Solutions to Selected Problems

– from Madhow: Fundamentals of Digital Communication

and from Johannesson & Zigangirov: Fundamentals of
Convolutional Coding –

Saif K. Mohammed

Department of Electrical Engineering

Linköpings universitet, SE-581 83 Linköping, Sweden

Linköping 2012
Note: This material is prepared for the 2012 version of the Master course TSKS04 Digital
Communication Continuation Course. For almost every task planned for tutorials you find
either hints and answers or complete solutions. For the tasks where we give complete solutions,
we have chosen to adjust the notations to concur with the notation that has been used in the
lectures.
This document will evolve during the course as the tutorials go by.
This version of the document was compiled February 12, 2012.

Solutions to Selected Problems – from Madhow: Fundamentals of Digital Com-

munication
and from Johannesson & Zigangirov: Fundamentals of Convolutional Coding

c 2010–2012 Mikael Olofsson, Saif K. Mohammed

Department of Electrical Engineering

Linköpings universitet
SE-581 83 Linköping
Sweden

This document was prepared using LATEX2ε with the aid of TeXnicCenter on an Dell PC
running CentOS 5. The figures were produced using Xfig (from xfig.org). Finally, the plots
were produced using Matlab (from MathWorks, Inc.).
Contents
Tutorial 3 1
Task 2.15 (Hints) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Task 2.17 (Solution) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Task 2.18 (Hints and answer) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Task 2.25 (Hints and answer) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Task 2.26 (Solution) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Task 2.27 (Solution) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Task 2.29 (Hints) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

Tutorial 4 6
Task 5.1 (Solution) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Task 5.2 (Hints and answer) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Task 5.3 (Solution) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Task 5.4 (Hints and answer) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Task 5.6 (Solution) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

i
ii
 1

Tutorial 3

Task 2.15

Hints
Follow the instructions given in the task.

Task 2.17

Task
Consider a pulse s(t) = sinc(at) sinc(bt) where a ≥ b.

a. Sketch the spectrum S(f ) of the pulse.

b. Suppose that the pulse is to be used over an ideal real baseband channel with
one-sided bandwidth 400 Hz. Determine a and b such that the pulse is Nyquist for
4-PAM at 1200 b/s and exactly fills the channel bandwidth.
c. Now, suppose that the pulse is to be used over a passband channel spanning the
frequency band 2.4–2.42 GHz. Assuming that we use 64-QAM at 60 Mb/s, deter-
mine a and b so that the pulse is Nyquist and exactly fills the channel bandwidth.
d. Sketch an argument showing that the magnitude of the transmitted waveform in
the preceding settings is always finite.

Solution
a. Define the signals

s1 (t) = sinc(at),
s2 (t) = sinc(bt).

Then we have
1
S1 (f ) = I (f ),
a {|f |<a/2}
1
S2 (f ) = I{|f |<b/2} (f ),
b
S(f ) = (S1 ∗ S2 )(f ).

The convolution results in the following graph, where also frequency-shifted ver-
sions according to Poisson are included.
2

S(f )
1
a

−b−a b−a a−b a+b

2 2 2 2

b. We have the bit rate R = 1200 b/s and the alphabet size M = 4. The symbol rate
1/T is then given by

1 R
= = 600 symbols/s.
T log2 (M )

This is our sampling frequency.

To fulfill the bandwidth demand, we have
a+b
= 400 Hz
2

To fulfill the Nyquist criterion (consider the frequency domain, eq. (2.80) in the
text book), we must have

a+b a−b
+ = 600 Hz.
2 2
The solution to these two equations is a = 600 Hz and b = 200 Hz.
c. We have a passband system with bandwidth 20 MHz, bit rate R = 60 Mb/s and
alphabet size M = 64. The symbol rate 1/T is then given by

1 R
= = 10 Msymbols/s
T log2 (M )

The one-sided bandwidth of the corresponding baseband system is 10 MHz. This

time, we have
a+b
= 10 MHz
2
a+b a−b
+ = 10 MHz
2 2
with solution a = b = 10 MHz.
falls as 1/t2 . The absolute value of
d. Both s1 (t) and s2 (t) falls as 1/|t|. Thus, s(t)P
A
the total signal is then bounded by the sum (kT )2
, where A is proportional to
the maximum symbol energy, and this sum is finite. So, the absolute value of the
total signal is finite.
 3

Task 2.18

Hints
Check the definitions of Nyquist and square-root Nyquist.

Answer
a. True (Because p(kT ) = δk0 ) R +∞
b. False (For p(t) to be square root Nyquist, we must have −∞ p(t)p∗ (t − kT )dt =0.
It is easy to see that it is not true for k = 1).

Task 2.25

Hints
Write down the complex envelops s̃0 and s̃1 of s0 and s1 w.r.t frequency reference
respectively. We know that

< s0 , s1 >= Re{< s̃0 , s̃1 >}

and compute the Re{< s̃0 , s̃1 >}. Then, find the minimum frequency separation for
φ0 = φ1 = 0 and for arbitrary φ0 and φ1 .

Answer
a. kf0 − f1 k = 1/2T
b. kf0 − f1 k = 1/T

Task 2.26

Task
a. Specify the Walsh-Hadamard codes for 8-ary orthogonal signaling with non-coherent
reception.
b. Plot the baseband waveforms corresponding to sending these codes using a square
root raised cosine pulse with excess bandwidth of 50 %.
c. What is the fractional increase in the bandwidth efficiency if we use these wave-
forms as building blocks for biorthogonal signaling with coherent reception?

Solution
a. To make the matrices below more readable, let + denote 1 and let − denote −1.
The Walch-Hadamard codes are given by the relations
4

 
Hm−1 Hm−1
H0 = (+), Hm = , for m > 0.
Hm−1 −Hm−1

Thus, we have
 
  + + + +
+ +  + − + − 
H1 = , H2 =  ,
+ −  + + − − 
+ − − +
+ + + + + + + +
 

 + − + − + − + − 


 + + − − + + − − 

 + − − + + − − + 
H3 =  

 + + + + − − − − 


 + − + − − + − + 

 + + − − − − + + 
+ − − + − + + −

The eight rows are then used as signals.

b. Omitted.
c. We note that the eight signals are orthogonal. Thus, we need N = 8 dimensions
for octal signaling (M = 8) and the bandwidth efficiency is then
log2 (M ) 3
=
N 8
At bi-orthogonal signaling we have two antipodal signals per dimension. Thus, we
have M = 16 signals. Using coherent detection, it is enough with N = 8 dimen-
sions, and the bandwidth efficiency is then
log2 (M ) 4
=
N 8
The increase in percent is then given by
4/8 − 3/8 1
= ,
3/8 3

i.e. 33 %.

Task 2.27

Task
We wish to send at a rate of 10Mb/s over a passband channel. Assuming that an
excess bandwidth of 50% is used, how much bandwidth is needed for each of the follow-
ing schemes: QPSK, 64-QAM and 64-ary noncoherent orthogonal modulation using a
Walsh-Hadamard code?
 5

Solution
Let Rb be the wanted data rate, i.e. Rb = 10 Mb/s. The smallest possible bandwidth
Bmin is given by the Nyquist criterion as
Rb
Bmin = ,
ηB
where ηB is the bandwidth efficiency expressed in bits/symbol/dimension, and we are
supposed to use 50% excess bandwidth. Our bandwidth is therefore
3 3Rb
B = Bmin = .
2 2ηB
The bandwidth efficiency is defined as
log2 (M )
ηB = ,
D
where M is the number of signals in the constellation and D is the number of complex
dimensions that are spanned by the signals. Totally, this gives us the bandwidth
3Rb D
B= .
2 log2 (M )

QPSK Here we have M = 4 and D = 1.

3Rb D
B= = 7.5 MHz.
2 log2 (M )
64-QAM Here we have M = 64 and D = 1.
3Rb D
B= = 2.5 MHz.
2 log2 (M )
64-Walsh-Hadamard Here we have M = 64. Non-coherent detection means that we
need one complex dimension per orthogonal signal. Thus, we have D = 64.
3Rb D
B= = 160 MHz.
2 log2 (M )

Task 2.29

Hints
Read up on DPSK. The information is represented by the phase differences.
a. According to Figure 2.22, the information bits 00, 10, 11, 01 corresponds to phase
shift 0, π/2, π, −π/2, respectively. Let θ(n) ∈ {±π/4, ±3π/4} (or alternatively θ(n) ∈
{π/4, 3π/4, 5π/4, 7π/4}.
Solution: −π/4, 3π/4, π/4, 3π/4, −3π/4, 3π/4, −π/4, −π/4, −3π/4, π/4
b. r[2]r∗ [1] = 1 + 3j. Its argument is closest to a phase difference of π/2, which
corresponds to information bits 10. This would be an error, since the sent bits are 11
at time instant n = 2.
6

Tutorial 4

Task 5.1

Task
Consider a digitally modulated system using QPSK at bit rate 2/T . The transmit filter,
channel and receive filter have the following impulse responses.

gTX (t) = I{0≤t<T /2} (t) − I{T /2≤t<T } (t),

 
1 T
gC (t) = δ(t) − δ t − ,
2 2
gRX (t) = I{0≤t<T /2} (t).

Let z[k] denote the receive filter output sampled ad time instance kTs + τ , where Ts is
a sampling interval to be chosen.

a. Show that ML sequence detection using the samples {z[k]} is possible, given an
appropriate choice of Ts and τ . Specify the corresponding choice of Ts and τ .
b. How many states are needed in the trellis for implementing ML sequence detection
using the Viterbi algorithm?

Solution
Let p(t) be the total impulse response of the cascade of the sender filter and the channel
i.e.
3 1
p(t) = (gTX ∗ gC )(t) = I{0≤t< T } (t) − I{ T ≤t<T } (t) + I{T ≤t< 3T } (t)
2 2 2 2 2

Its matched filter is then

3 1
pMF (t) = p∗ (−t) = I{0<−t< T } (t) − I{ T <−t<T } (t) + I{T <−t< 3T } (t)
2 2 2 2 2

Furthermore, let y(t) denote the output from the channel. We have the bit rate 2/T . In
QPSK, we have M = 4 signals, 2 bits (log2 (M )) per symbol. Thus, we have the symbol
rate 1/T .

a. The optimal (ML) case is if we use hMF (t) as our receiver filter. Let z0 [n] be the
output from that filter, sampled in the time instances nT . Then we have
Z ∞
z0 [n] = (y ∗ pMF )(nT ) = y(t)pMF (nT − t) dt.
−∞

The actual output from the receiver filter, sampled in the time instances kTs + τ
for all integers k is given by
Z ∞
z[k] = (y ∗ gRX )(kTs + τ ) = y(t)gRX (kTs + τ − t) dt.
−∞
 7

With τ = 0 and Ts = T /2 we get

∞  
T
Z
z[k] = y(t)gRX k − t dt.
−∞ 2

Furthermore, we have
   
T 3 1 3T
hMF (t) = gRX t + − gRX (t + T ) + gRX t + ,
2 2 2 2

which can be rewritten as

pMF (nT − t) =
     
T 3 T 1 T
= gRX (2n + 1) − t − gRX (2n + 2) − t + gRX (2n + 3) − t
2 2 2 2 2

From the above, we draw the conclusion that

3 1
z0 [n] = z[2n + 1] − z[2n + 2] + z[2n + 3]
2 2
holds.
b. The number of required states is M L , where L is the memory length. The finite
memory condition is
h[n] = 0, |n| > L.
It is clear that here we have

h[n] = 0, |n| > 1.

Thus, L = 1. The number of needed states is then

M L = 41 = 4.

Task 5.2

Hints
a. The minimum required bandwidth Bmin = Rb /ηB , where ηB = log2 (M )/D.

b. To
have the minimum number of states, we need to sample in the time instances
k + 21 T . In this case L = 2.

Answer
a. The minimum required bandwidth is 2 MHz
b. We need 64 states.
8

Task 5.3

Task
For BPSK-(±1) signaling in the standard MLSE setting, suppose that the channel
memory is L = 1, with h[0] = 1 and h[1] = −0.3.

a. What is the maximum pairwise error probability, as a function of the received

Eb /N0 , for two bit sequences that differ only in the first two bits? Express your
answer in terms of the Q function.
b. Plot the transfer function bound (log scale) as a function of Eb /N0 (dB). Also plot
the error probability of BPSK without ISI for comparison.

Solution
a. The pairwise error probability is according to page 233 given by
 
||s(b + 2e) − s(b)||
Pe = Q ,
2σ
where s(b) is the signal corresponding to the bit sequence b, i.e.
X
s(b) = b[n]p(t − nT ),
n

where p(t) is the used pulse. Furthermore, σ 2 = N0 /2 is the PSD of the noise and
e is a valid error sequence. According to the reasoning on page 233, we also have
r !
h[0]e2 [1] + h[0]e2 [2] + 2h[1]e[2]e[1]
 
||s(e)||
Pe = Q =Q
σ σ2
r !
2h[0] + 2h[1]e[2]e[1]
=Q ,
σ2

where we have used the fact that the error sequence is non-zero only in the first two
positions. Since the Q function is strictly decreasing, the expression is maximized
when the nominator inside the square root is minimized. We have four alternatives,
namely
(e[1], e[2]) = (1, 1) : 2h[0] + 2h[1]e[2]e[1] = 2h[0] + 2h[1] = 1.4
(e[1], e[2]) = (1, −1) : 2h[0] + 2h[1]e[2]e[1] = 2h[0] − 2h[1] = 2.6
(e[1], e[2]) = (−1, 1) : 2h[0] + 2h[1]e[2]e[1] = 2h[0] − 2h[1] = 2.6
(e[1], e[2]) = (−1, −1) : 2h[0] + 2h[1]e[2]e[1] = 2h[0] + 2h[1] = 1.4
Then we have
r ! s  ! r !
2h[0] + 2h[1] Eb h[1] Eb
Pe,max = Q =Q 4 1+ =Q 2.8 ,
σ2 N0 h[0] N0

where we have used that we have the bit energy

Eb = h[0]E{|b[n]|2 } = h[0].
 9

b. This task is almost identical to Example 5.8.2 on pages 238–240. We have the
same coefficients as we have there, i.e.
h[0] E
− Nb
a0 = e− 2σ2 = e 0 ,
 
h[0]+2h[1] E 2h[1] Eb
− − Nb 1+ h[0] −0.4
a1 = e 2σ 2 =e 0 =e N0 ,
 
h[0]−2h[1] E 2h[1] Eb
− − Nb 1− h[0] −1.6
a2 = e 2σ 2 =e 0 =e N0 .
We also have the same error probability expression as we have there,
E
1 1 − Nb
2 a0 e 0
2
Pe ≤  2 =  2
1 − 21 (a1 + a2 )
 E E
1 −0.4 Nb −1.6 Nb
1− 2 e 0 + e 0

For ISI-free communication, ~e = (±1, 0, ...)

ks(~e)k2 = h[0]e2 [1] = h[0].
We have
r !
h[0]
Pe ≤ Q
σ2
r !
2Eb
=Q .
N0

A plot of these two:

0
10
Transfer function bound
−2 ISI-free communication
10

−4
10

−6
10

−8
Pe

10

−10
10

−12
10

−14
10

−16
10
0 5 10 15
E b/N0 (dB)
10

Task 5.4

Hints
a. Simply check the Nyquist criterion.
b. Be inspired by task 5.1.
c. Start with the usual metric in Equation 5.13 on page 206, which is to be maximized,
and make use of the fact that all involved signals are real valued. Determine h[0] and h[1]
for the given situation. (h[n] = 0, for|n| > 1) Simplify the metric, remove unnecessary
constants and change its sign such that the maximization is replaced by minimization.
d. Use the definitions on pages 236–237.

Answer
a. No, the pulse is not Nyquist.
b. Choose T s = T and τ = T /2. ZMF [n] = 1/2(r[n] + r[n +
1])



c. Metric: mn b[n], s[n] = b[n] b[n − 1] − 2T r[n] + r[n + 1]

1−2T (r[1]+r[2]) 1−2T (r[2]+r[3])

+1 +1 +1

−1−2T (r[2]+r[3])
−1+2T (r[1]+r[2])
−1+2T (r[2]+r[3])

−1 −1
1+2T (r[2]+r[3])

d. η = 1

Task 5.6

Task
We would like to develop a model for simulating the symbol rate sampled matched filter
outputs for linear modulation through a dispersive
P channel. I.e. we wish to generate
the samples Z[k] = (Y ∗ pMF )(kT ) for Y (t) = n B[n]p(t − nT ) + W (t), where pMF (t)
is matched to p(t) and where W (t) is complex WGN.

a. Show that the signal contribution Zs to Z[k] can be written as

X
Zs [k] = B[n]h[k − n] = (B ∗ h)[k],
n

where we have h[l] = (p ∗ pMF )(lT ).

 11

b. Not part of the plan Show that the noise contribution ZW [k] to Z[k] is WSS, proper
complex Gaussian random process with zero mean and covariance function
∗
CZW [k] = E {ZW [l]ZW [k − l]} = 2σ 2 h[k].

For real-valued symbols, signals and noise, h[k] are real and

CZW [k] = E {ZW [l]ZW [k − l]} = σ 2 h[k].

c. Now specialize to the running example in Figure 5.1, with BPSK signaling (B[n] ∈
{±1}). We can now restrict Y , p and W to be real-valued. Show that the result
of a specializes to
3 1
Zs = B[k] − (B[k + 1] + B[k − 1]).
2 2
Show that the result of b specializes to
 
3 1
RZW [z] = σ 2 − (z + z −1 ) , T ask
2 2

where the PSD RZW [z] is the z-transform of CZW [k].

d. Suppose that W [k] are i.i.d. N(0, 1) random variables. Show that this discrete-time
WGN sequence can be filtered to generate ZW [k] as follows:

ZW [k] = g[0]W [k] + g[1]W [k − 1].

Find the coefficients g[0] and g[1] such that ZW [k] has statistics as specified in c.
Hint: Factorize RZW [z] = (a + bz)(a ∗ +b ∗ z −1 ) by finding the roots, and use one
of the factors to specify the filter.
e. Not part of the plan Use the results in a-d to simulate the performance of MLSE
for the running example. Compare the resulting BER (Bit-Error Rate) with that
obtained using the transfer function bound.
12

Solution
a. Let Ys (t) be the contribution of the signal to Y (t), i.e.
X
Ys (t) = B[n]p(t − nT ).
n

Then Zs [k] can be rewritten as

Z∞ X
Zs [k] = (Ys ∗ pMF )(kT ) = B[n]p(t − nT )pMF (kT − t) dt
−∞ n

X Z∞
= B[n] p(t − nT )pMF (kT − t) dt.
n −∞

The variable substitution τ = t − nT gives us

X Z∞ X

Zs [k] = B[n] p(τ )pMF ((k − n)T − τ ) dτ = B[n](p ∗ pMF ) (k − n)T
n −∞ n
X
= B[n]h[k − n] = (B ∗ h)[k],
n

which is what was supposed to be shown.

b. A linear transformation on complex WGN results in a proper Gaussian random
process (Chapter 4). Wide-sense stationarity is preserved as described in Appendix
A.
c. For the example in Figure 5.1, we have (according to the box on page 205)

3/2,
 k = 0,
h[k] = −1/2, |k| = 1,

0, else,


Consequently, we have
X
Zs = B[n]h[k − n] = h[1]B[k − 1] + h[0]B[k] + h[−1]B[k + 1]
n
3 1

= B[k] − B[k + 1] + B[k − 1] .
2 2
We have real-valued entities, which gives us the covariance function

CZW [k] = σ 2 h[k]

according to b. With our h[k], this becomes


3 2
2σ ,
 k = 0,
CZW [k] = − 21 σ 2 , |k| = 1,

0, else,

 13

Then we have the z-transform

 
2 3 1
X
−k −1


RZW [z] = CZW [k]z = σ − z+z .
2 2
k

which is what was supposed to be shown.

d. We want to factorize RZW [z] on the form
RZW [z] = (a + bz)(a∗ + b∗ z −1 ).
Then we can use one such factor as the transfer function of a filter. That filter has
the correct output according to the super formula. The first factor corresponds
to a non-causal filter, while the second factor corresponds to a causal filter. Of
course, we choose the second one. We solve the related quadratic equation
1 3 1
− x2 + x − = 0,
2 2 2
which has the two roots
√
3− 5
x0 = ,
2√
3+ 5
x1 = = x−1
0 ,
2
and we have the factorization
z 2 − 3z + 1 = (z − x−1
0 )(z − x0 ).

Then we have
σ 2 −1 2
 
2 3 1
RZW [z] = σ − (z + z −1 ) =−
z (z − 3z + 1)
2 2 2
σ 2 −1 σ2
= (x0 − z)(1 − x0 z −1 ) = (1 − x0 z)(1 − x0 z −1 )
2 2x0
σ√ σ√
  
σ σ −1
= √ − 2x0 z √ − 2x0 z .
2x0 2 2x0 2
Thus, we get
σ σ
a= √=p √
2x0 3− 5
√
q
σ
b=− 3− 5
2
The causal filter has the transfer function
G[z] = a∗ + b∗ z −1 = a + bz −1
and impulse response

 √ σ√ , n = 0,
 3−p
 5
√
g[n] = aδ[n] + bδ[n − 1] = − σ 3 − 5, n = 1,

 2
0, else.


e. —