Digital Signal Processing for Engineering

Applications

Laboratory Experiments

Prof. Dr. C. Roppel

Dipl.-Ing. (FH) M. Margraf
Faculty of Electrical Engineering
Laboratory for Telecommunications

Contents:

General Information ................................................................................................................ 1

Python and the Jupyter Notebook .......................................................................................... 2
1 Using the Function Generator and the Oscilloscope ...................................................... 6
1.1 Check basic settings ............................................................................................................ 6
1.2 Using BenchVue ................................................................................................................. 7
1.3 Fast Fourier Transform (FFT) with the oscilloscope .......................................................... 9
1.4 Aliasing ............................................................................................................................. 11
1.5 Spectrum of a Square Wave .............................................................................................. 12
1.6 Amplitude Modulation (AM) ............................................................................................ 12
2 Sine and Dual Tone Signal .............................................................................................. 13
2.1 Sine Signal ........................................................................................................................ 13
2.2 Dual Tone Signal............................................................................................................... 16
3 Spectrum of the Dual Tone Signal ................................................................................. 18
4 Bearing Vibration Analysis............................................................................................. 20
4.1 Spectral Analysis ............................................................................................................... 20
4.2 Filter Design ...................................................................................................................... 23
4.3 Envelope Detection ........................................................................................................... 24
Digital Signal Processing for Engineering Applications Page 1

General Information

• For experiment 1 you work in the laboratory, while for the other experiments you
work from home using Python.

• Handle the hardware in the laboratory (the measurement equipment, PCs, cables,
connectors etc.) with care!

• Username and password for the lab PCs is "student".

• All user files have to be stored in C:\usr. At the end of each lab session, send your
files to your own e-mail account, and delete all your files on the lab PC.

You can use your HSM mail account via webmail:

https://webmail.fh-schmalkalden.de/

• Do not connect USB devices to the PC. These devices will be scanned by the virus
scanner and will generate an alarm, if a virus is detected!

• Laboratory report: The Jupyter Notebook is the report. Generate a PDF (via HTML)
from the notebook, do not send the Notebook itself. Change the filename:

The filename has to be DSP_EA_xxxxxx.pdf, where xxxxxx is your

matriculation number.

• Send the pdf document per e-mail to [email protected]. The deadline for
the report will be announced in the lecture.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 2

Python and the Jupyter Notebook

In this Laboratory we use the Python scientific ecosystem. As a user interface we use Jupyter
notebooks, an interactive computing environment where you can combine text, code and
visualizations. We recommend the Anaconda distribution, which includes all the necessary
packages and libraries. Download and install Anaconda Individual Edition from
www.anaconda.com. Launch the Anconda Navigator:

Select Environments. Here you can see and manage the packages. We need Numpy, Matplotlib,
Scipy and Pandas, which are installed.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 3

Select Learning. Here you can access the documentation for Anaconda and the packages.

To get started, select Scipy Documentation, next select Scipy Lecture Notes, and work through
the lecture notes in order to learn the basics about Python, Numpy, Matplotlib and Scipy.

Also have a look at the full documentation for Python, Numpy, Matplotlib, Scipy and Pandas,
which you will need while working through the lab experiments.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 4

Launch either Anconda Navigator, and then Jupyter Notebook, or Jupyter Notebook directly
from the start menue. This starts the notebook server and opens the notebook dashboard in a
browser window:

Select New -> Notebook (Python 3), to create a new notebook.

New cells are by default code cells. If you change the cell type to Markdown, you can enter text
and insert images.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 5

To finish, go to the Home tab and select Quit. This stops the notebook server and you can close
the browser (just closing the bowser does not stop the notebook server).

In a markdown cell, enter the title "Digital Signal Processing for Engineering Applications
Laboratory". Add another markdown cell and enter your name, HSM ID and the date. Add
another markdown cell and enter the heading "Experiment 1". Under this heading, insert your
results for experiment 1. Proceed in a similar way for the other experiments.

To generate a PDF document from the notebook, select File -> Download as -> pdf. This
requires additional software like Latex. Alternatively, you can select Download as -> HTML,
open the HTML document in your browser and print to PDF.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 6

1 Using the Function Generator and the Oscilloscope

In the lab you work with the HP33120A function generator and the digital oscilloscopes
Keysight DSOX 2002A or DSO 6032A. The user manual for the HP33120A is available in
printed from in the lab. The user's guide for the oscilloscopes is available on the PCs in the
directory C:\usr\Keysight. Screen shots of the oscilloscope can be taken with the BenchVue
software.

Do not change the language settings (English) of both the oscilloscope and the BenchVue
software.

The following instructions refer to the DSOX 2002A. The operation of the DSO 6032A is
similar; the BenchVue software is used for both oscilloscopes. When different parameters for
the DSO 6032A have to be used, this is indicated like
[DSO 6032A: horizontal scale 2 ms/, span 25 kHz, sampling rate 50 kHz].

In the text below, [key] refers to a key and SK[key] refers to a softkey of the oscilloscope.
A softkey is one of the five keys below the screen. By rotating and pressing the entry knob to
the right of the screen you select entries in the softkey menus. With the round key (Back) to
the left of the softkeys you can step back in the softkey menu.

1.1 Check basic settings

1. Press [Default Setup] on the oscilloscope (DSOX 2002A only).

2. Connect the function generator output to channel 1 input of the oscilloscope. Generate a
1 kHz sine wave with amplitude 2 Vpp (peak-to-peak). Press [Autoscale] on the
oscilloscope.

The status line at the top of the display shows the vertical and the horizontal scale settings.
The vertical scale now is 500 mV/ and the horizontal scale is 200 µs/.

3. Change the amplitude and frequency of the function generator and observe the
oscilloscope display. When finished, set it again to 1 kHz and 2 Vpp.

4. Change the vertical and horizontal scale of the oscilloscope and observe the display.
When finished, press [Autoscale] again.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 7

5. Measure the amplitude (peak-to-peak) and the frequency of the sine signal:
− Key [Meas] → SK[Source] → Select channel 1
− SK[Type] → Select "Peak-Peak" with the entry knob
− SK[Add Measurement] → Select "Frequency" with the entry knob

The measurements of the sine signal should show an amplitude of 2 Vpp and a frequency of
1 kHz. If the amplitude is too large, the load setting of the function generator has to be
corrected. The function generator has a source impedance of 50 Ω. If the generator is
connected to a 50 Ω load, the amplitude is only one half compared to the case when the
generator is connected to a high impedance load. The generator can be configured for a
50 Ω load or a high impedance load in order to show the correct amplitude. The setting can
be modified in the menu. Press the keys [Shift] [Menu], navigate with the arrow keys
to D:SYS MENU and further to 1:OUT TERM, where you can select either 50 OHM or
HIGH Z. Since the oscilloscope is a high impedance load (approx. 1 MΩ), choose
HIGH Z.
Ua

HP33120A

Upp
Ri = 50 Ω

∼
t
RL Ua

Figure 1-1: The amplitude of the function generator depends on the load impedance RL

1.2 Using BenchVue

1. Start the BenchVue software on the PC.

2. To start the oscilloscope app, double-click on the oscilloscope symbol in the lower right
corner.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 8

3. Go to the Screen Image tab. Select "Invert Colors" (otherwise the graphics have a black
background color, which is unfavorable for printing). Click on "Get Current Screen".

4. Click on "Export" or the save symbol in order to save screenshots of the oscilloscope in
various formats.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 9

1.3 Fast Fourier Transform (FFT) with the oscilloscope

Measurements in the frequency domain are based on the Discrete Fourier Transform (DFT).
Fast Fourier Transformation (FFT) is a term for algorithms, which allow for a fast and efficient
computation of the DFT. In this section we use the oscilloscopes built-in FFT function to
display the spectrum of the signal.

1. Activate the FFT:

Key [Math] → SK[Operator] → select FFT → SK[Source] → select Channel 1,
SK[More FFT] → SK[Auto Setup] to show the entire available spectrum

On the screen the magnitude spectrum of the signal is displayed. The frequency range
(span) and the resolution are displayed above the softkeys.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 10

2. The span depends on the sampling rate. The sampling rate is shown in the upper right
corner and depends on the horizontal scale setting. A horizontal scale of 200 µs/ results in
a sampling rate of 250 MSa/s (Million Samples per second). However, the FFT sampling
rate is usually lower and is related to the span by
FFT sampling rate = 2 times Span
The resolution depends on the FFT sampling rate:
Resolution = FFT sampling rate / Number of samples
Change the horizontal scale and observe how this changes the signal in the time domain,
the sampling rate, the span and resolution of the FFT.

By manually setting the span and the center frequency you can zoom into the FFT (this
was done in the screenshot shown). But this does not change the resolution! With the
knobs to the right of the [Math] key you can change the vertical scale of the FFT.

3. Autoscale the FFT (press SK[More FFT] → SK[Auto Setup]). Set the horizontal
scale to 200 ms/. The span now is 15.6 kHz (sampling rate 31.2 kHz). The spectral line of
the sine wave at 1 kHz is clearly visible [DSO 6032A: horizontal scale 2 ms/, span
25 kHz, sampling rate 50 kHz]. Use the cursor to measure the frequency:
Key [Cursors] → SK[Source] → select Source: Math, SK[Cursor] → select X1,
move the cursor in x direction by turning the cursor knob. Move the cursor to the peak,
the frequency is displayed in the cursor box.

4. The FFT vertical units are dBV (Decibel-Volt referenced to an rms voltage of 1 V):
U rms
U dBV = 20 log
1V
A sine wave having an rms (root-mean square) voltage of 1 V has an amplitude of 0 dBV.
Use the cursor to measure the amplitude:
Key [Cursors] → SK[Source] → select Source: Math, SK[Cursor] → select Y1,
move the cursor in y direction by turning the cursor knob. Move the cursor to the peak,
the amplitude is displayed in the cursor box.
For precise measurements of the amplitude, zoom into the FFT by manually setting the
span and the center frequency. Then it will be easier to move the cursor to the tip of the
peak in the spectrum.
Test different window functions: Press SK[More FFT] → SK[Window], and select
different window functions with the entry knob. Observe the effect on the shape of the
FFT. Use the Flat Top window for best amplitude measurements.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 11

5. Select different amplitudes at the function generator, and note down in a table the
following values:
− The peak-to-peak amplitude of the function generator
− The calculated rms value
− The calculated dBV value
− The measured dBV value

6. Autoscale the FFT (SK[More FFT] → SK[Auto Setup]). When the amplitude of
the input signal exceeds the vertical range of the oscilloscope, the signal is clipped.
Increase the amplitude of the function generator until clipping occurs, while observing the
spectrum. In order to see more details in the spectrum, you can switch off the time-domain
signal by pressing key [1] for channel 1 two times.

1.4 Aliasing

1. Adjust the function generator to generate a sine wave with a frequency of 1 kHz and an
amplitude of 2 Vpp. Adjust the horizontal scale of the oscilloscope to 200 ms/, activate
the FFT and select auto setup (SK[More FFT] → SK[Auto Setup]). The span will
be 15.6 kHz [DSO 6032A: 2 ms/, Span 25 kHz].

2. Increase slowly the frequency of the function generator. Wait for a stable FFT display and
watch the peak in the spectrum moving to the right.

3. Increase the frequency beyond 15.6 kHz. Since the FFT sampling rate is 31.2 kHz, signals
above 15.6 kHz will be aliased. Watch the peak, while increasing the frequency further
[DSO 6032A: sampling rate 50 kHz, aliasing above 25 kHz].

4. What happens, when you reach a frequency of 31.2 kHz [DSO 6032A: 50 kHz]? What
happens, when you increase the frequency further?

5. Select different frequencies at the function generator, and note down in a table the
following values:
− Generator frequency f0
− Effective sampling rate (FFT sampling rate) fs
− FFT frequency (frequency of the peak in the spectrum)
− The difference fs − f0
− The sum −fs + f0

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 12

1.5 Spectrum of a Square Wave

1. Adjust the function generator to generate a square wave with a frequency of 10 kHz and
an amplitude of 2 Vpp. Press [Autoscale] and observe the time domain signal.
Measure the period of the square wave.

2. Activate the FFT. Adjust the horizontal scale of the oscilloscope in order to see a
frequency range of about 200 kHz.

3. Measure the frequency of the harmonics, and note down the values in a table.

1.6 Amplitude Modulation (AM)

1. Adjust the function generator to generate an AM signal with these parameters:

− Carrier frequency 5 kHz
− Carrier amplitude 1 Vpp
− Modulating frequency 200 Hz
− Modulation index 0.7 (70 %)

2. Press [Autoscale] and observe the time domain signal. Change the modulating
frequency and the modulation index and observe the effect on the.

3. Activate the FFT. Adjust the horizontal scale of the oscilloscope in order to see a span of
15.6 kHz.

4. Measure the frequency of the spectral components of the AM signal, and note down the
values in a table.

5. Change the modulating frequency and the modulation index and observe the effect on
both the time domain waveform and the spectrum.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 13

2 Sine and Dual Tone Signal

In this experiment we generate a discrete-time version of a sine wave with Python and we plot
the signal. By using the sound card of the PC we can generate an analog signal, which we can
hear and measure with the oscilloscope. Next we generate a dual tone signal, which is composed
of two sine waves having different frequencies and amplitudes.

2.1 Sine Signal

Generate a sine wave with the frequency f1 = 1000 Hz, amplitude A1 = 1 and the sampling rate
fs = 8000 Hz.

1. In the Jupyter notebook, add a section for Experiment 2.

2. Import numpy. Define a variable fs = 8000 (sampling rate). Generate a vector time with
the sampling times (0, 1/fs, 2/fs, ...) with a total length of 33 samples, using the numpy
function arange().

Note that we use arange(33.0) instead of arange(33). The latter results in an

array of integer numbers, which can cause overflow problems when we multiply the array
with a large number (try it!). arange(33.0) results in an array of floating point
numbers. You can check the data type of a variable with variablename.dtype.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 14

3. Define a variable f1 = 1000 and calculate the sine values sin(2 π f1 t) of the sampling
times defined by the vector time.

4. Import matplotlib.pyplot. Plot the sine signal using the functions plot() and stem().
Generate a plot with two subplots, with axis label and plot title, similar to the plot shown
in the screen shot below. For mathematical expressions in text like labels, the Latex style
is supported.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 15

5. Typically, we start with a few samples and check the output. Now we extend the number
of samples to 16000 (corresponding to a signal time of 2 seconds). Because a plot of the
2 seconds signals does not make sense, we pass only the first 33 samples of the time
vector and the signal vector to the plot function.

Check the signal length using the function len().

6. Import the Audio class from IPython.display, and use the Audio() function to listen to
the sound.

This is only possible, if you have access to the lab:

− Connect the oscilloscope to the audio output of the PC.
− Measure frequency and amplitude of the signal in the time domain.
− Measure frequency and amplitude of the signal using the FFT function. Convert the
dBV value of the amplitude to the peak-to-peak value.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 16

2.2 Dual Tone Signal

1. Add a subsection with the heading "Dual Tone Signal".

2. Although not necessary, copy the code from the previous section including the import of
the packages. This allows you to work only with the code in the current subsection.

3. Add a second sine wave to the 1 kHz signal. The second signal shall have the amplitude
A2 = 0.5 and the frequency f = 1.7 kHz. Again, generate a plot of the first 33 samples, and
use the audio function to listen to the signal.

4. Determine the peak amplitude, the power, the rms value and the crest factor of the signal
using the abs(), max() and mean() functions.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 17

5. Calculate the theoretical values for the power, the rms value and the crest factor. Assume
a peak amplitude of A1 + A2 = 1.5, and that the two sine waves are uncorrelated and that
the total power is the sum of the individual power of the two sine waves.

6. By default, the audio function normalizes the amplitude of the signal to the range ±1. Now
use the option normalize = False. This will result in an error message, because the
peak amplitude of the signal is larger then 1.
Scale the amplitude of the dual tone signal accordingly.
Note: When you import or generate signals, it is good practice to normalize the
amplitude to 1.

This is only possible, if you have access to the lab:

− Connect the oscilloscope to the audio output of the PC.
− Measure the frequencies and amplitudes of the two sine signals using the FFT
function.
− Convert the dBV values of the amplitudes to their peak-to-peak values, and confirm
that the ratio is 2:1.
− Calculate the ratio of the amplitudes using the dBV values.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 18

3 Spectrum of the Dual Tone Signal

In this experiment we will calculate the spectrum of the dual tone signal of Section 2.2 using
the discrete Fourier transform (DFT):
N −1
S DFT (k ) = ∑ x(n) e − j 2 π n k / N , k = 0 , 1, ..., N − 1
n =0

The spectrum calculated by means of an N-point DFT is composed of N values spaced ∆ f = fs /N

apart. All N values represent the spectrum in the frequency range from 0 to fs, so for the kth
value we have
𝑘𝑘
𝑓𝑓𝑘𝑘 = 𝑓𝑓𝑠𝑠 , 𝑘𝑘 = 0, 1, … , 𝑁𝑁 − 1
𝑁𝑁

1. In the Jupyter notebook, add a section for experiment 3. Copy the relevant part of the code
from Section 2.2 to generate the dual tone signal. Change the signal length to N = 4096
samples.

2. Use the function fft() in the numpy.fft module to calculate the DFT. Plot the whole
spectrum (more precisely, the magnitude of the DFT) over k (k = 0, 1, …, N − 1).

3. Plot the whole spectrum over frequency f k (k = 0, 1, …, N − 1). For this purpose define a
vector freq_axis using the numpy command arange() (see Section 2.1).

4. The previous plot shows the spectrum in the range from 0 to fs , where the range from fs /2
to fs is the periodic repetition of the spectrum due to the sampling. We obtain the two-
sided spectrum in the range − fs /2 ≤ f ≤ fs /2 by swapping the upper half of spectrum with
its lower half. This can be done using the function fftshift() in the numpy.fft
module. Define a new frequency axis for the range − fs /2 ≤ f ≤ fs /2 and plot the two-sided
spectrum over frequency.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 19

5. Because of the symmetry of the spectrum, it is often shown as a one-sided spectrum in the
range 0 ≤ f ≤ fs /2. Plot this one-sided spectrum over frequency.

6. Finally, plot the one-sided spectrum in Decibels (dB) over frequency. Use the numpy
function log10() to convert the spectrum to Decibels. Now you should clearly see the
leakage effect for the 1700 Hz sine wave.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 20

4 Bearing Vibration Analysis

Here we analyse the signals from bearings with different faults available from the Society For
Machinery Failure Prevention Technology (https://mfpt.org/fault-data-sets/). The bearings
have the following parameters:

Revolution per second of the axle: frev = 25 Hz

Number of balls: N=8
Ball diameter: d = 0.235 inch
Average bearing diameter: D = 1.245 inch
Contact angle: φ=0

From this data set, we use three files:

• baseline.xlsx
• OuterRaceFault.xlsx
• InnerRaceFault.xlsx
These Excel files were obtained from the original binary files and contain signals sampled
with a rate of 48828 Hz for 3 seconds.

4.1 Spectral Analysis

1. In the Jupyter notebook, add a section for experiment 4. Calculate the characteristic
frequencies of the bearing:
− Fundamental Train Frequency (FTF)
− Ball Spin Frequency (BSF)
− Ball Pass Frequency Outer Race (BPFO)
− Ball Pass Frequency Inner Race (BPFI)

2. We use the pandas function read_excel to import excel files. Import pandas and read
in baseline.xlc. This results in a pandas data frame. Convert the data frame to a numpy
array using to_numpy. Define a variable fs = 48828 (sampling rate) and check the
signal length (number of samples).

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 21

3. Now signal is a matrix with only one column, that is an array([[...]]). Change this
to a vector, that is to array([...]), using signal = signal[:, 0].

Plot the signal over time in seconds.

4. Normalize the amplitude of the signal to 1. Again, plot the signal over time in seconds,
and confirm that the signal is in the range ±1. Use the Audio function to listen to the
sound.

5. For the bearing vibration analysis we have to plot the spectrum of signals often. Because
the characteristic frequencies are quite low, we are interested in the spectrum up to a
frequency of 500 Hz. Therefore we define a function plot_spectrum_500, which
plots the one-sided spectrum of a signal up to 500 Hz, scaled in Decibels.

Use this function to plot the spectrum.

6. Calculate the crest factor of the signal.

7. Calculate the kurtosis of the signal.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 22

8. Now repeat steps 2 to 7 for OuterRaceFault.xlsx.

− When plotting the spectrum, add vertical lines at integer multiples of BPFO.
Can you see a spectral line at BPFO?
− Compare the crest factor and the kurtosis with the values for the baseline signal.

9. Now repeat steps 2 to 7 for InnerRaceFault.xlsx.

− When plotting the spectrum, add vertical lines at integer multiples of BPFI.
Can you see a spectral line at BPFI?
− Compare the crest factor and the kurtosis with the values for the baseline signal.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 23

4.2 Filter Design

Before we continue with the analysis of the vibration signal, we will consider the design of
FIR filters, which we will need for extracting the envelope of the vibration signal.

FIR filters are designed with the function firwin() from the scipy.signal module. A
lowpass filter is designed as follows:
import scipy.signal as scs
b = scs.firwin(n_coeffs, f0)
This is a window based design using the Hamming window as a default. n_coeffs is the
number of coefficients, and f0 is the bandwidth. f0 is a number in the range [0 … 1], where
1 corresponds to fs / 2. The function returns the array b, the filter coefficients.

A bandpass filter is designed using

b = scs.firwin(n_coeffs, [f0, f1], pass_zero = False)
The band edges of the passband are f0 and f1. With pass_zero = False the range
from 0 to f0 is a stopband.

The function
w, H = scs.freqz(b)
calculates the frequency response of the filter in the range from 0 to fs / 2. w is the frequency
vector and H is the frequency response vector. The frequencies w are given in radians in the
range from 0 to π, where π corresponds to fs / 2. The magnitude of H is the amplitude response
of the filter.

The function lfilter() calculates the filtered signal y, using the filter coefficients b and
the input signal x:
y = scs.lfilter(b, 1, x)

1. Design a lowpass filter with 127 coefficients and the bandwidth 0.1 fs .
Plot the impulse response of the filter.
Plot the amplitude response of the filter, scaled in dB, over frequency in the range
from 0 to fs / 2.

2. Design a bandpass filter with with 127 coefficients and the band edges 0.1 fs and 0.2 fs .
Plot the impulse response of the filter.
Plot the amplitude response of the filter, scaled in dB, over frequency in the range
from 0 to fs / 2.

Fakultät Elektrotechnik Prof. Dr. C. Roppel

Digital Signal Processing for Engineering Applications Page 24

4.3 Envelope Detection

When a ball hits a fault on the outer or inner race, a signal with BPFO or BPFI is generated,
respectively. However this signal modulates signals at other resonant frequencies. One method
for demodulating these amplitude modulated signals is envelope detection.

Because we do not now the resonant frequency, we can use a combination of filters, crest factor,
kurtosis or spectral kurtosis so select a promising frequency band. Here we filter the signal first
with different filters and calculate the crest factor. Use the following filters:

Lowpass 2 kHz
Bandpass 2 kHz to 4 kHz
Bandpass 4 kHz to 6 kHz
Bandpass 6 kHz to 8 kHz
Bandpass 8 kHz to 10 kHz
Bandpass 10 kHz to 12 kHz
Bandpass 12 kHz to 14 kHz

Choose the filter which results in the highest crest factor. To extract the envelope of the signal,
we have to take the absolute value of the signal, and we have to send this signal through a
lowpass filter with the bandwidth 500 Hz.

1. Import OuterRaceFault.xlsx as described in Section 4.1.

2. Design lowpass/bandpass filters as described above with 127 coefficients. Generate a

single plot showing the amplitude response (scaled in dB) of all filters.

3. Filter the vibration signal and calculate the crest factor. Select the filtered signal with the
highest crest factor.

4. Generate a plot with two subplots, showing the original and the filtered signal,
respectively, for 100 ms.

5. Take the absolute value of the filtered signal (function abs()), and apply a 500 Hz
lowpass filter with 127 coefficients. Plot the demodulated signal for 100 ms.

6. Use the function plot_spectrum_500 from Section 4.1 to plot the spectrum of the
demodulated signal. Add vertical lines at integer multiples of BPFO. Can you see a
spectral line at BPFO?

7. Repeat steps 3 to 6 for InnerRaceFault.xlsx.

Fakultät Elektrotechnik Prof. Dr. C. Roppel