ECE 3614: Introduction to

Communications Systems
Lecture 15: Quantization and
Waveform Encoding

Lingjia Liu, Ph.D.

Associate Professor
Electrical and Computer Engineering
Virginia Tech, Blacksburg, VA
 Overview

 Last class we introduced the concept of digital communications

 Digital communication of analog message signals requires
analog-to-digital conversion which includes
– Sampling
– Quantization
– Waveform encoding
 Today we discuss quantization and waveform encoding
 Reading
– 5.2-5.4
 Pulse Code Modulation (PCM)

 Pulse Code Modulation refers to a system that creates a digital baseband

signal from an analog signal using sampling, quantization and waveform
encoding.

PCM

Flat-top
x Sampling
Quantizer Waveform
encoding
n*fs bps
fs M=2n levels n bits/level
 Digital Representation of Analog
Signals
 Sampling analog signals makes them discrete in time:

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1
0 2 4 6 8 10
 Digital Representation of Analog
Signals
 Quantization of sampled analog signals makes the samples discrete in
amplitude:

2.5

1.5

0.5

-0.5

-1

-1.5

-2

-2.5
0 2 4 6 8 10
 Quantization

 Continuous time signals are sampled at discrete time intervals

 Sampling may be performed without distortion provided signal is
sampled at Nyquist rate
 Continuous-valued samples of data require an infinite # of bits to
represent with perfect precision
 Quantization is the process of approximating continuous-valued
samples with a finite # of bits
 Quantization always introduces some distortion
 Notation Associated with Quantization

 Let X be a random variable representing a sample of data

Then X  f Q  X  is the quantized value of X

~

 A quantizer has M quantization levels:

X~ ~
x1, ~ xM 
x2 ,, ~
 The M levels correspond to M quantization regions
 The endpoints of the quantization regions are specified by M  1 values:

x0 , x1 , , x M 
where x0  , x M  
 Then:
x  fQ  x  ~
xk 1  x  xk  ~ xk
Graphical Description of Quantization

Output Level ~
x
4
~
x8
~ 3
x7
~
x6
2

~
x5 1

~
x4 -4 -3 -2 -1 1 2 3 4 Input Level x
~ -1
x3
~ -2
x2
~ -3
x1 -4
Table Representation of Quantizer

k x k 1 xk ~
xk Output
Bits
1 . -3 -3.5 000
2 -3 -2 -2.5 001
3 -2 -1 -1.5 010
4 -1 0 -0.5 011
5 0 1 0.5 100
6 1 2 1.5 101
7 2 3 2.5 110
8 3 . 3.5 111
 Concise Representation of
Quantizer

 Usually, it is sufficient just to list the quantization levels.

 Example: {-3.5, -2.5, -1.5, -0.5, 0.5, 1.5, 2.5, 3.5}
 Why?
– We assume that all points are quantized to the nearest quantization level
– This determines where the borders of the quantization regions are
– Any other borders would increase the error introduced by the quantizer
 Practical Methods for Implementing
Analog to Digital Converters

 Counting or Ramp ADC

– Test value is incremented in equal steps until it is greater than
Lower Complexity

input sample
 Serial or Successive Approximation ADC

Faster
– Uses binary search to narrow range of input sample until desired
accuracy is reached
 Parallel or Flash ADC
– Input sample is compared with all possible quantization levels at
once
 Distortion

 Quantization introduces distortion into a signal.

 We want to minimize average distortion D, where

 This measure of distortion is sometimes also called mean square error

(MSE)
 MSE penalizes large errors more than small errors
Another Way of Viewing Quantization

 Quantization adds a random noise to the true value of the sample point

2
Then MSE  E nQ   may be thought of as noise power

 We can define a signal-to-noise ratio (SNR) to measure performance

Quantizer
x + x  f Q  x   x  nQ
~

nQ  x  ~
x
 Signal to Noise Ratio Calculations for
Quantizers

 Average SNR

 Note that the noise nQ here is not necessarily Gaussian. It is simply

modeling “distortion”.
 Example: SNR Calculation

 Let:

 Let:
ìï 1 8, -4 £ x £ 4
f ( x) = í
ïî 0, else
 Signal power:
 SNR for Uniform Quantization

 General result: S  M2

 
 N  avg
– Assumes uniform quantizer with M levels
– Assumes that input samples have uniform distribution with identical
range as quantizer
– Though some texts do not make it clear, this result applies only when
these special set of conditions hold. Otherwise, we have to use integral
formula
 However, a useful Rule of Thumb:
– Each additional bit (doubling M) increases SNR by 6 dB
S
   6.02n  a
 N avg
– Where a depends on the distribution of the signal
 Waveform Encoding

 Once the information is converted to bits, it must be mapped onto

waveforms
 If each bit is mapped to one of two different waveforms, we term
this binary encoding
 If m bits are mapped to M = 2m waveforms, we term this M-ary
encoding
 The mapping of bits to pulse-based waveforms (i.e., pulse trains)
is termed “line coding”
“1” “1”
{1, 0, 0, 1, 0…}
“0” “0” “0”

Example: Non-return to zero polar signaling

 Bandwidth of PCM Signals

 Sample rate: fs samples/second.

 Bit rate out of the quantizer:
f s log 2 M  f s  n bits / second
 Bandwidth of the resulting digital signal depends on the
waveform encoding used
 Minimum theoretical bandwidth with optimal waveform
(baseband waveforms) : f s  n 2 Hz
 First null bandwidth (with rectangular pulse waveforms) for
baseband waveforms: f s  n Hz
Note that the resulting bandwidth
depends on the digital waveform
that is used.
 Rectangular Pulse

Time Waveform Magnitude Spectrum

20log10(|X(f)|)
Voltage

1/T
 Example 3: PCM Calculation

 Problem:
– Suppose that an analog music signal is found to have a bandwidth of 15
kHz and that samples of the signal may be modeled as having a uniform
distribution.
– Find the minimum first-null bandwidth (assuming the use of square pulses)
at which it would be possible to transmit a PCM signal while maintaining
an average SNR of at least 58 dB.
Example 3 (cntd.)
 Bandpass Digital Modulation

 Baseband digital maps bits to pulses

 Bandpass digital maps bits to sinusoids with different
– phases
– frequencies
– amplitudes
 Bandpass digital techniques can be viewed as AM or FM with
digital baseband waveforms as the message signal
 Binary Frequency Shift Keying

 While FM is an analog communications technique, the basic idea

can also be used with digital modulation
 First we map bits to square pulses with amplitudes +1/-1
representing data bits 1/0.
 Second, we use the square pulse baseband signal as our message
and use kf = Df = Rb/2 where Rb is the bit rate.
 The result is a digital modulation scheme known as Binary
Frequency Shift Keying (BFSK)
 Example

Rb = Rs = 1bps fc = 5Hz  f1 = 4.5Hz, f2 = 5.5Hz

1 1

0.5 0.5

Volts
Volts

0 0

-0.5 -0.5

-1 -1

0 1 2 3 4 5 6 7 8 9 10 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3

Time (sec) Time (sec)

f1 = 4.5Hz , f2 = 5.5Hz
 Relationship between AM and
Digital Modulation

 DSB-LC AM  BASK
– Message is entirely in the amplitude
 DSB-SC AM  BPSK, BASK
– Message is in the phase and amplitude
– In BPSK, amplitude is always 1 since the message is square wave.
– In BASK the message signal doesn’t go negative so there are no
phase changes.
 Relationship Between Analog
and Digital Modulation Schemes
Message Signals
1 0 0 1
+A Amplitude
Unipolar NRZ Only
BASK
Phase
+A Only
Polar NRZ
BPSK

-A

Arbitrary Amplitude
DSB-SC
Analog signal and Phase

DSB-LC

+Ac Arbitrary Amplitude

Analog signal Only
with offset
What is
Message
 Summary

 Transmitting an analog signal over a Digital Communication system

involves
– Sampling
– Quantizing
– Waveform encoding (i.e., mapping the bits to transmit waveforms)
 Today we examined the last two items
– Sampling/Quantization convert analog signals to a stream of 1’s and 0’s
– Waveform encoding maps bits by modulating
 A pulse train in baseband digital communications
 A sinusoid in bandpass digital communications