The following figure shows the resulting FT of a “sinc” pulse for FB = 100 Hz and k = 2, 6, 25.

0
FT of Truncated 'sinc' Pulse, Taillength k ∗ TB , FB = 1/TB = 100 Hz
k=2
k=6
−10 k = 25

−20
[dB]
20 log 10 (|P(f)|)

−30

−40

−50

−60
0 20 40 60 80 100
f [Hz]

The truncation in the time domain to t = ±kTB leads to sidelobes in the frequency domain
which are spaced FB /(2k) apart. This is closely related to Gibbs phenomenon (trying to
approximate a discontinuity in one domain with a finite number of basis functions in the other
domain). As a consequence, the first sidelobe will always only be about 21 dBRdown from the
∞
main lobe, no matter how large k is chosen. Another problem is that, since  t−1 dt = ∞,
PAM signals based on the “sinc” pulse are very sensitive to timing errors of the sampling
circuit at the receiver.
A solution to these problems is obtained by making the transition from TB to 0 near |f | =
FB /2 = 1/(2TB ) in the frequency domain less abrupt. The graph below shows the FT of a
pulse p(t) which uses an extra amount of α/(2TB ) = αFB /2, with 0 ≤ α ≤ 1, of bandwidth
to accomodate a sinusoidal or raised cosine rolloff.

P (f )

TB

f
−1−α −1 −1+α 1−α 1 1+α
2TB 2TB 2TB 0 2TB 2TB 2TB

11
Analytically, this P (f ) can be expressed as
 1−α
 TB , |f | ≤ ,

 2TB

TB h 
i
P (f ) = 1 + cos πTαB |f | − 1−α
, 1−α
≤ |f | ≤ 1+α
,
 2

2TB 2TB 2TB

 1+α
0, |f | ≥ 2TB
.

In the time domain this corresponds to

sin(πt/TB ) cos(παt/TB ) p(t)

p(t) =
πt/TB 1 − (2αt/TB )2 1 α = 0.5

0≤α≤1 t
· · · −2TB −TB 0 TB 2TB · · ·

Now, for any α > 0, p(t) decreases like 1/|t|3 as |t| → ∞, which is a big improvement over the
1/|t| decrease of the“sinc” pulse. Typical values used in practice are α = 0.2 . . . 0.5. Because
of its shape in the frequency domain, this pulse is called raised cosine in frequency (RCf )
pulse. It is quite easy to see, either in the time or frequency domain, that the RCf pulse
satisfies Nyquist’s first criterion for no ISI.

1.5 Partial Response Signaling

Consider the general blockdiagram of a communication system using pulse amplitude mod-
ulation (PAM) with baud rate FB = 1/TB .

Noise n(t)

an PAM s(t) Channel r(t) Filter b(t) bn

+ • ↓ •
p(t) HC (f ) HR (f )
t = nTB

Define

q(t) ⇐⇒ Q(f ) = P (f ) HC (f ) HR (f ) and qn = q(nTB ) ⇐⇒ Q(z) .

The DT equivalent circuit of the PAM communication system is shown in the following
figure.

an PAM bn
qn

12
If Nyquist’s first criterion is satisfied then there is no intersymbol interference (ISI) at the
sampling times and thus
qn = δn =⇒ b n = an ∗ q n = an ∗ δn = an .
Many practical channels have spectral nulls at certain frequencies, e.g., at dc if transmission
media are transformer-coupled or magnetic recording and playback is used. Near the upper
frequency limit, practical channels usually have some rolloff characteristic, as opposed to the
“brickwall’ cutoff of ideal filters. In magnetic recording the finite gap width of the read/write
head actually leads to a spectral null which essentially defines the upper frequency limit of
the channel. In all these cases it is not very convenient and efficient to try to satisfy Nyquist’s
first criterion. A better alternative is to allow a controlled amount of ISI between adjacent
samples at the output of the DT equivalent PAM model qn ⇔ Q(z). This technique is called
partial response (PR) signaling or correlative-level coding.
Example: A simple example is the DT equivalent PAM model with

−1 1 , n = 0, 1 ,
Q(z) = 1 + z ⇐⇒ qn =
0 , otherwise ,
which places a spectral null at f = FB /2 and is called class I PR. The DTFT of this qn is
∞
X
Q(f TB ) = qn e−j2πf TB n = 1 + e−j2πf TB
n=−∞
= (ejπf TB + e−jπf TB ) e−jπf TB = 2 cos πf TB e−jπf TB .
The magnitude and the phase of this DTFT are shown in the following figure.

2.0
DTFT of Class I PR Signaling, Q(z) = 1 + z −1

1.5
|Q(f/FB )|

1.0

0.5

0.0
−2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0
100

50
∠ Q(f/FB ) [deg]

−50

−100
−2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0
Normalized Frequency f/FB

13
To obtain a corresponding CT q(t), make use of the fact that
∞
X
Q(f TB ) = FB Q(f − m FB ) .
m=−∞

To obtain a q(t) of minimum bandwidth set


TB Q(f TB ) , |f | < FB /2 ,
Q(f ) =
0, otherwise .

For class I PR this becomes


2TB cos πf TB e−jπf TB , |f | < FB /2 ,
Q(f ) =
0, otherwise .

Taking the inverse FT yields

sin πt/TB
q(t) = , with q(t) = 1 at t = 0, TB .
πt/TB (1 − t/TB )

This pulse is shown in the following figure for the range−4TB ≤ t ≤ 5TB (with FB = 100 Hz
and thus TB = 0.01 s), and it is easy to see that q0 = q(0) = q1 = q(TB ) = 1, and all other
qn = q(nTB ) = 0.

1.4
Partial Response Signaling Pulse q(t) for Q(z) = 1 + z −1
1.2
1.0
0.8
0.6
()
q t

0.4
0.2
0.0
−0.2
−4 −3 −2 −1 0 1 2 3 4 5
Normalized Time t/T B

More generally, if
X X
Q(z) = qn z −n =⇒ Q(f TB ) = qn e−j2πf TB n ,
n n

then the Q(f ) of minimum bandwidth is

( P
TB n qn e−j2πf TB n , |f | < FB /2 ,
Q(f ) =
0, otherwise .

14
Taking the inverse FT yields q(t) as
X sin(πt/TB − nπ) sin πt/TB X t/TB
q(t) = qn = (−1)n qn .
n
π(t/TB − n) πt/TB n
t/TB − n

Different classes of PR signaling schemes and associated Q(z) are shown in the following ta-
ble. The most popular schemes are PR I and PR IV. Magnetic recording systems increasingly
use the more advanced EPR4 and E2PR4 schemes.

Classes of Partial Response Signaling Schemes

Class Q(z) Remarks
I 1 + z −1 Duobinary coding
1 − z −1 “Dicode” channel
II (1 + z −1 )2
III (1 + z −1 )(2 − z −1 )
IV (1 + z −1 )(1 − z −1 ) Modified duobinary
(1 + z −1 )(1 + z −1 + z −2 ) (Optical storage)
EPR4 (1 + z −1 )2 (1 − z −1 ) Enhanced class IV
E2PR4 (1 + z −1 )3 (1 − z −1 ) Enhanced EPR4
V (1 + z −1 )2 (1 − z −1 )2

In general, when PR signaling is used, bn 6= an in the expression bn = an ∗ qn , even in the

absence of noise because of the ISI that was deliberately introduced by using a Q(z) 6= 1. A
simple solution to remove the effect of this ISI and recover the original data signal is to use
a DT filter whose z-transform is the inverse of the DT PAM model Q(z). For class I PR, for
example, one would use
1
PR I: Q(z) = 1 + z −1 (FIR) =⇒ [Q(z)]−1 = (IIR) .
1 + z −1
At first sight it appears to be a good idea to use a DT filter with system function [Q(z)]−1 at
the receiver to remove the effects of ISI. However, in order to introduce only a well controlled
amount of ISI, Q(z) is generally chosen to be a FIR filter, which implies that its inverse is
an IIR filter. But using an IIR filter at the receiver can lead to infinite error propagation if
the received signal contains errors due to noise. Therefore, rather than compensating at the
receiver for the ISI introduced by the PR signaling scheme, one generally uses precoding
at the transmitter, as shown in the following block diagram.

dn 1 an PAM bn
Q(z) Q(z)
| {z } | {z }
Precoder DT equiv PAM

15