L.
EEC025 - Fundamentals of Signal Processing (FunSP)
2023/2024 – 1st semester
Week09, 13 Nov 2023
Objectives:
-modifying and measuring the frequency response of FIR filters running in real-time on the
STM32F746G Discovery board:
designing an equiripple linear-phase FIR of order 80
creating two modified versions of the designed FIR filter
obtaining experimentally the frequency responses of all three filters and
comparing them to the theoretical ones
DSP Education Kit
LAB 8
Finite Impulse Response (FIR) Filters
Issue 1.0
Contents
1 Introduction ............................................................................................. 1
1.1 Lab overview ........................................................................................................................... 1
2 Requirements .......................................................................................... 1
3 The FIR filter to be tested and modified ................................................... 1
4 FIR Filter with Coefficients Specified in Separate Header File ................... 4
5 Completing the C code ............................................................................. 4
6 Measuring the Frequency Responses ....................................................... 7
6.1 Setting-up for this lab experiment .......................................................................................... 8
6.2 Sketching the frequency response of the baseline FIR filter .................................................. 8
6.3 Sketching the frequency response of the H(-z) FIR filter ........................................................ 9
6.4 Sketching the frequency response of the H(z2) FIR filter ...................................................... 10
7 Extra-class: Generating FIR Filter Coefficient Header Files Using MATLAB
11
8 Conclusions ............................................................................................ 13
9 Additional References ............................................................................ 13
1 Introduction
1.1 Lab overview
The examples in this exercise introduce some of the concepts of Finite Impulse Response (FIR)
filtering. Also explored are various methods of modifying and estimating the magnitude frequency
response of FIR filters implemented in real-time.
2 Requirements
To carry out this lab, you will need:
An STM32F746G Discovery board
A PC running Keil MDK-Arm
MATLAB
An oscilloscope
3.5 mm audio jack cables + BNC cables
An audio frequency signal generator
3 The FIR filter to be tested and modified
The FIR filter that we test and modify experimentally in this lab is an equiripple linear-phase filter of
order 80 (this means that the length of its impulse response is 81 samples). The filter has a peculiar
frequency response magnitude so as to facilitate its experimental assessment. The filter is designed
in Matlab using the Parks MacClellan optimization algorithm that is called by the Matlab command
firpm() :
order=80;
f=[0 0.465 0.535 1.0];
m=[1 1 0 1];
w=[1 1];
h=firpm(order,f,m,w);
frange=pi*[-1:1/256:1-1/256];
[H W]=freqz(h,1,frange);
plot(W/pi, abs(H), 'LineWidth', 2)
xlabel('\omega/\pi')
ylabel('GAIN')
Copyright © 2023 Arm Limited (or its affiliates). All rights reserved.
Page 1
The corresponding frequency response magnitude is as illustrated in Figure 1.
GAIN
Figure 1: Frequency response magnitude of the FIR filter to be tested in the lab
Question 1 [ 1 pt / 10 ]: Assuming that the sampling frequency is 8 kHz, represent graphically the
frequency response magnitude that you expect to observe in the lab. Represent frequency using two
different but equivalent frequency axes: / (in the range [0, 1]) and freq (in Hz, in the range [0,
FN] where FN is the Nyquist frequency).
In particular, specify the exact frequencies, in Hertz, that correspond to the upper limit of the lower
band ( / 0.465), and the lower limit of the upper band ( / 0.535) where, very
approximately, ≅ 0.
Copyright © 2023 Arm Limited (or its affiliates). All rights reserved.
Page 2
0.465 → (Hz)
0.535 → (Hz)
In this lab, we will also perform and assess experimentally the impact of two types of filter
modification that have been studied in Problem 2 of the week05 TP class (video is available on
Moodle). In fact, if the impulse response of our linear-phase FIR is ℎ , and if its Z-Transform is
, we will create two modified versions of ℎ such that their Z-Transforms are , and
. The impact of these transformations are as illustrated in Figure 2 (not necessarily according
to this order).
a) b) c)
GAIN
GAIN
GAIN
Figure 2: Frequency response transformations of an original FIR filter a).
Question 2 [ 1 pt / 10 ]: If the Z-Transform of ℎ is , indicate what transformation in ℎ
causes its Z-Transform to become . Specify that modification for all describing the modified
(or new) impulse response. Implement the modification in Matlab and check the resulting frequency
response. What frequency reponse ( b) or c) ) in Figure 2 corresponds to this modified filter ?
Answer:
Question 3 [ 1 pt / 10 ]: If the Z-Transform of ℎ is , indicate what transformation in ℎ
causes its Z-Transform to become . Specify that modification for all 0,1, ⋯ ! 1, where
N=81. Implement the modification in Matlab and check the resulting frequency response. What
frequency reponse ( b) or c) ) in Figure 2 corresponds to this modified filter?
Answer:
Copyright © 2023 Arm Limited (or its affiliates). All rights reserved.
Page 3
4 FIR Filter with Coefficients Specified in
Separate Header File
In previous labs, we tested FIR and IIR filters whose coefficients were specified explicitly on the C
code. This is quite acceptable when the number of coefficients is low, in the order of 10, or less.
When the number of coefficients is large, it is more practical to specify the filter coefficients in a
separate header file. That is the case of the current lab that uses a header file (FPS81.h) that is
included in the C code baseline to be used in this lab: stm32f7_fir_intr_FPS81.c . Both
files are included in a ZIP that is available on Moodle.
Thus, the C code stm32f7_fir_intr_FPS81.c implements an FIR filter for which the filter
coefficients ℎ are not specified within this source file but are read from a separate header file
which is included by using the preprocessor command
#include “FPS81.h”
This also means that in order to change the characteristics of the FIR filter implemented, it is
sufficient to change in the preprocessor command the name of the header file to be included, and
Rebuild the project. For more on this possibility, see Section 7.
5 Completing the C code
In this lab experiment, we will use the main() project file that is named
stm32f7_fir_intr_FPS81.c and that is available on the Moodle platform. Its C code, which
is not complete, is listed next.
// stm32f7_fir_intr_FPS81.c
#include "stm32f7_wm8994_init.h"
#include "FPS81.h"
#include "stm32f7_display.h"
#define SOURCE_FILE_NAME "stm32f7_fir_intr_FPS81.c"
extern int16_t rx_sample_L;
extern int16_t rx_sample_R;
extern int16_t tx_sample_L;
extern int16_t tx_sample_R;
#define Nm1 2*N-1
enum filtertype{Plain, Shifted, Upsampled};
float32_t x[Nm1];
Copyright © 2023 Arm Limited (or its affiliates). All rights reserved.
Page 4
void BSP_AUDIO_SAI_Interrupt_CallBack()
{
int16_t i,j;
float32_t yn = 0.0;
// uncomment just one of the following three lines
enum filtertype myfilter=Plain;
// enum filtertype myfilter=Shifted;
// enum filtertype myfilter=Upsampled;
x[0] = (float32_t)(rx_sample_L);
BSP_LED_On(LED1);
switch (myfilter)
{
case Plain:
for (i=0 ; i<N ; i++) yn += h[i] * x[i];
break;
case Shifted:
// please complete your code here
break;
case Upsampled:
// please complete your code here
break;
default:
for (i=0 ; i<N ; i++) yn += h[i] * x[i];
}
for (j=(Nm1-1) ; j>0 ; j--) x[j] = x[j-1];
BSP_LED_Off(LED1);
tx_sample_R = (int16_t)(yn);
tx_sample_L = tx_sample_R;
return;
}
int main(void)
{
stm32f7_wm8994_init(AUDIO_FREQUENCY_8K,
IO_METHOD_INTR,
INPUT_DEVICE_INPUT_LINE_1,
OUTPUT_DEVICE_HEADPHONE,
WM8994_HP_OUT_ANALOG_GAIN_0DB,
WM8994_LINE_IN_GAIN_0DB,
WM8994_DMIC_GAIN_9DB,
SOURCE_FILE_NAME,
NOGRAPH);
while(1){}
}
Before you compile and run this code, you need to complete it (before the lab class!).
Copyright © 2023 Arm Limited (or its affiliates). All rights reserved.
Page 5
Question 4 [ 1 pt / 10 ]: in the C code, under “case Shifted:” which corresponds to one of the
above filter transformations, replace the comment “// please complete your code here” by
one of the following 3 alternatives (only one is correct):
Alternative A:
for (i=0 ; i<N ; )
{
yn += h[i] * x[i]; i++;
}
Alternative B:
for (i=0 ; i<N ; )
{
yn -= h[i] * x[i]; i++;
}
Alternative C:
for (i=0 ; i<(N-1) ; )
{
yn += h[i] * x[i]; i++;
yn -= h[i] * x[i]; i++;
}
yn += h[i] * x[i];
Explain what alternative is correct taking into consideration your answer to either Question 2, or
Question 3.
Question 5 [ 1 pt / 10 ]: in the C code, under “case Upsampled:” which corresponds to one of
the above filter transformations, replace the comment “// please complete your code
here” by one of the following 3 alternatives (only one is correct):
Alternative A:
for (i=0, j=0 ; i<N ; i++)
{
yn += h[i] * x[j];
j+=2;
}
Copyright © 2023 Arm Limited (or its affiliates). All rights reserved.
Page 6
Alternative B:
for (i=0, j=0 ; i<N ; i++)
{
yn += h[2*i] * x[j];
j++;
}
Alternative C:
for (i=0, j=0 ; i<N ; i++, j++)
{
yn += h[2*i] * x[2*j];
}
Explain what alternative is correct taking into consideration your answer to either Question 2, or
Question 3.
6 Measuring the Frequency Responses
After you complete the C code as recommended in the previous section, proceed, as indicated next,
to compile the code, upload it to the STM32F7 board, and to run it.
After unzipping them, take the stm32f7_fir_intr_FPS81.c (which should be completed
according to the suggestions in Section 5) and FPS81.h files to the “src” directory that is located
under the folder:
C:\uvision\Keil\STM32F7xx_DFP\2.9.0\Projects\STM32746G-Discovery\Examples\DSP Education Kit\
Remember that the directory where you can find the the DSP_Education_Kit.uvprojx
project file is:
C:\uvision\Keil\STM32F7xx_DFP\2.9.0\Projects\STM32746G-Discovery\Examples\DSP Education Kit\MDK-ARM
You can copy-paste this link directly to File Explorer in Windows for a quick and easy access. For your
convenience, this link is also available on a TXT file on Moodle.
As in previous labs, we use the DSP_Education_Kit.uvprojx project file as our baseline
project. This project file is represented by the icon , or just
. Double-click on this file/icon. This starts the Keil MDK-Arm development
Copyright © 2023 Arm Limited (or its affiliates). All rights reserved.
Page 7
environment (µVision). Replace the existing main() file in that project by the new main() that is
stm32f7_fir_intr_FPS81.c .
Now, proceed as usual to compile the code, downloading it to the STM32F746G board (by starting
the debugger), and then to run the code.
6.1 Setting-up for this lab experiment
Set the function generator to generate a sine wave having 5 Vpp and 100 Hz. Using a “T” and a BNC-
BNC cable, take the output of the function generator to CHAN1 of the oscilloscope. Connect the
output of a sinusoidal signal generator (i.e. the function generator) to the (left channel of the) LINE
IN socket on the Discovery board (Remember: make sure that you use the adapter with the blue
mini-jack whose interface board has a resistor divider. It is meant to protect the analog input of the
kit against excessive voltage levels).
Then, using another BNC-BNC cable, take the LEFT channel (yes, LEFT channel) of the STM32F746G
LINE OUT output to the CHAN2 input of the oscilloscope.
Using the oscilloscope SETTINGS button and menu, make sure that the Vpp and frequency of both
input and output signals are being measured in real-time.
6.2 Sketching the frequency response of the baseline FIR filter
The frequency response of a filter reflects its gain at different frequencies. As seen in previous lab
experiments, one way of assessing the frequency response of the filter is simply to measure its gain
using a sinusoidal input signal at a few frequencies of interest.
In this section, we measure the magnitude frequency response of the baseline FIR filter that is
signaled in the C code as “Plain”, and whose theoretical frequency response is depicted in Figure 1.
In this experiment, you will vary the frequency from 100 Hz, up to 3.8 kHz, and take note of the Vpp
and frequency of the output wave represented on the oscilloscope. Only four frequencies should
suffice: 100 Hz (which we will take as an approximation for 0 Hz), 3800 Hz (which we will take as an
approximation for 4 kHz), and two other frequencies that are anticipated in Question 1.
Based on these measurements, sketch in the following plot the approximate frequency response
magnitude of the baseline FIR filter that is implemented by the above C code when you vary the
frequency from 100 Hz up to 3.8 kHz.
Copyright © 2023 Arm Limited (or its affiliates). All rights reserved.
Page 8
/
Question 6 [ 1 pt / 10 ]: Is this sketch consistent with the expected frequency response magnitude
that is represented in Figure 1 and in Question 1 ?
6.3 Sketching the frequency response of the H(-z) FIR filter
In this section, we measure the magnitude frequency response of the modified FIR filter whose
transfer function is H(-z) and that is signaled in the C code as one of the two “Shifted” or
“Upsampled ”cases. Its theoretical frequency response is depicted in Figure 2.
In this experiment, you will vary the frequency from 100 Hz, up to 3.8 kHz, and take note of the Vpp
and frequency of the output wave represented on the oscilloscope. As in the previous experiment,
only a few frequencies should suffice in addition to 100 Hz (which we will take as an approximation
for 0 Hz), and 3800 Hz (which we will take as an approximation for 4 kHz).
Based on these measurements, sketch in the following plot the approximate frequency response
magnitude of the H(-z) FIR filter when you vary the frequency from 100 Hz up to 3.8 kHz.
Copyright © 2023 Arm Limited (or its affiliates). All rights reserved.
Page 9
/
Question 7 [ 2 pt / 10 ]: Is this sketch consistent with the expected frequency response magnitude
that is represented in Figure 2 ?
6.4 Sketching the frequency response of the H(z2) FIR filter
In this section, we measure the magnitude frequency response of the modified FIR filter whose
transfer function is H(z2) and that is signaled in the C code as one of the two “Shifted” or
“Upsampled ”cases. Its theoretical frequency response magnitude is depicted in Figure 2.
In this experiment, you will vary the frequency from 100 Hz, up to 3.8 kHz, and take note of the Vpp
and frequency of the output wave represented on the oscilloscope. As in the previous experiment,
only a few frequencies should suffice in addition to 100 Hz (which we will take as an approximation
for 0 Hz), and 3800 Hz (which we will take as an approximation for 4 kHz).
Based on these measurements, sketch in the following plot the approximate frequency response
magnitude of the H(z2) FIR filter when you vary the frequency from 100 Hz up to 3.8 kHz.
Copyright © 2023 Arm Limited (or its affiliates). All rights reserved.
Page 10
/
Question 8 [ 2 pt / 10 ]: Is this sketch consistent with the expected frequency response magnitude
that is represented in Figure 2 ?
Question 9: All of the above implementations of linear-phase filters are not totally efficient. What
else could be done in order to improve the realization/implementation efficiency ?
7 Extra-class: Generating FIR Filter Coefficient
Header Files Using MATLAB
If the number of filter coefficients is small, a coefficient header file may be edited by hand. To be
compatible with program stm32f7_fir_intr.c and others, a coefficient header file must
define a constant N and declare and initialize the contents of an array h[], which contains N floating
point values.
For larger numbers of coefficients, the MATLAB function stm32f7_fir_coeffs(), defined in
file stm32f7_fir_coeffs.m (and that is included on the ZIP that is available on Moodle), can be
used. This function should be passed a MATLAB array of real-valued coefficient values and will
prompt the user for an output filename.
Copyright © 2023 Arm Limited (or its affiliates). All rights reserved.
Page 11
For example, the coefficient file maf5.h was created by typing the following at the MATLAB
command prompt.
>> x = [0.2, 0.2, 0.2, 0.2, 0.2];
>> stm32f7_fir_coeffs(x)
enter filename for coefficients maf5.h
The coefficient filename must be entered in full, including the suffix .h.
Alternatively, the MATLAB filter design and analysis tool fdatool can be used to calculate FIR filter
coefficients and to export them to the MATLAB workspace (File – Export… – Export To Workspace /
Export As Coefficients). Then, function stm32f7_fir_coeffs() can be used to create a
coefficient file compatible with programs including stm32f7_fir_intr.c. It is recommended
that the filter coefficients values passed to function stm32f7_fir_coeffs() are normalized
such that their gain is unity. fdatool does it automatically, but if you are designing filter
coefficients without the aid of fdatool, you should aim for a passband gain of 1.
Figure 3: Design of a bandpass FIR filter using MATLAB fdatool
Coefficient header file bp1750.h was generated using MATLAB function
stm32f7_fir_coeffs()after designing the filter using fdatool (as shown in 3).
Copyright © 2023 Arm Limited (or its affiliates). All rights reserved.
Page 12
Note: Additional header and source files are provided in the lab-specific folder provided.
The file maf5.h listed in the following code snippet contains filter coefficient values that will result
in implementation of a five-point moving average filter.
// maf5.h
// this file was generated using function stm32f7_fir_coeffs.m
#define N 5
float h[N]={0.2, 0.2, 0.2, 0.2, 0.2};
8 Conclusions
This laboratory exercise has introduced FIR filters and explored several different frequency response
modifications.
9 Additional References
Moving average filters:
https://www.analog.com/media/en/technical-documentation/dsp-book/dsp_book_Ch15.pdf
Copyright © 2023 Arm Limited (or its affiliates). All rights reserved.
Page 13
