ENGR 4323/5323

Digital and Analog Communication

Ch 6
Sampling and Analog-to-Digital Conversion

Engineering and Physics

University of Central Oklahoma
Dr. Mohamed Bingabr
 Chapter Outline

• Sampling Theorem
• Pulse Code Modulation (PCM)
• Digital Telephony: PCM IN T1 Carrier Systems
• Digital Multiplexing
• Differential Pulse Code Modulation (DPCM)
• Adaptive Differential PCM (ADPCM)
• Delta Modulation
• Vocoders and Video Compression
2
 Sampling Theorem

Sampling is the first step in converting a continuous signal to a

digital signal.

Sampling Theorem determines the minimum number of

samples needed to reconstruct perfectly the continuous signal
again from its samples.

Sampling Theorem: A continuous function x(t) bandlimited to B

Hz can be reconstructed from its samples if it was sampled at
rate equal or greater than 2 B samples per second. If the
sampling rate equals 2 B then it is called the Nyquist rate. 3
 Sampling Theorem

Sampling Rate
fs= 1/Ts

Nyquist Rate

fs = 2B

fs > 2B 4
 Sampling Theorem
𝛿  𝑇 ( 𝑡 )=∑ 𝛿 (𝑡 −𝑛 𝑇 𝑠)
𝑠
𝑛

 ´𝑔 ( 𝑡 ) =𝑔 (𝑡 ) 𝛿 ( 𝑡 )=∑ 𝑔 (𝑛 𝑇 𝑠)𝛿 (𝑡 −𝑛 𝑇 𝑠)

𝑇𝑠
𝑛
𝑛=∞
  1 𝑗𝑛 𝜔 𝑡
𝛿 𝑇 ( 𝑡 )= ∑ 𝑒 𝑠

𝑠
𝑇 𝑠 𝑛=− ∞
𝑛=∞
  1 𝑗𝑛𝜔 𝑡
´𝑔 ( 𝑡 ) =𝑔 (𝑡 ) 𝛿 𝑇 ( 𝑡 )= ∑ 𝑔 (𝑡 )𝑒 𝑠

𝑠
𝑇 𝑠 𝑛=− ∞
Use the frequency shifting property to find the spectrum of the
sampled signal 𝑛 =∞
  1
𝐺´ ( 𝑓 )= ∑ 𝐺(𝑓 −𝑛 𝑓 𝑠 )
𝑇 𝑠 𝑛=−∞ 5
 Reconstruction from Uniform Samples
To reconstruct the continuous signal g(t) from the samples, pass
the samples through a low-pass filter with cutoff frequency =B Hz.

 𝐻 ( 𝑓 ) =𝑇 𝑠 Π 𝜔
( 4𝜋 𝐵 )
h
  ( 𝑡 ) =2 𝐵 𝑇 𝑠 𝑠𝑖𝑛𝑐 (2 𝜋 𝐵𝑡 )

Sampling at Nyquist rate: 2BTs = 1

h
  ( 𝑡 ) =𝑠𝑖𝑛𝑐(2 𝜋 𝐵𝑡 ) h(t)
 ´𝑔 ( 𝑡 ) =  ´𝑔 ( 𝑡 ) LPF  𝑔 ( 𝑡 )
∑ 𝑔(𝑛𝑇 𝑠 )𝛿 (𝑡 −𝑛𝑇 𝑠 )
𝑛
 𝑔 ( 𝑡 ) =∑ 𝑔 ( 𝑘 𝑇 ) h(𝑡 − 𝑘𝑇 )=∑ 𝑔 ( 𝑘𝑇 ) 𝑠𝑖𝑛𝑐 [ 2𝜋 𝐵 (𝑡 −𝑘 𝑇 ) ]
𝑠 𝑠 𝑠 𝑠
𝑘 𝑘
6
 Reconstruction from Uniform Samples
 𝑔 ( 𝑡 ) = ∑ 𝑔 ( 𝑘 𝑇 𝑠 ) 𝑠𝑖𝑛𝑐 [ 2𝜋 𝐵(𝑡 − 𝑘 𝑇 𝑠)] Interpolation formula
𝑘

 𝑔 ( 𝑡 ) = ∑ 𝑔 ( 𝑘 𝑇 𝑠 ) 𝑠𝑖𝑛𝑐 [ 2𝜋 𝐵𝑡− 𝑘 𝜋 ]
𝑘 7
 Example 6.1

Find a signal g(t) that is band-limited to B Hz and whose

samples are g(0) = 1 and g(±Ts) = g(±2Ts) = g(±3Ts) =…= 0 where
the sampling interval Ts is the Nyquist interval, that is Ts = 1/2B.

8
 Practical Signal Reconstruction

~ 𝑔 ( 𝑡 )=∑ 𝑔 ( 𝑛𝑇 ) 𝑝 ( 𝑡 −𝑛𝑇 )=¿ 𝑝 ( 𝑡 ) ∗ ∑ 𝑔 ( 𝑛𝑇 𝑠 ) 𝛿 ( 𝑡 −𝑛𝑇 𝑠 ) ¿

𝑛
𝑠 𝑠
[ 𝑛 ]
~
 𝑔 ( 𝑡 )= 𝑝 ( 𝑡 ) ∗ 𝑔
´ (𝑡 )

 
~ 1
𝐺 ( 𝑓 ) =𝑃(𝑓 ) ∑ 𝐺(𝑓 − 𝑛 𝑓 𝑠)
𝑇𝑠 𝑛
  To recover g(t) from we pass it through an equalizer E( f )
9
 Practical Signal Reconstruction

  ~ 1
𝐺 ( 𝑓 )=𝐸 ( 𝑓 ) 𝐺 ( 𝑓 ) =𝐸(𝑓 )𝑃(𝑓 ) ∑ 𝐺(𝑓 −𝑛 𝑓 𝑠 )
𝑇𝑠 𝑛
  𝑇 𝑠|𝑓 |≤ 𝐵

{
𝐸( 𝑓 ) 𝑃( 𝑓 )= Flexible       𝐵< ¿ 𝑓 ∨¿(1/ 𝑇 𝑠 − 𝐵)
0|𝑓 |> 𝑓 𝑠 − 𝐵

The equalizer filter E( f ) must be low-pass in nature to stop all

frequency content above fs - B, and it should be the inverse of
10
P( f ) within the signal bandwidth of B Hz.
Practical Signal Reconstruction-Example

p(t)

  𝑡 −0.5 𝑇 𝑝
𝑝(𝑡 )=Π
(𝑇𝑝 )  𝑃( 𝑓 )=𝑇 𝑝 𝑠𝑖𝑛𝑐 ( 𝜋 𝑓 𝑇 𝑝 ) 𝑒 − 𝑗 𝜋 𝑓 𝑇 𝑝

 
~ 1
𝐺 ( 𝑓 ) =𝑃(𝑓 ) ∑ 𝐺(𝑓 − 𝑛 𝑓 𝑠)
𝑇𝑠 𝑛

  𝜋 𝑓 𝑇𝑠
𝐸 ( 𝑓 )=𝑇 𝑠 . ≈ When Tp is very small
𝑠𝑖𝑛 ( 𝜋 𝑓 𝑇 𝑝 ) 𝑇 𝑝
11
 Practical Issues in Sampling
Sampling at the Nyquist rate require ideal low-pass filter which is
unrealizable in practice.

fs=2B

fs > 2B

12
Practical Issues in Sampling

Aliasing
Practical signals are time-limited by
nature which means they can not be
band-limited at the same time.

13
 Maximum Information Rate
A maximum of 2B independent pieces of information per second
can be transmitted, error free, over a noiseless channel of
bandwidth B Hz.

14
 Nonideal Practical Sampling Analysis
Read details in textbook section 6.1.4

15
 Sampling Theorem and Pulse Modulation
The continuous signal g(t) is sampled, and sample values are
used to modify certain parameters (amplitude, width, position) of
a periodic pulse train.

Techniques for communication using pulse modulation:

1- Pulse Amplitude Modulation (PAM)
2- Pulse Width Modulation (PWM)
3- Pulse Position Modulation (PPM)
4- Pulse Code Modulation (PCM)
5- Time Division Multiplexing (TDM)
16
 Pulse Modulation

TDM

PAM

PWM

PPM
17
 Pulse Code Modulation (PCM)
PCM is widely used as a tool to convert analog signal to digital
signal.

The signal range

[-mp , mp] is divided
into L subintervals
(Δv = 2mp/L).
L levels require n
binary digits (bits)
where L = 2n. 18
 Telephone Vs. Music
Phone conversation for 5 minutes
1- Bandwidth 3500 Hz
2- sampling rate = 8000 samples/sec
3- number of samples 8 bits/sample

CD music recording
1- Bandwidth 20,000 Hz
2- sampling rate = 44,100 samples/sec
3- number of samples 16 bits/sample

Compare the channel bandwidth required to transmit speech

vs. music.
Compare the storage capacity required to store 5 minutes
phone conversation vs. 5 minutes music. 19
 Advantage of Digital Communication
1) Withstand channel noise and distortion much better than
analog.
2) With regenerative repeater it is possible to transmit over long
distance.
3) Digital hardware implementation is flexible
4) Digital coding provide further error reduction, high fidelity and
privacy.
5) Easier to multiplex several digital signals.
6) More efficient in exchanging SNR for bandwidth.
7) Digital signal storage is relatively easy and inexpensive.
8) Reproduction with digital messages is reliable without
deterioration.
9) Cost of digital hardware is cheaper and continue to decrease.
20
 Quantization Error Analysis

Original signal 𝑚
  (𝑡 )=∑ 𝑚 ( 𝑘 𝑇 ) 𝑠𝑖𝑛𝑐 ( 2 𝜋 𝐵𝑡 −𝑘 𝜋 )
𝑠
𝑘

Quantized signal 𝑚
 ^ (𝑡 )=∑ 𝑚
^ ( 𝑘 𝑇 𝑠) 𝑠𝑖𝑛𝑐 ( 2 𝜋 𝐵𝑡 −𝑘 𝜋 )
𝑘
  ( 𝑡 )=
Quantization noise 𝑞 ∑ [ 𝑚^ ( 𝑘 𝑇 𝑠) − 𝑚 ( 𝑘 𝑇 𝑠 ) ] 𝑠𝑖𝑛𝑐 ( 2 𝜋 𝐵𝑡 − 𝑘 𝜋 )
𝑘
𝑞  ( 𝑡 )=∑ 𝑞 ( 𝑘 𝑇 𝑠 ) 𝑠𝑖𝑛𝑐 ( 2𝜋 𝐵𝑡 − 𝑘 𝜋 )
𝑘
  𝑇 /2
~2 1 2
𝑞 (𝑡)= lim ∫ ∑ 𝑞 ( 𝑘 𝑇 𝑠 ) 𝑠𝑖𝑛𝑐 ( 2 𝜋 𝐵𝑡 − 𝑘 𝜋 ) 𝑑𝑡
Power of q(t)
𝑇 → ∞ 𝑇 − 𝑇 /2 [ 𝑘
21
]
 Quantization Error Analysis
 𝑇 /2 0𝑚≠𝑛
∫ 𝑠𝑖𝑛𝑐 ( 2 𝜋 𝐵𝑡 − 𝑚 𝜋 ) 𝑠𝑖𝑛𝑐 ( 2 𝜋 𝐵𝑡 −𝑛 𝜋 ) 𝑑𝑡 = 1 𝑚 =𝑛
−𝑇 / 2
2𝐵
{
  𝑇/2
~ 2 1 2 2
𝑞 (𝑡)= lim ∑ 𝑞 ( 𝑘 𝑇 𝑠 ) ∫ 𝑠𝑖𝑛𝑐 ( 2 𝜋 𝐵𝑡 −𝑘 𝜋 ) 𝑑𝑡
𝑇 →∞ 𝑇 𝑘 −𝑇 / 2
~
  2 1 2 2BT: number of samples over
𝑞 (𝑡)= lim ∑ 𝑞 ( 𝑘 𝑇 𝑠)
𝑇 → ∞ 2 𝐵𝑇 𝑘 averaging interval T
  ∆𝑣 /2 2 2
~ 2 1 2 (∆ 𝑣) 𝑚 𝑝
Mean square
𝑞= ∫ 𝑞 𝑑𝑞= = 2 quantization error
∆ 𝑣 −∆ 𝑣 /2 12 3 𝐿
~ 2
 𝑆 0 2 𝑚 (𝑡 )
Signal to Noise Ration (SNR) =3 𝐿 22
𝑁0 𝑚2𝑝
 Nonuniform Quantization
Nonuniform quantization reduces the quantization error by
reducing the quantization level where the signal is more frequently
exist (at low amplitude).
The µ-law (North America and Japan)
µ is the compression parameter

 𝑦= 1 𝜇𝑚
 0 ≤ 𝑚 ≤ 1
ln ⁡(1+𝜇) (
ln 1+
𝑚𝑝 ) 𝑚𝑝

The A-law (rest of the world)

  𝐴 𝑚  0 ≤ 𝑚 ≤ 1
𝑦=
{ 1+ln ⁡𝐴 𝑚 𝑝
1
(
1+ln ⁡𝐴
1+ ln
( )
𝐴𝑚
𝑚𝑝 ) 𝐴 𝑚𝑝
𝑚𝑝 𝐴
 1 ≤ 𝑚 ≤ 1
23
 Nonuniform Quantization
For L=128 (7-bit), µ = 100
For L=256 (8-bit), µ = 255

SNR for µ-law

𝑆
  0= 3 𝐿
2

𝑁 0 [ ln (1+ μ ) ]2

 2 𝑚 2𝑝
𝜇 ≫~2
𝑚 (𝑡 )

24
 Transmission Bandwidth
B : Signal bandwidth in Hz
L : Quantization Level
n : Number of bits per sample
fs : Samples per second
BT: Transmission Channel Bandwidth

fs = 2*B
n = log2 L
Number of bits per second = 2*B*n
BT = Number of bits per second /2;
BT = B*n

Usually the sampling rate is higher than the Nyquest rate 2B

to improve signal to noise ratio (SNR). 25
 Example
A signal m(t) band-limited to 3 kHz is sampled at a rate
33.333% 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
quantization 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.

26
 Channel Bandwidth and SNR
Output SNR increase exponentially with the transmission
bandwidth BT.

𝑐=¿
 
~ 2
 𝑆 0 𝑚 (𝑡 )
=3 𝐿2 2
𝑁0 𝑚𝑝

 𝑆 𝑜 =𝑐(2)2 𝑛  𝑆 𝑜 = 𝑐(2)2 𝐵 𝑇
/𝐵
𝑁𝑜 𝑁𝑜

  𝑆𝑜 𝑆𝑜   𝑆𝑜
( )
𝑁𝑜 𝑑𝐵
=10 𝑙𝑜𝑔10 ( )
𝑁𝑜 ( )
𝑁𝑜 𝑑𝐵
= ( 𝛼 + 6 𝑛 ) dB

Increasing n by 1 (increasing one bit in the codeword)

quadrables the output SNR (6 dB increase).
27
 Digital Telephony: PCM in T1 Carrier
Systems
T1 Specifications
24 Channels
Regenerative
Sampling: 2 µs pulse repeater every
Rate: 1.544 Mbit/s 6000 feet

DS1: digital signal level 1

ITU-T Specifications
30 Channels
Sampling: 2 µs pulse
Rate: 2.048 Mbit/s

28
 Synchronizing and Signaling

100011011100

125 μs
193 bits
0.4-0.6 ms to detect
50 ms to reframe

29
 Synchronizing and Signaling

8000 samples/sec
24 channels
1 frame bit
193 bits/frame
125 µs/frame
The framing bits pattern: 100011011100 (12 frame)
0.4 to 6 msec for frame detection
Up to 50 ms to reframe.
LSB of every sixth sample used for switching communication
(robbed-bit signaling).
30
Read the detail of frame signaling in textbook
 Digital Multiplexing (DM)
Digital interleaving
Word interleaving
Overhead bits
(synchronization)

Time division multiplexing of digital signals:

(a) digit interleaving;
(b) word (or byte) interleaving;
(c) interleaving channel having different bit rate;
(d) alternate scheme for (c).

31
North America Digital Hierarchy (AT&T)

32
 Signal Format DM 1/2
subframe

The F digits are periodic 01010101

Bit interleaving F digits identify the frames
4 channels The M digits 0111 identify subframes
Channel rate1.544 Mbit/s The C digits for bit stuffing and
Overhead frames are M,C,F Asynchronous
Each frame consist of four subframe
Each subframe has six overhead bits
Each subframe has six 48-interleaved data bits
The 48 interleaved data bits from the four channels (12 bits/ch).
Each frame has 1152 data bits (48*6*4) and 24 overhead bits(6*4).
33
Efficiency =1152/1176=98%
 Europe ITU System

34
Plesiochronous digital hierarchy (PDH) according to ITU-T Recommendation G.704.
 Differential Pulse Code Modulation
(DPCM)
DPCM exploits the characteristics of the source signals. It
reduce the number of bits needed per sample by taking
advantage of the redundancy between adjacent samples.
Instead of transmitting sample m[k] we transmit

d[k] = m[k] – m[k-1]

d[k] has lower amplitude so it require less bits per sample or

the size of quantization level will be smaller if we keep the
number of bits unchanged which reduces the quantization error.

  can improve the DPCM by estimating the kth value from

We
previous values and then transmit the difference
 𝑑 [ 𝑘 ] =𝑚 [ 𝑘 ] − 𝑚
^ [𝑘 ]
35
 Differential Pulse Code Modulation
(DPCM)
Linear predictor

Taylor Series
 𝑚 ( 𝑡 +𝑇 )=𝑚 ( 𝑡 )+𝑇 𝑚 ˙(𝑡 )+ 𝑇 2𝑠 𝑇 3𝑠
𝑠 𝑠 !
𝑚
¨ (𝑡 ) + ! ⃛ ( 𝑡 )+ … ≈ 𝑚 ( 𝑡 ) +𝑇 𝑠 𝑚 ˙(𝑡 )
𝑚
2 3

 𝑚 [ 𝑘 +1 ] ≈ 𝑚 [ 𝑘 ] +𝑇 𝑚 [ 𝑘 ] − 𝑚[ 𝑘 −1 ] For small Ts
𝑠
[ 𝑇𝑠 ]
 𝑚 [ 𝑘 +1 ] ≈ m [ k ] +(m [ k ] − m [ k − 1 ] ) ≈ 2 m [ k ] − m [ k − 1 ]
 
36
 Analysis of DPCM

Linear predictor
 𝑑 [ 𝑘 ] =𝑚 [ 𝑘 ] − 𝑚
^ 𝑞 [𝑘 ]

 𝑑 𝑞 [ 𝑘 ] = 𝑑 [ 𝑘 ] +𝑞 [ 𝑘 ]
 𝑚 𝑞 [ 𝑘 ] =𝑚
^ 𝑞 [ 𝑘 ] +𝑑 𝑞 [ 𝑘 ]

 𝑚 𝑞 [ 𝑘 ] =𝑚 [ 𝑘 ] − 𝑑 [ 𝑘 ]+𝑑 𝑞 [ 𝑘 ]
𝑚
  𝑞 [ 𝑘 ] =𝑚 [ 𝑘 ]+ 𝑞 [𝑘 ]
mq[k] is a quantized version of m[k]

DPCM system
37
(a) transmitter (b) receiver
 Adaptive Differential PCM (ADPCM)
Adaptive DPCM further improve the efficiency of DPCM encoding
by incorporating an adaptive quantizer (varied Δv) at the
encoder.
The quantized prediction error dq[k] is a good indicator of the
prediction error size. It can be used to change Δv to minimize
dq[k]. When the dq[k] fluctuate around large positive or negative
value then the prediction error is large and Δv needs to grow and
when dq[k] fluctuates around zero then Δv needs to decrease.
8-bit PCM sequence can be
encoded into a 4-bit ADPCM
sequence at the same sampling
rate. This reduce channel bandwidth
or storage by half with no loss in
quality.
 Delta Modulation (DM)
Delta modulation oversample the baseband signal (4 time the
Nyquist rate) to increase the correlation between adjacent
samples. The increase in correlation results in a small
prediction error that can be encoded using only one bit (L=2).
In DM the information of the difference between successive
samples is transmitted by a 1-bit code word.

 𝑚 𝑞 [ 𝑘 ] =𝑚 𝑞 [ 𝑘 − 1 ] +𝑑 𝑞 [ 𝑘 ]
 𝑚 𝑞 [ 𝑘 − 1 ] =𝑚 𝑞 [ 𝑘 − 2 ] +𝑑 𝑞 [ 𝑘 −1 ]

 𝑚 𝑞 [ 𝑘 ] =𝑚 𝑞 [ 𝑘 −2 ] +𝑑 𝑞 [ 𝑘 ] +𝑑 𝑞 [ 𝑘 − 1 ]
𝑘
 
𝑚𝑞 [ 𝑘 ] = ∑ 𝑑 𝑞 [ 𝑚 ]
𝑚= 0 0 1 2 3 4 2 1 -1 1
 Delta Modulator and Demodulator

a) Delta modulation
b) Delta demodulators
c) Message signal versus integrator
output signal
d) Delta-modulated pulse trains
e) Modulation errors
Threshold of Coding and overloading
1- small step size E causes slope
overload
2- Large step size (E) causes
granular noise.
|  𝑚 ˙(𝑡 )|< 𝐸 / 𝑇 𝑠 |  𝑚 ˙(𝑡 )|< 𝐸 𝑓 𝑠
|  𝑚 ˙(𝑡 )|< 𝐸 𝑓 𝑠

If m(t) = A cos ωt

 [ 𝐴 𝐸𝑓 𝑠
𝑚𝑎𝑥 ] 𝑣𝑜𝑖𝑐𝑒 ≅
𝜔
41
 Linear Prediction Coding (LPC) Vocoders

 𝐻 ( 𝑧 ) = 𝑔
𝐴 (𝑧)
𝑝 −1
 
(
𝐻 ( 𝑧 ) =𝑔 . 1− ∑ 𝑎 𝑖 𝑧
𝑖=1
−𝑖
)
The human speech production mechanism.

Table 6.1 Typical pressure impulses.

 Linear Prediction Coding (LPC) Vocoders

𝑝 −1
  𝑔
𝐻 (𝑧 )=
𝐴 (𝑧 ) (
=𝑔 . 1 − ∑ 𝑎𝑖 𝑧
𝑖=1
−𝑖
)
The LPC analyzer
- Estimate the all-pole filter coefficients in A(z).
- The optimum filter coefficients are determined by minimizing
the mean square error (MSE) of the linear prediction error.
Video Compression
 Video Compression
Pixel intensity Pixel intensity-128

DCT Coefficients Quantization Table

 Video Compression

Quantized DCT Coefficients

Video Compression