lEEE TRANSACTIONS ON COMMUNICATIONS, V O L . COM-20. NO.

6 , DECEMBER 1 9 7 2 1063

Considerations of Nonlinear Effects in

Phase-Modulation Systems
JHONG S. LEE AND DONALD G. WOODRING

Abstract-Aclass of “optimum” phase modulation (PM) system would represent a phase modulation (PM) system that has re-
treated in the past is one wherein a sinusoidal carrier is phase modulated
ceived a good deal of attention in the past (see [ 1 1 -[3], for
linear/.v by a Gaussian (baseband) random process and demodulation is
example). For convenience, we will call the system without
performed by a phase-locked loop (PLL). This paper is concerned with
the smooth limiter a linear phase modulation (LPM) system,
PM systems in which the phase modulation is allowed to be nonlinear.
In practice, most phase modulators are restricted to finite and the one with the smooth limiter a nonlinear phase modu-
ranges of
linear operations. The performance (SNR) of the PLL demodulator is lation (NPM) system.
degradedwhen the linearrange is exceededbyanamplitude of the
The phase modulators (or in general, angle modulators) are
modulating waveform. The occurrence of performance degradation is
a random event whose frequency depends
essentially voltage- or current-controlled oscillators. They are,
on the linear range and the
statistics of the waveform amplitude. as such, subjected tosaturationphenomena when driven
By interpreting the nonlinearphasemodulationasasaturation beyond the linear range .of operations. The smooth limiter in
phenomenon, we determine in this paper the performances of PLL de- the NPM systemshown in Fig. 1 is intended to account for
modulator as a function of varying degree of saturation (or limiting)
saturations occurring in the PM systems. In any practical PM
.level for several cases of baseband spectrums. We model the “satura-
tion” by an error function limiter. Our results are general, and it
system, in which the modulating waveform has an unbounded
is
amplitude range, such as a Gaussian process,’a saturation is
shown that the previous results of linear modulations are special cases
of OUT results. We then move on to assume that the baseband wave- certain to occur attimes during the operation.
form is a composite signal of multichannel frequency division multi- For an LPM system, the form of the optimum demodulator
plexed@DM) signals.As a result, we compute the lower bounds of
with an appropriate delay (under “linear assumption”) is, in
signal-to-crosstalk power ratio for several cases of baseband spectrums.
fact, the PLLfollowed by a linear filterf ’ ( t ) as shown in Fig. 1,
whose detailed structures depend on N o , A’ and the spectrum
of the message process m(t) [3]. For an NPM system, how-
I. INTRODUCTION ever, it is seldom possible to predict the degree of phase dis-

T HE COMMUNICATION system of interest in this paper tortion, ahd hence, the spectrum of the effective modulating
is shown in Fig. 1 . The waveform to be transmitted, m(t), process x ( t ) cannot be predicted. Thus, it is difficult t o specify
is a sample functionfrom a stationary zero-mean Gaussian the structure of the optimum demodulator foran NPM system.
random process. It is passed through a “smooth limiter” with The mainobjectiveof this paper is to determine the per-
ahardnessparameter a, whosetransfer function,denoted formance degradation of theoptimumdemodulator of an
symbolically L [rn(t);a], is an error function togive an output LF” system whenthe -message waveform is subjected to a
~ ( t )E L [m(t);a]. The function x(t),which is a nonlinearly distortion. We thus consider the effects of a nonlinear phase
transformed version of the “message” waveform m(t), then distortion on the performance of the demodulator that is de-
modulates a sinusoidal carrier of frequency w o and power A’ signed to be optimum for an LPM system. We shall investigate
to form the transmittedsignal this problem for a few important cases of different baseband
spectrumsforbothextremes ofzeroandinfinite delay de-
modulators. We will show graphically the performance (SNR)
of the NPM system :above the threshold” as well as “at the
threshold” for different degrees of phase distortions.
The channel corrupts the transmitted signal with an additive A certain consideration is also given in this paper under the
white Gaussian noise power of spectrum (two-sided) assumption that the modulatingwaveform m(t) representsa
~ N o / 2 W/Hz. Thedemodulator is aphase-locked loop (PLL) frequency division multiplexed (FDM) signal: the passage of
followed by a linear filter and the demodulator output is an FDM signal throughanamplitude limitingnonlinear device
estimate %(t - 6) of the message waveform m(t) with a delay such as the smooth limiterproducescrosstalk in each chan-
of 6 seconds. If the smooth limiter in Fig. 1 is eliminated or nel’s outputduetointermodulationsamong signals in all
replacedbyalinear gain device, the resultingconfiguration channels. We compute and presentgraphically in this paper
the signal power-to-crosstalkpower ratio in the center chan-
Paper approved by the Communication Systems Discipline Committee nel (the “worst” channel) as a function of the degree of phase
of the IEEE CommunicationsSocietyforpublicationwithoutoral
presentation: Manuscript received April 24, 1972.
J. S. Lee is with the Department of Electrical Engineering, Catholic ‘The Gaussian assumption for the message process is well justified in
University of America, Washington, D.C. 20017. realitywhen it representssuitablymultiplexedmultichannel signals.
D. G. Woodring is with the Maritime Administration, Office of Re- The assumption is valid to the extent that the central limit theorem can
search and Development, Washington, D.C. 20235. be invoked. For further discussion, see Section IV.

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on March 22,2023 at 03:47:15 UTC from IEEE Xplore. Restrictions apply.
 1064 IEEE TRANSACTIONS ON COMMUNICATIONS, DECEMBER ‘1972

AMODEL OF A
NONLINEARPHASE
MODULATOR
n(t)
1 I

I
f i ~ s i n [ w ~ t + x ~ t ~ ]

MODULATOR

Fig. 1. PM system under consideration.

distortion (limiting level) for various baseband spectrums such where

as the Markov, rectangular, and other spectrums. A
q= a/u,. (7)
Limiter Model Equation ( 6 ) is plotted in Fig. 2 for several values of 4 . The
The transfercharacteristicof the smooth limiterin Fig. 1 parameter a,which appeared in (2) for the first time, is called
is represented by an error function of the form a “limiter rms value” since it is identified with the variance
(or rms) of the normal distribution [ 6 ] . It is a parameter that
x ( t ) L [rn(t);a] A a
erf [rn(t)/flcu] , (2) determines the“hardness”of the smooth limiter, as can
noted from Fig. 2. Thus, the parameter q as defined by (7)
be
where
may be termed a normalized limiter hardness parameter.
Q normalization parameter for zero-memory lossless lim- It is easy to show that (6) becomes
iter (to be determinedbelow);
4 ) message waveform;
Q limiter “hardness” parameter;
and
It is thus seen that the smooth limiter defined by (2) or (6)
is capable of representing. not only a harder limiter (q = 0),
(2/fi) exp ( - u 2 (3)
)du. but a linear amplifier of unity gain (q + m) as well. Therefore,
the analysis and the results of the NPM system given in this
paper is applicable t o all cases of interests.
The saturation’effect of an amplifier is often modeled by an
error function limiter since it offers not only mathematically
tractable results, but also represents the behaviors of practical 11. ANALYSIS
amplifiers rather closely [ 4 ] , [ 5.] The receiver (ordemodulator) is aconventional PLL fol-
It can be shown [ 6 ] that the autocorrelation function of the lowed by a linear filter characterized by the impulse response
limiter output x ( t ) for the Gaussian input m(t) is given by of f’(t). The filter f ’ ( t ) is includedin the receiver to allow
delay of 6. seconds betweentheinputandtheoutput.The
R x ( 7 )= E [ x ( t ) x (+
t T ) ] = (2177) Q arc sin [R,(T)/(u; + a ’ ) ] ,
loop filter f ( t ) is of necessity zero delay [ 3 ] , and hence the
(4) case of zero delay demodulator does not require the filterf‘(t).
It is a characteristic of the PLL that when the error between
where theactualmodulating signal x ( t ) and theestimated signal
E = expectation operator k(t) is small, the equivalent form of the loop can be repre-
sented by a linear equivalent circuit as shown in Fig. 3, where
R,(7) = E { mm( t()t + T)} ( 4 4 n’(t) is an effective white Gaussian noise ofspectral density
N 0 / 2 A 2 W/Hz, and H o ( j o ) represents the closed loop transfer
u& A R , ( 0 ) = E { m Z ( t ) } . (4b) functionofthe PLL. Since the .difference %(t)- x ( t ) is a
We assume that the limiter is lossless, namely, that R,(O) = random variable whose range is unbounded,the “small dif-
R,(O), thereby obtaining ference,” and hence linear operation, is to be defined within
the meaning of the constraint given by
Q = u; ((2177) arc sin [u&/(u; + a 2 ) ]} - ’ . (5) 2 2
Ox& ocr, (9)
By putting ( 5 ) into ( 2 ) we obtainthe normalizedlimiter
where
transfer function
IJ& i! E { [ x ( t6- ) - k(t- 6 ) ] 2 } (10)

and u& is the critical value of the variance that is acceptable

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on March 22,2023 at 03:47:15 UTC from IEEE Xplore. Restrictions apply.
 LEE AND WOODRING: NONLINEAR EFFECTS IN I" SYSTEMS 1065

and
15

I.o

where S,(w) denotes the rational spectrum of the message

05 (baseband) process m ( t ) and [ 3 represents the upperhalf-
+

t plane poles and zeros of therationalfunctionwithinthe

b'
l o bracket.
-x

Mean Square Error

-0.5
From (12), the meansquare error can be expressed as
u i =R,(O)- 2E[m(t- 6 ) & ( t - 6 ) ] + E { [ h ( t - 6 ) ] ' ) . (16)
For stationary process m(t), the expectations in (16) are in-
- -2.0
-1.5 l .-1.5 o i
-1.0 z -0.5? !0 f Y
0.5 J 1.0 1.5 20
dependent of delay symbol 6 andthus (16)may be ex-
pressed as
m ( t )/urn-+

Fig. 2. Normalized transfer function of the smooth limiter. o2 =R,(O) - 2 E [ m ( t ) h ( t ) ]+ E { [ & ( t ) ] * ) . (17)
The demodulator output&(t)can be written as

$(t) =ll h(u) [x(t - u) + n'(t - u ) ] d u , (18)

where h(t) is the impulse response of a physically realizable

Fig. 3. Linearized equivalent demodulator and output filter for delayed filter H( j w ) defined by (1 1). -Our immediate objectives are to
estimate.
compute the termsin (17). To this end, we have from (18)
for the validity of the linear model of Fig. 3. When (9) is
%t)l
satisfied with equality, the system is said to be operating at
"threshold" [ 13,[3] . Beyondthis value o r , the periodic
nonlinearity of theloop becomes .a significant detrimental
factor in the system performance [ 3 ] .
= E [ m ( t ) J - l h ( u ) { L [m(t - u ) ;a ] t n'(t - u ) ) du
1 . (19)

Thisexpressioncanbeevaluatedina rather straightforward

Transfer Functionof the Demodulator manner. Interchanging the order of integration and expecta-
The two filters in Fig. 3 can be combined to form an equiv- tion operation, we have
alent filter
H ( j o ) =Hdjw)F W ) , (1 1) E [m(t)&(t)]= 1 -m
m

h(u) E { m ( t )L [m(t- u);a ] } du. (20)

which must be optimized so as to minimize

Note that the term E[m(t) n'(t - u ) ] in(19)becomeszero
u z = E { [ m ( t - 6 ) - & ( t - &)I2}. (12) since the signal and noise are assumed to be independent. The
expectation within the integrandin(20) is evaluated inAp-
This is the conventional minimum variance filtering problem. pendix I, and the result is given by
It is to be noted, however, that the input to the demodulator
is x ( t ) , which is anonlinear function of the actual message E { m ( t )L [m(t- u);a1 = K(a)Rrn(u), ( 2 1)
input m(t). Thus, the system we consider is optimumonly
where R,(u) is defined by (4a), and
for the case when
X ( t ) = m(t). (1 3) [W/arc sin W ] *I2 (22)

Equation (13) is true when4 .+ m, which is the case of LPM.

w 4 1/(1 t .
"la:)
= 1/(1 t q2). (23)
Forthe case of Gaussian signal and Gaussian noise, the Substituting (2 1) into(20), we get for the second term of (17)
optimum filtertransfer function is given by the well-known
solutions to the Wiener-Hopf equation. For rational spectra,
and for the important case of zero delay (6 = 0) and infinite
delay (6 + m), the solutions are simply given by [3, p. 1431
2E[m(t)h(t)= 2K(a)
I: h(u)R,(u) du.

For the third term in (17), a direct substitution of (18) is not

(24)

(N0/2A2)'/2 as rewarding as might be expected. It turns out that the fre-

H(jw) = 1 - (14)
[Sm(w)+ N o / 2 A 2 ] +' = quency domain approach is simpler in this case. Recognizing

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on March 22,2023 at 03:47:15 UTC from IEEE Xplore. Restrictions apply.
 1066 IEEE TRANSACTIONS O N COMMUNICATIONS, DECEMBER 1 9 1 2

the third term of (17) as the power(meansquare value) of where hk z k / ( 1 + q 2 ) k . Theconstraint term given by ( 9 )
&(t),we have can also be expressed in this form. Proceeding in the same
0 manner as above, ( 9 ) becomes (Appendix 11):
E { [&(t)l2> = J s, ( W ) d o / 2 n , (25) m

where S ~ ( Wdenotes
) the
-m

spectral density of &(t). Since we

U& - (2/?T)Q 5 J
k
hk
-m
R k ( u ) [h”) (U) - 2h(U)] du

are always restricted to a& < uzr (linear assumption), we can

odd

make use of
well-known
a result of
linear
system
a theory: t ( N 0 / 2 A 2 )h ? ) (0) < u$. (33)
+ S,,(W)l I H ( W ) I2 >
= [SX(W) (26) In theory (32) and (33) cannow be used toobtain amaxi-
where s , , ( ~=)N o / 2 A 2 and s , ( ~ is) the spectrum of x ( t ) . mum allowable value on the mean-square error given in (32).
Taking the inverse transform and employing the multiplication- From a practical .standpoint, however, it is far simpler to retain
convolution property of the Fourier transform, we obtain (33) and use those values of u2 in (32) such that (33) is
/.m. .
satisfied at the same time.
R;,(T) = J h(’) (u) [RX(7- u ) + ( N 0 / 2 A 2 )6 ( -~ u ) ] du,
-m
111. THE SYSTEM PERFORMANCE
We measure the performance of thesystemby the output
(27) signal-to-noise ratio (SNR),defined by
where
h(’)(u) 6 IH(jw)12 exp ( j w u )dw/2n. 1 (28)

Using (27) we have for the third termin (17) In view of (32), it is seen that the SNR depends on 4 , u&, the
signal spectrum,andthe carrier-to-noisepower ratio in the
channel.
i_
E { [h(t)l} = ~ ~ (= 0 )d 2 ) (u) ~ , ( u )du A commonlyencountered family of spectraldensities is
given by the set of rational functions whose denominator are
+ ( ~ , / 2 ~ h2( 2) ) (0). (29) Butterworth polynomials [3]. We will consider such spec-
trums in our study, given by
Now substituting ( 4 ) for R,(u) in (29),we have

Expanding the arc sine in the integrand of (30) in a Taylor S,(W; k -+ m) corresponds to arectangular spectrum.The
series, we have optimum transfer functionsforthe equivalentlinearsystem
of Fig. 3 for zero delay and infinite delay have been given by
(14) and (1 S), respectively. For the assumed baseband spec-
trum of the form given by (35), the optimum transfer func-
odd tions and the corresponding impulse responses are summarized
in Table I for a few cases of interest. We point out that for
+ ( N 0 / 2 A 2 )h(?)( 0 ) , (31) the spectra given in Table I the integrals appearing in (32) and
where’ (33) are determined in closed forms. The resultingsumma-
tions in (32) and (33) were further simplified where possible
Zk [(k - 2) ! !/k(k- 1 ) ! ! ] . andwere used in the evaluation of SNR for various values
By substituting (24) and (31) into (17) we obtain arefined of q , u h , and A 2 / N o B .
form of the mean square error ’ The calculation of the SNR in (34) behooves simultaneous
computations of (32) and (33). This is to ensure thatthe
constraintequation in (33) is always satisfied.Thecarrier-
to-noisepower ratio A Z / N o B that is obtainedfrom simul-
taneous solutions of (32) and (33) with equality, is defined as
“threshold.” The threshold depends, among others, upon the
choice of the value given to u$ in (33). As will be noted
shortly, in some cases there is no solution for the threshold.
The constraint equation (33) “defines,” for a given value of
’Notation:k!!=k(k-2)(k-4)...,withO!!=(-l)!!=l . u : ~ ,alinear operationofthePLL system. Thenormally used

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on March 22,2023 at 03:47:15 UTC from IEEE Xplore. Restrictions apply.
NONLINEAR
ECTS
WOODRING:
LEE AND IN PM SYSTEMS 1067

TABLE I

for k = I
2u& (2nE)
S,(w; 1) = Rm(r)= uh exp [-(2nB) 1711
( 2 7 4 2 + w2 ’
for Zero Delay

for Infinite Delay

(2nB)Z (p2 - 1)
H( j w ) =
.~p2 + wz ’
(2nB)Z
.

fork=2

for Zero Delay

h(t) = (2nB) - + a exp [-(2nBp) It I/<]

for k = m

sin (2nBr)
R,(r) = uh ~

Io; IWI > 2nB, (2nBr)

values of ufr (themaximum allowable mean-square phase One may wish to consider a value of u$ different from 0.25
errors for “linear”assumptions)arerestricted tothe range or 0.5 in (33). Insuch case, newsimultaneoussolutions of
of 0.25 < u$ < 0.5 [ 2 ] , [3]. Inourstudy we considered (32) and (33) with anew value of uZr will produce anew
two cases of ofr= 0.25 and u$ = 0.5. “dotted” line in each figure. Once a threshold curve (dotted
In Figs. 4-6, we have plotted SNR for a Markov baseband line) is drawn for a given choice of uZy value, the curves that
spectrum for u$ = 1, 4 , and 16. For each value of u h , both are below and on the left-hand side of the dotted line are to
zero-delay and infinite-delay cases are considered. The SNR’s be discarded from further consideration since the actual per-
are plotted as a function of A 2 / N o B with 4 as parameter. formances (SNR’s) belowthresholdare much worse (lower)
Explicitin each figure are theperformance differences for than indicated in each figure. For a further observation, con-
different choices of the value given to u$ in (33). The thresh- sider Fig. 4(a). The baseband spectrum is assumed to be
olds for two different values of uZr = 0.25 and uZr = 0.5 are Markov with mean-square value of uf = 1.0. For a linear op-
computed and plotted as dotted lines in each figure. Thus, the erationofthe PLL system with zero delay, carrier-to-noise
SNR’s computed for values of A 2 / N o B that satisfy the con- power ( A 2 / N o B ) of 20 dB (threshold) or more is required.
straint equation in (33) are the performances “above thresh- when q = 0.25.3 In Fig. 7(a)-(c) we have plotted SNR’s for
olds,” and these are the curves on the right-hand side of the arectangular baseband spectrum for several power levels. In
dotted line for each u& considered.Note in Fig. 5(a), for these figures we have shown only the infinite delay demodula-
example, that the curve for q = 0 (hard limiter case) is always tor performances. Throughout the figures, the curve identified
below thresholds. This means that in this case the mean-square with 4 -+ 00 is worthy of note. This is the case of the ideal
phase error cannot be less than 0.5 (let alone 0.25) and hence linear phase modulator (LPM) in which the modulated carrier
a linear operation of the PLL cannot be assumed if the modu-
latoracts as ahardlimiter (4 = 0). However, in the same 3The parameter 4 is actuallydetermined(measured) as theresponse
figure we see that for > 0.25, thresholds are well defined for characteristic of the modulator device as a function of the input signal
level. Seethe paragraph following (37) in [4] for a possiblemethod
both cases of u,$. of measuring the value q of a given device.

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on March 22,2023 at 03:47:15 UTC from IEEE Xplore. Restrictions apply.
1068 IEEE TRANSACTIONS ON COMMUNICATIONS, DECEMBER 1 9 7 2

I"

f 10 IO

I.o I.o
0 5 35 IO 3 0 12
55 20 0 5 IO 1535 20
30 25
A2/N0B ( d B ) AZ/NoE (dB)

Fig. 4. (a) SNR (not in decibels) of NPM system demodulator versus carrier-to-noise ratio for different limiting levels ( k = 1,
uf = 1.0, 6 = 0). (b) SNR (not in decibels) of NPM system demodulator versus carrier-to-noise ratio for different limiting
levels ( k = 1, uf = 1.0, s = -).

a
IO (b) 5 lo

I.o IO I I I I I I
0 5 IO. 125
5 20 30 35 0 5 IO 15 20
30 25 35
A2/NoB (dB) A2/N0E (dB)

Fig. 5. (a) SNR (not in decibels) of NPM system demodulator versus carrier-to-noise ratio for different limiting levels ( k = 1,
uf = 4.0, 6 = 0). (b) SNR (not in decibel's) of NPM system demodulator versus carrier-to-noise ratio for different limiting
levels (k = 1, u& = 4.0, 6 = -).

IO2

a
z IO IO
ln

I.o I.o I I I I I I
0 5 IO 15 20 25 30 35 0 5 35 IO30 125
5 20
A2/NoB

Fig. 6 . (a) SNR (not in decibels) of NPM system demodulator versus carrier-to-noise ratio for different limiting levels (k = 1,
u& = 16.0,s = 0). (b) SNR (not in decibels) of NPM system dem.odulator versus carrier-to-noise ratio for different limiting
levels ( k = 1, u& = 16.0, 6 = -).

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on March 22,2023 at 03:47:15 UTC from IEEE Xplore. Restrictions apply.
LEE AND WOODRING:NONLINEAREFFECTS IN PM SYSTEMS 1069

’ 0 3 3

10 I I I I I I
0 5 IO 15 20 25
35 30 0 5 35 1030 125
5 20
A2/N& (dB) A‘/NoB(dB)

(a) (b)

I.o
0
I
5
I
IO
I
15
I
20
I
25
I
30
J
35
A‘/N,B (dB)

(C)
Fig. 7. (a) SNR (not in decibels) of NPM system demodulator versus carrier-to-noise ratio for different limiting levels (k = -,
oh = 1.0, 6 = -). (b) SNR (not in decibels) of NPM system demodulator versus carrier-to-noise ratio for different limiting
levels (k = -, u h = 16.0, 6 = =). (c) SNR (not in decibels) of NPM system demodulator versus carrier-to-noise ratio for dif-
ferent limiting levels ( k = -, u h = 16.0,6 = -).

phase is always assumed to be linearly related to modulating for the case of u%: = uzr. The curves shown are forthe
process m(t) at all times, however large its amplitude may be. Markov spectrum, infinitedelay demodulator,fordifferent
Thus, the curve with q + m in each figure may be viewed as values of q as indicated. We have shown, in all of the figures
theupperboundonthe SNR. Theoperation of the system throughout, Viterbi’s [3] results for the performance of the
“at threshold” is illustrated in Fig. 8. This is a plot of SNR LPM systems. Our results for 0% = 0.25 and q + coincided

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on March 22,2023 at 03:47:15 UTC from IEEE Xplore. Restrictions apply.
 1070 IEEE TRANSACTIONS O N COMMUNICATIONS. DECEMBER 1 9 1 2

to3, talk is concerned, the demodulator-demultiplexor complex is

equivalent to a linear processor in which no intermodulation
is generated.
Cahn [7] has considered the problem of computing SCR
that is minimum among all channels when the composite FDM
signals of “Gaussian spectrum shape” is passed through an
ideal amplitude clipper. By assuming that the Gaussian spec-
trum is symmetric- about the center channel, Cahn has com-
puted minimum SCR by considering the center channel where
crosstalk is maximum. Lee [4] considered a similar problem
Lz
z
(0
except that the amplitude-limiting device was a smooth limiter.
Both Lee [4] andCahn [7] assumed the shape of the spec-
trumto be a Gaussian form andpresented SCR ingraphic
form as afunction of a parameter that is related tothe
10
limiting level.
Our purpose in this section is to show SCR curves for sev-
eral different shapes of basebandspectra. More specifically,
we will consider thespectrum cases showninTable 1 and
compare them in the same graph with SCR corresponding to
the Gaussian spectrumthat Lee [4]has considered. Lee’s
I.o general result on SCR is directly applicable in our study.’ The
5 IO 15 20 25 30 35 40
A2/NoB(dB) formula given by (34) in [4] is the SCR at the center channel.
Fig. 8. SNR (not in decibels) at threshold defined by o;,. = 0.25 and We may express it, using our notation, as
0.50 for different limiting levels (k = 1 , 6 = 0 ) . m

with Viterbi’s results. For high A 2/N0B, the SNRgiven.by

[
”0
Pm (7)d7
SCR = (3 7)
(34) approaches a limiting value for all q < 03. One can easily
show that
,=1

[+]
where
2u
[(2v 1) ! !I2
-
as A2/NoB+ w ,
x,= P v ! ( v + I)!
1V. CROSSTALKIN FDM-NPM SYSTEM
In the previous sections the waveform m(t) (see Fig. 1) was and p m ( 7 ) = R,,(7)/Rm(0) = Rm(7)/uL is the normalized
meantonly to be a sample functionfrom stationary
a autocorrelation function of the composite signal process nz(t).
Gaussian random process. In this section we will assume that We have computed (37) for cases of the input spectra (hence
it represents a sum of multichannel FDM signals. The correlation functions) compiled in Table I, and the results are
Gaussian assumption on the process m(t) can still be tenable plotted in Fig. 9, along with SCR for Gaussian spectrum shape
under thisassumption if the number of channels is so large considered earlier by Lee [4]. It is interesting to compare the
that the central limit theorem can be invoked. We have now different values of SCR corresponding to hard limiter (q = 0)
an FDM-NPM system. The passage of FDM signals through under various input power spectrum assumptions as indicated.
an amplitude limiting device, such as smooth limiter, produces Recently Bond and Meyer [8] considered theproblem of
crosstalkineachchannel duetointermodulationsamong computingthe signal-to-crosstalkpower ratio (S/IM in their
signals in all channels. notation) at the output of a bandpass hard limiter under the
In an FDM-NPM system an additional problem of great in- assumption that the input process consist‘s of N 2 3 constant
terest is to determine the signal power-to-crosstalkpower ratio amplitude sine waves, and hence, the spectrum is of rectangu-
(SCR) in each channel since it is a measure of signal degrada- lar shape. Although they [8] have employed a different “sta-
tion. Let us restrict our consideration to the linear model for tistical approach,” a comparison of their result with a value
the demodulator ‘(Fig. 3) as before, and .assume further that shown in our Fig. 9 is interesting. From Table IV in [8], we
thebandpass filtering atdemultiplexor4 does notintroduce read S/IM = 8.28 dB forN = 15. In ourterminology,this
additional crosstalk. We can thencompute SCR fromthe corresponds to SCR = 8.28 dB for q = 0. From our Fig. 9,
quantities at the output of the limiter since, insofar as cross- we read SCR = 8.10 dB at q = 0 for rectangular spectrum case.
One of the implications of this rather close agreement is that
41n our discussion it suffices to assume that the “demultiplexor” is
aset of parallelbank of filtersfollowing the demodulatorshown in SThe discussion given in this section is heavily based on the papersby
Fig. 1 . Cahn [ 7 ] and Lee [4].

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on March 22,2023 at 03:47:15 UTC from IEEE Xplore. Restrictions apply.
 LEE AND WOODRING: NONLINEAR EFFECTS I N PM SYSTEMS 1071

/ APPENDIXI
GAUSSIANSPECTRUM
(LEE 141)
FUNCTION
CROSS-CORRELATION O F INPUTA N D
OUTPUTOF SMOOTH LIMITER
We derive here (21) given i n thetext. Let the cross-
correlation function between the input m ( t ) and the output
x ( t ) of the limiter of Fig. 2 be defined as
k =I(MARKOV SPECTRU
R,,(u) = E { m ( t )x ( t - u ) ) . (39)
From (2) this can be written in terms of the input m ( t )alone
R m x ( u ) = E { m ( t ) Lu[ )m; a( t]-) . (40)
For convenience, let m l represent the value of m ( t ) at time t ,
and m 2 the value at time t - u. Thus,
R m x ( u ) = E { m l L[m2;al). (41)
l 2 . 3 FGAUSSIAN SPECTRUM (LEE C41)
k=2
IO
93
Using the joint probability density function p ( m l , m 2 ) of m l
AI and m 2 , (41) may be expressed as
I I I I I I I
0 1 2 3 4
p = a/u,
5 6 7

Fig. 9. Lower boundson SCR versus limiting levels inmultichannel

FDM-NPM systems for differentbaseband spectra.
Rm,(U)=J
-m
m

1,m

m l L [ m z ; a l p ( m l , m 2 ) d m l d m 2 . (42)

Since m ( t ) is a Gaussian process, the density function p(m 1 ,

m 2 ) is given by
the sum of 15 constant-amplitude sine waves can be reasonably
approximated as a Gaussian process.

V. SUMMARY
Thispaper has treated two problems of practical interests (43)
with regard to a phase modulation system in which the modu- where
lating waveform is drawn from a stationary Gaussian random A
Pm = P r n ( U ) = E { m l m z ) / Q &
=Rm(u)/oh. (44)
process. The optimum demodulator has been known to be a
phase-locked loop (PLL) for a linear phase modulation and for Substituting (43) into (42)and rearranging terms, we have
an additive Gaussian noise channel.
Our first concern was the investigation of the performance
of this very optimum demodulator when the phase modula-
tion became nonlinear due to a limited linear dynamic range
of themodulatorfor high level input. The nonlinear phase
modulation was viewed asasysteminwhich the effective
modulating waveform is a smoothly limited version of the
original “message” waveform. The smooth limiter was repre-
sented by an error function that contained a parameter whose
variation could represent from a linear device to a hard limiter.
We presented the performance (SNR) of the demodulator as The second integral in (45) is recognized to be the expression
a function of channel carrier to, noise power ratio ( A 2 / f V ~ B ) for the mean of Gaussian random variable m l , which is seen
for different degrees of nonlinear phase modulation for several to be just prn(u)m 2 . Thus, (45) reduces to
types of basebandspectrums. We considered both zero and
Rrnx(u)=Prn(u)K~(a), (46)
infinitedelay demodulation cases as well. Theperformance
“at threshold” for a Markov baseband spectrum was also pre- where
sented under two different threshold definitions. By regarding
themodulating waveform as the composite signal of multi-
channel FDM signals, we computed lower bounds of signal
power to crosstalkpower ratio (SCR) for several baseband
To evaluate K l ( a ) , we substitute into (47) the limiter output
spectrums. function L [ m 2; a ] ,defined in (2), expressed in the following
equivalent form:
ACKNOWLEDGMENT
The authors wish to thank the anonymous reviewers of this
paper for several helpful suggestions.

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on March 22,2023 at 03:47:15 UTC from IEEE Xplore. Restrictions apply.
 1072 IEEE TRANSACTIONS ONCOMMUNICATIONS,DECEMBER 1972

where second term in (57) may be expressed as

I,
m

E { x ( t ) A(?)} = E { x ( t ) h ( u ) [ x ( t - u ) + n’(t - u ) ] du 1

The substitution then results in

h ( u ) E { x ( t ) n’(t - u,} du. (59)

+ L

The last inte,gralin (59) is zero since the signal and the zero
Now let the last integral in ( 5 1 ) be viewed as follows: mean noise n ( t - u ) are independent. Thus, (59) becomes

exp ( - v 2 ) dv = cdp{$[ ;.’m2

e x p ( - v 2 ) dv
I +R ECx(t)A(t>} = 1, m

h ( u ) R x ( u ) @, (60)

=sin2 exp(-P2m~)d/3tR1, (52) where R,(u) is the autocorrelation function of the limiter out-
put, which is given by (4). Upon its substitution into (60), we
have
where R is a constant of integration, which is zero (note that
for nz2 = 0, (52) becomes 0 = 0 + R and hence R = 0).
Substituting ( 5 2 ) ,with R 1 = 0 into (5 l), we have

* [R,(u)/(u$ + a 2 ) ]d u . ( 6 1 )
Expanding arc sine in a Taylor series, ( 6 1 ) becomes

which is a standard integral,yielding

where hk is defined in (32). Substituting (31), (58), and (62)

into (57), we obtain
Substituting (49) and (50) into (54), we obtain

r 1 k
odd
J-m

k J-m
odd
Finally, if we put (55) into (46), we get
+ ( N 0 / 2 A 2 )h ( 2 )(0). (63)
Rmx(u)= R m ( u ) K(a), (56)
where R m ( u ) = u& p i (G) and Combining the second and third term, we have the expression
given in (3 3).
~ ( a= K
) (a)/& = [ W/arc sin W I‘I2
REFERENCES
as defined by( 2 2 ) and (23).
[ l ] A. J . Viterbi and C.R. Cahn, “Optimum coherent phase and fre-
quencydemodulation of a class of modulatingspectra,” IEEE
APPENDIXI1 Trans. Space Electron. Telem., vol. SET-10, pp. 95-102,Sept.
1964.
DERIVATIONOF THE CONSTRAINTEQUATION (33) [ 2 ] C. J . Boardrnan and H . Van Trees, “Optimum angle modulation,”
IEEE Trans. Commun. Technol., vol. COM-13, pp. 452-469, Dec.
Noting that for stationary process the expectations are in- 1965.
-.--.
dependent of delay 6, ( 1 0 ) becomes, 131 A. J . Viterbi, Principles of Coherent Communications. New York:
McGraw-Hill, 1966, ch. 6.
u;h=Rx(0)- 2 E { x ( t ) A@)} + E { [ & ( t ) 1 2 } , (57) 141 J. S. Lee, “Signal-to-crosstalk powerratioinsmoothlylimited
multichannel FDM signals,” IEEE Trans. Commun. Technol., vol.
where COM-16, pp. 63-67, Feb. 1968.
[5 1 R. F. Pawula,“Theeffects of quadratic AM-PM conversion in
RJO) = R m ( 0 )= :
0 (58) frequency-division multiplexed
multiple-access
communications
satellite systems,” IEEE Trans. Commun. Technol., vol. COM-19,
and the last termin (57) given by (31). Using (18),the pp. 345-349, June 1971.

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on March 22,2023 at 03:47:15 UTC from IEEE Xplore. Restrictions apply.
 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. COM-20, NO 6 , DECEMBER 1972 1073

161 R . F. Baum, “Thecorrelation function of smoothly limited where he workedondesign of radio-frequencyinterferencemeasure-

Gaussiannoise,” IRE Trans. Inform. Theory, vol. IT-3, pp. 193- mentsystems.From1965to1968hewasanAssistantProfessor at
197, Sept. 1957. George Washington University, teaching statistical communication and
[ 7 ] C. R.Cahn, “Crosstalk duetofinite limiting
of
frequency- coding theories. He was a consultant to Radiation Systems, Inc., from
multiplexed signals,”&-oc. IRE, vol. 48, pp. 53-59, Jan. 1960. 1966 to 1968. From 1968 to 1969 he wasemployedbytheFederal
[8] F. E.BondandH;,F.Meyer, “lntermodulationeffects in limiter
amplifier repeaters, IEEE TrQnS.Commun. Technol., vol. COM-18, Systems Division of the IBM Corporation, where he worked on satellite
pp. 127-135,Apr. 1970. communication systems. Since 1969 he has been an Associate Professor
of Elcctrical Engineering at the Catholic University of America, Wash-
ington,D.C. Heis aConsultanttothe Naval Research Laboratory,
Washington, D.C.
Dr. Lee is a member of the American Mathematical Society and the
American Association for the Advancement of Science.

Jhong S. Lee(”61)wasborn in Koreaon

December 20, 1935. He receivedtheB.S.E.E. Donald G . Woodring was born in Hazleton, Pa., on June 20, 1947. He
degree
from
the
University
Oklahoma,
of receivedthe B.S. degree in electricalengineeringfromPennsJ-lvania
Norman, in 1959, and the M.S.E. and D.Sc. de- State University, State College,in 1969 and the M.E.E. degreefrom
grees in electrical engineering from the George theCatholicUniversityofAmerica,Washington, D.C., in1971. He
WashingtonUniversity,Washington, D.C., in heldanNSFTraineeshipduringtheperiod 1969to1972. He is
1961 and 1967, respectively. currentlydoingworktowardthePh.D.degreeincommunications
He joinedWeinschelEngineering in 1959. theory.
From 1961 to 1963 he was employed by FXR, Since 1972 he has been employed by the Maritime Administration,
Inc.,whereheworked on radartransmitters Office of Research andDevelopment,Washington,D.C.,wherehe is
and millimeter-wave’ferrite component develop- involved
with the Maritime Satellite Communication/Navigation
ment. He joined F‘requencyEngineeringLaboratories,Inc.,in 1964, Program.

A Theoretical Analysis of Telephone Dial

Pulse Signaling
STEFAN J. MICHALAK, WILLIAM J. OLDER, AND GODFREY W. MUEHLE

Abstract-This
theoretical
paper
presents
a analysis ofdistortion
the I. INTRODUCTION
of the dial pulse waveform and its effects on the response characteristics
of the pulse-receiving circuit inthe telephoneoffice.
T o deriveexpressionsforthewaveformfordialspeeds
20 pulses per second (pps) Laplace transformation
W
of 10 and
HEN a subscriber dials a number, dial pulses are sent
from the subscribersset throughthe transmission line
and matrix algebra -tothetelephoneoffice. This is achieved by interruptinga
areapplied to the system, represented by two complementary linear direct current path by dial contact D,see Fig. 1.I

lumped-constanttime-invariantnetworksresponsibleforthegeneration Thisgenerated current waveform is distorted,duetothe

of the leading and trailing edgesof a dialpulse. number of extensions on the transmission line, the line param-
Based on the known waveform and the operational requirements of
the pulse-receivingcircuit,a function is derivedshowing thepercent eters? the subscribers set, and the pulse-receiving circuit
break value of a pulse train as it arrives at the central office, i d the parameters.
dependence of the percent break on the number of extension (branch) This distortion can cause severe malfunction of the pulse-
telephones connected and the subscribers’ loop parameters. Its inverse, receiving circuit.
Shown in graphical form, Presents the upper limits for the number of The problem was observed many years ago, but recent system
extensions to be associated with a transmission line of given electrical
parameters. The conditions under which abrupt changes of the percent developments have made its solution moreurgent.
break
value and quasi-pulses occurare analyzed. Empirical approaches have been made, but there has been a
considerable lack of theoretical study. This paper is intended
to fill this gap.

Communication
Paper
approved
theby Theory
Committee of the 11. DLALFUNCTION
IEEE CommunicationsSocietyforpublicationwithoutoralpresenta- Consider the network shown in Fig. It consists of,? tele-
tion. Manuscript received September 16, 1971.
The authors are with Bell-Northern Research, Bramalea, Ont., Canada. phones, atransmission line, a mercury type relay L , and a pulse-

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on March 22,2023 at 03:47:15 UTC from IEEE Xplore. Restrictions apply.