For the signals which does not change rapidly from one sample to next sample, the

PCM scheme is not preferred.

When such highly correlated samples are encoded the resulting encoded signal
contains redundant information.
By removing this redundancy before encoding an efficient coded signal can be
obtained: DPCM
Assume neighboring speech samples are similar to each other (large correlation)
By knowing the past behavior of a signal up to a certain point in time, it is possible to
make some inference about the future values. Predict future samples based on
previous samples
Record only difference between actual samples and predicted values

1/2/17

e(nTs) = x(nTs) - x^(nTs)

v(nTs) = Q[e(nTs)]
= e(nTs) + q(nTs)
u(nTs) = x^(nTs) + v(nTs)
u(nTs) = x^(nTs) + e(nTs) + q(nTs)
u(nTs) = x(nTs) + q(nTs)

1/2/17

 The Rx consists of a decoder to reconstruct the quantized error signal.

The quantized version of the original input is reconstructed from the decoder output
using the same predictor as used in the transmitter.
In the absence of noise the encoded signal at the receiver input is identical to the
encoded signal at the transmitter output.
Correspondingly the Rx output is equal to u(nTs), which differs from the input x(nts)
only by the quantizing error q(nTs).

1/2/17

For slowly varying signals, a future sample can predicted from past samples.
+

s(t)

Predictor

+
-

e(t)

e(t)

s(t)

Predictor

Receiver Side

Transmitter Side
Transversal filter can perform the prediction process.

1/2/17

DPCM System
Quantization error is accumulated.

1/2/17

Prediction Gain ( Gp)

The output signal-to-quantization noise ratio of a signal coder is defined as

where x2 is the variance of the signal x(nTs) and Q2 is the variance of the
quantization error q(nTs). Then

where E2 is the variance of the prediction error e(nTs) and (SNR)P is the
prediction error-to-quantization noise ratio, defined by

1/2/17

The Prediction gain Gp is defined as

The prediction gain is maximized by minimizing the variance of the prediction

error. The main objective of the predictor design is to minimize the variance of
the prediction error.
The prediction gain is defined by

and
where 1 Autocorrelation function of the message signal
1/2/17

Single-Tap Prediction
A single-tap predictor predicts the next input example x(kTs) from the
immediate previous input sample x([k-1]Ts).

1/2/17

1/2/17

1/2/17

10

Assignment Problems:
1. Consider a DPCM system whose transmitter uses a first-order predictor
optimized in the minimum mean-square sense. Calculate the prediction gain of
the system for the following values of correlation coefficient for the message
signal:

2. A PCM system uses a uniform quantizer followed by a 7-bit binary encoder.

The bit rate of the system is 56 Mbps. Find the output signal-to-quantization
noise ratio when a sinusoidal wave of 1MHz frequency is applied to the input.

1/2/17

11

Adaptive differential pulse code modulation(ADPCM)

Ref: P. Cummiskey, N. S. Jayant, and J. L. Flanagan, "Adaptive quantization
in differential PCM coding of speech," Bell Syst. Tech. J., vol. 52, pp. 1105
1118, Sept. 1973.

Adaptive similar to DPCM, but adjusts the width of

the quantization steps
Encode difference in 4 bits, but vary the mapping of
bits to difference dynamically
If rapid change, use large differences
If slow change, use small differences

Publication
US4513426 A
number
Publication
Grant
type
Application
US 06/451,487
number
Publication
Apr 23, 1985
date
Filing date

Dec 20, 1982

Priority
date

Dec 20, 1982

Fee status

Paid

Inventors

Nuggehally S. Jayant

Original
Assignee

At&T Bell Laboratories

Lecture Assignment to Chaitya!!!!!!

Refer above paper by N S Jayanth and discuss in
class.
1/2/17

12

Comparison and Sampling/Coding Techniques

1/2/17

13

Delta Modulation (DM)

Delta Modulation is a special case of DPCM.
Delta Modulation is the one-bit (or two-level) version of DPCM.
DM provides a staircase approximation to the over sampled version of an input BB
signal. Inexpensive & simple to implement.
The difference between the input and the approximation is quantized into only two
levels, namely, corresponding to positive and negative differences, resp.
The stair case approximation remains within of the input signal.

1/2/17

14

Block Diagram of DM
Tx/Rx

1/2/17

15

Complete Block diagram of a Delta Modulation (DM) System

1/2/17

16

DM Waveform

1/2/17

17

QUANTIZATION NOISE
Delta modulation systems are subject to two types of quantization error: slope
overload distortion, and granular noise.
Considering the maximum slope of x(t), & in order for {u(nTs)} to increase as fast
as the input sequence of samples {x(nTs)} in a region of maximum slope of x(t),
we require that the following condition be satisfied.

Otherwise, we find that the step size = 2 is too small for the stair case
approximation u(t) to follow a steep segment of the input waveform x(t), with
the result that u(t) falls behind x(t).
This condition is called slope-overload, and the resulting quantization error is
called slope-overload distortion(noise).
A DM using a fixed step size is often referred ton as linear delta modulation (LDM).
1/2/17

18

 The granular noise occurs when the step size is too large relative to the local
slope characteristics of the input wave form x(t), thereby causing the staircase
approximation u(t) to hunt around a relatively flat segment of the input waveform.
The choice of the optimum step size that minimizes the MSV of the quantizing
error in a LDM will be the result of a compromise between slope overload
distortion and granular noise.
Output SNR for Sinusoidal Modulation.
x(t) = A cos(2fot)
The maximum slope of the signal x(t) is given by

1/2/17

19

To Avoid Slope Overload Distortion,

Maximum permissible value of the output signal power equals

1/2/17

20

When there is no slope-overload, the maximum quantization error is .

Assuming that the quantizing error is uniformly distributed,

The Rx is a LPF with BW set equal to the message bandwidth, W such that f0 W. Assuming that
the average power of the quantization error is uniformly distributed over a frequency interval
extending from -1/Ts to 1/Ts, we get the result:
1/2/17

21

Average output noise power

Maximum value of the output SNR is:

Conclusion:
Under the assumption of no slope-overload distortion, the maximum output SMR
of a DM is proportional to the sampling rate cubed.
This indicates a 9db improvement with doubling of the sampling rate.

1/2/17

22

Simple Implementation of a DM system

1/2/17

23

Assignment Problems:
1. Consider a Delta modulator system designed to operate at 4 times the Nyquist
rate for a signal with a 4KHz bandwidth. The step size of the quantizer is 400mV.
Find the maximum amplitude of a 1KHz input sinusoid for which the delta
modulator does not show slope overload and also the post-filtered output SNR.
2. Consider a Speech Signal with maximum frequency of 3.4KHz & maximum
amplitude of 1volt. This speech signal is applied to a LDM whose bit rate is set at
20kbit/sec. Explain the choice of an appropriate step size for the modulator.
3. Determine the output SNR in a DM system for a 1KHz sinusoid sampled at 32KHz
without slope overload and followed by a 4KHz post reconstruction filter.

1/2/17

24

Adaptive Delta Modulation

Inappropriate step size creates noise.
Make adaptive.
number of successive 1s or 0s
1

3
2
4
4
.
.

1/2/17

step size

25

ADM Block Diagram.

1/2/17

26