260 Principles of Communication Systems

x() 1 ms

1 2 4 5 1 2
nl.
Fig. 5.20 A PAM-TDM systern for Example 4.3.
allocated tor each =1000/10 = 100 us.
It same guard time is to be maintained we re
qute narrower pulses of width = 100 50 =
W(0)
LOw () dt
Exanple S.4 Show that, PWM denodulation can 1JW()

be achiev ed by simple time averaging of PWM pulses

by an averaging low pass filter. W()
0dt =
Solution lfx()is the modulated signal, the width, W()
of which in each cycle of time period T is proportional
to iessage n(), then time averaged output of x() can Thus, x ) is directly proportional to W(i) and hea:
be written as, modulating signal m(1).

Self-test Questions
3. Instantaneous sampling is not prcferred as the sample energy becomes too low which maae
it susceptible to noise. Is that correct?
6 Crosstalk can be suppressed but not avoided with samples of finite duration. Is that true?
7 Which of these two approximates equalizer transfer function, x/sinr or sinx/x?
Is PPM signal related to differentiation of PWM or the reverse?

5.4 DIGITAL REPRESENTATION OF ANALOG SIGNAL

We return to our journey towards complete digital representation of analog signal. We hake
already gone through an important step of sampling the time axis, i.c. discrete-time represen:atit (uPw()w
of analog signal. Now we need to discretize the amplitude axis and then represent that sig
terms of binary digits to complete the process.
5.4.1 Quantization of Signals
The limitation of thesystem we have been describing for communicating over long channels is a
once noise has been introduced at any place along the channel, we are "stuck" with it. We au
describe how the situation is modified by subjecting a signal to the operation of quantizatil
When quantizing asignal m(), we create a new signal m,(!) which is an approxinmation to m
However, the quantized signal m() has the great merit that it is, in large measure, separable frum
the additive noise.
The operation of quantization is represented in Fig. 5.21. Here we contemplate a signal
whose cxcursion is contined to the range from V; to V We have divided this total range into MM
 Buanitation ivtenvals Coded
Jinag term, e
inteas will be Pwen 2
M= 2
it Code then ae 2S6
intevads enti?irg
261
Pulse Modulation and Digital Transmission of Analog Signal

6 l4

quantization.
() dt

of
nd he:0
operation

The
o h make 5.21
h :true?
r
Fig.

m)
m(),
le have
2ntatia
ignal

Lo1 VL
67 Las
s is that V
Ve noW
1zation.
to m(i).
za
le from
n

ial mc
into
 262
Principles of Communication Systems

Cytl niervals cach of size S. Accordingly S, called the step size, is S - (l /M. In Fig. 5..
e Show he specific example in which M = 8. In the center of cach of these steps we loca i
which the input is
quantization levels mo l, m. The quantized signal mn,() is generated in the following wz y: S/2, the noise will
Whenever m) is in the range , the supal » ) uintains the constant level m whenever m indicated in the fg
S in the range A,. m,) maintains the constant evel ,; and so on. Thus 0theTr0Msignal
limes be found al onc of the levels , m,, .... m-. The transition n ,0
n,4) will at ad noise is such that e
M, to m probability that fr
IS nmade abruply whenm) passes the transition level which is midway bctween m, ad possible to suppre:
mË and sO On. To state the matter in an alternative fashion, we say that, at every instant of tim a We have showr
m,A has the valueotf the quantization level to whichm) is closcst. Thusthe signal m) does t. * be significantly re
change au all with time or it makes a "'quantum" jump of step size S. Note the disposition of te suffered by m,().
quantizalion levels in the range from V, to These levcls are cach separated byan amount S, ta probability P, of :
the separalion of the cxtremes I; and V, cach from its nearest quantization level is only S/2. Alsa, incrcasing S resul
al eeV instant of time, the quantization error m() m() has a magnitude which is cqual to signal m,(). This
less than S2. Hencc, the receiv
e see, therefore, that the quantized signal is an approximation to the original signal. Tre between them is
quality ot the approximation mnay be improved by reducing the size of the steps,thercby increas1 g discussed further
the number of allowable levels. Evenually, with small enough steps, the human car or theeye wil
not be able to distinguish the original from the quantized signal. To give the rcader an idca of te 5.4.2 Quant
numter ef quantization levels required in apractical system, we note that 256 levelscan be used io It has been pointe
obtain the quality of commercial color TV, while 64 levels gives only fairly good color I differ from one ar
pertormance. These resuits are also found to be valid when quantizing voice. to the quantizatic
Now let us consider that our quantized signal has arrived at a repcater somewhat attenuated arn d quantization error
corrupted by noise. This time our repcater consists of a quantizer and an amplifier. There is noise Let us divide t
superimposcd on the quantized levels of my(). But supposc that we have placed the repeater at H each of magnituc
point on the conmunications channel wherethe instantancous noise voltage is almost always le:s m, m, ..., My a:
than half tihe separation between quantized levels. Then the output of the quantizer will consist ct message signal m
asuccession of levcls duplicating the original quantized signal and with the noise removed. In rare
instances the noise results in an error in quantization level. A noisy quantized signal is shown i3 quantizer output
Fig. 5.22a. The allowable quantizer output levels are indicated by the dashed lines separated by Let f(m) dm b
amount S. The output of the quantizer is shown in Fig. 5.22b. The quantizer output is the level ta the mean-square

S
Error in level
2
Large noise excursion Now, ordinarily
be constant. Hov
small in coipari
rcasonable to ma
the first term of
may now remov

m, Eq. (5.24)b.
(b
(a

Pig. 5.22 (a) A quantized signal with added noise. (b) The signal after requantization.
large that an error
One instance is recorded in which the noise level is so
results.
 Pulse Modulation ond Digital Transmission of Analog Signal 263

which the input is closest. Therefore, as lonp as the noise has an instantancous amplitudc less than
the noise will not appear at theoutput. One instance in which the noise does exceed S/2 is
indicated in the figure, and, corespondingly. an error in level does occur. The statistical naturc 01
noise is such that even if the average noise magnitude is much less than S/2, there is always a finite
probability that from time to time, the noise maonitude will exceed S/2. Note that it 1s nevcr
possible to suppress completely level errors such as the one indicated in Fig.
We have shown that through the method of signal quantization., the effect of.22.additive noise can
be significantly reduced. By decreasing the spacing of the
suffered by m(i). This effectively decreases the relative noiserepeaters, we decrease the attenuation
power and hence decreases the
probability P, of an error in level. P,can also be reduced by increasing
increasing S results in an increased discrepancy between the true signalthen(t) step size S. However.
signal m,(). This difference m(t) m(t) can be regarded as and the quantizcd
Hence, the received signal is not a perfect replica of the noise and is called quantizution noIse.
between them is due to errors caused by additive noise and transmitted signal m(t). The difference
discussed further in Chap. 12. auantization noise. These noises are

5.4.2 Quantization Error

It has been pointed out that the quantized
signal and the original signal from which it was derived
differ from one another in a random manner. This
to the quantization process and is called
difference or error may be viewed as anoise due
quantization error. We now calculate the mean-square
quantization error e,wherc e is the difference between the original and quantized signal voltages.
Let us divide total peak-to-peak range of the message signal m(t) into M equal
eact of magnitude S volts. At the center of each voltage interval we locate a voltage intervals,
m, m,, ..., my as
quantization level
shown in Fig. 5.23a. The dashed level represents the instantaneous value of the
message signal m(t) at a time t. Since, in this figure, m(t) happens to be closest to the level n, the
quantizer output will be m, the voltage corresponding to that level. The error is e = m(t) - m7
Let f(m) dm be the probability that m(i) lies in the voltage range m - dm/2 to m + dm/2. Then
the mean-square quantization error is
m+S/2

m-S/2 f(m)(m - m)' dm+ f(m)(m - m,) dm + (5.24)

Now, ordinarily the probability density function f(m) of the message signal m() will certainiy not
be constant. However. suppose that the number Mof quantization is large, so that the step size S is
Small in comparison with the peak-to-peak range of the message signal. n this caSe, l Is ce y
reasonable to make the approximation that f(m) is constant within each quantization tang
the Tirst term of Eq. (5.24) we set f(m) =), a constant, In the second ternm f(m) =/" e. We
may now removef .f , etc., from inside the integral sien, If we make the substitution r=m
m, Eq. (5.24) becomes
S/2
(5.25a)
S/2

(Z.2Sb)
12
 264 Principles of Communication Systems

MM

Mk+2 Error, e
ts
Peak-tO-peak mk +1
range of
signal = MS
ts_f
m

(b)
m(t)

(a) 2

Fg. 5.23 (a) A range of voltage over which a signal

m{t) makes excursions is divided
1nto M quantization ranges each of size S. The
quantization levels are
located at the center of the range. (b) The error voltage
the instantaneous value of the signal e(t) as a function of
m(t).
Nowf'S is the probability that the
1s the probability that n is in the signal voltage m(t) will be in the first quantization range, f'S
second
parentheses in Eq. (5.25b) has a total valuequantization
of
range, etc. Hence the sum of terms in the
1s unity. Therefore, the mean-square quantization error

12 (5.26)
5.4.3 Pulse Code Modulation (PCM)
A signal which is to be
tion is used to reduce thequantized
effects
prior to transmission is usually sampled as
of well. The quantiza
number of nessages if we choose tonoise, and the sampling allows us to
do so.
generate a quantized PAM waveform, that is,lhe combined operations of time-division
sampling and
multiplex a
quantizing
a train of pulses whose
number of discrete magnitudes. amplitudes are restricted to a
We may, if we choose, transmit
these
represent each quantized level by a codequantized
number
sample values directly. Alternatively we may
and
sample value itselt. The merit of doing so will be transmit the code number rather than the
frequently the code number is converted, betore developed in the subsequent discussion. Most
transmission, into its representation in binary