Pulse Shaping

The PSD S y (w) of a digital signal y(t ) can be controlled

by S x (w) the pulse pattern or by the pulse shape P(w) .

In this section, we discuss the control of PSD by pulse

shaping.
The basic pulse p(t ) was a simple rectangular pulse.
In this case
The bandwidth of the PSD S y (w) is infinite.

But its essential bandwidth of S y (w) was finite.

For example
Most of the power of a bipolar signal is contained within
the essential band (0 to fo ) .

1
However, that the PSD is small but is still nonzero in the
range ( f > fo ) .

Therefore, when such a signal is transmitted over a

channel of bandwidth ( fo ) , a significant portion of its

spectrum is transmitted, but a small portion of the

spectrum is suppressed.
This will cause distortion in the received signal, but since
the power in the suppressed components is small, the
distortion is not severe.
There will be intersymbol interference (ISI) caused by
the distortion which manifests itself as pulse spreading.

2
To Resolve the problem of ISI
If we are considering Time-Limited pulses
Such pulses cannot be band-limited.
Part of their spectra is suppressed by a band-limited
channel.
This causes pulse distortion and the consequent ISI.

If we are considering Band-limited pulses

Such pulses cannot be time-limited.
Obviously, various pulses will overlap and cause ISI.
Thus, whether we begin with time-limited pulses or band-
limited pulses, it appears that ISI cannot be avoided.
Pulse amplitudes can be detected correctly despite pulse
spreading (or overlapping), if there is no ISI at the
decision-making instant.
This can be accomplished by a properly shaped band-
limited pulse.
To eliminate ISI, Nyquist proposed three different criteria
for pulse shaping

3
 Nyquist's First Criterion for Zero ISI

In the first method, Nyquist achieves zero ISI:

By choosing the pulse shape such that:
It has a nonzero amplitude at the center, say at (t = 0) ,

and zero amplitudes at (t = ± nTo ) .

p (t ) = 1 t =0

p (t ) = 0 t = ± n To

Where, (n = 1, 2, 3...)
To is the separation between successive transmitted pulses.

4
It can be seen that each transmitted pulse causes zero ISI at
all the remaining pulse centers, or signaling instants.
f
If we restrict the pulse bandwidth to o , then only one pulse
2
Sinc ( f ot ) has this property as shown in figure.

Sinc ( f ot ) = 1 at t = 0

1
Sinc ( f ot ) = 0 at t = ± nTo (To = )
fo

The Sinc ( f t ) pulse.

o
Using this pulse, we can transmit at a rate of ( f o ) pulses/sec,
f
without ISI, over a bandwidth o (The Nyquist rate).
2

Because,
1 w
Sinc( f ot ) ←⎯→ Π ( )
fo 2πf o

5
The Sinc ( f ot ) can be generated as an impulse response of an
fo
ideal filter of bandwidth as shown in the following figure.
2

Fig. The minimum bandwidth pulse that satisfies Nyquist's first criterion.

The Sinc pulse have serious practical problems.

Any of these errors can cause sufficiently large ISI for the
scheme to fail.
1. Transmission rate

2. Sampling rate at the receiver

3. Small jitter in sampling instants

This is because Sinc ( f ot ) decays too slowly as (1 / t ) .

6
The solution is to find a pulse p(t ) that satisfies the above
equation but that decays faster than (1 / t ).
Nyquist has shown that such a pulse requires
fo
bandwidth (k ) with (1 < k < 2)
2

This can be proved as follows

The desired pulse p(t ) satisfies:

p (t ) = 1 t =0

p (t ) = 0 t = ± n To

Let p(t ) ←⎯→ P( w)

f
Where, the bandwidth of P(w) is in the range ( o , f o ) as
2
shown in fig.(a).

Fig. Spectrum of the pulse p(t ) .

7
Consider the vestigial spectrum shown in figure over the
range 0 < w < wo .

Fig. A vestigial spectrum.

w
The bandwidth of P(w) is ( o + wx ) .
2

If we define the roll-off factor (r )

wx
r=
wo / 2

Where, 0 ≤ r ≤ 1
fo
The bandwidth of P(w) is (1 + r ) .
2

8
The pulse p(t ) required has zero ISI at the centers of all other

pulses transmitted at a rate of f o pulses / sec .

Shaping the pulse p(t ) so that it causes zero ISI at the centers of
all the remaining pulses is the first of three methods proposed by
Nyquist for zero ISI. This criterion for zero ISI is known as
Nyquist's first criterion.

Because 0 ≤ r < 1

Case 1: wx = 0 ( r = 0)

w
Case 2: wx = o (r = 0.5)
4

w
Case 3: wx = o ( r = 1)
2

9
Three curves, corresponding to the three cases:
Fig.(a) shows three curves corresponds to the above three cases.
Fig.(b) shows their impulse responses.

Fig. Pulses satisfying Nyquist’s First Criterion.

It can be seen that increasing ( wx or r ) improves p(t ) , so in

case (3) r = 1 has the maximum value.

w
Case (3) wx = o ( r = 1) ( wx has the maximum value)
2
This characteristic is known as raised-cosine characteristics,
because it represents a cosine raised by its peak amplitude.

10
 Nyquist's Second Criterion for Zero ISI

This scheme doubles the transmission rate (or reduces the

bandwidth to half) by using a simple trick (doubling the dotting
speed) known to telegraphers.
Duobinary scheme uses Second Criterion for Zero ISI

Example
A Telegraph System is designed to transmit data at a rate ( f o )

could also receive data at a rate ( 2 f o ) unambiguously, provided

the received signal was interpreted correctly.

To explain this behavior

Consider a system designed to transmit data at a rate of
f o bits / sec using polar signaling.

When 1 is transmitted by a full-width rectangular pulse p(t )

The received pulse will rise to positive amplitude (k ) .

11
When 0 is transmitted by − p(t )
The received pulse will rise to negative amplitude (−k ) .

If the transmitted pulse rate is doubled

The transmitted pulse width is halved, and the received pulses
cannot reach their full values.
But if 1 is followed by 1
We have two half-width pulses in succession-making one full-
width pulse. This causes the received pulse to reach full positive
value (k ) .

Similarly, if 0 is followed by 0
The received pulse reaches full negative value (−k ) .

But if 1 is followed by 0 or vice versa

The received pulse will stay close to 0.

12
Thus, the received signal can be interpreted as shown
in Table (1)

Table (1)
Received amplitude Transmitted digits

(k ) 1 previous digit also 1

(−k ) 0 previous digit also 0

0 Complement of the previous digit

Thus, the data can be received unambiguously even when

the rate is doubled.

This trick uses Nyquist's Second Criterion pulse p(t ) :

T
p (± o ) = C
2
and,
nTo
p(± )=0 n = 3, 5, 7,....
2

13
This means that the pulse p(t ) has zero interference at
points midway between all the signaling instants
3To 5T 7T
(t = ± , ± o , ± o ,.........)
2 2 2

With the sole exceptions

At the points midway between itself and its immediate
T
neighbors (t = ± o ) , as shown in figure.
2

Fig. The minimum bandwidth pulse that satisfies Nyquist’s Second Criterion.

14
 fo
If we restrict the pulse bandwidth to
2

Fig. Spectrum of the minimum bandwidth pulse that satisfies Nyquist’s

Second Criterion.

The functioning of this scheme can be as follows:

A 1 is transmitted by p(t ) and a 0 is transmitted by − p(t ) , note
T f
that: p (± o ) = o
2 2
And,
nTo
p( ± )=0 (n = 3, 5, 7....)
2

15
When a 1 is followed by 0 or (vice versa)
We have two pulses of opposite polarity in succession p(t )
followed by − p(t ) (or vice versa). Hence, at the midpoint of the
two pulses, the pulse amplitudes are equal and of opposite signs,
and the sum is zero as shown in fig.(a).

Fig. Transmitting 1 followed by 0 for Nyquist's second criterion.

16
When a 1 is followed by 1
f
The amplitudes add to 2( o ) = f o at the midpoints of the two
2
signaling instants as shown in fig.(b).

Fig. Transmitting 1 followed by 1 for Nyquist's second criterion.

When a 0 is followed by 0
It yields a negative value (− f o ) .

Example
17
A typical sequence 110010 and the corresponding waveform
x(t ) are shown in fig.(c).

The rate of transmission is fo digits/sec using a

fo
bandwidth Hz .
2

Fig. Transmitting sequence 110010 for Nyquist's Second Criterion.

To detect the received sequence, we sample the waveform x(t )

at the midpoints of the signaling instants.

Three possible sample values exist f o , 0 , − f o .

18
1. If the sample value is f o , the digit detected is 1 and the

preceding digit is also 1.

2. If the sample value is 0 means a transition, that is, the

digit detected is 0 and the preceding digit is 1 and vice
versa.

3. If the sample value is (− f o ) , means that the digit

detected is 0 and the preceding digit is also 0.

Example
The error-detecting property of this scheme.

19
The following table shows that:
There are always an even number of zero-valued samples
between two full-valued samples of the same polarity
and
an odd number of zero-valued samples between two-full-valued
samples of opposite polarity.

Transmitted 1 1 0 1 1 0 0 0 1 0 1 1 1
sequence
Samples of fo 0 0 fo 0 − fo − fo 0 0 0 fo fo
x(t)
Detected 1 1 0 1 1 0 0 0 1 0 1 1 1
sequence

Thus, the first sample of x(t ) is f o .

The next full-valued sample (the fourth sample) is f o , between

these full-valued samples of the same polarity, there are an even

number (i.e., 2) of zero-valued samples.
If one of the sample values is detected wrong, this rule is
violated, and the error is detected.
In the Duobinary case

20
Binary data digits are transmitted by ternary (3-valued)
pulses.
Pulses of the same polarity are separated by an even
number of zero pulses, and those of opposite polarity are
separated by an odd number of zero pulses.

Use of Differential Coding

21
In the above table, a zero-valued sample implies transition, that
is, the digit is detected as 1 if the previous digit is 0 and vice
versa. This means that the digit interpretation is based on the
preceding digit.
If a digit were detected wrong, the error would tend to
propagate. Use of so-called differential coding eliminates this
problem.

In differential coding
A 1 is transmitted by a pulse identical to that used for the
previous bit.
A 0 is transmitted by a pulse negative of that used for the
previous bit.

22
This is shown in the following figure, using a half-width
rectangular pulse.

Fig. Differential coding.

This scheme not only simplifies the decision rule but also makes
the decision independent of the previous digit and eliminates
error propagation.

23
 Nyquist's Third Criterion for zero ISI

In this scheme, ISI is eliminated by shaping the pulse so

that the total area under the pulse in the signaling interval
is nonzero but is zero in any other signaling interval.

One pulse p(t ) that satisfies this criterion has the

following spectrum
1 w
P ( w) = .Π ( )
Sinc (
w
) 2πf o
2πf o

It can be shown that p(t ) , the inverse transform of this P(w)

satisfies the third criterion.

This scheme is inferior to the first and second from the point
of view of noise immunity.
Pulse Generation

24
A pulse p(t ) satisfying a Nyquist criterion can be generated as
the unit impulse response of a filter with transfer function P(w) .
A better method is to generate the waveform directly, using the
transversal filter as shown in figure.

Fig. Pulse generation by transversal filter.

From figure, sample values are obtained from the graph of

the pulse p(t ) to be generated and the constants of the

25
filter are set in proportion to these sample values in
sequence.

When a narrow rectangular pulse with the width of the

sampling interval is applied at the input of the transversal
filter, the output will be a staircase approximation of p(t ) .

When the output is passed through a low pass filter, the

output will be a smoother version. The approximation can
be improved by reducing the pulse sampling interval.

It should be stressed once again that the pulses arriving at

the detector input of the receiver need to have the Nyquist
form.

26
Hence, the transmitted pulses should be so shaped that
after passing through the channel, they are received in the
form of Nyquist pulses.

In practice, however, pulses need not be shaped rigidly at

the transmitter. The final shaping can be carried out by an
equalizer at the receiver.

27
 Scrambling

Scramblers tend to make the data more random by removing

long strings of 1's or 0's in binary data.

Figure shows a typical scrambler and unscrambler

Fig. A scrambler and an unscrambler.

28
The scrambler consists of a feedback shift register and the
matching unscramble has a feed-forward shift register.
Each stage in the shift register delays a bit by one unit.

To show that the matched unscrambler does indeed

unscramble, consider the output sequence T of the
scrambler.

If S is the input sequence, then,

S ⊕ D3T ⊕ D5T = T
Where, D represents the delay operator.

That is; D nT is the sequence T delayed by n units .

Remember that the modulo-2 sum of any sequence with
itself gives a sequence of all 0's.

29
Adding ( D 3 ⊕ D 5 )T in both sides of the above equation we
get:
S = T ⊕ ( D3 ⊕ D5 )T

S = [1 ⊕ ( D3 ⊕ D5 )] T

S = [1 ⊕ F ] T where, F = D 3 ⊕ D 5

To design the unscrambler at the receiver

we start with the received sequence T .
It follows that:

S = T ⊕ ( D3 ⊕ D5 ) T

This is readily implemented by the arrangement shown in the

above figure.
Note that a single detection error in the received sequence T
will be affect three output bits in R.
Hence, scrambling has the disadvantage of causing multiple
errors for a single received bit error.

30
 The Eye Pattern

The ISI can be conveniently studied on an oscilloscope through

what is known as the "eye pattern".

A random binary pulse sequence is sent over the channel. The

channel output is applied to the vertical input of an oscilloscope.

The time base of the scope is triggered at the same rate as that of
the incoming pulses, and it yields a sweep lasting exactly (To ) ,

the internal of one pulse.

The oscilloscope shows the superposition of several traces,

which is nothing but the input signal (vertical input) cut up
every (To ). , and then superimposed.

The oscilloscope pattern thus formed looks like a human eye,

and hence, the name eye diagram.

31
Example
Consider the transmission of a binary signal by polar rectangular
pulses.

If the channel is ideal with infinite bandwidth

Pulses will be received without distortion. the resulting eye
diagram will be as in fig.(a).

Fig. The eye pattern.

32
If the channel is not distortionless or has finite
bandwidth, or both
Received pulses will no longer be rectangular but will be
rounded and spread out.
If the equalizer is adjusted properly to eliminate ISI at the pulse
sampling instants, the resulting eye diagram will be rounded but
will still have full opening at the midpoint of the eye as shown
in fig.(b).

Fig. The eye pattern.

This is because the midpoint of the eye represents the sampling

instant of each pulse and the pulse amplitude is maximum also,
at this point there is interference from other pulses (because of
zero ISI).

33
If ISI is not zero
Pulse values at their respective sampling instants will deviate
from full-scale values by varying amount in each trace, closing
the eye partially at the midpoint as shown in Fig.(c).

Fig. The eye pattern.

In the presence of channel noise

The eye will tend to close in all cases. Smaller noise will cause
less closing.
The decision threshold as to which symbol (1 or 0) is
transmitted is the midpoint of the eye.

34
The eye diagram is useful in:

1. Decide the optimum sampling or decision-making instant.

2. Determine optimum tap settings of the equalizer so, taps are

adjusted to obtain the maximum eye opening.

3. To know the amount of noise that can be tolerated.

4. The width of the eye indicates the time interval over which

the decision can be made.

35
 Timing Extraction

The received digital signal needs to be sampled at precise

instants. This requires a clock signal at the transmitter (symbol
or bit synchronization).

Three general methods of synchronization exist

1. Derivation from a primary or a secondary standard (e.g.,
transmitter and receiver slaved to a master timing source.
In this method is suitable for large volume of data and high-
speed communication systems because of its high cost.
2. Transmitting a separate synchronizing signal (pilot clock).
In this method, part of the channel capacity is used to
transmit timing information and is suitable when the
available capacity is large compared to the data rate.
3. Self-synchronization, where the timing information is
extracted from the received signal itself. This method is a
very efficient method of timing extraction or clock recovery
because the timing is derived from the digital signal itself.

36
Example The Self-Synchronization Method
A digital signal such as an ON-OFF binary signal contains a
discrete component of the clock frequency itself as shown in
fig.(a). This can also be seen from the fact that such a waveform
can be expressed as sum of two waveforms:
1. A random component as shown in fig.(b).
2. A periodic component with the same fundamental
frequency as the clock frequency as shown in fig.(c).
Hence, when the ON-OFF binary signal is applied to a resonant
circuit tuned to the clock frequency, the output signal is the
desired clock signal.

Fig. ON-OFF signal and its components.

37
Not all the binary signals contain a discrete component of the
clock frequency.
For example, a bipolar signal has no discrete component of any
frequency. In such cases, timing can be extracted by using a
nonlinear operation.

38
 Detection Error Probability

The signal received at the detector consists of the desired pulse

train plus a random channel noise. This can cause error in pulse
detection.

Example
Consider the case of polar transmission using a basic pulse
p(t ) as shown in fig.(a).

Fig. Basic pulse.

This pulse has a peak amplitude A p .

39
A Typical received pulse train is shown in fig.(b).

Fig. Received pulse train.

Pulses are sampled at their peak values

If noise were absent, the sample of the positive pulse
(corresponding to 1) will be A p , and that of the negative pulse

(corresponding to 0) will be − A p .

Because of noise, these samples would be (± A p + n) where, (n)

is the random noise amplitude shown in the above figure.

40
The decision whether 1 or 0 is transmitted could be made readily
from the pulse sample, except that (n) is random, meaning its
exact value is unpredictable.

If 1 is transmitted but (n) at the sampling instant may has a

large negative value, this will make the sample value
( A p + n) small or even negative.

If 0 is transmitted, but ( n ) at the sampling instant may has a

large positive value, this will make the sample value (− A p + n)

can be positive and the digit will be detected wrong as 1. This is

clear from fig.(b).

The amplitude (n) of the so-called Gaussian noise ranges

( from − ∞ to ∞) although the probability that (n) will take on

n2
very large values decreases rapidly as exp( − ),
2
2σ n

where, σ n is the r.m.s value of the noise.

41
Still, occasionally, (n) can take on large positive or negative
value causing detection errors.
When 0 is transmitted
The sample value of the received pulse is (− A p + n) ,

if (n > A p ) , the sample value will be positive and the digit will

be detected wrongly as 1.

If P(ε 0) is the probability of error given that 0 is transmitted,

then P(ε 0) is the probability that (n > A p ) .

When 1 is transmitted
The sample value of the received pulse is ( A p + n) ,

if (n < − A p ) , the sample value will be negative and the digit will

be detected wrongly as 0.

If P(ε 1) is the probability of error given that 1 is transmitted.

P(ε 1) is the probability that (n < − A p ) .

42
The probability density function (PDF) of n for the
case of Gaussian noise is given by:
n2
−
1 2σ n2
P ( n) = e
2π σ n

Where, σ n is the r.m.s. value of the noise signal.

The PDF is shown in fig.(c). It is the relative

distribution of values of n.

Fig. Error probability in threshold detection.

43
The probability that an amplitude of ( n ) lies in a range (α , β )
is given by the area under PDF over this range.
It follows that:
Ap
P (ε 0) = Q ( )
σn

The PDF of ( n ) is symmetrical about the origin,

Ap
So, the probability that (n > ) is the same as the probability
σn

Ap
that (n < − ).
σn

Hence, P(ε 0) = P (ε 1)

A very good approximation to the Q-function is:

x2
1 0.7 −
Q(( x) = (1 − ). e 2 x>2
2π σ n x2

44
 Example
Consider the peak pulse amplitude A p to be k times the noise

r.m.s value, that is, A p = k σ n .

In this case
P (ε 0) = P (ε 1) = Q(k )

Table shows the error prob. for various value of k .

k 1 2 3 4 5 6
P (ε 0) 0.1587 0.0227 0.00135 3.16 x10-5 2.87x10-7 9.9x10-10

Thus, when A p is five times the noise r.m.s amplitude, the error

probability is 2.87x10-7.

45
For the polar case
The separation between pulse amplitudes to be distinguished is
Ap
2 A p . The error probability is: Q( )
σn

For the ON-OFF case

The separation between amplitudes to be distinguished is only
A p and consequently,

A
p
The error probability is: P(ε 0) = P (ε 1) = Q ( )
2σ
n

For the Bipolar case

The situation is slightly different
Because 1 is transmitted by a positive or a negative pulse and 0
is transmitted by no pulse.
The average error probability is:
1 3 Ap
P(ε ) = [ P(ε 0) + P(ε 1)] = .Q( )
2 2 2σ n

46
 M-ary Communication

Digital communication uses only a finite number of symbols for

communication, the minimum number being two (the binary
case). We shall briefly discuss some aspects of M − ary
communication (communication using M symbols).

1. Multi-amplitude scheme

It easy to show that the information transmitted by each

symbol increases with M .

Example
when M = 4 (4-ary or quaternary)
we have four basic symbols, or pulses available for
communication as shown in figure.

47
 Fig. 4-ary Multi-amplitude signal.

A sequence of two binary digits can be transmitted by just one

4-ary symbol (11, 10, 01, 00).
Because we have four distinct symbols available, we can assign
one of the four symbols to each of these communications.
This means that:
One 4-ary symbol can transmit the information of two binary
digits.
One 8-ary symbol can transmit the information of three binary
digits.

48
One 16-ary symbol can transmit the information of four binary
digits.
In general, the information IM transmitted by an M − ary

symbol is given by:

I M = log 2 M binary digits or bits

This means that:

We increase M
The rate of communication increases by a factor of log 2 M .

The Pulse amplitudes increase with M .

The transmitted power increases as M 2 .

the bandwidth remains constant.

49
2. Multi-Tone signaling (MFSK)

In this case, M symbols are transmitted by M orthogonal pulses

of frequency ( w1, w2 ,......., wM ) , each of duration To .

Thus, the M transmitted pulses are of the form:

φk (t ) = 2 p (t ) cos( wk t )

Where,
wk = 2π ( N + k ) To .

In M − ary Orthogonal scheme, we may use M orthogonal

pulses φ1(t ), φ2 (t ), .........., φM (t ) with the property:

To
∫ φi (t ) . φ j (t ) = c i= j
0
To
∫ φi (t ) . φ j (t ) = 0 i≠ j
0

50
 Fig. One possible set of M − orthogonal signal.

φk (t ) = sin(2π k f o t ) 0 < t < To (k = 1,2,3...., M )

φk (t ) = 0 otherwise

In an M − ary orthogonal scheme

The rate of communication is increased by a factor of (log 2 M )

The transmission bandwidth is increased by a factor of M .

The transmitted power is practically independent of M .

51
3. Multiphase Signaling

In binary PSK (BPSK), The two basic pulses are:

p′(t ) cos wc t p′(t ) cos( wc t + π ) .

We generalize the idea for the M − ary case (MPSK) using M

(
pulses with the k th pulse p′(t ) cos wct + ( 2π ) k
M
)
Thus, the phases of successive pulses are 2π radians a part as
M
shown in figure.

Fig. MPSK signals.

52
Thus, M − ary signaling allows us great flexibility in
exchanging signal power (or SNR) with transmission
bandwidth.

The choice of the appropriate system depends on:

1. If the Bandwidth is at a premium choice

We use multi-amplitude signaling.
In Multi-Amplitude or Multiphase Signaling
The bandwidth is independent of M .

But the transmitted power increases as M 2 .

2. If the Power is at a premium choice

We use orthogonal signaling.
In Multi-Tone Signaling (MFSK)
The transmitted power is practically independent of M .
But the transmission bandwidth increases as M , that is, the
bandwidth increases exponentially with the information rate.

53