Scribd
Upload
43 views4 pages

HW 3

This document contains 6 questions related to analyzing the capacity of MIMO wireless systems under different channel conditions and assumptions. Question 1 involves analyzing the capacity formula for a MIMO system with iid channel entries under fast fading with channel state information at the receiver (CSIR). Questions 2-4 analyze the capacity of different MIMO channel models under various assumptions about channel knowledge and reciprocity. Questions 5-6 analyze channel inversion techniques for a scalar Rayleigh fading channel and calculate the capacity of a specific 3x3 MIMO channel model with a given fading matrix and noise covariance matrix.

Uploaded by

durgeshsingh023
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
43 views4 pages

HW 3

This document contains 6 questions related to analyzing the capacity of MIMO wireless systems under different channel conditions and assumptions. Question 1 involves analyzing the capacity formula for a MIMO system with iid channel entries under fast fading with channel state information at the receiver (CSIR). Questions 2-4 analyze the capacity of different MIMO channel models under various assumptions about channel knowledge and reciprocity. Questions 5-6 analyze channel inversion techniques for a scalar Rayleigh fading channel and calculate the capacity of a specific 3x3 MIMO channel model with a given fading matrix and noise covariance matrix.

Uploaded by

durgeshsingh023
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Indian Institute of Technology Bombay

Department of Electrical Engineering


Handout 8
Homework 3

EE 764 Wireless and Mobile Communication


Mar 20, 2015

Question 1) We know that the capacity of a MIMO system with Gaussian iid channel
entries under fast fading and CSIR is,
C = E log(det(I + P

1
HH )),
Nt

(1)

where the expectation is over the fading matrix H. Let us assume that Nr Nt . Our
discussion applies to all kind of matrices with iid entries.
(a) Assume that H has arbitrary distribution, but fast-fading with CSIR is the underlying
model. What is the achievable rate if we still employ independent Gaussian distributions
at the transmit antennas.
(b) From (1),
Nr

C = E log(1 + P i ),
i=1

where the expectation is over the eigen-values (random) of N1t HH . We will now show that
C
Nr is easily evaluated in the limiting sense, i.e. by assuming Nr and Nt to be large. To this
end, denote
1 Nr
FH (x) =
1{i x} .
Nr i=1
Compute Fh ().
(c) Recall the idea of Riemann integral in calculus. Assume that the maximal Eigen value

of N1t HH , say max , is bounded almost surely (mild technical assumption), thus 0 i .
into n intervals, and call the lth interval Il . Argue that for any
Cut the interval [0, ]
continuous function g(),
Nr
n Nr

g(i ) g( l)1{i Il } ,
n
i=1
l=1 i=1
and
n

1 Nr

g(i ) g( l)(FH ((l + 1) ) FH (l ))


Nr i=1
n
n
n
l=1
where is used in the Riemann sense, i.e. the RHS becomes the LHS in the limit of large
l.
(d) Imagine a large, but fixed value of n. If you now increase Nr , what will happen. The
number of Eigen-values increases and the fraction of which falls in interval Il will give the
probability of obtaining an eigen value in Il . Intuitively, this is the frequency interpretation,
and it turns out that
x
FH (x) F (x) = f (u)du,
0

which is called the limiting cdf of the eigen values. This will tell us the asymptotic fraction
of Eigen values which lie below a given number. Argue that
max
1 Nr
g(x)dF (x)
g(i )
Nr i=1
0

(e) In the special case where Nr = Nt , the limiting Eigen-value distribution of N1t HH has a
very appealing form, a result shown by Marcenko and Pastur, known as the quarter-circle
law. In particular
x
F (x) = f (u)du,
0

where

1
4 u2 , 0 u 2.

Verify that f (u) is indeed a density function.


f (u) =

(f ) When Nr Nt , we can still have a closed form characterization of f (u), given by the
Marcenko-Pastur law (we do not state it here, amble references are available online.) In
general, show that
max
C

log(1 + P )f ()d.
Nr
min
Thus we can compute the rate per receive antenna by simply evaluating the line-integral.
Compare this with the complexity of integrating over the fading matrix.
C
for m m Gaussian
(g) Is the above formula good? Let us get a surprise by evaluating m
random matrices when m = 1, 2, . We expect the value to converge when m gets to a few
hundreds or thousands. Here is the plot with m, where P = 100 and the channel entries
are normalized Gaussians.

C
m

3.5

2.5

2
0

20

40

60

80
m

100

On the above plot, evaluate and draw the limiting value of

120

C
m.

140

160

Is infinity close to Eight?

Question 2) You have to submit Exercise 8.6 of the book by Tse and Viswanath, which

deals with the capacity of MIMO-CSIR systems.


Question 3) Imagine a time-varying MIMO system with CSIT and CSIR.
(a) If the channel remains constant for a block of n channel uses, and changes IID over
blocks, what is the capacity of the system. You can assume identical and independent
fading links for each tranmit-receive antenna pair.
(b) What is the capacity of the channel varies IID from instant to instant, i.e. under
fast-fading. Again assume identical and independent fading links for each tranmit-receive
antenna pair. Compare your answer with the previous part.
Question 4) Consider a fixed MIMO channel given by
y = Hu + z,
where H is an Nr Nt matrix, and z Nc (0, I) is independent of all inputs and the channel.
It is known that Nr Nt . Let C1 be the capacity of this channel under a power constraint
P.
Let us now construct a reciprocal system using the same apparatus. In particular, the
transmit antennas now become the receive antennas and vice versa. Let us represent this
MIMO by
w = H T v + z,

(2)

where z is Nc (0, I) independent of v. Let C2 be the capacity of the reciprocal system in


(2) under the same power constraint P .
If Nr = 2Nt , what is the relation between C1 and C2 ?
Question 5) Channel Inversion: Let us consider uncoded BPSK transmissions throughout this question. We know that the error probability behaves like exp() for an AWGN
channel where is the received SNR. On the contrary, for a coherent Rayleigh fading
channel given by y = hx + z, the error probability goes down as O( 1 ). A natural question
is, can we somehow convert the fading channel to an AWGN model. This seems viable
if the transmitter also knows the fading coefficient, known as a CSIT model. From now
on, assume a scalar, coherent, fast-fading Rayleigh channel where the transmitter somehow
has also obtained the fading coefficients accurately in advance before each transmission.
Assume also that the fading coefficients are normalized to have unit average gain.
(a) Using the known fading coefficient, the transmitter can try to invert (or equalize) the
effect of fading. Suppose we perform this by simply scaling the transmitted symbols by the
inverse of the fading coefficient, i.e. the new received symbols are
y =

x
h + z = x + z.
h

This is known as channel inversion. For the above model, we expect the same error
probability behaviour as an AWGN. If this surprises you, then you have some thinking to
do for the remaining part of the question.

Find the flaw in the argument which guarantees an exponentially decaying error probability by channel inversion.
Hint: Watch your transmissions, assuming that you are allowed to spend an average
transmit-power of P
(b) Consider the model in the previous part and assume that the fading distribution is that
of max h1 , h2 , where hi , i = 1, 2 is normalized complex Gaussian. Can we now achieve an
exponentially decaying error probability with SNR by channel inversion.
Question 6) Consider a 3 3 MIMO flat fading system, The communication model for
every channel use is y = Hx + z, with the fixed fading matrix given by
1
0

1 0
61
H=
4 2

5
6+1

0 4 2

6+1

4 2

61

4 2

Let the noise process be distributed as Nc (0, Q), where Q is the noise covariance matrix.
Consider an average transmit power constraint of
Ex2 100.
Find the capacity when

1 0 0

Q = 0 1 41
1

0
4 1

You might also like