Dr.

Ali Hussein Muqaibel

Ver. 3.1

Dr. Ali Muqaibel 1

 Introduction
•Pulse coded
PCM Modulation

 DPCM & Motivation DPCM •Differential

PCM

◦ Predictive DPCM Predictive

◦ How to predict? DPCM

◦ SNR Improvement ADPCM

•Adaptive
DPCM

◦ Adaptive DPCM (ADPCM) DM •Delta

Modulation

 Delta Modulation (DM) also ADM •Adaptive

 Bandwidth for PCM ADM DM

Dr. Ali Muqaibel 2

 When a signal is highly oversampled (sampled at several times the
Nyquist rate) , the signal does not change a lot from one sample to
another.
 Differential: quantize the difference between consecutive samples
instead of the original signal.
 Why?
◦ A quantizer with much less number of bits (less information to transmit) or 𝑛↓
◦ A quantizer with the same number of bits but much smaller quantization intervals ∆𝜈 ↓
(less quantization noise and much higher SNR).

Dr. Ali Muqaibel 3

 For 𝑥(𝑡) are sampled at a rate much higher than the Nyquist rate
𝑑 𝑘 =𝑥 𝑘 −𝑥 𝑘−1
 𝑑[𝑘] will be less than the range of values of 𝑥[𝑘]
 𝑑𝑞 [𝑘] quantized difference

+ d[k]
x[k] + Quantizer dq[k]

–
x[k-1]
Delay Ts

Dr. Ali Muqaibel 4

 +
dq[k] + xˆ [k ]

+
xˆ [k − 1] Delay Ts

 Approximation of 𝑥[𝑘] using

𝑥ො 𝑘 = 𝑑𝑞 𝑘 + 𝑥[𝑘
ො − 1].
 The advantage?
2
∆𝑣 2 𝑚𝑝 2𝑚𝑝
 Quantization noise 𝑃𝑞 = = 2 , ∆𝑣 = , by reducing 𝑚𝑝 , we can reduce
12 3𝐿 𝐿
the quantization distortion.
 The Cost: the reconstructed signal may suffer from one or two possible
problems.
1) Effect of error propagation.

2) Accumulation of quantization noise.

Dr. Ali Muqaibel 5

The quantizer used to quantize 𝑑[𝑘] is an 8–level quantizer with quantization
intervals [–4,–3), [–3,–2), [–2,–1), … , [3,4) and the output quantization
levels are the center points in each interval (–3.5, –2.5, –1.5, … , 3.5).

–1 0 1 2 3 4 5 6 7 8 9
0 0.3 1.5 0.7 1.0 2.3 3.7 2.8 3.5 2.8 0

0 0 0.3 1.5 0.7 1.0 2.3 3.7 2.8 3.5 2.8

0 0.3 1.2 –0.8 0.3 1.3 1.4 –0.9 0.7 –0.7 –2.8
Quantization U(or
U U U U U U U D U U
Up/Down D)

0.5 0.5 1.5 –0.5 0.5 1.5 1.5 –0.5 0.5 –0.5 –2.5

0 0.5 1.0 2.5 2.0 2.5 4.0 5.5 5.0 5.5 5.0

0.5 1.0 2.5 2.0 2.5 4.0 5.5 5.0 5.5 5.0 2.5

0.5 0.7 1.0 1.3 1.5 1.7 1.8 2.2 2.0 2.2 2.5

Err. Direction
U U U U U U U U D U U
Up/Down

𝑑 𝑘 =𝑥 𝑘 −𝑥 𝑘−1 Note that the error between the original and

reconstructed samples always increased except when
𝑥ො 𝑘 = 𝑑𝑞 𝑘 + 𝑥[𝑘
ො − 1] the quantization error switched direction at k = 7 (the
shaded box).
Dr. Ali Muqaibel 6
 Estimate (predict) the value of the 𝑘 𝑡ℎ sample 𝑚[𝑘] from knowledge of the previous
sample values. Let this estimate be 𝑚[𝑘],
ෝ then we transmit the difference (iteratively),
then we transmit the difference (perdition error)
𝑑 𝑘 = 𝑚 𝑘 − 𝑚[𝑘]
ෝ
 At the receiver, first predict and then add the difference (iteratively)
 DPCM is a special case of predictive DPCM where
𝑚[𝑘]
ෝ = 𝑚[𝑘 − 1]
 Example: linear
 (first order prediction 𝑚ෝ 𝑘 = 𝑚 𝑘 − 1 + (𝑚 𝑘 − 1 − 𝑚 𝑘 − 2 )

𝑘 –1 0 1 2 3 4 5 6 7 8 9
𝑚[𝑘] 0 0.3 1.5 1.7 1.9 2.3 2.8 3.2 3.5 3.1 2.8

𝑚
ෝ [𝑘] 0 0 0.6 2.7 1.9 2.1 2.7 3.3 3.6 3.8 2.7

𝑑[𝑘] 0 0.3 0.9 -1 0 0.2 0.1 –0.1 -0.1 –0.7 +0.1

Dr. Ali Muqaibel 7

 We can express the next sample as (Taylor Series)
𝑇𝑠2 𝑇𝑠3
𝑚 𝑡 + 𝑇𝑠 = 𝑚 𝑡 + 𝑇𝑠 𝑚ሶ 𝑡 + 𝑚ሷ 𝑡 + 𝑚
ഺ 𝑡 + ⋯ . ≈ 𝑚 𝑡 + 𝑇𝑠 𝑚ሶ 𝑡
2! 3!
 A full knowledge of 𝑚(𝑡) and its derivatives allow for future prediction
𝑚 𝑘𝑇𝑠 − 𝑚 𝑘𝑇𝑠 − 𝑇𝑠
𝑚ሶ 𝑘𝑇𝑠 ≈
𝑇𝑠
𝑚 𝑘 −𝑚 𝑘−1
𝑚 𝑘 + 1 ≈ 𝑚 𝑘 + 𝑇𝑠 = 2𝑚 𝑘 − 𝑚 𝑘 − 1
𝑇𝑠
 For higher order prediction, we require more past samples
𝑚
ෝ 𝑘 = 𝑎1 𝑚 𝑘 − 1 + 𝑎2 𝑚 𝑘 − 2 + ⋯ + 𝑎𝑁 𝑚[𝑘 − 𝑁]
 𝑁 𝑡ℎ order prediction Equation
 Larger 𝑁 ,better prediction

Dr. Ali Muqaibel 8

 𝑁 𝑡ℎ order prediction Equation
𝑚ෝ 𝑘 = 𝑎1 𝑚 𝑘 − 1 + 𝑎2 𝑚 𝑘 − 2 + ⋯ + 𝑎𝑁 𝑚 𝑘 − 𝑁
 𝑎1 = 1, 𝑁 = 1: First order prediction is a simple time delay (DPCM , 𝑚 ෝ 𝑘 =𝑚 𝑘−1 )
 Prediction coefficient (Statistical correlation between different samples)
 Linear Prediction: transversal filter ( a tapped delay line) where the tap gains=prediction
coefficients.

Dr. Ali Muqaibel 9

 𝑞𝑘
𝑚𝑘 𝑑𝑘 𝑑𝑞 𝑘

 We cannot determine 𝑚 ෝ 𝑘 at the receiver because

all we have is quantized samples. A solution is to 𝑚
ෝ𝑞 𝑘
DPCM System
find the difference with the quantized samples at the (a) Transmitter
(b) Receiver
transmitter.
 𝑑 𝑘 =𝑚 𝑘 −𝑚 ෝ𝑞 𝑘 𝑑𝑞 𝑘
 𝑑𝑞 𝑘 = 𝑑 𝑘 + 𝑞 𝑘
𝑚
ෝ𝑞 𝑘
 At the receiver
 𝑚𝑞 𝑘 = 𝑚 ෝ 𝑞 𝑘 +𝑑𝑞 𝑘
 𝑚𝑞 𝑘 = 𝑚 𝑘 − 𝑑 𝑘 + 𝑑𝑞 𝑘 = 𝑚 𝑘 + 𝑞[𝑘] • What if we do not include the quantizer in the
prediction at the transmitter?
• 𝑑 𝑘 =𝑚 𝑘 −𝑚 ෝ 𝑘
• 𝑑𝑞 𝑘 = 𝑑 𝑘 + 𝑞 𝑘
• At the receiver
• 𝑚𝑞 𝑘 = 𝑚 ෝ 𝑞 𝑘 +𝑑𝑞 𝑘
• 𝑚𝑞 𝑘 =? +𝑑𝑞 𝑘 = 𝑚 𝑘 +? +𝑞[𝑘]

Dr. Ali Muqaibel 10

 SNR improvement, 𝐺𝑝
𝑃𝑚
𝐺𝑝 =
𝑃𝑑
𝑃𝑚 and 𝑃𝑑 are the powers of 𝑚(𝑡) and 𝑑(𝑡)
 In practice, SNR improvements may be as high as 25dB (voice and low
activity images)
 Alternatively for the same SNR, the bit rate (DPCM) can be reduced
(compared with PCM) by 3 to 4 bits per sample.
◦ Thus, telephone systems using DPCM can often operate at 32 or even 24kbits/sec
(instead of 64kbits/s)

Dr. Ali Muqaibel 11

 Adaptivly change ∆𝑣
 𝑑𝑞 [𝑘] can be compared to ∆𝑣 or 0 , to see if we need to increase or
decrease the current ∆𝑣.
 8bit PCM may be reduced to 4bit ADPCM at the same sampling
rate.
 ITU-T standard G.726. used by Microsoft in WAVE players

Dr. Ali Muqaibel 12

 A loss of frame synchronization in PCM destroys everything!
 Delta Modulator (DM) is special case of the DPCM which uses a 2–level quantizer.
𝑑 𝑘 = 𝑚 𝑘 − 𝑚[𝑘 − 1]
 A lowpass filter after the accumulator will smooth the reconstructed signal.
 The sampling in the delta modulator must be performed at a very high rate (in many cases
sampling at 30 to 40 times the Nyquist rate is required). Why? To increase the correlation.

2-Level
Quantizer Sampler
(Comparator) Frequency fs 𝑘

+ g(t) g[k] gq(t)

𝑚
ෝ 𝑞 𝑘 = ෍ 𝑑𝑞 [𝑘]
x(k) + gq[k] 𝑚=0

–
xˆq (t )
gq[k] LPF xˆ (t )
xˆq (t )
Accumulator
(Integrator)

Accumulator
(Integrator)

Dr. Ali Muqaibel 13

 ෝ 𝑞 𝑡 is a stair case approximation of 𝑚(𝑡).
𝑚
 + pulse 𝑚 𝑡 > 𝑚 ෝ 𝑞 (𝑡).
 - pulse 𝑚 𝑡 < 𝑚 ෝ 𝑞 (𝑡).
 Threshold of coding (Step) and overloading:
◦ Variation in 𝑚(𝑡) smaller than the step value
(threshold of coding) are lost is DM (Granular Error)
◦ If 𝑚ሶ 𝑡 is too high, 𝑚
ෝ 𝑞 𝑡 cannot follow 𝑚(𝑡). (Slope
overload)
◦ Adaptive Step , Adaptive DM (ADM)
◦ Detect the pattern and decide how to adapt
accordingly
 (++++, or ----) increase step Granular Error
 +-+-+-+ decrease step
Slope overload

Dr. Ali Muqaibel 14

 In DM each digit has equal importance.
 In DM multiplexing each channel requires its own decoder while for
PCM one decoder for all (On the other hand, permit flexibility and
avoid cross talk (no need for stringent multiplexing design)
 DM outperforms at low SNR while PCM outperforms at high SNR.
 DM is very simple, inexpensive, and no need for framing bits (less
overhead)
 DM is suitable for voice & TV signals.

Dr. Ali Muqaibel 15

 Bandwidth required for binary baseband= 𝑅/2.
 Bandwidth required for binary passband= 𝑅.

 Example: For a given communication channel with bandwidth

1𝑀𝐻𝑧, how many voice signals (4𝑘𝐻𝑧, 64𝑘𝑏𝑖𝑡/sec ) can we
transmit? Consider AM and digital passband.
1𝑀𝐻𝑧
◦ For AM: 1𝑀𝐻𝑧/ 2𝐵 = = 125
8𝑘𝐻𝑧
1𝑀𝐻𝑧
◦ For digital passband= = 15.625
64𝑘𝐻𝑧
◦ We can see the cost for going digital.

Dr. Ali Muqaibel 16