Source coding

1. Suppose we use 𝑀 level quantizer to encode a length 𝐿 of uniformly

distributed random variable. Write the equation of entropy for this case.

2. Consider a discrete random symbol 𝑋 with 𝑀 + 1 symbols for which 𝑝1 ≥ 𝑝2 ≥

⋯ ≥ 𝑝𝑀 > 0 and 𝑝𝑀+1 = 0. Suppose that a prefix-free code is generated for 𝑋
and that for some reason, this code contains a codeword for 𝑀 + 1. Find 𝐿̅ for
the Huffman code including symbol 𝑀 + 1 in terms of 𝐿̅ for the Huffman code
omitting a codeword for symbol 𝑀 + 1.

3. A random symbol 𝑋 is drawn from symbol alphabet {1,2, … , 𝑀} with a pmf

{𝑝1 , 𝑝2 , … , 𝑝𝑀 }. Let 𝑍 be a random symbol that is 1 if 𝑋 = 1 and 0 otherwise.
Using the fact that 𝐻[𝑋𝑍] = 𝐻[𝑋] + 𝐻[𝑍|𝑋] = 𝐻[𝑍] + 𝐻[𝑋|𝑍], prove that
𝐻[𝑋] ≤ 𝐻𝑏 [𝑝1 ] + (1 − 𝑝1 ) log(𝑀 − 1), where 𝐻𝑏 [𝑝1 ] = −𝑝1 log 𝑝1 −
(1 − 𝑝1 ) log(1 − 𝑝1 ) .

4. Assume a discrete memoryless source and let 𝑁(𝑗, 𝑥 𝑛 ) be the number of

symbols in an 𝑛 −string 𝑥 𝑛 taking on the value 𝑗.

The set 𝑆𝜀𝑛 is defined as follows:

𝑁(𝑗,𝑥 𝑛 )
1. 𝑆𝜀𝑛 = {𝑥 𝑛 : 𝑝𝑗 (1 − 𝜀) < < 𝑝𝑗 (1 − 𝜀); 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑗 ∈
𝑛
𝑋}.
Show that every 𝑥 𝑛 in 𝑆𝜀𝑛 has the following property: 𝐻[𝑋](1 − 𝜀) <
− log 𝑝𝑋𝑛 (𝑥 𝑛 )
< 𝐻[𝑋](1 + 𝜀).
𝑛

5. The figure below presents the state diagram of a second-order Markov

source, where the transitions between the states is labelled in a ( ) and the
probabilities are written in bold font.

0.6
(0) 0.4
00 10
(1)
0.5
(0) (1) 0.5 (1)
0.5 (0) 0.5

01 11 (1)
(0) 0.4
0.6
a) Is the above process generated by the source ergodic? Justify it by first
defining the ergodic process clearly. (5 marks)
b) Find the steady state probabilities of the state. (10 marks)
c) Find the entropy of the above source. (15 marks)
6. Consider a DMS with a two symbol alphabet {𝑎, 𝑏}, where 𝑝𝑋 (𝑎) = 1/5 and
𝑝𝑋 (𝑏) = 4/5. Let 𝑋 𝑛 = 𝑋1 , … 𝑋𝑛 be a string of random symbols from the source
with 𝑛 = 1000.
a) Let 𝑊(𝑋𝑗 ) be the log pmf rv for the 𝑗𝑡ℎ source output. Find variance of
𝑊(𝑋𝑗 ). (3 marks)
b) For 𝜀 = 0.01, evaluate the bound on the probability of the typical set using
the Chebyshev inequality. (3 marks)
Quantization

1. The dynamic input range of a quantizer is 𝐿 = 𝐿1 + 𝐿2 , where the region with

length 𝐿1 has a uniform probability density function of 𝑔1 and the region with
length 𝐿2 has a uniform probability density function of 𝑔2 .
a) Find the entropy of the system if the step size in region 𝐿1 is chosen as ∆1
and the step size in region 𝐿2 is chosen as ∆2 . (5 marks)
b) Find the minimum squared error constrained to the entropy if 𝑀
representation points are to be used. (15 marks)
[Mention the optimization equation clearly. There will be no marks without the correct
optimization equation]

2. Consider a binary scaler quantizer that partitions the set of reals ℝ into two
subsets (−∞, 𝑏] and (𝑏, ∞), and represents (−∞, 𝑏] by 𝑎1 ∈ ℝ and (𝑏, ∞)
by 𝑎2 ∈ ℝ. Assume that the input has the pdf as shown below:

-1 0 1

Note that the pdf is built using two isosceles triangle and a rectangular function.
For 𝑏 > 0 and 𝜀 → 0, find the stable triple [𝑏, 𝑎1 , 𝑎2 ].
 Waveforms

1. If 𝑥(𝑡) and 𝑦(𝑡) are real signals, prove the following inequality
∞ ∞ ∞
|∫ 𝑥(𝑡)𝑦(𝑡)𝑑𝑡| ≤ √∫ 𝑥 2 (𝑡)𝑑𝑡 √∫ 𝑦 2 (𝑡)𝑑𝑡
−∞ −∞ −∞
State all the necessary facts in proving this.

2. Answer very precisely (maximum 2 sentences per answer, 10 marks)

a) State Riesz Fischer theorem. (2 marks)
b) Describe the two conditions under which 𝑥(𝑡) can be represented by its
samples 𝑥(𝑘𝑇𝑠 ) faithfully, where 𝑇𝑠 is some constant and 𝑘 denotes integer
values.(2 marks)
c) Give an example of an uncountable set which is of measure 0. (2 marks)
d) While an 𝐿1 signal guarantees a continuous Fourier transform, there is a
preference for 𝐿2 signals. Why is this the case? (2 marks)
e) Provide an example of a signal for which Riemann integration is
undefined, but Lebesgue integration is defined. (2 marks)

3. Let
2𝐸
√ cos(2𝜋𝑓0 𝑡) 𝑖𝑓 0 ≤ 𝑡 ≤ 𝑇;
𝑠0 (𝑡) = { 𝑇
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒,
and
2𝐸
𝑠1 (𝑡) = {√ 𝑇 cos(2𝜋𝑓1 𝑡) 𝑖𝑓 0 ≤ 𝑡 ≤ 𝑇;

0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒.
(a) Compute the energy of the signals 𝑠0 (𝑡) and 𝑠1 (𝑡), assuming 𝑓0 𝑇 ≫ 1 and
𝑓1 𝑇 ≫ 1. [6 marks]
(b) Find conditions on the frequencies 𝑓0 , 𝑓1 and the duration 𝑇 to ensure both
that the signals 𝑠0 (𝑡) and 𝑠1 (𝑡) are orthogonal and that 𝑠0 (0) = 𝑠0 (𝑇) =
𝑠1 (0) = 𝑠1 (𝑇). [6 marks]

4. Consider the signals 𝑆1 (𝑡) and 𝑆2 (𝑡) as shown below:

If we start the Gram-Schmidt orthogonalization procedure with signal 𝑆1 (𝑡), find the
orthonormal basis functions
 5. Given two waveforms 𝑢1 , 𝑢2 ∈ 𝕃2 . Let 𝑉 be the set of all waveforms 𝑣 that
are equidistant from 𝑢1 and 𝑢2 . Thus
𝑉 = {𝑣: ‖𝑣 − 𝑢1 ‖ = ‖𝑣 − 𝑢2 ‖}.
Show that in general 𝑉 is not the subspace of 𝕃2 .

6. Let us assume that we have two signals as following:

𝑠1 (𝑡) = 2𝜙1 (𝑡) + 𝜙2 (𝑡)
𝑠2 (𝑡) = 5𝜙1 (𝑡) + 𝜙2 (𝑡)
where 𝜙1 (𝑡) and 𝜙1 (𝑡) are the orthonormal function.
a. Draw the constellation diagram. (1 mark)
b. Find the distance between the two signals. (2.5 mark)
c. Find the angle between the two signals. (2.5 mark)

Each question is of 10 marks.

7. The function 𝑝(𝑡) satisfies the following properties:

𝑝(𝑡) = 𝑝2 (𝑡) for all 𝑡
𝑝(𝑡) = 0 for all |𝑡| > 1
𝑝(𝑡) = 𝑝(−𝑡) for all 𝑡 (symmetry)
𝑝(𝑡) = 1 − 𝑝(𝑡 − 1) for 0 ≤ 𝑡 < 1/2
Show that 𝑝(𝑡) is orthogonal to 𝑝(𝑡)𝑒 𝑖2𝜋𝑚𝑡 for integer 𝑚 ≠ 0.

8. Find the equivalent length of the signal ℎ(𝑡) = 𝑒 −2|𝑡| , such that it is processed
by the equipment which has a norm threshold of 𝜏.
 Modulation

1. Explain one extra constraint in carrier less amplitude and phase modulation
scheme compared to QAM.

2. Name one modulation scheme which is extremely energy efficient and one which
is extremely spectrally efficient.

3. Explicitly mention the utility of prolate spheroidal waveforms in the context of

defining degrees of freedom.

4. A communication system has to be established at 100 Mb/s over a passband

channel. Assuming that an excess bandwidth of 50% is used, how much channel
bandwidth is needed for the following modulation schemes: QPSK, 64-QAM, and
64-ary noncoherent orthogonal modulation scheme using a Walsh-Hadamard
code?

5. Consider binary DPSK with encoding as described in the following figure.

Coder Slicer
10 00 10 00

Adder (XOR) Adder (XOR)

…

−1
11 11 −1
01 01

−1 −1

Suppose we fix 𝑏[0] = −1, and that the information bits {𝑖[𝑛]} to be sent is
0110001101.
a) Find the transmitted symbol sequence corresponding to the given bit sequence.
b) Assuming that we use a rectangular time limited pulse, draw the corresponding
complex baseband transmitted waveform. Is the Q component being used?
c) Now, suppose that the channel imposes a phase shift of −𝜋/6. Draw the I and
Q components of the noiseless received complex baseband signal.
d) Suppose that the complex baseband signal is sent through a matched filter to
the rectangular timelimited pulse, and is sampled at the peaks. What are the
received samples {𝑟[𝑛]} that are obtained corresponding to the transmitted
symbol sequence {𝑏[𝑛]}?
e) Find 𝑟[2]𝑟 ∗ [1]. How do you figure out the information bit 𝑖[2] based on this
complex number?

6. Suppose there is a problem of phase offset in the designed communication system.

Let us assume that the phase offset is 𝜃, and the received baseband I and Q
component obtained is 𝑠𝑐 and 𝑠𝑠 . Find out the transmitted I and Q component.

7. An 8-QAM constellation is shown below:

 a. What is energy per bit for 𝑥 = 1 and = 1 ?
b. Find the optimum value of 𝑥 for 𝑥 > 1 at high SNR.
c. Find the best case power efficiency (𝑑 2 /𝐸𝑏 ).

8. A new modulation scheme is derived from the 4-ary orthogonal modulation scheme
by subtracting the mean of the orthogonal modulation scheme. The orthogonal
modulation scheme has an average energy per symbol of 1 Joule.
a) Find the average energy per symbol in the new modulation scheme.
b) What is the inner product between any two waveforms in the new modulation
scheme?
c) Find the spectral efficiency of the new modulation scheme assuming coherent
detection.

9. Let 𝑣(𝑡) be a continuous 𝐿2 waveform with 𝑣(0) = 1 and define 𝑔(𝑡) =

𝑣(𝑡)𝑠𝑖𝑛𝑐(𝑡/𝑇), where 𝑠𝑖𝑛𝑐(𝑥) = 𝑠𝑖𝑛(𝜋𝑥)/𝑥. From 𝐺(𝑓), the Fourier transform of
𝑔(𝑡), directly conclude that 𝑔(𝑡) is an ideal Nyquist pulse.

10. Let {𝑢𝑘 } be a complex data sequence and let 𝑢(𝑡) = ∑𝑘 𝑢𝑘 𝑝(𝑡 − 𝑘𝑇) be the
3 5
corresponding modulated output. Let 𝑃(𝑓) be equal to √𝑇 over 𝑓 ∈ [2𝑇 , 2𝑇] and be
equal to 0 elsewhere. At the receiver, 𝑢(𝑡) is filtered using 𝑝(𝑡) and the output 𝑦(𝑡)
is then 𝑇-spaced sampled at time instants 𝑘𝑇. Show that 𝑦(𝑘𝑇) = 𝑢𝑘 for all 𝑘 ∈ 𝑍.
Also, indicate for what values of 𝑓𝑐 does this scheme work?

11. The baseband signal of a modulation scheme is 𝑢(𝑡) = ∑𝑛 𝑏𝑟 [𝑛]𝑝(𝑡 − 𝑛𝑇) +

𝑇
𝑗𝑏𝑖 [𝑛]𝑝 (𝑡 − 𝑛𝑇 − ), where {𝑏𝑟 [𝑛], 𝑏𝑖 [𝑛]} are ±1.
2
4
a) Assuming 𝑓𝑐 = 𝑇 and the transmitted symbols are [00,11,01,10], draw the typical
passband waveform for a duration of 4𝑇. You can assume any reasonable 𝑝(𝑡).
This waveform is known as offset QPSK.
b) Is there any advantage of using the given baseband waveform compared to
conventional QAM?
𝜋𝑡
c) If 𝑝(𝑡) = sin ( 𝑇 ) 𝐼[0,𝑇] (𝑡), show that this waveform is the same as MSK.

12. Let 𝑠(𝑡) = ∑𝑛 𝑏[𝑛]𝑝(𝑡 − 𝑘𝑇) be a binary PAM waveform, where sequence 𝑏[𝑛] is
stationary with zero mean.
a) Show that 𝑠(𝑡) is a cyclostationary random process, that is, 𝑠(𝑡) and 𝑠(𝑡 − 𝑑)
are statistically indistinguishable, only if 𝑑 = 𝑘𝑇, where 𝑘 is an integer.
 b) Let us define 𝑅𝑠 (𝑡, 𝑡 − 𝜏) = 𝐸[𝑠(𝑡)𝑠 ∗ (𝑡 − 𝜏)]. Show that 𝑅𝑠 (𝑡, 𝑡 − 𝜏) depends on
both 𝑡 and 𝜏.
c) Now let us define 𝑅𝑠 (𝑡, 𝑡 − 𝜏) = 𝐸[𝑠(𝑡 − 𝐷)𝑠 ∗ (𝑡 − 𝜏 − 𝐷)], where 𝐷 is a uniformly
distributed random variable between [0, 𝑇]. Show that
𝑃 ∞
𝑅𝑠 (𝑡, 𝑡 − 𝜏) = 𝑇𝑑 ∫−∞ 𝑝(𝑢)𝑝(𝑢 − 𝜏)𝑑𝑢,
where 𝑃𝑑 = 𝐸[|𝑏[𝑛]|2 ]. Note that the function 𝑅𝑠 (𝑡, 𝑡 − 𝜏) now solely depends
on 𝜏, making it a valuable definition of the autocorrelation function. This specific
form is particularly useful for analyzing cyclostationary random processes and
is referred to as the stationarized autocorrelation function.
d) Show that
𝑃
𝐹[𝑅𝑠 (𝑡, 𝑡 − 𝜏)] = 𝑇𝑑 |𝑃(𝑓)|2,
where 𝐹[𝑥] is the Fourier transform of 𝑥, and 𝑃(𝑓) is the Fourier transform of
𝑝(𝑡).
e) Using the power spectral density (PSD) formula derived above, draw the PSD
of binary PAM assuming a rectangular pulse shape, and a sinc pulse shape.

13. Suppose a noncoherent vector binary modulation scheme is used, where 0 is

represented by {1,1,0,0} and 1 is represented by {0,0,1,1}, in the presence of
AWGN channel with the noise vector as {0.2, −0.8,1.2, −10}. After stating the
optimal detector rule, find out whether an error will occur in the transmission of 0
and 1. Also find out the bandwidth required if a sinc pulse with a first null at 𝑇 is
used to encode the amplitudes. [10 marks]

14. Determine the spectral efficiency for the following (baseband) modulation scheme
(where crosses represent a signal point).

15. Let us assume that we have a bandlimited AWGN channel with 100 MHz
bandwidth. A communication link needs to be set over this channel with a data rate
of 18 Mb/s. Name the modulation scheme that you will chose to ensure the largest
link length for a given transmitted power. Motivate your choice. [8 marks]

16. Assume that we use standard 64-QAM and 64-FSK coherent modulation schemes
with the minimum distance between any two symbols as 𝑑, and in the presence of
AWGN channel with the noise variance as 𝑁0 /2 in one dimension. For both these
modulation schemes, a root-raised cosine pulse is used with a roll-off factor of 0.2.
Derive for both these modulation schemes, the following parameters: [24 marks]
a) Energy per symbol in 2D and energy per bit [4 marks]
b) Spectral efficiency. [3 marks]
c) SNR. [2 marks]
d) Approximate probability of symbol error. [6 marks]
e) Approximate probability of bit error (assuming Gray coding where applicable).
[5 marks]
f) Find the approximate distance between the 𝐸𝑏 /𝑁0 required in dB for the same
low symbol error rate in these modulation schemes. [4 marks]
 Random process

1. A process 𝑿(𝑡) = ∑𝑛𝑘=0 𝑩𝑘 𝑝(𝑡 − 𝑘), where 𝑛 = 4 and where 𝑩𝑘 is iid Bernoulli
1, 0 ≤ 𝑡 < 1
random variable and 𝑝(𝑡) = { . Then, 𝑿(𝑡) has how many
0, otherwise
distinguishable sample functions?

2. If 𝑁 = 𝑁𝑟 + 𝑗𝑁𝐼 and 𝑌 = |𝑁| where 𝑁𝑟 and 𝑁𝐼 are independent normal random

variables. Find the probability density function of 𝑓𝑌 (1).

3. If 𝑉1 and 𝑉2 are two random variables related to the random variables 𝐻, 𝑁1 , and
𝑁2 by the following equations,
𝑉1 = 𝐻 + 𝑁1
𝑉2 = 𝐻 + 𝑁1 + 𝑁2
If 𝑁1 and 𝑁2 are the normal random variables then find 𝑓𝑉2 |𝑉1 ,𝐻 (𝑉2 = 0.5|𝑉1 =
0. , 𝐻 = 1).

𝑍1 1 2 𝑊1
4. If ( )=( ) ( ) where 𝑊1 and 𝑊2 are independent normal random
𝑍2 2 1 𝑊2
𝑍
variables, and 𝑍 = ( 1 ).
𝑍2
a) Find the covariance matrix 𝐾𝑍 = 𝐸[𝑍𝑍 𝑇 ].
b) If = [0. 0]𝑇 , then find 𝑇 𝑘𝑧−1 .
c) Find 𝑓𝑍 ( ).

5. A Gaussian random vector Z (with elements Z1 and Z2) is obtained by linear

transformation of Gaussian random vector W, as is given below
𝑍 2 𝑊1
[ 1] = [ ][ ]
𝑍2 − 2 𝑊2
where the elements of W are iid normal random variables.
a. Find the inverse covariance matrix of random vector Z.
b. Find the probability density function of random vector Z.
c. Find the mapping of the circle with radius √1 in W space to the 𝑍 space.

6. The random variables 𝑋1 and 𝑋2 are jointly Gaussian, with 𝐸[𝑋1 ] = 1, 𝐸[𝑋2 ] = −2,
𝑐𝑜𝑣[𝑋1 ,𝑋2 ]
𝑣𝑎𝑟[𝑋1 ] = 4, 𝑣𝑎𝑟[𝑋2 ] = 1, and correlation coefficient 𝜌(𝑋1 , 𝑋2 ) = =
√𝑣𝑎𝑟[𝑋1 ]𝑣𝑎𝑟[𝑋2 ]
−1.
a) Write down the mean vector and covariance matrix for the random vector 𝑋 =
[𝑋1 , 𝑋2 ]𝑇 .
b) If vector 𝑌 = 2𝑋1 − 𝑋2, then find the pdf of 𝑌.
c) Suppose that 𝑍 = 𝑋1 − 𝑎𝑋2. Find 𝑎 such that 𝑍 is independent of 𝑋1.

7. A white Gaussian noise 𝑁(𝑡) with one-sided power spectral density of 𝑁0 is fed to
𝑡
two correlators, one computing the inner product of the noise with 𝑠𝑖𝑛𝑐 (𝑇) and the
2𝑡
other computing the inner product of the noise with 𝑠𝑖𝑛𝑐 ( 𝑇 ) , and a matched filter
 𝑡
with a filter with an impulse response of 𝑠𝑖𝑛𝑐 (𝑇 − 𝑇) and with a sampler at T. In
this process, three random variables V1, V2 and V3 are obtained as shown in the
figure below.

T
Filter,
r

Find 𝐸[𝑉1 𝑉2 ], 𝐸[𝑉12 ], 𝐸[𝑉22 ] and 𝐸[𝑉32 ].

8. A random process 𝑍(𝑡) is given as:

𝑍(𝑡) = 𝑋(𝑡) + 𝑋(𝑡 + 𝜆)
If the auto-correlation function 𝑅𝑋 (𝜏) of the random process 𝑋(𝑡) is
𝑅𝑋 (𝜏) = exp(−𝜏 2 ), find the auto-correlation function 𝑅𝑍 (𝜏) of the random process 𝑍(𝑡).

9. Let a WSS random process 𝑿(𝑡), whose autocorrelation function is 𝑅𝑿 (𝜏) = 𝑒 −|𝜏|,
sin(2𝑡)
is passed through an LTI filter with impulse response ℎ(𝑡) = 𝜋𝑡 to produce 𝒀(𝑡)
process at the output.

sin(2𝑡)
𝑿(𝑡) 𝒀(𝑡) = ?
𝜋𝑡

Find 𝐸[𝒀2 (𝑡)].

10. Let random process 𝑿(𝑡) = 𝑨 cos(𝜔𝑡), where 𝑨 is a random variable with pdf
𝑓𝑨 (𝑎) = 𝜆𝑒 −𝜆𝑎 , 𝑎 ≥ 0, 𝜆 > 0, is passed through an LTI filter whose impulse
response is ℎ(𝑡) = 2sinc(𝑡) giving 𝒀(𝑡) process at the output. Diagrammatically:

𝑿(𝑡) = 𝑨 cos(𝜔𝑡) 2sinc(𝑡) 𝒀(𝑡) = ?

Then what is the mean of the output random process (𝑡) ?

11. Let 𝐍(𝑡), a white Gaussian noise process having double-sided power spectral
density of 10−6 Watts/Hz is passed through an LTI filter having transfer function
ℎ(𝑡) = rect(𝑡). Find the auto-correlation function of the signal at the output.

12. Let 𝑋 and 𝑌 be independent zero-mean Gaussian rvs with variance 𝜎𝑋2 and 𝜎𝑌2
respectively. For the random variable 𝑋, find the Fourier transform of its pdf. Also
find the density of 𝑋 + 𝑌, using Fourier transform of the densities.

13. A WSS process 𝑋(𝑡) is input to the LTI system with impulse response ℎ(𝑡) =
𝑠𝑖𝑛𝜃
𝑠𝑖𝑛𝑐(𝑡)𝑢(𝑡), where 𝑠𝑖𝑛𝑐(𝜃) = . The process 𝑋(𝑡) has mean 1 and
𝜃
autocorrelation function 𝑟(𝑡) = 𝑒 −|𝑡| .
a. What is the mean of the output process?
b. Find the autocorrelation function of the input and output process 𝑅𝑋𝑌 at
time 𝑡 = 1.
c. Find the autocorrelation function of the output process 𝑅𝑌 at time 𝑡 = 1.

14. Let 𝑍1 , 𝑍2 , … be a sequence of independent Gaussian rvs, 𝑍𝑘 ~𝑁(0, 𝜎𝑘2 ), and let
{𝜙𝑘 (𝑡): ℝ → ℝ} be a sequence of orthonormal functions.
(a) Argue from fundamental definitions that, for each 𝑡, 𝑍(𝑡) = ∑𝑛𝑘=1 𝑍𝑘 𝜙𝑘 (𝑡) is a
Gaussian rv. Find the variance of 𝑍(𝑡) as a function of 𝑡. [8 marks]
(b) For any set of epochs 𝑡1 , … , 𝑡𝑙 , let 𝑍(𝑡𝑚 ) = ∑𝑛𝑘=1 𝑍𝑘 𝜙𝑘 (𝑡𝑚 ) for 1 ≤ 𝑚 ≤ 𝑙. Explain
carefully from the basic definitions why {𝑍(𝑡1 ), … , 𝑍(𝑡𝑙 )} are jointly Gaussian and
specify their covariance matrix. Explain why {𝑍(𝑡); 𝑡 ∈ ℝ} is a Gaussian random
process. [10 marks]
 Detection

Problems

1. An OOK stream is sent, where 𝐴 is used to transmit 1 and 0 is used to transmit 0.

The noise is additive white Gaussian with a variance of 𝜎 2 . The received symbol is
𝑟.
a) Determine the log-likelihood ratio as a function of 𝑟.
b) Assuming apriori probability of 0 as 𝜂/(1 + 𝜂), what is the threshold of detection
in MAP detector?
c) Find the threshold of detection in ML detector approximately.

2. Two 8-PSK constellations are shown below:

A B
a) Find the power efficiency of constellation A and B.
b) What is the distance between the constellation A and B in terms of 𝐸𝑏 /𝑁0
requirements at high SNR in dB?

3. An OOK system is as shown in the figure below, where the received signal for bit
1 and bit 0 is assumed as following

𝑠(𝑡) + 𝑛(𝑡), bit 1 sent

𝑦(𝑡) = {
𝑛(𝑡), bit 0 sent

where 𝑛(𝑡) is an Additive White Gaussian Noise with one-sided power spectral
density of 𝑁0 . The received signal is passed through a filter and a sampler at 𝑡0 .

)
1

0 1 2

-1
a) Find the noise energy per degree of freedom and the BER at the output of the
sampler if 𝑡0 = 0 and 𝑡0 = 1.
 b) the impulse response of the filter is chosen as 𝑠(−𝑡 + 2), what is the optimum
sampling time instant 𝑡0 ?

4. Let us assume in the above question, ℎ(𝑡) = 𝐼[−1,0] (𝑡).

a) Determine part (a) again.
b) Find the BER at the output of the sampler if a linear combination of samples at
𝑡0 = 0 and 𝑡0 = 1 is used.
1
c) Assuming now that the noise is Laplacian noise with a pdf of 𝑓𝑁 (𝑥) = 2 𝑒 −|𝑥| ,
−∞ < 𝑥 < ∞, and the filter is a matched filter, determine the log-likelihood ratio
𝑝(𝑦⁄𝑥=0)
(where log-likelihood ratio can be computed as 𝑟 = ln (𝑝(𝑦⁄𝑥=1))) for 0 < 𝑦 < 2.
d) Assuming that the prior of symbol ‘1’ is 𝑃𝐻 (1) and that of symbol ‘0’ is 𝑃𝐻 (0),
and these are related as
∞
𝑃𝐻 (0)
𝑃𝐻 (1) = ∑
𝑛!
𝑛=0

where 𝐻 denotes the hypothesis, determine the threshold in terms of y.

5. The 𝐸𝑏 /𝑁𝑜 requirement of 2-PAM in AWGN channel to achieve an error rate of

10−8 is 12 dB. Determine the 𝐸𝑏 /𝑁𝑜 requirement of 16-PPM (baseband) to achieve
the similar error (symbol) rate in an AWGN channel.

6. The receiver in a binary communication system employs a decision statistic Z,

which behaves as follows:
𝑍 = 𝑁 if 0 is sent
𝑍 = 4 + 𝑁 if 1 is sent
where 𝑁 is modeled as Laplacian with density
1
𝑓𝑁 (𝑥) = 2 𝑒 −|𝑥| , −∞ < 𝑥 < ∞

𝑓 (𝑧/1)
a) Determine the log-likelihood ratio 𝐿𝐿𝑅( ) = log e 𝑓𝑍|1 (𝑧/0), as a function of z.
𝑍|0
b) Assuming apriori probability of 0 as 0.12, what is the threshold of detection in
MAP detector approximately?
c) Find the threshold of detection in ML detector approximately.

7. In 4-QAM system, a signal is received with an amplitude of (0.25 and 0.5) in the
complex dimension. All signals are equally probable, and the sum of likelihoods is
given as below.
𝑀

∑ fY (𝑦/𝑖) = 1.46
𝑗=1
where 𝑌 is the received random variable and 𝑖 is the Hypothesis that was transmitted.
 (s2,11) 1 (s1,00)

(0.25,0.5)

-1 1

(s3,10) -1 (s4,01)

a) Find the posterior probability of signal 𝑆1 and 𝑆4 .

b) Determine the log-likelihood ratio calculated for the soft decisions for the first
𝑃[𝑏(1)=0]
bit, i.e., log 𝑒 𝑃[𝑏(1)=1].

8. The receiver in an optical on-off keyed (OOK) system decides based on photon
counts. Let us assume that photon count is Z, where Z is a Poisson distributed
random variable with mean 100 if 1 is sent and mean 10 if 0 is sent. Also assume
that bit 1 is 100 times more probable than bit 0.
𝑝 (𝑧/1)
a) Find the log-likelihood ratio 𝐿𝐿𝑅( ) = log10 𝑍|1 , in terms of received number
𝑝𝑍|0 (𝑧/0)
of photons .
Hint: Note that the Poisson distributed random variable Z with mean m has the
probability mass function to receive n photons as:
𝑒 −𝑚 𝑚𝑛
𝑝𝑍 (𝑍 = 𝑛) =
𝑛!
b) Determine the threshold in MAP and in ML detection.

9. An octal QAM signal constellation has two types of symbols as shown below.
Symbols of the same type have equal probabilities but the square symbols are half
as equiprobable as the cross symbols.

a) Find the average symbol energy of this constellation.

b) Determine the average symbol error rate (obtained using intelligent union
bound) under the presence of AWGN with a variance of 1.
c) Find the average BER.
10. Consider a biorthogonal signalling scheme where 2 symbols are put on each
orthogonal dimension with a mean of 0 and a spectral efficiency of /2.
a) If the symbols are represented as 𝑅 𝑛 , what is 𝑛? (5 marks)
b) Assuming that the noise has uniform positive amplitudes in all dimensions
with a total energy of 0.25, and the symbol has an energy of 1 unit, find the
received vector when the symbol with a positive amplitude in the first
dimension is transmitted. (5 marks)
c) Describe the receiving strategy for the above case with a proper example.
(5 marks)

11. A 3 bits single parity check code is used, where 0 is encoded as 0 volts, and 1 is
encoded as 1 volts. The channel is additive white Laplacian, where noise is
modelled with density
1
𝑓𝑁 (𝑥) = 2 𝑒 −|𝑥| , −∞ < 𝑥 < ∞
The received vector is {0.2,5, −0.5}. Find out the likelihood ratio (probability of
zero/probability of one) of the first bit (intrinsic plus extrinsic). [10 marks]
Wireless Communication

1. What is the use of I-PN and Q-PN sequence used in CDMA IS-95?
2. Write one advantage of Walsh Hadamard waveforms compared to PPM.
3. Draw the schematic of CDMA IS-95, clearly highlighting the bit rates at
various steps.
4. Suggest one case where there can be significant delay spread without any
Doppler spread, and vice versa.
5. Draw the schematic of Rake receiver with some explanation.
6. When bandwidth is infinite, there will not be any fading. Justify.