CHAPTER 10

Problem 10.1 :
Suppose that am = +1 is the transmitted signal. Then the probability of error will be :
Pe|1 = P (ym < 0|am = +1)
= P (1 + nm + im < 0)
1
1
P (1/2 + nm < 0) + P
=
4

4




1
3
1
1
=
+ Q
+
Q
4
2n
4
2n

1
(3/2 + nm < 0) + P (1 + nm < 0)
2


1
1
Q
2
n

Due to the symmetry of the intersymbol interference, the probability of error, when am = 1 is
transmitted, is the same. Thus, the above result is the average probability of error.

Problem 10.2 :
(a) If the transmitted signal is :
r(t) =


n=

In h(t nT ) + n(t)

then the output of the receiving lter is :

y(t) =


n=

In x(t nT ) + (t)

where x(t) = h(t)  h(t) and (t) = n(t)  h(t). If the sampling time is o by 10%, then the
1
1
samples at the output of the correlator are taken at t = (m 10
)T . Assuming that t = (m 10
)T
without loss of generality, then the sampled sequence is :
ym =


n=

In x((m

1
1
T nT ) + ((m )T )
10
10


1
If the signal pulse is rectangular with amplitude A and duration T , then
n= In x((m 10 T
nT ) is nonzero only for n = m and n = m 1 and therefore, the sampled sequence is given by :

1
1
1
T ) + Im1 x(T T ) + ((m )T )
10
10
10
9
1
1
=
Im A2 T + Im1 A2 T + ((m )T )
10
10
10

ym = Im x(

225

The variance of the noise is :

N0 2
A T
2

2 =
and therefore, the SNR is :


9
SNR =
10

2

As it is observed, there is a loss of 10 log10

2(A2 T )2
81 2A2 T
=
N0 A2 T
100 N0

81
100

= 0.9151 dB due to the mistiming.

(b) Recall from part (a) that the sampled sequence is

ym =

9
1
Im A2 T + Im1 A2 T + m
10
10

The term Im1 A10T expresses the ISI introduced to the system. If Im = 1 is transmitted, then
the probability of error is
1
1
P (e|Im = 1, Im1 = 1) + P (e|Im = 1, Im1 = 1)
2
2
 A2 T
 8 A2 T
2
2
1
1
10
2
2
N0 A T
N0 A T

=
e
d
+
e
d
2 N0 A2 T
2 N0 A2 T

P (e|Im = 1) =

1 2A2 T 1
Q
+ Q
=
2
N0
2

8
10

2

2A2 T
N0

Since the symbols of the binary PAM system are equiprobable the previous derived expression
is the probability of error when a symbol by symbol detector is employed. Comparing this with
the probability of error of a system with no ISI, we observe that there is an increase of the
probability of error by

1
Pdi (e) = Q
2

8
10

2

2A2 T 1 2A2 T
Q
N0
2
N0

Problem 10.3 :
(a) Taking the inverse Fourier transform of H(f ), we obtain :
h(t) = F 1[H(f )] = (t) +

(t t0 ) + (t + t0 )
2
2

y(t) = s(t)  h(t) = s(t) +

s(t t0 ) + s(t + t0 )
2
2

Hence,

226

(b) If the signal s(t) is used to modulate the sequence {In }, then the transmitted signal is :

u(t) =

n=

In s(t nT )

The received signal is the convolution of u(t) with h(t). Hence,



In s(t nT )  (t) + (t t0 ) + (t + t0 )
y(t) = u(t)  h(t) =
2
2
n=
=

n=

In s(t nT ) +



In s(t t0 nT ) +
In s(t + t0 nT )
2 n=
2 n=

Thus, the output of the matched lter s(t) at the time instant t1 is :

w(t1 ) =

n=

In


In
2 n=

s( nT )s( t1 )d


s( t0 nT )s( t1 )d


+
In
s( + t0 nT )s( t1 )d
2 n=

If we denote the signal s(t)  s(t) by x(t), then the output of the matched lter at t1 = kT is :
w(kT ) =


n=

In x(kT nT )



In x(kT t0 nT ) +
In x(kT + t0 nT )
+
2 n=
2 n=

(c) With t0 = T and k = n in the previous equation, we obtain :

wk = Ik x0 +


n=k

In xkn


+ Ik x1 +
In xkn1 + Ik x1 +
In xkn+1
2
2 n=k
2
2 n=k







= Ik x0 + x1 + x1 +
In xkn + xkn1 + xkn+1
2
2
2
2
n=k
The terms under the summation is the ISI introduced by the channel. If the signal s(t) is
designed so as to satisfy the Nyquist criterion, then :
xk = 0, k = 0
227

and the aobove expression simplies to :

wk = Ik +

(Ik+1 + Ik1 )
2

Problem 10.4 :
(a) Each segment of the wire-line can be considered as a bandpass lter with bandwidth W =
1200 Hz. Thus, the highest bit rate that can be transmitted without ISI by means of binary
PAM is :
R = 2W = 2400 bps
(b) The probability of error for binary PAM trasmission is :


P2 = Q

2Eb
N0

Hence, using mathematical tables for the function Q[], we nd that P2 = 107 is obtained for :

2Eb
Eb
= 5.2 =
= 13.52 = 11.30 dB
N0
N0

(c) The received power PR is related to the desired SNR per bit through the relation :
PR
1 Eb
Eb
=
=R
N0
T N0
N0
Hence, with N0 = 4.1 1021 we obtain :
PR = 4.1 1021 1200 13.52 = 6.6518 1017 = 161.77 dBW
Since the power loss of each segment is :
Ls = 50 Km 1 dB/Km = 50 dB
the transmitted power at each repeater should be :
PT = PR + Ls = 161.77 + 50 = 111.77 dBW

Problem 10.5 :
xn =

h(t + nT )h (t)dt
228

vk =
Then :
1
E
2

[vj vk ] =
=

1
E
2

z(t)h (t kT )dt




z(a)h (a jT )z (b)h(b kT )dadb

  1
E [z(a)z (b)] h (a jT )h(b kT )dadb
2

= N0

h (a jT )h(a kT )da = N0 xjk

Problem 10.6 :
In the case of duobinary signaling, the output of the matched lter is :
x(t) = sinc(2W t) + sinc(2W t 1)
and the samples xnm are given by :
xnm = x(nT mT ) =

1 nm= 0
1 nm= 1

0 otherwise

Therefore, the metric CM(I) in the Viterbi algorithm becomes

CM(I) = 2


n

= 2


n


n

In rn
In rn


n

m

In2
n

In Im xnm


n

In In1

In (2rn In In1 )

Problem 10.7 :
(a) The output of the matched lter demodulator is :
y(t) =
=


k=

Ik

gT ( kTb )gR (t )d + (t)

Ik x(t kTb ) + (t)

k=

where,
x(t) = gT (t)  gR (t) =
229

cos t
sin t
T
T
t
t2
1

4
T
T2

Hence,
y(mTb ) =

Ik x(mTb kTb ) + v(mTb )

k=

= Im +
The term
mission.

1
I
m1

1
1
Im1 + Im+1 + (mTb )

+ 1 Im+1 represents the ISI introduced by doubling the symbol rate of trans-

(b) In the next gure we show one trellis stage for the ML sequence detector. Since there is
postcursor ISI, we delay the received signal, used by the ML decoder to form the metrics, by one
sample. Thus, the states of the trellis correspond to the sequence (Im1 , Im ), and the transition
labels correspond to the symbol Im+1 . Two branches originate from each state. The upper
branch is associated with the transmission of 1, whereas the lower branch is associated with
the transmission of 1.
(Im1 , Im ) Im+1
-1
-1 -1 P
P
-1 1
1 -1
11

PP
P
-1 1PP
P
P

PP
1

-1
P

P
1 PP

-1

Problem 10.8 :
(a) The output of the matched lter at the time instant mT is :
ym =


k

1
Im xkm + m = Im + Im1 + m
4

The autocorrelation function of the noise samples m is

E[k j ] =

N0
xkj
2

Thus, the variance of the noise is

N0
N0
x0 =
2
2

=
I
=
Eb
If a symbol by symbol detector is employed and we assume that the symbols
I
m
m1

have been transmitted, then the probability of error P (e|Im = Im1 = Eb ) is :

2 =

P (e|Im = Im1 =

Eb ) = P (ym < 0|Im = Im1 =

230

Eb )

5
2
5
1  4 Eb Nm
= P (m <
Eb ) =
e 0 dm
4
N0

1  4
=
2
5

2Eb
N0

2
2

5
d = Q
4

2Eb
N0

If however Im1 = Eb , then :



P (e|Im =
Since the two symbols

Eb , Im1

3
3
= Eb ) = P (
Eb + m < 0) = Q
4
4

2Eb
N0

Eb , Eb are used with equal probability, we conclude that :



P (e) = P (e|Im =


5
1
Q
=
2
4

Eb) = P (e|Im = Eb )


2Eb
3
1
+ Q
N0
2
4

2Eb
N0

(b) In the next gure we plot the error probability obtained in part (a) (log10 (P (e))) vs. the
SNR per bit and the error probability for the case of no ISI. As it observed from the gure, the
relative dierence in SNR of the error probability of 106 is 2 dB.
-2
-2.5
-3

log(P(e)

-3.5
-4
-4.5
-5
-5.5
-6
-6.5
-7
6

10
11
SNR/bit, dB

12

Problem 10.9 :
For the DFE we have that :
Ik =

0

j=K1

cj ukj +

231

K2

j=1

cj Ikj

13

14


2
We want to minimize J = E Ik Ik  . Taking the derivative of J, with respect to the real and
imaginary parts of cl = al + jbl , 1 l K2 , we obtain :
J
al

= 0 E Ikl Ik Ik Ikl
I Ik



 k

Ik Ik
=0
E Re Ikl

and similarly :



=0





J

= 0 E Im Ikl
Ik Ik
=0
bl

Hence,

E Ikl
Ik Ik



1 l K2

= 0,

(1)

Since the information symbols are uncorrelated : E [Ik Il ] = kl . We also have :



E [Ik ul ] = E Ik

= flk


L

m=0

fm
Ilm + nl



Hence, equation (1) gives :



E Ik Ikl
= E Ik Ikl
, 1 l K2



K 2
0

c
u
+
c
I
I

0=E
j
kj
j
kj
j=1
kl
j=K
1


0
0=
j=K cj flj + cl
0 1
cl = j=K1 cj flj , 1 l K2

which is the desired equation for the feedback taps.

Problem 10.10 :
(a) The equivalent discrete-time impulse response of the channel is :
h(t) =

1

n=1

hn (t nT ) = 0.3(t + T ) + 0.9(t) + 0.3(t T )

If by {cn } we denote the coecients of the FIR equalizer, then the equalized signal is :
qm =

1

n=1

cn hmn

which in matrix notation is written as :

0.9 0.3 0.
c1
0

0.3 0.9 0.3 c0 = 1

0. 0.3 0.9
0
c1
232

The coecients of the zero-force equalizer can be found by solving the previous matrix equation.
Thus,

c1
0.4762

c0 = 1.4286
0.4762
c1
(b) The values of qm for m = 2, 3 are given by
q2 =
q2 =
q3 =
q3 =

1

n=1
1

n=1
1

n=1
1

n=1

cn h2n = c1 h1 = 0.1429
cn h2n = c1 h1 = 0.1429
cn h3n = 0
cn h3n = 0

Problem 10.11 :
(a) The output of the zero-force equalizer is :
qm =

1

n=1

cn xmn

With q0 = 1 and qm = 0 for m = 0, we obtain the system :

1.0
0.1 0.5
c1
0

1.0
0.1 c0 = 1
0.2

0.05 0.2
1.0
0
c1
Solving the previous system in terms of the equalizers coecients, we obtain :

c1
0.000

c0 = 0.980
c1
0.196

233

(b) The output of the equalizer is :

qm =

0
c1 x2 = 0
c1 x1 + c0 x2 = 0.49
0
1
0
c0 x2 + x1 c1 = 0.0098
c1 x2 = 0.0098
0

m 4
m = 3
m = 2
m = 1
m=0
m=1
m=2
m=3
m4

Hence, the residual ISI sequence is

residual ISI = {. . . , 0, 0.49, 0, 0, 0, 0.0098, 0.0098, 0, . . .}
and its span is 6 symbols.

Problem 10.12 :
(a) If {cn } denote the coecients of the zero-force equalizer and {qm } is the sequence of the
equalizers output samples, then :
qm =

1

n=1

cn xmn

where {xk } is the noise free response of the matched lter demodulator sampled at t = kT .
With q1 = 0, q0 = q1 = Eb , we obtain the system :

Eb 0.9Eb 0.1Eb
c1
0

Eb 0.9Eb
0.9Eb
c0 = E b
0.1Eb 0.9Eb
Eb
c1
Eb
The solution to the system is :


c1 c0 c1

0.2137 0.3846 1.3248

(b) The set of noise variables {k } at the output of the sampler is a gaussian distributed sequence
with zero-mean and autocorrelation function :
$

R (k) =

N0
x
2 k

|k| 2
otherwise

Thus, the autocorrelation function of the noise at the output of the equalizer is :
Rn (k) = R (k)  c(k)  c(k)
234

where c(k) denotes the discrete time impulse response of the equalizer. Therefore, the autocorrelation sequence of the noise at the output of the equalizer is :

Rn (k) =

N0 Eb
2

0.9402
k=0
1.3577
k = 1
0.0546
k = 2
0.1956
k = 3
0.0283
k = 4
0 otherwise

To nd an estimate of the error probability for the sequence detector, we ignore the residual
interference due to the nite length of the equalizer, and we only consider paths of length two.
Thus, if we start at state I0 = 1 and the transmitted symbols are (I1 , I2 ) = (1, 1) an error is
made by the sequence detector if the path (1, 1) is more probable, given the received values of
r1 and r2 . The metric for the path (I1 , I2 ) = (1, 1) is :


2 (1, 1) = [ r1 2Eb r2 2Eb ]C

where :

N0 Eb
C=
2

r1 2Eb
r2 2Eb

0.9402 1.3577
1.3577 0.9402

Similarly, the metric of the path (I1 , I2 ) = (1, 1) is



2 (1, 1) = [ r1 r2 ]C

r1
r2

Hence, the probability of error is :

P2 = P (2 (1, 1) < 2 (1, 1))
and upon substitution of r1 = 2Eb + n1 , r2 = 2Eb + n2 , we obtain :
P2 = P (n1 + n2 < 2Eb )
Since n1 and n2 are zero-mean Gaussian variables, their sum is also zero-mean Gaussian with
variance :
N0Eb
N0 Eb
= 4.5958
2 = (2 0.9402 + 2 1.3577)
2
2
and therefore :



8Eb
P2 = Q
4.5958N0
The bit error probability is

P2
.
2

235

Problem 10.13 :
The optimum tap coecients of the zero-force equalizer can be found by solving the system:

1.0 0.3 0.0

c1
0

0.2 1.0 0.3 c0 = 1

c1
0.0 0.2 1.0
0
Hence,

c1
0.3409

c0 = 1.1364
0.2273
c1

The output of the equalizer is :

qm =

0
c1 x1 = 0.1023
0
1
0
c1 x1 = 0.0455
0

m 3
m = 2
m = 1
m=0
m=1
m=2
m3

Hence, the residual ISI sequence is :

residual ISI = {. . . , 0, 0.1023, 0, 0, 0, 0.0455, 0, . . .}

Problem 10.14 :
(a) If we assume that the signal pulse has duration T , then the ouput of the matched lter at
the time instant t = T is :
y(T ) =

 T

r( )s( )d

 T
0

 T

(s( ) + s( T ) + n( ))s( )d
2

s ( )d +

 T

n( )s( )d

= Es + n

where Es is the energy of the signal pulse and n is a zero-mean Gaussian random variable with
variance n2 = N02Es . Similarly, the output of the matched lter at t = 2T is :
y(2T ) =

 T

s2 ( )d +

 T
0

= Es + n
236

n( )s( )d

(b) If the transmitted sequence is :

x(t) =


n=

In s(t nT )

with In taking the values 1, 1 with equal probability, then the output of the demodulator at
the time instant t = kT is
yk = Ik Es + Ik1Es + nk
The term Ik1Es expresses the ISI due to the signal reection. If a symbol by symbol detector
is employed and the ISI is ignored, then the probability of error is :
1
1
P (error|In = 1, In1 = 1) + P (error|In = 1, In1 = 1)
2
2
1
1
=
P ((1 + )Es + nk < 0) + P ((1 )Es + nk < 0)
2

2
1 2(1 + ) Es 1 2(1 )2 Es
+ Q
Q
=
2
N0
2
N0

P (e) =

(c) To nd the error rate performance of the DFE, we assume that the estimation of the
parameter is correct and that the probability of error at each time instant is the same. Since
the transmitted symbols are equiprobable, we obtain :
P (e) = P (error at k|Ik = 1)
= P (error at k 1)P (error at k|Ik = 1, error at k 1)
+P (no error at k 1)P (error at k|Ik = 1, no error at k 1)
= P (e)P (error at k|Ik = 1, error at k 1)
+(1 P (e))P (error at k|Ik = 1, no error at k 1)
= P (e)p + (1 P (e))q
where :
p = P (error at k|Ik = 1, error at k 1)
1
P (error at k|Ik = 1, Ik1 = 1, error at k 1)
=
2
1
+ P (error at k|Ik = 1, Ik1 = 1, error at k 1)
2
1
1
P ((1 + 2)Es + nk < 0) + P ((1 2)Es + nk < 0)
=
2

2
1 2(1 + 2) Es 1 2(1 2)2 Es
Q
=
+ Q
2
N0
2
N0
237

and
q = P (error at k|Ik = 1, no error at k 1)


= P (Es + nk < 0) = Q

2Es
N0

Solving for P (e), we obtain :

P (e) =



2Es
N0

q






=


2(1+2)2 Es
2(12)2 Es
1p+q
1
2Es
1 12 Q

Q
+
Q
N0
2
N0
N0

A sketch of the detector structure is shown in the next gure.

Input rk

+ n
+

Estimate

Threshold
device

Output a
k

Delay

Problem 10.15 :
A discrete time transversal lter equivalent to the cascade of the trasmitting lter gT (t), the
channel c(t), the matched lter at the receicer gR (t) and the sampler, has tap gain coecients
{xm }, where :

0.9 m = 0
xm = 0.3 m = 1

0 otherwise
The noise k , at the output of the sampler, is a zero-mean Gaussian sequence with autocorrelation function :
E[k l ] = 2 xkl ,
|k l| 1
If the Z-transform of the sequence {xm }, X(z), assumes the factorization :
X(z) = F (z)F (z 1 )
then the lter 1/F (z 1 ) can follow the sampler to white the noise sequence k . In this case the
output of the whitening lter, and input to the MSE equalizer, is the sequence :
un =


k

Ik fnk + nk
238

where nk is zero mean Gaussian with variance 2 . The optimum coecients of the MSE equalizer, ck , satisfy :
1


n=1

where :

(n k) =
$

( k) =
With

k = 0, 1

cn kn = k ,

xnk + 2 n,k , |n k| 1
0
otherwise
fk , 1 k 0
0
otherwise

X(z) = 0.3z + 0.9 + 0.3z 1 = (f0 + f1 z 1 )(f0 + f1 z)

we obtain the parameters f0 and f1 as :

$
0.7854
f0 =
,
0.1146

f1 =

0.1146
0.7854

The parameters f0 and f1 should have the same sign since f0 f1 = 0.3. However, the sign itself
does not play any role if the data are dierentially encoded. To have a stable inverse system
system F (z 1 ) = f0 + f1 z
1/F (z 1 ), we select f0 and f1 in such a way that
the zero of the
is inside the unit circle. Thus, we choose f0 = 0.1146 and f1 = 0.7854 and therefore, the
desired system for the equalizers coecients is

0.9 + 0.1
0.3
0.0
c1
0.7854

0.3
0.9 + 0.1
0.3

c0 =
0.1146
0.0
0.3
0.9 + 0.1
c1
0
Solving this system, we obtain
c1 = 0.8596,

c1 = 0.0266

c0 = 0.0886,

Problem 10.16 :
(a) The spectrum of the band limited equalized pulse is
$

X(f ) =
$

=
$

1
2W
1
2W
1
W

n=




n
x( 2W
)ej
0

2 + 2 cos f
W
0

1 + 1 cos f
W
0
239

nf
W

|f | W
otherwise

|f | W
otherwise
|f | W
otherwise

where W =

1
2Tb

(b) The following table lists the possible transmitted sequences of length 3 and the corresponding
output of the detector.
-1 -1
-1 -1
-1 1
-1 1
1 -1
1 -1
1 1
1 1
As it is observed there are 5 possible output
P (bm = 2) = 14 and P (bm = 4) = 18 .

-1 -4
1 -2
-1 0
1 2
-1 -2
1 0
-1 2
1 4
levels bm , with probability P (bm = 0) =

1
,
4

(c) The transmitting lter GT (f ), the receiving lter GR (f ) and the equalizer GE (f ) satisfy the
condition
GT (f )GR (f )GE (f ) = X(f )
The power spectral density of the noise at the output of the equalizer is :
S (f ) = Sn (f )|GR (f )GE (f )|2 = 2 |GR (f )GE (f )|2
With
GT (f ) = GR (f ) = P (f ) =

T50 T50 |f |
e
2

the variance of the output noise is :

2 = 2
=

 W
W

|GR (f )GE (f )|2 df =

2
 

X(f
)



 df
2

GT (f ) 

|1 + cos f
|2
W
df
2
2 T50
W 2 e2T50 |f |

8 2
=
2
2 T50
W2

 W
0

f
1 + cos
W

2

e2T50 f df

The value of the previous integral can be found using the formula :


eax cosn bxdx


1
a
n1
2
(a
cos
bx
+
nb
sin
bx)e
x
cos
bx
+
n(n

1)b
eax cosn2 bxdx
= 2
a + n2 b2

240

Thus, we obtain :




2T50 + W 21T50
1
8 2
2T50 W
=

1
+
2
2
2
2 T50
W2
2T50
4 2 T50
+ 4 W 2



4T50
2T50 W
2 2
e
+
1
2
4 T50 + W 2

To nd the probability of error using a symbol by symbol detector, we follow the same procedure
as in Section 9.3.2. The results are the same with that obtained from a 3-point PAM constellation
(0, 2) used with a duobinary signal with output levels having the probability mass function
given in part (b). An upper bound of the symbol probability of error is :
P (e) < P (|ym| > 1|bm = 0)P (bm = 0) + 2P (|ym 2| > 1|bm = 2)P (bm = 2)
+2P (ym + 4 > 1|bm = 4)P (bm = 4)
= P (|ym| > 1|bm = 0) [P (bm = 0) + 2P (bm = 2) + P (bm = 4)]
7
P (|ym| > 1|bm = 0)
=
8
But
P (|ym | > 1|bm = 0) =
Therefore,

2
2

P (e) <

1
14
Q
8

ex

2 /2 2

dx

Problem 10.17 :
Since the partial response signal has memory length equal to 2, the corresponding trellis has 4
states which we label as (In1 , In ). The following gure shows three frames of the trellis. The
labels of the branches indicate the output of the partial response system. As it is observed the
free distance between merging paths is 3, whereas the Euclidean distance is equal to
dE = 22 + 42 + 22 = 24
(In1 , In )
-4 -4 -4
(-1,-1)

-2

(-1,1)

-2

(1,-1)

(1,1)

241

Problem 10.18 :
(a) X(z) = F (z)F (z 1 ) = 12 z + 1 + 12 z 1 . Then, the covariance matrix is :

1 + N0
1/2
0
1/2

1/2
1 + N0
= 1/2
and = 1/ 2
0
1/2
1 + N0
0

The optimum equalizer coecients are given by :

Copt = 1

(1 + N0 )2 1/4

1
12 (1 + N0 )
= det()

1/4

2
N0 + 32 N0 + 14

3
1
1
= 2 det()
N02 + 2 N0 + 2
N0
1
2 4


where det() = (1 + N0 ) (1 + N0 )2

1
2

12 (1 + N0 )
1/4
1/2

(1 + N0 )2
12 (1 + N0 )
1/ 2
1 (1 + N0 ) (1 + N0 )2 1/4
0
2

(b)

det( I) = (1 + N0 ) (1 + N0 )2 12
1 = 1 + N0 , 2 = 12 + 1 + N0 , 3 = 1 12 + N0
and the corresponding eigenvectors are :

1/2
1/2
1/ 2

0
v1 =
, v2 = 1/ 2 , v3 = 1/ 2
1/2
1/2
1/ 2
(c)
Jmin (K)|K=1 = Jmin (1) = 1  1 =
(d)
=

2N03 + 4N02 + 2N0 + 3/4

2N03 + 4N02 + 5N0 + 1

1 Jmin (1)
2N02 + 3N0 + 3/4
=
Jmin (1)
2N03 + 4N02 + 1/4

Note
that as N0 0, 3. For N0 = 0.1, = 2.18 for the 3-tap equalizer and =

1 + N20 1 = 3.58, for the innite-tap equalizer (as in example 10-2-1). Also, note that
= N10 = 10 for the case of no intersymbol interference.

242

Problem 10.19 :
For the DFE we have that :
Ik =

0

j=K1

cj ukj +

The orthogonality principle is simply :

K2

j=1

cj Ikj , and >k = Ik Ik

E >k ukl = 0, for K1 l 0

E Ik ukl = E Ik ukl , K1 l 0





E Ik I

kl = E Ik Ikl , 1 l K2

E >k Ikl
= 0, for 1 l K2

Since the information symbols are uncorrelated : E [Ik Il ] = akl , where a = E |Ik |2 is a
constant whose value is not needed since it will be present in all terms and thus, cancelled out.
In order to solve the above system, we also need E [uk ul ] , E [Ik ul ] . We have :


 

L
E [uk ul ] = E
n=0 fn Ikn + nk
L

= a m=0 fm
fm+kl + N0 kl

and


L

E [Ik ul ] = E Ik

= aflk

m=0

L
m=0

fm
Ilm + nl

fm
Ilm + nl





Hence, the second equation of the orthogonality principle gives :



= E Ik Ikl
, 1 l K2
E Ik Ikl



2
0=E
cu
+ K
j=1 cj Ikj Ikl
 j=K1 j kj

0
0=a
j=K1 cj flj + acl
0
cl = j=K1 cj flj , 1 l K2

which is the desired equation for the feedback taps. The rst equation of the orthogonality
principle will give :


0
j=K1 cj ukj

E Ik ukl = E Ik ukl , K1 l 0

=E
afl

afl
=

0

j=K1 cj



 
L

m=0

K2

K 2

j=1 cj Ikj

fm
fm+lj + N0 kl + a

ukl

j=1 cj fjl ,

K1 l 0

Substituting the expression for cj , 1 j K2 , that we found above :

fl
=

fl
=

0


L

0



j=K1 cj

j=K1 cj

m=0 fm fm+lj + N0 kl

L
m=0

fm
fm+lj + N0 kl

0

j=K1 cj lj

K 2 0
j=1

0

m=K1

j=K1 cj

cm fjm fjl
, K1 l 0

K 2

m=1

= fl
, K1 l 0

243

fmj fml
, K1 l 0

l

where lj =
the equalizer.

m=0

fm
fm+lj + N0 lj , which is the desired expression for the feedforward taps of

Problem 10.20 :
The tap coecients for the feedback section of the DFE are given by the equation :


ck = 0j=K1 cj fkj , k = 1, 2, ..., K2

= (c0 fk + c1 fk+1 + ... + cK1 fk+K1 )
But fk = 0 for k < 0 and k > L. Therefore :
cL = c0 fL , cL+1 = 0, cL+2 = 0, etc

Problem 10.21 :
(a) The tap coecients for the feedback section of the DFE are given by the equation : ck =

0j=K1 cj fkj , 1 k K2 , and for the feedforward section as the solution to the equations :
0
0

j=K1 cj lj = fl , K1 l 0. In this case, K1 = 1, and hence :

j=K1 cj lj = fl , l =
1, 0 or :
0,0 c0 + 0,1 c1 = f0
1,0 c0 + 1,1 c1 = f1
But lj =

so :

l

m=0

fm
fm+lj + N0 lj , so the above system can be written :





1
1
+
N
c
1/
0
0
2
2
2
=
1
1
+
N
c
1/ 2
0
1
2

c0
c1


= 
2 N02 + 32 N0 + 14

1
2

+ N0
N0

2
, for N0 << 1
2 2N0

The coecient for the feedback section is :

1
c1 = c0 f1 = c0 1, for N0 << 1
2
(b)
Jmin (1) = 1

0

j=K1

cj fj =

2N02 + N0

2 N02 + 32 N0 +
244

1
4

2N0 , for N0 << 1

(c)
=

1 + 4N0
1
1 Jmin (1)
=

, N0 << 1
Jmin (1)
2N0 (1 + 2N0 )
2N0

(d) For the innite tap DFE, we have from example 10-3-1 :
Jmin =
=

1+N0 +

1Jmin
Jmin

2N0

(1+N0 )2 1
1N0

2N0 , N0 << 1

(1+N0 )2 1
2N0

(e) For N0 = 0.1, we have :

Jmin(1) = 0.146, = 5.83 (7.66 dB)
Jmin = 0.128, = 6.8 (8.32 dB)
For N0 = 0.01, we have :
= 51 (17.1 dB)
Jmin (1) = 0.0193,
Jmin = 0.0174, = 56.6 (17.5 dB)
The three-tap equalizer performs very wee compared to the innite-tap equalizer. The dierence
in performance is 0.6 dB for N0 = 0.1 and 0.4 dB for N0 = 0.01.

Problem 10.22 :
(a) We have that :

= 900, 1+
= 1200
2T
1 + = 1200/900 = 4/3 = 1/3
1
2T

(b) Since 1/2T = 900, the pulse rate 1/T is 1800 pulses/sec.
(c) The largest interference is caused by the sequence : {1, 1, s, 1, 1, 1} or its opposite in
sign. This interference is constructive or destructive depending on the sign of the information

symbol s. The peak distortion is 3k=2,k=0 fk = 1.6
 5

(d) The probability of the worst-case interference given above is 12 = 1/32, and the same is
the probability of the sequence that causes the opposite-sign interference.

245

Problem 10.23 :
(a)

F (z) = 0.8 0.6z 1

X(z) F (z)F (z ) = (0.8 0.6z 1 ) (0.8 0.6z) = 1 0.48z 1 0.48z

Thus, x0 = 1, x1 = x1 = 0.48.
(b)






1 
2n 2
jT
H +
=
X
e
= 1 0.48ejT 0.48ejT = 1 0.96 cos T
T n= 
T 

(c) For the linear equalizer base on the mean-square-error criterion we have :
Jmin =

T
2

1
2

1
2

But :

1
2

Therefore :
Jmin =

 /T

N0
/T 1+N0 0.96 cos T d

N0
1+N0 0.96 cos d



1
1a cos d,

N0
1+N0

a=

0.96
1+N0

1
1
d =
, a2 < 1
2
1 a cos
1a

1
N0
N0
*
=


2
1 + N0
0.96
(1 + N0 )2 (0.96)2
1 1+N
0

(d) For the decision-feedback equalizer :

Jmin =

1 + N0 +

2N0

(1 + N0 )2 (0.96)2

which follows from the result in example 10.3.1. Note that for N0 << 1,
Jmin

1+

2N0

1 (0.96)2

1.56N0

In contrast, for the linear equalizer we have :

N0
Jmin 
3.57N0
1 (0.96)2
246

Problem 10.24 :
(a) Part of the tree structure is shown in the following gure :

I3 = 3

I3 = 1

I2 = 3

I3 = 1
I2 = 1

I3 = 3

I2 = 1

I2 = 3

I2 = 3

I2 = 1
I1 = 3

I1 = 1

I2 = 1

I2 = 3

I2 = 3
I1 = 1

I2 = 1

I2 = 1

I1 = 3

=
3
I
2

I2 = 3

I2 = 1

I2 = 1

I2 = 3

(b) There are four states in the trellis (corresponding to the four possible values of the symbol
Ik1 ), and for each one there are four paths starting from it (corresponding to the four possible
values of the symbol Ik ). Hence, 16 probabilities must be computed at each stage of the Viterbi
algorithm.
(c) Since, there are four states, the number of surviving sequences is also four.

247

(d) The metrics are

(y1 0.8I1 )2 , i = 1 and
1 (I1
1 (I1
1 (I1
1 (I1


i

(yi 0.8Ii + 0.6Ii1 )2 , i 2

= 3) = [0.5 3 0.8]2 = 3.61

= 1) = [0.5 1 0.8]2 = 0.09
= 1) = [0.5 + 1 0.8]2 = 1.69
= 3) = [0.5 + 3 0.8]2 = 8.41

2 (I2 = 3, I1 = 3) = 1 (3) + [2 2.4 + 3 0.6]2 = 5.57

2 (3, 1) = 1 (1) + [2 2.4 + 1 0.6]2 = 0.13
2 (3, 1) = 1 (1) + [2 2.4 1 0.6]2 = 6.53
2 (3, 3) = 1 (3) + [2 2.4 3 0.6]2 = 13.25
2 (1, 3) = 1 (3) + [2 0.8 + 3 0.6]2 = 12.61
2 (1, 1) = 1 (1) + [2 0.8 + 1 0.6]2 = 3.33
2 (1, 1) = 1 (1) + [2 0.8 1 0.6]2 = 2.05
2 (1, 3) = 1 (3) + [2 0.8 3 0.6]2 = 8.77
2 (1, 3) = 1 (3) + [2 + 0.8 + 3 0.6]2 = 24.77
2 (1, 1) = 1 (1) + [2 + 0.8 + 1 0.6]2 = 11.65
2 (1, 1) = 1 (1) + [2 + 0.8 1 0.6]2 = 6.53
2 (1, 3) = 1 (3) + [2 + 0.8 3 0.6]2 = 9.41
2 (3, 3) = 1 (3) + [2 + 2.4 + 3 0.6]2 = 42.05
2 (3, 1) = 1 (1) + [2 + 2.4 + 1 0.6]2 = 25.09
2 (3, 1) = 1 (1) + [2 + 2.4 1 0.6]2 = 16.13
2 (3, 3) = 1 (3) + [2 + 2.4 3 0.6]2 = 15.17
The four surviving paths at this stage are minI1 [2 (x, I1 )] , x = 3, 1, 1, 3 or :
I2
I2
I2
I2

= 3, I1 = 1 with metric 2 (3, 1) = 0.13

= 1, I1 = 1 with metric 2 (1, 1) = 2.05
= 1, I1 = 1 with metric 2 (1, 1) = 6.53
= 3, I1 = 3 with metric 2 (3, 3) = 15.17

Now we compute the metrics for the next stage :

3 (I3 = 3, I2 = 3, I1 = 1) = 2 (3, 1) + [1 2.4 + 1.8]2 = 2.69
3 (3, 1, 1) = 2 (1, 1) + [1 2.4 + 0.6]2 = 9.89
3 (3, 1, 1) = 2 (1, 1) + [1 2.4 0.6]2 = 22.53
3 (3, 3, 3) = 2 (3, 3) + [1 2.4 1.8]2 = 42.21

248

3 (1, 3, 1) = 2 (3, 1) + [1 0.8 + 1.8]2 = 0.13

3 (1, 1, 1) = 2 (1, 1) + [1 0.8 + 0.6]2 = 7.81
3 (1, 1, 1) = 2 (1, 1) + [1 0.8 0.6]2 = 12.29
3 (1, 3, 3) = 2 (3, 3) + [1 0.8 1.8]2 = 28.13
3 (1, 3, 1) = 2 (3, 1) + [1 + 0.8 + 1.8]2 = 2.69
3 (1, 1, 1) = 2 (1, 1) + [1 + 0.8 + 0.6]2 = 2.69
3 (1, 1, 1) = 2 (1, 1) + [1 + 0.8 0.6]2 = 7.17
3 (1, 3, 3) = 2 (3, 3) + [1 + 0.8 1.8]2 = 19.17
3 (3, 3, 1) = 2 (3, 1) + [1 + 2.4 + 1.8]2 = 10.37
3 (3, 1, 1) = 2 (1, 1) + [1 + 2.4 + 0.6]2 = 2.69
3 (3, 1, 1) = 2 (1, 1) + [1 + 2.4 0.6]2 = 7.17
3 (3, 3, 3) = 2 (3, 3) + [1 + 2.4 1.8]2 = 15.33
The four surviving sequences at this stage are minI2 ,I1 [3 (x, I2 , I1 )] , x = 3, 1, 1, 3 or :
I3
I3
I3
I3

= 3, I2 = 3, I1 = 1 with metric 3 (3, 3, 1) = 2.69

= 1, I2 = 3, I1 = 1 with metric 3 (1, 3, 1) = 0.13
= 1, I2 = 3, I1 = 1 with metric 3 (1, 3, 1) = 2.69
= 3, I2 = 1, I1 = 1 with metric 3 (3, 1, 1) = 2.69

2
(e) For the channel, min
= 1 and hence :

6
P4 = 8Q
av
15

Problem 10.25 :
(a)

But

bk =

1
K

1
K

1
K

K1


K1
n=0

En ej2nk/K

K1 K1
K1
l=0

cl ej2nl/K ej2nk/K

l=0

n=0

cl

K1 j2n(kl)/K
e
n=0

$
j2n(kl)/K

n=0

Hence, bk = ck .
249

0, k = l
K, k = l

(b)
E(z) =
=

K1
k=0

ck z k

K1  1 K1
k=0

K1

1
K

1
K

1z K
K

n=0

n=0

K1
n=0

En

En ej2nk/K z k

K1  j2n/K 1 k
e
z

k=0

1z
En 1exp(j2n/K)z
1

K1

En
n=0 1exp(j2n/K)z 1

(c) The block diagram is as shown in the following gure :

Parallel Bank of Single Pole Filters
y0 (n) 
X

+

Comb. Filter

x(n)


+
X

z K

z 1

ej2/K

+

E0

y1 (n) 

+

1/K

.
.
.
.

z 1

E1

yK1 (n)

1
ej2(K1)/K z


y(n)
+

EK1

(d) The adjustable parameters in this structure are {E0 , E1 , ..., EK1 } , i.e. the DFT coecients
of the equalizer taps. For more details on this equalizer structure, see the paper by Proakis (IEEE
Trans. on Audio and Electroacc., pp 195-200, June 1970).

250