c

c
c
c
c
c
c
c
c
c
c
c

cc

c
c
c
c
c
c
c
c
c
c
c
c
c
c
U. Uday Kumar
D. Sravan Paul
SK.Moinuddin
c
c
c
c



c
c
Pulse code modulation (PCM) is a common method of digitizing or quantizing an analog
waveform. For any analog-to-digital conversion process, the quantization step produces an
estimate of the waveform sample using a digital codeword. This digital estimate inherently
contains some level of error due to the finite number of bits available. In practical terms, there is
always tradeoff between the amount of error and the size of the digital data samples. The goal in
any system design is quantizing the data in smallest number of bits that results in a tolerable
level of error. In the case of speech coding, linear quantization with 13 bits sampled at 8 KHz is
the minimum required to accurately produce a digital representation of the full range of speech
signals. For many transmission systems, wired or wireless, 13 bits sampled at 8 KHz is an
expensive proposition as far as bandwidth is concerned. To address this constraint, a companding
system is often employed. Companding is simply a system in which information is first
compressed, transmitted through a bandwidth limited channel, and expanded at the receiving
end. It is frequently used to reduce the bandwidth requirements for transmitting telephone quality
speech, by reducing the 13-bit code words to 8-bit code words. Two international standards for
encoding signal data to 8-bit codes are A-law and µ-law. A-law is the accepted European
standard, while µ-law is the accepted standard in the United States and Japan.

cc

cc
For digital transmission, the analog signal is converted to a digital signal. Consider a
telephone system where human speech is converted to digital signal. In practice the converted
digital signal has a fixed precision. Pulse Code Modulation is a commonly used technique to
convert analog signals into equivalent digital sequences.
With reference to the quantization technique used, Pulse Code Modulation can be classified as
ac Uniform
ac àon Uniform

c c
In uniform quantization the value of the quantization interval is fixed. The entire range of
the analog signal is divided into a pre-determined value of fixed intervals. This can be a
disadvantage if the input signal has small signal excursion. To overcome this difficulty, non-
uniform quantization and companding have been proposed.

c ccc
In non uniform PCM the value of the quantization interval varies from one level
to another. àon-uniform quantization may be achieved by first passing the message through a
compressor, a nonlinear device which compresses the peak amplitudes. This is followed by a
uniform quantizer, such that uniform zones at the output correspond to non-uniform zones at the
input. At the receiving end, the compressed signal is passed through an expander, another
nonlinear device used to cancel the nonlinear effect of the compressor. The combined process is
known as companding.

cccc c

Consider a speech signal. Speech signal is composed of relatively fewer voiced
phonemes than unvoiced phonemes. Unfortunately, the uniform quantizer, which has equally
spaced zones, provides unneeded quality for large signals which are least likely to occur, and
pronounced truncation effects for the more frequent small amplitude signals. As a result, uniform
quantization does not perform as well as a quantizer with wider zones at high amplitudes and
narrower zones at lower amplitudes. Instead of employing uniform quantization, a natural non-
uniform substitute is observed in the human auditory system. It is believed that the human
auditory system is a logarithmic process in which high amplitude sound does not require the
same resolution as low amplitude sound [Give reference]. Conversion to a logarithmic scale
allows quantization intervals to increase with amplitude, and it ensures that low-amplitude
signals are digitized with a minimal loss of fidelity. Fewer bits per sample are necessary to
provide a specified signal-to-noise ratio (Sà ) for small signals and an adequate dynamic range
for large signals.
c

!!ccc
In addition to reducing quantization error, companding decreases the required bandwidth
of the system. That is, systems solely employing uniform quantization require 13-bit code words
for equivalent performance requirements of the telephone system. However, while increasing
performance, systems using nonlinear companding may reduce the required codeword length to
8-bits or less. Companding is simply a system in which information is compressed, passed
through a channel and then expanded on the other side. Companding may be accomplished in
hardware via a CODEC, or in software using a look-up table approach or a real time direct
calculation. However, if hardware companding is implemented and intermediate processing of
the signal is necessary, and then reverse companding is required. Two international companding
standards that retain (save) up to 5 bits of precision by encoding signal data into 8 bits are µ-law
and A-law.
"#$ccc c!!c
 c
PCM encoding is composed of three (four if non uniform) successive steps:
ac Sampling
ac Duantizing
ac Compression (if non uniform PCM)
ac Coding
"c
It is the determination of a signal¶s amplitude at regular time intervals. Since the
telephone network has a bandwidth of 4 kHz [Give reference], for accurate reproduction, a voice
signal must be sampled at a rate of at least 8 kHz, according to àyquist¶s theorem. That is the
amplitude of the signal is sampled every 125µs.
c
 c
Once the signal¶s amplitude is obtained, it is quantized into a discrete set of amplitude
levels for representation as a digital signal. Duantization is achieved by dividing the amplitude of
the signal into quantization intervals, also known as bins. All signal amplitudes falling within a
bin are represented by the midpoint of that quantization interval. The quantization process
introduces quantization error into the digital signal; however, the introduced error may be
minimized by minimizing the width of the bins with respect to the number of bits needed to
uniquely identify the quantization bins.
There are many ways to represent the quantization output. Theoretically we take the
centroid of a quantization interval which is calculated based on the probability distributions. One
simple method is simply to take the midpoint of the quantization interval because of the fact that
the quantization interval is too small that midpoint and the centroid will coincide in most of the
cases.
cc
Commonly used Companding standards are:cc
ac µ law
ac A law
%cc
The U.S. and Japan use µ-law companding. Limiting sample values to 13 magnitude bits, the D -
law compression portion of this standard is defined mathematically by the continuous equation:

where µ is the compression parameter (µ =255 for the U.S. and Japan), and x is the normalized
integer to be compressed. A piece-wise linear approximation to this compression equation is
illustrated in Figure. [Give figure number and reference].

.
 During compression, the least significant bits of large amplitude values are discarded.
The number of insignificant bits deleted is encoded into a special field of the compressed code
format, called the chord. Each chord of the piece-wise linear approximation is divided into
equally sized quantization intervals called steps. The step size between adjacent codeword¶s is
doubled in each succeeding chord. Also encoded is the sign of the original integer. The polarity
bit is set to 1 for positive integer values. Thus, an 8 bit Mu-255 codeword is composed of 1
polarity bit concatenated with a 3-bit chord concatenated with a 4-bit step. Before chord
determination, the sign of the original integer is removed and a bias of 33 is added to the
absolute value of the integer. Due to the bias, the magnitude of the largest valid sample is
reduced to 8159 and the minimum step size is reduced to 2/8159. The added bias enables the
endpoints of each chord to become powers of two, which in turn simplifies the determination of
the chord and step. Chord determination may be reduced to finding the most significant 1 bit of
the binary representation of the biased integer value, while the step equals the four bits following
the most significant 1

%cc
c A-law is the recommended companding standard used across Europe. Limiting sample
values to 12 magnitude bits, the compression portion of this standard is defined in the
continuous.

where A is the compression parameter (A=87.6 in Europe), and x is the normalized integer to be
compressed. A piece-wise linear approximation to this compression equation is illustrated in
Figure.
 A-law companding has the same basic features and implementation advantages as µ -law
companding. A-law companding is approximated by linear segments, with the first chord defined
to be exactly linear. A zero-level output for the first quantization interval is not defined.
"%c "&c!c%&c "%c
The µ-law algorithm provides a slightly larger dynamic range than the A-law at the cost
of worse proportional distortion for small signals. By convention, A-law is used for an
international connection if at least one country uses it.c

cc
Once the quantization is performed we have to code the output of the quantizer. The
coder output is a binary sequence which is going to be transmitted over the channel.
c
c
#cc c
Signal to quantization noise ratio is the measure of the performance of the pulse code
modulated system.

c
In case of PCM noise is due to the quantization error due to the quantizing process. So we can
calculate noise power as:
àoise power = (quantization error)2
= (signal value-quantized value)2
where is quantized value.
c
!cc!c
The principal advantage of Pulse Code Modulation (PCM) is the noise immunity. But it is
not the only technique used in practice. Other pulse modulation systems are also employed in
spite of the highly superior performance of PCM. The reasons are
ac PCM needs very complex encoding and quantizing circuitry
ac PCM requires larger bandwidth as compared to analog systems.

In spite of these PCM is fast gaining popularity and is being used increasingly.
The reasons are very simple.
ac Multiplexing equipment is much cheaper.
ac Further distance between repeaters is large because PCM tolerates much worse signal to
noise ratios.
ac Advent of very large scale integration (VLSI) has reduced the cost of complex circuits
needed in PCM.
ac Increased bandwidth requirements by PCM the problems is no longer a serious one
because of the advent of large band width fiber optic systems.
"#!

In telephony PCM is used to digitize the Analog Voice (or speech). Digitized voice channel
data rate is 64 kbps. (T1 System)

In Space communication, space craft transmits signals to earth. Here, the transmitted power
is very low i.e. 10 to 15 w and the distances are huge (a few million km).Still due to the high
noise immunity, only PCM Systems can be used in such applications. Way back in 1965 PCM
was used by Mariner space craft to transmit back pictures of Mars. Of course each picture took
several minutes of transmission.

Variants of PCM are differential PCM and delta modulation. Differential modulation is a
PCM with the modification that each word in this system indicates the difference in amplitude,
positive or negative, between this sample and the previous one. Thus this system indicates the
relative rather than the absolute value of each sample. When this system is applied to speech
samples, the high value of correlation between adjacent samples ensures that the amplitude of the
current sample is related to the previous one and large variations from one sample to the next are
unlikely. As a result fewer bits are needed to indicate the size of the magnitude change relative to
the case of absolute magnitude. Smaller bandwidth is therefore needed for transmission.