Final Exam
Name: Zhankun Luo
Problem 1
Given the sequence having 4 data points as , and
the sampling period of seconds, no window function is used.
a. Sketch the 4-points fast Fourier transform algorithm (Decimation in frequency FFT) to compute
the DFT coefficients: X(0), X(1), X(2), X(3)
b. Compute the amplitude spectrum
c. Determine the corresponding frequency for the amplitude spectrum .
d. Determine the frequency resolution.
e. Use the DFT formula to determine
(10 points)
solution
Thus
a. Sketch the 4-points fast Fourier transform algorithm (Decimation in frequency FFT) to compute
the DFT coefficients: X(0), X(1), X(2), X(3)
We compute X(k)
b. Compute the amplitude spectrum
The amplitude spectrum to a two-sided amplitude spectrum is
We compute
c. Determine the corresponding frequency for the amplitude spectrum .
For k, the corresponding frequency is
When k = 1, for the amplitude spectrum .
d. Determine the frequency resolution.
The frequency resolution is
e. Use the DFT formula to determine
When k=1
Problem 2
Given the following DSP system with a sampling rate of 8000 Hz
a. Obtain transfer function
b. Make a pole-zero plot and determine the stability.
c. Obtain the frequency response and then the magnitude response
d. Compute the filter gain at the frequency of 0 Hz, 1000 Hz, 2000 Hz, 3000Hz, 4000 Hz,
respectively. Make a plot of the magnitude frequency response.
e. Determine the filter type, that is, the lowpass filter, or high-pass filter, or bandpass filter, or
band-stop filter. (10 points)
solution
a. Obtain transfer function
Do transform to the DSP equation
b. Make a pole-zero plot and determine the stability.
zeros:
poles:
Because all poles:
The DSP system is stable.
c. Obtain the frequency response and then the magnitude response
Then the magnitude response
d. Compute the filter gain at the frequency of 0 Hz, 1000 Hz, 2000 Hz, 3000Hz, 4000 Hz,
respectively. Make a plot of the magnitude frequency response.
So, the filter gain at 0 Hz, 1000 Hz, 2000 Hz, 3000Hz, 4000 Hz are
e. Determine the filter type, that is, the lowpass filter, or high-pass filter, or bandpass filter, or
band-stop filter.
filter type: band-pass
Problem 3
Design a 5-tap bandpass FIR filter whose lower and upper cutoff frequencies are 800 Hz, and
1000, respectively using the Hamming window method.
Assume the sampling frequency is 4000 Hz.
a. List the FIR filter coefficients
b. Determine the transfer function
c. Determine the DSP equation
d. Set up MATLAB routine "freqz()" to obtain the frequency response plot.
(10 points)
solution
a. List the FIR filter coefficients
Here , then
So, we compute
Using the Hamming window method.
Then,
b. Determine the transfer function
c. Determine the DSP equation
d. Set up MATLAB routine "freqz()" to obtain the frequency response plot.
freqz([-0.007484, 0.008413, 0.1, 0.008413, -0.007484], [1], 4096, fs)
% B(z) = [-0.007484, 0.008413, 0.1, 0.008413, -0.007484]
% A(z) = [1]
% 4096 points for plot
% fs: 4000 Hz sampling rate
Problem 4
A DSP design engineer used the following MATLAB code to design FIR filter.
fs=8000;
f=[ 0 0.15 0.25 0.4 0.5 1]; % edge frequencies
m=[ 1 1 0 0 1 1]; % ideal magnitudes
w=[ 10 15 10 ]; % error weight factors
format long
b=remez(24,f,m,w) % Parks-McClellen algorithm and Remez exchange
(1) Determine the edge frequencies in Hz for passband and stopband
(2) Determine the filter type and number of taps.
(3) Weights for optimization, that Wp and Ws
(5 points)
solution
(1) Determine the edge frequencies in Hz for passband and stopband
passband: 0 - 600Hz and 2000 - 4000 Hz
stopband: 1000 - 1600 Hz
(2) Determine the filter type and number of taps.
filter type: band-stop
number of taps: 24+1=25
(3) Weights for optimization, that Wp and Ws
Problem 5
Design a second-order bandpass IIR digital Butterworth filter with the lower cut-off frequency of
100 Hz and upper cut-off frequency of 120 Hz at a sampling frequency of 1000 Hz using the
bilinear transformation method.
a. Determine the transfer function
b. Make a pole zero plot and determine the stability
c. Set up MATLAB routine "freqz()" to obtain the frequency response plot.
d. Determine difference equation in the direct-form I
e. Draw the realization block diagram using the direct-form II.
(10 points)
solution
a. Determine the transfer function
rad/s, Hz
order of Butterworth: 2 / 2 = 1
The 1st-order Butterworth filter
Then substitute , band-pass filter
Then with BLT, transfer function
b. Make a pole zero plot and determine the stability
poles: 0.7263 -j 0.5950, 0.7263 +j 0.5950
zeros: -1, +1
Because all poles,
The DSP system is stable
c. Set up MATLAB routine "freqz()" to obtain the frequency response plot.
freqz([0.0592, 0, -0.0592], [1, -1.4527, 0.8816], 4096, fs)
% B(z) = [0.0592, 0, -0.0592]
% A(z) = [1, -1.4527, 0.8816]
% 4096 points for plot
% fs: 1000 Hz sampling rate
d. Determine difference equation in the direct-form I
e. Draw the realization block diagram using the direct-form II.
the realization block diagram
Problem 6
For the following adaptive filter used for noise cancellation application,
and an adaptive filter with two taps:
with initial values , and
(a) determine the LMS algorithm equations
(b) perform adaptive filtering for each
(10 points)
solution
(a) Determine the DSP equations using the LMS algorithm
for k= 0, 1; that is, write the equations for all adaptive coefficients:
(b) perform adaptive filtering for each
Python script is below:
from rls.lms import Lms
from rls.rls import Rls
list_x = [-0.5, 1.2, 0.5]
list_d = [-1, 2, 1]
#####################
# LMS adaptive filter
#####################
N, mu, w_0 = 2, 0.1, [0.5, -0.5]
lms = Lms(mu, N, w_0)
list_y, list_e, list_w = [], [], [w_0]
for x, d in list(zip(list_x, list_d)):
lms.train(x, d)
list_y.append(lms.y)
list_e.append(lms.e)
list_w.append(lms.w)
print(list_y)
print(list_e)
print(list_w)
print('\n')
Then
Problem 7
Given a DSP system with a sampling rate set up to be 8,000 samples per second, implement
adaptive filter with 5 taps for system modeling.
Assume that the system as the following input and output:
tow taps:
a. Determine the DSP equations using the RLS algorithm with and .
b. Perform adaptive filtering for n=0, 1, 2.
(10 points)
solution
a. Determine equations using the RLS algorithm. [initialization for ]
b. Repeat (b) using the RLS algorithm with δ=1 and λ=0.96.
Python script is below:
from rls.lms import Lms
from rls.rls import Rls
def convert(L, n):
if isinstance(L, (float, int)):
return round(L, n)
list_new = []
for elem in L:
list_new.append(convert(elem, n))
return list_new
def print_approx(L, n=6):
L_new = convert(L, n)
print(L_new)
list_x = [2, -4, 4, -2]
list_d = [1, -1, 0, 1]
#####################
# RLS adaptive filter
#####################
delta, lambda_, N = 1, 0.96, 2
rls = Rls(delta, lambda_, N) # initial w = [0, ..., 0]
list_y, list_e, list_alpha, list_w, = [], [], [], [[0] * N]
for x, d in list(zip(list_x, list_d)):
rls.train(x, d)
list_y.append(rls.y)
list_e.append(rls.e)
list_alpha.append(rls.alpha)
list_w.append(rls.w)
print_approx(rls.Q)
print_approx(rls.k)
print('\n')
print_approx(list_y)
print_approx(list_e)
print_approx(list_alpha)
print_approx(list_w)
print('\n')
Then
Problem 8
The following Wiener filter is used to predict the sinusoid
Assume that where the Wiener filter predictor with
delays of , and is given as
Find the Wiener filter coefficients: , and , and the minimized the cost function of
(15 points)
solution
Here we define
Moreover, for , we conclude that ( are not equal each other)
Because we know that (for )
Thus for , we conclude
Find the Wiener filter coefficients , here if is invertible
Here we have
Finally, we compute with python Sympy library, is
Here is the python script to compute
import sympy as sym
from sympy import cos, sin
n1, n2, n3, omega1, omega2 = sym.symbols('n_1, n_2, n_3, Omega_1, Omega_2')
R = sym.Matrix([[1, 0.5*cos((n1-n2)*omega1) + 0.5*cos((n1-n2)*omega2)],
[0.5*cos((n1-n2)*omega1) + 0.5*cos((n1-n2)*omega2), 1]])
P = sym.Matrix([[0.5*cos(n1*omega1) + 0.5*cos(n1*omega2)],
[0.5*cos(n2*omega1) + 0.5*cos(n2*omega2)]])
det = R.det()
sym.trigsimp(sym.cancel(det)) # simplity det(R)
eigen = R.eigenvals()
sym.trigsimp(sym.cancel(eigen)) # find eigen value of R
w = R.inv()* P # find Wiener filter w_*
w = sym.trigsimp(sym.cancel(w)) # simplifty w_*
sym.cancel(w)
J_min = -P.transpose() * w + sig_sq # find J_min
J_min_simplify = sym.cancel(sym.trigsimp(J_min))
print(sym.latex(J_min_simplify))
As we can see, the minimized the cost function of is a very complex and
long expression
Actually, we need 4th-order difference equation to describe the ,
2nd-order is not enough
Problem 9
For the sampling conversion from 7 kHz to 3 kHz with the following specifications:
Passband frequency range = 0 – 500 Hz
Passband ripple = 0.02 dB
Stopband attenuation = 46 dB,
a. draw the block diagram for the interpolator;
b. determine the window type, filter length, and cutoff frequency if the window method is used
for the combined FIR filter H(z).
(10 points)
solution
a. draw the block diagram for the interpolator;
b. determine the window type, filter length, and cutoff frequency if the window method is used
for the combined FIR filter H(z).
For interpolation filter:
For Anti-aliasing filter:
Because 3 kHz < 4 kHz, we choose kHz
From table, Passband ripple<0.02 dB Stopband attenuation>46 dB,
window type: Hamming
select the closest odd number
cutoff frequency
Problem 10
For the design of a two-stage decimator (M1xM2=5x3) with the following specifications:
Original sampling rate = 15 kHz
Frequency of interest = 0 - 250 Hz
Passband ripple = 0.05 (absolute)
Stopband attenuation = 0.005 (absolute)
Final sampling rate = 1,000 Hz,
a. Draw the decimation block diagram;
b. Specify the sampling rate for each stage;
c. determine the window type, filter length, and cutoff frequency for each stage if the window
method is used for anti-aliasing FIR filter design
(10 points)
solution
a. Draw the decimation block diagram;
b. Specify the sampling rate for each stage;
Here we select the sampling rate for stage 1
the sampling rate for stage 2.
c.
A. determine the window type, filter length, and cutoff frequency for the first stage H1(z);
From table, Passband ripple<0.05 dB Stopband attenuation>46.02 dB,
window type: Hamming
filter length,
select the closest odd number
cutoff frequency
B. determine the window type, filter length, and cutoff frequency for the second stage H2(z)
From table, Passband ripple<0.05 dB Stopband attenuation>46.02 dB,
window type: Hamming
filter length,
select the closest odd number
cutoff frequency
