1

Digital Communications

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 2

Analog Signal and Digital Signal

Analog signal
t (continuous amplitude)

111010101100001 1 110 1
Digital signal
t (discrete amplitude)

Why digital?

• easier to be regenerated • complete theory

• flexible hardware implementation and efficient storage

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 3

Digital Communications
Analog Signal
Bit sequence
t 0001101110……

A-D Digital Digital

Source Baseband Bandpass
Conversion
SOURCE Modulation Modulation

t t

Baseband Bandpass
Channel Channel

D-A Digital Digital

User Baseband Bandpass
Conversion Demodulation Demodulation

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 4

Lecture 6. Digital Communications

Part I. Analog-to-Digital (A-D) and
Digital-to-Analog (D-A) Conversion

• Sampling
• Quantization
Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 5

A-D Conversion

Analog signal Bit sequence

A-D Conversion

Sampling Quantization

000010011010000…
t t

• Sampling
• A continuous-time analog signal is transformed into a discrete-time
signal.

• Quantization
• A discrete-time signal is transformed into a series of binary bits.
Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 6

Sampling

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 7

Sampling

Time Domain 
xS (t )  x(t )   (t  nTS )
n 
x(t)

Sampling

t t
Ts


  (t  nT )
n 
S

TS: Sampling period

fS=1/TS: Sampling rate
TS t

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 8

Sampling
 
Frequency Domain X S ( f )  X ( f )  ( fS   ( f  nf
n 
S ))  f S  X ( f  nf
n 
S )

Harmonic components
X(f)
XS(f)
Sampling

…... …...

0 B f -2fS -fS 0 B fS 2fS f


fS   ( f  nf
n 
S )
• How to recover the original signal x(t)?

Lowpass filtering
fS f

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 9

Aliasing

XS(f)

…... …...

-2fS -fS 0 B fS 2fS f

fS-B  B
XS(f)

Aliasing

…... -2fS -fS 0 B fS 2fS …... f

Distortion will be incurred if there’s aliasing.

• What is the requirement on the sampling rate to avoid aliasing?

f S  2B
Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 10

Nyquist Sampling Theorem

Nyquist Sampling Criteria : A baseband signal with limited

bandwidth B can be uniquely determined by its values at uniformly
spaced intervals if and only if the sampling rate f S  2 B.

fS=2B is called the Nyquist sampling rate, representing the minimum

requirement without introducing distortion.

• A signal with a larger bandwidth requires a higher sampling rate.

What about a signal which is not strictly bandlimited to B?

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 11

Practical Considerations on Sampling

X(f)
XS(f) Aliasing

…... …...

0 B f 0 B f

XS(f)
• Use a sampling rate fS>2B
…... …...

• Pre-filtering 0 B fS>2B f
X(f)
XS(f)

…... …...

0 B f -2B 0 B fS=2B f

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 12

Quantization

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 13

Quantization
xS(t)
1

0
t

-1 Quantization

11 11 11 11 11 10 10 10 11 10 10 10 00 00 00 10 10…

• Quantization: to transform a discrete-time signal into a series of

binary bits.
– Step 1: Use a finite number of values to represent the amplitude;
− Step 2: Assign binary codes to different quantization level.
Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 14

Uniform Quantization
xS(t)
xmax

Dynamic range
t of xs(t)

xmin

• Suppose the sampled signal amplitude x varies from xmin to xmax.

• Suppose the number of quantization levels L  2

b
(use b bits to represent
a sampling symbol)

Uniform Quantization: the dynamic range is divided into L equal-width

quantization regions.
Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 15

Uniform Quantization

• Uniform Quantization: the dynamic range is divided into L equal-width

quantization regions.

x0 x1 x2 x3 xL-1 xL
xmin
s0 ŝ0 …… xmax x

xmax  xmin
Step size  
L Quantization threshold

Let ŝ0 represent the quantized value of the input value s0.

Midriser: ŝ0 =xi+/2 if xi<s0≤xi+1.

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 16

Transfer Function of Midriser

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 17

Example 1: Midriser

Suppose that a signal x(t) with dynamic range (-4,4) volts is

applied to a 3-bit midriser.

1. Plot the transfer function;

2. Determine the quantized values of 0.15V and -3.1V.

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 18

Solution

The quantization step size at the output is   (4  (4)) / 23  1

1. y 2. Quantized values of
0.15V and -3.1V are
3.5
2.5
0.5V and -3.5V,
1.5 respectively.
0.5 1 2 3 4
x x
-4 -3 -2 -1 -0.5 x
-1.5
-2.5
-3.5

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 19

Quantization Error

x0 x1 x2 x3 xL-1 xL
xmin
s0 ŝ0
…… xmax x

xmax  xmin
Step size  
L Quantization threshold

• What is the maximal difference between the input s0 and its quantized
value ŝ0 ? /2

Let e  sˆ0  s0 .  / 2  e   / 2

Quantization Error (noise)

e can be regarded as a uniformly distributed random variable with pdf

1/   / 2  x   / 2
e  0,  e2  2 /12
f e ( x)  
 0 otherwise
Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 20

Signal-to-Quantization-Noise-Ratio (SQNR)

 xmax  xmin   xmax  xmin 

2 2

• Quantization error power =  e   /12  

2 2
12 L2 12  22b
xmax  xmin
Step size  
L
Number of quantization levels L  2b

 A larger b leads to a lower quantization error power.

(We can use more bits to improve the quantization precision.)

• SQNR:
SQNR  Px /  e2

Px and  e2 are the power of the signal and the quantization error, respectively.

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 21

Example 2: SQNR

A sinusoidal signal with peak amplitude Am is applied to a

uniform quantizer with a dynamic range of (-Am, Am).
Determine the output SQNR.

x  x   2 Am 
2 2
We have  max min 
Px  Am2 / 2 and  e2  2 /12
12  22b 12  22b
Px
Therefore, SQNR   1.5  22b
 2
e

Write it in the form of decibels:

10log10 (SQNR)  10log10 1.5  2b 10log10 2  1.8  6b (dB)

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 22

Non-Uniform Quantization

Human speech:
volts • Small samples have a
larger percentage;
• Small samples are more
0 susceptive to noise.
t (msec)

Give the small samples more priority!

(more bits!)
Use small intervals for signals with small amplitudes and large intervals for
signals with large amplitudes:

…… ……
xmin 0
xmax x
Step size  increases with the signal magnitude.
Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 23

Non-Uniform Quantization
Compress the original
How to perform non-uniform quantization? signal such that its
amplitude roughly
• non-linear compression on the analog signal;
follows a uniform
• uniform quantization. distribution!

-law :
ln 1   (| x | / xmax )
y  ymax sgn( x)
ln(1   )
sgn( x)  1 if x  0
sgn( x)  1 if x  0

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 24

SQNR of -Law Quantization

Uniform Quantization -law Quantization
y(t) y(t)

 

x(t) x(t)
1<  2 <  3 < 4

• For weak signals (with low power), -law has better SQNR as   law  uniform
• For strong signals (with large power), -law has lower SQNR as   law  uniform

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 25

Example 3: -Law Quantizer

Suppose that a signal x(t) with dynamic range (-4,4) volts is

applied to a 3-bit -law quantizer. Determine the quantization
thresholds at both input and output sides.

For simplicity, we assume that ymax=xmax and ymin=xmin, and

=255.

Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 26

Solution

The quantization step size at the output is   (4  (4)) / 23  1

The quantization thresholds at the output are then given by
yth  [4, 3, 2, 1,0,1, 2,3, 4]

ln 1   (| x | / xmax )
According to y  ymax sgn( x)
ln(1   )
x   xmax /   1     1  sgn( y)
| y|/ ymax
We have  

xth   xmax /   1     1 sgn( yth )

yth / xmax
Finally, according to  
The quantization thresholds at the input are then given by
xth  [4, 0.9882, 0.2353, 0.0471,0,0.0471,0.2353,0.9882, 4]
Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 27

Code Assignment

output
Gray Code
100
7
Generation: for an N-bit binary code,
5
101
111 bN bN-1 …… b3 b2 b1
3

1110
-8 -6 -4 -2010 2 4 6
   
8 input
-1
011 -3 gN gN-1 gN-2 …… g2 g1
001
-5
modulo-2 addition
000 -7
Properties:
1. The first bit is the sign bit;
2. Adjacent words differ only by one bit;
000000 001001 3. Except for the sign bit, the codewords are
010011 011010 …… mirror symmetrical about the horizontal axis.
Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6
 28

A-D and D-A

Analog
Baseband Signal 0001101110……

Source A-D Conversion

• Sampling
• Quantization

0001101110……
User Analog
D-A Conversion Baseband Signal

• Decoding (0001101110…  0.1 -0.3 0.2 …)

• Generate a pulse the amplitude of which is the quantized value (repeat)
• Lowpass filtering
Lin Dai (City University of Hong Kong) EE3008 Principles of Communications Lecture 6