Maharaja Institute of Technology Mysore

VTU Question Papers

“Mtech – Singal Processing”

III Semester
Dec.2019 - Jan.2020

LIBRARY & INFORMATION CENTER

Maharaja Institute of Technology Mysore
Belawadi, Naguvanahalli (P), S.R.Patna (Tq), Mandya - 571477
 INDEX
S/L Subject Code Subject Title Exam Date

1 18ESP31 Adaptive Signal Processing Dec.2019 - Jan.2020

2 18ESP321 Speech and Audio Processing Dec.2019 - Jan.2020

3 18ESP332 Pattern Recognition & Machine Learning Dec.2019 - Jan.2020

 18ESP31

pm
USN

3
Third Semester M.Tech. Degree Examination, Dec.2019/Jan.2020

:2
Adaptive Signal Processing

8
Time: 3 hrs. Max. Marks: 100

:1
Note: Answer any FIVE full questions, choosing ONE full question from each module.

H
01
2. Any revealing of identification, appeal to evaluator and /or equations written eg, 42+8 = 50, will be treated as malpractice.

-M
Module-1
1 a. Define Adaptive Systems? Mention the characteristics of Adaptive Systems.

9
(10 Marks)

H
b. With a neat diagram of adaptive linear combiner, derive the MSE and explain the

01
performance function.

-M
(10 Marks)
-2 OR
Important Note : 1. On completing your answers, compulsorily draw diagonal cross lines on the remaining blank pages.

H
12
2 a. Explain the general properties of adaptive systems. Briefly discuss application of closed-

-M
loop adaptation. (10 Marks)
1-

b. A simple example of a single–input adaptive linear combiner with two weights is shown in
Fig.Q2(b). The input and desired signals are sampled sinusoids at the same frequency, with
H
-3

N samples per cycle. Assume N > 2. Explain the performance surface for the same.
-M
U
VT

m
-M

6p

H
H

:0

-M
-M

37

H
H

-M
:
01
-M

Fig.Q2(b) (10 Marks)

H
19
H

-M
-M

Module-2
20

3 a. Briefly discuss about the “learning curve”. (10 Marks)

H

b. Explain the gradient search by Newton’s method for adaptive systems. (10 Marks)
H

-M
2-
M

OR
-1

4 a. Explain the gradient search by method of steepest descent. (10 Marks)

H

b. Compare the learning curves and comment on the same. (10 Marks)
31

-M

Module-3
5 a. List out the properties of the LMS/Newton algorithm and compare them with that of the
H

LMS algorithm. (10 Marks)

-M

b. Discuss the LMS/Newton algorithm that can be applied to the practical situations. (10 Marks)
H

OR
-M

6 a. Explain the four types of realizations of structures in adaptive processing. (10 Marks)
b. With a neat diagram, explain the adaptive filter with preprocessing to produce orthogonal
signals. (10 Marks)
H

1 of 2
-M
H
M
 18ESP31

pm
Module-4

3
7 a. Explain adaptive modeling of a multipath communication channel with a neat figure.

:2
(10 Marks)
b. Illustrate with a neat diagram the adaptive modeling to measure the earth’s impulse

8
response. (10 Marks)

:1

H
01
OR

-M
8 a. With a neat block diagram, explain the synthesis of FIR digital filters for adaptation.
(10 Marks)

9
b. Enumerate the adaptive process to adjust the linear-phase weights to minimize mean-square

H
01
error using a zero-phase diagram. (10 Marks)

-M
-2 Module-5
9 a. Explain general description of inverse adaptive modeling.

H
(10 Marks)
12
b. Briefly discuss about equalization and deconvolution achieve in telephone channels.

-M
(10 Marks)
1-

OR H
-3

-M
10 a. Explain the adapting poles and zeros for IIR digital filter synthesis. (10 Marks)
U

b. Describe the two approaches to match both amplitude and phase specification while
maintaining stable IIR filter. (10 Marks)
VT

m
-M

6p
*****

H
H

:0

-M
-M

37

H
H

-M
:
01
-M

H
19
H

-M
-M

20

H
H

-M
2-
M

-1

H
31

-M
H
-M

2 of 2
H
-M
H
-M
H
M
 18ESP321

pm
USN

3
Third Semester M.Tech. Degree Examination, Dec.2019/Jan.2020

:4
Speech and Audio Processing

5
Time: 3 hrs. Max. Marks: 100

:1
Note: Answer FIVE full questions, choosing ONE full question from each module.

H
01
Module-1
2. Any revealing of identification, appeal to evaluator and /or equations written eg, 42+8 = 50, will be treated as malpractice.

-M
1 a. With neat block diagram, explain pitch period estimation using parallel processing approach.

0
(10 Marks)

H
02
b. Derive and explain the process of uniform lossless tube. (10 Marks)

-M
-2 OR
2 a. Explain short time autocorrelation functions with necessary waveforms.
Important Note : 1. On completing your answers, compulsorily draw diagonal cross lines on the remaining blank pages.

(10 Marks)

H
01
b. Explain short time average zero crossing rate with neat block diagram. (10 Marks)

-M
2-

Module-2
3 H
a. Explain uniform quantization and so that SNR = 6B – 7.2. (10 Marks)
-0

-M
b. Explain the differential PCM with feed forward and feedback concepts. (10 Marks)
U

OR
VT

4 m
a. Explain filter bank summation methods of short time synthesis. (10 Marks)
-M

b. Discuss overlap addition method for short time analysis. (10 Marks)
6p

H
Module-3
H

:3

5 -M
a. Explain Durbin’s recursive algorithm for the autocorrelation equations. (10 Marks)
-M

b. Explain linear predictive synthesizer with neat block diagram. (10 Marks)
33

H
H

OR
-M
:
01
-M

6 a. Explain digital voice response system with neat block diagram. (10 Marks)
P
b. Show that the gain of a linear predictive model is G 2  R n (0)    K R n (K )  En . (10 Marks)
H
20
H

K 1
-M
-M

20

Module-4
H

7 a. With a neat diagram, explain parallel format synthesizer. (10 Marks)

H

-M

b. Explain the concept of adaptive noise cancellation with neat diagram.

1-

(10 Marks)
M

-0

OR
H

8 a. Explain speech enhancement techniques. (10 Marks)

02

-M

b. Explain spectral subtraction and filtering techniques for speech enhancement. (10 Marks)

Module-5
H

9 a. Explain hidden Markov model and its applications. (10 Marks)

-M

b. Explain the following under audio processing : i) Stereo ii) Multichannel surround sound.
(10 Marks)
H
-M

OR
10 a. Discuss speech recognition and speaker recognition. (10 Marks)
b. Explain the frequency masking techniques. (10 Marks)
H

*****
-M
H
M
 USN 18ESP332

pm
Third Semester M.Tech. Degree Examination, Dec.2019/Jan.2020
Pattern Recognition and Machine Learning

2
Time: 3 hrs. Max. Marks:100

:5
Note: Answer any FIVE full questions, choosing ONE full question from each module.

8
Module-1

:1
1 a. Explain Bayesian curve fitting function. (10 Marks)

H
01
b. Explain Dirichlet distribution used in multinomial variables and show its conjugate prior for
2. Any revealing of identification, appeal to evaluator and /or equations written eg, 42+8 = 50, will be treated as malpractice.

-M
multinomial. (10 Marks)

0
OR

H
02
2 a. Explain the loss functions for regression. (10 Marks)

-M
b. Explain any one non parametric methods.
-2 (10 Marks)
Important Note : 1. On completing your answers, compulsorily draw diagonal cross lines on the remaining blank pages.

H
01
Module-2

-M
3 a. What is meant by a linear regression model? Explain any one of the linear basis function
4-

model used in linear regression. (10 Marks)

b. What is meant by Linear Discriminate Function (LDF). Explain least squares for
H
-0

classification. (10 Marks)

-M
U

OR
VT

4 a. Explain Bayesian linear regression with predictive distribution method. (10 Marks)
H

b. Explain Fisher’s Linear Discriminant Function. m (10 Marks)

-M

9p
Module-3

H
H

5 a. Explain the techniques used for constructing new Kernals from simpler Kernals used as
:5

-M
-M

building blocks. (10 Marks)

32

b. Explain Support Vector Machines (SVM) for regression. (10 Marks)

H
H

-M
:

OR
01
-M

6 a. Explain error back propagation procedure taking a simple example. (10 Marks)
b. Explain Relevance Vector Machine (RVM) for regression. (10 Marks)
H
20
H

-M

Module-4
-M

7 a. Explain K-means clustering.

20

(10 Marks)
H

b. What is meant by Principal Component Analysis (PCA). Explain the methods used in PCA.
H

(10 Marks)
-M
1-
M

-0

OR
H

8 a. Explain EM (Expectation Maximization) algorithm. (10 Marks)

04

-M

b. Explain two methods of principal component analysis and applications of PCA. (10 Marks)

Module-5
H

9 a. Explain conditional independence properties using a 3 example graph. (10 Marks)

-M

b. Explain Viterbi algorithm. (10 Marks)

H

OR
-M

10 a. Explain polynomial regression using graphical model. (10 Marks)

b. Explain sum product algorithm for the HMM (Hidden Markov Model). (10 Marks)
H

*****
-M
H
M