SCHOOL OF ENGINEERING

SENSOR NETWORKS AND DATA ANALYSIS 2

ELEE08021

Exam Diet: May 2022 Duration: 1.5 hours plus 1 hour upload Exam Date: 23/05/2022

Exam starts: 13:00 Exam ends: 15:30 All times are BST (UTC+1)

Before commencing work, please read the academic, formatting, scanning and uploading guidance.
Examination information

• This exam paper consists of TWO sections

• Candidates should attempt ALL THREE questions, as follows:
• Section A: ONE question. Attempt the whole section.
• Section B: TWO questions. Attempt the whole section.

Specific instructions

• Students should assume reasonable values for any data not given in a question, or not available on a
datasheet, and should make any such assumption clear on their answer sheets.
• Students in any doubt as to the interpretation of the wording of a question, should make their own decision,
and should state it clearly on their answer sheet.
• Write concise, complete answers. If a length limit is given, stay within it. Produce equations and diagrams to
a good hand-drawn standard.
• This is an open book exam. This means you can freely access any printed or online materials to aid you in
your answers. Online materials can include text, images, videos and data. You may NOT engage in
interactions or discussions relating to the exam questions or examined subject matter in any form. Sharing
the answers to this exam in any way, by any means and in any form is STRICTLY NOT allowed.
• Use only a standard calculator. Do not use computer-based spreadsheets, mathematical solvers, simulation
tools, graphing calculators or any other tool which is interactive in nature, such as online mathematical
equation solvers.

Technical instructions (For full details see the formatting guidance and how to upload your exam to learn)
• Write in dark blue or black ink on white or light-coloured A4 paper, or the nearest equivalent size; unlined,
lined, and graph paper are all acceptable, as are pages with holes for binding.
• Write on one side of the paper only. Start every question on a new page. Use portrait orientation.
• On every page clearly write the QUESTION NUMBER (left side) and your EXAMINATION NUMBER (right
side).
• Arrange your pages so that your answers are given in the same order as the question paper.
• Take clear individual pictures of each page and combine in a single PDF document according to the scanning
instructions.
• Name your file with the course code and your examination number, e.g. ENGI00000-B123456.pdf
• Check your file carefully then upload it in the ONLINE EXAM area for this course on LEARN.
• If you require technical support, contact [email protected]

Special Items
• Data Sheets (4pages)

Convenor of Board of Examiners: Professor R Cheung

External Examiner: Professor Yang Hao & Dr T Tjahjadi
SECTION A

Question A1

a) (i) Explain the concept of modulation and why it is needed in a

communication system. (2)

(ii) A channel of bandwidth 4 kHz is available for signal transmission.

Determine the data rate that can be accommodated on this channel
with On-off Non-Return-to-Zero line code. (1)

b) (i) Give one key characteristic of an energy signal, expressed in a

mathematically well-defined manner. (1)

(ii) Give an example of a non-periodic signal that would be classified as

a power signal. (1)

(iii) Calculate the fundamental frequency of the following signal in hertz:

(3)
3
x(t) = 2 + 5 sin 4t + cos 7t

(iv) In a sampled data system for analogue-to-digital and

digital-to-analogue conversion, give TWO reasons why it is
impractical to sample a signal perfectly. (2)

c) (i) Consider the dataset consisting of the 2D points depicted in

Figure A1c overleaf. We can represent this dataset using a matrix
X ∈ R8×2 .

What is the first principal component of X, and why? You do not

need to perform an eigendecomposition to answer this question. (2)

(ii) You are given a training set of data points {xi }10 2
i=1 where xi ∈ R .
Each point has an associated class label yi ∈ {0, 1, 2}. These data
points and labels are depicted in Figure A1d overleaf.
"
(I) Use a 3-NN classifier to classify a test point at −4 −2 . (1)

"
(II) Use a 5-NN classifier to classify a test point at 3 3 . (1)

"
(III) Use a 9-NN classifier to classify a test point at −4 −1 . (1)

CONTINUED OVER

ELEE08021 Sensor Networks and Data Analysis – May 2022

SECTION A

(iii) Consider a training set of 1D data points and labels {xi , yi }N i=1
where yi ∈ {−1, 1}. A linear classifier takes in a point and computes
a prediction zi = wxi + b where w and b are learnable parameters.
To learn these parameters so that the predictions resemble the
labels we can minimise a loss function such as:
N
1
L= e−yi zi
N
i=1

∂L
(I) Derive expressions for ∂w and ∂L
∂b . (2)

(II) You are given a linear classifier with initial parameters

w0 = 0 and b0 = 1 and the following training set (N = 4):

x1 = −1 y1 = −1
x2 = −3 y2 = −1
x3 = 1 y3 = 1
x4 = 2 y4 = 1

Use your gradient expressions to perform a single iteration

of gradient descent to update the classifier’s parameters to
1
w1 and b1 . Use a learning rate of 10 . (3)

2 4
{x } : =0
3
i i yi

{x } :
i i yi =1
1 2 {x } :
i i yi =2

1
dimension 2

dimension 2

0 0
1
1 2
3
2 4
2 1 0 1 2 4 3 2 1 0 1 2 3 4
dimension 1 dimension 1

Figure A1c: A dataset consisting of 8 Figure A1d: A dataset consisting of

2-dimensional points. 10 2-dimensional points.

ELEE08021 Sensor Networks and Data Analysis – May 2022

SECTION B

Question B1

a) Explain the term ‘multiplexing’ in relation to communication systems and

state TWO common types of signal multiplexing. (5)

b) Explain why using double side band with suppressed carrier (DSB-SC) in a
communication system is more power efficient than using double side band
amplitude modulation (DSB-AM). (3)

c) An amplitude modulated (AM) signal, xAM (t), is represented by the

following expression:

xAM (t) = [Ac + m(t)] cos(2πfc t + π).

The carrier frequency, fc = 107 Hz and the power in the modulation signal,
m(t), is represented as Pm .

(i) Derive an expression for the power efficiency, η , of the AM signal,

xAM (t), in terms of Pm . (2)

(ii) Given that the modulation index, µ = 0.5, determine the power
efficiency, η , of the modulating signal m(t) shown in Figure B1c
overleaf. (4)

d) A radio antenna pointed in a direction of the sky has a noise temperature

of Ts = 50 K. The antenna feeds the received signal to a pre-amplifier
operating at an ambient temperature of T0 = 290 K. The pre-amplifier has
a gain of 35 dB over a bandwidth of 10 MHz and a noise figure of 2 dB.

(i) Determine the noise power at the output of the preamplifier in dBm. (4)

(ii) Determine the signal power Pr (in dBm) at the antenna terminals
such that the signal-to-noise ratio is 50 dB at the output of the
pre-amplifier. (2)

You may take the Boltzmann constant as k = 1.38 × 10−23 JK−1 .

CONTINUED OVER

ELEE08021 Sensor Networks and Data Analysis – May 2022

SECTION B

m(t)

Am

T/2 T 2T
T/4 3T/4 t

-Am

Figure B1c: Triangular waveform as modulating signal, m(t).

ELEE08021 Sensor Networks and Data Analysis – May 2022

Question B2

A continuous-time signal, x(t), has a Fourier transform denoted by X (ω), and is

expressed as: 
0, ω < −ωm


X (ω) = ω2 if −ωm ≤ ω < ωm


0, ω ≥ ωm

a) (i) Write down the definition of the Fourier transform, and calculate the
value of the integral: (2)
∞
I= x(t) dt
−∞

(ii) Using the definition of the inverse Fourier transform, calculate the
value of x(0); i.e. x(t) evaluated at t = 0. (2)

In this part of the question, you do not need to find the full expression
for x(t).

(iii) Determine whether energy or power provides the most suitable

measure of the size of the time-domain signal, x(t), justifying your
answer. (1)

(iv) Using the appropriate form of Parseval’s theorem, calculate the

value of the measure chosen to find the size of the signal, x(t). (4)

b) Find an expression for the signal, x(t), written in terms of t and trigonometric
functions of t. Show all your working. (6)

You may use the identity:

x2 2x 2
x2 ecx dx = ecx − 2 + 3 + constant
c c c

c) What is the Nyquist frequency, ωn , for the signal x(t)? (1)

d) The signal x(t) is now sampled at a rate of ωs .

(i) If ωs ≥ ωn , plot the resulting spectrum of the sampled signal,

labelling carefully key frequencies. (2)

(ii) If ωs < ωn , plot the resulting spectrum of the sampled signal,

explaining key characteristics of the resulting graph. (2)

END OF PAPER

ELEE08021 Sensor Networks and Data Analysis – May 2022

 ELEE08021 Sensor Networks and Data Analysis 2 (SNADA)
Formula and Tables of Transforms

Trigonometric Identities

ejx 2 e−jx ejx + e−jx

sin x = cos x =
2j 2

1 = cos2 x + sin2 x cos 2x = cos2 x 2 sin2 x

1 1
cos2 x = (1 + cos 2x) sin2 x = (1 2 cos 2x)
2 2
1 tan A ± tan B
cos3 x = (3 cos x + cos 3x) tan (A ± B) =
4 1 3 tan A tan B
cos(A ± B) = cos A cos B 3 sin A sin B sin(A ± B) = sin A cos B ± cos A sin B
A+B A2B A+B A2B
sin A + sin B = 2 sin cos sin A 2 sin B = 2 cos sin
2 2 2 2
A+B A2B A+B A2B
cos A + cos B = 2 cos cos cos A 2 cos B = 22 sin sin
2 2 2 2
1 1
cos A cos B = [cos(A + B) + cos(A 2 B)] sin A cos B = [sin(A + B) + sin(A 2 B)]
2 2
1
sin A sin B = [cos(A 2 B) 2 cos(A + B)]
2

b
a cos x + b sin x = c cos (x + ») where c= a 2 + b2 and » = 2 tan−1
a

b
a c Cosine Rule: c2 = a2 + b2 2 2ab cos ³
a b c
g a Sine Rule: = =
sin ³ sin ³ sin ³
b

Fourier Series Analysis of Periodic Waveforms

If g(t) is periodic with period T , then:

∞
a0
g(t) = + [an cos (nË0 t) + bn sin (nË0 t)]
2
n=1

T T
2 2 2 2
where an = g(t) cos (nË0 t) dt and bn = g(t) sin (nË0 t) dt
T − T2 T − T2
∞ T
jnω0 t 1 2
or: g(t) = cn e where cn = g(t) e−jnω0 t dt
n=−∞
T − T2

2π 1
where Ë0 = T = 2Ãf0 ; f0 = T is the fundamental frequency.
Fourier Transform Analysis of Aperiodic Signals

The Fourier transform of a signal g(t) is given by:

∞ ∞
1
G(Ë) = g(t) e−jωt dt and g(t) = G(Ë) ejωt dË
−∞ 2Ã −∞

Parseval’s theorem of energy conservation:

∞ ∞
1
|g(t)|2 dt = |G(Ë)|2 dË
−∞ 2Ã −∞

Selected Fourier Transforms

g(t) G(Ë)

1 (DC level) 2÷(Ë)

1
u(t) (unit step) ÷(Ë) +
jË

ejω0 t 2÷(Ë 2 Ë0 )

cos Ë0 t à [·(Ë 2 Ë0 ) + ·(Ë + Ë0 )]

Ã
sin Ë0 t [·(Ë 2 Ë0 ) 2 ·(Ë + Ë0 )]
j
∞ ∞
2Ã 2Ãm
·(t 2 nT ) (impulse train) · Ë2
n=−∞
T m=−∞
T

g(t 2 Ç ) (time shift) e−jωτ G(Ë)

1 Ë
g(at) (scale in time) G
|a| a

ejω0 t g(t) G(Ë 2 Ë0 ) (frequency shift)

g1 (t) æ g2 (t) (convolution) G1 (Ë)G2 (Ë) (multiplication)

1
g1 (t) g2 (t) (multiplication) G1 (Ë) æ G2 (Ë) (convolution)
2Ã

Duality: If g(t) transforms to p(Ë), then p(t) transforms to 2Ãg(2Ë).

Symmetry: If g(t) is real, then G(2Ë) = G∗ (Ë) (7 means complex conjugate).

If g(t) is real and even, then G(Ë) is real and even.
If g(t) is real and odd, then G(Ë) is imaginary and odd.
z-Transforms Communications Theory

The z-transform of a discrete-time causal sequence Amplitude Modulation An amplitude modu-

g[n] (defined for n = 0, 1, 2, . . . ) is given by: lated (AM) signal can be expressed as:

∞
xc (t) = Ac [1 + a mn (t)] cos (2Ãfc t)
G(z) = g[n] z −n
n=0

Angle Modulation The general angle-modulated

Selected z-Transforms signal is given by:
xc (t) = Ac cos (2Ãfc t + Ç(t))
g[n], (n g 0) G(z) For PM: Ç(t) = kp m(t)
t
·[n] (unit pulse) 1 For FM: Ç(t) = kp m(³) d³
t
·[n 2 m] z −m = 2Ãfd m(³) d³

z
1 (unit step)
z21 Angle Modulation
z
n (unit ramp) Ç(t) = ³ sin (2Ãfm t)
(z 2 1)2
z
rn the angle-modulated signal is:
z2r
∞
rz
n rn xc (t) = Ac Jn (³) cos [2Ã (fc + nfm ) t]
(z 2 r)2
n=−∞
z sin Ë0
sin(Ë0 n)
ù
z2 2 2z cos Ë0 + 1 úJ (³) , for n even
n
where J−n (³) =
z22 z cos Ë0 û2J (³)
n for n odd
cos(Ë0 n)
z2 2 2z cos Ë0 + 1
z r sin Ë0 The bandwidth is given by Carson’s Rule:
rn sin Ë0 n
z 2 2z r cos Ë0 + r2
2
B j 2 (³ + 1) fm
z 2 2 z r cos Ë0
rn cos Ë 0n
z 2 2 2z r cos Ë0 + r2
For a FM modulator with m(t) = A cos (2Ãfm t),
rn g[n] G r−1 z
³ = fd A/fm

g[n + 1] z G(z) 2 zg[0]

Carson’s Rule For Arbitrary FM Signals:
g[n 2 1] z −1 G(z) + g[21]
B j 2 (D + 1) W

Final Value Theorem: where W is the bandwidth of the message signal

m(t), and the deviation ratio is
lim g[n] = lim (z21)G(z) (discrete-time)
n→∞ z→1 peak frequency deviation
D=
W