IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 42, NO.

21314, FEBRUARYNARCHIAPRIL 1994 1325

Performance Analysis of a Coherent Frequency

Hopped Spread-Spectrum System with Multipath
Channel Equalization in the Presence of Jamming
Valentin Lupu and Laurence B. Milstein, Fellow, ZEEE

Abstract- A method of equalization for a coherent frequency 11. SYSTEM MODELS

hopped spread-spectrum receiver, operating in a heavily
multipath-contaminated channel, is presented. The perfor- A. Discrete Representation
mance in the presence of thermal noise and either noise or
multi-tone jamming is obtained and compared with the perfor-
Discrete representations of the transmitted waveforms, the
mance of a noncoherent frequency hopped spread-spectrum sys-
tem operating in the same environment. It is shown that channel impulse response and the received signals are used,
significant improvement in performance, in the presence of based on the dimensionality of these temporal functions
partial-band jamming, is achieved, primarily due to the equali- (time-bandwidth product). Any temporal function g ( t ) of
zation. duration Tg,essentially bandlimited to B and centered at the
angular center frequency wc, has a complex Ng = TgBg
1. INTRODUCTION dimensional vector representation, g(Ng is assumed to be an
integer). The components of g, g 1, g2...,g,, are the samples
The performance of noncoherent frequency hopped (FH) of the complex envelope of g(t), taken at the Nyquist rate
spread-spectrum systems has been extensively analyzed. [81. The orthogonal basis which approximately spans the
Coherent frequency hopped spread-spectrum (C-FHSS) sys- complex envelope of the temporal function g ( t ) is given by
tems, assumed to be not feasible for practical implementation
over realistic channels, have been considered and analyzed in
{@ i ( t ) } = a sinc [zBg(t- i/Bg)] exp (jw,t),
a sequence of papers published only in the last decade [l-71. i = (1,2, ...,Ng), (2.1)
A joint pha.se/timing tracking loop receiver, used in conjunc-
where sinc(x) = sin (x)/x. The following relationships are
tion with channel equalization for channels with phase distor-
defined:
tion, is proposed and analyzed in [6] and [7]. Negligible
intersymbol interference (ISI) is assumed in these works. TI
'
In this paper, a time-varying multipath channel, leading to <gm * gk >= de,(t)gk(rklr= (&I * g h (2.2)
significant intersymbol interference, is considered in conjunc-
tion with a C-FHSS system operating in the presence of both where c * > stands for the inner product in the temporal func-
thermal noise and either noise or multi-tone jamming. In tion space, * stands for the complex conjugate, and ( stands e )

what follows, we present the design and an approximate per- for the scalar product in the vector space.
formance evaluation of an equalized system for the multipath
channel. Simulated system performance of the equalized sys- B. TransmitterModel
tem operating in the presence of noise jamming is computed The information source delivers equiprobable symbols
and compared with the performance of a nonequalized C- [ui}, ui E k 1, at a rate B = 1/T. The symbols are assumed D
FHSS system, as well as with the performance of a non- be independent, identically distributed random variables.
coherent frequency hopped spread-spectrum system (NC- The corresponding impulse series ~ u ; & t - i T )excites the
FHSS). i
transmitter baseband filter whose impulse response, P T ( t ) , is
1,OltlT
Paper approved by Evaggelos Geraniotis, the Editor for Spread Spectrum (2.3)
of the IEEE CommunicationsSociety. Manuscript received January 8,1991;
revised February 20,1992 and March 1,1993.
This work was partially supported by the Office of Naval Research under The baseband filter output, Dip&- in, is translated to the
Grant "014-91-J-1234. i
V. Lupu is with Rafael, in Haifa, Israel. channel center frequency and spread by the coherent fre-
L. B. Milstein is with the Department of Electrical and Complter En- quency hopping waveform
gineering, University of California at San Diego, La Jolla, CA 92093.
IEEELog Number 9401031. v ( t )= z c o s [[ao
+ f ( i ) w ~ ] ( t - i T ) } p ~ ( f - i T )(2.4)
,
i
0090-6778/94$04.000 1994 IEEE

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on November 07,2023 at 17:42:18 UTC from IEEE Xplore. Restrictions apply.
1326 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 42, NO 21314, FEBRUARYMARCWAPRIL 1994

where w 0 &RM/T, M is an integer, W R A22r/T is the angular paths and we normalize the indirect path attenuations with
reference frequency, and f(i) is a function which maps the respect to this one;
integer sequence (i) into an integer pseudo-random b) the indirect path delays are related to the direct path delay,
sequence. That is, f ( i ) ~{ 1,2,3, ..., N ) , such that and the latter one is taken to be zero.
Prob c f ( i )= k ] = 1/N for k E { 1,2, 3, ...,N } , where N is the
number of the transmitted frequency slots, and equals 2" for
D. Noise and Jammer Model
some integer n. (N is also the system processing gain.) Thus,
for long sequences of consecutive integers, f(i) is uniformly Additive Noise. The channel noise is additive white Gaussian
distributed over the integer set (1,2, 3,...,I?). The noise (AWGN) with two-sided power spectral density No/2.
transmitter center frequency, fc, is given by
Jammer Model. The received signal is corrupted by a partial
fc = (M +N/2)/T, where M >N. (2.5) band jammer, covering some fraction of the band (not neces-
Each period T, the transmitter sends one of the signals from sarily contiguous), given by Bl = yB, 0 < y 5 1. The total
the following set, employing BPSK modulation: jammer power is given by Pi, and we distinguish two dif-
ferent jammer models.
) [OO+ i ~ ~ ) t i]=, 1,2,3 ,...,N.
si(t) = ' b E / T p ~ ( tCOS (2.6)
The set is obviously timelimited and has the energy approxi- a) Noise Jammer. The noise jammer has its power uniformly
mately contained within the frequency band B = N / T cen- distributed over each Jammed frequency slot, and its one-
tered at the angular frequency W, = kfc. The corresponding sided power spectral density is given by N j = Pj/yB.
vector representation sihas the k-th component s% given by b) Multi-Tone Jammer. The multi-tone jammer consists of
sik = ~ e x p ~ ~ ( N - - 2 i ) k / Nk] =, 1,2,3, ...,N. (2.7) yN jamming tones of the same power Pj/yN. The vector
representation of the complex envelope of the multi-tone
It is evident that <s, *sl> = (s, = 61
,,
as;)where Sm1is the jammer is of the form
Kronecker delta. The set {si} is orthogonal and spans the
vector space of the temporal bandpass functions having ji = U1oJ11 , . . . ,ji.N-11,
dimensionality N = BT and centered at a,.
where jik= d m p iexp Ijn(N-2i)/T+Oi1, (2.9)
N
C. Channel Propagation Model Bi = 0,l and C p i= yN. The Bi are independent random
i=l
The channel propagation model is derived according to variables, uniformly distributed over ( 0 , a ) .
the following assumptions:
a) the bandwidth of the channel is essentially limited to the E. Channel Model
transmitter frequency band of interest, B = N/T.
b) the physical channel impulse response, C(t), is timelimited The received signal r ( t ) is given by
to a duration T,, such that 0 I T, I L T . The following
definition for the channel impulse response is used: r(t)=s(t)@c(t)+n(t)+j(t), (2.10)

C ( t ) ; O I t I T, where 8 stands for the convolution operation. The channel

(2.8) response, at a given instant iT I t I (i+ 1)T, to the source
'@)={0 ; T, I t I L T . symbol ui is not only a function of the transmitted waveform
u , s ~ ( ~ ) (but
~ ) , depends also of the past history of the last L
In other words, we will design our equalizer to handle a max-
transmitted waveforms ui-ksf(z-kj(t-iT+kT), 1 I k I L. The
imum delay spread of LT seconds.
channel response to a transmitted signal ui-Lsf(i-Lj(t) can be
The channel impulse response, c (t),has LBT = LN dimen-
represented by a vector of dimension
sions, and is represented by an LN-component complex vec-
m ' = N + L N - l = N ( L + l ) - l . For ease of computation, we
tor c = ( a o , cyl , . . . , We note that the multipath
sin (x) will use for the channel response vector representation the
channel can be approximated by a s u m of delayed - dimensionality of m = N ( L + l), which is the sum of the vec-
X
functions with the minimum distinguishable differential path tor dimensions involved in the convolution operation. The
delay of 1/B [8]. The model can be described by LN com- convolution operation can be performed in the vector space
plex numbers {ai ] such that using the circular convolution operation [9]. The channel
a) I aiI represents the i-th path relative magnitude; response to the transmitted signal ui-~s,-+~)(t),in vector
b) the i-th path relative delay is given by zi= i/B = iT/M, representation, is then given by
c) the phase @ i =@(ai) = w,zi represents the relative RF car- ri-L = ui-Lsf(i-LjY +n+j,-(i-L), (2.11)
rier phase of the i-th path signal with respect to that of the
direct path (i = 0). where sf(i-L)is an m-component row vector obtained from
In the succeeding derivation, we assume the following: sf (i-L)by NL zero padding, Y is an m x m matrix called the
a) the direct path (i = 0, minimum delay path relative to the convolution matrix and is obtained by a cyclic shift to the
transmitter) has the minimum attentuation among all possible right of the m-component row vector yo ([91), wh&e

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on November 07,2023 at 17:42:18 UTC from IEEE Xplore. Restrictions apply.
 LUPU AND MILSTEIN: FWSS SYSTEM WITH MULTIPATH CHANNEL EQUALIZAnON
1327

a1 , .. ., CCNL-1, 0, 0, ..., 0).

yo = (ao, (2.12) where &m(is given by
Thus, yo is obtained from the channel impulse response vec-
tor c by padding N additional zeros. The convolution result
has one zero-component in excess as a result of using m In vector notation, (2.15) becomes
instead of m' dimensions. The excess zero-component does
' L
not affect the final result. 4= (ri.s;(i)) = ui-k&(0+ni, (2.17)
Consider the response of the channel to the signal k=O
Ui-LSf(i-L)(t)in the interval iT I t I (i+ 1)T. Only the last N where ni is a Gaussian random variable with conditional
vector components of ri-L are needed for this calculation, mean given by
and they are determined by the last N columns of the matrix
Y . Similarly, it is possible to compute the channel response E(ni)=
to each of the signals which affect the total channel output in
the interval iT I t I (i + 1)T. The only difference is in the
vector representation of the transmitted signals. Since the
subsequent signals are delayed in time by multiples of T, the and conditional variance
portion of interest of the channel response is shifted to the N o / E+ Bf oPj/yBE ;noisejammer
left by the appropriate amount of time (or, in the vector (2.19)
representation, by the appropriate number of components). a;(i)= {No/. ;multi-tonejammer
Using the properties of the circular convolution, the segment
of interest in the response can be found by either shifting to The notation r denotes the channel state, and thus, the last
the left the rows of the matrix W or by shifting to the right L + 1 transmitted frequencies. For any channel state r, there
the signal vector. We consider the second option, in which is a corresponding set of coefficients { z k m } ,
the transmitted vector, padded with NL zeros, contains the k = 0 , 1, 2, ...,L, such that Zo(r)denotes the desired input
signal vector at a shifted position. The padded and shifted to signal multiplier, corresponding to ui, and any other
the right vectors, originating from the transmitted signals coefficient Zk(r)relates to the intersymbol interference due
Sf(i-L)(t),0 I k I L, are denoted s)j;!$", where the super- to the transmitted symbol tlj-k.
script denotes the amount of right-shifting of the extended
vector (sf(i-k), 0, 0, ..., 0). Thus, the vector is 111. CHANNEL EQUALIZATION
(L-k+l)
S f(i-k) =(O, 0,+..v sf(i-k),O,..., O), (2.13) A. Equalization Method

where 0 denotes and N-component zero vector and sf (i-k) is As expression (2.15) suggests, a narrow-band equalization
preceded by L-k 0's and succeeded by k 0's. Since there are of each channel state is possible by using transversal filters.
N distinct transmitted vectors distinct extended vectors and Since the channel characteristic is unknown and time-
L+1 different positions in the extended vectors, there are varying, an adaptive equalizer filter is considered for each
N(L+ 1) distinct extended vectors orthogonal to each other state based on the LMS algorithm. The coefficients of each
and having norm E. equalizer are stored in a RAM (random access memory) and
updated each time the system revisits the respective state.
We denote the matrix formed by the last N columns of
The same coefficient adjustment algorithm is considered for
the circulant matrix and having dimension N (L + 1) x N by
all the state equalizers.
C. The signal, present in the time interval iT S t I (i + l)T at A simple method for estimating the number of channel
the receiver input, is the linear combination
states and, therefore, the RAM requirements, is as follows:
We consider a periodic pseudo-random generator
f ( i ) =f (i+ Q),Q being the generator period. Then, the total
number of states cannot exceed the number of generator
F. Receiver Model states, Q. A channel state is defined by the last L + l
transmitted frequencies and each frequency can take on N
The receiver to be analyzed uses a dehopping comelator, different values; therefore, the total number of states is upper
and we assume perfect bit-timing tracking. The perfect bounded by NL+'. Hence,
knowledge of T and its timing allows us to synthesize the N, I min (NL+', Q), (3.1)
dehopping replicas needed for the correlation implementa-
tion. The correlator output sampled at t = (i + l)T is where N, stands for the number of channel states. Usually,
(i+)T Q > N and, therefore, for systems of practical interest, the
L
r; =
IT
r(t)sf(i)(t)dt= c
k=O
ui-&zk(r)
number of states grows exponentially with the channel
memory (measured in hop time-duration units).
(i+l)T
+WE) [n (t)+j(t)lsf(i)(t)dt, (2.15)
I

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on November 07,2023 at 17:42:18 UTC from IEEE Xplore. Restrictions apply.
1328 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL 42, NO 21314, FEBRUARYMARCHIAPRIL 1994

B. Assumptions
The channel equalization is possible if the channel charac-
teristic and the channel impairments (noise, jammer) can be
considered long-term stationary. These processes are
assumed to be wide-sense stationary in a time interval longer
than the maximum convergence time of the equalization
algorithm.
Let c (t,z)be the channel impulse response at time t due
to an impulse applied at time z. Then, the interval zd is
defined as the decorrelation separation in time and represents
a measure of the fading rate, or a measure of the time interval
for which the channel can be considered as almost time-
invariant. Accordingly, let z j represent the time interval for Fig. 1. Decision Feedback Equalizer Schematic Diagram
which the jammer can be considered almost time-invariant.
The following relationship, derived from [lo], describes the iii is
stationarity assumption:
E [(Gi)2]= E [(l-p,)~?+Peu?]= E(u?)= 1. (3.6)
min(zd, Tj) * ?F[P (r)W)l, (3.2) Two decisions iiand ij,originating from the same state and
where p ( r ) stands for the average number of iterations, separated in time by the interval 0' - i)T, are, approximately,
needed during the learning process at state r, in order that the statistically independent, if a decision error occurring at sym-
state equalizer coefficients will converge to the optimal tap- bol u propagates through the equalizer in less_than (j - i)
weight vector within a certain degree of accuracy; T ( r ) symb~~s. n u s , for simplicity, we assume E(uiiij) = o for
denotes the average revisiting time at state r. Since we i ij. Since the &st and the second order moments of hi do
assume a generator f(i) such that each frequency has an ci
not depend on time, is approximately WSS. The equalizer
equal occurrence probability, each state has an equal coniiguration consists of an adaptive channel state estimator
occurrence probability too. Then, the average revisiting time modeling the channel transfer function, whose realization is
is the same for any possible state and an adaptive Wiener filter performing the LMS algorithm.
The role of the channel estimator, used in the feedback path
T(T)= EV,min (zd, zi)> maxp(T)7Ns. (3.3) of the decision feedback equalizer, can be easily understood
all r
from Fig. 2.
C. Equalizer Structure
Consecutive input observables {ri} originate from dif-
ferent channel states and, therefore, any vector of consecutive
input observables is nonstationary. The nonstationarity is n
"I
easily observed from either (2.18) or (2.19), depending upon ___t_ ( C h a n n a l estimator)
the jammer type. The conditional mean, in the case of
multi-tone jamming, or the conditional variance, in the case Fig. 2. Channel Estimator Modeling.
of noise jamming, depends on time (represented by indices i
or r). In this figure, both the unknown channel and the channel esti-
The H R filter approach to the equalizer realization, which mator include the dehapping process as well as the multipath
is basically an adaptive Wiener filter performing the LMS channel model. Under the assumption of low probability of
algorithm, has been analyzed in the past for stationary input error, the decision feedback equalizer is unconditionally
signals. For this system, the input vector is nonstationary; stable and, therefore, practically suited for adaptive equaliza-
however, the decision feedback equalizer, as shown in Fig. 1, tion of slowly varying multipath channels. When a decision
provides to the equalizer an approximately wide sense sta- error occurs, the error propagates because of the memory
tionary (WSS) input {izi}. involved in the estimator realization. Simulation results show
Considering a bit error rate P,, the conditional probability that the error propagation is not catastrophic [11]. Since the
density function of i i is modulation scheme is BPSK, a training period is required in
order to resolve the R radian ambiguity associated with the
f;,u(i-iI/uI)= (1- P e p ( & -ui> +P,6(Uj + UJ. (3.4) coherent demodulation.
The expected vdue of li; is therefore The algorithm which adjusts the tap-weight values of the
TDL filter is the LMS algorithm, also known as a stochastic
E(& = E&,&j) = 2,[(l -P,)u;-P,u;] gradient algorithm. The tap-weight values at state r are
updated using the recursive relationships
= (1-2Pe)Eu(u;) = 0, (3.5)
= wi +yeiRi,
where E denotes mathematical expectation. The variance of

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on November 07,2023 at 17:42:18 UTC from IEEE Xplore. Restrictions apply.
 LUPU AND MILSTEIN: FHISS SYSTEM WITH MULTIPATH CHANNEL EQUALIZATION
1329

ei = ri’-wjRT, (3.7) have the same probability of occurrence. The exact evalua-
where ei is the error between the correlator output, r/ is the tion of the probability of error, for a given state, should rake
channel estimator output, Ri is a row vector of the last L + 1 A

estimates { &], and p is a positive real constant. The indices

i + 1, i, i - 1, ...,i -L, refer to the revisiting time of state r,
and are not related to the transmitter source counting process.
The vector wi, referred to as the current channel estimate at
state r, converges in mean to the channel intersymbol
interference coefficients { z k ( r ) ] , for a suitable step-sizep, as
i approaches infinity.
The necessary and sufficient condition for the overall sta-
I YWR/-PVTl
Synthesi2er
Conirolled Clk
Conirolled Clk

bility of the LMS algorithm, in the case of a stationary input, Frequency

Synthesizer
is [123
2
O<P<<* (3.8)
Memory
x n k
k=l

where (2,) are the eigenvalues of the correlation matrix (R

having entries ( i ? ( l i , i k ) ) = { & k ] , assuming statistical Fig. 3. Suboptimal Receiver Block Diagram
independence of i, and i&. This assumption is true if the
average revisiting time T ( r ) is much larger than the channel in consideration all the possible intersymbol interference pat-
memory LT. Therefore, (R = 0, where 0 is the identity matrix terns, i.e., averaging over 2Ldifferent transmitted sequences.
of order (L+ 1) x (L+ 1). In this particular case, there is one The evaluation is complicated by occasional incorrect previ-
eigenvalue of multiplicity L + 1, i l k = 1, and the condition ous decisions, as well as by random fluctuations of the tap-
(3.8) becomes weight vector. This brute force method of evaluation will
yield an exact result but proves to be too cumbersome [13].
o < p <2/(L+1). (3.9) A simpler, but approximate, approach to bit error evaluation
The tapweight vector wi converges in mean to { z k ( r ) } . is proposed below.
. Using [12], the mean-squared error is (see the appendix)
A. P e v o m n c e in the Presence of Noise Jammer
(3.10)
We assume that the dehopping reference is phase rotated
by the estimate of the phase of Zo(T) and that the step-size
IV. C-FHSS SYSTEM PERFORMANCE ANALYSIS parameter p is small enough in order to assure negligible
fluctuations of the tap-weight component wio from the aver-
The effect of multipath propagation on the received sig- age value Zo(r). Then, the observable at the decision device
nal, after the equalization, is to multiply the incoming desired input xi, can be approximated as
signal by Z 0 ( r ) , which means to scale and to rotate the
incoming desired signal phasor. In order to make the best xi = wioiii+e;= I Z o ( r ) Iui +ei,(4.1)
decision, either the altered incoming desired phasor should be where ii= uifor small probability of error. Under the above
rotated back or the dehopping replica should be rotated ahead assumptions, the conditional mean and the conditional vari-
by the same amount. A practical C-FHSS received has to ance of xi are
keep the synthesized RF reference coherent with the
transmitter synthesizer (this could be done by tracking the E(ni 1ui)= Izo(r)lui+E(ei)=
izo(r)iui+E(ni)=
izo(r)iui
exact information rate R, as proposed in [4-71). A schematic and
block diagram of a suboptimal receiver is shown in Fig. 3.
The receiver two-dimensional correlator provides the i? [xi- 1 (zo(r)
1 ui I 2~ =E(eief)
complex observable to the state equalizers. The correct
phase for each state could be obtained by computing the esti-
mate of the Z0(r)phase from the tap-weight component, respectively. We define the signal-to-noise ratio at the deci-
wo(T),and averaging in order to reduce the variance due to sion device input to be the ratio of the square of the condi-
the presence of channel impairments (noise, jamming) and a tional expected value of xi to the conditional variance of xi,
random information sequence. The effect of the multipath approximated by
can be either constructive, if IZo(T)I > 1, or destructive, if
IZo(r)I < 1. Each state has a different probability of bit
error, and the average probability of bit error is the arithmetic (4.3)
average of the state bit error rate, provided that the states
The decision S N R may be rewritten as

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on November 07,2023 at 17:42:18 UTC from IEEE Xplore. Restrictions apply.
1330 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL 42, NO 21314, FEBRUARYIMARCHIAPRIL 1994

knowledge and, consequently, the imperfect RF synchroniza-

tion, affect the performance of the equalized receiver. The
where
performance degradation of the equalized receiver due to the
imperfect bit synchronization is beyond the scope of this
(44.4) work; however, the degradation can be lowered to a negligi-
ble value by narrowing the synchronization loop bandwidth.
The decision SNR, and subsequently the bit error rate, are In the convergence of the equalization algorithm, we
traded-off with the convergence rate, since both of them assume that the bit synchronizer is in the tracking mode. No
depend on the step-size factor p . The average probability of joint bit timing acquisition\equalizer convergence strategy is
error is obtained by averaging the state probability of error considered in this work, and therefore we assume that these
over all possible states. That is, two tasks are sequentially performed by the receiver shown
in Fig. 3. Since the bit timing acquisition requires a training
period [4],as the equalizer does, there is no additional
hardware needed in order to implement a practical receiver.
where N, is the total number of states and N s ( r ) represents
the occurrence number of state I- within a generator period Q. V. NC-FHSS SYSTEM PERFORMANCE ANALYSIS
The function BER [SNR]denotes the BER dependence on the
signal-to-noise at the decision device input. In Section VI, For comparison purposes, a non-coherent FH system is
BER [SNR] will be approximated by employing a Gaussian analyzed. We assume that the system bandwidth expansion,
assumption, and the resulting performance will be compared i.e., the number of frequency slots, N, is a system constraint.
to that obtained by computer simulation.
A. Frequency Allocation
B. Performance in the Presence of Multi-Tone Jammer We consider a partition of the total frequency bandwidth
In the presence of a multi-tone jammer, the conditional into two equal and contiguous parts, B l and B 2 , such that
expected value of xi is altered by the projection of the jam- a) mi = mo + k i / T E B 1 if i I N/2; otherwise m iE B z ;
mer phasor on the rotated dehopping replica: b) the pseudo-random generator f ( i ) maps the sequence of
integers { i } into the set of integers [ l ,2, ..., N/2], and N is
I ui)= Izoo-)I ui+ p sm“4 re(r)-q(r)i, assumed to be an even integer;
where c) if the symbol ui = - 1 is to be transmitted, then
m f ( i ) = 00 +%f (i)/T is chosen; otherwise
of(i) = wo+%If(i)+N/2]/Tis chosen.
No/E
Also, var ( X i IU i ) = (4.6) B. Noncoherent Receiver Structure
1 - 0.5p(L + 1)
The conditional signal-to-noise ratio at the decision device Two complex-valued correlators, using as references the
input depends on the sign of u;. That is, dehopping waveforms corresponding to f ( i ) and f ( i )+N/2,
respectively, and with additional phases $ f ( i ) ( t ) and
*
SNR (I- I ui = 1) (bf (i)+Nn(t), yield two-complex-valued observables ri’(Ho)
- { I Z 0 0 7 I + ~ P ( r ) c o rs~ ( n - 4 m 1 } 2 E and ri’(H1). The phases $ f ( i ) ( l ) and $f(i)+Nn(t) are two
independent random processes considered constant over the
NO symbol time duration T and uniformly distributed over
[1-0.5p(L + l)].
* (4.7) (O,%). The observablles ri’(Ho) and ri’(H1) correspond to
the two hypotheses H o = { ui= - 1 was sent) and
Finally, the average bit error rate is given by H1 = {ui= 1 was sent}, respectively. The envelopes of the
observables are used to make final decision.

C. NC-FHSS Receiver Performance

The observables can be expressed as
The theoretical performance, developed above for both L
noise and multi-tme jamming, is based upon the assumption ri‘(Hm) = Zk(ri,m)exp [ - j @ f ( i ) e d / 2 1 +ni(Hm), (5.1)
k=O
that the bit duration T, as well as the bit timing, are perfectly
known. A practical receiver has to perform both bit timing where m = 0, 1. The conditional probability density function
acquisition and tracking, using a bit synchronizer, as shown of ri’(Hm)is a normal p.d.f. having a conditional mean and a
on Fig. 3 . Inherently, the bit timing, under channel impair- conditional variance, conditioned on $ f ( i ) t m ~
and
n 8f(r)+mlvn,
ments like thermal noise and jamming, is a stochastic process which depend on whether the observable was obtained from
rather than a deterministic value. The imperfect timing the correct or the incorrect correlator. Let H , and Hhc denote

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on November 07,2023 at 17:42:18 UTC from IEEE Xplore. Restrictions apply.
 LUPU AND MILSTEIN: FWSS SYSTEM WITH MULTIPATH CHANNEL EQUALIZATION 1331

that the observable is obtained from the correct or incorrect

correlator, respectively. We have at the correct correlator
output
E(ni I#f(i.m),ef(i,m)*Hc) (5.8)
L
;noise jammer

(5.2)

and
o2tni I #f(i,m),ej(i,m),Hc)
- r / E +Pf(i,m)Pj/yBE ;noise jammer
- No/E

where f (i,m) = f (i) + mN/2.

;multi-tone jammer, (5.3)

and
I exp [ j O f ( i , I ; multi-tone

The output of the incorrect comlator has similar condi-

tional mean and conditional variance as does the correct Qi,inc(xinc I8j(i,1-m)

correlator output, but {Z(r)} is replaced by the set =02(ni I ef(i,1-m)r + f ( i , l-m), ~ i n c ~ 2 . (5.11)
{ uk(ri,l-m)}, obtained from
A decision error occurs when xi,, <xi,inc. The conditional
probability of error is then given by [141
BERNCO
The set { Uk(ri, represents the coefficients of the inter- = QM(u,b)- [c/(c+ 1)1exp [-(u2+b2)/2]Zo(ub), (5.12)
symbol interference due to the transmitted symbols ui-k at the
incorrect correlator output, and the expression (5.4) is similar where
to the expression (2.16). The conditional mean and variance 2 Afinc AZC OfC
at the incorrect correlator are then given by, respectively, u = ,b2= c=- , (5.13)
- 4 c +o?,inc ’
o 2 c +& UfinC
E(ni I# j ( i , 1-m)t 8 f ( i , l-m), Him) and QM(*)is the Marcum-Q function.
L The probability of error, as given in expression (5.12),
C.uk(ri,i-m)exp[ - j # f ( i , 1-m)I ;noise jammer represents the conditional bit error rate at srate r, defined by
the last L + 1 transmitted kequencies, conditioned on the jam-
= w p j ( i ,1-m) exp ULef(i.1-m) -#f(i, 1-m)11 (5.5) mer P ~ 8, (i,m)
S and 8, (i. 1-m) -
The average bit error rate, conditioned on the jammer set
l +k l uk(ri,1-m) exp [-j#,(i, ;multi-tone jammer of phases {8k}, k = 1,2, ..., N, is found by the average

and
0
2
(ni I#f(i,l-m)* 8f(i,l-m)9 Him) where N,(T)/N, is the relative frequency of occurrence of
state I‘‘in one pseudo-random generator period. In the partic-
No/E + P f ( i , l-m)Pj/yBE ;noise jammer
={NO/. ;multi-tone jammer. (5.6)
ular case of equiprobable frequencies and Q >NL+l, this
term can be replaced by l/Ns.
The envelopes of the observables are the random variables xc
and xhc having Rice-Nakagami probability density functions VI. SIMULATION
conditioned on 8, (i,m), and having the form
The simulation purposes are a) to prove the ability of the
L(x)= (x/o:>exp [-(A: +x2)/2a~lZ~(Axx/a~), (5.7) proposed state equalizer to improve the performance of a C-
where l o ( -is
) the modified Bessel function of the first kind FHSS system; b) to conlirm the theoretical performance
and zero-th order. The parameters A:, o:, under the correct obtained in the previous sections; c) to allow the performance
hypothesis, are comparison between C-FHSSand NC-FHSSsystems operat-
ing under the same channel impairments.
The bit duration T, as well as the bit timing, are assumed
to be perfectly known. These assumptions should not affect

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on November 07,2023 at 17:42:18 UTC from IEEE Xplore. Restrictions apply.
1332 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL 42, NO 21314, FEBRUARYIMARCHIAPRIL 1994

the performance of the implementation shown in Fig. 3 of -10, -6, -3 and 0 dB (larger values of JSR are not considered
Section IV if the bit synchronizer equivalent noise bandwidth because forward error correction is not used), and for E/No
is narrow enough. Consequently, the bit synchronizer and in the range 0 to 12 dB. As indicated above, an approximate
both the frequency and the carrier synthesizers shown in Fig. evaluation of the equalized receiver performance is computed
3 are not implemented in the simulation program. Also, the by using the relationship
channel impulse response is kept constant during the simula-
tion. The estimates of the channel intersymbol interference BER [SNR(I-)] =1
2
erfc m. (6.2)
coefficients {Z,(r)] and the state RF phases {$o(r)} are
acquired and tracked during the simulation, and the decision As will be seen, this approximation proves to be an accurate
device output is compared to the transmitted bit in order to predictor for the equalized C-FHSS system, at least in the
monitor the bit error rate. above range of JSR and E/No. The accuracy of expression
(6.2), implying that the output samples of the LMS algorithm,
in the presence of AWGN and a random binary sequence
A. Simulation Parameters input, have approximately Gaussian statistics, is in agreement
with the conclusion of [151.
The following system was simulate&
The equalized and nonequalized receiver performances
a) The pseudo-random generator f ( i ) is implemented by
are computed for two cases:
using a maximal-length linear feedback shift register of
a) partial band jamming with y = 0.5, slots #1, #2, #3 and #4
length 5, with feedback located at taps 5 and 3. The
jammed. Figures 4-7 correspond to noise jamming and Figs.
sequence period is Q = 31 and the number of frequency slots
8-11 correspond to multitonejamming.
isN=8.
b) partial band noise jamming averaged over all combinations
b) The channel memory is L = 2, hence the number of states
of slots as shown in Figs. 12 and 13. It is seen that the worst
N, = 31.
case y is about 0.5, and performance is approximately
c) The normalized center frequency is arbitrarily chosen to be
independent of y for y > 0.5.
fcT=855. The normalized lower frequency edge is
The computation results are plotted and compared with the
foT = M = 850.
simulation results. An interesting result is observed in the
d) There are NL = 16 complex coefficients, defining the vec-
case of noise jamming for the particular case of y = 0.5 and
tor representation of the channel impulse response
slolts #1, #2, #3 and #4 jammed (seeFig. 7, JSR = 0 dB). At
c = ( a 0 9 .**, a15), arbitrafily chosen as
low signal-to-noiseratios, the equalized channel performance
a0 = (0, 1); = (-.82, 0.); a3 = (.34-.769);
is .worse than the ideal (no multipath) channel performance;
a5= -.72,0.); all=(Os, -.382); als = (O., -.269). The at high
other coefficients are (O., 0.). - signal-to-noise
- ratios, the equalized channel performs
e) The multi-tone jammer set of phases is arbitrarily chosen loo
as follows:
el = .296; e2= 2.233; e3= 1.015; e4=4.341; es = 5.247;
66 = 5.614; 67 = 1.451; 6'8 = 0.701.
f ) The step-size parameter isp = 0.1.
g) The reference estimated phase of each state is obtained by
averaging over SO transmitted symbols.
h) 6200 symbols are used in the mining period (200 symbols
per state),
Although the channel impulse response is kept constant
during the simulation, a rough estimate of the allowable fad-
ing rate for this system can be easily obtained using expres-
sion (3.3) and the simulation results. The average number of
iterations, p ( T ) , needed for the learning process, is about 80.
\
Since the sequence perid equals the number of states, each
state occurs once every sequence period and, therefore, the
revisiting time T ( r ) is constant and equals the sequence
period (31T) for any state. Therefore, IO-^ 1
min ( z d / T ) > 80.31= 2480.
The value of p(r),used in expression (6.1), is somewhat
(6.1) No Multipath BPSK
Predicted Equalized Channel
Simulated Equalized Channel
\
conservative, since the learning process duration is measured 1 : I SimulatedNonequalized Channel
for an initial zero-valued tap-weight vector. A practical sys-
tem would tolerate a faster fading rate.
B. Simulation results The simulation results are plotted for Signal to Noise Ratio (dB)
four different levels of jammer-to-signal power ratio (JSR), Fig. 4. Bit Error Rate; Noise Jammer to Signal Ratio = -10 dB

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on November 07,2023 at 17:42:18 UTC from IEEE Xplore. Restrictions apply.
 LUPU AND MILSTEIN: FWSS SYSTEM WITH MULTIPATH CHANNEL EQUALIZATION

I I
IO'
o
l "/-

I0-
\

10-3 \

m
\
05
W w
m m
IO-^
\ Io-; \\A\ \

IO-^
\ q\
- \
No Mdtipath BPSK
---____ Predicted Equalized Channel No Multipath BPSK
Simulated Equalized Channel -______ Predicted Equalized Channel \
h Simulated Equalized Channel \
Simulated Nonequalized Channel A '\

0 2 4 6 8 1 0 1
IO-^ I I I I
I
I 2 4 6 8 IO 12
Signal t o Noise Ratio (dBJ Signal t o Noise Ratio (dB)

Fig. 5. Bit Error Rate; Noise Jammer to Signal Ratio = -6 dB Fig. 7. Bit Error Rate; Noise Jammer to Signal Ratio = 0 dB

I oo I I 8 I I
IO0 I I I I I

IO-'

to-' IO+

IO'?
CK

\
cc
w
m
lo-* w
m

I 0-3
IO-^ \
-______
A
No Multipath BPSK
Predicted Equalized Channel
Simulated Equalized Channel
\ \
.!'
I o-6 _--____
A
No Multipath BPSK
Predicted Equalized Channel
Simulated Equalized Channel \
I I I Simulated Nonequalized Channel ! ! SimulatedNonequalized Channel \
I 0-4 I I I I 9 to-' 8 I I I I

2 4 6 8 10 12 2 4 6 8 10 2
Signal t o Noise Ratio (dB) Signal t o Noise Ratio (dB)

Fig. 6. Bit Error Rate; Noise Jammer to Signal Ratio = -3 dB Fig. 8. Bit Error Rate; Multi-Tone Jammer to Signal Ratio = -10 dB

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on November 07,2023 at 17:42:18 UTC from IEEE Xplore. Restrictions apply.
1334 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL 42, NO 21314, FEBRUARYhlARCHIAPRIL 1994

rT
W
m

I \
Io5 1 IO-^ 1 - __ NoMultipathBPSK
\ ‘A,
\
2,

Predicted Equalized Channel

A Simulated Equalized Channel Simulated Equalized Channel
! I Simulated Nonequalized Channel Simulated Nonequalized Channel

io4
0 2 4 6 8 IO 12
Signal to Noise R a t i o ( d B )

Fig. 9. Bit Error Rate; Multi-Tone Jammer to Signal Ratio = -6 dB Fig. 11. Bit Error Rate; Multi-Tone Jammer to Signal Ratio = 0 dB

100

I Q 0 l

c
[L
I35
W -m
Lrl

10-2
m (3

\ 0
\

\
I

\
\
\
IG-5 -
------_
NoMultipathBPSK
Predicted Equalized Channel
\
A Simulated Equalized Channel

I d6
j

I
Simulated Nonequalized Channel
I I I I
tl L
6
+
X
q a m m a - ,25
aamma- ,375
aammo - .5
I
I
“2.0 i.0 d.0 $.O 8.0 li.0 tJ.0 I
SIGNAL TO NOISE RATIO (dB)

Fig. 10. Bit Error Rate; Multi-Tone Jammer to Signal Ratio = -3 dB Fig. 12. Bit Error Rate; Equalized Channel, Noise Jammer Predicted
Performance, JSR = OdB

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on November 07,2023 at 17:42:18 UTC from IEEE Xplore. Restrictions apply.
LUPU AND MILSTEIN: FWSS SYSTEM WITH MULTIPATH CHANNEL EQUALIZATION
1335

I I I
C I 1 I I 1 -
I I I
I
I I
I I I I
I
-2

a
-E

T-=--
-9

-12

- I5
0 No Jammer
0 JSR =-lo dB
-18 JSR=d dB
JSR=-3 dB
JSR=O
-21
5 IO 15 20 25
SIGNALTO NOISE RATIO (dB1 Signal to Noise Ratio ( d B )

Fig. 13. Bit Error Rate; Equalized Channel, Noise Jammer, Predicted Fig. 15. Bit Error Rate; Equalized Channel, Noise Jammer Predicted
Performance,JSR = OdB Performance
I I I I I I I I
better than the ideal channel. At low signal-to-noise ratios,
the noise is the dominant factor in the SNRo(r)expression
(4.4); the bit error rate is dominated by the destructive states
(IZ,(I') c 1I), which happen to originate from the non-
jammed slots #$, #6, #7, and #8. In this case, the bit error
rate of the equalized channel is higher than that of the ideal
channel. At very high signal-to-noise ratios, primarily the
jammer noise affects the performance; therefore, the channel
having higher average SNR at the jammed slots performs
better. The states originating from the jammed slots in this
particular example are, on the average, constructive, and
No Jammer therefore the equalized channel performs better than the
JSR =-lo dB ideal channel. The same tendency can be seen in Fig. 6; this
JSR=-6 dB
JSFb-3 dB
behavior is also clearly seen from the comparison of Fig. 14
lSR=O dB and Fig. 15. Figure 14 depicts the computed performance of
a C-FHSSchannel without multipath and using BPSK modu-
lation, according to the theoretical expression in (6.3) below.
Figure 15 presents the predicted performance of the equal-
ized multipath channel, with the same noise jamming,
according to (6.2).
0. . . ...o ...... 1
B E R =~ ~(1 - ~ y)~evc
~ 6E)
2
0
0 1
IO
1 I

20
I

Signal t o Noise Ratio (dB)

I

30
I I
40
I

!
+-yegc[d-]
21 -+- (6.3)
SNR JSR
Fig. 14. Bit h a Rate; No Multipath, Noise Jammer Predicted
Performance The performance of the NC-FHSSsystem is also com-

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on November 07,2023 at 17:42:18 UTC from IEEE Xplore. Restrictions apply.
 1336 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 42, NO 2/3/4, FEBRUARYiMARCHiAPRIL 1994

I 0' I I I I I

puted and simulated, for the same specific noise jammer and
for the same parameters; the results are presented in Fig. 16
and Fig. 17. The performance of the NC-FHSS system in the
presence of a noise jammer of JSR = - 10 dl3 (y=O.5,
jammed slots #1, #2, #3, #4) and multipatb is substantially
worse than the performance of an equalized C-FHSS system
in the presence of the same multipath and a similar noise
jammer having a higher power level (JSR = 0 B),as shown
in Fig. 16. Figure 17 shows good agreement between the
predicted performance, developed in Section V, and the
simulation results, for the NC-FHSS system operating in the E
m
10-
above scenario.
Finally, the performance of the equalized C-FHSS system,
as well as the performance of the NC-FHSS system, in the
presence of the worst-case average jammer (y=O.5), are
given in Fig. 18. Recall that in an ISI-freeAWGN channel, a
significant performance improvement could be obtained at
Predicted NC-FHSS system. J S R S dB
low S N R values by using antipodal signaling with coherent ---- -- Predicted NC-FHSS system, JSR=-3 dE
IO0 1 I I I
I I I Sb~ulatedNC-FHSS system, JSR= 0 dB
A Simulated NC-FHSSsystem, JSR=-3 dB

io-'
2 4 6 8 IO
Signal to Noise Ratio (dB)

Fig. 17. Bit Earor Rate; NC-FHSS System, Noise Jammer, Predicted
and Simulated Performance
LT
w
m
I
I I
-
Pr Et dic t e d €qua Iized C F HSS
0 Simulated E ualized C-FHSS
* Predicted Ne-FHSS System

-- Equalim3C-FHSS system, JSRSO dB

NC-FHSS system, JSR=-IO dB
. .. ...
\
NC-FHSS system, JSR=6 dB
_ _ . _ . _ NC-FHSS
"
--- system, JSR=-3 dB
NC-FHSS cystem,J S R d dB
\ c
II:
IO-~J I I I I I w
m
0 2 4 6 8 IO I u
W
0
J
Signal to Noise Ratio f d B )
Fig. 16. Bit Error Rate; NC-FHSS System, Noise Jammer, Predicted
Performance

detection instead of a NC-FSK scheme. However, in the case

of heavily IS1 contaminated channels, this improvement no
longer exists. As comparatively seen in Figs. 4, 5, 6, 7 (for
the nonequalized CH-FHSS system) and Fig. 16 (for a NC-
FHSS system), the BER is dominated by the IS1 effect and
shows low sensitivity to both the signaling scheme and the
JSR (in the -10 to 0 dB range). The results presented in Figs.
16 and 18 show that the significant performance improve-
ment achieved by the coherent system over the noncoherent Fig. 18. Bit &or Rate; Noise Jammer,
JSR = 0 dB, y = 0.5
system is primarily due to the channel equalization, rather

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on November 07,2023 at 17:42:18 UTC from IEEE Xplore. Restrictions apply.
LUPU AND MILSTEIN: FWSS SYSTEM WITH MULTIPATH CHANNEL EQUALIZATION 1337

than the coherent reception (the C-FHSS system performs Jmin

better than the NC-FHSS system with a 10 dB advantage in lim E(eief)= (A.4)
i+ l-OSp(L+l)'
JSR).
The minimum mean-squared error of the optimum Wiener
solution, Jmin,is obtained from [12]:
VII. CONCLUSIONS
J ~ =nvur (Ti') - vur (xo, (A.5)
It was demonstrated that a narrow-band method of equali- where xi' is the output of the channel model filter (refer to
zation is feasible for C-FHSS systems operating in a mul- Fig. 2 and replace the channel estimator by the optimum
tipath environment leading to severe intersymbol interference Wiener solution). Expression (AS) is true under the approxi-
effects. The proposed method of equalization uses a decision mation ii= u j , which holds for low probabilities of error.
feedback equalizer performing the LMS algorithm. While The optimum Wiener solution to the channel model is
the number of channel states grows exponentially with the exactly the transversal filter having the tap-weights (zk(r)}
channel memory, for practical systems with large pseudo- and, therefore, we have
random sequence period compared to the number of available L L
frequency slots, the hardware complexity, as well as the com-
putational effort, is not affected, except for the amount of
var (xi') = E [ z I:zk(r)ui-kui-m]
k=Om=O
memory required to store the state equalizer coefficients. - L
The required memory does grow exponentiallywith the chan- - IE [Ezk(r)ui-k I 'e (A.6)
k=O
nel memory.
A fairly accurate method for predicting the system perfor- using the properties of ui, namely E(&&) =,6 and
mance is developed, yielding good prediction over a E(ui) = 0,we can write
significant range of interest (-10s JSA I O dB and L
0 I E/No I 12 dB). vur (xi') = z 1zkm I'.
k=O
(A.7)
The performance of the equalized C-FHSS system is
significantly superior to the NC-FHSS system with the same In order to evaluate vur(ri'), we use expressions (2.15)-
occupied bandwidth, as shown in the comparative graphs. (2.18).
The equalization is the dominant factor affecting the substan- - L
tial performance improvement, and is realizable, provided E(ri')= E [zzk(r)ui-k] +E(ni) = E(ni)
k=O
(A.8)
that the hopping rate is much faster than the channel fading
rate, so that t4e adaptation process is possible. and
The FH code acquisition performance, as well as the code
tracking performance, are beyond the scope of this paper.
However, it is expected that the equalization scheme pro-
posed in this paper will improve the performance of a
receiver code tracking loop which uses decision-directed
measurements. We obtain the value of the minimum mean-squared error,
Jhn, by substituting expressions (A.7) and (A.9) into (AS):
-
APPENDIX J~~ = Iq n i ) I = vur (ni). (AJO)
Finally, we substitute expression (A.lO) into (A.4) and obtain
The excess mean-squared error, Jm(i), at iteration i, is the
difference between the mean-squared error produced by the the mean-squared error
adaptive algorithm and the minimum mean-squared error
value, Jfin, obtained by the optimum Wiener solution. Using
[12], we have
REFERENCES

M. K. Simon and A. Polydoros, "Coherent Detection of

Frequency-Hopped Quadrature Modulation in the Presence of
Jamming. Part I QPSK and QASK Modulations," IEEE
T r m . Commun., vol. COM-29, pp. 1644-1660, Nov. 1981.
i-1
G. Cherubini and L. B. Milstein, "Performance Analysis of
where Ai are the eigenvalues of the correlation matrix lR . Both Hybrid and Frequency Hopped Phase-Coherent Spread
Since, in our case, A = 1 for i = 1,2, ...,L + 1, we have Spectrum, Part I and II." IEEE Truns. Commun., vol. COM-
37, pp. 600-622, June 1989.
(A.3) G . Cherubini and L. B. Milstein, "Analysis of the Tracking
Loop of Phase-Coherent Frequency Hopped Spread Spechum
Systems," IEEE Mikom Proceedings, Oct. 1986, 31.3.1-
We substitute expression (A.3) into (A.1) and obtain 31.3.5.

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on November 07,2023 at 17:42:18 UTC from IEEE Xplore. Restrictions apply.
 Chun-Meng Su and L. B. Milstein, "Analysis of a Coherent Theory, and an Associate Technical Editor for the IEEE Communi-
Frequency Hopped Spread Spectrum Receiver in the Pres- catwm Magazine, and is currently a Senior Editor for the IEEE
ence of Jamming," vol. COM-38, pp. 715-726, May 1990. J o u d on Selected Areas in Communications. He was the Vice
Chun-Meng Su and L. B. Milstein, "Analysis of a Coherent President for Technical Affairs in 1990 and 1991 of the IEEE Com-
Frequency Hopped Spread Spectrum System," IEEE Milcom munications Society, and is currently a member of the Board of
Proceedings, Oct. 1988, 34.3.1-34.3.5. Governors of both the IEEE Communications Society and the IEEE
IMormarion Theory Society. He is also a member of Eta Kappa Nu
Chun-Meng Su and L. B. Milstein, "Comparison of Joint and Tau Beta Pi.
PhaseiTiming Tracking Loops for a Coherent Frequency
Hopped Spread Spectrum Receiver," IEEE Globecom
Proceedings, 1988, pp. 17.3.1-17.3.5.
Chun-Meng Su and L. B. Milstein, "Channel Equalization
and Performance Analysis of a Coherent Frequency Hopped
Spread Spectrum System," IEEE Journal on Selected Areas
in Commun., vol. SAC-7, pp. 548-560, May 1989.
H. Urkowitz, Signal Theory and Random Processes, i.lrtech
House, Inc., 1983.
A. P. Clark, Equalizers f o r Digital Modem. London,
Prentice-Hall, 1986.
P. Monsen, "Fading Channel Communications," IEEE Com-
mun Magazine, pp. 16-26, Jan 1980.
S . E. Benedetto, E. Biglieri and V. Castellani, Digital
Transmission Theory, New Jersey, Prentice Hall, 1987.
S. Haykin, Adaptive Filter Theory, New Jersey, Prentice-
Hall, 1986.
9. G. Proakis, Digital Commmications, New Yo&,
McGraw-Hill Book Co., 1983.
M. Schwartz, W. R. Bennett and S. Stein, Communication
System and Techniques, New York, McGraw-Hill Book Co.,
1966.
R. A. Iltis and L. B. Milstein, "An Approximate Statistical
Analysis of the Widrow LMS Algorithm'with Application to
Narrow-Band Interference Rejection, IEEE Tram. Commun.,
vol. COM-33, no. 2, pp. 121-130, Feb.'1985.

Valentin Lupu was bom in Bucharest, Romania in November 1947,

and emigrated to Israel in 1966. He received the B.Sc. (cum laude)
in electrical engineering from the Technion-Israel Institute of Tech-
nology, Haifa, Israel and the MSc. in electrical engineering from
San Diego State University, California in 1970 and 1989, respec-
tively.
From 1970 to 1975 he served on active duty with the Israeli Air
Force as a Communications Officer. Since 1975, he has been with
RAFAEL (the Israel Armament Development Authority), where he
is currently an R&D communications engineer.

Laurence B. Milstein (S'66-M'68-SM'77-F'85) received the

B.E.E. degree from the City College of New York, New York, NY,
in 1964, and the M.S. and Ph.D. degrees in electrical engineering
from the Polytechnic Institute of Brooklyn, Brooklyn, NY, in 1966
and 1968, respectively.
From 1968 to 1974 he was employed by the Space and Com-
munications Group of Hughes Aircraft Company, and from 1974 to
1976 he was a member of the Department of Electrical and Systems
Engineering, Rensselaer Polytechnic Institute, Troy, NY. Since
1976 he has been with the Department of Electrical and Computer
Engineering, University of Califomia at San Diego, La Jolla, CA,
where he is a Professor and former Department Chairman, working
in the area of digital communication theory with special emphasis
on spread-spectrum communication systems. He has also been a
consultant to both government and industry in the areas oE radar and
communications.
Dr.Milstein was an Associate Editor for Communication Theory
for the IEEE Transactions on Communications, an Associate Editor
for Book Reviews for the IEEE Transactions on Infomation

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on November 07,2023 at 17:42:18 UTC from IEEE Xplore. Restrictions apply.