Midterm Exam 2 KEY

ELEC456 / DIGITAL SIGNAL PROCESSING

Spring 2020

Date Start time Duration No. of pages

th
Wed. 29 April 2020 12:00 90 minutes 5
INSTRUCTIONS
Open Book and Notebook Exam.
1. Read and observe the following rules:
a) No candidate shall be permitted to enter the examination room after Question Mark
the expiration of one–half hour, or to leave during the first half–hour of
the examination. (1)
b) Candidates are not permitted to ask questions to the invigilators, except / 30
in cases of supposed errors or ambiguities in examination questions.
2. Show all your work. Justify your answers. Partial credit is possible for an
answer, but only if you show the intermediate steps in obtaining the answer.
(2)
/ 40
Caution – Candidates guilty of any of the following, or similar, dishonest
practices shall be immediately dismissed from the examination and shall be (3)
liable to disciplinary action:
▪ Making use of any books, papers or memoranda, cell phones, audio or / 30
visual cassette players or other memory aid devices, other than as
authorized by the examiners.
Total
▪ Speaking or communicating with other candidates. /100
▪ Purposely exposing written papers to the view of other candidates.
▪ The plea of accident or forgetfulness shall not be received.

STUDENT NAME :__________________________________________________________

STUDENT ID :_________________________________

Solve the following questions with justifications

 Student Name: Student ID:
-------------------------------------------------------------------------------------------------------------------------------

Question 1: a) 10pts b) 10pts c) 10pts [30pts]

Consider the discrete‐time signal, x[n], as shown below. We sampled its DTFT, X(Ω) , uniformly at every
2π/5 radians to obtain the DFT coefficients, 𝑋[𝑘] = 𝑋(Ω𝑘 ) = 𝑋(2𝜋𝑘/5). The discrete‐time signal, 𝑥̃[n],
is obtained by taking the 5-point inverse DFT, i.e., 𝑥̃[𝑛] = 𝐼𝐷𝐹𝑇(𝑋[𝑘]). So, answer the following with
proper justifications:

a) Why DFT coefficients cannot approximate the DTFT spectrum properly? What kind of
problem occurs?
b) Plot 𝑥̃[n].
c) How to solve the problem in (a) so that the DFT coefficients can be used to recover:
i) some part of the signal from the DFT coefficients, 𝑋[𝑘] = 𝑋(Ω𝑘 ) = 𝑋(2𝜋𝑘/5)?
ii) all of the signal with a new sampling scheme?

Sol:

a) Because the recovered signal, 𝑥̃[𝑛], is periodic with period, N=5 samples, the IDFT operation will
cause “Time Aliasing” which will alias x[0,1] with x[5,6].

b)
c) Time aliasing can be avoided by:

i) Applying the 5-point time window to the original signal before taking DTFT and then sampling
the DTFT uniformly at every 2π/5 radians to obtain the DFT coefficients, 𝑋[𝑘] = 𝑋(Ω𝑘 ) = 𝑋(2𝜋𝑘/5).

ii) Sampling the DTFT uniformly at every 2π/7 radians to obtain the DFT coefficients, 𝑋[𝑘] =
𝑋(Ω𝑘 ) = 𝑋(2𝜋𝑘/7).

------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------

Page 2 of 5
Student Name: Student ID:
-------------------------------------------------------------------------------------------------------------------------------
Question 2: (A-D) 10pts [40pts]
For the DT system shown below the input signal, xn(t) is corrupted by a high frequency additive noise, n(t)
i.e., xn(t)=s(t) + n(t) where s(t) is a bandlimited audio signal. The magnitude spectrum of both the signal
and noise, S(f) and N(f), are plotted below. Since the input signal is not bandlimited due to the noise, a
continuous-time anti-aliasing filter, Haa(f) is used before sampling (magnitude spectrum shown below).
xn (t) xc (t ) x[n] r[n] y[n ] yr (t)
Haa(w) C/D H () 2 D/C

T1 T2
S( f ) N( f )

f f
-10KHz 10KHz -10KHz 10KHz

Haa(f ) H ()
1

f 
-20KHz -10KHz 10KHz 20KHz − C C

Solve the following with justifications:

A) What is the maximum sampling period, T1 so that the aliasing can be avoided?
B) With the sampling rate found in part (A), propose a suitable DT filter with a frequency response,
𝐻(Ω) and cut-off frequency, Ω𝐶 , and T2 so that yr(t) = Ks(t) with some constant, K.
C) In order to simplify the discrete time system, we want to remove the down-sampling block (by
M=2) so that y[n]=r[n]. In this case repeat part (B).
D) In the simplified discrete time system (no down-sampling), with some aliasing allowed during
sampling, can yr(t) = Ks(t) still be achieved by using a higher sampling period, T1 than the one
found in part (A)? If so, plot DTFT of x[n], 𝑋(Ω)and 𝑋(𝑓) (in both radians and Hz together in the
same plot) to show the aliasing clearly. Then compute the highest possible T1, the corresponding
T2 and the cut-off frequency of the LPF, Ω𝐶 (in radians) with the corresponding fc (in Hz.). If this
is not possible, justify your answer.

Sol:
A) 1/ T1=2x20KHz = 40KHz
B) 1/T2=20KHz and the discrete time filter is a LPF shown below with cut-off frequency, Ωc = π/2.

H ()


− C C
C) Same LPF as in part (B) and 1/ T1=1/ T2=40KHz
D) 1/ T1=1/ T2=30KHz and aliasing is obvious as shown below. In this case the discrete time filter is
a LPF shown above with cut-off frequency, Ωc = 2π/3 rad. or fc=10 KHz.

Page 3 of 5
Student Name: Student ID:
-------------------------------------------------------------------------------------------------------------------------------
X()

-30KHz -10KHz 10KHz 15KHz 20KHz 30KHz

f
2
−2  2

3

------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------

Page 4 of 5
Student Name: Student ID:
-------------------------------------------------------------------------------------------------------------------------------
Question 3: (a-c) 10pts [30pts]
For the system described by the difference equation:
y[n] = x[n - 1] + 4x[n-2] - 5y[n - 1] - 6y[n-2]

A) Find the transfer function, H(z). Draw the pole-zero diagram.

B) What kind of a system (filter) is this (IIR or FIR)? Is it causal? Is it stable? Justify.
C) From H[z] find the impulse response, h[n].

Sol.
A) Take Z transform of both sides:
(1 + 5𝑧 −1 + 6𝑧 −2 )𝑌(𝑧) = 𝑧 −1 (1 + 4𝑧 −1 )𝑋(𝑧)
𝑌(𝑧) 𝑧 −1 (1 + 4𝑧 −1 ) (𝑧 + 4) (𝑧 + 4)
𝐻[𝑧] = = −1 −2
= 2 =
𝑋(𝑧) 1 + 5𝑧 + 6𝑧 𝑧 + 5𝑧 + 6 (𝑧 + 2)(𝑧 + 3)

ROC for H(z)

o x-3 x-2 0
1 3
-4

B) This is an IIR filter due to the feedback from the past output samples. It is a causal system (current
output depends on current and past input values). It is not stable (unit circle is not is not in ROC).
C)

𝑧 −1 (1 + 4𝑧 −1 )
𝐻(𝑧) = −1 −2
= 𝑧 −1 𝐻1 (𝑧)
1 + 5𝑧 + 6𝑧
(1 + 4𝑧 −1 ) 2 1
𝐻1 (𝑧) = = −
−1
1 + 5𝑧 + 6𝑧 −2
1 + 2𝑧 −1 1 + 3𝑧−1
Taking the inverse Z-transform by inspection and delay 1 sample:

ℎ[𝑛] = 2(−2)𝑛−1 𝑢[𝑛 − 1] − (−3)𝑛−1 𝑢[𝑛 − 1] 𝑤𝑖𝑡ℎ |𝑧| > 3

Page 5 of 5