Indian Institute of Technology Kharagpur

Mid-semester Exam
Subject: Communication - II (EC31204), Digital Comm. (EC31002) [Backlog]
Date: 20 February 2023, Session: AN, Time: 14:00 PM–16:00PM (2 hours)
Department of Electronics and Electrical Communication Engineering
Instructor: Amitalok J. Budkuley
Total marks: 30 marks (*up to 3 bonus marks)

Instructions: There are 5 questions. Answer to a single question should be contiguous and not
scattered across the answer script. Bonus marks (are optional and indicated by *) for a specific
question will be available only if the answer(s) to the rest of the said question is fully correct.

[1] A “two-look” channel: One of two equally likely signals, s0 = −1 or s1 = +1 is sent over a noisy
channel (see figure below). The two noise variables N1 , N2 are independent (of each other and signal) with
1
identical density function fN (n) = e−|n| , n ∈ R. Analyse the maximum aposteriori (MAP) detector
2
and determine (and draw) the optimum decision/decoding regions for each signal.
Hint: There is a “minimum distance decoder”-like interpretation here but unlike the Gaussian case done
in class, the “distance” notion is different. Draw carefully. 3+3=6 Marks

[2] “Hyping” the Cubes: Consider a centered two-dimensional cube with vertices at s0 = (−a, −a), s1 =
(a, −a), s2 = (−a, a), and s4 = (a, a) (draw this and see why it is ‘centered’). We transmit M = 22 = 4
equally likely messages with this constellation over an additive Gaussian noise channel where each of the
two noise components, say N1 , N2 are independent and identically distributed (i.i.d.) with distribution
N1 , N2 ∼ N (0, σ 2 ), where σ 2 is finite.

(a) Determine the average probability of error for the 2-D constellation.

(b) Extend the probability of error calculation in part (a) to a centered 3-dimensional hypercube with
vertices similarly placed (draw it).

(c) Generalize the calculation to an centered k-dimensional constellation, where k is any finite positive
integer.

(d) (Bonus) What happens to the probability of error if one of the messages is removed from the constel-
lation. Argue clearly.

2+2+2+1*=6+1* Marks

1
 2
[3] Thinking Gaussians: Let X ∼ N (0, σX ) and Z ∼ N (0, σZ2 ) be two independent Gaussian random
variables. Let Y = X + Z be a function of the two Gaussian random variables. Answer the following
questions (note: if any distribution is Gaussian in nature, you may express it in the form N (µ, σ 2 ), where
µ, σ 2 are mean and variance.)

(a) What is the distribution of Y ?

(b) What is the conditional distribution of Y when X = x?

(c) What is the conditional distribution of X when Y = y?

(d) (Bonus) For this part (note the variables carefully), assume that Y1 = X + Z1 and Y2 = Y1 + Z2 , where
Z1 , Z2 are i.i.d. with distribution identical to Z. Are the pair (Y1 , Y2 ) jointly Gaussian? Argue clearly
with analysis (and not just words).
Hint: In the standard approach, it will help to recall the linear algebraic definition of a jointly Gaussian
pair of random variables.

1+2+3+2*=6+2* Marks

[4] The “centrality” of E[·]: Consider a non-negative random variable X with finite E[X] = µ. Show that

E[X]
P (X : X ≥ a) ≤ .
a
What is the probability that X takes a value at least X = k · E[X], where k is a positive finite integer?
4+2=6 marks

[5] Consider a random source which outputs i.i.d symbols from the set X = {α1 , α2 , · · · , αm }, where the
m
X
probability of symbol αi is pi ∈ [0, 1]. Note that pi = 1.
i=1

(a) Devise a (i) prefix-free code (ii) uniquely decodable code (which is not prefix-free) for the source.
m
X
(b) Let E[L] := pi li . Inspired by Kraft’s inequality on lengths, we seek to solve the following constrained
i=1
optimization problem using the basic Lagrange multiplier approach:

min E[L]
(l ,l ,···l ):
P1m 2 −lm
i ≤1
i=1 2

where the minimization is over all lengths l1 , l2 , · · · , lm satisfying Kraft’s inequality. Solve it to de-
termine the optimum E[L] and an optimum collection of lengths (l1∗ , l2∗ , · · · , lm ∗
)? (Relax the fact that
lengths need to be integers to solve the optimization problem). Comment meaningfully (1 out of 4
marks reserved) on the purpose of the problem, what it formalizes and the result (esp. optimal lengths).
Hint: : for the given objective function f (l1 , l2 , · · · , lm ), construct the the Lagrangian augmented func-
tion F (l1 , l2 , · · · , lm , λ) = f (l1 , l2 , · · · , lm ) + λ(--constraint--) where λ is the Lagrangian multiplier.
Then follow standard calculus.

2+4=6 marks

Wish you the best of luck!