Quantization

Prepared by
Dr. P. Nirmala
SENSE

1/29
 Objectives
Quantization-Introduction

Uniform quantization

Quantization error (noise)

analysis

Non uniform quantization

Analog Companding

Digital Companding
2/29
 Quantization
• Quantizing: Conversion of an analog sample of the input
into a digital form
• Human ear/eye – finite intensity differences
• Process:
• Input sample value is divided into a finite set of decision
levels (thresholds)
• Output is assigned a discrete value from finite set representation levels
• Step size: separation between decision thresholds/ representation
levels.

3/29
 Scalar Quantization
•Memory less
•Instantaneous

Mid tread:
•Decision thresholds: ±∆/2, ±3∆/2, ±5∆/2…
•Representation levels: 0, ±∆, ±2∆, ±3∆…

Mid rise
•Decision thresholds: 0, ±∆, ±2∆, ±3∆…
•Representation levels: ±∆/2, ±3∆/2, ±5∆/2…
•Quantization error: ±∆/2

•∆-step size (2Vmax/L)

•Overload=(Vmax-Vmin)/2
5/29
 Quantizer types

a) Mid tread b) Mid rise

6/29
 Quantization Noise Level
Let thequantization error be denoted by
therandom variable Q of sample value q
qm
Q  M  V, (E[M]  0)
Assuming a uniform quantizer of the midrise type

2m
the step - size is   L
max

 m m  m , L : total number of
max  max

levels 
 1   q
f Q (q)   , 2
2 
  0,
otherwise

1 
2
 Q  E[Q ]   2 q f Q (q)dq
2 2 2
 2
 q 2
dq

2
  2

12
7/29
 Quantization Noise Level
When the quantized sample is expressed in binary
form,
L  2R
where R is the number of bits per sample
R  log 2 L
2m max

R
1 2
 Q2  mmax
2
2 2
3 R
Let P denote the average power of m(t)

 (SNR)o  P2
Q

3P
( )2 2R

m 2max
(SNR) o increases exponentia lly with increasing R
(bandwidth
SNR  1.8 ). 6dB per bit
6R
8/29
 Robust Quantization(Non Uniform)

Motivation
• Speech signals have the
characteristic that small-
amplitude samples occur more
frequently than large-amplitude
ones
• Human auditory system exhibits a
logarithmic sensitivity
• More sensitive at small-
amplitude range (e.g., 0 might
sound different from 0.1)
• Less sensitive at large-amplitude histogram of typical speech signals
range (e.g., 0.7 might not sound
different much from 0.8)

9/29
Non Uniform Quantization- Companding

Compressor Expander

10/29
 -law algorithm

• It has continuous compressor characteristics (low-linear, high-

logarithmic)
• The -law algorithm (μ-law) is a companding algorithm,
primarily used in the digital telecommunication systems of
North America and Japan.
• Its purpose is to reduce the dynamic range of an audio signal.
• In the analog domain, this can increase the signal to noise ratio
achieved during transmission, and in the digital domain, it can
reduce the quantization error (hence increasing signal to
quantization noise ratio).
•  = 255

11/29
 -law algorithm

log(1 
vm)
log(1
)

12/29
 A-law algorithm
• It has piecewise compressor characteristics (low-linear
segment, high-logarithmic segment)
• A-law algorithm provides a slightly larger dynamic range
than the  -law at the cost of worse proportional distortion
for small signals.
• PCM telephone systems in Europe.
• A=87.56

 Am
 , 0  m  1A
1 log A
v  1 m
1 log A m
 , 1
 1 log A
13/29
A
 Digital Companding
• Digital companding involves compression in the transmitter
after the input sample has been converted to a linear PCM
code and then expansion in the receiver prior to PCM
decoding.
• Recent digitally compressed PCM systems use a 12-bit
linear PCM code and an eight-bit compressed PCM code.
• There are 16 segments (eight positive and eight negative).
• However it is often called as 13-segment compression
(because the curve for segments +0, +1, -0, and -1 is a
straight line with a constant slope and is considered as one
segment).

14/29
Digital Companding

15/29
 Digital Companding
• The eight-bit compressed code consists of a sign bit, a three-bit
segment identifier, and a 4-bit magnitude code that specifies the
quantization interval within the specified segment.
• The sign bit is transferred directly to an eight-bit compressed
code.
• The segment number in the eight-bit code number of leading 0s
in the 11-bit magnitude portion of the linear code beginning with
the most significant bit.
• Subtract the number of leading 0s (not to exceed 7) from 7.
• The result is the segment number, which is converted to a three-bit
binary number and inserted into the eight-bit compressed code as
the segment identifier.
• The four magnitude bits (A, B, C, and D) represent the
quantization interval (i.e., subsegments) and are substituted into
the least significant four bits of the 8-bit compressed code.
17/29
Digital Companding

1
Digital Companding

19/29
Digital Companding

20/29
 Test your understanding
• Derive the signal to quantization noise ratio for a PCM system
with linear quantization technique. Assume that the input to the
PCM is a sinusoidal signal.

21/29
 Summary

• Basics of quantization and its types are discussed

• Algorithms used for non uniform quantization are also
presented.
• Digital companding used in PCM systems has
been illustrated with examples

22/29