C h a p t e r N n e

Modulation Techniques

While it is conceptually possible to transmit baseband signals directly, it is usually much

more effective to transmit data by modulating a higher frequency carrier wave. This allows
precise control of the radiated frequency spectrum, more efficient use of the allocated Rf
bandwidth, and flexibility in accommodating djfferent baseband signal formats. In general, a
sinusoidal carrier wave may be modulated by varying any of its three degrees of freedom: run-
plitude, frequency. or phase. Most modern wireless systems use digitaJ modulation, where the
modulation variables change in discrete steps, in contrast to older methods of analog modulation
where earlier amplitude, frequency, or phase may vary continuously. 1n contrast to analog mod-
ulation. digital modulation makes more efficient use of the radio spectrum. and usually requires
less prime power. In addition, digital modulation performs better over a fading communications
channel, and is more compatible with the use of error-correcting codes. We will begin with
a short review of analog modulation, then focus most of our attenlion on digital modulation
techniques. Our objective is to study circuits for modulation and detection, the effect of noise
on the detection process, and to derive expressions for the bit error rates of digital modulation
methods in the presence of noise. These results will be useful for our study of receiving system
perfonnance in Chapter I 0.

9.1 ANALOG MODULATION

We begin with a review of basic analog modulation techniques. including single-
sideband and double-sideband amplitude modulation. and frequency rnodulalion. We will
describe circuits for modulation and demodulation, and derive expressions for the signal-
to-noise ratio of these methods. Although some wireless systems employ other analog
modulation methods, such as angle (phase) or pulse position modulation, we will not cover
such topics here because of space limitations and the much greater importance of dlgi-
tal modulation methods. The subject of analog modulalion, however, remains useful a~
288
 9.1 Analog Modulation 289

RF
Ba,ebanu
data in

Modulotor Upconverlcr RF ~,age

(a)

RF IF

RF ,tage Down convener Dcmo<lula1or

(b)

FI GURE 9.1 Illustrating lhe function of modulation and demodulation in trnnsmitter and receiver
--------- systems: (a) modulator. (b) t.lcmodulaior.

an introduction to the analysis of modulation methods, and as a comparison with digital

modulation methods. References LI H 3] can be consulted for further detail on these topics.
Receiver circuits were introduced in Chapter I, and are discussed in more detail in
Chapter 10. but it is useful here to review the basic heterodyne transmitter and receiver
in order to place our discussion of modulation in its larger context. Figure 9.1 shows
block diagrams for lypical transmitter and receiver systems. The fundamental purpose or
the transmiller is 10 modulate an RF carrier signal with the baseband data thlll is to be
trarn,mitted. while the receiver must recover the baseband data from the received modulated
carrier wave. In both cases we see that the modu lator/demodulator provides the interface
between the baseband information and the IF signals. Because the IF signal in a receiver
contains noise from the antenna and receiver circuitry. as well as the desired signal. the
characteristics of the demodu lator play a critical role in the overall performance of the
wireless system.
In each case considered next we assume a bandlimitcd modulating waveform 111(1).
This baseband signal may represent voice. music, computer da1a. or other information
10 be transmilled over the wireless channel, but for simplicity we assume 111(1) to be a
=
single sinusoid of the form 111(1) co~ 21rf111 1. The maximum frequency of the modulating
waveform is /;1,1. so that O :'.S J,11 :'.S f,11-

Single-Sideband Modulation
Figure 9.2 !.hows a block diagram for SSB modulation and demodulation. This diagram
only shows the relevant IF and baseband stage~. and omit~ the RF and other components
that .u·e required in a practical transmitrer and receiver. The l ocal oscillator of the SSB
modulator is modulated by the bandlimited baseband signal 111(1). forming a DSB signal
at the output of the mixer. The upper sideband is selected from the mixer output with a
bandpass filler. although the lower sideband could be selected. if preferred. To avoid the
requirement for the sharp-cutoff sideband filter. the single-sideband mixer discul>::.ed in
Section 7.4 could insteud be u~ed to generate either a USB or LSB signal directly. In either
290 Chapter 9: Modulation Techniques

SSB modula1or Channel & receiver SSB demodulator

______
F_lG_U_R_E_9_._2 Block diagram of a SSB modulator and demodulator.

case the spectrum occupied by the SSB signal extends from .fu: to .fir:+ /;11 , where JM is
the maximum frequency of m(t).
Gaussian white noise. wilh a two-sided power spectral density n 0 /2, is added to lhe IF
signal to model noise contributed by the transmission channel. as we.II as noise generated
by the input stages of the receiver. The IF bandpass filter has a passband extending from
+
/ rr to /JF IM, which passes the received SSB signal as welJ as noise power within this
band. The demodulator mixes the input SSB ff signal with a local oscil lator of frequency
/rr- Note that the demodulator LO is idenlical in frequency and phase with the modulator
LO; this is called a synchronous. or coherent, demodulator. The difference frequency from
the mixer is low-pass fillered to recover the baseband signal, m.(t).
The SSB input voltage lo the demodulator can therefore be expressed as

U;(f) = A cos(WJF + W111)l + n(I), (9.1)

where Wtr = 2rrf1r, w111 = 2rcJ;,,, and 11(1) is gaussian white noise that has been passed
through the IF bandpass filter. Since 11(1) is limited to a narrow frequency band near wrF,
it can be expressed using the narrowband representation of noise dfacussed in Section 3.3.
Thus we let n(t) = x(t) cos wrFt - y(t) sin wrr:t, where x(t) and y(r) arc the bandlimited
quadratme components of the noise, and rewrite (9.1) as

V; (t) = A COS(W rF + w,,, )I + x(t) cos WrFI - y(t) sin WtFf' (9.2)

For a local oscillator signal cos 2n.fTFt, the output voltage of the mix.e r is proportion;il
to

v1(t) = v;(t)cosw[l,t
A A l I
= 2 cosw,,,t + 2 cos(2wrr + w 111 )t +
2 x(t)(J + cos 2cvwt) -
2y(t) sin 2w1r-t
(9.3)
Since our fu1al result will involve the ratio of powers, we have ignored the mixer
conversion constant. After low-pass filrering the final output voltage from the demodulator is
A I
U0 (!) =
2 COSW111 / + x(t). (9.4)
2
which contains the original modulating waveform and a bandpass noise signal. Note that,
because it is orthogonal to the synchronous LO, the quadrature component of the noise
voltage does not appear in the output of (9.4).
We now find d1e signal and noise powers in order to evaluate the output signal-to-noise
ratio. The average input signal power of the SSB signal of (9. l) is found by time-averaging:
A2
S; =2· (9.5)
 9.1 Analog Modulation 291

Similarly, the average power of the output signal in (9.4) is

s,, = ~( { y = ~2 = d (9.6)

The power of the narrowband input noise of (9.1) and (9.2) is given by

N, = E(n 2 (t)I = E{.r2(t)} = £{/(t)}. (9.7)

according to (3.47 ). Then the noise power of the output voltage of (9.4) is

2
N11 = E { [~x{t)r } = ~£{.r (t)} = :; . (9.8)

So the output SNR is, from (9.6) and (9.8).

So S; 4 S;
(9.9)
N0 = 4 N1 = N1'
This result shows that the SSB demodulator does not degrade the input SNR. The frequency
translation pe,formcd by the SSB demodulator affects the input signal and noise equally,
leaving their ratio unchanged. Ofcourse. practical mixers and filters have losses and generate
noise. and this will lead Lo a degradation of the output SNR. But the process of SSB
demodulation itself is inherently lossless in tcm1S of the SNR.
While the above mathematical derivation provides the desired result, we may get a
bener understanding of the effect of mixing and filtering on lhe noise power by considering
the power spectral densi ty from a graphical point of' view. We assume the power spectral
density of the input white noise before the I F filter is 110/2. As shown in Figure 9.3a,

_.....__.__n-

- /IF
I 111,
f,11
i~n
-+----------'--'---+

(a)
.rw I
•

fu, + f.,1
.r

"o

□ □
8

I - ZfiF
cfi 2fiF I •f
- 2/,~ f.,1 2Jjp +!11
(bl

' l

Fl CURE 9.3 Power density spectrums of noise in a SSB demodulator. (a) Power spectral density
--------- or the noise at the demodulator input (b) Power ~pcctral density of the noise at the
output or the mixer. (c) Power spe.:tral density or the noise after low-pass filtering.
292 Chapter 9: Modulation Techniques

the lF bandpass filler passes noise power only over its passband. from /iF to /JF + .fA,.
Including the negative part of the frequency spectrum thus gives the input noise power to
the demodulator as

(9.10)

Mix..ing the bandpass noise with the LO has the effect of shifting the noise spectrum up and
down by /1F, as shown in Figure 9.3b. As derived in Chapter 3, mixing noise has the effect
of reducing its power by haJf. Thus, from (3.42). the power spectral density of the noise
after the mixer must be 11 0 /8, so that the total noise power after the mixer is

N, = 4 .fM ('~o) = no{M. (9.11)

Finally, as shown in Figure 9.3c. the low-pass baseband filter selects only the noise spectrum
from -.f.H to fM• so the output noise power is

. (110) nofM N;
No =2)M S
= - - =
4 4. (9. 12)

The Last step follows from (9. 10), and is in agreement with our earlier result of (9.8). Then
we can rewrite the output SNR of (9.9) in terms of the power spectral density of the input
white noise:

(9.13)

Double-Sideband Suppressed-Carrier Modulation

Next we consider a double-sideband amplitude modulation system, with the block dia-
gram shown in Figure 9.4. The mixer of the modulator forms the product of the modulating
signal and the LO. producing a DSB waveform; unlike the SSB case, both sidebands are
retained. ff the mixer were ideal, the ca1Tier frequency (fir) would not be present in Lhe
output, so U1is modulation is referred to as double-sideband suppressed carrier (DSB-SC).
As we have seen in Chapter 7, realistic mixers produce virtually all combinations of the
input frequencies. so fi lte1ing must be used to actually achieve a true DSB-SC wavefonn.
The spectrum of tbe DSB-SC signal extends from .fir - h1 to .fir+ ./M, where .fM is
again the maximum frequency of the moduJating input. This is twice the bandwidth of the
corresponding SSB waveform.
As in the previous case, we assume Lhat gaussian white noise is added lo the IF signal
before demodulation. This noise is again bandlimited by Lhe IF bandpass filler, which now
has a bandwidth of '2f,11. centered at f ir , to pass both sidebands of the DSB signal. The

m (t )

-cp IF
fi Iler ,.,._, Ji
Baseband
filter
v,,(1)
SQ, N0

11 012
rr
(AWGNJ

DSB modulator Channel and receiver DSB demodu lator

_ __ _ _F_I_G_U_J._{E_9._4 Block diagram of a DSB-SC modulator and tlemodulator.

 9.1 Analog Modulation 293

input voltage 10 the demodulator can then be expressed as

A A
u,(r) = .J2 COl>(wrF - w,,,)r + ../2 cos(wlF + w,,,)r + 11(r). (9.14)

We have normalized the input voltage so that the total DSB-SC input power is idemical
to lhe input power of the SSB case (corresponding lo equal transmitter powers):

(9.15)

Again using the narrowband representation for n(t), we can rewrite (9.14) as
A A
v,(r) = ,r.:;cos(w1p-w,,,)r + r,,cos(w,r+w,,,)r +x(t)cosw,Fr - v(r)sinwwr.
v2 v2 •
(9.16)
For an LO of 2,r/1Fr the output of the mixer is

11,(t) = u;(r )coswrrr

A A 1 1
= r,, co~ w,,, t + r,, cos(2w 1r - w 111 )r + -x(r )( 1 + cos 2w1Ft) - - y(f) sin 2w1rr
v2 v2 2 2·
(9. I 7)

Aft.er the low-pass baseband filter, the final output is

A I
llo(r) = M COSWml + - x(t). (9.18)
v2 2
Again, the quadrature component of the noise drops out because it is onhogonal to lhe
synchronous LO. The output signal power i s

(9.19)

Observe thnt we now have S0 = S, /2, whereas in the SSB ca ewe had S11 = S; /4. This
is because the two sidebands of the DSB signal add in phase after mixing with the LO,
doubling the receive voltage, and increasing the output power by a factor of four. However,
since each sidchand of the DSB waveform has half the power of the SSB waveform, the
overall increase in power is two. relative to the SSB case.
The noise power of the output voltage of (9.18) is
2
N,, = E { [ 1x(t)] } = N; . (9.20)
2 4
where N, = £(11 2(1)} = £{x 2(r)} = E(y2 (r)} as before. Then the output SNR is
Su S; 4 ? S;
(9.21)
N0 = 2 N, = - N, •
This result suggests that the DSB-SC demodulator improves the input signal-to-noise ratio
by a factor of two. This improvement is due to the effective doubling of signal power noted
above. I Iowever, a more realistic result is found by expressing the input noise power in
terms of the power spectral density of the white noi~e at lhe input lo the IF stage. Since the
IF bandwidth is 2JM, we have

N, = 4/M ( no) = 211oh,1. (9.22)

2
294 Chapter 9: Modulation Techniques

where the factor of four accounts for positive and negative frequency ponfons of the filter
response. Then the output SNR of (9.21) is rewritten as

So S;
(9.23)
No = 110.fM'
which i.s seen to be identical to the result for the SSB case given in (9. 13). This implies
that the coherent SSB and DSB-SC demodulators have the same SNR performance, when
expressed in terms of a uni fom1 input white noise level; this is a more meaningful comparison
lhan that of (9.21 ).

EXAMPLE 9.1 SNR FOR A DSB-SC DEMODULATOR

Der.ive the output SNR for a DSB-SC demodulator with an arbiu·ary modLtlating
signal, m(r ), where the input signal and noise is represented by

v;(t) = m(t)coswrrt + n (t ).

Consider m(f) to be a random process with £{111(/)) = 0 and E{m 2(t)} = m 2/2.

Solution
Assuming an LO vol tage of cos WLFf , the output of the mixer is

v,(t) = V1(r) COSW1pl

= 111(1) cos2 WIFI + x(I) cos 2 WLFI - y(I) cos u) Ipt sin WIFf
I l 1
= m(I)( I + cos 2<uLFI) + x(t)( I + cos 2w11-t) - y(1) sin 2w 1pt
2 2 2
After low-pass filtering the output voltage is

The average power of the input signal is found by time-averaging the expected
value of the square of the input signal voltage:

The output signal power is

2
So = E { [ -Lm(f)] } = -m2 = -.
S;
2 • 8 2

The noise power of the demodulator output is

Then the output signal-to-noise ratio is

S0 _ S; 4 _ S;
2
N0 - 2 N; - N;'
which is the same result as was obtained for a sinusoidal modulating sigmtl. 0
 9.1 Analog Modulation 295

Double-Sideband Large-Carrier Modulation

If the douhlc-sidcbnnd signal of the previous case is transmiued without suppression of
the carrier wave. it is referred to as double-sideband large-carrier (DSB-LC) modulmion.
This is an advantage in that the carrier signal. even if much lower in amplitude than I.he
sidebands, can be used as a reference signal to phase-lock the local oscillator to synchroniza-
tion with the incoming signal. The block diagram of the DSB-LC modulator/demodulator
is essencially the same as the DSB-SC ~ystern shown in Figure 9.4. but now I.he input signal
is expressed as

u,(t) = A[I +m(t))cosw 1pt +n(t)

111A
= A cosw,Ft +
2 lcos(wIF -w,,,)t + cos(w,r +w,,,)tl + 11(1)
(9.24)

where the modulating signal i~ m(t) =

m cos w 111 1. The amplitude, m, of the modulating
signal relative to I.he carrier is called the 111od11/atio11 index. The signal power associated
with this input voltage is

(9.25)

where the first term is the carrier power and the second is the power in the sidebands. The
carrier power thus increases the total input power, but does not directly contribute to the
output power after demodulation since it does not contain any modulation information.
Again we use I.he narrowband representation of noise to write 11(t) 1 (t) cos w 11, 1 - =
y(t) sin <•>1Ft. Mixing the input vollagc of (9.24) with the LO and low-pass filtering gives
the demodulator output voltage ns

v0 (t) = 2111A cosw,,,c + 1I x(t). (9.26)

Thus the output signal power is

= ') . (9.27)
2(2 + 111- )

where we used (9.25) to represent S0 i n terms of S,. Because the IF bandwidth for the DSB-
LC case is the same as that for the DSB case. the input and output noise powers are the
same. Usi11g (9.20) and (9.22) with (9.27) gives the output SNR of the DSB-LC modulator
as

So S, 2m2 111
2
S,
-=----.,= ,. (9.28)
No Ni 2 + 111· 11oh1(2 + 1w)

Note that if 111 » I, so that the carrier power is negligible relative to the sideband power,
(9.28) reduces Lo (9.23). W hen the carrier power is not negligible. however, the output SNR
is less than that for either SSB or DSB-SC. For example. when 111 = I (a modulation index
of 100%), we have
>
111·

2+m2 =3.
which implies a reduction in SNR of 4.8 dB.
296 Chapter 9: Modulation Techniques

v,(r)

_ _ _ _ _F_•1_G_U_RE
__ 9_
.5 M odulation envelope o r a double-sideband umplitude modulated carrier.

Envelope Detection of Double-Sideband Modulation

As we have seen, SSB amlDSB-SC demodulators require a synchronous local oscillator
for proper operation. An advantage of DSB with a carrier component (DSB-LC) is that
detection can be done without a local oscillator and mixer, by using an envelope detector,
Since it does not rec.iuire knowledge of the phase of Lhe incoming signal. an envelope detector
is an example of a 11011coherent demodulator. Because DSB-LC with an envelope detector
results in a much simpler receiver circuit. i( is the preferred method for broadcast AM radio.
where it is desired to make the receiver as inexpensive as possible.
The DSB-LC signal wavefonn of (9.24) can be written as

u;(t) =A LI + m(l) I coswrr,/ = r(l)cosw1Ff, (9.29)

where r (r) is tbe amplitude variation, or envelope, of the carrier. Figure 9.5 shows a plot
of the waveform of (9.29) for a typical modulating waveform 111(1). Note Lhat m(I) differs
from r (1) by a constant term.
The basic envelope clet.eL:tor c ircuit is shown in FigUJ·e 9.6a. When the DSB-LC signal
of (9.29) is applied lo the input of the circuit, Lhe diode allows charging of the capacitor
<luri11g U1e positive portions of the carrier cycle. During the negative p01tions the diode
does not conduct. so the capacitor begin~ to discharge through the resistor. The resulting
outpllt voltage is sketched in Figure 9 .6b. and is seen to approximate rhe acrual envelope
function. In reality. the carrier frequency is many times greater than the highest freque11cy
component of the modulation, so the output of a properly designed detector can track the
envelope much more closely than is indicated in the sketch. The RC time constant should be
large enough so that the capacitor voltage does not decay too quickly before the next carrier
peak an-ives, but small enough so the output can track the e nvelope when it is decreasing.
Also, rhe output of the envelope detector shown in Figure 9.6a has a DC level that must be
removed with a series capacitor. This may limit the low frequency response of the detector.
Also note that r(t) as defined in (9.29) may become negative if 1m l > l , a condition
referred 10 as overmodulation. The synchronous demodulator discussed above will $till
correctly recover Lhe modulating waveform in this case. bul the enve lope detector will not.

►I

(a) (b)

_ _ _ _ _F_I_G_U_RE
_ i' _
9._6 Envelope dc1cction of an A M signal. (a) Basic envelope detector circuit. (b) Output
wavefom1 of the envelope detector.
 9.1 Analog Modulation 297

DSB-LC

111(1)
lhl
So, No
IF Envelope
11rJ2 filter detector
(AWGN)

DSB modulator Channel and recei,c:r Envelope detector

FrGURE 9.7 Block Lllagram of a DSB-LC modulator and demodulator using envelope Jeteclion.
---------
This is because the output voltage of the envelope detector of Figure 9.6a is always positive,
and cannot Lrack a negative value of r(t). Thus broadcast radio systems must ensure U1at
me modulation index i:, always less than I 00% in order to avoid signal distortion. This is
also the reason why envelope detection cannot be used with a DSB-SC ~ignal.
The block diagram of a DSB-LC system using envelope detection is shown in Figure 9.7.
As in tbe case of 1he synchronous DSB demodulator, the IF bandwidth is 2/,~1 , where /M
is the maximum frequency range of the modulating waveform, m(I) = 111 cos w111 r. We can
derive the output SNR of the envelope detec1or by writing the input voltage of (9.24) as

V1(r) = Al.I +m(l)]COSWlf/ +n(r)

=[A+ Am COS Wm I+ x(t)] COSW1Ff - y(r)sinct11FI

= r(t) cos[w1Ff + 0(t)l (9.30)

where r(r) i:, the envelope of the carrier. as defined by

r(t) = JIA + Am cos w 111 t + x(f )]2 + y 2(t). (9.31)

The phase function 0(r) is defined by

y(t)
0(1) =tan- I - - - -- - - - (9.32)
A +Amcosw111 t+x(t)
The signal input power is

(9.33)

and the output signal power is given by the time-average of the square of the envelope
voltage with the noise terms set to zero:

" 1112A2
S" =A·+~- (9.34)

Thc first rerm in (9.34) represents power of a DC component, which would generally be
filtered with a DC block.
Evaluation of the noise power comained in the envelope voltage is complicated by me
pre~ence or the square root in (9.31 ). We can obtain an approximation for the special case
when the input SNR is large by assuming x(t) «
A and y(I) «
A. Then (9.31) reduces 10

r(I) =[A+ Amcosw111 1 +x(t)l l + .y2(r)

A+ Am cosw,,,t + y(t)
~ A+ Am cos w,,,r + x(f) (9.35)
298 Chapter 9: Modulation Techniques

This gives the output noise power as

2
N 0 = E{x (r)) = N 1, (9.36)
so the out_put SNR becomes
s,, 2ni2
(for large Si/ N;) (9.37)
N12+111 2 •
This is identical to the result of (9.28) for the synchronous demodulation of a DSB-LC
signal, showing that no degradation in SNR occurs with the much simpler method of
envelope detection for high input SNR.
When the input SNR is not large, however, the output SNR for envelope detection will
be reduced. ln fact, if we expand (9.3 1) as

r(t) = j A2 + 2A 2m(1) + 2Ax(t) + 2Am(t)x(t) + x 1(t) + A2m 2(1) + y 2 (t), (9.38)

we see that the e□velope voltage does not contain a term proportional to the modulating
signal m(f), but involves products of the signal and noise, the square of the signal, and
the square root funclion. This means that the output of the envelope detector is seriously
distorted and unusable for small input SNR. Figure 9 .8 shows a graph of the output SNR
versus input SNR. for both envelope and synchronous demodulation. For an input SNR
greater than about IO dB, the two cases are comparable. As the input SNR drops below
about 10 dB, the output SNR for envelope detection begins to decrease at n faster rate than
the unity-slope response of the synchronous case. This is called the threshold e.f/'ect.

Frequency Modulation
ln an FM waveform the instantaneous frequency of lhe carrier wave is varied in pro-
ponion to the modulating signal. We will see that the SNR for FM radio can be much better
than either SSB, DSB-SC, or DSB-LC, bul at Lhe expense of increased channel bandwidih.
This improved perfom1,mce has led Lo the applicalion oJ frequency modulation in broadcast
radio, Le!evision sound, two-way voice radio, and the AMPS cellular telephone system.

';0
~ (dB)
I)
30

20

10

- 20
Synchronous DSB
(slop~= I) '-..

Slope= 2 I.
I.
/4

_ _,_9_.8
_ _ _ _ _FI_G_URE Output versus i11pul SNR for synchronous and envelope detect.ion of a DSB-LC AM
signal.
 9.1 Analog Modulation 299

vco
~-------
S,, N,
L1mi1cr Discrim- Envelope
inutor de1.e ctor
m(t)
(AWGN)
'----r--'
PM modulutor Channel FM dcmouulutor

FIGURE 9.9 Block diagram of an FM modulator and demodulator.

---------
Figure 9.9 shows a block diagram for an FM radio system. A direct metJ10d ofgenerating
an FM signal is to control the l'requency of oscillation of a VCO with the modulating signal,
111(1 ). Demodulation can be accomplished by first passing tJ1e received (and downconverted)
[F signal tJ1rough a11 amplitude limiter, followed by a djscriminator and an envelope detector.
TheopernIio11 of the tlemodulator will becomecle,u·whcn we analyze thcSNRperformance.
A phase-locked loop can also be used to directly demodulate an FM signal, as discussed in
Chapter 8.
An FM waveform has the general form

ll(I) = A cos0(t), (9.39)

where the phase function 0(1) depends on the [F frequency and an arbitrary modulating
wavefom1111(1) as

0(1)=ww1 + k 1~ 0
m(r)dr. rad (9.40)

The instantaneous frequency (the rate of change of phase) of (9.39) i s given by

d0(1)
w = - - = cvIF + km(f) rad/sec. (9.41)
d1
For the speciaJ ca~e of a sinusoidal modulating signal 111(1) = m cos w,,,r. (9.40) reduces to

k111
0(t) = WIFI + - sinw,,,t. (9.42)
w,,,

If we define /',,.w = km as lhe maximum frequency deviation. and define /3 = /::,,(~>/w111 as

Lhc modulation i ndex, then (9.39) can be wrillcn simply as

v(r) = A cos(w11 ..r + f3 sin w,,,r). (9.43)

The FM signal of (9.43) has an infinite number of sidebands, in contrast to the finite-
bandwidth AM signals di scussed above. We can analyze the frequency spectrum of (9.-1-3)
by expanding the wavefonn into a Fourier series that i • peri odic in multiples of Wm- The
easiest way to do this is to write (9.43) as the real part of a complex exponential.

{9.44)

and then expand the second exponential as a complex Fourier series:

eJfhmw,.I
""
= L C,,eJIIW,.I.
11=-oc
300 Chapter 9: Modulation Techniques

The Fourier coefficients can be founJ as

c,, = _I l r/2 . . .
eJfl,11,w,.,1e- ;11w,,.1 dt, (9.45)
T l =-T/2

where the period is T = 2TC/w,,,. Changing variabies with x = w 111 1. and using an integral
identi ty for the Bessel function from Appendix B, reduces this result to

C,, = _I_
2TC
1"
.•·=-1r
ej((J sin., - 11x) dx = J,,(/3 ), (9.46)

where J,,(/J) is Lhe Bessel function of the first kind of order 11. Then the Pmirier series for
(9.43) can be written as

v(t) = A Re l e.l"'"'' , , too J,.(fJ )ei""'"'' ) = A

11
f;, 00
J,, (/3) cos(WIF + nw,,,)t. (9.47)

Thus there are sidebands spaced at multiples of w,,, on either side of the carrier at WIF, with
amplitudes given by A J,, (/3). Typical spectra for three values of fJ are shown in Figure 9.10
(note that we may have O ~ /3 < oo). Observe that Lhe sideband magnitude decreases for
large n., so the practical bandwidth is not infinite. ln fact, good transmission fidelity generally
only requires those sidebands whose amplitude is I% or larger relative to the unmodulated
carrier (Lhe 11 = 0 term); that is, for aJl 11 ~uch that IJ,,(fJ)I ~ 0.01. For large fJ, it can be
shown Lhat n. ~ /3 will satisfy this condition. In this case the required IF bandwidt11 is

B = 2ncv,,, ~ 2{J(v,,, = 2~w. (large fJ)

~10
/3 =0.1
________.,__.L
__.,_______-+-1 f
fu,
(a)

/3 = 1,0

•I
i ll'
(b)

r- ] .0
I
.i1 ~
I
I /3 = 5.0
I II III II fjp
I I
' •I
(c)

FIG URE 9.10 Magni Lude of the specLrum of a frequency modulated signal, for various valuesofU1e
---------- moduIation index /3. (a) f3 = 0.1, (b)/3 = 1.0, and (c) f3 = 5.0. The center frequency
is frr = wrr-/2,r. and the spacing of all sidebands is ;;,. = w,,, /2:rr.
 9.1 Analog Modulation 301

since f3 = t::.w/w111 • In lhe case of small {3, 11 ~ I for the last required sideband, so the total
IF bandwidth in this case is

B = 2nw,,, ~ 2w 111 • (small {3)

A convenient way to combine these two results is with the approximation known as Carson 's
rule:

B ~ 2(li.w + w111 ) = 2wm( I + /3). (9.48)

We can now derive the SNR for FM modulation. assuming a sinusoidal modulating signal.
The voltage ai the inpui to the FM demodulator shown in Figure 9.9 can be w1itten, for a
sinusoidal modulaling signal, as

v;(t) = A cos 0(t) = A cos(wll'l + /3 sin CtJ 111 l ). (9.49)

Then the input power of the carrier wave can be found from (9.49) with /3 = 0 as
S; = A 2 /2. (9.50)

The purpose of the limiter is to remove any possible amplitude variation from the input
to the discriminator, since this stage functions as a differentiator. If we assume the limiter
constant is unity, lhen the output voltage of the discriminator can be wrirten as

t11 (t ) = !!_V;(t ) = -(WrF + {3w 111 COS W111 f ) Sin(WJFI + /3 Slll W 111 1). (9.51)
c/1
After envelope detection (and DC blocking of the w1F term) the output voltage is

Vo(£)= f3w111COS W111I , (9.52)

wbich is seen to be proportional to the original modulating sig11al, 111(t). The output signal
power is
~2 2 (9.53)
So = p w,,J2.

Nexl we consider the presence of noise, and set /3 = 0 to compute the output noise
power. As we did for the case of envelope detection of AM, we express the noise voltage
using the narrowband representation of noise and convert to polar form:

s,(1) + 11/ (1) = A cos W1Ff + x(t) cos (V 1FI - y(r) sin WJPl
= r (t )coslwrFt -</J(t ) I (9.54)

where the envelope functi on is

r(1) = J[A + x(r)J 2 + y 2(r ),

and the phase functi on is

If we assume the noise voltage is small relative to the carrier amplitude. so that x(t) « A
and y(t ) « A, then we can use a small argument approximation to write the phase term as

(9.55)
302 Chapter 9: Modulation Techniques

The limiter will remove the effect of the time variation of r(t), so the noise output of the
discrimjnator will be
d
n 1(1) = - cos[wwl - ¢(/)J
dt
d¢(r )) .
= - ( W1F - --
dt
sm[wn:t - ¢(t)l (9.56)

After envelope detection and DC blocking. the output noise voltage is

cl¢(1) I dy(t)
no(/)= - -
dt
= -A - dt-. (9.57)

Since H (w) = jw for a differentiator (see Appendix B), the power spectral density of the
output noise can be found as
~ w2
S,,,.(w) = IH (wWSy(w) = A2 no for lwl < Wm- (9.58)

Then the output noise power over the IF passband from - wm LO w111 is
3
1 l w'" now 1 (9.59)
N0 = 2rr -w,. S,,o(w) dw = 3n ~ ; .

The input noise power over the same passband is

no
N 1 = -(2cvm )-
I
= -now,,,-. (9.60)
2 2rr 2rr
Finally, the output SNR of the FM demodulator is

(9.61)

where (9 .50) and (9.60) have been used to express this resttlt in terms of S1 and N1• Note thal
(9.61) indicates that the SNR of a demodulated FM signal can be improved by a factor of
3fJ2 /2. We can compare this with the corresponding result for AM demodulation by using
(9.28) with a 100% modulation index (111 = I),
2nS,

and taking the following ratio:

(9.62)

This comparison is made on the basis of equal transmit power, since we have expressed
each SNR in tenns of S1. The bandwidth of the FM signal, from !IF± /3.f;,,, is

while the bandwidth of the AM signal, from frF ± _f,,,, is

 9.2 Binary Digital Modulation 303

This allows (9.62) to be written in terms of bandwidth as

S I
0

N" FM =~ (BFM) 2 (9.63)

s(t [ 2 BA,,vr.
No AM
The results of (9.62) and (9.63) clearly iJJusti-ate the main advantage of FM relative to
AM, which is Lhat FM allows an improvement in SNR at Lheexpenseofincreased bandwidth,
white AM does not. For example. with a modulation index of fl = 4 the SNR for FM is
72 times better than that for AM, while the bandwidth is fo ur times larger.

9.2 BINARY DIGITAL MODULATION

Digital modulation methods are quickly surpassing analog techniques for modem wire-
less systems because of their improved pert·ormance in the presence of noise and fading,
lower transmit power requirements, and belier suitability for transmission of digital data
with error correction and encryption. ln this section we consider binary modulation methods
where a sinusoidal carrier wave is switched between one of two possible states according
to the binary symbols "O" and ·'J ." A carrier of the form A cos(wl + ¢) has three degrees
of freedom: the amplitude (A), the frequency (w), and the phase(¢). T hese correspond to
three fundamental binary modulation methods:

• Amplitude shiJt keying (ASK)

• Frequency shift keying (FSK)
• Phase shift keying (PSK)
FigUJe 9.11 shows ex_amples of waveforms for ASK, FSK, and PSK. in each case a
carrier wave is either switched on or off (ASK), switched between two disti11ct frequencies

"''" l
0 0 0
• I

ASK - ~ ~Nij \Mi

0
FSK fu~~WAWb~ ~WJWJMWW • I

PSK - ~
1
MA VIN NV\~JJ AM • r

FIGURE 9.11 Binary baseband data and modulated wavefom1s for ASK, FSK, and PSK.
304 Chapter 9: Modulation Techniques

0 0 () 0 0

FIGURE 9.12 Binary signaling methods. (a) On-off. or return-to-zero (RZ). coding. (b) Non-return-
- -- - - - -- - - ro -zero (NRZ) coding. (c) Polar NRZ coding.

(FSK), or switched between two distinct phase states (PSK), according to the binary data,
We will discuss methods of modulation and demodulation for each of these. and in the
following section derive expressions for the error rates of these methods in the presence of
noise. We wi II also briefly discuss the requi.rement for carrier recovery and synchronization.

Binary Signals
The baseband data source may be voice. fax, music, telemetry data. or computer data.
In the case of an analog source, such as voice, music, or telemetry, an analog-to-digital
converter can be used to obtain a digital representation. Several sources might be multiplexed
together for transmission over a single channel. and a parallel-to-serial converter can be
used to obtain sequentia l, or line coded, binary data. Thus we can assume baseband data
consisting of a serial bit sequence as shown in Figure 9. 12a. This representation involves a
return of the signal vol cage to zero before the end of Lhe bit period, and is called rerurn-10-
zero (RZ) signali.ng. A method that uses less bandwidth is the non-return-to-zero technique
of Figure 9.12b, where the signal voltage remains high for the duration of a binary ''I."
Another variation is to use the polar 11.on-ret11rn-to-zero (_polar NRZ) of Figure 9. I 2c, where
a binary "0'' is represented by a negative vo.ltage. Polar NRZ thus has an average DC value
of zero, which is an advantage in many practical implementations.

Amplitude Shift Keying

In amplitude shift keying the carrier wave is turned on and off according to the binary
baseband data sequence. ASK is also known as 0 11-0.ff keying. As shown in Figure 9.13a, an
ASK modulator can simply consist of a local oscillator and a mixer driven by NRZ binary
data, and is virtually identical to the DSB-SC modulator discussed in the previous section.
AlternaLively. an ASK modulator can be made with an LO switched on and off with a switch
controlled by NRZ data. The ASK waveform has a double-sideband suppressed-carrier
spectrum. As in the case of DSB-SC modulation, ASK can be demodulated coherently
using a synchronous local oscillator and mixer, but it is also possible to demodulate ASK
with an envelope detector.
 9.2 Binary Digital Modulation 305

bD
111(/) 111(1) COS Wol 111(1) co~ w0 1
t
1---- 111(/)

(a) (hJ

11
o 0
oI ___ ,
►I

(c)

FJGURE 9.13 Modulation and demodulation of ASK signals. (a} ASK modulator. (b) Synchronou~
demodulation or ASK. (c) Envelope detection of ASK.

A coherent ASK demodulator is shown in Figure 9.J3b. Lfwe write the ASK modulated
signal as

v(f) = m(r) cos<vot, (9.64)

where 111(1) = 0 or I, then the output of the mixer is

v, (t) = v(t) cos wot
I
= 3:1n(t)(J + cos2wor)

where we again ignore the mixer conversion const,mt. After low-pass filtering this reduces
LO

I
llo(I) = m(t). (9.6S)
2
which is seen to be on exact reproduction of the original baseband signal. Of cour!>e. thb
requires the LO to have precisely the same phase and frequency as lhe incoming signal. When
the local oscillator does not have perfect phase coherence or frequency synchronization.
distortion may be introduced. This is explored further in Problem 9.7.
Because the envelope of an ASK signal varies according to the binary baseband data, it
is possible to demodulate ASK noncoherenlly with an envelope detector. and thereby avoid
the requirement for a coherent LO. The basic scheme is shown in Figure 9. 13c. Although
envelope detection cannot be used to demodulate a DSB-SC signal. it can be used with an
ASK DSB-SC waveform because the modulating signal 111(1) is never negative. As discussed
in Section 9.1, the envelope detector functions by extracting the envelope waveform from an
AM signal. Equation (9.64) is of the form of an envelope 111(1) multiplying a high-frequency
carrier, so the same argument u ed in Section 9.1 leads to the conclusion that the output of
the envelope detector will be the envelope 111(, ).
306 Chapter 9: Modulation Techniques

FIGURE 9.14 Circuit for u rectifier detector.

- -- - -- - - --

EXAMPLE 9.2 SQUARE-LAW DETECTOR

Another noncoherent demodulator is the rectifier. or square-law. detector shown
in Figure 9. 14. It is similar to the envelope detector. wiLh the important difference
that a capacitor is not used. If the diode is biased to operate in its square-law region,
we can assume that the voltage across the resistor is proportional to the square of
the input voltage. [f the ASK waveform of (9.64) is applied to the input or the
square-law detector, derive the output voltage from the low-pass filter.

Solution
Assuming a square-law dependence, the resistor voltage can be written as

VN(I) = Cv 2 (t) = Cm2(r)cos2 wo1

C ,,
= m·(t)(l + cos 2w0 t)
2
After low-pass filtering. we then have the output

iJ0 (1) = 2C m2(t) = 2C m(t).

where the last step follows because 111 2(1) = 111(1) when 111 = 0 or I. 0

Frequency Shift Keying

Frequency shift keying involves switching a sinusoidal carrier wave between Lwo fre-
quencies, cv 1 and (J>i, , as illustrated in Figure 9.11. While w 1 and w2 can be selected arbitrarily,
in practice the frequencies are usually symmetrically chosen so that

= CVQ - D.W
<v 1
W2 = Cl!Q + /::,.(J)

where D.w is known as the ji-equen.cy de1 1ia,io11. The FSK modulated waveform can be
written as
for m(I) = I
v(t) = cos wt . OJ= { (VI (9.66)
CV2 for 111(1) = 0

In general the spectrum of m1 FSK signal is complicated because of the essentially

random switching between two frequency states. lf 2(t:,.f) = 2(t:,.r,>J2n) is Lhe difference
in carrier frequencies, and T is the bit period of the binary data, then it can be shown that
the effective bandwidth of an FSK signal is B = 2(1::,.f + 2/ T ).
An FSK signal can be generated using the modulator circuit of FiguJe 9.15a. which
employs a tunable oscillator to switch between cv 1 and w2. Coherent detection of FSK is
 9.2 Binary Digital Modulation 307

~□NRZ
_____
;,.__,,__ COS WI
cos "'' ,-........--½111( 1)
W= <t>1, Wz

nt(I)
(a) (bJ

FIGURE 9.15 M odulation and demodulation of FSK signals. (Ll) FSK modulator. (b) Synchronous
demodulation or FSK.

accomplished wilh the synchronous c.lemodulator of Figuro 9.15b, which requires two co-
herent local oscillators operating at w 1 and w 2. To analyze Ll1e operation of the demodulator,
=
assume we have the incoming FSK waveform of (9.66) with w w I . Then the outputs of
Ll,e top and bottom mixers arc

' 1t = I (1 +cos2w11)
u1(t) = v(t)cosw 1t =cos·tv (9.67a)
2
v 2(t) = u(r )coswit = cosw 1t cosn>it
1
=
2[cos(w1 - wi)t + cos(wI + cv2)t] (9.67b)

After low-pass filtering only the DC term from (9.67a) remains, resulting in a positive
pulse at the output of the summer. indicating that a "1" has been received. Similarly. ii
the incoming FSK signal has w = «>2, the output~ of the mixers will be reversed from that
of (9.67), and the output of the summer will be u negative pulse, indicating that a "0" has
been received.
FSK can also be demodulated using an envelope detector. thereby avoiding the require-
ment for two coherent local oscillators. TI1e FSK envelope demodulator circuit is shown
in Figure 9.16a. Operation is based on LJ,e fact that bandpass filtering can be used lo de-
compose the FSK signal into two ASK components. as shown in Figure 9.16b. These ASK
signals can then be individually demodulated using envelope or square- law detectors. The
required bandpass filtering is usuall y done at the IF stage. as opposed to RF filtering, to
ease the filter cutoff requiremcnL.~. The outputs of the envelope detectors are combined with
a summer to form a polar NRZ output.

Phase Shift Keying

In PSK modulation Lhe phase or the carrier wave is switched between two states, usually
0° and 180" . Tn this case the PSK waveform can be written as

u(t) = m(t)coswot, (9.68)

wherem(t) = I or-I. The PSK waveform can be generated by mixing a polar NRZ ver sion
of LJ,e binary data witJ, a local oscillator, as shown in Figure 9.I 7a. T he spcctmm of tJ1e
PSK waveform is relatively wide in bandwidth due to the sharp Lransitions caused by phase
reversal. These are usually smoothed by filtering, but the resulting bandwidth is usually still
wide enough that PSK is impractical for multichannel wireless systems.
Synchronous demodulation of PSK can be accomplished using the demodulator of
Figure 9. I7b. After mixing Lhe PSK si gnal of (9.68) with the local oscillator, the output of
308 Chapter 9: Modulation Techniques

cos (,J/
l---- .-111(1)

r-,...,, v,(I)
r-.....,, -
r-,._,
BPF-w2

(a)

Acos wt I1vnvmMmn
~mm onvrnnnnnmMn~n
mvvvvmm vv
no
• I

> I

" 1(/) 1
--l+J-l-l-lQOu.--4--1-nl--+1-on__._____'-1-1-on+l-o~. I
VV /0 0 VO

(bJ

FIGURE 9.16 Envelope detectil~n of FSK. (a) FSK envelope detector. (b) Decomposition of FSK
--------- signa I into two ASK signals.

the mixer is
I
v 1 (t ) = v(t)coswot =
2111(1)[ ! +cos2wotJ,
so, after low-pass filtering, the output voltage is

(9.69)

which is proportional to the original polar-NRZ data. Since the PSK waveform h as a constant
envelope, it cannot be demodulated with an envelope detector_ Note that it is possible to
add cos wol to the PSK waveform of (9.68) to obtain an ASK waveform, which could then
be demodulated with an envelope detector. But this procedure offers no advantage ove.r
coherent detection because it still requires a synchronous local oscillator.

m(f) 111(1) COS Wol 111(1) COS Wal

f m(t)

(pol ar NRZ) (Pol ar NRZ)

(b)

FIGURE 9.17 Modulation and demodulation of PSK signals. (a) PSK modulator. (b) Synchronous
- - -- - - -- - demodu lation of PSK.
 9.3 Error Probabilities for Binary Modulation 309

± cos (w0 r + <f>) cos 2(w0 t + 1M ,.....__, cos2(w01 + <f,)

- - - . - - - + - 1 ( · ,2 t - - - - - ' " ,.....__,
i ~ - - - - - - < ' " i +2

Squarer BP- 2fc

LO

.___► 111(1\ =± I

FJGURE 9.18 A carrier-recovery circull suitable for synchronou~ PSK demodulation.

Carrier Synchronization
We have seen that ASK. FSK. and PSK can each be demodulated using cohcrcm detec-
tors, where the local oscillator is in synclu·onism wilh the incoming carrier. Synchronism
implies that both frequency and phase are identical. a re!>ult that is generally difficult 10
achieve in practice. The effect of a phase error 6.¢ is that the output signal is reduced in
amplitude by cos 6.</J. while an error 6.w in frequency introduces a factor of co~ 6.wt (sec
Problem 9.7). As an example. if we set a criteria of requiring less than 45° phase error at
a carrier frequency of I GHz. synchronilation of the LO to the carrier must be better than
=
T/8 0. 125 nS. A free-running local oscillator will virtually never exhibit such synchro-
nism because of frequency drift. Doppler effects, and U1e arbitrary (and sometimes variable)
distance between the u·ansmiuer and receiver. An advantage of ASK and FSK i s U,at Lhey
can be demodul ated without a synchronous LO by using envelope detection, but we will
see that a penalty is paid in Lhese cases, since the bit error rates are not as good as those
obtained with coherent de1ecLion.
In general, there are Iwo ways in which a local oscillator can be synchronized with an
incoming carrier wave: transmit a pi lot. carrier, or use a carrier-recovery circuit. Transmitting
a low-level carrier is probably the easiest way. as this signal can be used to phase-lock the
local oscillator. The transmitted carrier passes through the same propagation delays as the
modulated signal. and so automatically arrives at the receiver with the same phase and
frequency.
Carrier recovery circuits use phase or frequency information from the received signal 10
i.ynchronize the local oscillator. Many of the variations of these circuits use a phase-locked
loop. which we study in detail i.n Chapter 8, but a clever alternative carrier-recovery circuit
for PSK demodulation is shown in Figure 9.18. A PSK signal of the form of (9.68), with
m(t) = ±I, is applied lo the input of the circuit. The phase ¢ of the waveform accounts
for the arbitrary phase delay between transmitter and receiver. A squarer circuit (such as
a diode operating in its square-law region) squares the input and effectively removes the
phase modulation. A bandpass filter is used co select the 2w0 component of the squared
signal, and a divide-by-two circuit yields an output of cos(wot + ¢). which is the desired
local oscillator signal synchronized to the incoming PSK waveform. Mixing with the PSK
signal then produces a polar-NRZ version of 111(1).
M odem digital demodulators w,ually operate at the IF stage, and employ digital signal
processing (DSP) circuits to perfon11 all functions or signal conditioning. carrier recovery
and synchronization. demodulation, and signal formatting.

9.3 ERROR PROBABILITIES FOR BINARY MODULATION

In an ideal situation a receiver will detect the same binary digit that was transmitted.
but the presence or noise in a communication channel introduces the possibility that eITors
will be made during the detection process. Here we derive expressions for the probability
of error for the binary modulations schemes discussed in U1e previous section. We will do
310 Chapter 9: Modulation Techniques

TABLE 9.1 PCM Signals for ASK, FSK, and PSK

Modulation s,(I) Detection

Type ul" "O'" Threshold

ASK V 0 V/2
FSK V -V 0
PSK V -V ()

this for synchronous detection of ASK. FSK, and PSK, and envelope detection of ASK and
FSK. Although envelope detection is simpler to implemenr than synchronous detection,
a price is paid in the error rates for envelope detection because of the absence of phase
information.

PCM Signals and Detectors

ASK, FSK, and PSK s ignals can all be considered as special cases o r pulse code
modulation (PCM). Ln general. PCM codes u binary " I" as a signal voltage St (1), and a
binary ·•o•· as a signal voltage s 2 (t). each of bit duration T. In the transmitter these PCM
signals are used to modulate a carrier by amplitude, frequency, or phase modulalioo. ln the
receiver, as we have shown, synchronous demodu lation or envelope detection (for ASK or
=
FSK) can be used Lo recover s(t) s 1(1) or s2(1 ). The PCM signals for the cases of ASK,
FSK, and PSK modulation can be summarized as shown in the Table 9.1 . The demodulated
signal is then used to make the decision as to whether a binary" 1" or "O" bas been rnceived.
In the absence of noise this can simply be done by setting a detectinn threshold, where
•· r• or "O" is decided using a comparator circuit to determine whether s(t) is greater or
less than the tlu·eshold value. For FSK and PSK, which use symmetric voltages of ±V, the
optimum threshold is obviously zero, while for ASK the optimum threshold is V/2. The fact
that the optimum threshold depends on the actual value of the received signal is a sedo~s
disadvantage or ASK.
When noise is present the problem of detection is more difficult. It can be shown in
this case that optimum detection of PCM signals can be made wiLb a correlc11io11 recefrer.
consisting of an integrator (or low-pass filter) followed by a sampler and a comparaLOr. (11
can be shown that this circuit is equivalent to the optimum matched filter for PCM signals.)
The basic circuit is shown in Figure 9.19a. As illusu·ated in Figure 9.1 %, the integrator
integrates tJ1e incoming signal and noise over the bit duration, T. The output of the integrator
is sampled at the end of U1e bit period, and a comparator is used to determine whether lhe
output voltage is above or below the threshold value. The result is a close approximaUon
of the original binary message m(l). Although tJ1e effect of the (zero mean) noise voltage
is minimized by averaging over the bit period, the random nature of the noise results in
occasional errors, and the likelihood of an eITor increases witJ1 the power of the input noise.
The li.kelihood that a single bit is received inco1Tectly is called the probabili1y of error. or
the bit error rare (BER).
Tf we assume that the input signal and gaussian noise voltage is of the form s(t} + 11(1),
=
where s(t) s 1(1) or s2(1), then the sampled output of the integrator in Fig-ure 9. 19a can be
written a!:I

v,iT) = f T [s(t)
1= 0
+ n(l)J di = s11(T) + 11,iT), (9.70)
 9.3 Error Probabilities for Binary Modulation 311

.1·(1) + 11(1)
- - ' - -- -a~
f 1
( ) ,/1'
s,,(I) + 11,,(1)
T
s,,(T) + 11.,(T) §
Threshold
111(1)
o Sampler I. I. 0. I

Jntegra1or Comparutor
(a)

.l s(I)
0
+1• 1---'1_....:...- VT

T 2T 3T
-v - VT

(b)

_. E
_ _ _ _ _F_I_G_U_R _ 9_.1_9 Optimum correlation receiver for PCM detection. (a) Correlation dctcclor circuit.
(b) Input and output signal voltage waveforms.

where lhe integrated signal voltage is

T I VT
if s(t)=V
So(T)= i s(1)dt = -\IT ifs(t) =-V (9.71)
1=0 0 ifs(t)=O

and the integrated noise voltage is

11 0 (T) =J T n(t)dt. (9.72)

l:aO

We further define the bit energy as the energy of the (non-zero) signal voltage over one bit
period:

Eb=
! ,T

• t= O
2
s (t)dt =VT.
2
(9.73)

The noise power outpul from an integrator with a bandl imited white noise .input was derived
Lil (3.38) a~

noT ,
N,, = =cr. (9.74)
2
where 110/2 is tbe two-sided power spectral density of the white noise, and a 2 is the variance
of the gaussian probability distribution functjon (don ·t confl.1se 11 0 (T ) with 11 0). We can now
derive the probability of en-or for the various binary mbdulation methods separately.

Synchronous ASK
Assume that a binary "O" has been transmitted in an ASK system, in the presence of
bandlimited gaussi an white noise. Thus we have s(t) s2(t) 0 and s0 (T) 0, wid1 a = = =
detection threshold of VT/2 (the threshold musl also be integraLed over tJ1e bit period).
Correct detection occurs when the output signal and noise voltage from the sampl er is less
than the threshold. Conversely, an e-rror will occur i f lhe output signal and noise voltage is
312 Chapter 9: Modulation Techniques

greater than the threshold. Thu~ we can write lhe probability of error when a '"ff. is sent as

7
P?>=P{su(T)+n 0 (T)> \ } =P {11 0 (T)> V/ }

(9.75)

since s,,(T) = 0. This integral can be expressed in term.~of the complementary error function
by using U1e change of variable x = no//2.a:

p"(Ol = -I
.fir
1"'° .,=~r,;
,
e- , · dx = -l e,fc ( - VT-
2 2../?.a
) .

The argument of the complementary e1Tor function can be expressed in terms of£,,, the bit
energy, and the power specLral density of the white noise by using (9.73) and (9.74):

Then the final expression for the probability of error when a ·•o•· is sent via ASK is

pW> = ~e,fc ( {e;:) . (9.76)

, 2 Y¼
It is left as a problem Lo show that Lhe probability of error when a " I" is transmitted, PJI',
is equal to p_!O>when lhe threshold is set at VT/2. ln this case Lhe total probability of error
of a particular bit in a message having equal occurrences of "0" and •• 1•• bits is

p
e
= ~p (O) + ~p( I ) =
2 ,, 2 ,.
p(O)
~
= p(I)
,
= ~erfc (
2 • Y¼
(E;;) · (9.77)

Synchronous PSK
Now assume Lhat a binary "0" is transmitted in a PSK system. i.n the presence of
bandlimited gaussian whhe noise. Then we have s(r) = s2(1) = - V and s0 (T) = -VT,
with a detection threshold of zero. An error will occw· if the sampled output voltage from
U1e integrator is greater than the threshold. so we can write

P;°l = P{s,,(T) + 11 0 (T) > 0) = P(n,,(T) > VT)

(9.78)

where the change of variable x = no/ ../?.a was again used lO evaluate the integral, and
(9.73)-(9.74) were used to express the argument of the complementary error function in
terms of Lhe bit energy and lhe power spectral density of d1e noise. Due to the symmetry of
the PSK signal and the demodulator, P?> = PJDl.
Observe lhat the bit energy-to-noise ratio in the argument of Lhe complementary error
function for the PSK case cWfers by a factor of four (6 dB) compared to the result for ASK.
This implies that. for the same probabil ity of etTor, PSK requires only one-fourlh the pow~r
of an ASK system . Since a~ ASK signal is off half the time, however, the average transmit
power of an ASK system is half that of a PSK system, for the same peak power (same signal
 9.3 Error Probabilities for Binary Modulation 313

voltage, V). 11ws, in terms of average transmit power, the PSK result is better by a factor
of two (3 dB), compared wilh ASK.

Synchronous FSK
Nexl we consider a synchronous FSK receiver, a11d assume lhat a binary "0" is transrnil-
ted in the presence ofbandlimited gaussian white no ise. Then we haves(, ) = s2(t ) = -V
and s,,(T ) = - \IT, with a detection threshold of zero. These signal levels are similar to
the PSK case. but in the synchro nous FSK demodulator output noise consists of the dif-
ference between noise that has passed through both the cv 1 and cui channels. As shown
in Figure 9. l 5b, if n 1(t) and n2(r) are the noise voltages due to the u>1 and cv2 channels,
respectively, then the total output noise voltage from the demodulator is
(9.79)
2
The noise voltages n 1(t ) and 112 (1) are both gaussian wilh zero mean and variance a , but
are uncorrelated because they are taken ove r different filter passbands from a white noise
spectrum. The variance of n(I) can be computed as

E [n 2(t )} = E j n~(I) - + n~(I)}

2 11 1(f )112(!)

= E lnf(t) } + E (n~{t ) } = 2a 2 (9.80)

since E {n 1(1)112(t )) = E {n 1 (t)} £(112(1)) = 0 because n 1(t) and n2(t ) are uncon-elated and
have zero mean. The result of (9 .80) shows that the total noise power of the FSK demodulator
is doubled relative lo the synchronous ASK or PSK demodulator.
The probability of error for synchronous FSK is calculated i.t1 the same way as for PSK,
but with a 2 replaced with 2a 2 :

P}Ol = Pl s,,(T) + 11 (T ) > 0 } = P(n 1(T) - t12(T ) > VT)

= J oo e-11lf4a 2 d no = ~ e,:f'c (VT) = ~e,fc ( {E;) (9.8 1)

no=VT J4rra2 2 2cr 2 V¼
The change of variable x = n.o/2a was used to evaluate the integral, and (9.73)-(9.74) were
used to express the argument of the complementary error function in tenns of the bit energy
and the power spectral density of the noise. As before, P,( 1) 0
P,~ ). =
Observe Uiat synchronous FSK requires 3 dB more signal power than an equivalent
PSK system for the same probability of error, and 3 dB less power than an ASK system
on a peak power basis. FSK and ASK, however, have equal error rates when compared in
terms of average transmit power, since an ASK system lransmits power only half U1e time.

Envelope Detection of ASK

Although envelope detection of ASK is simpler than synchronous detection, the deriva-
tion of the probability of error is mo re diffi cull because of the effect of the nonlinearity
of the envelope detector on the noise. Let the i.t1coming ASK signal and noise voltage be
wtitten as

v(t ) = m(t) cos w 0t + n(t ), (9.82)

where m(I) = V or 0, for a trans mitted " 1" or "0.'' respectively. The bandpass gaussian
white no ise 11(1) cai1 be expressed using Lhe narrowband representation of (3.43) as n.(t) =
x(I) cos wot - y(t ) sin w0 t. Then (9.82) can be expressed as an envelope function r (t) times
314 Chapter 9: Modulation Techniques

the carrier:

v(1) = [m(t) + x(r)J cos wol - y(t) sin wol

= r(l)cos[wo1 + 0(1)] (9.83)

The envelope and phase funcLions are defined as

r 2 (1) = [m(1) + x(t)J2 + y1(1), (9.84)

y(I)
tan 0(t) = -m(t)- --
+ x(t)
(9.85)

As shown in Figure 9.13c, the output of the envelope detector will be Lhe envelope
function r(t). T he data m(t) can be recovered from the output of Lhc envelope detector by
using the PCM threshold detector of Figure 9.19a. The probability of error thus depends on
the statistics of r (t), which are not gaussian and. in fact. depend on whether m(I) = 0 or V,
First consider the case where m(r) = 0 . Since x(t) and y(t) are independent gaussian
random variables. we can form the joint probability dist1ibution function as

(9.86)

for -oo < x.y < oo. Since in this case r 2(1) = x 2 (t) + y 2(f), we use the change of
variables

x(l) = r(t)cos0(1)
y(t) = r(t)sin0(t)
to convert from rectangular to polar coordinates. The differential element is dxdy = rdrd0.
Applying this transformation to (9.86) gives the joint pelf of r(I) and 0(t) as
_ ,2 /2al

J,-o(r. 0) = fn(X,
-
y) = _
e _. - 2- ,
2ircr
(9.87)

where O :5 r < co. T hen the pdf of r(/) is computed as

(9.88)

for O :::: r < oo. This is the Rayleigh probability distribution function, which occurs in
many problems of probability that involve circular symmetry; a common example is lhat oi
throwing darts at a circular target. The probability that a particular sample lies in the annular
ring from r 10 r + cir is given by .f,.(r)dr. We also used lhe Rayleigh pdf in Section 4.5
to describe the statistics of fading in a multipaLh propagation environment. If the envelope
detection lhresbold is set at r0 , the probability of error when sending m(t) = 0 is

(9.89)

Now consider transmission of m(r) = V, and let x' = x + V . Then x' is a gaussian
random variable with mean V , and the pdf for x' is

(9.90)
 9.3 Error Probabilities for Binary Modulation 315

The envelope and phase funcLions in this case are

r 2(t) = [ V + x(r)]2 + y2(1) = x'2 (1) + y 1(t ), (9.9 1)

v(I) )'(I)
tan 0(1) = • = -- (9.92)
V + x(t) x'(f)
If we use Lhe change of variables

x'(t) = r(T) cos 0(r)

y(f) = r(t ) sin 0(t).

Lhen the joint probability distribution function can be found as

- [(,•'-Vll-1 ,,l J/2t'Tl
J~o(r, 0) = J,,ix',
•
y) = f~ •(x')fiy) = e
• 21ra-
?

1
e- "' 212rr e-(r1-2rco,11)/2az
= - - - - -2- - - - (9.93)
2no-

since (x' - V)2 + y 2 = (r cos0 - V)1 + r 2 sin.:?0 = V 2 + r 2 - 2Vr cos 0. Then Lhe pdf
of r (f) can be calculated as

2,,. r<,-v'r1.,,1 1 2" '

fr(r) =
111=0
}~o(r. 0)r d0 = ,
11Ta- o-o
e-Vrco,il/ 2<,. d0. (9.94)

The integral in (9.94) can be wri1ten in terms of the modified Bessel function of zero
order, lo(:c) (sec Appendix 8 ):

/~(r) = ;2 e-cvi+, 2)/20 1 l o ( :~). (9.95)

for O ~ r < oo. This is known as the rician probability distribution function. named after
S. 0. Rjce, of Bell Labs. who in 1he 194Os derived this result in connection with early
communications systems. The peak of (9.95) occurs when r =
V. Note that the rician
distribution reduces to the Rayleigh pdf when V =
0, a!> expected; it is left as a problem 10
show Lhal the rician pdf approaches a gaussinn distribution when the argument is large.
The probability of error wlien scndfog m(/) = V is then found as

P?> = P{r(t < ro)} = '" [ 'o ,- e •<V'+r'>tio' lo ( Vr) dr.

[ ,=O fr(r)dr = •=OCT
1 2a
(9.96)

where ro is Lhe detection threshold. The integral in (9.96) cannot be evaluated in cloi.ed
fom1. but must be calculated numerically.
The expressions for P?>and P:1> given in (9.89) and (9.96) differ because the statistics
of the noise arc different for m(t) =
0 and 111(1) = V. If Po is the probability Lhat a given
transmitted message bit is a ''O,'' and P1 is the probability ofa" 1:· then the overall probability
of error for envelope detected ASK is

Pe = Pio P'oi
~
+ P1 pe<I> • (9.97)

Tbe bit error probabilities P;01 and P:' 1 depend on the detection threshold r0 , as well as the
ratio V /a; these expression~ can be written in 1crm s of the (peak) bit energy-to-noise ra tio
using the relation that V 2 /a 2 = 2£1,/110 .
Figure 9.20 shows a graphical interpretation of the detection process. The Rayleigh and
rician probability distribution functions are plotted versus r(t ), and the error probabilities
when sending a "O" or a ·' 1" arc shown as the appropriate areas under Lhese curves. Por
316 Chapter 9: Modulation Techniques

FIGURE 9.20 Envelope detection or ASK The optimum detection threshold occurs at the intersec-
- -- -- - -- - - lion of the Rayleigh and rician puf.

lhe usual case where Po = P 1 = 0.5, the probability of enor will be minimized when I.he
detection lhreshold is chosen at the intersection point of the probability distribULion functions
for Pf1l and P} 1l (see Problem 9.20). ln general, the optimum threshold is a function of V
and a. and must be fou nd numerically, but for lhe case where lhe inpLtt SNR is very large
it can be shown that an approximate value is ro = V /2. The fact that the threshold depe11ds
on the received signal I.eve! is a disadvantage of ASK. particularly when fading is present.

Envelope Detection of FSK

Derivation of the probability of error for noncoherent FSK is si milar to the procedure
used for noncoherenl ASK, since lhe two channels of the FSK demod ulator of Figure 9.16a
essentially decompose the FSK signal into two ASK signals. If frequency w 1 is transmitted
for a "l :• then the output of the envelope detector for w1 wi 11 consist of the detected signal
plus noise:

(9.98a)

while the output of the envelope detector for cu2 will consist of only noise:

(9.98b)

The converse results apply when frequency w2 is transmitted for a ··O." This symmetry
implies that the error probabilities are thesnme for sending a "1 "ora "O" and. since the output
of the demodulator is fom1ed as r 1(t) - r2(t ). the detection Lhreshold can be set at zero.
The probability of error when a'· J" is sent can be written as

(9.99)

Jn this expression botb r 1(1) and r2 (t ) are random variables. Since 1·1(1) consists of signal
plus noise, it has a rician pdf; r2(1) has a Rayleigh pdf since it consists of noise only. lf we
temporarily assume a fixed value of r 1(t), the probability of error can be computed as

We must now integrate this result over all possible values of r 1, weighted by the pdf of r 1(i).
 9.3 Error Probabilities for Binary Modulation 317

to find the overall probability of error:

(9.1 00)

where we have substituted the Rayleigh and rician density functions from (9.88) and (9.95).
While this expression might appear formidable, in fact it can be reduced to a very simple
result. First, the inner integral can be evaluated directly as

which reduces (9. 100) to

Pe( I ) = l oo ~ 2 e - v i12n2
e- ,-f!a2 / O( Vr1? ) d·/J . (9. 101)
r,=0 a a~

Now use Lhe change o f variable x = fir1, which gives

Pe( I ) -- ~ e - V 2/ 4q 2
2
1"" !......
,,•=O a
2e
-(ll
2
/2+x' )/2a~ , o ( ~ ;;;
v 2a 2
) t,/ ,t.
·• (9. I02)

2
The integrand of (9. 102) is identical to the rician pdf of (9.9S), if V 2 is replaced with V / 2.
Thus the illtegral i.n (9. l02) must be unity. This gives the final result that

(9. 103)

Bit Rate and Bandwidth Efficiency

Each of the above expressions for probability of error is expressed in terms of Eh/ no,
the ratio of bit energy to noise power spectral density. T he dimensions of Eb are W-sec,
while the dimensions of no are W / Hz , so the ratio is dimensionless. In practice it is more
convenient to write the bil energy in terms of signal power and the data rate. Let R11 be
the bit rate of the binary message signa l, with dimensions of bits per second (bps). Then
the signal power is S = R1,E1,, i.n Watts. The bit energy-to-noise density ratio can the□ be
written as

-= - - (9.1 04)
no noRb
Thus tJ1e probability of error is determined solely by the carrier power, the bit rate, and the
PSD of the input noise. Since all our expressions for probability of error are monotonically
decreasing with an increase in E1,/ no, (9. 104) shows that the en or rate will increase with an
increase in bit rate. for a fixed noise level. Note that (9. 104) is independent of tJ1e receiver
bandwidth.
We can also express the bit energy-to-noise density ratio in terms of the SNR of tJ1e re-
ceiver. Tf the receiver has an IF bandwidth t::.J, then the noise power is N = (2.6.f)(no/2) =
no6.f, for noise with a two-sided PSD of no/2. T hen we have
E1, S .6.f
-=-- (9.105)
318 Chapter 9: Modulation Techniques

Depending on the type of modulation, the required receiver bandwidth may range from one
to several times the bil rate.
Each of 1he above binary modulation methods transmit one bit during each bit period,
and are therefore said to have a bandwidth efficiency of I bps/Hz. Thus. for example, an
analog telephone circui1 having a bandwidth of 2400 Hz (600 Hz-3 kHz) is limited to a
maximum data rate of (2400 Hz)(l bps/Hz)= 2400 bps when used with binary modulation.
This rate can be approached with FSK, but we will see later in this chapter that substantially
greater data rates can be achieved with multilevel modu lation methods that transmit more
than one bit per period. Such methods have bandwidth efficiencies greater than l bps/Hz.
Thus the terms data rare and bandwidth are nor synonymous, in spite of the often-heard
(but incorrect) use of the word "bandwidth" when referring to high data rates.

Comparison of ASK, FSK, and PSK Systems

Now that we have derived expressions for the probability of error for ASK, FSK,
and PSK systems, we can compare the relative advantages of each. Figure 9.21 shows a
comparison of bit error rates for ASK, FSK, and PSK. Note that the lowest error rate occurs
for the coherent detection of PSK, but the price paid for this performance is the need for a
synchronized LO and a relatively wide signal bandwidth. Nexl in order of performance is
coherent FSK. which requires 3-4 dB more power than PSK for the same error rate. The
error rate for noncoberent (envelope detected) FSK is only slightly worse (about 1 dB) than
for coherent FSK. and is better than coherent ASK. This conclusion onJy applies, however,
when using the peak power for ASK: when the average power of the ASK signal is used
(he error rate for coherent ASK and coherent FSK are identical. The worst perfom1ance is
obtained with noncoherent ASK.
The error rate that is required in practice depends on the application, as well as 1rade-
offs between bandwidth, Lransmitter power, range, and receiver complexity and cost. It
may range from 10- 2 (im.age data from early space probes) to I0- 8 for high-data rate
computer networks. A typical error rate that is often used as a goal for modern wi'reless

10°..---- - - - -- - - -- - - - - -- - ---,

~~ I 0- 2
...
I

e~ 10- 3
--
._
0

Coherent
PSK

o-
I 7 L-J.-1-..1....JL.....L-'-'-'--'--'-'-'-..,_L.....L-'-.L.-l.- ' - - ' - 'J....L....,_u....u..__,__.u.....--'--'
- 10 -5 0 5 10 15 20
E1,tu0 (dB)

FIGURE 9.21 Comparison of bit error rates for coherent (ASK, FSK. and PSK) and noncoherent
- - - -- - - -- - (ASK and FSK) demodulation. The results for ASK are based on peak. bit energy.
 9.3 Error Probabilities for Binary Modulation 319

TABLE 9.2 Eb/110 for P, ::::: 10-s for Various Modulation Methods

Modulation Method

coherent PS K 9.6 dB
coherent FSK 12.6 dB
noncoherent FS K 13.4 dB
coherent ASK 15.6 dB
noncohcrent ASK 16.4 dB

o-
systems is I 5 : the Table 9.2 lists the required E1,/ II() for various binary modulation meth-
ods. ASK transmitters are very simple, and efficient since no power is r:idiated when no
data is being sent. ASK receivers arc also simpl e. if envelope detection is used. Because
the bit error rate is poor in comparison to other modulation methods, ASK is limited to
low data rares. In addition, the fact that the detection threshold depends on the received
signal level makes ASK pe1formance very poor in a fading environment. For these rea-
~ons. ASK applications are usually limited to shon-range, low-cost telemetry and RFID.

FSK systems can use a zero threshold, regardless of signal strength. An FSK transmitter
is only slightly more complicated than for ASK, and an FSK receiver using envelope
detection can be made l>imply and inexpensively. In addition. the error rate for noncoherent
FSK is comparable to cohercm FSK, and much better than noncoherent ASK. Because
of these features, noncoherent FSK has found widespread historical appl ication in a wide
variety of both baseband and modulated data transmission systems, such as data modems.
teletype, and fax.
PSK give~ better error rate performance than any other binary modulation method, but
requires synchronous detection. PSK has a constant envelope, and uses a zero detection
threshold. This cases requirements on the transmitter power amplifiers. and makes the
performance of PSK in a fading environment bcrter than ASK or FSK. PSK also requires
a relatively wide signal spe1.:trum, typically ranging from 2R11 to 4R1,. For these reasons,
applications of PSK are generally limited to high performance systems such as space and
satellite communications, such as interplanetary space missions and the Global Positioning
Satellite (GPS) system.

EXAMPLE 9.3 COMPARISON OF FSK AND ASK MODULATION

A wireless local area network operating at 2.44 GHz transmits data at the rate of
1.15 Mbp!>, with a desired error rute of I 0- 5 . The transmitter and receiver each
use a monopole amcnna wilh a gain of 4.5 dB. The transmit power is 0.5 W, and
the receiver and receive antenna have a combined system noise temperature of
600 K. Compare the maximum possible operating range for coherent FSK and
noncoherent ASK modulation

Sol11tio11
From Table 9.2. £ 1,/110 = 12.6 dB= 18.2 for coherent FSK with P~ = I0- 5 .
The wavelength is

300
'A.=
2440
= 0.123 m.
320 Chapter 9: Modulation Techniques

Using (9.104) LO find the required receive carrier power gives

• - 2.1 6 0
P, = -E1, (noRb) = -kT,ysRb
Eb
= ( 18.2)(1 .38 X 10 )( 0 )(l.15 x I 06 )
n.o 110
= 1.73 X 10-!3 W

Then the Friis formula gives U1e range as

P1G, G ,.).2 (0.5)(2.82)2(0. 123)2

R= ( rr)2P,
4
=\ (4rr)2(1. 73 x _ ) =47km.(coherentFSK)
10 13
5
For the case of noncoherent ASK. Eb/no= 16.4 dB= 43 .6 for P, = L0- . The
required receive carrier power is now

£1>
P,= -(110R1,)= -Eb k1,ysRb=
'. (43. 6)(1.38 x lO- 23)(600)(l. 15 x I 06 )
no no
= 4. 15 X l0- 13
W

Then the maximum range is

(0.5)(2.82) 2 (0.1 23)2

R= = 30 km. (noncoherent ASK)
(4n)2 (4.15 X 10- 13)

We see that the use of noncoherent ASK modulation leads to a reduction of

about 25% i.n maximum range, compared with coherent FSK. Because of less than
ideal propagalion conditions, such as blockage, attenuation, and fading effects.
these ranges would not likely be achieved in practice. 0

9.4 EFFECT OF RAYLEIGH FADING ON BIT ERROR RATES

As we discussed in Section 4.5, the presence of multiple signal paths between trans-
mitter and receiver lead to Rayleigh fading. Thjs causes sharp reductions in received signal
power over short time intervals. or with movement over small di.stances, due to phase cancel-
.1 ation. Fading has the effect of significantly increasing the error rate for digital modulation

o r, equivalently, decreasing Lhe usable range of a wire.less system (for a fixed ElRP, error
rate, and system noise level). Although the average receive power is not greatly reduced.
fading has a major impact on error rates because the presence of brief but deep drops in
received power results in very large error rates for short periods, which has the effect of
greatly increasing the average error rate.
Figure 9.22 shows an idealized model for a propagation c hannel with fading and addi-
tive noise. The input signal is s;(t), and the multjpJicative factor r (t ). with O::: r(t) < oo,
represents the change in amplitude due lo Rayleigh fading. White gaussian noise is added
after the fading factor, and may include noise received by U1e receiver amenna as well ns
noise generated in the receiver itself. As derived in Section 4.5, r(I ) is a random variable

r
· (1) - n (1) +11(1)
'

FIGURE 9.22 Model for a propagation channel having additive white noise and fading.
- - - - -- - - - -
 9.4 Effect of Rayleigh Fading on Bit Error Rates 321

with a Rayleigh PDF given by

(9.106)

for O ~ r(t) < oo. The rrns value of the distribution of r(t) is ,/fa (we use a here to avoid
confusion wilh a. the rms noise voltage). We can modify the expressions for probability
of error derived in the previous section to uccounl for Rayleigh fading by considering
the conditional probability of error for a fixed signal amplitude, then integrating over the
Rayleigh probability distribution function for L11i!> amplitude. This can be done for coherent
ASK, FSK, and PSK, and for noncoherenl FSK. The results show that realistic Rayleigh
fading can significantly increase the average bit error rate. and thus have a major impact 011
ilie performance of wireless systems.

Effect of Rayleigh Fading on Coherent PSK

First consider the case of coherent phase shift keying with Rayleigh fading. Lf we
consider r(t ) to he a fixed multiplier, then the probability or error derived in (9.78) can be
modi lied to give

(9.107)

The overall probability of error is now found by weighting (9.107) with the Rayleigh PDF
of (9. 106). and integrating over the range of r:

To evaluate this integral we first use me change of variable 11 = r 2 £,./ 110 . and the definition
of the complementary error function

erfc(.1) = 2
ft fI
"C
e - x2 d r.

lo rewrite (9. I08) as

(9. 109)

Next we define
2a 2 £b
r=-- (9.l lO)
no
which is the average received bit energy-lo-noise power spectral density ratio of the faded
received signal. Then (9. 109) can be simplified to

(9.1 1l)

2
The inner integral can be evaluated by using integration by parts. Let U = f :,.;;; e- " dx,
322 Chapter 9: Modulation Techniques

and dV = e- 11
/ r du, then .f UdV = UV - .f VdU reduces (9. 11I) to

= --
l { for
---r
Joo e - (1 + 1/ n11
----du
}
fir 2 11=0 2./u

(9.112)

The last integral in (9. 112) is a standard integral, aud is given in Appendix B. Because of
Lhe similarity of (9.76), (9.78), and (9.8 1), virtually identical derivations can be made for
the cases or coherent ASK and FSK; lhese are left as problems.
Equation (9.1 12) is the final result for the probability of error for PSK with Rayleigh
fading. Note Lhal this result is expressed in terms of r, defined in (9. 110) as the average
energy-Lo-noise ratio. While the probabi lity of error for large £1,/ 11.0 for the nonfaded case
decreases exponentially (see Problem 9. 17), (9. l 13) decreases much more slowly. Also
observe that (9. I 12) does not reduce to the non-faded result of (9. 78) for limiting values
of r = 0 or I; this is because Rayleigh fading does not include a line-of-sight term, which
is the assumption for the nonfaded case. When the fadi ng environment includes a strong
line-of-sight component the fading statistics become rician distributed, and in this case the
expression for probability of error will reduce to the nonfaded resul t when V = 0 [4J-[5l

Effect of Rayleigh Fading on Noncoherent FSK

Next consider the effect of Rayleigh fading on the error rate of noncoherent ASK. In
this case the derivation is easier because the non-faded expression for probability of error
in (9. I 13) is simpler.
Following the same procedure as in U1e previous case, the conditional probability of
error for noncoherenl FSK for a fixed amplitude factor r is

pC (E/; I r) -- ~e-
2 r i E,,/110
• (9.1 13)

Weighting this result with the Rayleigh PDF of (9.106) and integrating over the range of r
gives

I
= -- - ~- = - - (9.114)
2( 1+ a::1,) 2 +r
The required integral i.n (9.1 l 4) is listed in Appendix B. As before, we defu1e r in (9.110)
as the average bit. energy-to-noise ratio.

Comparison of Faded and Nonfaded Error Rates

Figure 9.23 shows the effect of Rayleigh fading on the bit e1rnr rates of coherent PSK
and noncoherent FSK modulation. The results are p lotted versus r = 2a 2 £1,/ no, the rms
average of the received bit energy-to-noise power density ratio. The error rates fo r Lhe faded
 9.4 Effect of Rayleigh Fading on Bit Error Rates 323

,oo

10-•

Q.,.~ 10 l
..
0
I:: 10- 3
'IJ
'-
0
....
,:5 I OJ
...
.D

..
.0
0
C. I0-5
- - Coheren t PSK
- - - Noncohercnl FSK

I0 7 L......-'--'---'------'-'---'---'-L.:...L-'---'---'-.L.....'-'--'---'----'-l........-'
- 10 0 10 20 30 40

_ _ __ _ F_I_G_U_R
_E_ 9_.2_3 The effect of Rayleigh fading on lhe bll error rares of coherent PS Kand noncoherent
FSK.

cases are given by (9. 11 2) and (9.114). For the nonfaded ca:.e~ we lei 2cx 2 = I, so that
r = E b/11 0, and use the expressions given in (9.78) and (9. 103).
Note the dramatic increase in the probability of error when fading is present. For
example, for Pr = 10- s, fading has the effect of increasing the required biL energy-to-
noise ratio by approximately 30 dB for both PSK and FSK. Since transmit power is usuaUy
limited by regulation, fading has the ultimate effect of reducin g the usable range of a wireless
sy~tem, for a given data rate. error rate. and noise level. Fortunately, because most of the
errors on a l"ading channel occur in short bursts, error-correcting codes can be used very
effectively to improve 1he error rate.

EXAMPLE 9.4 EFFECT OF RAYLEIGH FADING

A cellular phone base station using BPSK transmits at 882 MHz with an EIRP of
50 W, and an antenna height of 50 m. The mobile receive antenna i s at a height of
2 m. with a gain of - 1.0 dBi and a noise temperature of 300 K. Find the maximum
operaLing range if the required error rate is Io- ~, for a nonfadcd and a llnt Rayleigh
faded channel with 2cr 2 = - 10 dB. A ssume a data rate of24.3 kbps, and a receiver
noi se figure of 8 dB.

Solutio11
The equivalent noise temperature of the receive antenna and the recei ver i s

T,y, = TA + ( F - I )To = 300 + (6.3 - 1)(290) = I 837 K

since the receiver noi se figure is F = 8 dB = 6.3 .

Without fading, £1,/11 0 = 9.6 dB = 9. 12 for PSK with P, = 10- 5. So frnm (9. 104)
the required receiver input power is

Eb Eb - 23 3
P, = - (noR1>) = - (kT_,y, R1, )
II () 11 Q
= (9.12)(1.38 x 10 )( 1837)(24.3 x 10)

= 5.6 X 10- t 5 W
324 Chapter 9: Modulation Techniques

Since the propagatio11 channel invo lves multipath scattering. we will use the Unk
formula for ground reflections to include a realistic path loss factor. So from (3.58)
we calculate the maximum range a~
14 2 2 114
R --[P,C 1 C,hih~] l -
- --~~ - [ (50)(0.8)(50) 15(2) ] _
- . x I 04 m.
92
P, 5.6 X JO-

For the faded case, (9.112) gives r =

44 dB for Pe = I 0- 5 . Since 2a2 = - IO dB,
we have Eb/no= 44+ 10 dB= 54 dB= 2.5 1 x 105 . Then 1be receiver input
power is
3
Pr= (2.51 X 105)(1.38 X 10- 23 )(1837.)(24.3 X 10 )
= 1.55 X ] ()-IO W

and the maxinrnm range is now

4
(50)(0.8)(50)2(2)2 ] I/
R= [ =7.1 x 103 m,
J.55 X IQ- 10

which is reduced by a factor of more than IO from the non faded result. 0

9.5 M-ary DIGITAL MODULATION

As we have seen in t.be preceding sections, binary modulation methods transmit one bit
per signaling interval, with a bru1dwidth efficiency of I bps/Hz. Here we consider lhe more
general case of M-ary modulation methods, where more than one bit may be transmitted
per signaling interval. This allows greater bit rates for the same bandwidth, al the expense
of a more complex system.
If we transmit M = 2" symbols for each signaling interval, a bandwidth efficiency of
11 bps/Hz can be achieved. This can be done by using mulliple discrete amplirude level~, or
with multiple phase states, or with a combination of these. lf the symbol rate is Rs, then the
effective bit rate is Rb= nR,.. ln the binary case we have 11 = I, with M = 2 symbols (0
and I). Figure 9.24 shows an example for 11 = 5. Jn this case there are M = 25 = 32 distinct
symbols that can be transmitted, each consisting of n = 5 bits. lf the symbol transmission
rate is Rs, then the bit rate is Rb= 5R,,.. The symbol period is T = 1/ R~ = 5/ R1,. and lhe
bit period is I/ Rb- Note that if an error is made upon receiving a symbol, it is possible that
all the bits in that symbol may be in error.
We will first consider the M = 4 extension of PSK. then discuss M-ary p hase shift
keying and lhe general case of quadrature amplitude modulation.

I+-- - - r = l. - - --+I
Rb

FIGURE 9.24 Example of an M-ary signa l with 11 = 5. There are M = 2 5 = 32 symbols, each
- - - - - - - - - consisting of 5 bits. The sy,nboJ period is 7' = 5/ Rb, and Lhe bit period is T / 5 =
I/Rt,.
 9.5 M-ary Digital Modulation 325

Q Q

0s,

Ca) (b)

FIGURE 9.25 Pha,e ~tatcs for QPSK modulation. (a) Plrnsor representation. (b) Constellation
--------- uiagram.

Quadrature Phase Shift Keying

It is easy to sec that phase shift key ing can be generalized to divide the aUowable phase
states of I.he carrier into more than the two values (0° and I 80 ) used in BPSK. If we use
four slates, where 11 = 2 and M = 4, then we can transmit two bits. or four symbols. for
each signaling interval. This is called quadrature phase shift keying (QPSK).
The four phase states of a QPSK modulated carrier can be written as

so(/)= A COS(Wol + 45 ) (9. II 5a)

s,(t) = Acos(cvot + 135°) (9. I I5b)
s2(f) = A cos(wot - 135 ) (9.1 15c)
s3(t ) = A cos(wor - 45°) (9. 115d)

These can be written more succinctly as

7T
= A cos(wot + ¢, ), ¢, = (2i + 1)
4. = 0, I. 2, 3.
s;(I) for i (9. 11 6)

These four phase slates can be represented in phasor form. as in Figure 9.25a. Assuming
cosine-based phasers, the hori.wnlal axis represents the in-phase (I) component, while the
vertical axis represents the quadrature (Q). component that is shifted 90° in phase. Each of
the four signals of (9. 115) is represented by a separate phaser. with constant magnitude and
a phase of ±45" or± 135°. A related graphical representation is the constellatio11 diagm111,
shown in Figure 9.25b. In this case only t.he endpoint of the phaser is shown. Constellation
diagrams can be useful for visualizing M-ary modulaLion methods with either amplitude
or phase changes. The in-phase and quadrature components of the QPSK signal set can be
more clearly seen by using a tri gonomet.ric identity to expand (9.1 I6):

(9.117)

where A I and A Q are given in Table 9.3. In I.his table we have ignored a common factor
of .J?./2 in the values of A I and A Q · The above results suggest that the QPSK signal can
be generated by using binary NRZ data to modulate the in-phase (co:-ine) and quadrature
(sine) components of the carrier wave. Each QPSK phase state can then be used to represent
two bits of data, as shown in the above table. The incoming serial binary data sLream must
be multiplexed into groups of two bits, which separately modulate the land Q components
of I.he carrier. The QPSK modulator thus takes the form shown in Figure 9.26. A 90 hybrid
divider can be used to provide the I and Q components or the LO from a single oscillator.
The low-pass filters serve to limit 1hc bandwidth of the radiated signal; this filtering is
326 Chapter 9: Modulation Techniques

TAHL!i: 9.3 l and Q Components of a QPSK Signal

</); A, AQ Binary Data

0 45J J' I
135° - I 0,1
2 - 135' - I - I 0,0
3 -45° l - J 1.0

not easily performed after up-conversion due to filter limitalions. Because the average
transition between phase states is 90" , the bandwidth of the QPSK spectrum is narrower
than the spectrum of a B PSK signal. The output of the QPS K modulator is a double sideband
suppressed carrier signal.
FigLLre 9.27 shows a typical input binary data stream and the resultant I and Q data
for the QPSK modulator. Each pl'lir of input bits is coded into distinct I and Q symbols
according to Table 9.3. These symbols are then used to fom1 the QPSK signal as shown in
the figure. If the bit nue of the input data is R11 bils per second. the output symbol rate is
Rs = R1,/2 ~ymbols per second. Thus QPSK requires only half the channel bandwidth of
BPSK to transmit the same data rate, and therefore bas a bandwidU1 efficiency of 2 bps/Hz.
Note that the QPSK output is a constant envelope signal.
Demodulation of QPSK requires coherent detection. A block diagram of a QPSK
demodulator is shown in Figure 9.28. Tbe demodulator uses two mixers with quadrature
LO components to recover the I and Q signals. PCM detectors, each consisting of an
integrator and a sampler. are used to detect the I and Q NRZ data, which is then decoded
with a parallel-to-serial converter to provide serial binary output. Since QPSK is a constant
envelope modulation, the detectors can use a zero threshold.
To analyze the operation of the QPSK demodulator, assume that the QPSK signal s1(1)
of (9.116) is applied to the input. Then the I and Q outputs of the demodulator ,u·e
T
V, =f A 2 cos(wot + </>;) cos w0 t d t
1= 0

(9.118a)

T
VQ = - f =o A 2
cos(wol + ¢ 1) sin o.>0t dt
-A 2
= --
2
!r
1=0
Lsin(2wot + ¢,) - sin¢;l dt
A2T
= --sin¢;
2
(9.J 18b)

Binary Store ~ c o s w0,

daLa (NRZ) two bits
(serial 10
Ru bits/s ~ . R
parallel) ~ S H1Wo/ R., = {' symbols/s

______
F_IG
__UR_E_9_._26 Block diagram ofa QPSK modulator.
 9.5 M-ary Digital Modulation 327

8
~~~;Y -+--l-0~!_1~
! _1~!_1_.!~o__.__o ~o~ 1 ~'-+-
! _1~1_!~ • 1

/-data
I
t
O [T 1 J 0 0 RI
(NRZ) Ot---- - + r 7 - - - + - - - - - t -- - -
- 11------' .___ __ _
• 1
-+---
_,

Q-daia
(NRZ) O
I

-I
t ~o I I I

• I

Data and timing diagrams for the QPSK modulator.

--- ---FIGURE
----9.27

where T is the symbol peiiod, and ¢, is the phase of the ith QPSK sig nal. Table 9.4 lists
the specific outputs for the four possible input symbols. where the constant C is defined as
A 2 T /2../2. Comparison with Tnble 9.3 shows that the correcl binary data is recovered by
the QPSK demodulator.
The assignment of bits pairs to symbols given in Table 9.3 1s not unique, as we could
have chosen to assign the bit pair I , I to St , s2, or s,. instead of s0 . for example. But the
above choice has the very useful property that when an error occurs in the detection of a
symbol. it is most likely that only one of the bits will be in error, rmher than both bits.
This is because an eITor is most likely to result in a shift from the correct phase state to the
immediately adjacent phase, rather than the diametrically opposite phase (see Figure 9.25),
and the assignment of bits has been made so that there is a difference of only one bit
between consecutive QPSK phase states. This is called Gray coding, and is an example of
how judicious coding of the transmitted data can lead LO improved bit error rates.
For example. if the binary input is 0.1 the St symbol is sent. The presence of noise may
cause the demodulator to incorrectly indicate reception of either s0 (for the bit pair I , I). or
s 2 (for the bit pair 0,0). In either case there is only one bit that is in error. It is possible that
an error is severe enough so that the detector indicates SJ, for which both bits would be in
error, but this is much less likely than either of the first Lwo events.

Probability of Error for QPSK

We can derive the probability of error for QPSK using the same method as was used for
BPSK. By symmetry the error rates for all s1 are equal, so let us assume that St is received

,.
1----..i 1()
0
dr' 1----o---r o - - Vr
J(/)

>--- Q-'""' j ' l dr' ~--o---,I

( 0--\IQ
0
I
Sample
at r = T

--- ---FIGURE
---9.28
- Coherent QPSK demodulator.
328 Chapter 9: Modulation Techniques

TABLE 9.4 I and Q Outputs of a QPSK Demodulator

S/ (t) <Pi v, VQ Binary Output

So 45° C C I.I
S1 135° -C C 0.1
Sz - 135' -C -C 0.0
SJ -45° C -C 1.0

by the demodulator a long with additive gaus:,ian noise having a power spectral density of
11 0 /2:

s(I ) +11(1) = s1(1) + 11(/) = A cos(wot + 135°) + n(I). (9. l 19)

Then from (9.11 8) the I and Q outputs from the detector are

-A2T - E,.
V,(T) = ~ +n 0 (T) = r;; +n 0 (T) (9. l20a)
2v2 v2
A 1T E
VQ(T) = r;; + n,,(T) = ;_ + n 11 (T) (9.120b)
2v2 v2

where Es = A? is the symbol energy. The noise wmponent of the outputs is

n,,(T ) = f
1=0
00

An(1)coswol dt. (9. 121)

which is a gaussian random va1iable. We can find the variance of 11 0 (T) as follows:

= A2 f~o l :o E(n(l )n(s)) cos Wol cos w 0s di ds

=- -
A2110
2
f T
1=0 s=O
17'
i5(f - s) cos wol cos wos dt ds

~
2 2
A
= --
no 1T cos-o>o.tds A T no
=- - - = -E8 1.10
- (9 ..122)
2 .1=0 4 2

The detection threshold for both detectors is zero, so the probability of a symbol error at
the I correlator is

(9.123a)
 9.5 M-ary Digital Modulation 329

since from (9.122) ~ _r,

ff: .
= y ~ By symmetry we will also obmin the same expression for
the probability of a symbol (.lffOr at Lhe Q correlator:

p~Q) = ~e,fc(
2 v~
{£:'). (9.123b)

Since the overall probability that a symbol is received correctly is the product of the
probabjlitics that each correlator operates col'l'ectly. the overall probability of c1Tor for a
symbol is

(9.124)

where we made the approximation that P,. « I.

H, as is the usual case in practice. we use QPSK with Gray coding. then we can assume
that a symbol error i~ most likely to cause only a single bit error. Then since each symbol
contain~ two bits. the bit error rate for QPSK will be one-half the symbol error rate:

P/l = ~e,fc( ~ ) -
2 y~ (9. 125)

Note that because the symbol period Tis twice the bit period, E,, = 2E1,, where Ei, is the
bit energy. This shows that the expression of (9.125) is cquivalcm to the probability of error
for BPSK a:-. given in (9.78). Thus with QPSK it is possible to achieve twice the data rate as
for BPSK. with the s,une bandwidth and error rate. This fact has led LO the extensive use of
QPSK modulation (and its several variations) in a wide variety of applications. including
CDMA-PCS telephone systems, the Iridium LEO satellite telephone system. and the direct
broadcast television system (DBS).

EXAMPLE 9.5 COMPARISON OF BPSI< A D QPSK MODULATION

A QPSK system trans mits data at 20 Mbps over a radio link with an average
u·ansmil power of 2 W. The total link loss is I IO dB. and the two-sided power
spectral density of the noise is 110 /2 = 4 x 10- 20 W/Hz. If Gray coding is used,
find the bit error rate. and compare with a similar system using BPSK.

So/111/011
The receive power is P, = 2 x I0- 110110 = 2 x 10- 11 W. For QPSK the symbol
energy-to-noise ratio i~

£, Pr 2P, 2(2 X IQ- 11 ) -,-

tlQ = noR., = 110 Rh = (8 x 10-20 )(20 x 106 ) = _:,_
The probability of a bit error is given by (9. 125)

I
Pr= -e,j'c: ([;;:·,)
- = -e1:fc
I (f!-5)
- = -e1:fc(3.535)
I = 2.87 x 10- 1.
2 2110 2 2 2

For BPSK the bit energy i!,

E,, P, 2 x 10- 11
_ = -- = ---- 2
---- 6
= 12.5,
110 110 R1) (8 X IO 0)(20 x I 0 )
330 Chapter 9: Modulation Techniques

---<:;>-------if-------0,. . _ I

FIGURE 9.29 Constellati.on diagram for 8-PSK.

- -- - - - - - -
and the probability of a bit error is given by (9.78):

Pe = -lerfc 12.5) = - e,:fc(3.535) = 2.87 x io- 7 .

( ~ -I , = -l e,Jc( v r;-;:;--;:l
2 no 2 2

As expected, the bit error rates for QPSK and BPSK are identical. 0

M-ary Phase Shift Keying

We can generalize phase shift keying to M different phase states. By extension of
(9.116), an M-PSK signal can be defined as

s;(!) = il cos(oJol + </>; ). (9. l26)

with ¢; = 2';./, for i = 0, I , 2 . ... M - I. This definition divides the unit circle into M
equally spaced phase states. Thus M = 2 corresponds to BPSK, while M = 4 co1Tesponds
to QPSK. Figure 9.29 shows a constellation diagram for 8-PSK. If M = 2". then each ofU1e
M M-PSK symbols corresponds to 11. bits of binary data. lf Rs is the symbol rate of M-PSK,
then the effective bit rate wiU be Rh = 11 R8• Thus lhe bandwidth efficiency of M-PSK is
11 bps/Hz.
It is straightforward to show that the probability of a symbol error for M-PSK is
approximately given by

o
Pes =et.fc. ( . _:_srn
Er • 2 -rr ) . (9.127)
no M

This result is valid for M > 2 and P, < 10- 3 . lf Gray coding is used for M-PSK, the bit
error rate is given to a close approximation by

(9.128)

Quadrature Amplitude Modulation

We can further generalize M-ary modulation by allowing the amplinides of the 1 and
Q components of the carrier as given in (9.117) to vary arbitrarily. Thus we let

s; (t) = a; cos wot + b; sin wo1. (9. 129)

Tltis is called quadrawre amplilude modulcuion (QAM), because the quadrature components
of the carrier are independently controlled. This representation encompasses ASK and M-
PSK modulations. For example, for M = 4, QPSK modulation will result if we select the
 9.5 M-ary Digital Modulation 331

0 0 0 0

0 0 0 0

0 0 0 0

0 0 0 0

FIGURE 9.30 Consrellation diagram for 16-QAM.

coefficients as

We can therefore say that4-QAM is identical to QPSK, if the signal amplitudes are constant.
Although it is not necessary. the a; and b1 coefficients are usually chosen to form symmetrical
signal sets. The constellation diagram for J.6-QAM is shown i11 Figure 9.30. lt can be shown
that an approximate expression for the symbol error rate of 16-QAM is given by

P8 = -32 erjc· ( ~Ev)

-- .
5no
(9. 130)

Because of the high bandwidth efficiency that can be obtained with QAM, it is increasingly
being used in modem wireless systems, including point-to-point microwave radios, LMDS
systems, and the DY B-C digital video cable broadcasting system. Table 9 .5 summarizes the
ideal performance of several types of coherent digital modulation methods.

Channel Capacity
We have seen in the above sections lhal bit error rates decrease exponentially fast with
an increase in En/ 11 0 . This trend continues with higher-order M-ary modulation methods,
so it is possible to achieve as low an error rate as is desired o nce £ 1,/ no is above a critical
value. Since E,, = S / R1,. this also implies that. for a fi xed signal power S, there is a critical
value of the data rate R1, for which the error rate can be made as small as desired. T his
particular value is called the channel capacity, and is given by a formula derived by Claude
Shannon:

C = B log2 (1 + ~),
n B
(9. 13 1)
0
332 Chapter 9: Modulation Techniques

TABLE 9.5 Summary of Performance of Various Digital Modulation Methods

£1, / 110 (dB) for Bandwidth

Modulation Type P, = 10 5 Efficiency
binary ASK 15.6
binary FSK 12.6
binary PSK 9.6
QPSK (4-QAM) 9.6 2
8-PSK 13.0 3
16-PSK 18.7 4
16-QAM 13.4 4
64-QAM' 17.8 6

where C is the maximum data rate capacity of the chan nel (bps), B is the bandwidth
(Hz), S is the signal power (W), and 110/2 is U1e two-sided power spectral density of the
gaussian noise ( W/Hz). The Shannon channel capacity formula gives the upper bound on
the maximum data rate that can be achieved for a given channel in the presence of additive
gaussian noise. Practical modulation methods usually perform at only a traction of this
value, but the use of error correcling 1.:odes can provide performance close to the Shannon
limit.

REFERENCES
[1 1 F. G. Stremler, Jntroduction to Communication Systems. Addison-Wesley. Reading. MA, 1977.
[2] B. P. Lathi. Modern DigitaJ and Analog Communications Systems. 3rd edition. Oxford University
Press, New York, 1998.
[31 M. Schwartz. Information Transmission, Modulation, and oise, 3rd edit.ion, McGraw-Hill, New
York, 1980.
r4l A. Melu·otra. Cellular Radio Performance Engineering, A1tech House, Dedham. MA, J994.
[51 W. C. Jakes, .Microwave Mobile Commu.nications. lEEE Press. Piscataway, NJ , 1974.

PROBLEMS
9.1 An SSB demodulator has an IF bandwidrh of 30 kHz. and an input noise power spectral densit~
(two-sided) of 10- s W/Hz. If the required output SNR is 25 dB. find the minimum input signal power
and the output noise power. Repeat for a DSB-SC clem<)dulalor.
9.2 The DSB-SC demodulator requires a local oscillator that is synchronized in phase and frequency with
the carrier of the received signal. Consider the effect of an LO phase c1Tor b.rf> by letting the local
oscillator voltage be cos(,t11r,t + 6</>). and deriving an expression for the demodulated signal voltage.
Will this error distort the received signal?
9.3 Repeat Problem 9.2 for a local oscillator having an error 6w in Frequency. Let the local oscillator
voltage be cos(WtF + ti.w)1, and derive an expression for the demodulated signal voltage. Will this
error distort the received signal?
9.4 A broadcast radio station transmits a DSB-LC signnl with a total power of 30 kW, using a modulation
index of 70%. Find the power in the carrier and sidebands of the radiated signal.
 Problems 333

9.5 A square wave is applied to an envelope detector circuit. as shown below. Assuming an ideal diode
characteristic, sketch the oulplIt vol tage waveform.

11,(1)
1.0 +
4 5 6 V;(t)
-+---1,'---+"---l-"--4-'---4':........i"'---+'----1--- r( 111sec)

- 1.0

9.6 Repeal Example 9. 1 for a synchronous DSB-LC demodulator.

9.7 Derive the output voltage for synchronous demodulation of ASK, FSK, and PSK when the locul
oscillat0r has a phase en-or t:,.rj>, and a frequency error t:,.J. That is, lei the local oscillator be represented
as cos[(w + b.1v)/ + b.<p].
9.8 Can the square-law detecLOr described in Exampl e 9.2 be used lo demodulate a DSB-LC amplitude
modulated signal ? Derive a resu lt lo justify your answer.
9.9 An FM radio transmitter uses a can-ier frequency or 90 MHz. with a sinusoidal modulating signal at a
frequency of J,11 = 20 kHz. The voltage amplitude of Lhe carrier is 30 V (peak), the modulation index
is /J = 5. and the Lransminer drives a 50 r.? load. (a) What is the maximum frequency deviation from
the carrier frequency? (b) What is the total power delivered by the transmitter? (c) What fraction of the
total power is generated at the carrier freq11ency? (d) What is the approximate bandwidth according
to Carson's rnle? (e) What fraction of the total power is generated within this bandwidth?
9.10 Consider the special case of an FSK signal generated from a square-wave modulating wavefom1, wilh
f 1 = m / T and h = 11 / T, where T is the bil period. Fine! the Fourier spectrum of the FSK signal,
anJ sketch the form of tJ1e spectrum.
9.11 T he coherem FSK demodulator circuit shown below avoids tbe need for lwo mixers and liltcrs, as
used in the demodulator of Figure 9. 15b. Analyze this circuit to verify that it operates as shown.

9.12 Derive an expression for P,(I), the probability of en-or when a "I" is transmitted, for synchronous
ASK. PSK, and FSK demodulators.
9.1 3 Calculate Lhe E b/ 11 0 ratio. in dB, that is required for P, = IQ- 2 und for P, = I0- s, For binary ASK,
FSK, and PSK modulation using coherent dctccIion.
9.14 Binary data is received by a coherent FSK receiver. The input signal power is 1 1; W, and the two-sided
PSD of the input noise is 10- 14 W/1-17- Find the probability of e1Tor if the bit rate is 5 Mbps. How
docs this result change if noncoherent FS K is used?
9.15 The Mariner 10 satellite used to explore the planet Mercury in 1974 used PSK with P, 0.05 =
(Eb/no= 1.4 dB) to transmit image data back to Earth (a distance of about 1.6 x 108 km). The
satellite transmi tter operated at 2.295 GHz, with an antenna g,tin of 27.6 dB and a carrier power of
16.8 W. The grnund station had an antenna gain of 61.3 dB, and an overall system noise temperature
of J3.5 K.. Compute the maximum possible data rate.
9.16 The probability of error for coherent detection of ASK. FSK. and PS K was derived by assuming that
noise was added after down converting to the [F stage. For the down-converter and PSK demodulator
shown below. show that the same exprc~sion for P, is obtained when noise is added before Jown
conversion (as actually occurs in practice). Let the input signaJ be of the form s(I) 111(1) cos COnl , =
=
where 111(1) ± V, and let the two-sided PSD of the noise be no/2. Note thal the energy of the input
signal is now given as Eb = V 2 T /2, where T is the bit period.
334 Chapter 9: Modulation Techniques

.1"{1) + 11(/)

□
T
1-----..i 1<) dt' 1------<:r'
o T

9.17 Use the large argumem form of e,fc(x) given in Appendix D lo derive expressions for the probability
of error for coherent ASK. FSK, and PSK modulation when E,,/11 0 is very large. Compare your results
with exact values for Eh /110 = 5 dB and 15 dB. Give a.ii estimate for the range of E1,/no for which
the large argument expressions for P, give accurate results.
9.18 For the Rayleigh distribution function of (9.88) show that J;'::,0 f,(r) dr = I, and LhaL Lhe maximum
value of f,.(r) occurs at r = (l.
9. t 9 Use the large argument fom1 for /0 (x) given in Appenclix B to show that the rician pdf approaches u
gaussian distribution when ,. V » a 2.
9.20 The general expression for the probability of error for binary 111odulaLion is given by

P, = Po
1"" fo (x)dx + P1
1•'0
f1 (x)dx.

where Po and P 1 arc the probabilities of' occurrence for a ··o" or" I," respectively. and J o and /1 are
xo
-0,::,

the probability density functions for the occurrence of an error when receiving :1 ··o" or ··1 :· Minimiu
this expression to find the opti111um value of the de.rector threshold. x0 . Show that for the usual case
where Po = P 1 = 0.5, the result reduces to fo(xo) = /1 (x0). Hint: use the fact that
1�( f( )
ilx ,. f y)dy = x .
-;--
iJ
9.21 Consider a PSK receiver using a local oscillator with uL0(1) = cos(wo/ + </>), where</> represen1,
departure from ideal phase coherence with the incoming signal. (a) derive an expression for the
probability of error as a function of the phase error¢. (b) consider¢ as a random variable distributed
unifor111ly from -rc 10 ;r, and derive an e;-.pre�sion for the average probability of en·or. Hint: use the
even and odd properLies of e1fc(x) given in Appendix D.
9.22 Follow the procedure in Section 9.4 to derive expressions for the error rates of coherent ASK and
FSK with Rayleigh fading.
9.23 Evaluate the requited value off for P. = I0-2. 10-s, am! 10- 8 for faded and nonfadecl BPSK.
9.24 An TS-54 cellular phone system uses QPSK (Gray coded) and operates in a propagation environment
witl1 a ground reflection. The base station antenna transmits at 882 MHz with an ElRP of 30 W, with
an antenna that is 40 m high. The receive antenna has a gain of -1.0 dB, and is 1.5 111 high. The receil'e
antenna noise 1emperature is 200 K, and the receiver noise figure is 7 dB. The channel bandwidth
is 30 kHz. and the data rate is 46.6 kbps. If tile required bil error rate is I0-5 , find the 111ax.i01um
operating range of Lhe system.
9.25 Reconsider the interplanetary communication problems of Chapter4 with the use of digital modulation
methods. Let a microwave signal at 2 GHz be transmi11ed to the nem·est star (Alpha Centuri), with a
transmitter power of I kW. Assume large dish antennas with G = 60 dB ror transmit and receive. 3
4 K background noise temperature, and a receiver bandwidth of I kH7.. u· coherent FSK is used with
P, = 10-s is used, what is Lhe maximum data rate?
9.26 Calculate the required Eh/ n0 ratio for M-PSK with P, = tQ-5, for M = 2, 4, 8, I6, 32. and 64.
9.27 A standard wired telephone circuit has a usable frequency r:mge extending from 600 Hz LO 3000 Ju.
with a signal-to-noise ralio of about 30 dB. According to the Shrumon channel capacity expression,
what is the maximum data rate tlml can be achieved with this system?
9.28 Evaluate the Shannon channel capacity for an [S-54 cellular phone system. The channel bandwidth
is 30 kHz. Assume a receive signal power of -60 dBm and 110/2 = I x tQ-18 W/Hz.