Scribd
Upload
24 views

ADSP Compre

This document contains instructions for a 90 minute comprehensive examination in advanced digital signal processing. It consists of 7 questions worth a total of 70 marks. The questions cover topics like generating power spectral densities in MATLAB, determining autocorrelation functions, analyzing minimum phase models, applying the steepest descent algorithm to Wiener filters, modifications to LMS filters, true/false questions about adaptive filtering concepts, and deriving signal and observation models based on a scenario involving moving objects. Students are instructed to upload their answers as a single PDF file to Google Classroom and Mettle.

Uploaded by

Nivedita Acharyya 2035
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
24 views

ADSP Compre

This document contains instructions for a 90 minute comprehensive examination in advanced digital signal processing. It consists of 7 questions worth a total of 70 marks. The questions cover topics like generating power spectral densities in MATLAB, determining autocorrelation functions, analyzing minimum phase models, applying the steepest descent algorithm to Wiener filters, modifications to LMS filters, true/false questions about adaptive filtering concepts, and deriving signal and observation models based on a scenario involving moving objects. Students are instructed to upload their answers as a single PDF file to Google Classroom and Mettle.

Uploaded by

Nivedita Acharyya 2035
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Birla Institute of Technology and Science Pilani

KK Birla Goa Campus


Comprehensive Examination
Semester II 2020-2021
EEE G613 Advanced Digital Signal Processing
Duration: 90 minutes Max. Marks: 70
General Instructions:
1. All Questions are mandatory.
2. You can refer to any online/offline materials but strictly no contact with humans.
3. You have to scan your answers and upload it on Google classroom and Mettle in the form of pdf.
4. Upload single pdf only.
5. Keep cameras on during the examinations.
6. Refer other instructions mentioned on google classroom.

Q1.
a) Write a MATLAB code to generate the power spectral density of two cascaded blocks. They
are AZ (1) and AZ (2) blocks. The programmer can choose any value of variables required in
the programing. [5]
b) Write a MATLAB code to determine the autocorrelation functions of AP(3) model. The value
gain and coefficients can be decided by the programmer. [5]

Q2.
a) A minimum phase model is expressed as 𝑦(𝑛) = 𝑤(𝑛) + 𝑐 𝑤(𝑛 − 1). Here, 𝑤 is white
Gaussian noise with zero mean and unit variance and 𝑐 is a constant. The normalized
autocorrelation 𝜌(1) is one fourth of the value of 𝜌(0). Determine the constant 𝑐 expressed
in the model equation. [5]
b) Consider a process 𝑦(𝑛) = 𝑤(𝑛) − 0.3 𝑤(𝑛 − 1) + 0.2 𝑤(𝑛 − 2).
i) Is this process stationary? Justify your answer.
ii) Is the model minimum phase? Justify your answer. [2+3]

Q3. Consider an AP (3) model which has system transfer function


𝑑0 𝑑0
𝐻(𝑧) = =
1 + 𝑎1 𝑧 −1 + 𝑎2 𝑧 −2 + 𝑎3 𝑧 −3 (1 − 𝑝1 𝑧 −1 )(1 − 𝑝2 𝑧 −1 )(1 − 𝑝3 𝑧 −1 )
If 𝑝1 = 𝑝2 , then Determine
a) Is the given model minimum phase if 𝑎1 = −2 and 𝑎3 = 0.5. [5]
b) Is the given model minimum phase if 𝑎1 = 2 and 𝑎3 = 2. [5]

Q4. The steepest descent algorithm is applied to the wiener filter. The input to the filter is 𝑥(𝑛) and its
1 0.8
autocorrelation matrix 𝑅𝑥 = [ ]. Also, the cross-correlation between 𝑥(𝑛) and the desired
0.8 1
0.7 0.7
response 𝑑(𝑛) is 𝑅𝑥𝑑 = [0.8 0.5]𝑇 . The 𝑅𝑥 = 𝑄𝛬𝑄 𝐻 where 𝑄 = [ ] is unitary matrix and 𝑄 𝐻
0.7 −0.7
is Hermitian of 𝑄. Assume initial tap-weight vector is zero. Write a MATLAB code to determine the
range of step size parameter μ for which the steep descent algorithm is stable. Assume the value of μ
as the mean of its range, write a MATLAB code to determine the natural mode (𝑣(𝑛)) of steepest
descent method after 100 iterations. Where, 𝑣(𝑛)= 𝑄 𝐻 𝑐(𝑛). The 𝑐(𝑛) is weight error vector. [10]

Q5.
a) The LMS filter is widely used because of its simple implementation. However, several variants
of LMS filters are proposed to overcome its limitation. For example, normalized LMS filter is
proposed in which the step size in the algorithm is normalized to overcome this limitation.
Can you think of any other modification that can be made in LMS filter to overcome its
limitation? If you are suggesting any modification, kindly give justification for it. A brief but
logical explanation is enough to answer this question. [5]
b) What will be the effect on LMS algorithm if you set error 𝑒(𝑛) = 𝑦(𝑛) − 𝑑(𝑛). Here, 𝑑(𝑛) is
the desired response and 𝑦(𝑛) is the output of the filter. [5]

Q6.
a) As per the Wold decomposition theorem, a general non-stationary random process can be
expressed as the sum of a regular process and a predictable process. True or False?
b) The All zero model is also known as ____ model.
c) The duration of the impulse response of All Zero model is finite. True or False?
d) When steepest descent algorithm is applied to Wiener filter, the computation of correction
vector to weights 𝛿𝒘(𝑛) requires correlation matrix of tap input vector. True or False?
e) In steepest descent method, the positive value of step size parameter ensures the
convergence or stability of method. True or False?
f) What is regularizing term in the cost function of RLS adaptive filter?
g) The NLMS filter has time-varying ______________ parameter.
h) What is principle of orthogonality in Wiener filter?
i) The LMS filter doesn’t suffer from gradient noise. True or False?
j) The forgetting factor value in the cost function of RLS filter is between 0 and 1. True or
False? [1×10=10]

Q7. There are two objects moving in the same direction in a straight line. The instantaneous position
of first object and second object is 𝑦1𝑐 (𝑡) and 𝑦2𝑐 (𝑡), respectively. The true position of first and
second objects are given as 𝑦1𝑝 (𝑛) = 𝑦1𝑐 (𝑛𝑇) and 𝑦2𝑝 (𝑛) = 𝑦2𝑐 (𝑛𝑇), respectively. Here, 𝑇 is the
sampling interval and 𝑛 is the instant in time. A sensor records the mean value of the positions at 𝑛𝑡ℎ
instant. The recorded data 𝑥(𝑛) is equal to the mean of the true positions plus some observational
error 𝑣(𝑛) for 𝑛 ≥ 0. The 𝑣(𝑛) is white Gaussian noise with zero mean and unit variance. Both objects
move with random accelerations. Let 𝑦1𝑎 (𝑛) and 𝑦2𝑎 (𝑛) are the random acceleration of first and
second object, respectively. Determine the signal model equation and observation model equation
for the given scenario. [10]

*** ALL THE BEST ***

You might also like