CSE 415: Signal Processing

Unit 4. DISCRETE FOURIER TRANSFORM (DFT)

Discrete Time Fourier Transform (DTFT), connections: DTFT and Fourier transform,
Properties of DTFT, Discrete Time Fourier Transform Formulas.

Introduction to Discrete Fourier Transform (DFT), Difference between FT and DFT,

Calculation of DFT & IDFT, Properties of DFT ,Circular Convolution, Linear convolution
of sequences using DFT, Computation of DFT, Relation between Z-transform and DFT.

Fast Fourier Transform (FFT), Radix-2 FFT Algorithms, Computational complexity (FFT vs
Direct computation), Bit Reversal, Decimation-In-Time (DIT) and Decimation-In-
Frequency (DIF) FFT Algorithms, Goertzel Algorithm, Exercises

Some Terminologies for refreshing:

Signals
• x(t) --continuous time signal
• x(n)—discrete time signal (sampled version of x(t))

Domains
• Time Domain (any naturally occurring signal is seen in time domain)
• Frequency Domain- (all Fourier transforms convert a signal whether continuous time
signal or discrete time signal to the frequency domain so that the spectrum of the signal
is got and can be analysed)
• z-domain (discrete time signal & system analysis in z-domain)
• s-domain (analog system analysis)-Laplace Transform

A time-domain graph shows how a signal changes over time, whereas a frequency-
domain graph shows how much of the signal lies within each given frequency band over a
range of frequencies. The "spectrum" of frequency components is the frequency-
domain representation of the signal

System Responses
• Impulse response of a system-h(n)
A system's impulse response (h(t) for continuous-time systems or h[n] for discrete-
time systems) is defined as the output signal that results when an impulse is applied to
the system input. Since the impulse function contains all frequencies, the impulse
response defines the response of a linear time-invariant system for all frequencies.
The impulse response of a system is important because the behavior of the system to any
arbitrary input can calculated from the system impulse response easily using a convolution
summation.
• Transfer function of a System-H(z)
If we take the z-transform of h(n) we get H(z). H(z) is a function of input and output. H(z)=
Y(z)/X(z). H(z) is also called as System function
• Frequency response of the system-H(ej ˆω)
Frequency response is the output spectrum of a system or device in response to a stimulus, and
is used to characterize the dynamics of the system. It is a measure of magnitude and phase of
the output as a function of frequency, in comparison to the input.
By analyzing the output spectrum, design of the system and adjusting the parameters of the
system can be easily carried out. Corrective measurement for noise disturbance generated in
the system and parameters variation can be easily determined using frequency analysis.
• Impulse response of a system h(n) is not same as frequency response H(ejˆω) of a
system. Impulse response is in time domain. Frequency response is in frequency
domain.
• The impulse response h[n] can be transformed into the frequency response H(ejˆω)
mathematically using various Transforms.

Transforms
Most common transforms, and the fields in which they are used:
• Fourier series – periodic signals, oscillating systems.
• Fourier transform – non-periodic signals, transients.
• Laplace transform – electronic circuits and control systems.
• Z transform – discrete-time signals, digital signal processing systems.
• Wavelet transform — image analysis, data compression.
The above transforms can be interpreted as capturing some form of frequency, and hence the
transform domain is referred to as a frequency domain.

Signals and Transform

Any signal (input or output to a system) can be transformed in frequency domain for analysis
so that a proper system can be designed to handle the signals for processing.

Systems and Transform

Any system’s response (commonly the impulse response) can be transformed and analysed to
understand the system characteristics or behaviour for a signal input. Based on the analysis if
needed the system design can be altered for better performance to a signal.

Introduction
Need for Frequency Analysis

of a given signal using the basic mathematical tools described in this part is known as frequency
or Spectral Analysis.

Time domain Analysis provides some information like amplitude at sampling instant but does
not convey frequency content & power, energy spectrum hence frequency domain analysis is
used.

FIGURE 1 | A 1-Hz sinusoid. In the spectrum, the signal power concentrates at 1 Hz.

Spectral analysis is an important aspect of signal processing. The strength or amplitude of the
sinusoids are displayed as a function of frequency called the spectrum in the frequency domain.
Any signal can be decomposed in terms of sinusoidal (or complex exponential function ejwt)
components. – This is the essence of transforms, and it is how we convert from one domain to
another. Sinusoids form the “building blocks” of signals in frequency domain.

Figure 2. Sinusoids

Thus, the analysis of signals can be done by transforming time domain signals into frequency
domain and vice-versa. This transformation between time and frequency domain is performed
with the help of Fourier Transform (FT) for continuous-time signals. Analog systems process
analog signals (Continuous Time, Continuous Amplitude-CTCA) directly. Example:
photocopiers, old land-line telephones, audio tapes, old televisions, VCRs (same as TV).
Analog signal is represented as x(t).

But it is not convenient for computation by DSP processors as FT is for continuous-time.

Hence, we go for Discrete Fourier Transform (DFT) for discrete time signals.

Modern electronic products such as computers and mobile phones depend on digital signals.
Digital signal is represented as x(n). Sampling and Quantization converts CTCA signal to
Discrete Time Discrete Amplitude Signal (DTDA).

What are the various mathematical tools Fourier analysis provides for determining the
frequency content of various time-domain signals?

Fourier analysis converts a signal from its original domain (often time or space) to a
representation in the frequency domain and vice versa.

Below figure shows which Transform to choose for which type of signals.

Fourier series (FS)

The basic mathematical representation of periodic signals is the Fourier series which is a linear
weighted sum of harmonically related sinusoids or complex exponentials. The amplitude and
phase of a sinusoid can be combined into a single complex number, called a Fourier coefficient.
The weights or the coefficients Cn are one to one mapping of the original input. The Fourier
series is for periodic signals and spectrum of Fourier series is always discrete. There is always
a duality present between periodicity and discreteness, or periodicity and discreteness are
duals.
Figure 2: One cycle of a 5-Hz sinusoid repeating every 1 s. The spectrum of the signal shows power at
multiple frequencies, at both the fundamental frequency, i.e., 1 Hz, and harmonics, i.e., multiples of 1 Hz.
The strongest power appears at the 4th harmonic rather than at the fundamental frequency.

FIGURE 3 | A 20-Hz sinusoid amplitude modulated at 1 Hz. The dashed black curve shows the envelope
and the gray curve shows the waveform. The 1-Hz envelope imposes a clear 1 Hz rhythm in how the signal
power fluctuates in time. The spectrum of the modulated signal, however, shows no power at 1 Hz but instead
power at 20 Hz and 20 ± 1 Hz.

Fourier Transform (FT)

Continuous aperiodic signal input and the spectrum
Discrete Time Fourier Transform (DTFT)
This is for discrete Aperiodic signals. DTFT describes discrete signals by periodic spectra.

But, to be noted, the Fourier series describes periodic signals by discrete spectra.

These results are a consequence of the fact that sampling in one domain induces periodic
extension in the other domain.

Fourier Transform Techniques used for discrete signal Analysis:

Assume that x(t), shown in Figure 1, is the continuous-time signal that we need to analyze.

Figure 1. A continuous-time signal for which we need to determine the frequency content

Obviously, a digital computer cannot work with a continuous-time signal and we need to take
some samples of x(t) and analyze these samples instead of the original signal. Moreover, while
Figure 1 shows only the first 5 millisecond of the signal, x(t) may continue for hours, years, or
more. Since our digital computer can process only a finite number of samples, we have to make
an approximation and use a limited number of samples. Therefore, generally, a finite-duration
sequence is utilized to represent this analog continuous-time signal which may extend to
positive infinity on the time axis. Now we may wonder how many samples, L, we need in order
to estimate the frequency content of a given signal. We will see that later. For the time being,
assume that we sample x(t) in Figure 1 with a sampling rate of 8000 samples/second and take
only L=8 samples of this signal. The result is shown in red in Figure 2.
Figure 2. Sampling allows us to analyze continuous-time signals in a digital computer.

!
If we normalize the time axis to the sampling period Ts=" , we will obtain the discrete
!
sequence x(n), representing x(t), and let the values be as follows from the figure above:

n 0 1 2 3 4 5 6 7
x(n) 0.2165 0.8321 0.7835 0.5821 0.2165 −0.5821 −1.2165 −0.8321

There are only two techniques from the Fourier analysis family which target discrete-time
signals x(n):
(i) the discrete-time Fourier transform (DTFT) and
(ii) the discrete Fourier transform (DFT).

(i) DISCRETE TIME FOURIER TRANSFORM:

First, Let us see about Discrete-Time Fourier Transform (DTFT):

The DTFT of an input sequence, x(n), is given by

We can use Equation 1 to find the spectrum of a finite-duration signal x(n).

According to equation 1, the spectrum is got by multiplying each sample of x(n) with a
sinusoidal function ejwn and summing all together.

Figure below shows the discrete input x(n) and its DTFT X(w) or X(ejw)
Therefore, the inverse of the DTFT is given by

However, it should be noted that X(ejω) given in the above DTFT Equation 1 is a continuous
function of ω as shown in the above figure.

Hence, a digital computer cannot directly use Equation 1 to analyze x(n) using X(ejω) and re-
construct back the x(n). This is the drawback of DTFT.

This is why we go for the second technique called DFT for discrete signal analysis instead
of DTFT.

(ii) DISCRETE FOURIER TRANSFORM

Let us see the second technique Discrete Fourier Transform (DFT). DFT can be formulated in
two different ways:
o The idea of sampling X(ejω) at equally spaced frequency points forms the basis in the
first approach. (Note that this sampling is taking place in the frequency domain as
X(ejω) is a function of frequency).
o Signals that are both discrete and periodic in one domain are also periodic and discrete
in the other. This forms the basis for the formulation of the DFT in the second approach.

Mathematical Formulation of DFT based on the two different ways:

As explained earlier this is done following two approaches:
1) Frequency-Domain Sampling and Reconstruction of Discrete-Time Signal x(n)
2) Deriving the DFT Equations using periodicity property

1) Frequency-Domain Sampling and Reconstruction of Discrete-Time Signals

The DTFT equation of x(n) given in equation 1 of the previous section is written below:

#$
Therefore, w is replaced by %
∗𝑘

"#$∗&
Once sampled, X(𝑒 ' ) can be analyzed by the computer.
Before sampling x(n) is as in equation 2.

"#$∗&
Now after sampling of X(w) we have frequency domain samples X(𝑒 ' ). Therefore, to get
x(n) from frequency domain samples the inverse formula is,

"#$∗&
and X’(k) = X(𝑒 ' ) .
L is the number of samples of x(n)
"#$∗&
Equation 3, shows that selecting a set of samples, denoted by X(𝑒 ' ) or X’(k) taken from
X(ejω) or X(w) in the frequency domain and summing all samples together, the original time
domain signal x(n) can be got.

Equation 3, if expanded is equivalent to the following set of equations. We get n=L-1 equations
starting from n=0, as we selected only L samples from x(n).

The above set of Equations 4 can be solved for the unknowns X’(k) using matrix form in
MATLAB.
 𝑋 & (0) '#$( 𝑥(0)
& 𝑒 % ⋯ 𝑥(1)
& 𝑋… (1) . = &
⋮ ⋱ ⋮ .∗& … .
𝑋 & (𝑁 − 1) ⋯ 𝑥(𝐿 − 1)

Example:
Let’s proceed with the example started in the previous section. The discrete signal x(n) is given
below:

n 0 1 2 3 4 5 6 7
x(n) 0.2165 0.8321 0.7835 0.5821 0.2165 −0.5821 −1.2165 −0.8321

We have an L-sample-long sequence x(n), representing the analog continuous-time signal x(t).
Once frequency Transformation of x(n) is done using DTFT , then the goal is to find a set of
sinusoid samples in the frequency domain which can be added together to produce back the
x(n).

In this example, the no of input signal x(n) samples are 8. Therefore, L=8
We assume, the number of samples taken by sampling X(ejω) is 8, therefore N=8
"#$(')*)(,)*)
Given the values of L and N=8, the complex exponential values 𝑒 ' in the set of
Equations 4 can be easily calculated. Refer below the set of equations for L=N=8.

x(0) = X’(0)+ X’(1) + X’(2) +X’(3) + X’(4) +X’(5) +X’(6) + X’(7)

"#$ "#$(#.*) "#$(/.*) "#$(0.*)
x(1) = X’(0)+ X’(1) 𝑒 - +X’(2) 𝑒 - + X’(3) 𝑒 - +….X’(7) 𝑒 -

.
.
.
"#$0 "#$(#.0) "#$(/.0) "#$(0.0)
x(7) = X’(0)+ X’(1) 𝑒 - +X’(2) 𝑒 - + X’(3) 𝑒 - +….X’(7) 𝑒 -

As x(n) values are given in this example as below,

n 0 1 2 3 4 5 6 7
x(n) 0.2165 0.8321 0.7835 0.5821 0.2165 −0.5821 −1.2165 −0.8321
we can find the values of the coefficients X’(0) to X’(7) , which are the 8 unknowns from the
above set of equations using the MATLAB.

The calculated numerical values of X’(k) using MATLAB are given below:

Just for simplification, now, consider the last equation from the expanded version of Equation
3 which is the set of Equations 4. Means we take the equation for the last sample (i.e) for k=
L-1, which is x(L-1) given as

Using the X’(k) coefficients which has values and omitting the exponential term with zero
value for coefficients X’(k), we obtain

Now applying the numerical values of X’(k) we obtain,

"#$ 2 )"#$ 2 "3$ 2 )"3$ 2 "3$ 2
x(n) = [-0.5j 𝑒 - + 0.5j 𝑒 - ] + [0.1083𝑒 - +0.1083𝑒 - ] + [-0.0625j 𝑒 - +0.0625j
)"3$ 2
𝑒 - ]

Rearranging and multiplying numerator and denominator by 2, we get

"#$ 2 )"#$ 2 "3$ 2 )"3$ 2 "3$ 2
x(n) = [ 2*0.5j (𝑒 - -𝑒 - )/2] + [2*0.1083 (𝑒 - +𝑒 - )/2] + [2 *0.0625j( 𝑒 - -
)"3$ 2
𝑒 - )/2]

Now applying the trigonometric simplification and applying the Euler’s formula for
simplification,

we get,
Using this equation 5, we can reconstruct all the 8 samples of x(n). (i.e) from x(0), x(1), x(2),
x(3)…..x(7), for n=0 to n=7.

Also equation 5 shows that x(n) can be represented by two frequency components, one at the
normalized frequency 2p/8 and other at 4p/8 .

there is one point which needs further attention. The analysis started with a sequence of length
eight which is shown in Figure below.

Figure. The analysis started using only these eight samples.

Clearly, we know the value of x(n) for n=0,…,7, but we don’t know x(n) outside of this range.
The above example clearly shows that the analysis is actually looking for a sum of complex
exponentials which can reproduce the values of x(n) for n=0,…,7.

But, if we examine Equation 5 outside of this range, we see x(n) will have periodic repetition
of x(n) as shown in figure 4 below. We can call it p(n).

Figure 4. The analysis leads to p(n)which is the periodic form of x(n).

p(n), shown in Figure 4, is a periodic function with N=8. As shown in this figure, the values of
the original x(n) will be repeated every 8 samples.

This proves the common property of the Transforms, that discreteness in any domain leads to
periodicity in the other domain. Here as we sampled the DTFT signal in frequency domain, the
Inverse DFT to get x(n) from frequency domain samples gave the periodicity of x(n).

The above property allowed to develop the next approach of mathematically formulating the
DFT.

2) Deriving the DFT Equations using periodicity property

Note that the periodic behaviour of p(n) can also be understood by recalling that we are
sampling X(ejω), the DTFT of x(n), in the frequency domain. We discussed already how
sampling a signal in the time domain leads to replicas of the original signal in the frequency
domain. We observe a similar phenomenon here: sampling X(ejω) in the frequency domain
leads to replicas of x(n) in the time domain. The periodic form of x(n) is shown as p(n) in
Figure 4 above.

This second method for calculating the spectrum of a finite-duration sequence is simple and
intuitive. It clarifies the inherent periodic behavior of DFT representation. However, it is
possible to use the above discussion and derive closed-form DFT equations without the need
to calculate the inverse of a large matrix.
o To this end, we only need to make a periodic signal out of the N samples of the
finite-duration sequence x(n).
When developing this always remember that we are achieving a representation of the finite-
duration sequence using a periodic sequence, where the values in one period of this periodic
sequence are equal to those in the finite-duration sequence.
o Then, applying the discrete-time Fourier series expansion, we can find the
frequency domain representation of the periodic signal.
o The obtained Fourier series coefficients ak are the same as the DFT coefficients
X’(k)s except for a scaling factor.

Assume that the finite-duration sequence that we need to analyze is as shown in Figure 5 (a).
To calculate the N-point DFT, we need to make a periodic signal, p(n), from x(n) with
period N, as shown in Figure 5(b).

Considering the fact that p(n)=x(n) for n=0,1,…,N−1, we obtain the discrete-time Fourier
series of this periodic signal.

The frequency domain signal is represented by

Where N denotes the period of the signal.

The time-domain signal can be obtained as follows:

Now to obtain the actual DFT we have to multiply Equation 6 by N to remove the periodicity
effect introduced by us and hence divide by N to Equation 7.

Equation 9
Please note that while the discrete-time Fourier series of a signal is periodic, the DFT
coefficients, X(k), are a finite-duration sequence defined for 0 ≤ k ≤ N−1.

Summary of this method

• The DFT is one of the most powerful tools in digital signal processing; it enables us to
find the spectrum of a finite-duration signal x(n).
• Basically, computing the DFT is equivalent to solving a set of linear equations.
• The DFT provides a representation of the finite-duration sequence using a periodic
sequence, where one period of this periodic sequence is the same as the finite-duration
sequence. As a result, we can use the discrete-time Fourier series to derive the DFT
equations.

DFT/ IDFT pair (Assuming L=N)

Zero Padding, Aliasing and DFT Resolution

The relationship in equation (9) provides the reconstruction of the periodic signal x(n) from
the samples X(k) of the spectrum X(ejw). However, it does not imply that we can perfectly
recover X(ejw) or x(n) from samples. To accomplish this we need to consider the relationship
between p(n) denoted by xp(n) & x(n). Since, xp(n) is the periodic extension of x(n), it is clear
If more than equally spaced frequency samples of a length- signal are desired, they can easily
be obtained by zero-padding the discrete-time signal and computing a DFT of the longer length.
Note that zero-padding interpolates the spectrum. One should always zero-pad (by about at
least a factor of 4 when using the DFT to approximate the DTFT to get a clear picture of the
DTFT.

What is zero padding in DFT and in DSP?

Zero padding consists of extending a signal (or spectrum) with zeros. It is a technique
typically employed to make the size of the input sequence equal to a power of two. In zero
padding, you add zeros to the end of the input sequence so that the total number of samples is
equal to the next higher power of two. By appending artificial zeros to the signal, we obtain a
denser frequency grid when applying the DFT.

But, the number of samples taken from the input signal and not the zero padding decides the
frequency resolution of the spectrum obtained from DFT.

Does zero padding alter the frequency resolution?

Answer: No
Zero padding enables you to obtain more accurate amplitude estimates of resolvable signal
components. On the other hand, zero padding does not improve the spectral (frequency)
resolution of the DFT. The resolution is determined by the number of samples N and the
sampling rate T.
The DFT as a Linear Transformation
We know the DFT & IDFT formulas given below as in equation (8) and equation (9).
These may be expressed as given below,
Here N=4; W4kn ; k= 0 to 3 (row); n= 0 to 3 (column)

Refer Aside and calculate the values:

N=4
W4kn = 𝑒 )*(#,/.)01 =𝑒 )*(,/#)01
Apply, k= 0 to 3 (row); n= 0 to 3 (column)

The different values of W4kn are calculated as shown below:

If k=0 along row and n=0,1,2,3, or for n=0 along column, and k=0,1,2,3; the result is 0.
Because,
W04 = 𝑒 )*(,/#)∗3 =𝑒 3 = 1
Thus, the values of first row and column of W4kn matrix are 1

If k=1 and n=1,2,3;

𝑒 )*(,/#)01
k=1 and n=1; 1*1 =1
W14 = 𝑒 )*(,/#)∗! = 𝑒 )*(,/#) = -j

k=1 and n=2; 1*2 =2

W24 = 𝑒 )*(,/#)∗# = 𝑒 )*(#,/#) =𝑒 )*(,) = -1

K=1 and n=3; 1*3 =3

W34 = 𝑒 )*(,/#)∗4 = 𝑒 )*(4,/#) = +j (270 degrees)

The above is same for

k=2 and n=1,2,3
k=3 and n=1,2,3
These values should be calculated around a circle and based on that it repeats.

[Aside…

Euler Equation

Euler Identity
If we substitute the value into Euler's equation, then we get:

If we substitute to Euler's equation, then we get:

Note:
While you do the calculation for using the Euler identity, the negative sign in it
should be considered.
Aside ends…]

The DFT and IDFT are computational tools that play a very important role in many digital
signal processing applications, such as frequency analysis (spectrum analysis) of signals,
power spectrum estimation, and linear filtering. The importance of the DFT and ID FT in such
practical applications is due to a large extent on the existence of computationally efficient
algorithms, known collectively as fast Fourier transform (FFT) algorithms, for computing the
DFT and IDFT.

PROPERTIES OF THE DFT

In this section we present the important properties of the DFT. A good understanding of
these properties are extremely helpful in the application of the DFT to practical problems.

(i) Periodicity, Linearity, and Circular Symmetry Properties

 Circular Symmetry / Shift.

The circular shift equivalent to the above linear shift is as follows:

(ii) Multiplication of Two Sequences

1. Using Circular Convolution

X3(k) = X1(k) * X2(k)

Where X1(k) is the DFT of the input sequence x1(n) and X2(k) is the DFT of the input
sequence x2(n).

The expression in (5.2.39) has the form of a convolution sum. However, it is not the
ordinary linear convolution that we studied earlier, which relates the output sequence y(n)
of a linear system to the input sequence x(n) and the impulse response h(n).
Instead, the convolution sum in (5.2.39) involves the index ((m-n))N and is called circular
convolution.

Thus, we conclude that in the frequency domain it is the multiplication of the DFT’s of two
input sequences and is equivalent to the circular convolution of the two input sequences in
the time domain.
The following example illustrates the operations involved in circular convolution.

Example:
From this example, we observe that circular convolution involves basically the same four
steps as the ordinary linear convolution: Given two sequences
o folding (time reversing) one sequence,
o shifting the same folded sequence,
o multiplying the first sequence and the folded/shifted sequence to obtain a product
sequence, and
o finally, summing the values of the product sequence.
The basic difference between these two types of convolution is that, in circular
convolution, the folding and shifting (rotating) operations are performed in a circular
fashion by computing the index of one of the sequences modulo N. In linear convolution,
there is no modulo N operation.

Note: More examples on circular convolution is given at the end of this handouts.

2. Using the DFT of the sequences

Example:
By means of the DFT and IDFT, determine the sequence x3(n) corresponding to the
circular convolution of the sequences x1(n) and x2(n) given in the previous
(…

)
(iii) Additional DFT Properties

Hence reversing the N-point sequence in time is equivalent to reversing the DFT
values.
Applications of DFT&FFT
o DFT plays an important role in many applications of digital signal processing, including
linear filtering, correlation analysis, and spectrum analysis.
o They can be used in applications like image, video or audio compressing. Image
processing, Digital filtering, computation of convolution and so on.
o Let’s take an image which when converted from spatial domain to frequency domain,
it becomes narrower (it gets compact). Natural images contain most of the information
in low pass region in frequency domain. There will be more number of repeating zeroes
and low values. This can be encoded. Also, some of the high frequency components
can be discarded. This is what is used in image/video compression. For example: JPEG
o There are many circumstances in which we need to determine the frequency content of
a time-domain signal. For example, we may have to analyze the spectrum of the output
of an LC oscillator to see how much noise is present in the produced sine wave. This
can be achieved by the discrete Fourier transform (DFT).
 o One more advantage in frequency domain is convolution becomes multiplication which
is easier. Convolution is used in lot of applications.
o For an Engineer Power, Size and Speed are the bottlenecks. Now DFT/FFT can reduce
size and make things faster. This shows how powerful tool it is for an Engineer.

Need for other Efficient Computation methods of DFT:

Computation Complexity of Direct computation of DFT

For the efficient computation of DFT two different approaches are used;
1) First is

2)

****************

Circular Convolution more Examples:

Circular Convolution: (Multiplication of two DFT sequences can be achieved by the circular
convolution of time domain sequences)

Examples:
Circular shift of a sequence is defined using a modulo operation as show below:
Concept of circular shift in terms of discrete data sequences:

For example, for a a discrete signal with N=4,

Concept of circular shift in terms of concentric circles:

Example 1 :
Perform circular convolution of the given two sequences.

The procedure in diagrammatic form is given below:

Example 2:
Convolve the two sequences x(n) ={2,0,3,-1} and h(n)={10,20,30,40}
Illustration of Circular convolution for N=8 in reverse shifting:
Example 3:

Same Example 3 in Tabular form

Circular versus Linear Convolution:

Example 4
Find the Circular convolution and linear convolution of the sequences x(n)=[ 1 2 3 ] and
h(n)= [1 2 1]
Example 5:
Convolve the two 4 point sequences g(n)={1 2 0 1} and h(n) = {2 2 1 0}. Compare it with
Linear Convolution output.

Example 6:
Circular convolution of a finite-length sequence x2[n] with a single delayed impulse,
x1[n]=d[n-1]
Difference between FT and DFT

DTFT versus DFT

DTFT DFT
DTFT (Discrete Time Fourier DFT is the sampled version of DTFT spectrum.
Transform) is nothing but a Fourier
transform of a discrete sequence.
DTFT is difficult to evaluate on a So to make the evaluation of the DTFT possible on
computer, since a computer works only a computer, we choose a finite number of
on finite number of points. frequency points. This is equivalent
to sampling the Fourier transform at a certain
number of points. This is called the Discrete
Fourier Transform (DFT).
DTFT of a discrete time signal is DFT is basically the sampled version of the DTFT.
continuous. When signal becomes As the input taken (the number of samples taken)
discrete in time domain its spectrum is made periodic, with period N in time domain,
(DTFT) becomes continues and the spectrum (DFT) is discrete and finite.
periodic with period 2p. Periodic and Discrete are dual
Discrete and periodic are dual

Common Transforms Summary

i) Fourier methods (FT, FS, DTFT, DFT, FFT)
ii) Laplace transforms
iii) z-transforms
Sl.No. Transform Signals/Systems Description
Fourier Fourier methods are used for two primary purposes:
Methods Fourier methods are natural tools for understanding and
modeling the effects of the physical world on signals, and for
designing and characterizing common signal processing systems.
1.
Mainly focusses on Signal analysis.
Fourier • Analyze problems Converts a continuous (analog)
Transform (FT) involving continuous- signal from time domain to
time signals frequency-domain.
 • Mathematical analysis
tool and cannot be Describes signals as weighted
evaluated exactly in a combinations of continuous-
computer time complex-valued sinusoids.
• Is used in a wide range of
applications, such as
image analysis, image
filtering, image
reconstruction and image
compression.
Fourier Series • Analyze problems Converts periodic signal
(FS) involving Periodic signals (analog) from time domain to
• Broadly used in frequency domain.
telecommunications The fundamental frequency f0 =
system, for modulation 1/T0, the period of the signal, and
and demodulation of all other components are called
voice signals, also the harmonics, and they are integral
input, output and multiples of f0.
calculation of pulse and
their sine or cosine graph.
Discrete Time • Analyze problems Describes signals as weighted
Fourier involving discrete-time combinations of discrete-time
Transform signals complex sinusoids.
(DTFT) • Mathematical analysis
tool and cannot be
evaluated exactly in a
computer
Discrete Fourier • A computational signal Extracts a portion of a discrete
Transfer (DFT) processing tool used time signal to analyse, this is
solely for numerical done by effectively multiplying
analysis of data. the signal with a rectangular
• Handles discrete-time window.
signals
Fast Fourier • It is not a distinct Fourier A faster method of evaluating
Transform method, but is an efficient DFT
(FFT) computational technique
for evaluating the
Discrete Fourier
transform.

2. Laplace • Used to analyse The Laplace transform is used to

Transform Continuous Time Signals map the time domain (t-
as well as Systems. domain) representation to
• Laplace is used for frequency domain (S-domain),
stability studies in representation.
designing a system and
Fourier is used for
sinusoidal responses of
systems.
 • Is used extensively in
mechanical engineering
and electrical
engineering.

3 Z-Transform The z-transform is used to map

Used to analyse Discrete the time domain (t-domain)
Time Signals as well as representation to
Systems. frequency domain (z-domain),
Testing stability of systems. representation.