EEE543 – LECTURE 2 PART III

Pulse Code
Modulation - Part III

1
OUTLINE
 Differential Pulse Code Modulation
 Adaptive DPCM

 Delta Modulation

 Adaptive Delta Modulation

2
DIFFERENTIAL PULSE CODE MODULATION (DPCM)
 PCM is not a very efficient system because it generates so
many bits and requires so much bandwidth to transmit.
 DPCM exploits the characteristics of the signals to improve
the efficiency of PCM
 In analog messages the sample values are not independent
we can make a good guess about a sample value from
knowledge of past sample values
 There is a great deal of redundancy in the Nyquist samples.

 Proper exploitation of this redundancy leads to encoding a

signal with fewer bits.

3
DPCM CONT’D
 Consider a simple scheme; instead of transmitting the
sample values, we transmit the difference between the
successive sample values
 For the k-th sample 𝑚[𝑘] we transmit the difference
𝑑 𝑘 = 𝑚 𝑘 − 𝑚[𝑘 − 1]
 At the receiver, knowing 𝑑 𝑘 and several previous sample
value 𝑚[𝑘 − 1], we can reconstruct 𝑚 𝑘 iteratively
 The difference between successive samples is generally
much smaller than the sample values – the peak amplitude
𝑚𝑝 of the transmitted values is reduced considerably
𝑚𝑝
 The reduction in 𝑚𝑝 reduces the quantization step, Δ𝜈 =
𝐿
Δ𝜈 2
which in turns reduces the quantization noise 𝑞𝑒 = for a
𝐿
given L
 This means that for a given number of bits per sample n (or
transmission bandwidth) we can increase the SNR similarly 4

for a given SNR we can reduce the transmission bandwidth

DPCM CONT’D
 We can improve upon this scheme by estimating
(predicting) the value of the kth sample 𝑚 𝑘 from a
knowledge of several previous sample values
 If this estimate is 𝑚
ෝ 𝑘 , then we transmit the difference
(prediction error) 𝑑 𝑘 = 𝑚 𝑘 − 𝑚[𝑘]
ෝ
 At the receiver also, we determine the estimate 𝑚[𝑘]
ෝ from
the previous sample values, and then generate 𝑚 𝑘 by
adding the received 𝑑 𝑘 to the estimate 𝑚[𝑘]
ෝ
 In this way we reconstruct the samples at the receiver
iteratively
 The predicted (estimated) value 𝑚[𝑘]
ෝ will be close to 𝑚 𝑘 ,
and their difference (prediction error) 𝑑 𝑘 will be even
smaller than the difference between the successive samples.
 This scheme, known as the differential PCM (DPCM)

 To predict future samples, we use Taylor Series described

5
next
DPCM CONT’D
 Consider, for example, a signal m(t) ,which has derivatives of
all orders at t.
 If we know the signal and its derivatives at instant t, we can
predict a future signal value at 𝑡 + 𝑇𝑠
 This is possible even if we know just the first derivative

 The future signal 𝑚(𝑡 + 𝑇𝑠 ) can be obtained by the he Taylor

series as follows

6
DPCM CONT’D
 For the discrete samples we can use the Taylor series to
predict the future samples
 Let the kth sample of m(t) be given by: 𝑚 𝑘 = 𝑚 𝑘𝑇𝑠 then
𝑚 𝑘𝑇𝑠 ± 𝑇𝑠 = 𝑚 𝑘 ± 1 and so on
 If w set 𝑡 = 𝑘𝑇𝑠 and also noting that
𝑚 𝑘𝑇𝑠 − 𝑚 𝑘𝑇𝑠 − 𝑇𝑠
𝑚ሶ 𝑘𝑇𝑠 =
𝑇𝑠
 Then we can find 𝑚 𝑘 ± 1 as follows
𝑇𝑠 𝑚 𝑘 − 𝑚 𝑘 − 1
𝑚 𝑘+1 ≈𝑚 𝑘 +
𝑇𝑠
= 2𝑚 𝑘 − 𝑚 𝑘 − 1
 This shows that we can find a crude prediction of the (k + 1 )th
sample from the two previous samples
 The approximation improves as we add more terms in the
series on the right-hand side.
7
DPCM CONT’D
 To determine the higher order derivatives in the series, we
require more samples in the past
 We can express the prediction formula as

 The right-hand side is 𝑚

ෝ 𝑘 , the predicted value of 𝑚 𝑘

 This is the equation of an Nth-order predictor with larger N

would resulting in better prediction
 The input

 consists of the previous samples m[k - 1 ] , m[k - 2] , . . . , m[k -

N ] . although it is customary to say that the input is m[k]
and the output is 𝑚 ෝ 𝑘
8
ANALYSIS OF DPCM
 Recall: In DPCM we do not transmit the present sample 𝑚 𝑘 ,
but 𝑑 𝑘 (the difference between 𝑚 𝑘 and its predicted value
𝑚[𝑘]
ෝ
 At the receiver, we generate 𝑚[𝑘]
ෝ from the past sample
values to which the received 𝑑 𝑘 is added to generate 𝑚 𝑘
 One difficulty associated with this scheme is that 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 𝑘 − 2 , . . . .Hence, we cannot determine 𝑚 𝑘
 The quantized samples 𝑚𝑞 𝑘 − 1 , 𝑚𝑞 𝑘 − 2 , … can only
determine 𝑚 ෝ 𝑞 𝑘 , the predicted version of 𝑚𝑞 𝑘 - This will
increase the error in reconstruction
 A better strategy is to determine 𝑚 ෝ 𝑞 𝑘 , the estimate of 𝑚𝑞 𝑘
(instead of 𝑚 𝑘 ), at the transmitter, from the quantized
samples 𝑚𝑞 𝑘 − 1 , 𝑚𝑞 𝑘 − 2 9
ANALYSIS OF DPCM CONT’D
 The difference 𝑑 𝑘 = 𝑚 𝑘 − 𝑚 ෝ 𝑞 𝑘 is now transmitted via
PCM.
 At the receiver, we can generate 𝑚
ෝ 𝑞 𝑘 and from the received
𝑑 𝑘 we can reconstruct 𝑚𝑞 𝑘 .
 In the DPCM transmitter, the predictor input is 𝑚𝑞 𝑘 , its output is
the predicted value of 𝑚
ෝ𝑞 𝑘
 The difference 𝑑 𝑘 = 𝑚 𝑘 − 𝑚ෝ 𝑞 𝑘 is quantized to yield
𝑑𝑞 𝑘 = 𝑑 𝑘 + 𝑞[𝑘]
 Where 𝑞[𝑘] is the quantization error
 The predictor output 𝑚
ෝ 𝑞 𝑘 is fed back to its input so that the
predictor input 𝑚𝑞 𝑘 is

10
ANALYSIS OF DPCM CONT’D
 The figure below shows a DPCM transmitter

Figure 2.1 a)

 The transmitter transmits the quantized signal 𝑑𝑞 𝑘 over the

channel

11
ANALYSIS OF DPCM CONT’D
 The receiver (figure 2.1b) is the same as the shaded region of the
transmitter

Figure 2.1 b)
 The inputs and predictor output in both cases are the same also,
namely, 𝑑𝑞 𝑘 and 𝑚ෝ 𝑞 𝑘 respectively
 From the receiver output is 𝑚𝑞 𝑘 = 𝑚 𝑘 + 𝑞 𝑘 ,we can obtain the
desired signal 𝑚 𝑘 plus the quantization noise 𝑞 𝑘
 The quantization noise 𝑞 𝑘 is associated with the difference signal
𝑑 𝑘 which is generally much smaller than 𝑚 𝑘 12
ANALYSIS OF DPCM CONT’D
 To determine the improvement in DPCM over PCM, let 𝑚𝑝 and 𝑑𝑝
be the peak amplitudes of m(t) and d(t) , respectively.
 If we use the same value of L in both cases, the quantization step Δ𝜈
in DPCM is reduced by the factor 𝑑𝑝 /𝑚𝑝 .
 The quantization noise power is Δ𝜈 2 Τ12 the quantization noise in
2
DPCM is reduced by the factor 𝑚𝑝 Τ𝑑𝑝 and the SNR is increased
by the same factor
 The signal power is proportional to its peak value squared
(assuming other statistical properties invariant), the SNR
improvement due to prediction, 𝐺𝑝 , is at least
𝑃𝑚
𝐺𝑝 =
𝑃𝑑
 Where 𝑃𝑚 and 𝑃𝑑 are the powers of m(t) and d(t) , respectively.

 Notes:
 For the same SNR, the bit rate for DPCM could be lower than that for PCM
by 3 to 4 bits per sample. Thus, telephone systems using DPCM can often
operate at 32 or even 24 kbit/s. 13
ADAPTIVE DPCM
 Adaptive DPCM (ADPCM) can further improve the efficiency of
DPCM encoding by incorporating an adaptive quantizer at the
encoder.

Figure 2.2
 For practical reasons, the number of quantization level L is fixed

 When a fixed quantization step Δ𝜈 is applied, either

 The quantization error is too large because Δ𝜈 is too big or
 The quantizer cannot cover the necessary signal range when Δ𝜈 is too small

 The quantization step is made adaptive either large or small 14

depending the prediction error size

ADAPTIVE DPCM CONT’D
 The quantized prediction error 𝑑𝑞 𝑘 can be a good indicator
of the prediction error size
 When the quantized prediction error samples vary close to
the largest positive value ( or the largest negative value), it
indicates that the prediction error is large and Δ𝜈 needs to
grow
 Conversely, if the quantized samples oscillate near zero, then
the prediction error is small and Δ𝜈 needs to decrease.
 It is important that both the modulator and the receiver have
access to the same quantized samples.
 The adaptive quantizer and the receiver reconstruction can
apply the same algorithm to adjust the Δ𝜈 identically
 Compared with DPCM, ADPCM can further compress the
number of bits needed for a signal waveform
For example:
 It is very common in practice for an 8-bit PCM sequence to be
encoded into a 4-bit AD PCM sequence at the same sampling 15
rate. This easily represents a 2:1 bandwidth or storage
reduction with virtually no loss.
DELTA MODULATION (DM)
 Delta Modulation (DM) is a subclass of differential pulse code
modulation, can be viewed as simplified variant of DPCM, in which
1-bit quantizer is used with fixed first order predictor it was
developed for voice telephony applications.
 Principle of DM : DM output is 0 if waveform fall in value, 1 represents rise
in value, each bit indicates direction in which signal is changing (not how
much), i.e. DM codes the direction of differences in signal amplitude
instead of value of difference (DPCM).

 DM provides a staircase approximation to the over sampled version

of an input base band signal.
 The difference between the input and the approximation is quantized into
only two levels, namely, ±𝛿 corresponding to positive and negative
differences, respectively.

16
DM CONT’D
 Once the quantisation operation is performed, transmission
of the signal can be achieved by sending a zero for a negative
transition, and a one for a positive transition.

 Note that this means that the quantised signal must change
at each sampling point.

 The demodulator for a delta-modulated signal is simply a

staircase generator. If a one is received, the staircase
increments positively, and if a zero is received, negatively.
This is usually followed by a lowpass filter.

 The key to using delta modulation is to make the right choice

of step size and sampling period
17
DM CONT’D
 Below is a DM block

Figure 2.3

18
DM CONT’D
 The figure below illustrate the DM system and the input output
signals

Figure 2.4
19
DM CONT’D
 The input to the comparator is 𝑒 𝑡 = 𝑚 𝑡 − 𝑚෥ 𝑡
where 𝑚 𝑡 is the message signal and 𝑚 ෥ 𝑡 is a reference
signal

 Theanalog signal is quantized by a one-bit ADC (a

comparator implemented as a comparator)

 Thecomparator output is converted back to an analog signal

with a 1-bit DAC, and subtracted from the input after passing
through an integrator

 The shape of the analog signal is transmitted as follows: a "1"

indicates that a positive excursion has occurred since the last
sample, and a "0" indicates that a negative excursion has
occurred since the last sample. 20
DM CONT’D
 For small Δ, when the input signal changes rapidly (i.e. slope of
input waveform is high), it takes time for the output of the
modulator to follow the input this is called slope overload distortion
 For large Δ, if the input varies slowly, a large quantization noise
occurs known as granular noise

21
Figure 2.6
DM CONT’D

Example: For the figure below determine the transmitted DM

sequence

Figure 2.7
Sol: The transmitted bit train would be 1 1 1 1 0 0 0 1 0 1 1 1 1 1 0 22
DM OUTPUT WAVEFORMS
 The figure below show the original signal, the integrated signal and
the output digital data

Figure 2.8
23
DM LIMITATIONS
 The resulting output suffers from slope overload distortion which
can be reduced by increasing the step size
 When the analog signal has a high rate of change, the DM can “fall
behind” and a distorted output occurs
 This may result to granular noise as you increase the step size

 Has a small dynamic range

 Overcome by the use of adaptive delta modulator

24
Figure 2.9
DM TRADE OFFS
 Simplicity versus Quality
 In order to obtain high quality DM requires very high sampling
rates, typically 20× the highest frequency of interest, as opposed to
Nyquist rate of 2×

25
Figure 2.10
ADAPTIVE DELTA MODULATION (ADM)
 In conventional DM the problem of keeping both quantisation noise
and slope overload noise acceptably low is solved by oversampling
 i.e keeping the DM step size small and sampling at many times the
Nyquist rate
 The penalty incurred is the loss of some, or all, of the saving in
bandwidth which might be expected with DM
 An alternative strategy is to make the DM step size variable,
 Making it larger during periods when slope overload noise would
otherwise dominate – to prevent the step overload
 Making it smaller when quantisation noise might dominate – to prevent
granular noise
 The use of an adaptive delta modulator required that the receiver
be adaptive also, so that the step size at the receiver changes to
match the change in ∆ at the modulator.

26
ADAPTIVE DELTA MODULATION CONT’D
 The figure below shows the performance of adaptive delta
modulation

Figure 2.11

27
ADM CONT’D

Figure 2.12 28
COMPARISON OF PCM AND DM SYSTEMS
 The principal advantage of DM over PCM is the simplicity of its
implementation
 DM has worse SNR compared to PCM
 PCM requires more bandwidth
 In general, PCM exhibits better SNR characteristics at the same data rate.
Good voice reproduction via PCM can be achieved with 128 quantization
levels, or 7-bit coding (27 = 128).
 A voice signal, conservatively, occupies a bandwidth of 4 kHz. Thus,
according to the sampling theorem, samples should be taken at a rate of
8000 samples per second. This implies a data rate of 8000  7 = 56 kbps for
the PCM-encoded digital data.
 PCM is more preferred than DM for analog signals
 Because repeaters are used instead of amplifiers, there is no cumulative
noise
 The conversion to digital signaling allows the use of the more efficient
digital switching techniques.

29
SNR CONSIDERATIONS IN DM SYSTEMS
 Consider a sinusoid signal
𝑚 𝑡 = 𝐴 cos 𝜔𝑚 𝑡
 In DM systems slope overload distortion will occur if
∆ ∆ 𝑓𝑠
𝐴< =
𝜔𝑚 𝑇𝑠 2𝜋 𝑓𝑚
where 𝑓𝑠 = 1/𝑇𝑠

 The signal-to-quantizing-noise ratio in DM system under the

assumption of no slope overload is given by
𝑆 3𝑓𝑠3
𝑆𝑁𝑅 0 = = 2𝑓
𝑁𝑞 8𝜋2 𝑓𝑚 𝑀
0

 The maximum allowable amplitude of the input sinusoid is

∆ ∆
𝐴𝑚𝑎𝑥 = = 𝑓𝑠
𝜔𝑚 𝑇𝑠 𝜔𝑚

30
EXAMPLES
Example 1
 For a sinusoidal modulating signal

𝑚 𝑡 = 𝐴 cos 𝜔𝑚 𝑡 , 𝜔𝑚 = 2𝜋𝑓𝑚

show that the maximum output signal-to-quantizing noise ratio in a

DM system under the assumption of no slope overload is given by

𝑆 3𝑓𝑠3
(𝑆𝑁𝑅)0 = = 2𝑓
𝑁𝑞 8𝜋2 𝑓𝑚 𝑀
0

1
where 𝑓𝑠 = 𝑇 is the sampling rate and 𝑓𝑀 is the cut-off frequency of a
𝑠
low-pass filter at the output of the receiver. For no slope overload
condition we have
∆ ∆ 𝑓 31
𝐴< = (𝑓𝑠 )
𝜔𝑚 𝑇𝑠 2𝜋 𝑚
EXAMPLES CONT’D
Example 2
A DM system is designed to operate at 3 times the Nyquist rate for a
signal with a 3-kHz bandwidth. The quantizing step size is 250mV.

(a) Determine the maximum amplitude of 1-kHz input sinusoid for

which the delta modulator does not show slope overload.
(b) Determine the postfiltered output signal-to-quantizing noise ratio
for the signal in part (a)

32
EXAMPLES CONT’D
 Example 3
 Determine the output SNR in a DM system for a 1-kHz sinusoid,
sampled at 32kHz, without slope overload, and followed by a 4-kHz
postreconstruction filter.
𝑆 3𝑓𝑠3
(𝑆𝑁𝑅)0 = = 2𝑓
𝑁𝑞 8𝜋2 𝑓𝑚 𝑀
0
Example 4
 The data rate of Example 2 is 32 kb/s, which is the same bit rate
obtained by sampling at 8 kHz with 4 b per sample in a PCM
system. Find the average output SNR of a 4-b PCM quantizer for the
sampling of full scale sinusoid with fs = 8 kHz, and compare it with
the results of Example 2

33