ADVANCED DIGITAL COMMUNICATIONS

(CDA-SISTEL) and DIGITAL

COMMUNICATIONS (DC- MET Bridge )

Margarita Cabrera, Jaume Riba and Gregori Vázquez

February 8, 2021

1
Introduction: Some basic equations

+∞
1. Q (x) = .
2 /2 2 /2
√1
2π x
e−λ dλ < e−x

+∞ √ 1
2. 0
Q ( x) dx = .
2
3. For a real Gaussian random variable x ∼ N (0, σ 2 ), the p.d.f. is given by
1 x2
f (x) = √ e− 2σ2
2πσ

4. For a complex Gaussian random variable x ∼ CN (0, σ 2 ) and z = |x|2 , the p.d.f.
are given by
1 − |x|22
f (x) = e σ
πσ
1 − z2
f (z) = e σ ; z ≥ 0.
σ2
5. The unitary N xN Fourier matrix is dened as
 
1 2π
F = [fm,n ]0≤m,n≤N −1 with fm,n = √ exp −j mn .
N N

6. Basic series
M
X M (M + 1)
m=
m=1
2
M
X M (M + 1) (2M + 1)
m2 = .
m=1
6

2
1 Matched-Filter and Scalar Products

In this exercises we will study the relationship between conventional linear ltering
and signal scalar products. The aim of the exercise is to provide a practical descrip-
tion/implementation of the formal signal scalar product. In this sense, we dene the
scalar or inner product between two complex signals a (t) and b (t) as follows

 +∞
H
a b= a∗ (t) b (t) dt.
−∞

1. Consider a linear lter with impulse response h (t) = a∗ (td − t) where td is a

time-delay to guarantee that the lter response is causal. Find the output of
the former
lter to an input signal b (t) as a function of the cross-correlation
Rba (τ ) = −∞ b (t + τ ) a∗ (t) dt.
+∞

2. Show that the direct implementation of the scalar product as a linear lter is
given by the following scheme.

NOTE: The lter with response h (t) = a∗ (td − t) is known as the matched-lter
to the signal a (t). Formally speaking, this is the matched-lter under Gaussian,
white-noise detection conditions.
3. Consider that the lter input b (t) is a stationary random signal with power
spectral density Sbb (f ). Give the expression of the power spectral density at the
lter output Syy (f ).
4. If b (t) is spectrally white, that is, for Sbb (f ) = N0 /2, nd the signal mean
 power
at the system output E |y (td )|2 . Under what conditions E |y (td )|2 = N0 /2?
Explain.

2 Signals and Vectors

Consider the following eight equally likely symbols {sm (t)}1≤m≤8 . We will analytically
represent this modulation in two dierent orthonormal basis:
 q  
 φ1 (t) = 2 Q t−T /4
BASIS #1: q T Q  T /2 
 φ2 (t) = 2 t−3T /4
T T /2
 Q  t−T /2 
 φ1 (t) = √1T
BASIS #2:
T
  Q  t−3T /4 
 φ2 (t) = √1
Q t−T /4 1
T T /2
− T
√
T /2

3
 1. Provide the symbol coordinates and modulation constellation in basis #1. Pro-
vide the symbol coordinates and modulation constellation in basis #2.
2. Compute the average symbol energy Es and average bit energy Eb and verify
that they do not depend on the adopted basis in the vector representation of the
signals.

3 Binary Transmission in a Multidimensional Signal

Space

A sequence of statistically independent and equally likely bits is transmitted through

an AWGN channel. The average bit energy is given by Eb and the symbol rate r
symbols/sec. The size of the alphabet is M = 2 such that the bit-rate rb can be
b

modied through the proper selection of parameter b. The basis of the signal space is
composed by L = b orthonormal signals {φl (t)}1≤l≤L and the symbols of the alphabet
in this basis, as follows

4
  
±d/2
..
sm =  . .
 
±d/2
The symbols are transmitted through an AWGN channel with a noise term w (t) with
a power spectral density Sww (f ) = N0 /2 Watts/Hz.
1. Find the average bit energy Eb as a function of the symbols minimum distance
d.

2. Depict the system constellation and optimal ML/MAP receiver for L = b = 3.

3. Find the exact bit error rate (BER ) of the system for an arbitrary L as a function
of the Eb /N0 ratio.
4. Find the exact symbol error rate Pe of the system for an arbitrary L as a function
of the Eb /N0 ratio.
5. Repeat (1.) to (4.) for a new constellation such that
 
±d/2 ± jd/2
..
sm =  .
 

±d/2 ± jd/2

where the symbols are transmitted through an AWGN channel with a complex
noise term w (t) with a power spectral density Sww (f ) = N0 Watts/Hz.
NOTE: Many real systems can be understood by following the methodology described
in this exercise, based on M-QAM modulations on a multidimensional signal space.

4 Non-Gaussian MAP Detection and Diversity (Wozen-

craft and Jacobs 1965, 1990)

Consider a binary transmission with equally likely and independent symbols s1 and
s2 with amplitudes an = +1 for symbol s1 and an = −1 for symbol s2 . For a more
robust detection, a redundant scheme is adopted at the receiver (diversity ). The trans-
mitted symbol is received at two antennae receiver with relevant information vector
υ = [υ1 υ2 ]T according to the gure below (left) where w1 and w2 are non-Gaussian,
statistically independent noise terms with statistical distribution given by
1
fn1 (n) = fn2 (n) = e−|n| .
2
The signal model admits the following vectorial representation
     
υ1 an n1
υ= = sm + n = +
υ2 an n2
.

5
 1. Find the joint probability density function of the noise components fn (n).
2. Find the likelihood functions f (υ/s1 ) and f (υ/s2 ) as a function of the p.d.f.
found in (1.).
3. Show that the MAP (Maximum a Posteriori ) criterion is given by
ŝm = argmaxsm f (υ/sm ) P (sm ) .

4. Show that the optimal decision regions for the problem at hand are given by

R1: |υ1 − 1| + |υ2 − 1| ≤ |υ1 + 1| + |υ2 + 1|

R2: |υ1 − 1| + |υ2 − 1| ≥ |υ1 + 1| + |υ2 + 1| .

5. According to (4.), the optimal decision regions are the ones depicted in the gure
above (right). Notice that two regions are ambiguous in the decision making,
that is, the receiver can decide in a random manner any of the two symbols. If
we replace the decision rule in (4.) by the following new one
R1 : υ1 + υ2 > 0

R2 : υ1 + υ2 < 0.

Is this new receiver also optimal under a MAP sense? Explain.

6. According to the new decision rule in (5.), the relevant information becomes
υ = υ1 +υ2 . Find the new likelihood functions f (υ/s1 ) and f (υ/s2 ). NOTE: The
probability density function of the sum of two statistically independent random
variables is given by the linear convolution of the two involved probability density
functions.
7. Find the BER of this communication system.

5 Thin-Film Hard Disk

We consider a magnetic hard disk (HDD ) reading scheme. The modulation is real
valued, binary and unipolar {sm (t)}m=1,2 . Bits are equally likely and statistically
independent. The modulation is as follows

6
 +∞
X
x (t) = an p (t − nT )
n=−∞

where am ∈ {0, +A}, T Q is the
bit/symbol period and p (t) is a rectangular pulse of
length T , that is, p (t) = t
T
.
We also take into account that the reading noise is AWGN but the noise power
spectral density depends on the read symbol. The signal model is as follows
r (t) = sm (t) + w (t)
such that the noise spectral density Sww (f ) is given by
(
σ12 f or s1 (t)
Sww (f ) =
σ22 f or s2 (t)
where σ22 > σ12 .
1. Find an orthonormal basis for this modulation. Represent the symbols {sm (t)}m=1,2
in the proposed basis as a function of the average bit energy Eb and plot the mod-
ulation constellation.
2. Find the likelihood functions for the two symbols.
3. Find the optimal decision rule that minimizes the symbol/bit error probability.
4. Find the average bit/symbol error probability for an arbitrary decision threshold.

6 Square QAM Modulations

6.1 QPSK/ 4-QAM

1. Dene a QPSK modulation with rectangular shaping.
2. Give the analytical expression of the four symbols as a function of the bit energy
Eb .
3. Depict the optimal MAP/ML receiver (symbols equally likely).
4. Find the exact symbol error rate as a function of the Eb /N0 ratio.
5. Find the exact bit error rate as a function of the Eb /N0 ratio.

6.2 16-QAM
1. Dene a 16-QAM modulation with rectangular shaping.
2. Give the analytical expression of the symbols as a function of the bit energy Eb
for equally likely symbols.
3. Depict the optimal MAP/ML receiver. Compute a good approximation of the
symbol error rate as a function of the Eb /N0 ratio.
4. Find a good approximation of the bit error rate as a function of the Eb /N0 ratio.

7
6.3 M-QAM
1. Dene a M-QAM modulation with rectangular shaping.
2. Give the analytical expression of the symbols as a function of the bit energy Eb
for equally likely symbols.
3. Depict the optimal MAP/ML receiver. Compute a good approximation of the
symbol error rate as a function of the Eb /N0 ratio.
4. Find a good approximation of the symbol error rate as a function of the Eb /N0
ratio.
5. Find a good approximation of the bit error rate as a function of the Eb /N0 ratio.

7 Some Case Studies

1. Consider the following four symbols with equal probabilities dened over an or-
thogonal basis
       
0 1 −1/2
√ −1/2
√
s1 = ; s2 = A ; s3 = A ; s4 = A .
0 0 3/2 − 3/2
Find a tight upper-bound of the symbol error probability as a function of the
Es /N0 .
2. Two QPSK-Rings : Consider the following octal modulation with equally likely
symbols dened on an orthonormal basis
    
  
1 −1 1 −1
s1 = A ; s2 = 2A s3 = A s4 = 2A
1 1 1 1
       
1 1 −1 −1
s5 = A ; s6 = 2A s7 = A s8 = 2A .
−1 −1 −1 −1
Find a tight upper-bound of the symbol error probability as a function of the
Es /N0 . Propose a modication of the former constellation for improving the
symbol error probability under the same working conditions.
3. Unbalance-QPSK : Consider a quaternary digital modulation UQPSK (Unbalanced-
QPSK ) with the following alphabet

8
 a. Find an orthonormal basis and represent this modulation in this basis. Draw
the modulation constellation.
b. Find the average bit energy Eb .
c. Find the optimal MAP /ML receiver. Symbols are equally likely.
d. Find the symbol error probability Pe .
4. Consider the following binary PDM (Pulse Duration Modulation ) modulation

a. Find an orthonormal basis and represent this modulation in this basis with
respect to the average bit-energy Eb for this modulation.
b. Draw the modulation constellation. Find the optimal and simplest MAP /ML
receiver. Symbols are equally likely and the noise component is real and dis-
tributed according to n ∼ N (0, N0 /2).
c. Find the exact bit-error rate BER for the receiver in (b.). Is this modulation
power/energy ecient?. Explain.
Consider that the symbols are now transmitted through a complex Rayleigh
channel with h ∼ CN (0, 1)). The receiver input noise is complex AWGN with
noise power spectral density Sww (f ) = N0 , such that the relevant informa-
tion components at the matched lter outputs are given by the following model
= hsm + n. The receiver has a perfect channel-state information
T
ν = ν1 ν2
(CSI ).
d. Find the optimal MAP /ML receiver.
e. Find an upper-bound for the average bit-error rate BER for the receiver in
(e.).
f. Find the optimal MAP /ML receiver if the receiver does not have CSI.
g. Study the average bit-error rate BER for the receiver in (g.).
5. Consider the following quaternary modulation:(4-PPM )

9
 Symbols are all equally likely, statistically independent and the transmission is
through a real AWGN with noise power spectral density N0 /2 Watts/Hz.
a. Describe the four symbols in an orthonormal basis as a function of the average
bit energy (Eb ).
b. Analyze the optimal MAP /ML receiver and depict it.
c. Compute the exact symbol error rate for the optimal receiver as a function of
the received Eb /N0 .
d. The transmission is now through an unknown multiplicative Rayleigh channel
(F 3 ) with h ∼ CN (0, 1). The noise component is complex with noise spectral
density N0 Watts/Hz. Analyze and depict the optimal receiver with minimum
symbol error probability for this new scenario.

8 Binary Orthogonal MAP and ML Detection

Consider a binary orthogonal modulation dened from the two symbols

   
1 0
s1 = A ; s2 = A
0 1

with probabilities P (s1 ) and P (s2 ). The channel is a real AWGN with noise spectral
density N0 /2 and the signal model
   
υ1 n1
υ= = sm + .
υ2 n2

1. Find the MAP and the ML optimal decision rules of this system.
2. Check that, in general, the optimal detector can be expressed as follows
ŝm = s1 if αυ1 + βυ2 + γ > 0

ŝm = s2 if αυ1 + βυ2 + γ ≤ 0

and nd constants α, β and γ for the MAP and ML cases.

9 Impulse Radio: Ultra-Wideband (UWB ) Commu-

nications

In an Impulsive-Radio UWB (Ultra-Wide Band ) system, a burst of equispaced short

pulses (monocycles ) are transmitted. In this example and for the shake of simplicity, a
binary modulation is considered following a 2PPM (Binary Pulse-Position Modulation )
scheme.
Considering a simplied analysis, we can model the two UWB symbols according
to

10
 The symbols are sequentially transmitted following the general expression given by
+∞
X
xT (t) = si[n] (t − nT )
n=−∞

with statistically independent symbols and equally likely. At the receiver, the noise
term w (t) is a real AWGN component with power spectral density N0 /2.
1. Provide the signal space size K and an orthonormal basis for such space.
2. Represent the two symbols s1 and s2 in the orthonormal basis given in (1.).
Depict the modulation constellation.
3. Obtain the MAP /ML decision rule of the two symbols and depict the two decision
regions in the modulation constellation in (2.).
4. Depict the optimal MAP /ML receiver for this modulation. In order to make the
receiver design canonical, make the design such that a unique lter is used. Give
the impulse response of this lter and be sure that the energy of the lter impulse
response is normalized.
5. Compute the BER of the system as a function of the received signal Eb /N0 .
6. Consider now an alphabet with four equally likely symbols, such that s3 =
−s1 and s4 = −s2 . Find the new optimal MAP /ML receiver and the exact
BER as a function of the received signal Eb /N0 .

10 MAP and Minimum Symbol Error Probability

Criteria

What is the relationship between the Maximum A Posteriori (MAP ) and the Minimum
Symbol Error Probability criteria. Provide a mathematical proof of your answer.

11 Orthogonal Modulations and Some Bounds

We consider a complex valued orthogonal M-ary modulation scheme. All symbols are
equally likely and given by
Es φm (t), for m = 1, 2, . . . , M,
p
sm (t) =

11
where {φm (t)}1≤m≤M is an orthonormal basis and Es is the average symbol energy. The
received signal is r(t) = sm (t) + w(t), with a noise power spectral density at receiver
of Sw (f ) = N0 .
1. Give the vectorial representation of theM -ary modulation in this basis as a func-
tion of the average bit energy Eb .
2. Find the optimal MAP/ML detector for this modulation.
3. Calculate an upper-bound for the symbol error probability as a function of Eb /N0 .
4. Considering Gray coding for providing an approximate expression of the BER.
Study the behavior of the BER when the constellation size is arbitrarily large
(M → ∞). Find a minimum value for the Eb /N0 that makes the BER arbitrarily
small (BER → 0) as M → ∞.

12 Quasi-Orthogonal Modulation

We consider a modulation with M equally likely symbols sm such that

sm (t) = A pT /M (t − mT /M ) − 1/M pT (t) ; m = 0, 1, 2, ..., M − 1,

where pT (t) is a rectangular pulse of length T within the interval 0 ≤ t ≤ T . As an

example, the following gure illustrates the alphabet symbol waveforms for M = 4.

The signal model is a real AWGN channel with noise power spectral density N0 /2.
1. Find an orthonormal basis and the symbol components of the M -ary constellation
in this basis as a function of the Eb .
2. Find an upper-bound of the symbol error probability as a function of the Eb /N0 .
3. Give an approximation of the BER based on a Gray's encoding. Study the
behavior of the BER for an arbitrarily large constellation size (M → ∞). Find
the minimum Eb /N0 such that BER → 0 when M → ∞.
4. Modify the given modulation denition to make it (full-)orthogonal. Repeat (3.)
for this new case.

12
13 Quaternary Modulation and CSI
Consider a digital quaternary modulation with equally likely symbols given by
so (t) = Acos (2πf0 t) p (t)

s1 (t) = Acos (2πf1 t) p (t)

s2 (t) = Acos (2πf2 t) p (t)
s3 (t) = Acos (2πf3 t) p (t)
where p (t) is a rectangular pulse of duration T , frequencies fm = (N + m) /T for
m = 0, 1, 2, 3 and N an integer. We consider an AWGN signal model with a real noise
power spectral density N0 /2.
1. Find the average bit energy Eb .
2. Find an orthonormal basis and represent the four symbols in this basis.
3. Derive the minimum symbol error rate optimal receiver and derive a tight upper-
bound for the symbol error probability.
We now consider that the channel has a dierent response for each symbol. The
model is such that r (t) = hm sm (t) + w (t) with h0 = 0, h1 = h2 = h3 = 1.
4. The receiver does not know the channel response (no CSI at the receiver) and
the decision rule is the one found in (2.). Is the receiver still able to work under
these conditions if the received bit energy is arbitrarily high?
5. We now consider that the receiver has a perfect CSI. The design of the receiver
is modied for an optimal minimum symbol error probability including this new
knowledge. Depict the received signal constellation and derive the new optimal
MAP /ML receiver and a tight upper-bound for the symbol error probability.

14 Quaternary Orthogonal Modulation and CSI

Assume a digital transmission using the following four equiprobable waveforms
 
2πt
s0 (t) = A sin pT (t)
T
 
4πt
s1 (t) = A sin pT (t)
T
 
4πt
s2 (t) = −A sin pT (t)
T
 
6πt
s3 (t) = A sin pT (t)
T
where pT (t) is a rectangular pulse of unit amplitude and duration T .
Suppose rst that the received signal is r(t) = hsm (t) + w(t), with m = 0, 1, 2, 3,
and h a real, non-negative channel. The noise term w(t) is real and Gaussian with
spectral density Sw (f ) = N0 /2

13
 1. Find the optimal symbol detection decision rule for a known channel h. Evaluate
the symbol error probability of the optimal rule.
Suppose from now on that the received signal is r(t) = hm sm (t) + w(t), with
h0 = 0 and h1 = h2 = h3 = h.
2. Evaluate a lower-bound of the symbol error probability using the same rule given
in (1.). Explain the achieved result. Hint: For nk ∼ N (0, σ 2 ), k = 0, 1, 2 and
B>0
 
B
P r (max (B + n1 , −B + n1 , n2 ) > n0 ) > P r (n0 − n1 < B) = 1 − Q √ .
σ 2
3. We now assume that the receiver has full CSI, that is, it knows the values of
hm for m = 0, 1, 2, 3. Specify the new ML decision rule. Draw the new decision
regions on a plane. Evaluate the symbol error probability associated with the
optimal detector and critically compare it with that obtained in the previous
sections.

15 Transmission Through Two Parallel AWGN Chan-

nels

Consider the transmission of two information bits (b0 and b1 ) through two parallel
AWGN channels with dierent gains (h0 = 1 and h1 = α) as illustrated in the gure

.
The transmitter assigns one symbol x = [xo x1 ]T to each
 combination of the equally
likely input bits such that the mean bit energy is Eb = E kxk /2. The channels aect
2

the symbol with a dierent gain per component. The received relevant information is
y = [yo y1 ]T = [xo αx1 ]T where α is the gain of channel h1 with 0 ≤ α < 1.
The two noise components are statistically independent with mean power σ 2 = N0 /2
each.
The purpose of the problem is to study in detail two dierent options for the
transmission of information, each one characterized by a particular constellation of
transmitted symbols x = [xo x1 ]T and a particular detection strategy. Finally, we want
to fully understand the advantages and drawbacks of every option.

15.1 Option 1: Binary transmission scheme, independent per

canal.

Consider the design of the transmitted constellation, specied in the table

14
 1. Specify the optimal detector of the symbols. Does the receiver need to know α?
2. Specify the optimal detector for the bits b̂0 and b̂1 from z0 and z1 . Find the exact
symbol error probability and an upper bound on the BER as a function of the
Eb /N0 and α.

3. If α = 0, can we transmit reliably?

15.2 Option 2: Quaternary transmission scheme with repeti-

tion over each channel.

Consider now the design of the transmitter constellation specied in the table

1. Specify the optimal detector of the symbols. Does the receiver need to know α?
2. Find the exact symbol error probability as a function of the Eb /N0 and α.
3. Prove that the detector can be implemented from c = z0 + αz1 .
4. If α = 0, can we transmit in a reliably manner?
5. Which option would you prefer depending on the value of α?

16 Transmission Through Three Parallel AWGN Chan-

nels

Consider binary antipodal transmission through three independent real channels. Iden-
tical bit is transmitted through the three channels, i.e.
υ1 = a + n 1

υ2 = a + n 2
υ3 = a + n 3
√ √
where a may take the values + Eb (for bit=0) and − Eb (for bit=1) with identical
probabilities. The three noise terms are independent and identically distributed as
N (0, N0 /2).

1. An engineer applies his intuition and denes the receiver rule as follows: Use
independent ML receivers per channel, and then use a majority vote rule to take
. This means, for example, that, if receiver
the nal decision of the transmitted bit
1 and 2 decide bit=0 and receiver 3 decide bit=1, the nal decision will be bit=0,
because it is the majority decision. Compute the bit error probability by using
this intuitive rule.

15
 2. An engineer who knows very well the detection theory and signal space concepts
(i.e. yourself) says that the previous strategy is wrong or at least, not-optimal.
Find the optimal detection rule for that problem. Find the resulting bit error
probability and check that the intuitive solution in (1.) performs worse than the
optimal one.

17 Robust Frequency Domain Transmission

√ √
We consider a QPSK transmission with equally likely symbols sm = ± Eb ± j Eb .
Symbols are simultaneously transmitted at two frequencies such that the received signal
model is given by
r (t) = (h1 φ1 (t) + h2 φ2 (t)) sm (t) + w (t)
√  
T Q t
where φk (t) = exp (j2πfk t) for k = 1, 2. We consider fk T arbitrarily large.
T T
The two channel responses at the two frequencies hk , k = 1, 2 are complex valued and
known by the receiver. The noise term w (t) is complex, Gaussian and white with noise
spectral density N0 .
1. Provide a vectorial model for the relevant information at the receiver front-end.
2. Find the optimal MAP/ML receiver for this communication system.
3. Find the exact BER for the optimal receiver in (2.).

18 Slow-Variant Frequency Flat Fading Channels

Assume a MISO system with two transmitting antennas and one receiving antenna.
The received signal is given by
r[k] = h1 x1 [k] + h2 x2 [k] + w[k] = hT x[k] + w[k]

where hT = (h1 , h2 ) with hl , l = 1, 2 complex channel gains, both known by the receiver
and the noise sequence w[k] ∼ CN (0, N0 ). The goal is to transmit two bits using two
channel uses, that is, at an average rate of one bit per channel use. For this purpose,
the transmitter adopts the following coding for each of the four possibilities of the two
bits to be transmitted
m x[1] x[2]
   
A 0
0
 0   A 
−A 0
1
 0   −A 
0 −A
2
 A   0 
0 A
3
−A 0

16
 We dene the received vector in two channel uses as
 
r[1]
r= = sm + w
r[2]

for m = 0, 1, 2, 3 and w =

T
w[1] w[2] .

1. Analyze the components of r[k] for all m. Specify the components of vectors sm
in the model given above r = sm + w and represent them graphically. Show the
decision regions of the maximum likelihood detector for this problem.
2. Find a bound on the average symbol error probability for all the possible channels
states and indicate the diversity factor of the proposed code. Hint: First analyze
the probability for a particular known channel state h and nd the mathematical
expectation of this probability with respect to the channel statistic.
3. For an average symbol error probability of 10−4 , evaluate the savings in trans-
mission power compared to a SISO scheme r[k] = hx[k] + w[k] with a binary
antipodal signaling x[k] ∈ {−A, A}.

19 Coherent and Non-Coherent Binary Detection in

Rayleigh Channels

Consider a binary transmission system based on two orthogonal equally likely symbols
   
A 0
s1 = ; s2 =
0 A

The signal propagates through a multiplicative Rayleigh channel h ∼ CN (0, 1).

The relevant information υ = [υ1 υ2 ]T at the matched lter outputs is given by
υ = hsm + n.

The two noise components in n are Gaussian, complex, zero mean with mean
powerN0 and statistically independent between them. There is no channel state infor-
mation at the receiver (no CSI at the receiver).
1. Find the likelihood functions f (υ/sm ) for m = 1, 2.
2. Find the optimal MAP /ML receiver for this communication system.
3. Find the exact bit error rate (BER ) for the former optimal / L receiver
MAP M
as a function of the Eb /N0 .
4. Find the exact BER of the optimal MAP /ML receiver for an AWGN channel
with a full channel state information at the receiver as a function of the Eb /N0
and compare it with the obtained in (3.). Find an upper-bound of the average
BER for all the channel states.

17
 5. The comparison of (3.) and (4.) predicts a severe degradation of the performance
in both the coherent and non-coherent detection because the Rayleigh channel
statistics. We will now try to generalize this result to any channel with channel
gain h such that z = |h|2 is distributed according to fz (z) with fz (0) > 0. Show
that under these conditions, the average BER or BER slowly decreases in the
Eb
high-SNR regime in proportion to the inverse of the working only, that is
N0
Eb fz (0)
limEb /N0 →∞ BER → = ctn.
N0 2

+∞ √ 1
HINT: Make use of that 0
Q ( x) dx = .
2

20 ML/MAP in a Multiplicative Channel. Bhat-

tacharya Bound

Consider a binary system with symbols s0 = 0 and s1 = 1 and associated probabilities

P (s0 ) = p and P (s1 ) = 1 − p. The relevant information is given by υ = h (Asm + 1)
where h is a frequency-at fading channel gain uniformly distributed in |h| ≤ 1/2 and
A is a known constant.

1. Find and plot the two likelihood functions f (υ/s0 ) and f (υ/s1 ) .
2. Determine the ML decision rule in terms of the relevant information υ .
3. Find the error probability of the decision rule given in (2.).
4. Derive the general expression for the Bhattacharya bound.
5. Compute the Bhattacharya bound of the bit-error rate (BER ) from the likelihood
functions in (1.) and compare it with the error probability derived in (2.) for
P (s0 ) = P (s1 ) = 1/2.

6. Determine the MAP decision rule.

21 Space Shift-Keying (SSK ) Modulation

In the design of a new communications system for sensor networks we try to technically
simplify the transmitting terminal to reduce the power consumption and cost as much
as possible. The entire computation burden is moved to the receiver, with a much
higher technical complexity than the transmitter. For this purpose, we adopt the so
called SSK (Space Shit-Keying ) modulation. In this case study we will assume a two
antennae transmitter as for the gure.
For each symbol, a single antenna transmits amplitude +A during T seconds. Dur-
ing this time interval, the other antenna is not transmitting. The receiver must decide
which of the two antennae was transmitting for the symbol detection. As we see, this
system is binary and one bit per symbol is transmitted. The receiver has complete
channel state information (CSI ). Symbols are independent and equally likely.

18
 The channels between the transmitting antennae and the receiver antenna are both
complex and multiplicative (frequency-at fading F 3 ) channels, with responses h0 =
α0 ejθ0 and h1 = α1 ejθ1 .
The noise term n follows a CN (0, N0 ) distribution. We dene bit or symbol s0 as
the one such that the transmitting antenna is #0 and symbol or bit s1 the one for the
opposite case where the transmitting antenna is #1.

1. Draw the modulation constellation at the receiver for arbitrary values of h0 and
h1 .

2. Compute the two likelihood functions f (υ/sm ) for m = 0, 1 as a function of and

the two channel gains.
3. Give the optimal MAP/ML receiver for this communication system.
4. Compute the bit-error rate (BER ) for the optimal MAP/ML receiver. Give the
solution as a function of the SNR dened as the symbol energy per symbol period
with respect the noise power ratio.
5. For time-variant independent Rayleigh channels with h0 and h1 distributed as
CN (0, 1) compute the average bit-error rate.

6. If the CSI is also available at the transmitter, how it is possible to modify the
former SSK modulation for achieving a minimum bit-error rate with the same
transmission power budget. Hint: Use a Lagrange multiplier technique.

22 Sensor Network

Consider a sensor network that it is monitoring a given environmental magnitude.

The network is composed by L sensors. Each sensor is associated to an alarm (one
bit) xl ∈ {0, +A} for l = 1, 2, ..., L. Amplitude A is a known real magnitude. The
information provided by the sensors are i.i.d. and they propagate through known
complex frequency-at fading channels with responses given by {hl }1≤l≤L to a single
receiving monitoring node. The noise at the receiver w is complex AWGN with mean
power N0 . Thus, the receiver must to cope with a total number of network states given
by 2L and each one of these states is labeled as a symbol sm with m = 1, 2, ..., 2L .
1. Derive a vector signal model for this scenario.

19
 2. Derive the optimal MAP receiver/detector for this problem. Notice that the
receiver has a complete CSI associated to all the sensing nodes.
3. Derive an upper-bound for the network symbol error probability.
4. Derive an upper-bound for the network false-alarm probability, that is, for the
probability that the receiver wrongly detects at least one active sensing node
when in fact all them are inactive.
5. We now consider that the channel responses are known Rayleigh i.i.d. with
distribution CN (0, σn2 ) for n = 1, 2, ..., L. Give an upper-bound for the network
average false-alarm probability.

23 'Just-Gain' vs Diversity Gain

Consider a repetition code of length L through independent and identically distributed

Rayleigh channels {hl }1≤l≤L with distributions CN (0, 1). The noise at the receiver is
a complex AWGN term with mean power N0 . Give all the results as a function of the
SN R, dened as the average symbol energy per symbol period T to noise power ratio,
at detection.
1. Derive the√MAP detector
√ for the repetition code if we transmit a QPSK signal
sm = ±A T ± jA T . Consider full CSI at the receiver.

2. Derive an upper-bound of the average symbol error probability as a function of

the SN R.
3. We now consider that the L channels are full-correlated and that they only dier
on the (carrier ) phase response, such that |h| = |hl | for l = 1, 2, ...L. Derive an
upper-bound of the average symbol error probability as a function of the SN R
for this new scenario.
4. Analyze the dierences between the independent and full-correlated channel sce-
narios in the expected average performances.

24 Space Diversity: Receiver Beamforming

For the communications system of the gure, the transmitted symbol sm belongs to
an alphabet of arbitrary size with independent, equally likely symbols. The receiver
consists of two antennae. The signal model at the receiver for a transmitted symbol
sm is given by
υ0 = h0 sm + n0
υ1 = h1 sm + n1
where n0 and n1 are two complex, Gaussian, independent, zero mean noise terms,
with power spectral density N0 Watts/Hz.

20
 1. The outputs of the two antennae are linearly combined according to υ = λ0 υ0 +
λ1 υ1 . In order to nd the so-called Maximum-Ratio Combiner (MRC ), nd λ0
and λ1 such that the output signal-to-noise ratio (SNR ) is maximized, with a full
channel state information (CSI ) and the receiver
E |λ0 h0 sm + λ1 h1 sm |2
 
{λ0 , λ1 } = argmaxλ0 ,λ1 SN R = argmaxλ0 ,λ1 =
E |λ0 n0 + λ1 n1 |2
 

2
E |sm |2 λH h
  
= argmaxλ0 ,λ1 . ,
N0 kλk2
with: λ = [λ∗0 λ∗1 ]T and h = [h0 h1 ]T . (Hint: Make use of the Cauchy-Schwartz's
inequality).
2. Find the optimal MAP/ML receiver for a generic transmitted constellation if we
have a full CSI and the receiver..
√ √
3. If we adopt a QPSK modulation with symbols sm = ± Eb ± j Eb , nd the
exact bit-error rate of the optimal receiver in (1.) and in (2.)
4. For two independent Rayleigh channels with hk ∼ CN (0, σk2 ) with k = 0, 1, nd
the average bit-error rate for all the channel states.
5. Consider that the two Rayleigh channels are unknown by the receiver (no-CSI )
and that now they show the same distribution CN (0, 1). The modulation
√ is
changed to a binary unipolar modulation such that s0 = 0 and s1 = 2Eb . Find
the new optimal MAP/ML receiver.

25 Space Diversity: Transmission Beamforming

For the communications system of the gure, the transmitted symbol sm belongs to an
alphabet of arbitrary size with independent, equally likely symbols. The transmitter
consists of two antennae and the receiver of a single antenna. The signal model at the
receiver for a transmitted symbol sm is given by
υ = (h0 w0∗ + h1 w1∗ ) sm + n

where n is a complex, Gaussian, zero mean noise terms, with power spectral density
N0 Watts/Hz.

21
 1. In order to nd a similar solution that for the so-called Maximum-Ratio Combiner
but now at the transmitter end, nd w0 and w1 such that the output signal-to-
noise ratio (SNR ) is maximized. Consider a full channel state information (CSI )
at the transmitter. In the design of the transmitting strategy consider that the
total transmitted energy remains unmodied, that is
E |sm |2 = |w0 |2 + |w1 |2 E |sm |2 ⇒ |w0 |2 + |w1 |2 = 1.
 
 

2. For the solution found in (1.), nd the optimal MAP/ML receiver.
√ √
3. If we adopt a QPSK modulation with symbols sm = ± Eb ± j Eb , nd the
exact bit-error rate of the optimal receiver in (1.) and in (2.)
4. For two independent Rayleigh channels with hk ∼ CN (0, 1) with k = 0, 1, nd
the average bit-error rate for all the channel states.
5. Compare and describe the advantages and drawbacks of the receiver and trans-
mitter beamforming solutions.

26 Alamouti's Space-Time Code

We consider the following two transmitting antennae and a single receiver antenna
communication system as for the gure.

Each antenna is transmitting a dierent QPSK symbol that they interchange in

the next channel-use. Thus, after the reception of two consecutive symbols, the signal
model becomes
υn = υ (nT ) = h0 s1 + h1 s2 + n0
υn+1 = υ ((n + 1) T ) = −h0 s∗2 + h1 s∗1 + n1

22
 1. Show that the former signal model admits the following matrix representation
     
υn s1 n0
r= ∗ =M +
υn+1 s2 n∗1

and show that M H M = M M H is a diagonal matrix.

2. Find the optimal and simplest MAP/ML receiver for the detection and decision
of the two symbols. Consider that the two independent noise components are
distributed as CN (0, N0 ). Recommendation: Try to nd the optimal receiver
as a function of the unitary channel matrix H = khk
1
M with channel vector
h = [h0 h1 ] .
T

3. Investigate the need of the channel state information (CSI ) at the receiver and/or
at the transmitter for this communication scheme.
4. We now consider that the transmitted symbols are QPSK with sm = ±A ± jA.
Find the amplitude A such that the total transmitted bit energy is Eb .
5. Give the optimal receiver for the QPSK symbols in (4.) and the associated BER
as a function of the Eb .
6. For two independent Rayleigh channels with hk ∼ CN (0, 1) with k = 0, 1, nd
the average bit-error rate for all the channel states.
7. Identify the advantages and drawbacks of the Alamouti's solution with respect
to the use of transmission beamforming and receiving beamforming.

27 Extended Alamouti's Space-Time Code

We now consider a simple extension of the Alamouti's space-time code for a four trans-
mitting antennae and a single receiving antenna system. The noise at the receiver is
AWGN with distribution n ∼ CN (0, N0 ) and the four channels {hk }1≤k≤4 . We con-
sider four options or cases depicted below. The matrix columns represent the antenna
space symbol allocation and the rows the time channel uses. As an example, for case
A, in the rst channel use antenna #1 transmits symbol s1 , antenna #2 symbol s2
and antennae #3 and #4 are not used. Antennae are spatially sorted from antenna
#1 to antenna #4.

1. If we consider the channel coherence-time and the channel coherence-distance,

choose the best transmission strategy among the four options. Give a detailed
justication of the adopted method.
2. Give the optimal MAP/ML receiver for the options given in (1.).

23
 3. We now consider that the transmitted symbols are QPSK with sm = ±A ± jA.
Find the amplitude A such that the total transmitted bit energy is Eb .
4. From (3.), nd the exact BER at the output of the optimal receiver as a function
of the Eb .
5. For four independent Rayleigh channels with hk ∼ CN (0, 1) with k = 1, 2, 3, 4,
nd the average bit-error rate for all the channel states.
6. ¾Can we modify the adopted solution in (1.) for a better use of the channel
diversity? Give an example for a positive answer and a justication, otherwise.

28 Two Unbalanced Receiver-Diversity Scenarios

1. Consider a communication system in which the same information is transmitted

through two independent Rayleigh channels {hk }k=1,2 . The transmitted format
is a binary polar modulation with s0 = +A/2 and s1 = −A/2 and equally likely.
The signal model at the receiver is given by

νk = hk sm + nk ; k = 1, 2 m = 0, 1.
The two channels show dierent average power attenuation such that the associ-
ated distributions are given by hk ∼ CN (0, σk2 ) for k = 1, 2. The two independent
noise components follow the same distribution nk ∼ CN (0, N0 ). We consider a
perfect channel-state information at the receiver (CSI ).
a. Find the likelihood functions f (νk /sm ) k = 1, 2 m = 0, 1.
b. Find the optimal minimum bit-error rate receiver for this problem.
c. Find the bit-error rate for the receiver in (b.) for two known channel states
h1 and h2 .
d. Find an upper-bound of the average bit-error rate for all the channel states
h1 and h2 .

2. Consider a communication system in which the same information is transmitted

through two independent Rayleigh channels {hk }k=1,2 . The transmitted format
is a binary polar modulation with s0 = +A/2 and s1 = −A/2 and equally likely.
The signal model at the receiver is given by

νk = hk sm + nk ; k = 1, 2 m = 0, 1.
The two channels show the same average power attenuation such that the associ-
ated distributions are given by hk ∼ CN (0, 1) for k = 1, 2. The two independent
noise components follow dierent distributions nk ∼ CN (0, σk2 )for k = 1, 2. We
consider a perfect channel-state information at the receiver (CSI ).
a. Find the likelihood functions f (νk /sm ) k = 1, 2 m = 0, 1.
b. Find the optimal minimum bit-error rate receiver for this problem.
c. Find the bit-error rate for the receiver in (b.) for two known channel states
h1 and h2 .
d. Find an upper-bound of the average bit-error rate for all the channel states
h1 and h2 .

24
29 Time Diversity in Rayleigh Channels

Consider the transmission of a codeword of length L symbols x = [x[1], . . . , x[`], . . . , x[L]]T

through a frequency-at fading channel h = [h[1], . . . , h[`], . . . , h[L]]T , with h` ∼
CN (0, 1) independent and identically distributed. The `-th received symbol is given
by
y[`] = h[`]x[`] + w[`], for ` = 1, . . . , L,
where the noise terms w[`] ∼ CN (0, N0 ) are independent and identically √ distributed.
Transmitter employs an antipodal BPSK modulation, i.e. x[`] = ± Eb , with inde-
pendent symbols over time. Symbols are equally likely with average bit energy Eb .
1. Find the exact and an upper-bound for the bit error probability as a function of
Eb /N0 assuming perfect channel state information at the receiver (CSIR) for any
single symbol x[`] detection within the codeword.
Assume now that transmitter is equipped with a perfect interleaver. In order
to exploit any potential diversity, we decide to use a repetition code, such that
x` = x for ` = 1, . . . , L. The vector signal model is now given by

y = hx + w,

where w = [w1 , . . . , w` , . . . , wL ]T is the additive noise vector with w` ∼ CN (0, N0 )

independent and identically distributed. Assume perfect channel state informa-
tion at receiver (CSIR).
2. Find the optimal decision rule for this problem.
3. Calculate an upper-bound for the average error probability. Which is the diversity
gain the optimal receiver can achieve? Compare the performance with that of
(1.).
For implementation reasons, we decide to consider now a selection combiner (SC)
instead of the optimal receiver. The new receiver performs the detection based
on the received signal along the branch with the strongest gain only, and ignores
the rest. We would like to compare the performance of the optimal combiner
in (2.) with that of the SC. At high Eb /N0 (Eb /N0 → ∞), the typical error
is known as deep fading. The probability of a deep fading event is given by
Pr {khk2 < 1/SNR}.
4. Find the probability of a deep fading event for the selection combining and op-
timal combining cases (use the given hints). Does SC perform better than the
No-diversity case in (1.)? How much diversity gain can SC achieve? Is there any
performance loss with respect to optimal maximum ratio combining in (2.)? If
so, quantify this performance loss.
Hint (I): Let {zi }1≤i≤L be i.i.d. exponential random variables with with parameter
1. Let z = maxi zi . The pdf of z is approximated by

f (z) ≈ Lz L−1 , for small z.

PL
Hint (II): Let zi ∼ CN (0, 1) i.i.d. for i = 1, . . . , L. Then, z= i=1 |zi |2 ∼ χ22L
with
1
f (z) ≈ z L−1 , for small z.
(L − 1)!

25
30 Optimality of Time-Diversity Codes

We consider L time-interleaved channels {hl }1≤l≤L which are independent and identi-
cally distributed hl ∼ CN (0, 1). We transmit codewords of length L through these
channels. The m-th codeword would be of the form
 T
xm = xm,1 xm,2 ... xm,L ; m = 1, 2, ..., M.

The relevant information at each channel-use output is given by

νl = hl xm,l + nl ; l = 1, 2, ..., L.

The following stacked vectorial model is recommended for the study

ν = Hxm + n

with ν = ν1 ν2 ... νL , H = diag h1 h2 ... hL and n = n1 n2 ... nL .

 T
 T

The M codewords are equally likely and the codebook {xm }1≤m≤M is normalized
such that
M
1 X
kxm k2 = 1.
M L m=1
All the noise components are independent, and identically distributed with nl ∼
CN (0, N0 ).

1. Use the union-bound to show that the codeword error probability is upper bounded
by
M M
1 X X
Pe ≤ P (xm → xm0 ) .
M m=1 m0 =1;m0 6=m

Find the expression of the pairwise probability terms P (xm → xm0 ).

1. Show from the Cherno bound that the average codeword error probability for
all the channel states is upper bounded by
L
Y 1
P (xm → xm0 ) ≤
SN R
l=1 1+ |xm,l − xm0 ,l |2
4
and give the expression of the SN R term.
2. Show that the average probability in (3.) can be upper bounded by
(M − 1) 4L
Pe ≤ SN R−L
minm6=m0 δm,m0

where δm,m0 = |xm,l − xm0 ,l |2 is the codeword .

QL
l=1 product-distance

26
31 Optimality of Space-Time Codes.

In a space-time scheme, we consider L transmission antennae and one receiving an-

tenna with marginal frequency-at fading channels all them i.i.d. hl ∼ CN (0, 1) and
stationary during the block transmission.
The information is transmitted in blocks of N symbols. The noise term is dis-
tributed according to wn ∼ CN (0, N0 ). Each codeword is represented by a matrix of
size LxN (rows-columns) X m .
1. Provide a vector/matrix model for this scenario and derive the exact pairwise
error probability between two codewords X m and X m0 . for a given known channel
state.
2. Derive an upper bound for the average error probability for the pairwise terms
in (1.)
h for all possible Rayleigh channel
i states as a function of the determinant
det (X m − X m0 ) (X m − X m0 ) . Explain the determinant criterion in the de-
H

sign of the codebook.

3. Adapt the result in (2.) for the study of repetition codes of length L.

Now we want to add a design constrain to the wireless system with a reduced
latency. We adopt a MISO (Multiple-Input Single-Output ) transmission sys-
tem with L transmitting antennae. For each symbol or channel-use we transmit
through a single antenna, being the other L − 1 antennae inactive. All the L
channels are i.i.d. Rayleigh distributed as CN (0, 1). In this way, we avoid the
use of long time-interleavers in slow time-variant scenarios in order to reduce la-
tency. The M codewords are equally likely with length L and the codebook is the
following {xm }1≤m≤M . The receiver noise term is AWGN with mean power N0 .
The receiver has a complete CSI and it knows the active transmitting antenna
in all the channel uses.
4. Derive an upper bound for the average pairwise probability term P (xm → xm0 )
for all the possible channel states.
5. Q
Adapt the solution in (4.) as a function of the product-distance terms δm,m0 =
l=1 |xml − xm0 l | , where xml stands for the l−th element of the m−th codeword.
L 2

32 Fast Frequency Flat Fading Channels

Assume a fast fading channel with the following signal model

r(n) = h(n)xm (n) + w(n),

where w(n) ∼ CN (0, N0 ) is a white noise term and h(n) is an i.i.d. channel at
fading with distribution CN (0, 1). All terms are mutually statistical independent.
The objective is to transmit two bits every four channel uses (rate of 0.5 bits /cu), for
which the following code is proposed with equally probable codewords (indexed with
m = 0, 1, 2, 3), with four channel uses each (indexed with n = 0, 1, 2, 3). We dene the

27
average bit energy as follows
3 3
!
1 1 XX
Eb = |xm (n)|2
2 4 m=0 n=0

and the SNR as γ = NEb0 . We want to analyze the following two codes, CODE #1 and
CODE #2, dened as follows
m xm (0) xm (1) xm (2) xm (3)
0 A A A A
CODE#1: 1 −A −A −A −A ;
2 A A −A −A
3 −A −A A A
m xm (0) xm (1) xm (2) xm (3)
0 A A A A
CODE#2: 1 −A −A −A −A .
2 A A −A −A
3 −A −A A A
We rst consider CODE#1 for questions 1,2,3, and 4.
1. Find the average bit energy Eb from the transmitted codewords as a function of
A2 .
2. Find the minimum product-distance between the transmitted codewords as a
function of Eb (dened as the product of squares of the dierence between two
codewords scalar magnitudes within the codeword). Justify the result and explain
what you can expect from this code in a fast fading environment.
3. Assume that the receiver implements the optimal detector with full knowledge of
h(n) for n = 0, 1, 2, 3. Observing the Euclidean distances between the received
codewords nd an upper bound (the tighter the better) of the average codeword
error probability, P̄e .
4. What is the diversity gain of the proposed code?
We now adopt CODE# 2 for questions 5,6,7 and 8.
5. Find the average bit energy Eb from the transmitted codewords as a function of
A2 .
6. Find the minimum product-distance between the transmitted codewords as a
function of Eb (dened as the product of squares of the dierence between two
codewords scalar magnitudes within the codeword). Justify the result and explain
what you can expect from this code in a fast fading environment.
7. Assume that the receiver implements the optimal detector with full knowledge of
h(n) for n = 0, 1, 2, 3. Observing the Euclidean distances between the received
codewords nd an upper bound (the tighter the better) of the average codeword
error probability, P̄e .
8. What is the diversity gain of the proposed code?
We now want to perform a comparison between the two proposed codes.

28
 9. If the target average error probability is P̄e < 10−4 , nd the minimum γ required
by each of the proposed codes. If we consider a high-SNR regime γ  1, explain
the reasons that make one of the codes much better than the other for a fast
fading channel.
10. Propose a code of the same rate (0.5 bits /cu) that require a smaller γ for the same
average error probability target P̄e < 10−4 . You can use of complex codewords if
you think it is necessary.

33 SVD in Frequency Selective Channels

Consider the following time-invariant frequency-selective channel with channel matrix

given by  
1 0
H = 1 1 .

0 1
Channel-State Information (CSI ) is complete and available at both transmitting
and receiving ends.
1. Use the fact that vectors v1 = √12 1 1 and v2 = √12 1 −1 preserve
 T  T

orthogonality after passing the channel for nding out the singular value decom-
position (SVD ) of the channel matrix H.
2. Is multipath a drawback in digital communications? Hint: Compare your result
in (1.) with the AWGN model with identical power gain.
3. Consider
√ that we want to transmit √two simultaneous
√ data streams, one BPSK (
±A T ) and a second QPSK ( ±A T ± jA T ) following a bit-loading scheme.
0 0

Find the relationship between A and A0 for the two streams to show the same
transmitted power. Depict the transmission-reception scheme and the BER for
each of the data streams.
4. Consider now the transmission through a frequency-at fading 3x2 MIMO chan-
nel with channel matrix  
1 1 0
H= .
0 1 1

Explain the dierences and similarities with respect to the systems designed in
(1.).

34 SVD in Frequency Selective Channels

We consider the transmission through a linear, time-invariant channel with impulse

response h (t) = δ (t) + 0.5δ (t − T ) where T is the symbol period. In order to avoid
the use of a linear equalizer for the inter-symbol interference compensation, we adopt a
design base on the singular value decomposition (SVD ) of the channel matrix H. The
parallel transmission is done with blocks of length N = 2, with s = s1 s2 .
 T

29
 The receiver noise is an AWGN term with distribution given by w (n) ∼ CN (0, N0 I).
1. Derive the channel matrix H. Verify that vectors v1 = √12 1 1 and v2 =
 T

preserve orthogonality after passing the channel for nding out

1
 T
√
2
1 −1
the singular value decomposition (SVD ) of the channel matrix H = UΛVH .
2. Give the general expression for the transmitting matrix F (n) and receiving matrix
G (n). In the solution, include the possibility of performing power control on the
two symbols/channels.
3. Depict the simplied diagram of the equivalent independent multiplicative chan-
nels.
4. Derive the expression of the average transmitted power as a function of F (n).
5. Derive the expression of the peak transmitted power as a function of F (n).
√
6. We√now transmit equally likely and independent QPSK symbols sk (n) = ±A T ±
jA T , k = 1, 2. Derive the exact expression of the global average BER at the
system output if we transmit the same power in the two channels and the total
transmitted power is ST = 2. Analyze if the obtained BER is better or worst that
the achieved in the same conditions if the signal propagates through an AWGN
channel.
7. Give the condition on matrix F (n) to guarantee the same BER at the output
of the two channels. In this case, the global BER will be equal to the marginal
BER for any channel. Compare this gure with the one derived in (6.). Explain
the result.
8. We now adopt a repetition code with length such as L = N = 2 with s (n) =
s1 (n) = s2 (n). Derive the optimal MAP/ML detector of the system and the
associated BER.

35 2x2 MIMO
Consider a 2x2 MIMO systems with channel matrix
 
1 α
H= .
α 1
There is a full channel state information at the transmitting and receiving ends.
The noise power at the two receiver channels are i.i.d. with nl ∼ CN (0, N0 ) for l = 1, 2.
We transmit two independent symbols with the two transmitting antennae being s1m
the symbol at antenna #1 and s2m the symbol at antenna #2.

30
 1. Design and depict a transmission-reception scheme such that the information of
the two transmitting antenna symbols are uncoupled at the receiver output. Do
not consider power allocation at the transmitting terminal.
2. Derive the equivalent end-to-end 2x2 multiplicative model.
√ √
3. If the symbols slm , l = 1, 2 are QPSK with slm = ± Eb ± j Eb , derive the exact
average bit error rate at the MIMO system output.

36 Analysis of an OFDMA System

Consider an ODFMA system where a given user's trac is allocated in three contiguous
carriers. The linear channel matrix before the cyclic prex removal is given by
 
h0 0 0

 h1 h0 0 
H=
 h2 h1 h0 
.
 0 h2 h1 
0 0 h2

1. Find the expression of the circular channel matrix after the removal of the cyclic
prex. Assume that the length of the cyclic prex was correctly dened for this
channel.
Consider all along the rest of the exercise that after the removal of the cyclic
prex, the channel matrix for this user is given by
 
1 0 1
H =  1 1 0 .
0 1 1

2. Evaluate the minimum power/energy loss because of the use of a cyclic prex in
the transmission for this scenario.
3. The three allocated carriers are contiguous according to the problem denition.
How can you modify the carrier allocation criterion if you know the maximum
coherence-bandwidth of the channel?
4. Depict the optimal transmission/reception scheme for this system if there is no
channel-state information at the transmitter.
5. Modify the design in (3.) if there is a full channel-state information at the
transmitter and provide dierent end-to-end design criteria.
6. Depict an equivalent end-to-end parallel equivalent TX-Channel-RX system show-
ing the absence of channel memory. Compute the equivalent channel gains.
7. For a complex Gaussian
√ noise
√ with spectral density N0 , nd the exact system
BER if a QPSK ± Eb ± j Eb modulation is transmitted in the three carriers.
Take into account the bit energy penalty due to the removal of the cyclic prex.

31
37 Frequency Selective Channels

We consider a digital transmission through a frequency selective multipath channel

r[k] = x[k] + x[k − 1] + x[k − 2] + w[k]
where the noise term w[k] is white and with distribution N (0, N0 /2). The transmitter
works in blocks of two channel uses. Consider the transmission of a single block formed
by x[0] and x[1] and that the receiver observes the sequence r[k], for k = 0, 1, 2, 3. We
dene a[k] as the binary information to transmit. Two binary symbols are transmitted
in a block with a[0], a[0] ∈ A, −A and we dene vectors
 T  T
x= x[0] x[1] , a= a[0] a[1]
Note that when transmitting two bits of information, the previous two vectors belong
to an alphabet of size four as the result of the four combinations of the two transmitted
bits, that is, x ∈ {x0 , x1 , x2 , x3 } and a ∈ {a0 , a1 , a2 , a3 }. Due to the length of the input
sequence and the channel length, we consider the following channel output observation
vector  
r[0]
 r[1] 
 r[2]  .
r= 

r[3]
1. Consider the transmission of xm = am with m = 0, 1, 2, 3. Find the optimal
sequence detector and evaluate its probability of error.
2. Based on the SVD of the channel matrix
 
1 0
 1 1 
H=  = USVH
 1 1 
0 1
with
√1 √1
 
10 2  √ " #
√2 0 √1 √1

5 0
 
10 2 2
U= S= V=
 
√2 √1 −1
0 0 1 √

 10  2 2
√1 −1
√
10 2
We assume the transmitted linear precoded information xm = Vam , with m =
0, 1, 2, 3. Prove a model the entire
TX-channel-RX system using two parallel
virtual AWGN channels and evaluate the average error probability of the system.
3. You will have noticed that the two bits transmitted through the two parallel
virtual channels suer a dierent attenuation. To improve the quality the encoder
we proposed to transmit xm = VDam where
 √
√

β √ 0
D= 2
0 1−β
and 0 < β ≤ 1/2. Notice that for β = 1/2 the new scheme is the same that
the one adopted in (2.) with uniform power distribution. Find the optimal value
of parameter β which equates the reception quality of both bits and nd the
associated probability of error.

32
  
1
4. We now consider the following encoder xm = A(2m−3)
√ ,with m = 0, 1, 2, 3.
2 1
Find the optimal detector and the associated error probability.
5. Finally, consider the general case of a matrix H with the same singular vectors
than before but with arbitrary singular values λ0 and λ1 . Find the condition
on the ratio λ0 /λ1 so that the encoder proposed in (4.) is better than the one
proposed in (3.)?

38 DS-CDMA and the RAKE Receiver

1. Give the general expression of a QPSK modulated DS-CDMA signal with a

processing gain L = T /Tc , chip pulse-shape c (t), symbol and chip periods T , Tc
and chips ±1.
2. Depict the general scheme of a RAKE receiver with P ngers or coecients.
Analytically study the selection and design criterion for the RAKE ngers.

k=0 hk δ (t − kTc ) with KTc < T ,

3. For a channel with impulse response h (t) = K−1
P
provide the exact BER at the RAKE output. Consider a complete RAKE with
P = K and the noise term as AWGN with power spectral density N0 Watts/Hz.

4. Find an upper-bound for the average BER if the channel coecients are i.i.d.
with hk ∼ CN (0, 1) for k = 0, 1, ..., K − 1.

33
(intentionally blank page)

34