Presentation 3

Analog and Digital Communication
Waqas Ahmed
Pakistan Institute of Engineering and Applied Sciences

[email protected]
December 15, 2014
Waqas Ahmed (PIEAS) Short title December 15, 2014 1 / 41

Digital Communication

Principles of Digital Data Transmission
Source: The input to a digital system is in the form of sequence of

digits, such sources are computer, digitized voice signal (PCM,
DPCM) .
Multiplexer: The capacity of practical channel transmitting data is
much larger than the data rate of individual sources. Several sources
are combined to achieve the available capacity effectively.
Line Coder: The output of a multiplexer is coded into electrical
pulses or waveforms for the purpose of transmission over the channel.
The process is called line coding/ transmission coding.

Line Coder(1)
Pulse Code Modulation (PCM) waveforms falls into four categories
Nonreturn to Zero (NRZ)
Return to Zero (RZ)
Phase encoded
Multilevel binary

Line Coder(2)

Line Coder(2)

Line code(3)-Properties
Transmission Bandwidth-As small as possible

Power Efficiency- Transmission Power should be as small as possible,
for a given bandwidth and specified detection error probability
Error Detection and Correction Capability
Favorable Power Spectral Density It is desirable to have zero PSD
at ω = 0, because ac coupling and transformers are used at repeaters.
Adequate Timing Content It should be possible to extract timing or
clock information from the signal
Transparency It should be possible to transmit a digital signal
regardless of the pattern 1’s and 0’s.

PSD of various line codes(1)
The pulses are Tb seconds

apart, the transmission rate
is Rb = 1/TB
The kth pulse in the pulse
train y (t) is ak p(t). The
values ak are arbitrary and
random . The on-off, and
biploar line codes are all
special cases of this pulse
train y (t), where ak take on
values 0, 1, -1 randomly.
ak can be separated from
p(t) by using figure (d).

We need to derive Sx (ω) = F {Rx (τ )}
The time auto-correlation function of the impulse train x(t). This can
be conveniently done by considering the impulses as a limiting form of
the rectangular pulses. Each pulse has a width → 0 and kth pulse
height is hk . The strength of kth impulse is ak
hk = ak
T /2
Z
1
then by definition Rx (τ ) = lim x(t)x(t − τ )dt
T →∞ T
−T /2

Rx (τ ) is an even function τ , we need consider only positive τ ,
consider τ < the area associated with kth pulse is hk2 ( − τ ) and
1 X 2
Rx (τ ) = lim hk ( − τ ) replacing hk by ak /k ,
T →∞ T
k
1 X 2 −τ R0 τ
Rx (τ ) = lim ak , R x (τ ) = lim 1 − ,
T →∞ T 2 T →∞ Tb
k
Tb X 2
where R0 = lim ak
T →∞ T
k
During the averaging interval T(T → ∞) there are N pulses
1 X 2
(N → ∞) where N = fracT Tb and from R0 = lim ak
N→∞ N
k
Summation is over N pulses. Hence, R0 is time average of the square
of pulse amplitude ak .
1 X 2
R0 = lim ak = aek
N→∞ N
k

R0 |τ |
Rx (τ ) = Tb 1− , |τ | <
As we increases τ we find that the kth pulse of x(t − τ ) will start

overlapping the (k+1)th pulse of x(t) as τ approaches Tb . Repeating the
above argument
Tb X
R1 = lim ak ak+1
T →∞ T
k
1 X
R1 = lim ak ak+1
N→∞ N
k
R1 = a^k ak+1 ,
Similarly
Rn = a^k ak+n

Now letting → 0, the width of each triangular pulse → 0 and the

height → infty in such a way area is still infinite and triangular pulses
become impulses
∞
1 X
Rx (τ ) = Rn δ(τ − nTb )
Tb n=−∞
∞
1 X
Sx (ω) = Rn e −jnωTb
Tb n=−∞
∞
!
1 X
Sx (ω) = R0 + 2 Rn cos(nωTb )
Tb
n=1
∞
!
|P(ω)|2 X
Sy (ω) = R0 + 2 Rn cos(nωTb )
Tb
n=1

PSD polar signaling
In polar signaling 1 is transmitted by a pulse p(t) and 0 is transmitted

by −p(t), ak is equally likely to be 1 or -1. Given Rn = a^ k ak+n
1 X 2 1
R0 = lim ak = lim (N)
N→∞ N N→∞ N
k
Moreover, both ak and ak+n are either +1 or -1. Hence ak ak+n is
either 1 or -1. Assuming pulse amplitudes ak is equally likely to be +1
and -1, on average half the times product ak ak+n will yield +1 or -1.

1 X 1 N N
Rn = lim ak ak+n = lim −
N→∞ N N→∞ N 2 2
k
Consider a half width rectangular pulse

t Tb ωTb
p(t) = rect ⇔ P(ω) = sinc
Tb /2 2 4

Tb ωTb
Sy (ω) = sinc2
4 4
PSD polar signaling(1)
The essential bandwidth is
2Rb Hz, four times that of
theoretical limitRb /2. Even with
the full width pulse the essential
bandwidth is Rb Hz (twice).
No error detection or correction
capability.
It has nonzero PSD at (ω = 0),
rule out the use of ac coupling
during transmission.
For a given power, it can be shown that the detection error
probability for a polar scheme is smallest possible.
Transparent because of presence of pulse regardless of bit sequence.
No discrete clock frequency component in the spectrum, yet clock
can be extracted after rectification.
PSD On-Off signaling
In on-off 1 is transmitted by a pulse p(t) and 0 is transmitted by no

pulse, ak is equally likely to be 1 or 0. Given Rn = a^
k ak+n

1 X 2 1 N N 1
R0 = lim ak = lim (1) + (0) =
N→∞ N N→∞ N 2 2 2
k
Moreover, both ak and ak+n are either +1 or 0. Hence ak × ak+n is
likely to be 1 × 0, 0 × 0, 0 × 1, 1 × 1. On average half the times
product ak ak+n is equal to N4 and 0 for 3N4 terms

1 X 1 N 3N 1
Rn = lim ak ak+n = lim (1) + = n≥1
N→∞ N N→∞ N 2 4 4
k

t Tb ωTb
p(t) = rect ⇔ P(ω) = sinc
Tb /2 2 4

Tb ωTb
Sy (ω) = sinc2
4 4

PSD On-Off signaling(1)
n=∞ ∞
1 1 X −jnωTb 1 1 X −jnωTb
Sx (ω) = + e = + e
2Tb 4Tb n=−∞ 4Tb 4Tb n=−∞
n6=0
∞ ∞
X 2π X 2πn
e −jnωTb = δ ω−
n=−∞
Tb n=−∞ Tb
∞
1 2π X 2πn
Sx (ω) = + δ ω −
4Tb 4Tb2 n=−∞ Tb
∞
" #
|P(ω)|2

2π X 2πn
Sy (ω) = 1+ δ ω−
4Tb Tb n=−∞ Tb
For a half width rectangular pulse

" ∞ #
Tb ωT b 2π X 2πn
Sy (ω) = sinc2 1+ δ ω−
16 4 Tb n=−∞ Tb
PSD On-Off signaling(2)
Both discrete (clock

frequency 1/Tb ) and
continuous component are
present.
Logical result, On-off signal
can be expressed as the sum
of a polar signalling and a
periodic component.
For a given transmit power,
it is less immune to noise
interference than the polar
scheme.
On-off signalling requires
twice as much as power/bit
as polar component
Bipolar Signalling
A signal 0 is transmitted by no pulse and a 1 is transmitted by the
pulse p(t) and −p(t). In an equally likely
scenario
1 X 1 N N 1
R0 = lim ak2 = lim (±1)2 + (0) =
N→∞ N N→∞ N 2 2 2
k
To compute R1 , consider four possible scenarios 00, 01 ,10, 11. Since
0 is not encoded by ak = 0, the product ak ak+1 is 0 for 00, 01, 10.
On average,
1 X 1 N 3N 1
R1 = lim ak ak+1 = lim (+1)(−1) + (0) = −
N→∞ N N→∞ N 4 4 4
k
Similarly for ak ak+2 , we need to consider 8 possible scenarios 000,
001, 010, 011, 100, 110, 101, 111. Former 6 will yield 0, whereas the
remaining 2 will yield -1 and +1 respectively.

1 X 1 N N 3N
R2 = lim ak ak+1 = lim (1) + (−1) + (0) = 0
N→∞ N N→∞ N 8 8 8
k
In general
Rn = 0, n > 1
Bipolar Signalling(1)
|P(ω)|2 2 |P(ω)|2 2

ωTb
Sy (ω) = sin (1 − cos(ωTb )) = sin
2Tb 2Tb 2
For half width rectangular pulse

Tb ωTb ωTb
Sy (ω) = sinc2 2
sin
4 4 2
For bipolar signalling the bandwidth is Rb Hz regardless the pulse is full
width or half width.
Its spectrum has a dc null
Bandwidth is not excessive
Error detection probability
If we rectify the bipolar signal is rectified, we get on-off signal that
has a discrete component at the clock frequency.
It requires twice as much power as of polar signalling.
Detection probability is same as on-off signalling
It is not transparent
High Density Bipolar Signaling

High Density Bipolar Signaling(1)
The problem of the bipolar signal being nontransparent is eliminated
by adding pulses when the consecutive 0’s exceeds n. Such modified
coding is called high-sensity bipolar (HDB) coding and denoted by
HDBN, where N can take on value 1, 2, 3, . . .
Most important is the HDB3 format.
HDBN code is that when a run of N+1 zeros is replaced by one of
the special N+1 binary digit sequences. The sequences are chosen to
include some binary 1’s in order to increase the timing content of the
signal. The 1’s included deliberately violate the bipolar rule for easy
identification of the substituted sequence.
In HDB3 coding, the special sequences used are 000V and B00V,
where B=1 conforms to the bipolar rule and V=1 violates the bipolar
rule.
The choice of sequence 000V or B00V is made in such a way that
consecutive V pulses alternate signs in order to avoid dc wander and
maintain dc null.
High Density Bipolar Signaling(2)
This requires that the sequence B00V be used when there is an even
no of 1’s following the last special sequence 000V be used when there
is an odd number of 1’s following the last sequence.

T1 Carrier System(4)
At the receiver framing pattern is detected to determine any

synchronization loss
Signalling bits is also required for dialling, as well as on-hook and
off-hook signals. Rather than creating another time slot for signalling,
one information bit (the least significant bit) of every sixth sample of
a signal to transmit the information.
This means every sixth sample of each voice signal will have a
possible error corresponding to least significant bit. Every sixth frame
has 168 information bits, 24 signalling bits and 1 framing bit. This is
called 7 65 encoding or robbed-bit signalling. The signalling rate is
8000/6 = 1333bits/sec

Old 7 bit framing format
Super framing structure was introduced to identify the frame

boundaries and signalling frames (every sixth frame).
Superframe consisted of 12 frames, in which framing bits form a
special pattern 100011011100
Two bits per superframe are available for signalling, so it allows
maximum four state signalling.

Extended Superframe (ESF)
New superframe format- Extended Superframe which is 24 frames in

length and carries signalling bits in the eight bit of each channel is
frames 6, 12, 18 and 24. Allowing maximum 16 state signalling
The 8kbits/sec overhead capacity of ESF signal is divided into three
channels; 2kbits/sec for framing, 2kbits/sec for CRC and 4kbits/sec
for a data channel( transport information on signal performance as
received by the distant terminal)
Highly secure– because of the ESF format carries a pattern that
repeats in 24 frames (a long sequence)
In next generation networks signalling is carried out on different
channels, development of signalling networks like common channel
interoffice signalling (CCIS) and many others.

Differential Pulse Code Modulation-DPCM(1)
Generally, the sample values obtained from the analog signals are not
independent.
Rather than encoding each sample as an independent value, a
significant reduction can be achieved by using the degree of
redundancy present in the successive samples.
A simple example is one in which we can transmit the difference
between the successive samples. If m[k] is the k th sample, instead of
transmitting m[k], we transmit the difference d[k] = m[k] − m[k − 1].
At the receiver knowing d[k] and previous value m[k − 1], we can
construct m[k] iteratively at the receiver.
Since the d[k] will be generally smaller, the peak value of the
transmitted signals will be reduced significantly, as ∆υ = mp /L, this
reduces the quantization interval thus quantization noise.
Meaning for a given SNR requirement we can reduce L(or n).

This scheme can be further improved by a estimating the value of the

k th sample m[k] from a knowledge of previous sample values.
If estimate m̂[k], then we transmit the difference of T prediction error
d[k] = m[k] − m̂[k].
At the receiver, the estimate m̂[k] is generated from the previous
sample values, and m[k] is determined by adding the received d[k].
A good predictor will generate a small d[k].
This scheme is known as DPCM

Using Taylor series
Ts2 .. T 2 ...
m(t + Ts ) =m(t) + Ts ṁ(t) + m(t) + s m(t) + . . . (1)
2! 3!
≈m(t) + Ts ṁ(t) for small Ts
A knowledge of the signal and its derivatives at instant t can

predict/predict approximately a future signal t + Ts . Let us denote
the k th sample of m(t), thats is, m(kTs ) = m[k] and m(kTs ±
s −Ts )
Ts ) = m[k ± 1]. Setting t = kTs and ṁ(kTs ) ≈ m(kTs )−m(kT
h i Ts
m[k + 1] ≈ m[k] + Ts m[k]−m[k−1]
Ts
m[k + 1] = 2m[k] − m[k − 1]

The approximation improves as we add more terms on the right hand
side( larger the number of past samples)

m[k] = a1 m[k − 1] + a2 m[k − 2] + · · · + aN m[k − N]
The RHS is m̂[k], the predicted value of m[k]

m̂[k] = a1 m[k − 1] + a2 m[k − 2] + · · · + aN m[k − N]
The input of this predictor is m[k] and output is m̂[k]. Example: first
order predictor m̂[k] = m[k − 1] and a1 = 1.
Minimum mean squared error criterion for best prediction, the
prediction coefficients aj from statistical correlation between various
samples. The predictor is called a linear predictor.

Reminder: In DPCM we transmit not the present sample m[k], but

d[k] = m[k] − m̂[k] (the difference between m[k] and its predicted
value m̂[k].
Receiver: Generate m̂[k] from the past sample values to which the
received d[k] is added to generate m[k].
Problem: At the receiver, instead of past samples m[k − 1],
m[k − 2],... as well as d[k], we have quantized versions mq [k − 1],
mq [k − 2], . . .. We can only determine m̂q [k], the estimate of the
quantized sample mq [k], in terms of the quantized samples mq [k − 1],
mq [k − 2], . . .. This will increase the error in reconstruction.
A better strategy is to determine m̂q [k], the estimate of mq [k]
(instead of m[k]), at the transmitter also from the quantized samples
mq [k − 1], mq [k − 2], . . ..

The difference d[k] = m[k] − m̂q [k] is now transmitted using PCM.
At the receiver, we can generate m̂q [k], ,and from the received d[k],
we can reconstruct mq [k].
d[k] = m[k] − m̂q [k] dq [k] = d[k] + q[k]
|{z}
Quantization Error
The predictor output m̂q [k]

is fed back to its input so
that the predictor input
mq [k]

 m̂q [k] + dq [k]
mq [k] = m[k] − d[k] + dq [k]
m[k] + q[k]


If mp and dp are the peak amplitudes of m(t) and d(t), and same no
of quantization level L are used, then step size ∆υ is reduced by
factor mp /dp
Since noise power is (∆υ)2 /2, the quantization noise reduces by a
factor (mp /dp )2
SNR improvement
Pm
Gp =
Pd
SNR improvement can be as high as 24dB in short-term.
The bit rate of DPCM could be lower than that of PCM by 3-4 bits

Delta Modulation(1)
Sample correlations are further exploited in delta modulation by

oversampling the baseband signal (adjacent sample correlation
increases) frequency.
This small prediction eror can be encoded using 1-bit. DM is 1-bit
DPCM for quantization of m[k] − m̂q [k]. Advantage no need for
framing. DM use a first order predictor m̂q [k] = mq [k − 1]
dq [k] = mq [k] − mq [k − 1]
mq [k] = mq [k − 1] + dq [k]
Assuming mq [0] = 0
mq [k − 1] = mq [k − 2] + dq [k − 1]
Pk
mq [k] = dq [m]
m=0
Demodulator is an Accumulator

Delta Modulation(2)
The demodulator output is mq [k], which when passed through a low

pass filter yields the desired signal reconstructed from the quantized
samples.
The modulator consists of a

comparator and a sampler in
the direct path and
integrator-amplifier in the
feedback path.
Analog signal m(t) is
compared with the feedback
signal-the predicted signal
m̂q (t).
The error signal
d(t) = m(t) − m̂q (t) is
applied to a comparator.
Delta Modulation(3)
If d(t) is applied comparator output is a constant signal of amplitude

E , and if d(t) is negative, the output is −E . The difference is a
binary signal L = 2.
The comparator output is

sampled to produce a train
of narrow impulses. Each
sample is coded as binary
pulse (1-bit DPCM), dq [k] is
delta-modulated pulse train.
The modulated signal dq [k]
is amplified and integrated
in the feedback path to
generate m̂q (t), which tries
to follow m(t).

Delta Modulation(4)
Each pulse dq [k] at the input of integrator gives rise to a step
function (E or −E ). If d(t) > 0, a positive pulse is generated in
dq [k], which give rise to positive pulse in m̂q (t), trying to equalise
m̂q (t) to m(t) in smaller steps.
m̂q (t) is a stair case approximation of m(t). When m̂q (t) is passed
through a low pass filter, the coarseness is eliminated.

Delta Modulation(5)-Threshold of Coding and Overloading
Threshold of Coding- Variation in m(t) smaller than the step size are
lost in DM. Granular noise similar to that of quantization noise
Overloading- If m(t) changes too fast ṁ(t) is too high, m̂q (t) cannot
follow m(t) and overloading (slope-overload) occurs. One of the basic
limiting factors in DM.
The slope overload noise can be reduced by increasing step size σ (E
in the previous block diagram). This increases granular noise
(threshold of coding effect). There is an optimum value of σ, which
gives the best compromise giving the minimal overall noise.
During sampling interval Ts , m̂q (t) is capable of changing σ (step
size).
|ṁ(t)| < σfs
Example: m(t) = A cos(ωt)
σfs
|ṁ(t)|max = ωA < σfs , Amax =
ω
Delta Modulation(6)
For voice signals which contain frequency components upto 4kHz,

calculating Amax by using ω = 2π × 4000 will give overly conservative
value.
σfs
De jager shows that the [Amax ]voice ≈ ωr (2π×4000)
Adaptive Delta Modulation. In DM, the dynamic range of signal
amplitudes is too small because of the threshold and overload effects.
In DM suitable method appears to be the adaptation of step size σ
according to the level of input signal.
Signal m(t) falling rapidly chances of overlaod-Increasing step size to
avoid overload
Signal m(t) slowly varying-decrease step size to reduce granular noise
Overload causes dq [k] several pulses of same polarity in succession
calls for increased step size.
Pulses dq [k] alternating continuously in polarity indicated small
amplitude variation, requiring a reduction in step size.
Delta Modulation(7)-Output SNR
The error d(t) caused by granular noise in DM lies in the range of
(−σ, σ). Similar to PCM where quantization error was in the range of
(−∆υ/2, ∆υ/2). Replacing ∆υ/2 by σ in Noise power equation of
PCM(∆υ 2 /12) gives
σ2
¯2 =
3
The noise PSD has a continuous spectrum with most of the power
concentrated in frequency range extending well beyond the sampling
frequency fs . Most of this will be suppressed by the baseband filter
bandwidth B
Making assumption that the PSD of quantization noise is uniform
and concentrated in the band 0 to fs Hz.
σ2 B
N0 =
3 fs
S ^
3fs m 2 (t)
Output Signal power So = m ^ 2 (t), gives 0 =
N0 2
σ B
Comparison with PCM

The End

Presentation 3

Uploaded by

Copyright:

Available Formats

Presentation 3

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Presentation 3

Uploaded by

Copyright:

Available Formats

Analog and Digital Communication

Pakistan Institute of Engineering and Applied Sciences

December 15, 2014

Waqas Ahmed (PIEAS) Short title December 15, 2014 1 / 41

Waqas Ahmed (PIEAS) Short title December 15, 2014 2 / 41

Source: The input to a digital system is in the form of sequence of

Waqas Ahmed (PIEAS) Short title December 15, 2014 3 / 41

Waqas Ahmed (PIEAS) Short title December 15, 2014 4 / 41

Waqas Ahmed (PIEAS) Short title December 15, 2014 5 / 41

Waqas Ahmed (PIEAS) Short title December 15, 2014 6 / 41

Transmission Bandwidth-As small as possible

Waqas Ahmed (PIEAS) Short title December 15, 2014 7 / 41

The pulses are Tb seconds

Waqas Ahmed (PIEAS) Short title December 15, 2014 8 / 41

Waqas Ahmed (PIEAS) Short title December 15, 2014 9 / 41

As we increases τ we find that the kth pulse of x(t − τ ) will start

Waqas Ahmed (PIEAS) Short title December 15, 2014 11 / 41

Now letting  → 0, the width of each triangular pulse → 0 and the

Waqas Ahmed (PIEAS) Short title December 15, 2014 12 / 41

In polar signaling 1 is transmitted by a pulse p(t) and 0 is transmitted

In on-off 1 is transmitted by a pulse p(t) and 0 is transmitted by no

Waqas Ahmed (PIEAS) Short title December 15, 2014 15 / 41

For a half width rectangular pulse

Both discrete (clock

Waqas Ahmed (PIEAS) Short title December 15, 2014 20 / 41

Waqas Ahmed (PIEAS) Short title December 15, 2014 22 / 41

At the receiver framing pattern is detected to determine any

Waqas Ahmed (PIEAS) Short title December 15, 2014 23 / 41

Super framing structure was introduced to identify the frame

Waqas Ahmed (PIEAS) Short title December 15, 2014 24 / 41

New superframe format- Extended Superframe which is 24 frames in

Waqas Ahmed (PIEAS) Short title December 15, 2014 25 / 41

Waqas Ahmed (PIEAS) Short title December 15, 2014 26 / 41

This scheme can be further improved by a estimating the value of the

Waqas Ahmed (PIEAS) Short title December 15, 2014 27 / 41

Using Taylor series

A knowledge of the signal and its derivatives at instant t can

m[k + 1] = 2m[k] − m[k − 1]

Waqas Ahmed (PIEAS) Short title December 15, 2014 28 / 41

m[k] = a1 m[k − 1] + a2 m[k − 2] + · · · + aN m[k − N]

The RHS is m̂[k], the predicted value of m[k]

Waqas Ahmed (PIEAS) Short title December 15, 2014 29 / 41

Reminder: In DPCM we transmit not the present sample m[k], but

Waqas Ahmed (PIEAS) Short title December 15, 2014 30 / 41

The predictor output m̂q [k]

Waqas Ahmed (PIEAS) Short title December 15, 2014 31 / 41

Waqas Ahmed (PIEAS) Short title December 15, 2014 32 / 41

Sample correlations are further exploited in delta modulation by

Waqas Ahmed (PIEAS) Short title December 15, 2014 33 / 41

The demodulator output is mq [k], which when passed through a low

The modulator consists of a

If d(t) is applied comparator output is a constant signal of amplitude

The comparator output is

Waqas Ahmed (PIEAS) Short title December 15, 2014 35 / 41

Waqas Ahmed (PIEAS) Short title December 15, 2014 36 / 41

For voice signals which contain frequency components upto 4kHz,

Waqas Ahmed (PIEAS) Short title December 15, 2014 40 / 41

Waqas Ahmed (PIEAS) Short title December 15, 2014 41 / 41

You might also like

Now letting → 0, the width of each triangular pulse → 0 and the