21/04/2025

EE322
Analog & Digital Communication
Dr. Imran Javed
(Lecture-12)

Department of Electrical Engineering,

UET Lahore

4/21/2025 EE UET Lahore 1

Quantization
• After sampling, we get a discrete time signal
• However, the amplitude of the signal is still continuous
• By quantization, we discretize the amplitude of signal
• Performing both sampling and quantization, we finally
get a discrete-time finite-amplitude signal, where each
sample can be represented by finite number of bits
• In scalar quantization, each sample is quantized
individually
• In vector quantization, a block of samples is quantized
together at a time

4/21/2025 EE UET Lahore 2

1
 21/04/2025

Scalar Quantization
• Each sample is quantized into one of the chosen
finite number of levels
• Basically, each sample is rounded to the nearest
value from the set of possible quantization levels
• The amplitude axis is partitioned into N disjoint
subsets denoted by 𝑅𝑘 , 1 ≤ 𝑘 ≤ 𝑁
• Each subset is called as quantization region
• Corresponding to each 𝑅𝑘 , a quantization level
𝑥ො𝑘 is chosen which is the quantized version of
actual sample amplitude 𝑥 in that region.
4/21/2025 EE UET Lahore 3

4/21/2025 EE UET Lahore 4

2
 21/04/2025

Scalar Quantization (cont’d)

• Usually, the quantized amplitude is chosen as the
center of corresponding quantization region
• Then, each quantized sample is represented by a
binary sequence and transmitted
• This binary representation of quantized sample is
called as encoding
• If there are N quantization levels, then number of bits
needed to encode each sample are:
𝑛 = 𝑙𝑜𝑔2 (N) bits
• The data rate can be computed as:
𝑅 = 𝑛 × 𝑓𝑠 bits/sec
4/21/2025 EE UET Lahore 5

Scalar Quantization (cont’d)

• The quantization function Q is defined as:

• The square-error quantization distortion for a

single sample can be computed as:

• Average mean square quantization error for all

samples can be computed as

4/21/2025 EE UET Lahore 6

3
 21/04/2025

4/21/2025 EE UET Lahore 7

4/21/2025 EE UET Lahore 8

4
 21/04/2025

4/21/2025 EE UET Lahore 9

Normal Random Variable

4/21/2025 EE UET Lahore 10

10

5
 21/04/2025

Q-function

4/21/2025 EE UET Lahore 11

11

4/21/2025 EE UET Lahore 12

12

6
 21/04/2025

SQNR

4/21/2025 EE UET Lahore 13

13

4/21/2025 EE UET Lahore 14

14

7
 21/04/2025

Optimal Uniform Quantizer

• In uniform quantizer, all the quantization regions
(except the first and last) are of equal length,
which is denoted by Δ
• This means for
• Further assume that quantization levels are at a
distance of
• Then mean-square quantization error is given by

4/21/2025 EE UET Lahore 15

15

Optimal Uniform Quantizer (cont’d)

• To minimize D, we need to differentiate D with

respect to design parameters 𝑎1 and Δ , put equal to
zero and solve the two equations to find optimal
values of these parameters

4/21/2025 EE UET Lahore 16

16

8
 21/04/2025

Optimal quantization
spacing for standard
Gaussian random
source

4/21/2025 EE UET Lahore 17

17

PCM
• It is a baseband scheme that can be used to represent any
analog signal in digital form

4/21/2025 EE UET Narowal 18

18

9
 21/04/2025

4/21/2025 EE UET Lahore 19

19

PCM (cont’d)

4/21/2025 EE UET Lahore 20

20

10
 21/04/2025

Q-7.10
• Using Table 7.1, design an optimal quantizer
for the source given in Example 7.2.1
• Compare the distortion of this quantizer to the
distortion obtained there.

4/21/2025 EE UET Lahore 21

21

Solution

∆
𝑥ො𝑖 = 𝑥ො𝑖−1 + ∆= 𝑎1 + 𝑖 − 1 ∆ −
2

4/21/2025 EE UET Lahore 22

22

11
 21/04/2025

Q-7.9

4/21/2025 EE UET Lahore 23

23

Solution

4/21/2025 EE UET Lahore 24

24

12
 21/04/2025

Solution (cont’d)

4/21/2025 EE UET Lahore 25

25

13