Exp#02 Analysing Biomedical Signal Using DFT and Reconstruct The Signal Using IDFT

BME 3112
Biomedical Signal Processing Laboratory

Experiment # 02
Name: Analyzing Biomedical signals using Discrete Fourier Transform (DFT) and
reconstruct the original signal using Inverse Discrete Fourier Transform (IDFT)
Objective:
To implement DFT and IDFT and analyze Biomedical Signal using them.
Basic Theoretical Principles

The main points to understand about a Discrete Fourier Transform (DFT) are:
1. The Discrete Fourier Transform of a time signal, if done "correctly," is simply a sampled version
of the spectrum of the signal over some finite frequency range.
2. The Discrete Fourier Transform is very useful in computationally intensive filtering,

convolution, or correlation applications where it is often more efficient to do calculations in the
frequency domain instead of the time domain.
3. The Discrete Fourier Transform can be efficiently computed using a technique called a Fast
Fourier Transform.
We begin the mathematical development of the DFT with the Fourier Transform:
𝑥(𝑡) = ∫ 𝑋(𝑓)𝑒 𝑗2𝜋𝑓𝑡 𝑑𝑓

Fourier Transform (FT):
𝑋(𝑓) = ∫ 𝑥(𝑡)𝑒 −𝑗2𝜋𝑓𝑡 𝑑𝑡
The signal x(t) is some general analog time signal and X(f) is the spectrum of the time signal. Both
are complete representations of the signal, x. Obviously, neither one of these representations are
useful for DSP since both are analog.
Let us implement our idea of sampling the time signal, x(t), possibly processing those sampled
values in some manner, then reconstructing an analog signal from the sampled values. It is logical
to suggest that the same thing could be done using sampled values of the spectrum rather than
sampled values of the time signal. The obvious question is how to obtain sampled values of the
spectrum of a signal. Sampling the analog time signal leads to the Discrete-Time Fourier
Transform (DTFT):
The time signal, x(t), must be sampled at a sufficient rate to avoid aliasing. Xp(f) is periodic but
still continuous over frequency which is unsuitable for DSP. Sampling Xp(f) over one period or an
integer number of periods leads to the Discrete Fourier Transform (DFT):
The signal x(n) is created by taking N samples of the original signal x(t) over some finite duration
of time, D. The signal XT(k) is the DFT of the analog time signal, x(t). The previous equation
shows it can be computed by taking the N samples of x(t), multiplying each sample by a weighted
exponential function, then summing the results. The DFT will have N values spaced apart in the
frequency domain by 1/D. The following important result is typically proved in an engineering
course but simply stated in our course:
In other words, the DFT, which can be computed from a finite number of samples of x(t), is
approximately equal to the sampled spectrum of x(t) scaled by a constant,1/ts.
The accuracy of the DFT depends on two factors: the sampling rate chosen for x(t) and the duration
of time over which x(t) is sampled. The important factors to stress are:
1. If x(t) is not sampled sufficiently fast, aliasing will occur causing the DFT to be a poor match
to the original signal spectrum.
2. If x(t) is not sampled for a long enough duration of time, the resolution of the DFT will be poor.
The effect of sampling rate and the duration of time over which x(t) is sampled are explored using
several MATLAB exercises.
MATLAB Exercises to Explain the DFT
MATLAB program to find DFT without using Matlab function

Source Code: We can implement the DFT in straight forward way by using above equation using
loop and sum.
%For 4 Point DFT
N= 4;
x=1:4;
for k=0:3
for n = 0:3;
y(n+1) = x(n+1).*exp(-(1j*2*pi*k*n)/N);
end
xdft(k+1)= sum(y);
end
To check use matlab built in fft.m function
fft(x)
It will produce the same result as above
But loop execution time is higher Here is alternative ways to avoid loop.
DFT function is here. Write the complete code for getting the output. Similarly, you can write the
functions of inverse DFT.
%%DFT.m
function [Xk] = dft(xn)
N=length(xn);
n = 0:1:N-1; % row vector for n
k = 0:1:N-1; % row vector for k
WN = exp(-1j*2*pi/N); % Twiddle factor (w)
nk = n'*k; % creates a N by N matrix of nk values
WNnk = WN .^ nk; % DFT matrix
Xk = (WNnk*xn ); % xn is the input.
Example 1: Effect of Sampling Rate, Sf, on the DFT
In this example, the effect of sampling rate on the DFT is explored. The time signal is chosen to
be a decaying exponential. The signal and its spectrum are given by:
The signal and its spectrum are first plotted using MATLAB.
MATLAB Code:
t=0:0.01:8;
x=t.*exp(-t);
subplot(2,1,1);
plot(t,x);
title('x(t)=t*exp(-t)');
grid
f=-2:0.01:2;
xf=1./(1+2*pi*j*f).^2;
subplot(2,1,2);
plot(f,abs(xf));
title('Magnitude ofX(f)');
grid;
xlabel('Frequency (Hz)');
ylabel('Amplitude (A)');
In order to compute a DFT, the time signal x(t) must be sampled over some duration of time.
According to the Nyquist Theorem, the sampling frequency should be at least twice the bandwidth
of the signal. As seen from the plot of the spectrum, the spectral components above 1 Hz are very
small. In order to see the effects of sampling rate, two different sampling frequencies are chosen:
2 Hz and 16 Hz. The duration of time over which the signal is sampled is chosen to be 8 seconds.
The table below shows some of the key parameters for this example.
An increase in sampling frequency or duration will increase the size of the DFT which in turn
increases computational complexity. The following figures illustrate the effect of increasing the
sampling rate on the accuracy of the DFT. At the lower sampling rate of 2 Hz, the accuracy of the
DFT begins to degrade around 0.5 Hz due to aliasing. A sampling rate of 16 Hz results in a very
accurate sampled spectrum. The DFT in the second case actually ranges from –8 to 8 Hz but was
truncated to 2 Hz to allow a nice comparison with the lower sampling rate over the frequency range
of interest.
MATLAB Code:
% (a) Sf= 2 Hz, D=8 sec, N=16 samples
Sf=2; D=8; N=16;
t=0:1/Sf: (N-1)*1/Sf;
x=t.*exp(-t);
ya=fft(x,N);
ya=fftshift(ya);
fo=1/D; %Spectral Resolution;
fa=-(N/2)*fo:fo:(N/2-1)*fo;
subplot(2,1,1);
stem(fa,1/Sf*abs(ya));
title('1/Sf*DFT: Sf=2Hz and D=8 sec');
grid;
hold on;
plot(f,abs(xf));
hold off;
% (b) Sf=16 Hz; D= 8 sec; N=128 samples
Sf=16; D=8; N=128;
t=0:1/Sf: (N-1)*1/Sf;
x=t.*exp(-t);
yb=fft(x,N);
yb=fftshift(yb);
fo=1/D; %Spectral Resolution;
fb=-(N/2)*fo:fo:(N/2-1)*fo;
% Truncate the frequency range to 2 Hz.

% Find the entry in the frequency vector corresponding to -2 and +2 Hz:
f1= (-2-(-N/2*fo))/fo + 1; % (Desired - Start)/Increment + 1
f2=(2 - (-N/2*fo))/fo + 1;
subplot(2,1,2);
stem(fb(f1:f2),1/Sf*abs(yb(f1:f2)));
title('1/Sf*DFT: Sf=16Hz and D=8 sec');
grid;
hold on;
plot(f,abs(xf));
hold off;
Implement the Task:
1. Perform a 4-poiunt DFT of the number x(n)={1, 0, 0, 1}. Write down the complex values,
magnitude, and phase angle.
2. Proof your conversion by applying the inverse DFT of the sequence.
3. Now apply your written DFT function on your ECG signal at normal condition and after
exercise of duration 5 seconds each and plot the frequency spectrum (Freuency vs
Aplitude) in the frequency range 0-100 Hz. Apply IDFT to reconstruct your original ECG
from DFT coefficient and plot it.
4. Now apply your written DFT function on your EMG signal at normal condition and using
hand dynamometer for 1 cycle of data each and plot the frequency spectrum (Freuency vs
Aplitude). Apply IDFT to reconstruct your original EMG from DFT coefficient and plot it.
5. Apply DFT on Heart rate signal and integrated EMG and plot frequency spectrum.

Exp#02 Analysing Biomedical Signal Using DFT and Reconstruct The Signal Using IDFT

Uploaded by

Copyright:

Available Formats

Exp#02 Analysing Biomedical Signal Using DFT and Reconstruct The Signal Using IDFT

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Exp#02 Analysing Biomedical Signal Using DFT and Reconstruct The Signal Using IDFT

Uploaded by

Copyright:

Available Formats

BME 3112

Biomedical Signal Processing Laboratory

Basic Theoretical Principles

2. The Discrete Fourier Transform is very useful in computationally intensive filtering,

𝑥(𝑡) = ∫ 𝑋(𝑓)𝑒 𝑗2𝜋𝑓𝑡 𝑑𝑓

MATLAB program to find DFT without using Matlab function

%For 4 Point DFT

To check use matlab built in fft.m function

It will produce the same result as above

% Truncate the frequency range to 2 Hz.

2. Proof your conversion by applying the inverse DFT of the sequence.

You might also like