Scribd
Upload
22 views

Introduction To DFT: Deployment of Telecommunication Infrastructures

The document provides an introduction to discrete Fourier transforms (DFT). It discusses: 1) How DFT can be used to calculate the discrete frequency spectrum of a discrete-time signal by taking samples of its discrete-time Fourier transform (DTFT). 2) Key properties of DFT including linearity, circular shift, and duality with circular convolution. 3) How to calculate DFT and its inverse using functions like fft and ifft in Octave/Matlab. Linear convolution can also be computed using fft by zero-padding and multiplying frequency responses.

Uploaded by

Abhilash
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
22 views

Introduction To DFT: Deployment of Telecommunication Infrastructures

The document provides an introduction to discrete Fourier transforms (DFT). It discusses: 1) How DFT can be used to calculate the discrete frequency spectrum of a discrete-time signal by taking samples of its discrete-time Fourier transform (DTFT). 2) Key properties of DFT including linearity, circular shift, and duality with circular convolution. 3) How to calculate DFT and its inverse using functions like fft and ifft in Octave/Matlab. Linear convolution can also be computed using fft by zero-padding and multiplying frequency responses.

Uploaded by

Abhilash
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 16

Introduction to DFT

Deployment of Telecommunication
Infrastructures

Azadeh Faridi
DTIC – UPF, Spring 2009

1
Review of Fourier Transform

 Many signals can be represented by a fourier integral of the


following form:
(1)

 Where X(jw) is given by


(2)

➤ (2) is the inverse Fourier Transform equation


➤ (1) is the Fourier Transform equation
 Note that X(jw) is called the spectrum or the frequency domain
representation of x(t)
 Notation: F{x(t)} = X(jw) OR x(t) ← F.T. → X(j!)
2
LTI, Continuous Time Systems

 For an LTI (Linear Time-Invariant) continuous-time


system, with the impulse response h(t) as shown
below
x(t) h(t) y(t)
 y(t) is given by

➤ which is called the convolution of x and h


 because of F.T. properties, in frequency domain we
will have:

➤ analysis in frequency domain can be a lot easier 3


Fourier transform properties

Signal Fourier transform


ax(t)+by(t) aX(jw)+bY(jw)

x(t{¿) e{j!¿ X(jw)

ej'tx(t) X(j(w{'))

x(at) 

x(t)∗y(t) X(jw)Y(jw)

x(t)y(t)
x(t) Real X(jw) = X∗({jw) 4
Fourier transform properties

Signal Fourier transform


ax(t)+by(t) aX(jw)+bY(jw)

x(t{¿) e{j!¿ X(jw)

ej'tx(t) X(j(w{'))

x(at) 

x(t)∗y(t) X(jw)Y(jw)
∣X(jw)∣ = ∣X({jw)∣
x(t)y(t) ∢X(jw) = {∢X({jw)
x(t) Real X(jw) = X∗({jw) 5
Basic Fourier Transform Pairs

Signal: x(t) Fourier transform: X(jw)

6
Basic Fourier Transform Pairs

Signal: x(t) Fourier transform: X(jw)

-T T

7
Calculating FT on computers

 We can't numerically represent c.t. signal x(t) on


computer (need infinite points)
➤ ⇨ use sampling and convert to d.t. signal x[n]
➤ ⇨ use Discrete Time FT (DTFT) on x[n]
 DTFT:

 NOTE: X (ejw) is periodic in frequency, with period 2¼


➤ Low frequencies around w = 0

➤ High frequencies around w = ¼

8
LTI, Discrete Time Systems

 For an LTI (Linear Time-Invariant) discrete-time


system, with the impulse response h[n] as shown
below
x[n] h[n] y[n]
 y[n] is given by

➤ which is called the convolution of x and h


 Similarly to c.t. case, in frequency domain we have:

9
Convolution Properties

 x(t) ∗ ±(t { to) = x(t – to) (Similarly for DT systems)


➤ What is x(t) ∗ ±(t)?

 continuous-time, finite duration signals:


➤ If
Duration(x(t))= Tx and Duration(y(t))= Ty
➤ then
⇨ Duration(x(t) ∗ y(t))= Tx + Ty

 discrete-time, finite duration signals:


➤ If
Duration(x[n]))= Nx and Duration(y[n])= Ny
➤ then
⇨ Duration(x[n] ∗ y[n])= Nx + Ny – 1
10
DTFT on computer?

 x[n] is representable on computer


 but X (ejw) is still a continuous function of w
➤ can't be represented on computer

 ⇨ Use DFT (Discrete Fourier Transform)


➤ based on Fourier series
➤ works on finite duration signals. Basically,
◆ take a finite duration sequence (of length N),
◆ make it periodic,
◆ calculate the Fourier series → N values
➤ generates a discrete frequency response
➤ for N points need N Fourier samples

11
DFT

 For a signal x[n] of duration N

➤ Where
 DFT is basically a sampling of DTFT in frequency

 It is derived using the Fourier series of the periodic


version of x[n] 12
DFT in Octave/Matlab

 DFT can be calculated in Octave using the fft function


➤ fft(x,N0)
➤ x is the signal
➤ N0 is the number of points in the DFT
◆ It works in the same way as appending zeros to your signal
◆ larger N0 will give nicer graphs
◆ but N0 = N is enough to get correct inverse fft
◆ Be careful when choosing N0!!!
➤ ifft is the Octave function for calculating the inverse DFT
➤ by default fft generates the spectrum in [0,2*pi], but we are
usually interested in [-pi,pi]. use fftshift for this!
 Exercise:
➤ x[n] = [1 1 1 1]
➤ calculate the fft of x with different values of N0 and plot the
abs() of the frequency responses on the same plot.
13
Compare the results.
DFT Properties

 Linearity

➤ N3 = max [N1,N2]

 Circular shift

➤ This means that


 Duality

14
DFT properties - continued

 Circular convolution

➤ where

◆ for
➤ This is because a shift in the periodic version of x[n] is
equivalent to a circular shift in x[n]
 Note: (linear convolution)
➤ to calculate linear convolution using fft and ifft you need to
use a trick! (see next page)
15
Linear convolution using fft

 Let
➤ x1[n] be of length N1
➤ x2[n] be of length N2
 We can calculate the linear convolution of two signals by
multiplying their frequency responses and then taking the
inverse FT. To do this properly using fft function, you need to
choose N0 carefully, as follows:
➤ X1 = fft(x1,N1+N2-1)

➤ X2 = fft(x2,N1+N2-1)

➤ x1_conv_x2 = ifft(X1.*X2)

 Exercise: let x1 = x2 = [1 1 1 1]. calculate conv(x1, x2) directly


and using the method described above. You should get the
same result using either method.
16

You might also like