=
}
DIGITAL TRANSMISSION THROUGH BANDLIMITED CHANNELS
If the channel is a baseband that is bandlimited to B
c
,then
C(f)=0 for |f|> B
c
Suppose that the input to a bandlimited channel is a signal
waveform g
T
(t). Then the response of the channel is the
convolution of g
T
(t) with c(t) ;i.e.,
Expressed in the frequency domain, we have
H(f)=C(f)G
T
(f)
( ) ( ) ( ) ( ) ( )
T T
h t c g t d c t g t t t t
= =
}
DIGITAL TRANSMISSION THROUGH BANDLIMITED CHANNELS
The signal at the input to the demodulator is of the form h(t)+n(t),
where n(t) denotes the AWGN. Let us pass the received signal
h(t)+n(t) through the matched filter that has a frequency response
where t
0
is some nominal time delay at which we sample the filter
output
The signal component at the output of the matched filter at the
sampling instant t=t
0
is
The noise component at the output of the matched filter has a zero
mean and a power-spectral density
0
2
( ) ( ) 8.1.4
j ft
R
G f H f e
t -
=
2
0
( ) ( ) 8.1.5
s h
y t H f df c
= =
}
2
0
( ) ( ) 8.1.6
2
n
N
S f H f =
DIGITAL TRANSMISSION THROUGH BANDLIMITED CHANNELS
The noise power at the output of the matched filter has a
variance
The SNR at the output of the matched filter is
Compared to the previous result, the major difference in this
development is that the filter impulse response is matched to the
received signal h(t) instead of the transmitted signal
2
2
0 0
( ) ( ) 8.1.7
2 2
h
n n
N N
S f df H f df
c
o
= = =
} }
2
0 0
2
8.1.8
/ 2
h h
o h
S
N N N
c c
c
| |
= =
|
\ .
DIGITAL TRANSMISSION THROUGH BANDLIMITED CHANNELS
Example 8.1.1
The signal pulse g
T
(t), defined as
The channel output is corrupted by AWGN with power spectral
density . determine the matched filter to the received
signal and the output SNR
1 2
( ) 1 cos ( ) , 0
2 2
T
T
g t t t T
T
t
(
= + s s
(
0
/ 2 N
DIGITAL TRANSMISSION THROUGH BANDLIMITED
CHANNELS
DIGITAL TRANSMISSION THROUGH BANDLIMITED CHANNELS
The spectrum of the signal is
2 2
2 2
sin
( )
2 2 (1 )
sin
=
2 (1 )
j fT
T
j fT
T fT
G f e
fT f T
T c fT
e
f T
t
t
t
t
t
( ) ( ) ( )
( )
0 ,
T
T
H f C f G f
G f f W
otherwise
=
s
}
}
}
2
2
0 0
( )
2 2
W
h
n T
W
N N
G f df
c
o
= =
}
0 0
2
h
S
N N
c
| |
=
|
\ .
DIGITAL TRANSMISSION THROUGH BANDLIMITED CHANNELS
The amount of signal energy at the output of the matched filter
depends on the value of the channel bandwidth W.
The maximum value of
h
, obtained as W, is
???
2 2
0
max ( ) ( )
T
h T T
G f df g t df c
= =
} }
8.1.1Digital PAM Transmission through Bandlimited Baseband Channel
Let us consider the baseband PAM communication system
illustrated by the functional block in Figure 8.3.
Digital PAM Transmission through Bandlimited Baseband Channel
First we consider the M-ary PAM, the input binary data sequence
is subdivides into k-bit symbols and each symbol is mapped into a
corresponding amplitude level. The amplitude modulates the
output of the transmitting filter. The baseband signal at the output
of the transmitting filter may be expressed as
where T=k / R
b
is the symbol interval, R
b
is the bit rate and {a
n
} is a
sequence of amplitude corresponding to the sequence of k-bit
blocks of information bits
The received signal at the demodulator, may be expressed as
is the impulse response of the channel, and n(t)
represents the AWGN
( ) ( ) 8.1.9
n T
n
t a g t nT u
=
=
( ) ( ) ( ) 8.1.10
n
n
r t a h t nT n t
=
= +
( ) ( ) ( )
T
h t c t g t =
Digital PAM Transmission through Bandlimited Baseband Channel
The output of the receiving filter may be expressed as
To recover the information symbols {a
n
}, the output of the
receiving filter is sampled periodically, every T seconds. The
sampler produces
( ) ( ) ( ) 8.1.11
n
n
y t a x t nT v t
=
= +
( ) ( ) ( ) ( ) ( ) ( ) and ( ) ( ) ( ) denotes
the additive noise at the output of the receiving filter
R T R R
x t h t g t g t c t g t v t n t g t = = =
( ) ( ) ( ) 8.1.12
n
n
y mT a x mT nT v mT
=
= +
=
= +
+ +
= = =
2
2
0
W
2 2
T T
-W
( ) ( )
= G ( ) C ( ) 8.1.14
h
x h t dt H f df
f f df c
= =
=
} }
}
Digital PAM Transmission through Bandlimited Baseband Channel
The second term on the RHS of Equation (8.1.13) represents the
effect of the other symbols at the sampling instant t = mT, called
the intersymbol interference (ISI).
v
m
is a zero-mean Gaussian random variable with variance
By appropriate design of the transmitting and receiving filter, it
is possible to satisfy the condition x
n
=0 for , so that the ISI
term vanishes
2
0
/ 2
n h
N o c =
0 n =
8.1.2 Digital transmission through Bandlimited Bandpass Channels
The baseband PAM given by in Equation (8.1.9) modulates the
carrier, so that the transmitted signal u(t) is simply
A QAM signal in its simplest form, may be viewed as two amplitude-
modulated carrier signals in phase quadrature. That is, the QAM
signal may be expressed as
and {a
nc
} and {a
ns
} are the two sequence of amplitudes carried on
the two quadrate carriers.
( ) t u
( ) ( ) cos 2 8.1.15
c
u t t f t u t =
( ) ( ) cos 2 ( ) sin 2 8.1.16
where
( ) ( )
8.1.17
( ) ( )
c c s c
c nc T
n
s ns T
n
u t t f t t f t
t a g t nT
t a g t nT
u t u t
u
u
=
= +
=
=
=
=
=
=
=
2
( ) Re ( ) 8.1.19
c
j f t
u t t e
t
u
(
=
Digital transmission through Bandlimited Bandpass Channels
When transmitted through the bandpass channel, the received bandpass
signal may be represented as
where r(t) is the equivalent lowpass (baseband) signal, which may be
expressed as
In the case of baseband transmission, h(t) is the impulse response of
the cascade of the transmitting filter and the channel
2
( ) Re ( ) 8.1.21
c
j f t
w t r t e
t
(
=
( ) ( ) ( ) 8.1.22
n
n
r t a h t nT n t
=
= +
( ) ( ) ( )
T
h t c t g t =
Digital transmission through Bandlimited Bandpass Channels
The received bandpass signal can be converted to a baseband signal by
multiplying w(t) with the carrier signals and and
eliminating the double frequency terms by passing the two quadrature
components through separate lowpass filters, as shown in Figure8.4
The two quadrature components at the output of these lowpass filters
can be expressed as an equivalent complex-valued signal of the form
cos 2
c
f t t sin 2
c
f t t
( ) ( ) ( ) 8.1.23
n
n
y t a x t nT v t
=
= +
=
=
[ ( )] ( ) ( )
( ) 8.2.2
n T
n
a T
n
E V t E a g t nT
m g t nT
=
=
=
= =
+
= =
=
= +
+
+
+
}
}
}
}
8.2.8
8.2.1 The Power Spectrum of the baseband signal
The time-autocorrelation function of g
T
(t) is defined as
The average autocorrelation function of V(t) becomes
The Fourier transform of Equation(8.2.10) becomes
( ) ( ) ( ) 8.2.9
g T T
R g t g t dt t t
= +
}
1
( ) ( ) ( ) 8.2.10
V a g
m
R R m R mT
T
t t
=
=
2
2
2
( ) ( )
1
= ( ) ( )
T
1
= ( ) ( ) 8.2.11
T
j f
V V
j f
a g
m
a T
S f R e d
R m R mT e d
S f G f
t t
t t
t t
t t
=
=
}
2 2
| ( ) |
j fmT
T
G f e
t
8.2.1 The Power Spectrum of the baseband signal
S
a
(f) is the power-spectrum of the information sequence {a
n
}
and G
T
(f) is the spectrum of the transmitting filter. is the Fourier
transform of
The power-spectral density S
a
(f) is periodic in frequency with period 1/T,
we note that S
a
(f) has the form of an exponential Fourier series with
{R
a
(m)} as the Fourier coefficients.
We consider the case in which the information symbols in the sequence
{a
n
} are mutually uncorrelated
2
( ) ( ) 8.2.12
j fmT
a a
m
S f R m e
t
=
=
2
( )
T
G f
( )
g
R t
1/ 2
2
1/ 2
( ) ( ) 8.2.13
T
j fmT
a a
T
R m T S f e df
t
=
}
2 2
2
, 0
( ) 8.2.14
, 0
a a
a
a
m m
R m
m m
o + =
=
=
=
= +
2
2
( ) ( ) 8.2.16
a
a a
m
m m
S f f
T T
o o
=
= +
=
= +
2
2
( )
a
T
G f
T
o
2
2
( ) ( ) 8.2.18
a
V T
S f G f
T
o
=
8.2.1 The Power Spectrum of the baseband signal
Example 8.2.1
Determine the power-spectral density in Equation (8.2.17), when g
T
(t) is
the rectangular pulse
8.2.1 The Power Spectrum of the baseband signal
The Fourier transform of g
T
(t) is
2
2 2 2 2
sin
( )
Hence
sin
( ) ( ) ( ) =( ) sin ( )
j fT
T
T
fT
G f AT e
fT
fT
G f AT AT c fT
fT
t
t
t
t
t
=
=
8.2.1 The Power Spectrum of the baseband signal
By substitution for into Equation (8.2.17)
2
( )
T
G f
2 2 2 2 2
2 2 2 2 2
sin
( ) ( ) ( )
= sin ( ) ( )
V a a
a a
fT
S f A T A m f
fT
A T c fT A m f
t
o o
t
o o
= +
+
8.2.1 The Power Spectrum of the baseband signal
Example 8.2.2
Consider a binary sequence {b
n
}, from which we form the symbols
The {b
n
} are assumed to be uncorrelated binary valued ( ) random
variable, each having a zero mean and a unit variance. Determine
the power-spectral density of the transmitted signal
The autocorrelation function of the sequence {a
n
} is
1 n n n
a b b
= +
1
1 1
( ) [ ]
[( )( )]
2 =0
= 1 = 1
0 otherwise
a n n m
n n n m n m
R m E a a
E b b b b
m
m
+
+
=
= + +
( ) ( ) ( ) 8.3.2
m
n
r t a h t nT n t
=
= +
=
>
( ) (0) ( ) ( ) 8.3.5
m n
n
y mT x a a x mT nT v mT
=
= + +
* *
( ) ( )
T
H f G f =
8.3 SIGNAL DESIGN FOR BANDLIMITED CHANNELS
or more simply,
The middle term on the RHS of Equation(8.3.5) represent the ISI.
The resulting oscilloscope display is called an eye pattern
0
8.3.4
m m n m n m
n m
y x a a x v
=
= + +
( ) ( ) ( )
m
n
r t a x t nT n t
=
= +
1, 0
( ) 8.3.7
0, 0
n
x nT
n
=
=
=
=
=
( ) 8.3.9
m
m
X f T
T
=
+ =
2
( ) ( ) 8.3.10
j ft
x t X f e df
t
=
}
2
( ) ( ) 8.3.11
j fnT
x nT X f e df
t
=
}
8.3.1 Design of Bandlimited Signals for Zero ISIThe Nyquist Criterion
Let us break up the integral in Equation (8.3.11) into integral
covering the finite range of 1/T. thus we obtain
Where we have defined Z(f) by
(2 1) / 2
2
(2 1) / 2
1/ 2
2
1/ 2
1/ 2
2
1/ 2
1/ 2
2
1/ 2
( ) ( )
( )
[ ( )]
( ) 8.3.12
m T
j fnT
m T
m
T
j fnT
T
m
T
j fnT
T
m
T
j fnT
T
x nT X f e df
m
X f e df
T
m
X f e df
T
Z f e dt
t
t
t
t
=
= +
= +
=
}
}
( ) ( ) 8.3.13
m
m
Z f X f
T
=
= +
=
=
1
2
2
1
2
( ) 8.3.15
j nfT
T
n
T
z T Z f e df
t
=
}
( ) 8.3.16
n
z Tx nT =
, 0
8.3.17
0, 0
n
T n
z
n
=
=
=
=
+ =
s s
+
= + s s
>
0, -1
8.3.27
0
n
T n
z
otherwise
=
2
( ) 8.3.28
j fT
Z f T Te
t
= +
8.3.2 Design of Bandlimited Signals with Controlled ISI Partial
Response Signal
For , we obtain
Therefore, x(t) is given by
x(t)=sinc(2Wt)+sinc(2Wt - 1) 8.3.30
This pulse id called a duobinary signal pulse.
1
2
T
W
=
2
1
[1 ],
( )
2
0 otherwise
1
cos ,
= 8.3.29
2
0 otherwise
f
j
W
f
j
W
e f W
X f
W
f
e f W
W W
t
t
t
+ <
=
| |
<
|
\ .
= = =
1
sin ,
2 ( ) 8.3.33
0,
f f
j j
W W
j f
e e f W
W W W X f
f W
t t
t
| |
= s
|
=
\ .
>
/
1
( ) ,
2 2 ( ) 8.3.35
0 ,
jn f W
n
n
x e f W
W W X f
f W
t
>
= =
=
}
8.4 PROBABILITY OF ERROR IN DETECTION OF DIGITAL PAM
The probability of error for digital PAM in a bandlimited, additive
white Gaussian noise channel, in the absence of ISI, is identical to the
Section 7.6.2.
Hence
In the treatment of PAM given this chapter we imposed the additional
constraint that the transmitter signal is bandlimited to the bandwidth
allocated for the channel.
0
2
2( 1)
8.4.4
g
M
M
p Q
M N
c
(
=
(
(
2
3 /( 1),
g av av bav
M k c c c c = =
2
2
0
6(log ) 2( 1)
8.4.5
( 1)
bav
M
M M
p Q
M M N
c
(
=
(
8.4 .2 Symbol-by Symbol Detection of Data with Controlled ISI
In particular, we consider the detection of the duobinary and the
modified duobinary partial response signals.
In the both case, we assume that the desired spectral characteristic X(f)
for the partial response signal is split evenly between the transmitting
and receiving filters; i.e.
For the duobinary signal pulse, the samples at the output of the
receiving filter have the from
Let us ignore the noise for the moment and consider the binary case
where with equal probability. Then b
m
takes on one of three
possible values, b
m
= -2, 0, 2 with corresponding probabilities 1/4, 1/2 ,
1/4.
1/ 2
( ) ( ) ( )
T R
G f G f X f = =
1
8.4.6
m m m m m m
y b v a a v
= + = + +
1
m
a =
8.4 .2 Symbol-by Symbol Detection of Data with Controlled ISI
If is the detected symbol from the (m-1)st signal interval, its
effect on b
m
, the received signal in the m th signal interval, can be
eliminated by subtraction, thus allowing a
m
to be detected. This
process can be repeated sequentially for every received symbol.
The major problem with this procedure is that errors arising from the
additive noise tend to propagate.
Error propagation can be avoided by precoding the data at the
transmitter instead of eliminating the controlled ISI by subtraction at
the receiver.
The precoding is performed on the binary data sequence prior to
modulation. From the data sequence {d
n
} of 1s and 0s that is to be
transmitted, a new sequence {p
m
} is called the precoded sequence
1
, 1, 2,... 8.4.7
m m m
p d p m
= =
1 m
a
8.4 .2 Symbol-by Symbol Detection of Data with Controlled ISI
Then , we set a
m
= -1 if p
m
= 0 and a
m
= 1 if p
m
= 1; i.e., a
m=
2p
m
-1.
The noise free samples at the output of the receiving filter are given
as
Consequently,
Since d
m
=p
m
p
m-1
, it follows that the data sequence d
m
is obtained
from b
m
by using the relation
Consequently, if b
m
= 2, d
m
= 0 and b
m
=0, d
m
=1.
1
-1
1
(2 -1) (2 -1)
2( -1) 8.4.8
m m m
m m
m m
b a a
p p
p p
= +
= +
= +
1
1 8.4.9
2
m
m m
b
p p
+ = +
1 1
( ) ( ) ( mod 2)
1(mod 2) 8.4.10
2
m m m m m
m
d p p p p
b
= = +
= +
=
>
= +
1
2 ( 1)
(mod )
m m
m m m
a p M
p d p M
=
=
1
-1
2[ - ( -1)] 8.4.15
m m m
m m
b a a
p p M
= +
= +
0,1,..., ( 1)
m
p M =
8.4 .2 Symbol-by Symbol Detection of Data with Controlled ISI
Since d
m
=p
m
+ p
m-1
(mod M), if follows that
An example illustrating multilevel precoding and decoding is given in
table 8.2
( 1)(mod ) 8.4.17
2
m
m
b
d M M = +
8.4 .2 Symbol-by Symbol Detection of Data with Controlled ISI
In the case of the modified duobinary pulse, the controlled ISI is
specified by the values x(n/2W) = -1, for n=1, x(n/2W) = 1 for n=-1, and
zero otherwise, the noise-free sampled output from the receiving filter
is given as
According to the relation Equation (8.4.14) and
The detection rule for recovering the data sequence {d
m
} from {b
m
} in
the absence of noise is
2
8.4.18
m m m
b a a
=
2
(mod ) 8.4.19
m m m
p d p M
=
( 1)(mod ) 8.4.20
2
m
m
b
d M M = +
8.4 .3 Probability of Error for Detection of Partial Response Signals
We determine the probability of error for detection of digital M-ary
PAM signaling using duobinary and modified duobinary pulses. The
model for the communications system is shown in Figure 8.14
8.4 .3 Probability of Error for Detection of Partial Response Signals
Symbol-by-Symbol Detector
The precoder output is mapped into one of M possible amplitude levels.
Then the transmitting filter with frequency response G
T
(f) has an
output
The partial-response function X(f) is divided equally between the
transmitting and receiving filters.
The matched filter output is sampled at t= nT = n/2W. For the
duobinary signal , the output of the matched filter at the sampling
instant may be expressed as
( ) ( ) 8.4.21
n T
n
t a g t nT u
=
=
( ) ( ) ( ) 8.4.22
T R
G f G f X f =
1
m m m m
m m
y a a v
b v
= + +
= +
, 3 ,..., ( 1)
0, 2 , 4 ,..., 2( 1)
m
m
a d d M d
b d d M d
=
=
8.4 .3 Probability of Error for Detection of Partial Response Signals
And for the modified duobinary signal is
The input transmitted symbols {a
m
} are assumed to be equally probable.
Then, for duobinary and modified duobinary signals, the received
output levels have a (triangular) probability mass function of the form
where b denotes the noise-free received level and 2d is the distance
between any two adjacent received signal levels.
We assume that a symbol error is committed whenever the magnitude of
the additive noise exceeds the distance d. The noise component v
m
is zero-
mean, Gaussian with variance
2
m m m m
m m
y a a v
b v
= +
= +
2
( 2 ) , 0, 1, 2,..., ( -1) 8.4.25
M m
p b md m M
M
= = =
2
2
0
0
0
( )
2
= ( ) 2 / 8.4.26
2
W
v R
W
W
W
N
G f df
N
X f df N
o
t
=
=
}
}
, 3 ,..., ( 1)
0, 2 , 4 ,..., 2( 1)
m
m
a d d M d
b d d M d
=
=
8.4 .3 Probability of Error for Detection of Partial Response Signals
For both the duobinary and the modified duobinary signals. An upper
bound on the symbol probability of error is
But
2
( 2)
-1
0
2
( 2 2 ) ( 2 )
2 ( 2( -1) -2( -1) ) ( -2( -1) )
( 0) 2 ( 2 ) ( 0) ( 2( 1) )
1
(1- ) ( 0)
M
M
m M
M
m
p p y md d b md p b md
p y M d d b M d p b m d
p y d b p b md p b p b M d
p y d b
M
=
=
< > = =
+ + > = =
(
= > = = = =
(
= > =
8.4.27
2 2
/ 2
2
0
2
( 0)
2
=2Q 8.4.28
2
v
x
d
v
p y d b e dx
d
N
o
to
t
> = =
| |
|
|
\ .
}
8.4 .3 Probability of Error for Detection of Partial Response Signals
The average probability of a symbol error is upper-bounded as
For the M-ary PAM signal in which the transmitted levels are equally
probable, the average power at the output of the transmitting filter is
where is the mean square value of the M signal levels,
2
2
0
1
2(1 ) 8.4.29
2
M
d
p Q
M N
t
| |
< |
|
\ .
2
2
2
2
( )
( )
( ) 4
( ) ( ) 8.4.30
w
m
av T
w
w
m
m
w
E a
p G f df
T
E a
X f df E a
T T t
=
= =
}
}
2
( )
m
E a
2 2
2
( 1)
( ) 8.4.31
3
m
d M
E a
=
8.4 .3 Probability of Error for Detection of Partial Response Signals
therefore,
By substituting the value of from Equation (8.4.32) into Equation
(8.4.29), we obtain the upper-bound for the symbol error probability as
Where
av
is the average energy/transmitted symbol, which can be also
expressed in terms of the average bit energy as
av
=k
bav
=(log
2
M)
bav
.
The expression in Equation (8.4.33) for the probability of error of M-ary
PAM holds for both a duobinary and a modified duobinary partial
response signal.
2
2
3
8.4.32
4( 1)
av
p T
d
M
t
=
2
d
2
2 2
0
1 6
2(1 )
4 1
av
M
p Q
M M N
c t
| |
| |
|
<
|
|
\ .
\ .
2
2
0
6(log ) 2( 1)
8.4.5
( 1)
bav
M
M M
p Q
M M N
c
(
=
(
8.4 .3 Probability of Error for Detection of Partial Response Signals
If we compare this result with the error probability of M-ary PAM with
zero ISI, which can be obtained by using a signal pulse with a raised
spectrum, we note that the performance of partial response duobinary
or modified duobinary has a loss of (/4)
2
or 2.1db.
To observe the memory in the received sequence, let us look at the noise-
free received sequence for binary transmission given in Table 8.1.
The sequence {b
m
} is 0,-2,0,2,0,-2,0,2,2,... We note that it is not possible to
have a transition from -2 to +2 or from +2 to -2 in one symbol interval.
In other words, it is not possible to encounter a transition form -2 to +2
or vice versa between two successive received samples from the matched
filter.
8.5 DIGITALLY MODULATED SIGNALS WITH MEMORY
We observed that we can shape the spectrum of the transmitted signal
by introducing memory in the modulation,. The two examples cited in
that section are the duobinary and modified duobinary partial
response signal.
Signal dependence among signals transmitted in different signal
intervals is generally accomplished by encoding the data at the input to
the modulator by means of a modulation code.
Such a code generally places restrictions on the sequence of symbols
into the modulator and introduces memory in the transmitted signal.
Signal dependence among signals transmitted in different signal
intervals can also result from intersymbol interference introduced by
channel distortion.
8.5.1 Modulation Codes and Modulation Signals with Memory
Modulation codes are usually employed in magnetic recoding, in optical
recording, and in digital communications over cable systems to achieve
spectral shaping of the modulated signal that matches the passband
characteristics of the channel.
In magnetic recoding, we encounter two basic problems . One problem
is concerned with the packing density that is used to write the data on
the magnetic medium (disk or tape).
8.5.1 Modulation Codes and Modulation Signals with Memory
The binary data sequence to be stored is used to generate a write
current. This current may be viewed as the output of the modulator.
The two most commonly used methods to map the data sequence into
the write current waveform ate the so-called NRZ (non-return-to -zero)
and NRZI (non-return-to zero-inverse) methods.
We note that is identical to binary PAM in which the information bit 1
is represented by a rectangular pulse of amplitude A and the
information bit 0 is represented by a rectangular pulse of amplitude
A .
In contrast, the NRZI signal waveform is different from NRZ in that
transitions from one amplitude level to another (A to A or A to
A ),the amplitude level remains the same as the previous signal level
The positive amplitude pulse results in magnetizing the medium on one
(direction) polarity and the negative pulse magnetizes the medium in
the opposite (direction) polarity.
8.5.1 Modulation Codes and Modulation Signals with Memory
8.5.1 Modulation Codes and Modulation Signals with Memory
Since the input data sequence is basically random with equally probable
1s and 0s, whether we use NRZ of NRZI, we will encounter level
transitions for A to A or A to A with probability for every data bit
The readback signal for a positive transition (-A to A) is a pulse that is
well modeled mathematically as
Where T
50
is defined as the width of the pulse at its 50% amplitude level
2
50
1
( ) 8.5.1
1 (2 / )
p t
t T
=
+
8.5.1 Modulation Codes and Modulation Signals with Memory
The readback signal for a negative transition (A to - A) is the pulse p(t).
The value of T
50
is determined by the characteristics of the medium and
the read/write heads.
Now, suppose we write a positive transition followed by a negative
transition, and let us vary the time interval between the two transitions,
which we denote as T
b
(the bit time interval). Figure 8.18 illustrates the
readback signal pulses, which are obtained by a superposition of p(t)
with p(t - T
b
).
The parameter, =T
50
/T
b
, is defined as the normalized density.
We notice that as is increased, the peak amplitudes of the readback
signal are reduced and are also shifted in time from the desired time
instants. In the other words, the pulses interfere with one another, thus,
limiting the density with which we can write.
8.5.1 Modulation Codes and Modulation Signals with Memory
8.5.1 Modulation Codes and Modulation Signals with Memory
This problem serves as a motivation to design modulation codes that
take the original data sequence and transform (encode) it into another
sequence that results in a write waveform in which amplitude
transition are spaced further apart.
For example, if we use NRZI, the encoded sequence into the modulator
must contain one or more 0s between 1s.
The second problem encountered in magnetic recording is the need to
avoid (or minimize) having a dc content in the modulated signal (the
write current), due to the frequency-response characteristics of the
readback system and associated electronics
This problem can also be overcome by altering (encoding) the data
sequence into the modulator.
8.5.1 Modulation Codes and Modulation Signals with Memory
Runlength-Limited Codes
Codes that have a restriction on the number of consecutive 1s or 0s in
a sequence are generally called runlenght-limited code. These codes ate
generally described by two parameters, say d and , where d denotes
the minimum number of 0s between 1s in a sequence, and denotes
the maximum number of 0s between two 1s in a sequence.
When used with NRZI modulation, the effect of placing d zeros
between successive 1s is to spread the transition farther apart, thus,
reducing the overlap in the channel response due to successive
transition. By setting an upper limit on the runlength of 0s ensures
that transitions occur frequently enough so that symbol timing
information can be recovered from the received modulated signal.
Runlength-limited codes are usually called (d , ) codes.
8.5.1 Modulation Codes and Modulation Signals with Memory
The (d, k)-code sequence constraints may be represented by a finite-
state sequential machine with +1 states, denoted as
i
, 1 i +1, as
shown in Figure 8.19
We observe that an output data bit 0 takes the sequence from state
i
to
i+1
, i . The output data bit a takes the sequence to state
i
, the
output bit from the encoder may be a 1 only when the sequence is in
state
i
, d i +1. when the sequence is in state
i+1
, the output bit is
always 1.
8.5.1 Modulation Codes and Modulation Signals with Memory
The finite-state sequential machine may also be represented by a state
transition matrix, denoted as D, which is a square (+1) (+1) matrix
with elements d
ij
, where
d
ij
= 1, i d +1
d
ij
= 1, j = i +1 (8.5.2)
d
ij
= 0, otherwise
Example 8.5.1
Determine the state transition matrix for a (d , k) = (1 , 3) code
The (1,3) code has four states, we obtain its state transition matrix
which is
0 1 0 0
1 0 1 0
1 0 0 1
1 0 0 0
D
(
(
(
=
(
(
8.5.1 Modulation Codes and Modulation Signals with Memory
An important parameter of any (d,)code is the number of sequences
of a certain length, say n, which satisfy the (d, ) constraints. The
number of information bits that can be uniquely represented with N(n)
code sequences is
k = [log
2
N(n)]
where [x] denotes the largest integer contained in x. The maximum
code rate is then R
c
=k/n.
The capacity of a (d, ) code is defined as
C (d, ) is the maximum possible rate that can be achieved with the (d,
) constraints.
Where is the largest real eigenvalue of the state transition matrix
D.
2
1
( , ) lim log ( ) (8.5.4)
n
C d k N n
n
=
2 max
( , ) log (8.5.5) C d k =
max
max
= 0.5515.
capacities of (d, ) codes for 0 d 6 and 2 k 15, are given in Table 8.3.
We observe that C (d, ) < for d 3, and any value of . The most
commonly used codes for magnetic recording employ a d 2, hence, their
rate R
c
is at least 1/2
4 2
- 1 0 0
1 - 1 0
det(D I) det
1 0 - 1
1 0 0 -
= 1 0 8.5.6
(
(
(
=
(
(
=
8.5.1 Modulation Codes and Modulation Signals with Memory
8.5.1 Modulation Codes and Modulation Signals with Memory
Now, let us turn our attention to the construction of some runlength-
limited codes.
In general, (d, ) codes can be constructed either as fixed length codes
or variable-length codes. In a fixed-length code, each bit or block of k
bits, is encoded into a block of n > k bits.
For a given block length n, we may select the subset of the 2n
codewords that satisfy the specified runlength constraints. From this
subset, we eliminate codewords that do not satisfy the runlength
constraint when concatenated. Thus, we obtain a set of codewords that
satisfies the constraints and can be use in the mapping of the input
data bits of the encoder.
8.5.1 Modulation Codes and Modulation Signals with Memory
Example 8.5.3
Construct a d = 0, = 2 code of length n = 3, and determine its efficiency.
By listing all the codewords, we find that the following five codewords
satisfy the (0,2) constraint: (010), (011), (101), (110), (111).
We may select any four of these codewords and use then to encode the
pairs of data bits (00, 01, 10, 11). Thus, we have a rate k/n = 2/3 code that
satisfies the (0,2) constraint.
The fixed-length code in this example is not very efficient. The capacity
of a (0,2) code is C(0,2) = 0.8791, so that this code has an efficiency of
Surely, better (0,2) codes can be constructed by increasing the block
length n
2/ 3
0.76
( , ) 0.8791
c
R
efficiency
C d k
= = =
8.5.1 Modulation Codes and Modulation Signals with Memory
Example 8.5.4
construct a d = 1 , = code of length n = 5.
In this case, we are placing no constraint on the number of
consecutive zeros.
There are eight such codewords , which implies that we can encode
three information bits with each codeword. The code is given in Table
8.4
8.5.1 Modulation Codes and Modulation Signals with Memory
This code has a rate R
c
= 3/5. when compared with the capacity C (1,)
= 0.6942, obtained from Table 8.3, the code efficiency is 0.864, which is
quite acceptable.
The code construction method described in the two example above
produces fixed-length (d, ) codes that are state independent. By state
independent, we mean that fixed-length codewords can be
concatenated without violating the (d, ) constraints.
In general, fixed-length, state-independent (d, ) coeds require large
block lengths, except in cases such as those in the examples above,
where d is small.
8.5.1 Modulation Codes and Modulation Signals with Memory
Example 8.5.5
Construct a simple variable-length d = 0, = 2 code.
A very simple uniquely decodable (0,2) code is the following:
001
1010
1111
The code in above example has a fixed output block size but a variable
input block size. In general, both the input and output blocks may be
variable.
Example 8.5.6
Construct a (2,7) variable block size code.
The solution to this code construction is certainly not unique nor is it
trivial.
8.5.1 Modulation Codes and Modulation Signals with Memory
The code is listed in Table 8.5. we observe that the input data blocks of
2, 3, and 4 bits are mapped into output data blocks of 4, 6, and 8
bits ,respectively.
8.5.1 Modulation Codes and Modulation Signals with Memory
The code rate is R
c
= 1/2. this code rate for all codewords, the code is
called a fixed-rate code.
Another code that has been widely sued in magnetic recording is the
rate 1/2, (d, ) = (1,3) code given in Table 8.6
We observe that the first bit of the two-bit block is redundant and may
be discarded.
The second bit is the information bit . This code is usually called the
Miller code.
8.5.1 Modulation Codes and Modulation Signals with Memory
We observe that this is a state-dependent code which is described by
the state diagram shown in Figure 8.20
When the encoder is a state S
1
, an input bit 1 results in the encoder
staying in state S
1
and outputs 01. this denoted as 1/01. If the input bit
is a 0 the encoder enters state S
2
and output 01. This is denoted as 0/00.
Similarly, if the encoder is in state S
2
, an input bit 0 causes no
transition and the encoder output is 10. On the other hand , if the
input bit is a 1, the encoder enters state S
1
and output 01.
8.5.1 Modulation Codes and Modulation Signals with Memory
Trellis Representation of State- Dependent (d, ) Codes.
The state diagram provides a relatively compact representation of a
state-dependent code.
Another way to describe such codes that have memory is by means of a
graph called a trellis.
A trellis is a graph that illustrates the state transitions as a function of
time.
For example, Figure 8.21 shows the trellis for the d = 1, = 3 Miller
code whose state diagram is shown in Figure 8.20
8.5.2 The Maximum-Likelihood Sequence Detector
When the transmitted signal has memory; i.e., the signals transmitted
in successive symbol intervals are interdependent, the optimum
detector bases its decisions on observation of a sequence of received
signals over successive signal intervals.
Consider as an example the NRZI signal described in Section 8.5.1.
The signal transmitted in each signal interval is binary PAM. Hence,
there are two possible transmitted signals corresponding to the signal
points s
1
= -s
2
= , where
b
is the energy/bit. The output of the
matched-filter or correlation-type demodulator for binary PAM in the
kth signal interval may be expressed as
where n
k
is a zero-mean Gaussian random variable with variance
b
c
8.5.10
k b k
r n c = +
2
0
/ 2
n
N o =
8.5.2 The Maximum-Likelihood Sequence Detector
The conditional PDFs for the two possible transmitted signals are
8.5.11
Now, suppose we observe the sequence of matched-filter outputs r
1
,r
2
,...,r
k
.
The channel noise is assumed to be white and Gaussian, and E(n
k
n
j
)=0,
kj, the noise sequence n
1
,n
2
,...,n
k
is also white.
2 2
2 2
( ) / 2
1
( ) / 2
2
1
( )
2
1
( )
2
k b n
k b n
r
k
n
r
k
n
f r s e
f r s e
c o
c o
to
to
=
=
8.5.2 The Maximum-Likelihood Sequence Detector
Consequently, for any given transmitted sequence s
(m)
, the joint PDF
of r
1
,r
2
,...,r
k
. May be expressed as a product of k marginal PDFs; i,e.,
Where either s
k
= or s
k
=- . Then , given the received sequence
r
1
,r
2
,...,r
k
at the output of the matched filter or correlation type
demodulator, the detector determines the sequence
that maximizes the conditional PDF p( ). Such a detector
is called the maximum-likelihood (ML) sequence detector.
( ) 2 2
1 2
1
( ) / 2
1
( ) 2 2
1
( , ,..., ) ( )
1
2
1
( ) exp[ ( ) / 2 ] 8.5.12
2
m
m
k n k
K
m
k k k
k
K
r s
k
n
K
k m
k k n
k
n
p r r r s p r s
e
r s
o
to
o
to
=
=
=
=
=
=
[
[
b
c
b
c
( )
1 2
{ , ,... }
m m m m
K
s s s s =
1 2
, ,...,
m
k
r r r s
8.5.2 The Maximum-Likelihood Sequence Detector
By taking the logarithm of Equation (8.5.12) and neglecting the terms
that are independent of ( r
1
,r
2
,...,r
k
) we find that an equivalent ML
sequence detector selects the sequence s
(m)
that minimizes the
Euclidean distance metric
In searching through the trellis for the sequence that minimizes the
Euclidean distance D( r, s
(m)
), it may appear that we must compute the
distance D( r, s
(m)
) for every possible path (sequence). For the NRZI
example given above, which employs binary modulation, the total
number of paths is 2
k
, where K is the number of outputs obtained from
the demodulator.
We may reduce the number of sequences in the trellis search by using
the Viterbi algorithm to eliminate sequences as new data is received
from the demodulator.
( ) ( ) 2
1
( , ) ( ) 8.5.13
K
m m
k k
k
D r s r s
=
=
=
+ =
( ) ( ) ( ) ( )
T R rc
G f C f G f X f =
System Design In The Presence Of Channel
Distortion
Two types of distortion
amplitude distortion
phase distortion
Amplitude distortion result when is not constant for
Phase distortion result when is nonlinear
envelope delay:
when is linear, is constant.
( ) C f
f W s
( )
C
f u
( ) 1
( )
2
C
d f
f
df
u
t
t
=
( )
C
f u ( ) f t
System Design In The Presence Of Channel
Distortion
However, when is nonlinear and the various frequency
component in the input signal undergo different delays in passing
through the channel. In such a case we say that transmitted signal
has suffered from delay distortion.
Both amplitude and delay distortion cause ISI in received signal.
( )
C
f u
System Design In The Presence Of Channel
Distortion
For example, let us assume that we have designed a pulse with a
raised cosine spectrum that has zero ISI at the sampling instants.
An example of such a pulse is illustrated in Figure 8.33(a).When the
pulse is passed through a channel filter with for and
a quadratic phase characteristic (linear envelop delay).
The received pulse at output of the channel is shown in Figure
8.33(b).
( ) 1 C f = f W <
Note that the periodic zero crossing have been shifted by the delay distortion.
8.6.1 Design of Transmitting and Receiving Filters
for a known channel
We assume that is known and design
that maximize the SNR at the output of the receiving filter
and result in zero ISI.
( ) C f
( ), ( )
T R
G f G f
0
( ) ( ) ( ) ( ) ,
j ft
T R rc
G f C f G f X f e f W
t
= s
( )
rc
X f
where is the desired raised cosine spectrum that
yields zero ISI at sampling instants.
: time delay
The noise at the output of the receiving filter:
2
( ) ( ) ( )
( ) ( ) ( )
R
v n R
v t n t g d
S f S f G f
t t t
=
=
}
0
t
Design of Transmitting and Receiving Filters
for a known channel
For binary PAM transmission, the sampled output of the
matched filter is
where
is normalized to unity,
Probability of error
0 m m m m m
y x a v a v = + = +
0
x
2
2
: the noise term, zero-mean Gaussian with variance
( ) ( )
m
m
v n R
a d
v
S f G f df o
=
=
}
2
2
2
2
2
1
2
v
y
d
v
d
P e dy Q
o
o
t
| |
= = |
|
\ .
}
Design of Transmitting and Receiving Filters
for a known channel
Now suppose that we select the filter at the transmitter to have the
frequency response
where is a suitable delay to ensure causality.
The filter at receiver is designed to be matched to the receiver
signal pulse.
where is an appropriate delay.
( )
( )
( )
0
2
rc
j ft
T
X f
G f e
C f
t
=
0
t
( ) ( ) ( )
0
2
j ft
T rc
G f C f X f e
t
==> =
( )
2
( )
r
j ft
R rc
G f X f e
t
==> =
r
t
( ) ( )
( )
( )
( )
( )
( )
( )
( )
( )
2
2
2
2
0 0 0
2
2
2
2
1
2
1
2
0
2 2 2
.
2 .
v
W
v R rc
W
W
m
rc
av T
W
W
rc
av
W
W
rc
av
W
d
SNR
N N N
G f df X f df
E a
X f
d
P g t dt df
T T
C f
X f
d P T df
C f
X f
P T
SNR df
N
C f
o
o
=
= = =
= =
(
(
=
(
(
(
==> =
(
} }
} }
}
}
( )
( )
10
2
10log
W
rc
W
X f
df
C f
}
( )
( )
1 , in
1 ,
C f for f W loss performance
C f for f W noperformanceloss
s s =>
= s =>
We note that the term,
Design of Transmitting and Receiving Filters
for a known channel
example 8.6.1 : Determine and for a binary
communication system that transmits data at a rate of 4800 bits/sec .
( )
T
G f
( )
R
G f
( )
( ) ( )
2
15 0
1
, , 4800
1
: , , , 10 /
2
n
C f f f W Hz
f
W
N
n t zero mean white Gaussian S f W Hz
= s =
| |
+
|
\ .
= =
( )
( )
( )
( ) ( )
2
2
:
1
4800
1 cos( ) cos
2 9600
, 1 cos , 4800
9600
cos , 4800
9600
0 4800
rc
T
R
T R
Solution
W
T
f
T
X f T f T
f
f
Then G f T f Hz
W
f
G f T f Hz
and G f G f for f Hz
t
t
t
t
= =
| |
= + ( =
|
\ .
(
| |
| |
= + s
( | |
\ .
( \ .
| |
= s
|
\ .
= = >
8.6.2 Channel Equalization
In practice we often encounter channel whose frequency
response are either unknown or change with time. We may
design the transmitting filter to have a square root raised
cosine frequency response ; i.e.,
and the receiving filter, with frequency response , to be
matched to .
( )
( )
0
2
,
0 ,
j ft
rc
T
X f e f W
G f
f W
t
=
>
( )
R
G f
( )
T
G f
( ) ( ) ( )
T R rc
G f G f X f ==> =
Channel Equalization
The output of the receiving filter is
( ) ( )
( ) ( ) ( ) ( )
( )
n
n
T R
y t a x t nT v t
where x t g t c t g t
=
= +
= - -
( )
0
, 0, 1, 2,..
m n n m m
n
m n n m m
n
n m
ISI
n
y a x v
x a a x v
where x x nT n
=
=
= +
= + +
= =
Channel Equalization
In any practical system, it is reasonable to assume that the ISI affect
a finite number of symbols.
Hence, we may assume for and ,where
and are finite, positive integer.
Consequently, the ISI observed at the output of the receiving filter
may viewed as being generated by passing the data sequence
through an FIR filter with coefficients , as
shown in Figure 8.35.
This filter is called the equivalent discrete-time channel filter.
0
n
x =
1
n L <
2
n L >
1
L
2
L
{ }
m
a
{ }
1 2
,
n
x L n L s s
Channel Equalization
Maximum-Likelihood Sequence Detection
The optimum detector for the information sequence based
on the observation of the received sequence is a ML
detector.
The Viterbi algorithm provides a method for searching through the
trellis for the ML signal path.
Linear Equalizer
To compensate for the channel distortion, we may employ a linear
filter with adjustable parameters. These adjustable filter are called
channel equalizers or simply equalizers
First, we consider the design characteristic for a linear equalizer
from a frequency domain viewpoint.
{ }
m
a
{ }
m
y
( ) ( ) ( )
( )
( )
( )
1
C
T R rc
j f
E
G f G f X f
G f e f W
C f
u
=
= s
Channel Equalization
In this case, the equalizer is said to be the inverse channel filter to
the channel response. Since it forces the ISI to be zero at the
sampling times , the equalizer is called zero-forcing
equalizer.
t nT =
Channel Equalization
Hence, the input to the detector is of the form
where is the noise component, which is zero-mean Gaussian
with a variance
When noise is white
m
v
m m m
y a v = +
( ) ( ) ( )
( ) ( )
( )
2 2
2
2
v n R E
W
n rc
W
S f G f G f df
S f X f
df
C f
o
=
=
}
}
( )
( )
( )
2
0 0
2
,
2 2
W
rc
n v
W
X f
N N
S f df
C f
o
= =
}
Channel Equalization
example 8.6.2 : The channel given in Example 8.6.1 is equalized by
a zero-forcing equalizer.
Assume ,determine the value of the noise
variance at the sampling instants and probability of error.
( ) ( ) ( )
T R rc
G f G f X f =
Channel Equalization
( )
( )
( )
( )
( )
2 0
2
2
2 0
1
2 2
0
0
0
2
2 2
2
:
2
1 cos
2 2
1 cos
2
2 1
3
1
3
W
rc
v
W
W
W
av T
solution
X f
N
df
C f
f
TN f
df
W W
x
N x dx
N
M d
P G f
T
o
t
t
t
=
(
| |
| |
= +
( | |
\ .
( \ .
| |
= +
|
\ .
| |
=
|
\ .
=
}
}
}
( )
( )
( )
( )
( )
( )
2 2
2 2
2
2 0
1
3
1
3
2 1
3
2 1
1
3
W W
rc
W W
av
M
M d
df X f df
T
M d
T
M
P T
P Q
M
M N
t
=
| |
|
=
|
|
\ .
} }
Channel Equalization
Let us now consider the design of a linear equalizer from a time-
domain view-point.
In practice the channel equalizer is approximated by a finite duration
impulse response (FIR) filter, or a transversal filter, with adjustable
taps coefficients , as illustrated in Figure 8.37.
{ }
n
c
Channel Equalization
In practice , is often selected as . Notice that, in this case,
the sampling rate at the output of the filter is .
where are the equalizer coefficients, and .
and is the signal pulse corresponding
to . Then the equalized output signal pulse is
t
2
T
t =
( )
R
G f
2
T
( ) ( )
2
( )
N N
j fn
E n E n
n N n N
g t c t n G f c e
t t
o t
= =
= =
{ }
n
c
( )
2 1 N+ 2 1 N L + >
( ) ( ) ( ) ( )
T R
X f G f C f G f =
( )
x t
( )
X f
( ) ( )
N
n
n N
q t c x t nt
=
=
Channel Equalization
The zero-forcing condition can now be applied to the samples of
taken at times .
where is a matrix with element
is the coefficient vector and is the
column vector with one nonzero element.
Solution
( )
q t
t mT =
( ) ( )
1 , 0
0 , 1, 2,......
N
n
n N
m
q mT c x mT n
m N
t
=
=
= =
=
==>
Xc = q
X
(2 1) (2 1) N N + + ( ) { }
, x mT nt
c
(2 1) N + q
(2 1) N +
1
= c X q
Channel Equalization
Example 8.6.3: Consider a channel distorted pulse
where is the symbol rate. The pulse is sampled at
the rate and equalized by a zero-forcing equalizer.
Determine the coefficient of a five-tap zero-forcing equalizer.
( )
2
1
2
1
x t
t
T
=
| |
+
|
\ .
1
T
2
T
Channel Equalization
( )
( )
2
2
:
1 , 0
2
0 , 1, 2
1 1 1 1 1
5 10 17 26 37
1 1 1 1
1
2 5 10 17
1 1 1 1
X 1
5 2 2 5
1 1 1 1
1
17 10 5 2
1 1 1 1 1
37 26 17 10 5
n
n
solution
m
nT
g mT c x mT
m
=
=
= =
=
(
(
(
(
(
=
2
1
0
1
2
-1
opt
0
0
c q 1
0
0
2.2
4.9
c X q 3
4.9
2.2
c
c
c
c
c
(
(
(
(
(
( (
( (
( (
( ( = =
( (
( (
( (
(
(
(
( = =
(
(
(
Equalized
Output
Channel Equalization
One drawback to the zero-forcing equalizer is that it ignores the
presence of additive noise.
Let us consider the noise-corrupted output of the FIR equalizer which
is , where is the input to the equalizer,
given by
( ) ( )
N
n
n N
z t c y t nt
=
=
( )
y t
( ) ( ) ( )
.
N
n
n N
y t a x t nT v t
=
= +
( ) ( )
( )
( )
( ) ( ) ( ) ( )
2
2
2
, :
2 8.6.33
N
n
n N
m m
N
n m
n N
N N N
n k Y k AY m
n N k N k N
z mT c y mT n
J MSE E z mT a a transmitted symbol
E c y mT n a
c c R n k c R k E a
t
t
=
=
= = =
- =
= = (
(
=
(
= +
Channel Equalization
The MMSE solution is obtained by differentiating Equation(8.6.33)
( ) ( ) ( )
( ) ( )
Y
AY m
R n k E y mT n y mT k
R k E y mT k a
t t
t
= (
= (
( ) ( )
( ) ( ) ( )
( ) ( )
1
1
0
, 0, 1, 2,...,
1
1
k
N
n Y YA
n N
K
Y
k
K
AY k
k
J
c
c R n k R k k N
R n y kT n y kT
K
R n y kT n a
K
t
t
=
=
=
c
=
c
==> = =
=
=
Channel Equalization
Adaptive Equalizer
where is a matrix, is a column vector
representing the equalizer coefficients, and is a
dimensional column vector.
( ) ( )
( ) ( )
1 , 0
Zero Forcing:
0 , 1, 2,...,
MSE: , 0, 1, 2,....,
N
n
n N
N
n Y YA
n N
m
q mT c x mT n
m N
c R n k R k k N
t
=
=
=
= =
=
= =
==>
Bc = d
B
(2 1) (2 1) N N + + c
2 1 N+
d
(2 1) N +
-1
opt
c B d =
The equalizer coefficients can be updated step-by-step.
Choose arbitrarily the initial coefficient vector .
At the (m+1)th step the filter coefficients are updated
according to
The gradient vector
is the step-size parameter for the iterative procedure
As and
as illustrated in Figure 8.38
0
c
1 m m m +
= A c c g
, 0,1, 2,...
m m
i = = g Bc d
A
0
m
g
m
opt m
c c
Channel Equalization
Let be the nth coefficient of the m-th iteration,
for
From
the nth coefficient of the gradient vector can be expressed as
The term represents the filter output at
time m.
m,n
c
n=-N,-N+1,...,N
, 0,1, 2,...
m m
i = = g Bc d
,
,
[ ( ) ( )] [ ( ) ]
( ) ( )
N
m,n m k m
k N
N
m k m
n N
g E y mT n y mT k c E y mT n a
E y mT n y mT k c a
t t t
t t
=
=
=
(
| |
=
( |
\ .
,
( )
N
m m k
n N
z y mT k c t
=
=
(
y
[ ]
m m m
E e = g y
m
a
m
z
The algorithm for adjusting the equalizer tap coefficients may
be express as
The gradient vector can be approximated by
stochastic gradient algorithm (LMS algorithm)
m+1 m m
A c = c - g
m m m
g e = y
m+1 m m
m+1 m m m
m m m
e
g e
A
==> A
`
=
)
c = c - g
c = c + y
y
Channel Equalization
At the demodulator, the equalizer employs a known pseudo random
sequence to adjust its coefficients.
Upon initial adjustment, the adaptive equalizer switches from a
training mode to a decision-directed mode, in which case the error
signal is .
where is the output of the detector.
the step-size parameter:
where denotes the received signal-plus-noise power, which can
be estimated from the received signal.
{ }
m
a
m
e
m m
a z
. .,
m m m
i e e a z =
m
a
1
5(2 1)
R
N P
A =
+
R
P
Channel Equalization
The convergence characteristics of the stochastic gradient algorithm
is illustrated in Figure 8.40.
Channel Equalization
The block diagram of an adaptive zero-forcing equalizer is shown in
Figure 8.41.
Channel Equalization
Decision-Feedback Equalizer
The severity of the ISI is directly related to the spectral
characteristics and not necessarily to the time span of the ISI.
For example, consider the ISI resulting from two channel which are
illustrated in Figure 8.42.
Channel Equalization
In spite of the shorter ISI span , Channel B results in more severe ISI.
This is evidenced in frequency response characteristics of these
channels, which are shown in Figure 8.43.
Channel Equalization
The noise in Channel B will be enhanced much more than in Channel
A.
This implies that the performance of the linear equalized for Channel
B will be significantly poorer than that for Channel A.
This fact is borne out by the computer simulation results for the
performance of the linear equalizer for two channels, as shown in
Figure 8.44.
Channel Equalization
Hence, the basic limitation of a linear equalizer is that it performs
poorly on channels having spectral nulls.
Channel Equalization
A decision-feedback equalizer (DFE) is a nonlinear equalizer.
where , are the previously detected symbols.
The tap coefficients of the feedforward and feedback filters are
selected to optimized some desired performance measure.
For mathematical simplicity, the MSE criterion is usually applied
and a stochastic gradient algorithm is commonly used to implement
an adaptive DFE.
( ) y t
{ },
n
c
1
N
m
z
m
a
{ },
n
b
2
N
( )
1 2
1 1
N N
m n
m n n
n n
z c y mT n b a t
= =
=
m n
a
2
1, 2,..., n N =
Channel Equalization
Figure 8.46 illustrated the block diagram of an adaptive DFE whose
tap coefficients are adjusted by means of LMS stochastic gradient
algorithm.
Channel Equalization
Figure 8.47 illustrated the probability of error performance of the
DFE, for binary PAM transmission over Channel B
Channel Equalization
Although the DFE outperforms a linear equalizer, it is not the optimum
equalizer from the viewpoint of minimize the probability
of error.
The performance of the ML sequence detector is about 4.5 dB
better than that of the DFE at an error probability of .
4
10
K
1
T
8.7.1 An OFDM System Implemented via
the FFT Algorithm
Hence, the number of signal points for the th subchannel is
.
Let us denote the complex-valued signals points corresponding
the information on the subchannels by
.
represent the values of the DFT of a OFDM signal .
Since must be real-valued signal, its -point must satisfy
the symmetry property .
Therefore, we create symbols from information
symbols by defining
i
2
i
b
i
M =
, 0,1,..., 1
k
X k K =
{ }
k
X
( )
x t
( )
x t
N k k
X X
-
=
N
2 N K =
K
( )
( )
'
0 0
0
1, 2,...., 1
Re
Im
N k k
N
X X k K
X X
X X
-
= =
=
=
8.7.1 An OFDM System Implemented via
the FFT Algorithm
If we denote the new sequence of symbol as
, the -point IDFT yields the real-valued sequence.
where is simply a scale factor.
This sequence corresponds to samples of .
{ }
'
, 0,1,... 1
k
X k N =
N
1
2
'
0
1
, 0,1, , , , 1
N
k
j n
N
n k
k
x X e n N
N
t
=
= =
1
N
{ }
, 0 1
n
x n N s s
( )
x t
( )
( )
1
2
'
0
1
0
: signal duration
N
t
j k
T
k
k
n
x t X e t T
N
nT
T x x
N
t
=
= s s
=