SIGNAL PROCESSING

DEPARTMENT OF ELECTRICAL ENGINEERING

E 102 Digital Signalbehandling EQ2300/ 2E1340
Final Examination 20071215, 14.0019.00

Literature: Hayes: Statisti al Digital Signal Pro essing and Modeling

or
Proakis, Manolakis: Digital Signal Pro essing
Bengtsson: Complementary Reading in Digital Signal Pro essing
Copies of the slides and two le ture hand outs
Beta  Mathemati s Handbook
Colle tion of Formulas in Signal Pro essing, KTH
Unprogrammed po ket al ulator.

Noti e: Answer in Swedish or English.

At most one problem should be treated per page.
Motivate ea h step in the solutions (also for the multi- hoi e questions).
Answers without motivation / justi ation arry no rewards
Write your name and personnummer on ea h page.
Write the number of solution pages on the over page.

The exam onsists of ve problems with a maximum of 10 points ea h.

For a passing grade, 24 points are normally required.

Conta t: Bhavani Shankar, Signal Pro essing, 08-790 84 35

Results: Will be reported within three working weeks at My pages.

Solutions: Will be available on the ourse homepage after the exam.

Good lu k!

• Don't forget to ll in the ourse evaluation form! Follow the link at
the ourse WWW page.

• Don't forget to pi k up your graded proje t report for Proje t 2 from

STEX on 21 De ember 2007.

1
1. (a) In the lass, we had onsidered zero padding as a tool to onvert ir ular on-
volution into linear onvolution. We shall now onsider the onverse. Let x(n)
be a length 4 sequen e (i.e, x(n) = 0, n > 3 or n < 0) and h(n) is a length 3
sequen e. Dene sequen es

z1 = {x(3), x(0), x(1), . . . , x(3)}

↑
z2 = {x(2), x(3), x(0), x(1), . . . , x(3)}
↑
z3 = {x(1), x(2), x(3), x(0), x(1), . . . , x(3)}
↑

i. Dene y(n) = z3 ∗h(n) (∗ denotes Linear Convolution). Show that {y(3), y(4),
y(5), y(6)} = x(n) h(n)
4 ( 4 denotes 4 point ir ular onvolution). (2p)

ii. Can you extra t x(n) h(n)

4 from

• z1 ∗ h(n)
• z2 ∗ h(n) (1p)

1
(b) If thez transform of x(n) = ρn u(n), |ρ| < 1 is X(z) = 1−ρz −1
, nd the sequen e
1
whose z transform is, Y (z) = . (3p)
(1−ρz −1 )2

( ) A student is asked to verify the DFT and IDFT operations. He orre tly obtains
the DFT of {a, b, c, d, e} as {A, B, C, D, E}. Instead of omputing the IDFT, he
↑ ↑
omputes the DFT of {A, B, C, D, E} again. What would his answer be? (4p)
↑

2. The re eived signal at a radar is given by,

y(n) = Aej2πf0 n + w(n)

where w(n) is a zero mean white noise with varian e σw2 . A genie provides the following
∗
values for the auto- orrelation fun tion, ryy (k) = E {y(n + k)[y(n)] },

ryy (0) = 3.0300

√ 
ryy (1) = 1.5 3+i

From these data, determine the following using Pisarenko method,

(a) frequen y of operation, f0

(b) Amplitude, A
( ) Noise varian e, σw2

(10p)

2
3. For the following signal model,

y(n) = A sin(ω0 n + φ) + w(n), n = 0, 1, 2, ..., N − 1

where w(n) is a white noise, onsider the problem of estimating the amplitude A and
phase φ of the signal (ω0 is assumed to be known). This is a nonlinear estimation
problem that an be solved using a nonlinear least squares approa h. However, the
signal model an be reparameterized using the following trigonometri rule

sin(α + β) = sin(α) cos(β) + cos(α) sin(β)

su h that the estimation problem be omes linear in two new parameters.

(a) Do the appropriate reparameterization and show how to al ulate A and φ from
the new parameters.

(b) Write the estimation problem as a linear least squares problem in the new pa-
rameters (use ve tors and matri es). Find the least squares solution.

( ) Simplify the least squares solution for ω0 = π/2.

(10p)

3
4. A lter has been implemented digitally as shown in Figure 1. The dis rete-time
input signal x(n) [Volts℄ is A/D onverted (quantized) using quantization step size
∆ [Volts℄. The signal is ltered using nite pre ision arithmeti s (the pre ision is
∆). Finally, the signal is D/A onverted ba k to analog dis rete-time domain, y(n),
again using step size ∆ [Volts℄. We wish to determine the unknown gain, k , and the
PSfrag repla ements
Finite Pre ision

x(n) A/D D/A y(n)

z −1

Figure 1: The digital lter.

quantization step-size, ∆, by feeding the lter with a sinusoid

x(n) = cos(ω0 n + φ) [Volts℄, n = 0, ..., N − 1,

that has a known frequen y ω0 = π/3 [rad/sample℄, and an unknown phase, φ,
uniformly distributed in [0, 2π). Figure 2 shows the estimated spe trum of y(n)
using
PSfrag periodogram averaging: K = 100 periodograms were averaged using L = 1000
repla ements
samples for ea h periodogram. The total number of output samples is N = KL.

30
Peak: P̂ (ejπ/3 ) = 20 dB.
20

10
P̂ (ejω ) [dB℄

-10

-20

-30
Peak: P̂ (ejπ ) = −12 dB.

-40
-π 0 π
ω [rad/sample℄

Figure 2: The averaged periodogram.

(a) Determine the lter transfer fun tion, H(z), as a fun tion of k, disregarding the
quantization ee ts. (1p)

(b) Show that the periodogram peak at frequen y ω0 an be approximated as

L|H(ejω0 )|2
P̂ (ejω0 ) ≈
4
Hint: Note that the quantization noise power is negligible at frequen y ω0 .
PL−1
1
Further more, you may assume that
L n=0 exp(jωn) ≈ 0, for all ω 6= 0. (3p)

( ) Using the result of part (b), obtain the value of k. (3p)

(d) Using suitable data provided in Figure 2, estimate the quantization step size, ∆.
Assume that no overow o urs in the summation. (3p)

4
 5. Consider the system in Figure 3, where the input to the system is a sinusoidal signal
with two odd overtones (harmoni s). That is

3
X
x(n) = Ak sin(2π(2k − 1) fx,0 n)
k=1

where fx,0 = 1/20. Figure 4 shows the periodogram of the output y(m) al ulated
using N = 103 data samples.

PSfrag repla ements

x(n) y(m)
D

Figure 3: Downsampling

DFT{y(m)}N −1 2
1
 
Pper (f ) = N m=0

PSfrag repla ements

fy
−fy,0 fy,0
-0.5 0.5

Figure 4: Periodogram of the down sampled signal.

Determine the smallest down sampling fa tor D that has been used in gure 3. What
are the other possible values of D? (10p)

Hint: Due to aliasing aused by down sampling, all signals have the same frequen y
after down sampling.