Digital Communications (EECS-4214)

[Fall-2021]
Hina Tabassum
Department of Electrical Engineering and Computer Science
York University, Canada
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Week 0

Introduction to Course
Communication System
Analog vs Digital Communication
Transmitter
Channel
Receiver

(Hina Tabassum) 2 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Week 1

Review of Probability Concepts

Discrete Random Variables
Continuous Random Variables
Introduction to Random Processes
Stationary vs Non-stationary Random Processes

(Hina Tabassum) 3 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Summary: Random Variables

Random variable X

Characteristic Function
of X is similar to LT of X
with s= jw

Moment Generating Function

Probability Density Cumulative Density Moments of
(MGF) or
Function (PDF) Function (CDF) X
Laplace Transforms (LTs)

1st Moment of X,
1st Central Moment
i.e., Mean
around X, i.e., E[X-E[x]]
E[X]

2nd Central Moment

2nd Moment of X around mean of X, i.e.,
E[X^2] Variance Standard Deviation of X
E[(X-E[x])^2] is Square root of
Variance
Nth Central Moment
Nth Moment of X around mean of X, i.e.,
E[X^n] Variance
E[X-E[x]]

(Hina Tabassum) 4 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Week 2

Two Dimensional Transformation

Autocorrelation
Power Spectral Density
Random Process through Linear Systems
Practice Problems

(Hina Tabassum) 5 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Summary: Random Process

Random Process

Wide-Sense Auto Correlation

Ergodicity Strict Sense Stationary
Stationary (Time Domain)
Wiener-Khintchin
Theorem

Power Spectral Density

(Frequency Domain)
Ensemble Averaging Time Averaging

(Hina Tabassum) 6 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Week 3

Quiz-1
PCM: Sampling
PCM: Quantization

(Hina Tabassum) 7 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Introduction to PCM

PCM = ’Pulse Code Modulation’

Method of analog –to-digital conversion in which the analog signal is
converted into an electrical waveform of two or more levels.
PCM = Sampling + Quantization +Encoding

Sampling Quantization Encoding Modulation

Input Transmitter Reciever Output

Channel
Transducer (TX) (RX) Transducer

Source Destination

(Hina Tabassum) 8 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Sampling Illustration
Nyquist Criterion: Fs ≥ 2W

x (t ) X( f )

t f

Ts 2 Ts 3 Ts t Fs 2 Fs
f

t ~ f
X (f)
~
x (t ) H( f )

t f

(Hina Tabassum) 9 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Sampling Steps

Original signal x(t)

Original signal spectrum X (f )
Sampled signal
P∞ P∞
xδ (t) = x(t) k=−∞ δ(t − kTs ) = k=−∞ x(kTs )δ(t − kTs )

Sampled signal spectrum

P∞
Xδ (f ) = Fs ∗ n=−∞ X (f − nFs )

(Hina Tabassum) 10 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Sampling Steps

Spectrum Filtering
(
1 −W ≤ f ≤ W
H (f ) = (1)
0 otherwise

Reconstructed signal spectrum after filtering

X̃ (f ) = Xδ (f )H (f )

Reconstructed signal
P∞
x̃(t) = 2W k=−∞ x(kTs )sinc(πFs (t − kTs ))

(Hina Tabassum) 11 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Week 4

. PCM: Quantization

. PCM: Uniform Quantization

. PCM: Non-Uniform Quantization

(Hina Tabassum) 12 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Quantization

The process of transforming sampled amplitude values of a message

signal into a discrete amplitude value is referred to as Quantization.

Quantization

Scalar Uniform
SQNR
Non-
Vector
Uniform

(Hina Tabassum) 13 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Quantization Types

Uniform Quantization: when the signal amplitude is discretized

uniformly, i.e., the difference between any two quantization levels is
same.

Non-Uniform Quantization: when the signal amplitude is not

discretized uniformly, i.e., the difference between any two
quantization levels is not the same.

(Hina Tabassum) 14 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Scalar Quantization

A scalar value is selected from a finite list of possible values to

represent a sample.

Example: Landline Telephony

Message signal bandwidth = 4kHz
Sampling frequency = 8 kHz
Levels = 256 (8 bits per sample)
Bit rate = ?

(Hina Tabassum) 15 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Vector Quantization

A vector is selected from a finite list of possible vectors to represent an

input vector of samples.

Example: NA-TDMA
Message signal bandwidth = 4kHz
Sampling frequency = 8 kHz
Levels = 8192 (13 bits per sample)
Bit rate = ?

Vector quantization is a lossy compression technique used in speech

and image decoding.

(Hina Tabassum) 16 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Vector Quantization

Key Operation: quantization of a random vector by encoding it as

a binary codeword. Each input vector can be viewed as a point in
an n-dimensional space.

Let k be the number of quantization regions in the n-dimensional

space. Every region has its own codevector, k is the number of
codevectors in the codebook, and n represents the number of
samples in a codevector.

The vector quantizer can thus be perceived by partitioning the

space into a set of non-overlapping n-dimensional regions. The
vector is encoded by comparing it with a code book consisting of a
set of stored reference vectors known as code vectors.

(Hina Tabassum) 17 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Uniform Quantization

Output Output
7Δ/2

5Δ/2
2Δ
3Δ/2
Δ/2 Δ

Δ 2Δ 3Δ 4Δ Input
3Δ/2 Input

Mid-Rise Quantizer Mid-Tread Quantizer

Input-Output Characteristics of Uniform Quantizers

Mid-Rise: Even number of quantization levels
Mid-Tread: Odd number of quantization levels

(Hina Tabassum) 18 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

N -Level Quantizer (Mid-Rise Type) Analysis

The mid-point of each interval is selected as the quantized output

of each interval. At this point the quantization error is zero.

Quantization error varies from −∆/2 and +∆/2

Quantization error continues to grow at the extreme ends.

(Hina Tabassum) 19 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Signal-to-Quantization Noise Ratio (SQNR)

X = Input signal is a random variable and follows a PDF of fX (x)

Xq = Quantized output signal
nq = X − Xq = Quantization noise error

E[X 2 ] 12 2
SQNR = E[nq2 ] = ∆2 E[X ]

When ∆ → 0, E[X 2 ] ≈ E[Xq2 ]

(Hina Tabassum) 20 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Signal-to-Quantization Noise Ratio (SQNR)

E[X 2 ] = x 2 fX (x)dx
R

∆2
E[nq2 ] = (X − Xq )2 fX (x)dx =
R
12

To derive the quantization noise power E[nq2 ], we consider following

simplifying assumptions.

Assuming that X0 − ∆/2 < X < XN −1 + ∆/2 implies that

fX (x) = 0 for X0 − ∆/2 < X < XN −1 + ∆/2.

If ∆ → 0, fX (x) ≈ fX (xi )

(Hina Tabassum) 21 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Signal-to-Quantization Noise Ratio (SQNR)

12
x 2 fX (x)dx
R
SQNR= ∆2 x

For large vales of ∆, SQNR tends to zero.

2U0 U0
∆ = q−1 ≈ 2Uq 0 = 2n−1
q=2 n
2 U02
E[nq2 ] = ∆ 12 = 3.22n

] 2
SQNR= 10 log10 E[X
U2
+ 6n + 4.8[dB]
0

(Hina Tabassum) 22 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Example: Uniformly Distributed Signal

1
fX (x) = 2U0 , |x| < U0

SQNR = (q − 1)2 dB
SQNR = 20log10 (q − 1) dB
SQNR = 48 dB (when q = 256)

q >> 1 and q = 2n

SQNR[dB] = 20nlog10 2 = 6n

BitRate = Fs SQNR[dB]
6

(Hina Tabassum) 23 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Example: Gaussian Distributed Signal

2
fX (x) = √1 exp(− x 2 ), σ = U0 /4
σ 2π 2σ

2
SQNR = 3 (q−1)
16

SQNR[dB] = 10log10 3 + 20log10 (q − 1) − 10log10 16

SQNR[dB] = 48 − 7.25 (when q = 256)

Uniform quantizer underperforms for Gaussian distributed signal

(Hina Tabassum) 24 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Summary: Uniform PCM

Input signal amplitude range [-U0 U0 ]

Number of quantization levels: q = 2n
2U0
Quantization step: ∆ = q−1
Quantization noise power: E[nq2 ] = ∆2 /12
] 2
SQNR = 10log10 E[X
U2
+ 6n + 4.8[dB]
0

Uniformly Distributed Signal: SQNR= 20log10 (q − 1)

q >> 1, SQNR= 6n
Gaussian Distributed Signal: SQNR= 20log10 (q − 1) − 7.25dB

(Hina Tabassum) 25 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Example 1

An analog signal is sampled at the Nyquist rate Fs = 20KHz and quantized into
L=1024 levels. Find bit-rate and the time duration Tb of one bit of the binary
encoded signal.

(a) Rb = n.Fs = Fs log2 L

1
(b) Tb =
Rb

(Hina Tabassum) 26 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Example 2

A PCM system uses a uniform quantizer followed by a 8-bit binary encoder. The bit
rate of the system is 56Mega bits/sec. Find the output signal-to-quantization noise
ratio when a sinusoidal wave of 1MHz frequency is applied to the input.

(a) Rb /n = Fs
2
0.5xmax
(b) SQNR = ∆2 /12

(Hina Tabassum) 27 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Non-Uniform Quantizer
In non–uniform quantizer, the step size varies.
The use of a non–uniform quantizer is equivalent to passing the baseband signal
through a compressor and then applying the compressed signal to a uniform
quantizer.
The resultant signal is then transmitted.
At the receiver, a device called Expander is used to restore the signal samples to
their correct relative level.

Non-Uniform Quantization

Transmitter
COMPRESSOR A/D Converter

D/A Converter EXPANDER Receiver

(Hina Tabassum) 28 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Compression Laws: µ - Law

Commonly used logarithmic compression law

The µ-law is used for PCM telephone systems in the USA, Canada and Japan.
A practical value for µ is 255.
ln(1+µ|x|)
F(x) = ln(1+µ)
sgn(x), −1 ≤ x ≤ 1

(Hina Tabassum) 29 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Compression Laws: A- Law

Commonly used logarithmic compression law
The A-law is used for PCM telephone systems in the Europe.
A practical value for A is 100.
( 1+ln(A|x|)
1+ln(A)
sgn(x), 1/A ≤ |x| ≤ 1
F(x) = A|x|
1+ln(A)
sgn(x), |x| ≤ 1/A

(Hina Tabassum) 30 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Signal-to-Quantization Noise Ratio (SQNR)

∆2 fX (x)
E[nq2 ] =
R
12 x [F 0 (x)]2 dx

What happens when F(x) = x?

E[X 2 ]
SQNR= E[nq2

(Hina Tabassum) 31 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

E[nq2 ] Expression: µ - Law

∆2 fX (x)
E[nq2 ] =
R
12 x [F 0 (x)]2 dx

2
∆2 ln (1+µ)
R1
E[nq2 ] = 12 µ2 −1 (1 + µ|x|)2 fX (x)dx

0
This is a general expression bu using the compressor function F (x)
and thus can be used in conjunction with any type of signal
distributions fX (x).

(Hina Tabassum) 32 / 34
PCM: Quantization PCM: Uniform Quantization PCM: Non-Uniform Quantization

Summary: Non-Uniform PCM

Input signal amplitude range [-U0 U0 ]

Number of quantization levels: q = 2n
2U0
Quantization step: ∆ = (q−1)F 0 (x)
∆2 fX (x)
Quantization noise power: E[nq2 ] =
R
12 x [F 0 (x)]2 dx

(Hina Tabassum) 33 / 34
Questions?