Vous êtes sur la page 1sur 38

Communication numérique 2 (M9)

Chapter 3 : PCM and DPCM


Master Systèmes de Télécommunications et Réseaux Informatique
(STRI)

Pr Said SAFI
Université Sultan Moulay Slimane
Faculé Polydisciplinaire
Département de Maths&info
Béni Mellal

19 mai 2017

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and19DPCM


mai 2017 1 / 37
Contenu

Contenu

1 Introduction générale

2 exemple : Disque audio numérique

3 Quantication non-uniforme
Quantication non-uniforme avec la loi-A et la loi µ

4 Transmission bandwidth and the output SNR

5 Dierential Pulse Code Modulation (DPCM)

6 Analysis of DPCM

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and19DPCM


mai 2017 2 / 37
exemple : Disque audio numérique

Les disques audio numériques emploient la modulation PCM avec


quantication uniforme sur 16 bits et une fréquence d'échantillonnage de
44.1 KHz pour numériser les signaux audio.
Soit un signal m(t) = Am cos(2πfm t) (en volts). Quel est le rapport SNR
de quantication lorsque Am = mmax (amplitude maximale du signal à
l'entrée du quantication) ? La puissance Pm du signal (référence 1Ω est) :
A2m m2max
Pm = 2 = 2 [watts]

Le quanticateur est uniforme avec M = 2n niveaux et un pas ∆ de :

∆= 2mmax
M = 2mmax
2n [volts]

La puissance du bruit de quantication Pbruit (en supposant une


distribution uniforme du bruit) est :
( 2m2max )2 4m2max
Pbruit = ∆2
12 = n
12 = 12×(2n )2
[watts]

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and19DPCM


mai 2017 3 / 37
exemple : Disque audio numérique

Le rapport SN Rq de quantication est :


m2max )
Pm ( 3×2n
SN Rq = Pbruit = m2
2
= 2 = 3 × 2(2n−1)
max )
( 3×2 n

En représentation décibels, on a :

SN Rq = 10log10 (3 × 22n−1 )

A.N : Avec n = 16bits (M = 216 = 65536 niveaux de quantication) :

SN Rq = 3 × 2(2n−1) = 3 × 231 = 6.442 × 109

SN Rq (dB) = 10log10 (6.442 × 109 ) = 98.09dB

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and19DPCM


mai 2017 4 / 37
Quantication non-uniforme

Les signaux analogiques représentant la voix ont plus de chance d'avoir de


faibles valeurs que des valeurs près du maximum permis.
Le bruit de grenaille devient alors important. On peut réduire la pas de
quantication pour des valeurs du signal proches de 0 : On parle de
quantication non-uniforme.
In the following gure we represent the no-uniform quantication consist of
a compression circuit in the transmitter and the decompression in the
receiver.

Figure  no-uniform quantication : compression at the transmitter and the


decompression at the receiver

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and19DPCM


mai 2017 5 / 37
Quantication non-uniforme Quantication non-uniforme avec la loi-A et la loi µ

Among several choices, two compression laws have been accepted as


desirable standards by the CCITT (Consultative Committee for
International Telephony and Telegraphy) : the µ-law used in North America
and Japan, and the A-law used in Europe and the rest of the world and
international routes. Both the µ-law and the the A-law have odd symmetry
about the vertical axis. The µ-law (for positive amplitudes) is given by
1 m m
y= ln(1 + µ 0≤ ≤1 (1)
ln(1 + µ) mp mp

The A-low (for positive amplitude) is given by


  
A m m 1

1+ln(A) mp 0≤ mp ≤ A
y= m
1+ln(A m ) (2)
p 1 m

1+ln(A) A ≤ mp ≤1

The compression µ(or A) determines the degree of compression. To obtain


a nearly constant S0 /N0 over an input signal power dynamic range of
40dB, µ should be greater than 100.
Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and19DPCM
mai 2017 6 / 37
Quantication non-uniforme Quantication non-uniforme avec la loi-A et la loi µ

Early North American channel banks and other digital terminals used a
value of µ = 100, which yielded the best results for 7-bit (128-level)
encoding. An optimum value of µ = 255 has been used for all North
American 8-bit (256-level) digital terminals, and the earlier value of µ is
now almost extinct. For the A-law, a value of A = 87.6 gives comparable
results and has been standardized by the CCITT.
The compressed samples must be restored to their original values at the
receiver by using an expander (decompression) with a characteristic
complementary to that of the compressor. The compressor and the
expander together are called the compander.
Generally speaking, compression of a signal increases its bandwidth. But in
PCM, we are compressing not the signal m(t) but its samples. Because the
number of samples does not change, the problem of bandwidth increase
does not arise here.
We can demonstrate that when a µ-low compandor is used, the output
SNR is given by
S0 3L2 m2p
' µ (3)
N0 (ln(1 + µ))2 m2 (t)
Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and19DPCM
mai 2017 7 / 37
Quantication non-uniforme Quantication non-uniforme avec la loi-A et la loi µ

The output SNR for the cases of µ = 255 and µ = 0 (uniform


quantization) as a function of m2 (the message signal power) is shown if
the following gure.

Figure  (a) µ-law characteristic. (b) A-law characteristic

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and19DPCM


mai 2017 8 / 37
Quantication non-uniforme Quantication non-uniforme avec la loi-A et la loi µ

Figure  Signal-to-noise-quantization-noise ratio in PCM with and without


compression

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and19DPCM


mai 2017 9 / 37
Quantication non-uniforme Quantication non-uniforme avec la loi-A et la loi µ

Figure  Non uniform quantization

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 10 / 37
Quantication non-uniforme Quantication non-uniforme avec la loi-A et la loi µ

Figure  Non uniform quantization

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 11 / 37
Quantication non-uniforme Quantication non-uniforme avec la loi-A et la loi µ

 The compandor :
A logarithmic compressor can be realized by a semiconductor diode,
because the V-I characteristic of such a diode is of the desired form in the
rst quadrant :
KT I
V = ln(1 + ) (4)
q Is
Two matched diodes in parallel opposite polarity provide the approximate
characteristic in the rst and third quadrants (ignoring the saturation
curent).
In practice, adjustable resistors are placed in series with each diode and a
third variable resistor is added in parallel. By adjusting various resistors, the
resulting characteristic is made to t a nite number of points (usually
seven) on the ideal characteristics.
An alternative approach is to use a piecewise linear approximation to the
logarithmic characteristics. A 15-segmented approximation (see gure
below) to the eight bit (L=256) with µ = 255 law is widely used.
Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and
19 DPCM
mai 2017 12 / 37
Quantication non-uniforme Quantication non-uniforme avec la loi-A et la loi µ

Figure  Piecewise linear compressor characteristic

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 13 / 37
Quantication non-uniforme Quantication non-uniforme avec la loi-A et la loi µ

 The encoder :
The multiplexed PAM output is applied at the input of the encoder, which
quantizes and encodes each sample into a group of n binary digits. A
variety of encoders is available.
We shall discus here the digit-at-a-time encoder, which makes n
sequential comparisons to generate an n-bit code word. The sample is
compared with a voltage obtained by a combination of referenced voltages
proportional to 27 , 26 , ..., 20 .
The reference voltage are conveniently generated by a bank of resistors R,
2R, 22 R,..., 27 R.
The encoding involves answering successive question, beginning with
whether or not the sample is the upper o lower half of the allowed range.
The rst code digit 1 or 0 is general, depending on whether the sample is
in the upper or the lower half of the range.

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 14 / 37
Quantication non-uniforme Quantication non-uniforme avec la loi-A et la loi µ

In the second step, another digit 1 or 0 is generated, depending on whether


the sample is in the upper or the lower half of the subinterval in which it
has been located. This process continues until the last binary digit in the
code generated.
Decoding is the inverse of encoding. In this case, each of the n digits is
applied to a resistor of dierent value. The kth digit is applied to a resistor
2k R. The currents in all the resistors are added. The sum is proportional to
the quantized sample value.
For example, a binary code word 10010110 will give a curent proportional
to 27 +0+0+24 +0+22 +21 +0 = 150. This completes the D/A conversion.

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 15 / 37
Transmission bandwidth and the output SNR

For a binary PCM, we assign a distinct group of n binary digits (bits) to


each of the L quantization levels. Because a sequence of n binary digits
can be arranged in 2n distinct patterns,

L = 2n or n = log2 (L) (5)


Each quantized samples is, rhus, encoded into n bits. Because a signal
m(t) band-limited to B Hz requires a minimum of 2B samples per second,
we require a total 2nB bits per second (bps), that is 2nB pieces of
information per second.
Because a unit bandwidth (1Hz) can transmit a maximum of two pieces of
information per second, we require a minimum channel of bandwidth BT
Hz, given by

BT = nBHz (6)
This is the theoretical minimum transmission bandwidth required to
transmit the PCM signal. In practical reasons we may use a transmission
bandwidth higher than this.
Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and
19 DPCM
mai 2017 16 / 37
Transmission bandwidth and the output SNR

Example

A signal m(t) band-limited to 3kHz is sampled at a a rate 33 13 % higher


than the Nyquist rate. The maximum acceptable error in the sample
amplitude (the maximum quantization error) is 0.5% of the peak amplitude
mp . The quantized samples are binary coded.
Find the minimum bandwidth of a channel required to transmit the
encoded binary signal. if 24 such signals are time-division-multiplexed.
Determine the minimum transmission bandwidth required to transmit the
multiplexed signal.
Response :

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 17 / 37
Transmission bandwidth and the output SNR

Example

A signal m(t) band-limited to 3kHz is sampled at a a rate 33 13 % higher


than the Nyquist rate. The maximum acceptable error in the sample
amplitude (the maximum quantization error) is 0.5% of the peak amplitude
mp . The quantized samples are binary coded.
Find the minimum bandwidth of a channel required to transmit the
encoded binary signal. if 24 such signals are time-division-multiplexed.
Determine the minimum transmission bandwidth required to transmit the
multiplexed signal.
Response :
The Nyquist sampling rate is RN = 2 × 3000 = 6000Hz (samples per
second). The actual sampling rate is RA = 6000 + 6000 × 13 = 8000Hz .
The quantization step is ∆ν , and the maximum quantization error is
±∆ν/2, therefore we have
∆ν mp 0.5
2 = L = 100 mp ⇒ L = 200

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 17 / 37
Transmission bandwidth and the output SNR

For binary coding, L must be a power of 2. Hence, the next higher value of
L that is a power of 2 is L = 256.
From the above, we need n = log2 (256) = 8 bits per sample. We require to
transmit a total of C = 8 × 8000 = 64kbits/s. Because we can transmit
up to 2bit/s per hertz of bandwidth, we require a minimum transmission
bandwidth BT = C/2 = 32KHz .
The multiplexed signal has a total of CM = 24 × 64000 = 1.536 Mbit/s,
which requires a minimum of 1.536/2 = 0.768 Mhz of transmission
bandwidth.
 Exponential increase of the output SNR
We have L2 = 22n , and the output SNR can expressed as
S0
= c(2)2N SNR=cxL^2 (7)
N0
where
3m2 (t)
uncompressed case
(
c= m2p (8)
3
(ln(1+µ))2
compressed case
Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and
19 DPCM
mai 2017 18 / 37
Transmission bandwidth and the output SNR

Substutution of Eq.(7) into Eq. (8) yields


S0
= c(2)2BT /B (9)
N0
From Eq.(9) we observe that the SNR increases exponentially with the
transmission bandwidth BT . This trade of SNR with bandwidth is
attractive and comes close to the upper theoretical limit. A small increase
in bandwidth yields a large benet in terms of SNR. This relationship is
clearly seen by rewriting Eq. (9) using the decibel scale as
S0 S0
( )dB = 10log10 ( )
N0 N0
= 10log10 (c(2)2n )
= 10log10 (c) + 2n10log10 (2)
= (α + 6n)dB

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 19 / 37
Transmission bandwidth and the output SNR

Where α = 10log10 (c). This shows that increasing n by 1 (increasing one


bit in the code word) quadruples the output SNR (6-dB increase). Thus, if
we increase n from 8 to 9, the SNR quadruples, but the transmission
bandwidth increases only from 32 to 36kHz (an increase of only 12.5%).
This shows that in PCM, SNR can be controlled by transmission
bandwidth. We shall see later that frequency and phase modulation also do
this. But it requires a doubling of the bandwidth to quadruple the SNR. In
this respect, PCM is strikingly superior to FM or PM.
Exemple 2
A signal m(t) of bandwidth B = 4kHz is transmitted using a binary
companded PCM with µ = 100. Compare the case of L = 64 with the case
of L = 256 from the point of view of transmission bandwidth and the
output SNR.

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 20 / 37
Transmission bandwidth and the output SNR

Response

For L = 64, n = 6, and the transmission bandwidth is nB = 24 kHz.

= (α + 36)dB
S0
N0
3
α = 10log [ln(101)]2 = −8.51

Hence,
S0
N0 = 27.49dB

For L = 256, n = 8, and the transmission bandwidth is 32 kHz,


S0
N0 = α + 6n = 39.49dB

The dierence between the two SN Rs is 12dB, which is a ratio of 16.


Thus, the SNR for L = 256 is 16 times the SNR for L = 64. The former
requiers just about 33% more bandwidth compared to the latter.
Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and
19 DPCM
mai 2017 21 / 37
Transmission bandwidth and the output SNR

The commutators represented in the following gure are not mechanical


but are high-speed electronic switching circuits.
Sampling is done by electronic gates (such as a bridge diode circuit, as
shown in gure) opened periodically by narrow pulses of 2 − µs duration.
The 1.544 − M bit/s signal of the T1 system is called digital signal level
1 (DS1) which is used further to multiplex into progressively higher level
signals DS2, DS3 and DS4 (as seeing in SDH).
After the Bell System introduced the T1 carrier system in the United
States, dozens of variations were proposed or adopted elsewhere before the
CCITT standardized its 30-channel PCM system with a rate of 2.048
M bit/s (in contrast to T1, with 24 channels and 1.544 M bit/s)

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 22 / 37
Transmission bandwidth and the output SNR

Figure  T1 carrier system

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 23 / 37
Transmission bandwidth and the output SNR

The 30-channel system is used all over the world, except in North America
and Japan. Because of the widespread adoption of the T1 carrier system in
the United States and Japan before the CCITT standardization, the two
standards continue to be used in dierent parts of the world, with
appropriate interfaces in international communication.
Synchronizing and signaling
Binary code words corresponding to samples of each of the 24 channels are
multiplexed in a sequence, as shown in gure below.
A segment containing one code word (corresponding to one sample) from
each of the 24 channels is called a frame. Each frame has 24 × 8 = 192
information bits.
Because the sampling rate is 8000 samples per second, each frame takes
125µs. At the receiver, it is necessary to be sure where each frame begins
in order to separate information bits correctly. For this purpose, a framing
bit is added at the beginning of each frame.
This makes a total of 193 bits per frame. Framing bits are chosen so that a
sequence of framing bits, one at the beginning of each frame, forms a
special pattern that is unlikely to be formed in speech signal.
Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and
19 DPCM
mai 2017 24 / 37
Transmission bandwidth and the output SNR

The sequence formed by the rst bit from each frame is examined by the
logic of the receiving terminal. If this sequence does not follow the given
coded pattern (framing bit pattern), then a synchronization loss is
detected, and the next position is examined to determine whether it is
actually the framing bit. It takes about 0.4 to 6ms to detect and about
50ms (in the worst possible case) to reframe.
In addition to information and framing bits, we need to transmit signaling
bits corresponding to dialing pulses, as well as telephone on-hook/o-hook
signals.
when channels developed by this system are used to transmit signals
between telephone switching systems, the switches must be able to
communicate with each other to use the channels eectively.
Since all eight bits are now used for transmission instead of the seven bits
used in the earlier version.

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 25 / 37
Transmission bandwidth and the output SNR

Figure  T1 system signaling format

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 26 / 37
Dierential Pulse Code Modulation (DPCM)

In analog messages we can make a good guess about a sample value from a
knowledge of the past sample values.
In other words, the sample values are not independent, and generally there
is a great deal of redundancy in the Nyquist samples. Proper exploitation of
this redundancy leads to encoding a signal with a lesser number of bits.
Consider a simple scheme where instead of transmitting the sample value,
we transmit the dierence between the successive sample values.
Thus, if m(k) is the kth sample, instead of transmitting m(k), we transmit
the dierence d(k) = m(k) − m(k − 1).
At the receiver, knowing d(k) and the previous sample value m(k − 1) we
can reconstruct m(k). Thus, from the knowledge of the dierence d(k), we
can reconstruct m(k) iteratively at the receiver.
Now, the dierence between successive samples is generally much smaller
than the sample values. Thus, the peak amplitude mp of the transmitted
values is reduced considerably.

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 27 / 37
Dierential Pulse Code Modulation (DPCM)

Because the quantization interval ∆ν = mp /L, for a given L (or n), this
reduces the quantization interval ∆ν , thus reducing the quantization noise,
which is given by ∆ν 2 /12.
This means that for a given n (or transmission bandwidth), we can increase
the SN R, or for a given SN R, we can reduce n (or transmission
bandwidth).
We can improve upon this scheme by estimating (predicting) the value of
the kth sample m(k) from a knowledge of the previous sample values. If
this estimate is m(k)
b , then we transmit the dierence (prediction error)
b .
d(k) = m(k) − m(k)
At the receiver also, we determine the estimate m(k)b from the previous
sample values, and then generate m(k) by adding the received d(k) to the
estimate m(k)
b . Thus, we reconstruct the samples at the receiver iteratively.
If our prediction is worth its salt, the predicted (estimated) value m(k)
b will
be close to m(k), and their dierence (prediction error) d(k) will be even
smaller than the dierence between the successive samples.
Consequently, this scheme, known as the dierential PCM (DPCM).
Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and
19 DPCM
mai 2017 28 / 37
Dierential Pulse Code Modulation (DPCM)

Before describing DPCM, we shall briey discus the approach to signal


prediction (estimation). To an uninitiated, future prediction seems a
mysterious stu t only for psychics ...etc.
Consider, for example, a signal m(t) which has derivatives of all orders at t.
Using the Taylor series for this signal, we can express m(t + Ts ) as
Ts2
m(t + Ts ) = m(t) + Ts ṁ(t) + m̈(t) + ...
2!
≈ m(t) + Ts ṁ(t) for smallTs (10)
This equation shows that from a knowledge of the signal and its derivatives
at instant t, we can predict a future signal value at t + Ts . In fact, even if
we know the rst derivative, we can still predict this value approximately, as
shown in the last equation.
Let us denote the kth sample of m(t) by m(k), that is, m(kTs ) = m(k),
and m(kTs ± Ts ) = m(k ± 1), and so on.

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 29 / 37
Dierential Pulse Code Modulation (DPCM)

Setting t = kTs in equation above, and recognizing that


ṁ(kTs ) ≈ [m(kTs ) − m(kTs − Ts )]/Ts , we obtain
m(k) − m(k − 1)
m(k + 1) ≈ m(k) + Ts ( ) (11)
Ts
= 2m(k) − m(k − 1) (12)
The last equation (Eq. 11) shows that we can a crude (brut) prediction of
the (k + 1)th sample from the two previous samples. The approximation in
Eq.(10) improves as we add more terms in the series on the right-hand
side. To determine the higher order derivatives in the series, we require
more samples in the past. The larger number of past samples we use, the
better will be the prediction. So, in general, we can express the prediction
formula as
m(k) ≈ a1 m(k − 1) + a2 m(k − 2) + ... + aN m(k − N ) (13)
The right-hand side is m(k)
b , represent the predicted value of m(k). Thus,
m(k)
b = a1 m(k − 1) + a2 m(k − 2) + ... + aN m(k − N ) (14)
Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and
19 DPCM
mai 2017 30 / 37
Dierential Pulse Code Modulation (DPCM)

This is the equation (eq.14) of an N th-order predictor. Larger N would


result in better prediction in general. The output of this lter (predictor) is
b , the predicted value of m(k).
m(k)
The input is the previous samples m(k − 1) and the output is m(k) b . The
input is the previous samples m(k − 1), m(k − 2), ..., m(k − N ). Observe
that this equation reduces to m(k)
b = m(k − 1) for the rst-order predictor.
It follows from Eq. (11), where we retain only the rst term on the
right-hand side. This means that a1 = 1 and the rst-order predictor is a
simple time delay.
We have outlined here a very simple procedure for predictor design. In a
more sophisticated approach, in which we use the minimum mean squared
error criterion for best prediction, the prediction coecients aj in Eq.(14)
are determined from the statistical correlation between various samples.
The predictor described in Eq.(14)is called a linear predictor. It is
basically a transversal lter (a tapped delay line), where the tap gains are
set equal to the prediction coecients, as shown in gure below.

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 31 / 37
Dierential Pulse Code Modulation (DPCM)

Figure  Transversal lter (tapped delay line) used as a linear predictor

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 32 / 37
Analysis of DPCM

As mentioned earlier, in DPCM we transmit not the present sample m(t),


but d(k) (the dierence between m(k) and its predicted value m(k) b ). At
the receiver, we generate m(k)
b from the past sample values to the received
d(k) is added to generate m(k).
There is, however, one diculty in this scheme. At the receiver, instead of
the past samples m(k − 1), m(k − 2), ..., as well as d(k), we have their
quantized versions mq (k − 1), mq (q − 2),.... Hence, we cannot determine
b .
m(k)
We can only determine m cq (k), the estimate of the quantized sample
mq (k), in terms of the quantized samples mq (k − 1), mq (k − 2), .... This
will increase the error in reconstruction. In such a case, a better strategy is
to determine m cq (k), te estimate of mq (k) (instead of m(k)), at the
transmitter also from the quantized samples mq (k − 1), mq (k − 2), ....
The dierence d(k) = m(k) − m cq (k) is now transmitted using PCM. At
the receiver, we can generate m cq (k), and from the received d(k), we can
reconstruct mq (k). The gure below shows a DPCM transmitter.

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 33 / 37
Analysis of DPCM

Figure  DPCM sstem. (a) Transmitter, (b) Receiver.


Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and
19 DPCM
mai 2017 34 / 37
Analysis of DPCM

We shall soon show that the predictor input is mq (k). Naturally, its output
is m
cq (k), the predicted value of mq (k). The dierence

d(k) = m(k) − m
cq (k) (15)
is quantized to yield
d(k) = m(k) − m
cq (k) (16)
where q(k) is the quantization error. The predictor output m
cq (k) is fed fed
back to its input so that the predictor input mq (k) is
mq (k) = m
cq (k) + dq (k)
= m(k) − d(k) + dq (k)
= m(k) + q(k) (17)

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 35 / 37
Analysis of DPCM

This shows that mq (k) is a quantized version of m(k). The predictor input
is indeed mq (k), as assumed. The quantized signal dq (k) is now
transmitted over the channel.
The receiver shown in the gure (b) above is identical to the shaded
portion of the transmitter. The inputs in both cases are also the same, viz.
(à savoir), dq (k). Therefore, the predictor output must be m
cq (k) (the same
as the predictor output at the transmitter).
Hence, the receiver output (which is the predictor input) is also the same,
viz., mq (k) = m(k) + q(k), as found in Eq.(17). This shows that we are
able to receive the desired signal m(k) plus the quantization noise q(t),
This is the quantization noise associated with the dierence signal d(k),
which is generally much smaller than m(k).
The received samples are decoded and passed through a low pass lter for
D/A conversion.

Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and


19 DPCM
mai 2017 36 / 37
Analysis of DPCM

SNR Improvement(Amélioration du bruit)

To determine the improvement in DPCM aver PCM, let mp and dp the


peak amplitudes of m(t) and d(t), respectively. If we use the same value of
L in both cases, the quantization step ∆ν in DPCM is reduced by factor
dp /mp .
Because the quantization noise power is (∆ν)2 /12, the quantization noise
in DPCM reduces by factor (mp /dp )2 , and the SNR increases by the same
factor. Moreover, the power is proportional to its peak value squred
(assuming other statistical properties invariant). Therefor, Gp (SNR
improvement due to prediction) is
Pm
Gp = (18)
Pd
where Pm and Pd are the powers of m(t) and d(t), respectively. In terms of
dB units, this means that the SNR increases by 10log10 (Pm /Pd )dB.
Therefore, Eq. (10) applies to DPCM also with a value of α that is higher
by 10log10 (Pm /Pd )dB.
Pr Said SAFI USMSFPBM Communication numérique 2 (M9) Chapter 3 : PCM and
19 DPCM
mai 2017 37 / 37