-25,

= m g W 2 I 8, l ) ] . Y

(A.8)

M r 1990. a.

p r----J------

--__

~ _ _

--

__

__

2914

IEEE TRANSACTIONS ON COMMUNICATIONS. VOL. 42, NO. 10, OClDBER 1994

[SI

M.N r and R Halbert, Principlesof modulation and channel coding ad

for digital broadcasting for mobile receivers, EBU Rev., no. 224, pp. 3-25, Aug. 1989. [6] B. LeFloch, R. Halbert-Lassalle, and D. Castelain, i i a sound Dgtl bmdcashg t mbde receivers, IEEE Trans. Conswn Elec.. vol. 35. o no. 3, pp. 493-503, Aug. 1989. [fl M. K. Simon and D. Divsalar, Doppleramted diffmntial detection of MPSK, IEEE Tram. Commun.. vol. 37, no. 2, pp. 99-109, Feb. 1989. [8] D. C. Cox, W. S. M o d , and H. Sherry, Low-power digital radigas a ubiphus subscriber loop, IEEE Commun. Mag., vol. 29, pp. 92-95, Mar. 1991. [9] A. P. Sage and J. L. Melsa, Esrimarion Theory with Applications to YO& MCCraw-Hill, 1971. Communications nnd C~nhOl. NCW

Paul M. Moose (M79) was born in Omaha, NE, on July 22. 1938. He received the B.S.,M S , . . and Ph.D. d e w s in electrical engineering i 1960, n rm 1966. and 1970, respectively, f o the University of W s i g o .Seattle. ahntn Since 1977 he has been with the Department of
Electrical and Computer Engineering. Naval Postgraduate School, Monterey, CA, where he is an A& s Professor of Ekctrical and Computer Engineeaing. His research inters& are in digital communications. He is a -founder of and consultant ~ Digital Co-~ni~ationS h . Monterey, CA. r y c,

for ~