Page 1 of 9

 
Department of Electronics and Telecommunications

Examination paper for TTT4120 Digital Signal Processing

Academic contact during examination:: Torbjørn Svendsen

Phone: 930 80 477

Examination date: Thursday, December 18, 2014

Examination time (from - to): 09.00 - 13.00
Permitted examination support material: D – No printed or handwritten material
allowed. Specified, simple calculator
allowed

Other information:
• The examination consists of 4 problems where
– problem 1 concerns LTI systems
– problem 2 concerns digital filters
– problem 3 concerns filter realization
– problem 4 concerns signal generation
• Weighting of each sub-problem is given in parenthesis at the start of each problem.
• All problems are to be answered
• Grades will be announced 3 weeks after the examination date.

Language: English
Total number of pages: 9
Of this, number of enclosure pages: 1

Checked by:

Date Signature

Note! Students will find the examination results in Studentweb. If you have questions about your results, you
must contact your department. The Examination Office will not be able to answer such inquiries.
 Page 2 of 9

Problem 1 (3+3+4+2+3+4=19)

1a) Which properties need to be fulfilled in order to describe a time discrete system by its unit
pulse response h(n)?
Define the properties stability and causality using h(n).

1b) A time discrete signal y(n) is formed by processing another time discrete signal x(n). The
relationship between the signals is expressed by

y(n) = x(n) + 2x(n − 1) + x(n − 2)

Find the transfer function, H(z), of an LTI system that recovers x(n) from y(n).

1c) Find the causal unit pulse response, h(n), of the system in problem 1b.

1d) We have a stable and causal LTI system with real coefficients and transfer function

1 − βz −1
H(z) =
1 − αz −1
Define the legal region for poles and zeros in the z plane.
Sketch the region of convergence (ROC) for the system.

1e) Show that the time discrete Fourier transform (DTFT), X(f ), of a sequence x(n) has the
following properties

i) X(−f ) = X ∗ (f ) when x(n) is real

ii) X(f ) = −X(−f ) when x(n) = −x(−n)

1f ) Suppose you have an FIR filter with real coefficients, h(n), and a finite duration real se-
quence, x(n) We shall use this filter to produce a filtered version of x(n) that has the same
phase as the original signal.

1. We first filter x(n) with h(n) to form s(n) = x(n) ∗ h(n).

2. Thereafter, we filter the time reversed sequence using the same filter, i.e., v(n) =
s(N − 1 − n) ∗ h(n), where N is the length of s(n).
3. Finally, the filtered output is obtained by time reversing v(n), y(n) = v(K − 1 − n).
K is the length of v(n)

Show that the DTFT of y(n), Y (f ), has the same phase as X(f ).
 Page 3 of 9

Problem 2 (2+4+2+2+3=13)
We have a linear, time invariant time discrete system defined by the difference equation:
∞
X 1
y(n) = (− )k x(n − 2k); n ≥ 0
k=0
2

2a) Calculate the 8 first values of the unit pulse response of this system.

2b) Show that the system can be implemented with the filter depicted in Figure 1.
(Hint: Start by finding y(n) for the first values of n )

Figure 1: Filter implementation

2c) Find the z transform H(z) for this filter?

2d) What are the locations of the poles and zeros of H(z) ?

2e) Suppose we input a signal x(n) = cos(2πf n) to the filter.

For which frequency will the amplitude of the output signal y(n) be the largest?
How big is the maximum amplitude?
 Page 4 of 9

Problem 3 (3+2+4+4+4=17)

3a) A causal digital filter is given by the block diagram in i Figure 2.

Figure 2: Digital filter

Find the transfer function, H(z) of the filter.

Sketch the location of poles and zeros in the z plane.

3b) The filter in Figure 2 can be realized using other structures.

Draw a Direct Form I realization of the filter (DF I).

3c) Show that the filter in Figure 3 is equivalent to the filter in Figure 2. Based on the filter
structure in Figure 3, find the unit pulse response of the filter, h(n).

Figure 3: Alternative filter structure

 Page 5 of 9

Our filter is to be implemented in fixed point arithmetic, and we wish to ensure that we will
not have overflow in the summations, and that we have control of the round-off errors from the
multiplications.

3d) Find how we need to scale the input signal in order to avoid summation overflow in the two
filter structures of Figures 2 and 3.
Hint: If a sequence, x(n),Pincreases monotonously, such that x(n) < 0 for n < K and
x(n) ≥ 0 for n ≥ K, then ∞
P∞ PK−1
n=0 |x(n)| = n=0 x(n) − 2 n=0 x(n)

3e) Assume that the round-off error after each multiplication can be modeled as an additive
noise source with zero mean and variance σq2 .
Find the power of the total round-off noise at the output of the two filters expressed by σq2 .
 Page 6 of 9

Problem 4 (2+4+2+2=10)
We are to make a causal 2nd order IIR filter with an unit pulse response that is a pure sinusoid.
For simplicity, we let the unit pulse response be a zero phase cosine function. The filter will
operate on a sampling rate of Fs = 48kHz, and the sinusoid is to have a frequency of 8kHz.

4a) Find an expression for the unit pulse response, h(n), of the filter expressed as a sum of
exponential functions. (Remember that ejω = cos ω + j sin ω)

4b) Show that the z transform of the filter’s unit pulse response is

1 − 21 z −1
H(z) =
1 − z −1 + z −2
Draw the location of poles and zeros in the z plane.

4c) What is the expression for the difference equation describing the system?

4d) Explain how a filter like this can be used as a computationally efficient signal generator for
a digital sinusoid with normalized frequency f = 61 .
 Page 7 of 9

Appendix page 1 of 3

Appendix: Some basic equations and formulas.

A. Sequences:
N −1
X 1 − αN
n
α =
n=0
1−α
∞ −∞
X 1 X 1
|α| < 1 ⇒ αn = and − αn =
n=0
1−α n=−1
1−α
N −1
X 1 − αN N αN
(n + 1)αn = − ; α 6= 1
n=0
(1 − α)2 1 − α
∞
X 1
|α| < 1 ⇒ (n + 1)αn =
n=0
(1 − α)2

B. Linear convolution:
X X
y(n) = h(n) ∗ x(n) = h(k)x(n − k) = x(k)h(n − k)
k k

Y (z) = H(z)X(z) ⇒ Y (f ) = H(f )X(f ) ⇒

Y (fk ) = H(fk )X(fk ) fk = k/N for k = 0, . . . , N − 1 where we write Y (k) = Y (fk )

C. Transforms:
X X
Z : H(z) = h(n)z −n ⇒ H(f ) = h(n) e−j2πnf
n n

L−1
X
DFT : H(k) = h(n) e−j2πnk/N k = 0, ..., N − 1
n=0
N −1
1 X
IDFT: h(n) = H(k) ej2πnk/N n = 0, ..., L − 1
N k=0
 Page 8 of 9

Appendix page 2 of 3

D. The sampling (Nyquist) theorem:

Given an analog signal xa (t) with bandwidth ±B which is sampled by Fs = 1/Ts :

x(n) = x(nTs ) = xa (t)|t=nTs n = −∞, ...., ∞

X
X(f ) = X(F/Fs ) = Fs Xa [(f − k)Fs ]
k

xa (t) can be recovered from x(n) ⇔ Fs ≥ 2B

E. Autocorrelation, energy spectrum and Parseval’s theorem:

Given a sequence h(n) with finite energy Eh :

X
Autocorrelation: rhh (m) = h(n)h(n + m) m = −∞, ...., ∞
n

Energy spectrum: Shh (z) = H(z)H(z −1 ) ⇒ Shh (f ) = |H(f )|2

X Z 2π
2
Parseval’s theorem: Eh = rhh (0) = h (n) = |H(f )|2 df
n 0

F. Multirate formulae:

Decimation where Tsy = DTsx :

X
v(mTsy ) = h[(mD − k)Tsx ] x(kTsx ) m = −∞, ...., ∞
k

Upsampling where Tsx = U Tsy :

X
y(lTsy ) = h[(l − nU )Tsy ] x(nTsx ) l = −∞, ...., ∞
n

D
Interpolation where Tsy = DTsv = Tsx :
X U
y(lTsy ) = h[(lD − mU )Tsv ] x(mTsx ) l = −∞, ...., ∞
m
 Page 9 of 9

Appendix page 3 of 3

G. Autocorrelation, power spectrum and Wiener-Khintchin theorem:

Given a stationary, ergodic sequence x(n) with infinite energy :

Autocorrelation: γxx (m) = E[x(n)x(n + m)] m = −∞, ...., ∞

Power spectrum: Γxx (z) = Z[γxx (m)] ⇒

X
Wiener-Khintchin: Γxx (f ) = DT F T [γxx (m)] = γxx (m) e−j2πmf
m

H. The Yule-Walker and Normal equations where a0 = 1:

p
X
Yule-Walker equations: ak γxx (m − k) = σf2 δ(m) m = 0, ..., p
k=0

p
X
Normal equations: ak γxx (m − k) = −γxx (m) m = 1, ..., p
k=1

I. Some common z-transform pairs

Signal, x(n) X(z) ROC

1 δ(n) 1 Alle z
1
2 u(n) 1−z −1
|z| > 1
1
3 an u(n) 1−az −1
|z| > |a|
az −1
4 nan u(n) (1−az −1 )2
|z| > |a|
1
5 −an u(−n − 1) 1−az −1
|z| < |a|
az −1
6 −nan u(−n − 1) (1−az −1 )2
|z| < |a|