Homework 6

Course Title: Digital Signal Processing I (Spring 2020)

Course Number: ECE53800

Instructor: Dr. Li Tan

Author: Zhankun Luo

Chapter 7:

Problems 7.1, 7.2 (b), 7.5 (b), 7.6 (b), 7.7, 7.8, 7.9, 7.10, 7.11, 7.15, 7.19, 7.20

MATLAB Project 7.36

Advanced Problem 7.38

Problems

Problems 7.1
Design a three-tap FIR lowpass filter with a cutoff frequency of 1500 Hz and a sampling
rate of
8000 Hz using

1. Rectangular window function

2. Hamming window function.

Determine the transfer function and difference equation of the designed FIR system, and
compute and plot the magnitude frequency response for Ω=0, π/4, π/2, 3π/4, and π (rad).

solution
h(-M)~h(M) are

1. Rectangular window function

Transfer function

Difference equation

2. Hamming window function

Transfer function

Difference equation
Problem 7.2
Design a three-tap FIR high-pass filter with a cutoff frequency of 1600 Hz and a sampling
rate of
8000 Hz using

2. Hamming window function.

Determine the transfer function and difference equation of the designed FIR system, and
compute and plot the magnitude frequency response for Ω=0, π/4, π/2, 3π/4, and π (rad).

solution

h(-M)~h(M) are

2. Hamming window function.

Transfer function

Difference equation
Problem 7.5
Design a five-tap FIR bandpass filter with a lower cutoff frequency of 1600 Hz, an upper
cutoff
frequency of 1800 Hz, and a sampling rate of 8000 Hz using

2. Hamming window function.

Determine the transfer function and difference equation of the designed FIR system, and
compute and plot the magnitude frequency response for Ω=0, π/4, π/2, 3π/4, and π (rad).

solution

h(-M)~h(M) are

2. Hamming window function.

Transfer function

Difference equation
Problem 7.6
Design a five-tap FIR band-reject filter with a lower cutoff frequency of 1600 Hz, an upper
cutoff
frequency of 1800 Hz, and a sampling rate of 8000 Hz using

2. Hamming window function.

Determine the transfer function and difference equation of the designed FIR system, and
compute and plot the magnitude frequency response for Ω=0, π/4, π/2, 3π/4, and π (rad).

solution

h(-M)~h(M) are

2. Hamming window function.

Transfer function

Difference equation
Problem 7.7
Given an FIR lowpass filter design with the following specifications:

Passband=0~800 Hz
Stopband=1200~4000 Hz
Passband ripple=0.1 dB
Stopband attenuation=40 dB
Sampling rate=8000 Hz

Determine the following:

1. Window method
2. Length of the FIR filter
3. Cutoff frequency for the design equation

solution

Look up Table 7.7 in page 254, based on the requirements:

Passband ripple
Stopband attenuation

1. Window method: Hanning

2. Compute the transition band

Then use formula in Table 7.7

Find the closest odd number 63

3. Then estimate the cutoff frequency

The cutoff frequency are Hz

Problem 7.8
Given an FIR high-pass filter design with the following specifications:

Stopband=0~1500 Hz
Passband=2000~4000 Hz
Passband ripple=0.02 dB
Stopband attenuation=60 dB
Sampling rate=8000 Hz

Determine the following:

1. Window method
2. Length of the FIR filter
3. Cutoff frequency for the design equation.

solution

Look up Table 7.7 in page 254, based on the requirements:

Passband ripple
Stopband attenuation

1. Window method: Blackman

2. Compute the transition band

Then use formula in Table 7.7

Find the closest odd number 89

3. Then estimate the cutoff frequency

The cutoff frequency are Hz

Problem 7.9
Given an FIR bandpass filter design with the following specifications:

Lower cutoff frequency=1500 Hz

Lower transition width=600 Hz
Upper cutoff frequency=2300 Hz
Upper transition width=600 Hz
Passband ripple=0.1 dB
Stopband attenuation=50 dB
Sampling rate=8000 Hz

Determine the following:

1. Window method
2. Length of the FIR filter
3. Cutoff frequencies for the design equation

solution

Look up Table 7.7 in page 254, based on the requirements:

Passband ripple
Stopband attenuation

1. Window method: Hamming

2. Compute the transition band

Then use formula in Table 7.7

Find the closest odd number 45

3. Then estimate the cutoff frequency

The cutoff frequency are Hz

Hz
Hz
Problem 7.10
Given an FIR band-stop filter design with the following specifications:

Lower passband=0~1200 Hz
Stopband=1600~2000 Hz
Upper passband=2400~4000 Hz
Passband ripple=0.05 dB
Stopband attenuation=60 dB
Sampling rate=8000 Hz

Determine the following:

1. Window method
2. Length of the FIR filter
3. Cutoff frequencies for the design equation

solution

Look up Table 7.7 in page 254, based on the requirements:

Passband ripple
Stopband attenuation

1. Window method: Blackman

2. Compute the transition band

Then use formula in Table 7.7

Find the closest odd number 111

3. Then estimate the cutoff frequency

The cutoff frequency are Hz

Problem 7.11
Given an FIR system

realize H(z) using each of the following specified methods:

1. Transversal form, and write the difference equation for implementation

2. Linear phase form, and write the difference equation for implementation

solution

1. Transversal form

difference equation for implementation

2. Linear phase form

difference equation for implementation

Problem 7.15
Determine the transfer function for a five-tap FIR lowpass filter with a cutoff frequency of
2000 Hz and a sampling rate of 8000 Hz using the frequency sampling method.

solution

To reduce this ripple effect, the modified specification with a smooth transition band

for k=0~M

Then we can calculate

Thus, h(-M) ~ h(M) are

The transfer function

If we don't consider the ripple effect:

for k=0~M

Thus, h(-M) ~ h(M) are

The transfer function

Problem 7.19
A lowpass FIR filter to be designed has the following specifications:
Design method: Packs-McClellan algorithm

Sampling rate=1000 Hz
Passband=0~200 Hz
Stopband=300~500 Hz
Passband ripple=1 dB
Stopband attenuation=40 dB

Determine the error weights Wp and Ws for passband and stopband in the Packs-
McClellan
algorithm.

solution

Map

Thus , now determine the error weights Wp and Ws

So, ; Set minimal Filter order = 12

% Problem 7.19
clear; clc; close all
fs = 1000;
f = [ 0 0.4 0.6 1]; % edge frequencies
m = [ 1 1 0 0] ; % ideal magnitudes
w = [ 1 12 ]; % error weight factors [W_pass, W_stop]
b = firpm(12, f, m, w); % (12+1)Parks-McClellen algorithm and Remez
exchange
format long
vpa(b, 4)
freqz(b, 1, 512, fs) % plot the frequency response
axis([0 fs/2 -80 10]);
KEY VALUE KEY VALUE
0.03034 0.04032
-0.03153 -0.1051
0.01087 0.3023
0.4654
Problem 7.20
A bandpass FIR filter to be designed has the following specifications:
Design method: Packs-McClellan algorithm

Sampling rate=1000 Hz
Passband=200~250 Hz
Lower stopband=0~150 Hz
Upper stopband=300~500 Hz
Passband ripple=1 dB
Stopband attenuation=30 dB

Determine the error weights Wp and Ws for passband and stopband in the Packs-
McClellan
algorithm

solution

Map

Thus , now determine the error weights Wp and Ws

So, ; Set minimal Filter order = 18

% Problem 7.20
clear; clc; close all
fs = 1000;
f = [ 0, 0.3, 0.4, 0.5, 0.6, 1]; % edge frequencies
m = [ 0 0 1 1 0 0]; % ideal magnitudes
w = [ 39 10 39 ]; % error weight factors [W_stop, W_pass, W_stop]
b = firpm(18, f, m, w); % (18+1) taps Parks-McClellen algorithm and
Remez exchange
format long
vpa(b, 4)
freqz(b, 1, 512, fs); % plot the frequency response
axis([0 fs/2 -80 10])

KEY VALUE KEY VALUE

0.0329 0.01892

-0.06725 -0.05156
0.08012 0.1148
-0.07239 -0.1604
0.02714 0.1867
Advanced Problems

Problem 7.38
The frequency response of a half-band digital differentiator is given below:
for |Ω| < π / 2:
Design the FIR differentiator with (2M+1) coefficients using the Fourier transform method

solution

Using the Fourier transform design method, we can obtain the filter coefficients below:

The is an odd function, :

Converting non-causal half-band differentiator to causal one with (2M+ 1) coefficients

yields the following relation:
MATLAB Projects

Problem 7.36
Speech enhancement:
A digitally recorded speech in the noisy environment can be enhanced using a lowpass
filter if the recorded speech with a sampling rate of 8000 Hz contains information within
1600 Hz.
Design a lowpass filter to remove the high-frequency noise above 1600 Hz with following
filter specifications:

passband frequency range: 0~1600 Hz;

passband ripple: 0.02 dB;
stop-band frequency range: 1800~4000 Hz;
stop-band attenuation: 50 dB.

Use the designed low-pass filter to filter the noisy speech and adopt the following code to
simulate the noisy speech:

load speech.dat
t=[0:length(speech)-1]*T;
th=mean(speech.speech)/4; %Noise power =(1/4) speech power
noise=sqrt(th)*randn([1,length(speech)]); %Generate Gaussian noise
nspeech=speech+noise; % Generate noisy speech

In this project, plot the speech samples and spectra for both noisy speech and the
enhanced speech and use MATLAB sound() function to evaluate the sound qualities. For
example, to hear the noisy speech:

sound(nspeech / max(abs(nspeech)), 8000);

solution

Look up Table 7.7 in page 254, based on the requirements:

Passband ripple: 0.02 dB;

Stop-band attenuation: 50 dB.

1. Window method: Hamming

2. Compute the transition band
 Then use formula in Table 7.7

Find the closest odd number 133

3. Then estimate the cutoff frequency

The cutoff frequency are 1700 Hz

4.

are

[0.000754, -0.004524, -0.002923, 0.003281, 0.004574,

-0.001218, -0.005305, -0.001259, 0.00489, 0.003627, -0.003341,
-0.005347, 0.000922, 0.005987, 0.001892, -0.005322, -0.004502,
0.003394, 0.006307, -0.000531, -0.006835, -0.002707, 0.005853,
0.005629, -0.003441, -0.007549, -0.0, 0.007936, 0.003803,
-0.006542, -0.007153, 0.00348, 0.009247, 0.000757, -0.00946,
-0.005365, 0.007503, 0.009359, -0.003513, -0.011753,
-0.001915, 0.011763, 0.007796, -0.008988, -0.012892, 0.003538,
0.015915, 0.003911, -0.015756, -0.01216, 0.011694, 0.019605,
-0.003557, -0.02441, -0.008197, 0.024673, 0.022508, -0.01848,
-0.037841, 0.003568, 0.052398, 0.024362, -0.06438, -0.080682,
0.072255, 0.309515, 0.425, 0.309515, 0.072255, -0.080682,
-0.06438, 0.024362, 0.052398, 0.003568, -0.037841, -0.01848,
0.022508, 0.024673, -0.008197, -0.02441, -0.003557, 0.019605,
0.011694, -0.01216, -0.015756, 0.003911, 0.015915, 0.003538,
-0.012892, -0.008988, 0.007796, 0.011763, -0.001915,
-0.011753, -0.003513, 0.009359, 0.007503, -0.005365, -0.00946,
0.000757, 0.009247, 0.00348, -0.007153, -0.006542, 0.003803,
0.007936, -0.0, -0.007549, -0.003441, 0.005629, 0.005853,
-0.002707, -0.006835, -0.000531, 0.006307, 0.003394,
-0.004502, -0.005322, 0.001892, 0.005987, 0.000922, -0.005347,
-0.003341, 0.003627, 0.00489, -0.001259, -0.005305, -0.001218,
0.004574, 0.003281, -0.002923, -0.004524, 0.000754]

5. Hamming window function

 are

[6e-05, -0.000364, -0.00024, 0.000278, 0.000404, -0.000113,

-0.000523, -0.000133, 0.000552, 0.000441, -0.000438,
-0.000757, 0.000141, 0.00099, 0.000337, -0.001024, -0.000932,
0.000755, 0.001506, -0.000136, -0.001867, -0.000789, 0.001814,
0.001853, -0.0012, -0.002786, -0.0, 0.003257, 0.001641,
-0.002963, -0.003394, 0.001727, 0.004791, 0.000409, -0.005316,
-0.003132, 0.004543, 0.005868, -0.002278, -0.00787, -0.001322,
0.008363, 0.005699, -0.006748, -0.009927, 0.002791, 0.012841,
0.003224, -0.013255, -0.010427, 0.010208, 0.017403, -0.003207,
-0.022328, -0.007598, 0.023153, 0.021357, -0.017711,
-0.036594, 0.003477, 0.051422, 0.024046, -0.063844, -0.080304,
0.072104, 0.309354, 0.425, 0.309354, 0.072104, -0.080304,
-0.063844, 0.024046, 0.051422, 0.003477, -0.036594, -0.017711,
0.021357, 0.023153, -0.007598, -0.022328, -0.003207, 0.017403,
0.010208, -0.010427, -0.013255, 0.003224, 0.012841, 0.002791,
-0.009927, -0.006748, 0.005699, 0.008363, -0.001322, -0.00787,
-0.002278, 0.005868, 0.004543, -0.003132, -0.005316, 0.000409,
0.004791, 0.001727, -0.003394, -0.002963, 0.001641, 0.003257,
-0.0, -0.002786, -0.0012, 0.001853, 0.001814, -0.000789,
-0.001867, -0.000136, 0.001506, 0.000755, -0.000932,
-0.001024, 0.000337, 0.00099, 0.000141, -0.000757, -0.000438,
0.000441, 0.000552, -0.000133, -0.000523, -0.000113, 0.000404,
0.000278, -0.00024, -0.000364, 6e-05]

The magnitude and phase of are

6. Speech with noise:

 Speech filtered:

comment

1. The low-pass filter eliminates the high-frequency component of noise.

2. The quality of speech improves
3. The results and figures (except Problem 7.19, 7.20) are generated by Python
scripts
I used the python library written on my own to replace the function in MATLAB.
Here is the main script for problem 7.36

from fir_filter.choose_window_type import *

from fir_filter.calc_window_len import *
from fir_filter.calc_freq_cutoff import *
from fir_filter.fir_filter import *
from fir_filter.window import *
from fir_filter.calc_mag_angle import *
from fir_filter.filter import *
from fir_filter.fft1d import *

# choose window type

passband_ripple = 0.02
stopband_attenuation = 50
str_window_type = choose_window_type(passband_ripple,
stopband_attenuation)
print(str_window_type)
# calc length of window
f_s = 8000
list_transient_band = [ [1600, 1800] ]
window_len = calc_window_len(str_window_type,
list_transient_band, f_sample=f_s)
print(window_len)
# calc the cutoff frequency & normalized cutoff frequency
list_freq_cutoff = calc_freq_cutoff(list_transient_band)
print(list_freq_cutoff)
print_approx([calc_omega(list_freq_cutoff[0], f_s)])
# calc h(-M)~h(M) list of filter
list_filter = fir_filter(list_freq_cutoff, f_s, window_len,
"low_pass")
print_approx(list_filter, precision=6)
# Hamming window: calc h_w(n) = h(n) * w(n)
str_window_type = "Hamming"
path_fig = "../p7_36_hamm.png"
list_filter_window = window(list_filter,
str_window_type=str_window_type)
print_approx(list_filter_window, precision=6)
# plot Magnitude & Phase of H(e^{j Omega})
list_mag, list_angle, list_omega =
calc_mag_angle(list_filter_window)
plot_mag_angle(list_mag, list_angle, list_omega,
path_fig=path_fig)
# read data: "speech.dat"
import numpy as np
filename = "./speech.dat"
speech = np.loadtxt(filename)
th = np.mean( speech**2 ) / 4 # Noise power =(1/4) speech
power
noise = np.sqrt(th) * np.random.randn( len(speech) ) #
Generate Gaussian noise
nspeech = speech + noise # Generate noisy speech
# low-pass filter to filter the noisy speech
nspeech_filtered = filter(nspeech, list_filter_window)
nspeech_filtered = np.asarray(nspeech_filtered)
# save sounds
from scipy.io.wavfile import write
write("../nspeech.wav", f_s, nspeech)
write("../nspeech_filtered.wav", f_s, nspeech_filtered)
# plot speech with noise & filtered speech
plot_spectrum(nspeech, f_s, path_fig="../p7_36_noise.png",
str_title="speech with noise")
plot_spectrum(nspeech_filtered, f_s,
path_fig="../p7_36_filtered.png", str_title="speech filtered")