Differential Pulse Code Modulation (DPCM)

• When a voice / video signal is sampled at a rate slightly higher than the Nyquist rate, the
sampled signal is found to exhibit high degree of correlation between adjacent samples.

• As a result, the values of near by samples will be close to each other

• In standard PCM systems, these highly correlated samples when encoded contains lot of
redundant information.

• Removing the redundancy before encoding, we can obtain an efficient coded signal, i.e. the
number of bits to be sent can be reduced → Basic Idea Behind DPCM
How does DPCM achieve this ?

➢ By predicting the message sample 𝑚[𝑛] and then ‘encoding the difference between
the actual value and the predicted value’ Hence the name Differential PCM

Prediction algorithms are available which can predict an estimate of a future value of the
message signal if it knows the past behavior up to that point in time

1
Assume 𝑚(𝑡) is the input message signal sampled at 𝑓𝑠 = to produce the sequence
𝑇𝑠
{𝑚[𝑛]} whose samples are 𝑇𝑠 seconds apart
• The quantizer in DPCM quantizes the difference between the message sample
and its predicted value
• The input to the quantizer is given by Prediction of input
sample 𝑚[𝑛]
Prediction error 𝑒[𝑛] = 𝑚 𝑛 − 𝑚
ෝ 𝑛
• The quantizer output is given by
Quantization error
𝑒𝑞 𝑛 = 𝑒 𝑛 + 𝑞 𝑛

• This quantizer output is added to the predicted value 𝑚

ෝ 𝑛 to produce the prediction
filter input
𝑚𝑞 𝑛 = 𝑚
ෝ 𝑛 + 𝑒𝑞 𝑛 = 𝑚
ෝ 𝑛 + 𝑒 𝑛 + 𝑞[𝑛]
• But since 𝑚 𝑛 = 𝑒 𝑛 + 𝑚
ෝ 𝑛 , we have

𝑚𝑞 𝑛 = 𝑚 𝑛 + 𝑞[𝑛]

which represents a quantized version of the message sample 𝑚 𝑛

If the prediction is good ,
• 𝑒[𝑛] is likely to be small and hence it’s variance would be much smaller compared to
the variance of the input message signal
• DPCM would require fewer number of bits per sample compared to standard PCM

BW requirement of DPCM < BW requirement of PCM

• At the receiver if we can reproduce 𝑚𝑞 𝑛 , it will be close to 𝑚 𝑛 since we know
that 𝑚𝑞 𝑛 = 𝑚 𝑛 + 𝑞[𝑛] and good prediction would ensure that 𝑞[𝑛] will be
small

Reconstruction formula: 𝑚𝑞 𝑛 = 𝑚
ෝ 𝑛 + 𝑒𝑞 𝑛
Waveform Coding Techniques

1) Pulse Code Modulation or PCM

• Standard method for conversion of speech and video signals to digital form
• Robust to noise but demanding in bandwidth and computational requirements
2) Differential Pulse Code Modulation or DPCM
• A method for reducing bandwidth
• Highly computationally complex
3) Delta Modulation or DM
• Simple to implement
• Requires high bandwidth
Delta Modulation (DM)

• Has lower system complexity compared to standard PCM

• Transmission bandwidth is traded off for reduced system complexity
• DM exploits bandwidth-complexity tradeoff

• In DM, the incoming message signal 𝑚(𝑡) is oversampled (at a rate higher than
Nyquist rate) so that adjacent samples of the signal would be highly correlated or will
have values that are close in range
• This is done to simplify the quantization process
• DM allows to quantize each sample by using only a single bit
• DM produces a staircase approximation to the original message signal
• The difference between the input and the approximation can take only two values ±∆
corresponding to positive and negative differences
• If the approximation falls below the signal at any sampling instant, it is increased by ∆ and
if it lies above the signal, it is decreased by ∆
• If the signal does not change too much from sample to sample, the staircase
approximation always lies within ±∆ of the input signal
 Error signal 𝑒[𝑛] = 𝑚 𝑛 − 𝑚𝑞 𝑛 − 1
where 𝑚 𝑛 is the sample at 𝑛𝑇𝑠 , 𝑚𝑞 𝑛 − 1 is the latest approximation and 𝑒[𝑛] is the
error signal representing the difference between the two

+∆ 𝑖𝑓 𝑒 𝑛 > 0
Quantized version of 𝑒[𝑛] 𝑒𝑞 [𝑛] = ∆ sgn 𝑒 𝑛 =ቊ
−∆ 𝑖𝑓 𝑒 𝑛 < 0
Approximation at 𝑛𝑇𝑠 𝑚𝑞 [𝑛] = 𝑚𝑞 [𝑛 − 1] + 𝑒𝑞 [𝑛]

Approximation at (𝑛 − 1)𝑇𝑠

𝑚𝑞 (𝑡) is the staircase approximated signal

At the output of the encoder, for every +∆ , symbol 1 is sent out and for every −∆, symbol
0 is sent out
• Instead of quantizing the actual sample values, the difference between the actual
message sample and the staircase approximation is quantized
• Hence there are only two possible output representation levels, i.e. ±∆ which
requires only a one-bit quantizer

𝑒[𝑛] = 𝑚 𝑛 − 𝑚𝑞 𝑛 − 1
𝑒𝑞 [𝑛] = ∆ sgn 𝑒[𝑛]
𝑚𝑞 [𝑛] = 𝑚𝑞 [𝑛 − 1]+ 𝑒𝑞 [𝑛]
• Sampled message signal is applied to a modulator containing a comparator, quantizer

and accumulator

• 𝑧 −1 represents a unit delay, i.e., a delay equal to one sampling period

• One bit quantizer quantizes the error signal to ±∆

• Encoder encodes +∆ by symbol 1 and for every −∆ with symbol 0

• Bandwidth requirement of DM is very much higher than standard PCM since the input

signal is sampled at a frequency much higher than the Nyquist frequency and hence

more number of samples per second and hence more number of bits per second
 𝑚𝑞 [𝑛] = 𝑚𝑞 [𝑛 − 1]+ 𝑒𝑞 [𝑛]
= 𝑚𝑞 [𝑛 − 2]+ 𝑒𝑞 [𝑛-1]+𝑒𝑞 [𝑛]
=σ𝑛𝑖=1 𝑒𝑞 [i]

• To reconstruct the approximated value 𝑚𝑞 [𝑛], the decoded signal (𝑒𝑞 [𝑛] ) has to be
continuously added

At the output of the accumulator , the staircase approximation 𝑚𝑞 (𝑡) is reconstructed and the
sharp edges are smoothened out by passing through a LPF
DM is a special case of DPCM with two important differences:
1. Use of one bit quantizer in DM
2. Replacing the prediction filter by a delay element
“ DM is the one-bit version of DPCM”
Delta modulation is subject to 2 types of Quantization errors:
1. Slope Overload Distortion
2. Granular Noise

1. Slope Overload Distortion

When the step-size ∆ is too small and the signal 𝑚(𝑡) changes too fast, i.e., rises
too fast or falls too fast, then the staircase approximation may fall short of the
original signal and may not catch up to it
For overcoming slope overload, the step size ∆ should be chosen such that
Maximum slope
Slope of
of message signal
staircase
approximation

Otherwise ∆ is too small for the approximation to follow a steep segment of the
input waveform 𝑚(𝑡) resulting in 𝑚𝑞 (𝑡) falling behind 𝑚 𝑡 . This condition is called
as slope overload distortion
2. Granular Noise

Occurs when step size ∆ is too large relative to the slope of the input message signal
𝑚(𝑡) , i.e. the signal is relatively flat
This causes the staircase approximation to hunt around the flat segment of the input
waveform
Small step size Slope overload error
Large step size Granular noise

• We need to choose a large enough step size to avoid slope overload distortion/error
and a small enough step size to avoid granular noise
• It requires a delta modulator that is “adaptive” in the sense that the step size is made
to vary/adapt in accordance to the input signal
• We go for Adaptive Delta Modulation (ADM)
• The one we studied just now is called as linear delta modulation to differentiate
between the two
Q)

A sinusoidal voice signal g(t)=cos(6000πt) is to be transmitted using either PCM or DM.

The sampling rate for PCM is 8kHz and for transmission with DM, the step size is decided
to be 31.25mV. Slope overload distortion is to be avoided in DM.

Assuming that the number of quantization levels for PCM is 64, determine the bit rate.

Which scheme is to be chosen for a bandwidth constrained application ? DM or PCM ?