WHY IT IS REQUIRED TO DIGITIZE ANALOG SOURCES?

The various advantages that the transmission of digital information has over
analog.

 Digital systems are less sensitive to noise than analog. For long
transmission lengths, the signal may be regenerated effectively error-free at
different points along the path, and the original signal transmitted over the
remaining length.
 With digital systems, it is easier to integrate different services, for
example, video and accompanying soundtrack, into the same transmission
scheme.
 The transmission scheme can be relatively independent of the source.
For example, a digital transmission scheme that transmit voice at 10 kbps
could be used to transmit data at 10 kbps.
 Circuitry for handling digital signals is easier to repeat and digital
circuits are less sensitive to physical effects such as vibration and
temperature.
 Digital signals are simpler to characterize and typically do not have the
same amplitude range and variability as analog signals. This makes the
associated hardware easier to design.

While most all media (e. g., cable, radio, waves, optical fiber) may be used for
either analog and digital transmission, digital techniques, are more efficient
use of those media.

 Various media sharing strategies, known as multiplexing techniques are

more easier implemented with digital transmission strategies.
 There are techniques for removing redundancy from a digital
transmission, so as to minimize the amount of information that has to be
transmitted. These techniques fall under the broad classification of source
coding.
 There are techniques for adding controlled redundancy to a digital
transmission, such that errors that occur during transmission may be
corrected at the receiver without any additional information.
 Digital techniques make it easier to specify complex standards that may
be shared on a worldwide basis. This allows the development of
communication components with many different features (e.g., a cellular
handset) and their interoperation with a different component (e.g., a base
station) produced by different manufacturer.
 Other channel compensations techniques, such as equalization,
especially adaptive versions, as easier to implement with digital
transmission techniques.

It should be emphasized that the majority of these advantages for digital

transmission rely on the availability of low-cost microelectronics.

6.1 SAMPLING THEOREM

A signal g(t) whose spectrum is band-limited to B Hz, that is,

G(f) = 0 for |f| > B

can be reconstructed exactly from its discrete time samples taken uniformly at a
rate of R samples per second. The condition is that R > 2B. In other words, the
minimum sample frequency for perfect signal recovery is fs = 2B Hz.

We consider a signal g(t) shown in figure 6.1a whose spectrum is band-limited

to B Hz shown in figure 6.1b. Spectra are shown as functions of f as well as of
ꞷ. Sampling g(t) at a rate of fs Hz means that we take fs uniform samples per
second.
This uniform sampling can be accomplished by multiplying g(t) by an impulse
train ∂Ts (t) shown in figure 6.1c, consisting of unit impulses repeating periodically
every Ts seconds where Ts = 1/fs. This results in the sampled signal gʹ(t) shown in
figure 6.1d. The sampled signal consists of impulses spaced every Ts seconds (the
sampling interval).

The nth impulse, located at t = n Ts , has a strength g(nTs) which is the value of
g(t) at t=nTs. Thus the relationship between the sampled signal gʹ(t) and the
original analog signal g(t) is

gʹ (t) = g(t) ∂Ts (t) = ∑n g(nTs) ∂ (t - nTs)……………(6.1)

Because the impulse train ∂Ts (t) is a periodic signal of period Ts, it can be
expressed as an exponential Fourier series,

∂Ts (t) = 1/Ts ∑ Ꞷs = 2ℼ/Ts = 2ℼfs ……….(6.2)

Therefore, gʹ(t) = g(t) ∂Ts (t)

= 1/Ts ∑ (6.3)
To find Gʹ(f), the Fourier transform of gʹ(t), we take the Fourier transform of the
summation in Eq. (6.3). Based on frequency-shifting property, the transform ofn the nth
term is shifted by nfs. Therefore,

Gʹ(f) = 1/Ts ∑∞n=-∞ G(f-nfs)………………………. (6.4)

This means that the spectrum Gʹ(f) consists of G(f), scaled by a constant 1/Ts, repeating
periodically with periods fs = 1/Ts Hz as shown in figure 6.1e.

Figure 6.1e clearly shows that this requires

fs > 2B…………. (6.5)

Also the sampling interval Ts = 1/fs. Therefore,

Ts< 1/2B …………(6.6)

Thus as long as the sampling frequency fs is greater than twice signal bandwidth B
(in Hertz), Gʹ(f) will consists of non-overlapping repetitions of G(f). When this is
true, figure 6.1e shows that g(t) can be recovered from its samples gʹ(t) by passing
the sampled signal gʹ(t) through an ideal low pass filter of bandwidth of B Hz.

The minimum sampling rate fs = 2B required to recover g(t) from its samples
gʹ(t) is called the Nyquist rate for g(t), and the corresponding sampling interval
Ts=1/2B is called the Nyquist interval for the low –pass signal g(t).

If the spectrum G(f) has no impulse (or its derivatives) at the highest frequency B,
then the overlap still zero as long as the sampling rate is greater than or equal to the
Nyquist rate, that is

Fs ≥ 2B

An example is a sinusoid g(t) =sin2ℼB(t – t0). This signal is band limited to B Hz,
but all its samples are zero when uniformly taken at the rate of fs =2B (starting at
t=t0) and g(t) cannot be recovered from its Nyquist samples. Thus for sinusoids,
the condition of fs > 2Bmust be satisfied.

6.1.1 Signal Reconstruction from Uniform Samples

The process of reconstructing a continuous time signal g(t) from its samples
is also known as interpolation. In figure 6.1, we used a constructive proof to
show that a signal g(t) band-limited to B Hz can be reconstructed
(interpolated) exactly from its samples. Nyquist rate preserves not only all the
signal information, but also that simply passing the sampled signal through an
ideal low-pass filter of bandwidth B Hz will reconstruct the original message.
From equation (6.3), the sampled signal contains a component (1/Ts)g(t) and to
recover g(t) [or G(f)], the sampled signal gʹ(t) = ∑ g(nTs) ∂(t – nTs)

Must be sent through an ideal low-pass filter of bandwidth B Hz and gain Ts. Such
an ideal filter response has the transfer function

H(f) = Ts ∏(ꞷ/4ℼB) = Ts ∏(f/2B)……………..(6.7)

Ideal Reconstruction

To recover the analog signal from its uniform samples, the ideal
interpolation filter transfer function found in Eq. (6.7) is shown in figure 6.2a.
The impulse response of this filter, the inverse Fourier transform of H(f) is

h(t) = 2BTs sinc (2ℼBt) ……………………..(6.8)

Assuming the use of Nyquist sampling rate, that is 2BTs = 1, then

h(t) = sinc(2ℼBt) ………………….(6.9)

The h(t) is shown in figure 6.2b. Observing that h(t) = 0 at all Nyquist sampling
instants (t = ±n/2B) except t = 0, When the sampled signal gʹ(t)

is applied at the input of this filter, the output is g(t). Each sample in gʹ(t), being an
impulse, generates a sinc pulse of height equal to the strength of the sample, shown
in figure 6.2c. The process is identical to that shown in figure 6.6, except that h(t)
is sinc pulse instead of a rectangular pulse.

The kth sample of the input gʹ(t) is the impulse g(kTs) ∂(t - kTs); the filter output of
this impulse is g(kTs) h(t - kTs).

Hence, the output to gʹ(t), which is g(t), can be expressed as a sum,

g(t) = ∑k g(kTs) h(t - kTs)

= ∑k g(kTs) sinc[2ℼB(t - kTs)]……………. (6.10a)

= ∑k g(kTs) sinc(2ℼBt - kℼ) ……………… (6.10b)

Equation 6.10 is the interpolation formula, which yields values of g(t) between
samples as a weighted sum of all the samples values.
Example 6.1

Find the signal g(t) that is band-limited to B Hz and whose samples are
g(0) = 1 and g(±Ts) = g(±2Ts) = g(±3Ts) = …. = 0; where the sampling interval Ts
is the Nyquist interval for g(t), that is Ts = 1/2B.

Solution: We use the interpolation formula (6.10b) to construct g(t) from its
samples. Since all but one of the Nyquist samples are zero, only one term
(corresponding to k = 0) in the summation on the right-hand side of Eq. (6.10b)
survives. Thus

g(t) = sinc (2ℼBt) …………………(6.11)

This signal is shown in figure 6.3. (same as 6.2b).

Observe that this is the only signal that has a bandwidth B Hz and sample values
g(t) = 1 and g (nTs) = 0 (n ≠ 0). No other signal satisfies these conditions.

6.1.5 Some Applications of the Sampling Theorem

The sampling theorem is very important in signal analysis, processing, and

transmission because it allows us to replace a continuous time signal by a discrete
sequence of numbers. In the field of communication, the transmission of
continuous time message reduce to the transmission of a sequence of numbers,

The continuous time signal g(t) is sampled, and sample values are used to modify
certain parameters of a periodic pulse train. We may vary the amplitudes (figure
6.11b), widths (figure 6.11c), or positions (figure 6.11d) of the pulses in
proportion to the sample values of the signal g(t). Accordingly we have pulse
amplitude modulation (PAM), pulse width modulation (PWM), or pulse
position modulation (PPM). The most important form of pulse modulation today
is pulse code modulation (PCM).

One advantage of using pulse modulation is that it permits the simultaneous

transmission of several signals on a time sharing basis – time division
multiplexing (TDM). Figure 6.12 shows the TDM of two PAM signals.

Another method of transmitting several baseband signals simultaneously is

frequency division multiplexing (FDM). In FDM, various signals are
multiplexed by sharing the channel bandwidth. The information of various
signals is located in nonoverlapping frequency bands of the channel. In a way
TDM and FDM are dual of each other.
6.2 PULSE CODE MODULATION (PCM)

PCM is the most useful and widely used of all the pulse modulations.
Figure 6.13 is shown, PCM basically is a tool for converting an analog signal into a
digital signal (A/D conversion). An analog signal is characterized by an
amplitude that can take on any value over a continuous range of infinite number
of values. On the other hand, digital signal amplitude can take on only finite
number of values. An analog signal can be converted into a digital signal by
means of sampling and quantizing, that is, rounding off its value to one of the
closest permissible numbers (or quantized levels), as shown in figure 6.14.

The amplitude of the analog signal m(t) lie in the range (-mp, mp), which
is partitioned into L subintervals, each of magnitude ∆v = 2mp/L. Figure 6.14
shows for L = 16. Thus the signal is digitized with quantized samples taking on
any one of the L values. Such a signal is known as an L-ary digital signal.

We can convert an L-ary signal into a binary signal by using pulse coding.
Such a coding for the case of L = 16. This code, formed by binary
representation of 16 decimal digits from 0 to 15, is known as the natural binary
code (NBC). The analog signal m(t) is now converted to a binary digital signal.
 The audio signal bandwidth is about 15 kHz. However, for speech,
subjective tests show that signal articulation (intelligibility) is not affected if all the
components above 3400 Hz are suppressed. Since the objective in telephone
communication is intelligibility rather than high fidelity, the components above
3400 Hz are eliminated by a low pass filter. The resulting signal is then the
sampled at the rate of 8000 samples per second (8 kHz). This rate is
intentionally kept higher than the Nyquist sampling rate of 6.8 kHz so that
realizable filters can be applied for signal reconstruction. Each sample is
finally quantized into 256 levels (L = 256), which requires a group of eight binary
pulses to encode each sample (28 = 256). Thus a telephone signal requires 8x8000
= 64,000 binary pulses per second.

The compact disc (CD) is a more recent application of PCM. This is a

high-fidelity situation requiring the audio signal bandwidth to be 20 kHz.
Although the Nyquist sampling rate is only 40 kHz, the actual sampling rate
of 44.1 kHz is used. The signal is quantized into a rather larger number (L =
65,536) of quantization levels, each of which is represented by 16 bits to reduce
the quantization error. The binary-coded samples (1.4 million bit/s) are then
recorded on the compact disc.

6.2.2 Quantizing
For quantization, we limit the amplitude of the message signal m(t) to
the range (-mp, mp), as shown in figure 16.4. Note that mp is not necessarily the
peak amplitude of m(t). The amplitude of m(t) beyond ±mp are simply chopped off.
The amplitude range (-mp, mp) is divided into L uniformly spaced intervals,
each of width ∆v = 2mp/L. The quantized samples are coded and transmitted as
binary pulses. At the receiver some pulses may be detected incorrectly. Hence,
there are two sources of error in these scheme: quantization error and pulse
detection error.
In almost all practical schemes, the pulse detection error is quite small compared to
the quantization error and can be ignored. In the present analysis, the error in the
received signal is caused exclusively by quantization.
If m(kTs) is the kth sample of the signal m(t), and if mˆ (kTs) is the corresponding
quantized sample, then the interpolation formula in Eqn. (6.10),

m(t) = ∑k m(kTs) sinc (2ℼBt - kℼ)

and mˆ(t) = ∑k mˆ(kTs) sinc (2ℼBt - kℼ) where mˆ(t) is the signal reconstructed
from quantized samples. The distortion component q(t) in the reconstructed signal
kℼℼi8s q(t) = mˆ(t) - m(t) . Thus ,

q(t) = ∑k [mˆ(kTs) - m(kTs)] sinc (2ℼBt - kℼ)

= ∑k q(kTs) sinc(2ℼBt - kℼ)

where q(kTs) is the quantization error in the kth sample. The signal q(t) is the
undesired signal and, hence, acts as noise, known as quantization noise.

To calculate the power, or the mean square value of q(t), we have

q2(t)= lim 1/T∫T/2-T/2 q2(t)dt

= lim 1/T∫T/2-T/2[ ∑ ………..(6.29a)

We know that the signals sinc (2ℼBt - mt ) and sinc (2ℼBt - nt) are orthogonal,
that is ,

-∞∞ sinc (2ℼBt - mℼ) sinc (2ℼBt - nℼ)dt = 01/2B m =n

m≠n
……(6.29b)

q2(t)= limT ∞ 1/T∫T/2-T/2[ ∑

= limT ∞ 1/T∑

From orthogonality relationship

q2(t)= limT ∞ 1/2BT[ ∑ ………………… (6.30)

Because the sampling rate is 2B, the total number of samples over the
averaging interval T is 2BT. Hence, the RHS of eq. 6.30 represent the average or
the mean of the square of the quantization error. The quantum levels are separated
by ∆v = 2mp/L. The quantization error lies in the range (-∆v/2, ∆v/2), where ∆v
= 2mp/L. Assuming that the error is equally likely to lie anywhere in the range
(-∆v/2, ∆v/2), the mean quantization error q2 is given by

q2(t) = 1/∆v ∫∆v/2-∆v/2[ ∑

= (∆v)2/12 ………….(6.32)

= mp2/3L2 …………………(6.33)

The power of the quantization noise, Nq = q2(t) = mp2/3L2

Assuming that pulse detection error is negligible, the reconstructed signal mˆ(t) at
the receiver output is

mˆ(t)= m(t) + q(t)

The desired signal at the output is m(t), and the quantization noise is q(t).

Since the power of the message signal m(t) is m2(t), then

S0 = m2(t)

N0 = Nq = m2p / 3L2

And S0/ N0 = 3L2 m2(t) /m2p …………………..(6.34)

This means S0/N0, the SNR, is a linear function of the message signal power
m2(t) shown in figure 8.18 with  = 0.
6.2.3 Principal of Progressive Taxation: Nonuniform Quantization

The SNR is an indication of the quality of the received signal.

Unfortunately, the SNR is directly proportional to the signal power m2(t), which
varies from speaker to speaker. The signal power can also vary because of different
lengths of connecting circuits. The quantization noise Nq = (∆v)2 /12 is directly
proportional to the square of the step size . The problem can be solved by using
small steps for smaller amplitudes (nonuniform quantizing), as shown in figure
6.15a. The same result is obtained by first compressing signal samples and
then using a uniform quantization. The input-output characteristics of a
compressor are shown in figure 6.15b. The horizontal axis is the normalized input
signal. The vertical axis is the output signal y. The compressor maps input signal
increments ∆m into a larger increments ∆y for small input signals, and vice versa
for large input signals. The quantization noise is lower for small input signal
power.
The -law for positive amplitudes is given by

y = 1/ln(1+) ln(1 + m/mp ) 0 m/mp1 …………..(6.35a)

The A-law for positive amplitude is

y=  A/ 1+lnA(m/mp) 0 m/m
p 1

1/1+lnA(1+ ln Am/mp) i/Am/mp 1 (6.35b)

These characteristics are shown in figure 6.16.

The compression parameter  ((or A) dtermines the degree of compression.

To obtain a nearly constant S0/N0 over a dynamic range of for input signal power
40 dB,  should be greater than 100.
The compressed samples must be restored to their original values at the receiver by
using an expander with a characteristic complementary to that of compressor. The
compressor and expander together is called compandor. Figure 6.17 describes
the use of compressor and expander along with a uniform quantizer to
achieve nonuniform quantization.

Generally speaking, time compression of a signal increases its bandwidth.

But in PCM, we are compressing not the signal m(t) in time but its sample values.

When a -law compandor is used, the output SNR is

S0/N0 = 3L2 / [ln( 1 + )]2 2  mp2/ m2(t)

The output SNR for the cases of  = 255 and  = 0 (uniform quantization) as a
function of m2(t) (the message signal power) is shown in figure 6.18.

6.2.4 Transmission Bandwidth and the Output SNR

For a binary PCM, we assign a distinct group of n binary digits (bits) to each
of the L quantization levels. Because n- binary digits can be arranged in 2n
distinct patterns

L = 2n or n= log2L

Each quantized sample is, thus, encoded into n bits. Because a signal m(t)
band limited to B Hz requires a minimum of 2B samples per second, we require a
total of 2nB bit/s.We require a minimum channel of band width BT Hz, given
by

BT = nB Hz

This is the theoretical minimum transmission bandwidth required to transmit

the PCM signall.

Examples: 6.2 and 6.3 to be solved.

Exponential Increase of the output SNR

 We know, L2 = 22n and the output SNR in equation 6.34 or eq. 6.36 can be
expressed as S0/N0 = c (2)2n ……………….(6.39)

Where, c= 3m2(t)/ mp2 [uncompressed case, in Eq. 6.34)

= 3/[ln(1 + )]2 [compressed case in Eq. (6.36)]

Substituting of Eq. (6.38) into Eq. (6.39) yields

S0 / N0 = c(2)2BT/B …………………… (6.40)

We observe that the SNR increases exponentially with the transmission band width
BT..

(S0 / N0)dB = 10 log10(S0 / N0)

= 10log10[c(2)2n]

= 10log10c + 2nlog10 2

= ( + 6n) dB…………………… (6.41)

Where  = 10log10c. This shows that increasing n by 1 (increasing one bit in the
codeword) quadruples the output SNR (a 6 dB increase). Thus if we increase n
from 8 to 9, the SNR quadruples, but the transmission bandwidth increases only
from 32kHz to 36 kHz(an increase of only 12.5%). In this respect, PCM is
strikingly superior to FM or PM.

Example: 6.2 A signal m(t) band-limited to 3kHz is sampled at a rate 331/3 %

higher than the Nyquist rate. The maximum acceptable error in the sample
amplitude (the maximum quantization error) is 0.5% of the peak amplitude mp.
The quantized samples are binary coded. Find the minimum bandwidth of a
channel required to transmit the encoded signal. If 24 such signals are time-
division-multiplexed, determine the minimum transmission bandwidth required
to transmit the multiplexed system.

Solution: The Nyquist sampling rate RN= 2x3000 =6,000 Hz (samples per
second). The actual sampling rate is RA = 6000x (11/3) =8000 Hz.

The quantization step is ∆v, and the maximum quantization error is ±∆v/2.
Therefore from eqn. we get, ∆v/2 = mp/L = 0.5/100 mp  L = 200

For binary coding, L must be a power of 2. Hence the next higher value of L that is
a power of 2 is L=256.

From Eqn. we need n=log2 256 = 8 bits per sample. We require to transmit a total
of C = 8 x 8000 = 64,000bit/sec. Because we can transmit 2 bit/s per hertz of
bandwidth, we require a minimum transmission bandwidth BT = C/2 = 32kHz. The
multiplexed signal has a total of CM = 24 x 64000 = 1.536Mbit/s, which requires a
minimum of 1.536/2 = 0.768 MHz of the transmission bandwidth.

Example 6.3 A signal m(t) of bandwidth B = 4kHz is transmitted using a binary

companded PCM with  = 100. Compare the case of L = 64 with the case of L=256
from the point of view of transmission bandwidth and output SNR.

Solution: For L =64, n=6, and the transmission bandwidth is nB = 24 Hz,

S0/N0 = ( + 36) dB

 = 10log3/[ln (101)]

Hence, S0/N0 = 27.49dB

Again, for L = 256, n=8, and the transmission bandwidth is 32 kHz,

S0/N0 = ( + 6n)n = 39.49 dB

The difference between the two SNRs is 12 dB, which is the ratio of 16. Thus, the
SNR for L=256 is 16 times the SNR for L = 64. The former requires just about 33%
more bandwidth compare to the later.