Associate Prof. Tamer M.

Barakat
Electronics & Communications Dept.
3/28/2023 Associate Prof.Tamer M. Barakat 2
  Experiments shows that real speech has pdf
close to Laplacian pdf:
 This types of signals need a quantizer with
non-linear (non-uniform) characteristics which
can be achieved using the compandig
technique.

3/28/2023 Associate Prof.Tamer M. Barakat 3

  For most types of signals such as
speech/video, a linear quantizer is not the
optimum choice in the sense of minimizing
mean square error.
 The linear quantizer provides minimum
distortion only for signals with a uniform pdf.
 Speech signals, exhibit non-uniform statistics
with a smaller amplitudes are more likely than
larger amplitudes.

3/28/2023 Associate Prof.Tamer M. Barakat 4

 The most amount of the information is lost because of using the
uniform quantizer.
3/28/2023 Associate Prof.Tamer M. Barakat 5
  An alternative approach is to divide the input
amplitude range into non-uniform steps by
increasing the number of quantization steps in
the region around zero and correspondingly
decreasing the number around extremes of the
input range.
 The result of this non-uniform code is an input
/ output characteristics that is a stair case with
N steps of unequal width.

3/28/2023 Associate Prof.Tamer M. Barakat 6

  It used to minimize the mean square error:
S
𝑒2 ↓ ↑
N

3/28/2023 Associate Prof.Tamer M. Barakat 7

 ∆𝑦 = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 = 𝑞
∆𝑥 = ∅ 𝑥 = 𝑘𝑥
∆𝑦 𝑞 𝐴
= =
∆𝑥 𝑘𝑥 𝑥
𝐴
∆𝑦 = ∆𝑥
𝑥
1
𝑦= 𝐴‫׬‬ 𝑑𝑥 = 𝐴𝑙𝑛 𝑥 + 𝑐 [logarithmic]
𝑥
 This means that the non-uniform quantizer can
be achieved using the logarithmic compressor.

3/28/2023 Associate Prof.Tamer M. Barakat 8

  The non-uniform quantizer can be achieved by
first compressing the samples of the input
signal and then linearly quantizing the
compressed signals.
 At the receiver, a linear decoder is followed by
an expander that provides the invers
characteristic of the compressor…… this
technique is called companding.

3/28/2023 Associate Prof.Tamer M. Barakat 9

  The non-uniform quantizer can be achieved by
compressor and uniform quantizer.

3/28/2023 Associate Prof.Tamer M. Barakat 10

 So, the compressor used to amplify the low amplitude in the signal and
attenuate the high amplitudes

3/28/2023 Associate Prof.Tamer M. Barakat 11

  Logarithmic companding is used for speech signals.
 A logarithmic compressor curve widely used for speech
digitization is:
1. 𝝁 − 𝑳𝒐𝒘 Compander: (North-American & Japan)

3/28/2023 Associate Prof.Tamer M. Barakat 12

 𝒙
𝒍𝒐𝒈𝒆 [𝟏 + 𝝁 ]
𝒙𝒎𝒂𝒙
𝒚 = 𝒄 𝒙 = 𝒚𝒎𝒂𝒙 𝒔𝒈𝒏(𝒙)
𝒍𝒐𝒈𝒆 (𝟏 + 𝝁)

 Where: 𝒔𝒈𝒏(𝒙) is the polarity of x

 𝝁 is a parameter indicating the amount of

companding (compression/expansion) used.
3/28/2023 Associate Prof.Tamer M. Barakat 13
 2. A-Low Compander: [Europe-CCITT]

𝑦=𝑐 𝑥
𝑥
𝐴[ ] 1
𝑥𝑚𝑎𝑥
𝑦𝑚𝑎𝑥 𝑠𝑔𝑛 𝑥 0≤𝑥≤
(1 + 𝑙𝑜𝑔𝑒 𝐴) 𝐴
=
𝐴𝑥
1 + 𝑙𝑜𝑔𝑒 1
𝑥𝑚𝑎𝑥
𝑦𝑚𝑎𝑥 𝑠𝑔𝑛 𝑥 ≤𝑥≤1
(1 + 𝑙𝑜𝑔𝑒 𝐴) 𝐴

3/28/2023 Associate Prof.Tamer M. Barakat 14

  In practice, logarithmic companding laws are
approximated by non-linear devices such as diodes or
by piecewise linear segment

3/28/2023 Associate Prof.Tamer M. Barakat 15

  As shown in the fig., the piecewise linear
approximation using 15-segments, 𝜇 = 255 PCM.
 For this 𝜇-law curve, there are eight segments on each
side of zero.
 The encoding of a 8 bits -𝐵1 𝑀𝑆𝐵 𝑡ℎ𝑟𝑜𝑢𝑔ℎ 𝐵2 𝑙𝑆𝐵 of
each PCM word is accomplished as follows:

3/28/2023 Associate Prof.Tamer M. Barakat 16

  Ex: Assume a 8 bit PCM and 𝑊= 4kHz, find
the PCM bit rate?
 Solution:

Sampling rate= 2 𝑊=8 kHz

PCM rate = 𝑛𝑓𝑠 = 64 𝑘𝑏/𝑠

3/28/2023 Associate Prof.Tamer M. Barakat 17

3/28/2023 Associate Prof.Tamer M. Barakat 18
  Assume that:
1. M bit /sample
2. R = transmission rate = N𝑓𝑠 sample/sec
3. R = M N𝑓𝑠 bit /sec
𝑹
4. 𝑩𝑾 = = 𝑴𝑵𝒇𝒎 for TDM-PCM
𝟐

Note that: for TDM-PAM:

𝑩𝑾 = 𝑵𝒇𝒎 , where R = N𝑓𝑠

3/28/2023 Associate Prof.Tamer M. Barakat 19

  It is known as 30-channels PCM system.
 Audio 𝐵𝑊 = 𝑓𝑚 = 4 𝑘𝐻𝑧.
𝑠𝑎𝑚𝑝𝑙𝑒𝑠
 Sampling frequency = 𝑓𝑠 = 8 𝑘
𝑠𝑒𝑐
 A – law compander
 Sample rate = 8 bit/sample
 30 audio channels.
 32 time slots
 Multi-frame = 16 frames.
 Transmission rate = 8 ∗ 8 × 103 ∗ 32 = 2.048 𝑀 𝑏𝑖𝑡/𝑠𝑒𝑐
 Signaling on the time slot 16 by a rate = 2 Kb/sec

3/28/2023 Associate Prof.Tamer M. Barakat 20

  In this system, each frame is made-up of a set of 32 channels time
slots numbered from 0 to 31.
 Each channel time slot comprises 8-digit time slots.
 Each channel time slot is sampled at rate of 8000 samples /sec.
1
 The complete frame (frame length) repeats every = 125 𝜇 𝑠𝑒𝑐.
8000
 The comutator must operate by this rate to transmit all coming
samples from all channels.
125
 Each channel time slot duration is = 3.9 𝜇 𝑠𝑒𝑐.
32
3.9
 Each digit time slot duration is = = 0.488 𝜇 𝑠𝑒𝑐.
8
 The gross digit rate generated by the system is = 8000 × 32 × 8 =
2.048 𝑀𝑏/𝑠𝑒𝑐

3/28/2023 Associate Prof.Tamer M. Barakat 21

3/28/2023 Associate Prof.Tamer M. Barakat 22
  24 channels.
 8 bit /channel

 Frame length = 125 𝜇 𝑠𝑒𝑐

 There is no Multi-frame.

 Channel rate = 1.544 Mb/sec

 Alignment frame = bit No. 1

 No. of bits /frame = 8 * 24 +1 (frame sync.)

= 193 bit/frame
Signaling, on the bit No. (8) for each user and it will be
sent every 6 frames.

3/28/2023 Associate Prof.Tamer M. Barakat 23

3/28/2023 Associate Prof.Tamer M. Barakat 24
3/28/2023 Associate Prof.Tamer M. Barakat 25