Lecture Notes:

Subject: Digital Signal Processing

UNIT I
FAST FOURIER TRANSFORM
Introduction to DSP :
A signal is any variable that carries information. Examples of the types
of signals of interest are Speech (telephony, radio, everyday communication),Biomedical signals
(EEG brain signals), Sound and music, Video and image,_ Radar signals (range and bearing).
Digital signal processing (DSP) is concerned with the digital representation of signals and the
use of digital processors to analyse, modify, or extract information from signals.Many signals in
DSP are derived from analogue signals which have been sampled at regular intervals and
converted into digital form. The key advantages of DSP over analogue processing are
Guaranteed accuracy (determined by the number of bits used), Perfect reproducibility, No drift
in performance due to temperature or age, Takes advantage of advances in semiconductor
technology, Greater exibility (can be reprogrammed without modifying hardware),Superior
performance (linear phase response possible, and_ltering algorithms can be made adaptive),
Sometimes information may already be in digital form.There are however (still) some
disadvantages, Speed and cost (DSP design and hardware may be expensive,especially with high
bandwidth signals) Finite wordlength problems (limited number of bits may cause degradation).
Application areas of DSP are considerable: _ Image processing (pattern recognition, robotic
vision, image enhancement, facsimile, satellite weather map, animation), Instrumentation and
control (spectrum analysis, position and rate control, noise reduction, data compression) _
Speech and audio (speech recognition, speech synthesis, text to
speech, digital audio, equalisation) Military (secure communication, radar processing, sonar
processing, missile guidance) Telecommunications (echo cancellation, adaptive equalisation,
spread spectrum, video conferencing, data communication) Biomedical (patient monitoring,
scanners, EEG brain mappers,ECG analysis, X-ray storage and enhancement).
Discrete-time signals
A discrete-time signal is represented as a sequence of numbers:
Here n is an integer, and x[n] is the nth sample in the sequence. Discrete-time signals are often
obtained by sampling continuous-time signals. In this case the nth sample of the sequence is
equal to the value of the analogue signal xa(t) at time t = nT:
The sampling period is then equal to T, and the sampling frequency is fs = 1=T .
x[1]

For this reason, although x[n] is strictly the nth number in the sequence, we often refer to it as
the nth sample. We also often refer to \the sequence x[n]" when we mean the entire sequence.
Discrete-time signals are often depicted graphically as follows:

Discrete-time systems
A discrete-time system is de_ned as a transformation or mapping operator that maps an input
signal x[n] to an output signal y[n]. This can be denoted as

Example: Ideal delay

Linear time-invariant systems

If the linearity property is combined with the representation of a general sequence as a linear
combination of delayed impulses, then it follows that a linear time-invariant (LTI) system can be
completely characterised by its impulse response. Suppose hk[n] is the response of a linear
system to the impulse h[n -k]
at n = k. Since

If the system is additionally time invariant, then the response to _[n -k] is h[n -k]. The previous
equation then becomes

This expression is called the convolution sum. Therefore, a LTI system has the property that
given h[n], we can _nd y[n] for any input x[n]. Alternatively, y[n] is the convolution of x[n] with
h[n], denoted as follows:

The previous derivation suggests the interpretation that the input sample at n = k, represented
by
is transformed by the system into an output sequence. For each k, these
sequences are superimposed to yield the overall output sequence: A slightly di_erent
interpretation, however, leads to a onvenient computational form: the nth value of the output,
namely y[n], is obtained by multiplying the input sequence (expressed as a function of k) by the
sequence with values h[n-k], and then summing all the values of the products x[k]h[n-k]. The
key to this method is in understanding how to form the sequence h[n -k] for all values of n of
interest. To this end, note that h[n -k] = h[- (k -n)]. The sequence h[-k] is seen to be equivalent to
the sequence h[k] reected around the origin

Since
sequences are non-overlapping for all negative n, the output must be zero y[n] = 0; n < 0:

the

The Discrete Fourier Transform

The discrete-time Fourier transform (DTFT) of a sequence is a continuous function of !, and
repeats with period 2_. In practice we usually want to obtain the Fourier components using
digital computation, and can only evaluate them for a discrete set of frequencies. The discrete
Fourier transform (DFT) provides a means for achieving this. The DFT is itself a sequence, and
it corresponds roughly to samples, equally spaced in frequency, of the Fourier transform of the
signal. The discrete Fourier transform of a length N signal x[n], n = 0; 1; : : : ;N -1 is given by

An important property of the DFT is that it is cyclic, with period N, both in the discrete-time and
discrete-frequency domains. For example, for any integer r,

DFT only depends on samples in the interval 0 to N Since

1, it is implicitly assumed that the signals repeat with period N in both the time and frequency
domains. To this end, it is sometimes useful to de_ne the periodic extension of the signal x[n] to
be To this end, it is sometimes useful to de_ne the periodic extension of the signal x[n] to be
x[n] = x[n mod N] = x[((n))N]: Here n mod N and ((n))N are taken to mean n modulo N, which
has the value of the remainder after n is divided by N. Alternatively, if n is written in the form n
= kN + l for 0 < l < N, then n mod N = ((n))N = l:

Linear Convolution:
Linear convolution of two finite-length sequences Consider a sequence x1[n] with length L
points, and x2[n] with length P points. The linear convolution of the sequences,

Therefore L + P 1 is the maximum length of x3[n] resulting from the

linear convolution. The N-point circular convolution of x1[n] and x2[n] is

It is easy to see that the circular convolution product will be equal to the linear onvolution
product on the interval 0 to N 1 as long as we choose N - L + P +1. The process of augmenting
a sequence with zeros to make it of a required length is called zero padding.
Fast Fourier transforms
The widespread application of the DFT to convolution and spectrum analysis is due to the
existence of fast algorithms for its implementation. The class of methods are referred to as fast
Fourier transforms (FFTs). Consider a direct implementation of an 8-point DFT:

If the factors have been calculated in advance (and perhaps stored in a lookup table), then the
calculation of X[k] for each value of k requires 8 complex multiplications and 7 complex
additions. The 8-point DFT therefore requires 8 * 8 multiplications and 8* 7 additions. For an Npoint DFT these become N2 and N(N - 1) respectively. If N = 1024, then approximately one
million complex multiplications and one million complex additions are required. The key to
reducing the computational complexity lies in the observation that the same values of x[n] are
efectively calculated many times as the computation proceeds | particularly if the transform is
long. The conventional decomposition involves decimation-in-time, where at each stage a N-

point transform is decomposed into two N=2-point transforms. That is, X[k] can be written as
X[k] =N

The original N-point DFT can therefore be expressed in terms of two N=2-point DFTs.
The N=2-point transforms can again be decomposed, and the process repeated until only 2-point
transforms remain. In general this requires log2N stages of decomposition. Since each stage
requires approximately N complex multiplications, the complexity of the resulting algorithm is
of the order of N log2 N. The difference between N2 and N log2 N complex multiplications can
become considerable for large values of N. For example, if N = 2048 then N2=(N log2 N) _ 200.
There are numerous variations of FFT algorithms, and all exploit the basic redundancy in the
computation of the DFT. In almost all cases an
Of the shelf implementation of the FFT will be sufficient | there is seldom any reason to
implement a FFT yourself.
S ome forms of digital filters are more appropriate than others when real-world effects are
considered. This article looks at the effects of finite word length and suggests that some
implementation forms are less susceptible to the errors that finite word length effects introduce.
In articles about digital signal processing (DSP) and digital filter design, one thing I've noticed is
that after an in-depth development of the filter design, the implementation is often just given a
passing nod. References abound concerning digital filter design, but surprisingly few deal with
implementation. The implementation of a digital filter can take many forms. Some forms are
more appropriate than others when various real-world effects are considered. This article
examines the effects of finite word length. It suggests that certain implementation forms are less
susceptible than others to the errors introduced by finite word length effects.
Introduction to DFT:

The discrete Fourier transform (DFT) derived from the Fourier series The exponential
Fourier series of a continuous time periodic signal x(t) with fundamental period T 0 is given
by the synthesis equation

where the Fourier coefficients X k are given by the analysis equation

with the fundamental frequency F 0 and the period T 0 related by F 0 (Hz) = 1/T 0 (sec).
To obtain finite-sum approximations for the above two equations, consider the analog
periodic signal x(t) shown in Figure and its sampled version x s (nT). Using x s (nT), we can
approximate the integral for X k by the sum

k =
0, 1,
,
N1

x
(
n
)
e

j
2
k
n
/
N

k =
0, 1,
,
N1

where we used the relation F 0 T = 1/N, and approximated dt (or t) by T, and have used the
shorthand notation x(n) = x s (nT). (This procedure is similar to that used in a typical
introduction to integral calculus).

is sometimes
It
better to reason in terms of these periodic extensions when dealing with the DFT. Specifically, if
X[k] is the DFT of x[n], then the inverse DFT of X[k] is ~x[n]. The signals x[n] and ~x[n] are
identical over the interval 0 to N 1, but may differ outside of this range. Similar statements can
be made regarding the transform Xf[k].
Properties of the DFT
Many of the properties of the DFT are analogous to those of the discrete-time Fourier
transform, with the notable exception that all shifts involved must be considered to be
circular,
or
modulo
N.
Defining
the
DFT
pairs
and

A finite series
approximation for x(t) is obtained by truncating the series for x(t) in equation (1) to N
terms and substituting t = nT and F 0 = 1/TN. This will necessarily give the discrete
sequence x(n) instead of the continuous function x(t):
N 1

x(n) or x n =

Xk
k=0

j 2 k n / N

, n = 0, 1, , N1

x(t)

t
0

T0

2T 0

x(t)

x s (nT)

T 0 = NT

2T

(N1)T

N samples at n = 0, 1, , N1
t or dt = T = T 0 /N
The above two equations define the discrete Fourier transform (DFT) pair. A slight
adjustment of the (1/N) factor is needed so as to conform to standard usage. The
adjustment consists of moving the (1/N) factor from one equation to the other. Then the
direct DFT of the time series x 0 , x 1 , , x N-1 is defined as
N 1

Xk =

x n

n=0

j 2 k n / N

, k = 0, 1, , N1

And the inverse DFT is defined as

1 N 1
X e j 2 k n / N
xn = N k
,
k=0

(3)

n = 0, 1, , N1

(4)

It can be shown that substituting equation (3) into equation (4) produces an identity, so that
the two equations are indeed mutually inverse operations and therefore constitute a valid
transform pair.
(End of Omit)
The discrete Fourier transform as a discretized (sampled) version of the DTFT A
finite-duration sequence x(n) of length N (the length N may have been achieved by zeroj
padding a sequence of shorter length) has a Fourier transform denoted X() or X(e ),
N 1

X() =

x(n) e j n

0 < 2

X() is a continuous function of and has a period of 2. If we take N samples of

X() at equally spaced frequencies { k = 2k/N, k = 0, 1, ..., N1}, along the
interval [0, 2), the resulting samples are (Figure)

N 1
X(k) X 2 k =
x(n) e j 2 k n / N ,

N
n=0
Since these frequency samples are obtained by evaluating the Fourier transform X() at N
equally spaced discrete frequencies, the above relation is called the discrete Fourier
transform (DFT) of x(n). In other words, X(k) are discrete samples of the continuous X().

N samples of X()
The corresponding inverse discrete Fourier transform (IDFT) is given by
1 N 1
j 2 k n / N
, n = 0, 1, , N1
X (k ) e
x(n) =

k = 0

Example 1 Find the DFT of the unit sample x(n) = {1, 0, 0, 0}. (Aside. What is the DTFT
of x(n) = {1, 0, 0, 0}?)
x(n) = {1, 0, 0, 0}
1

n
1

3
N1

4
N

Sequence
Solution The number of samples is N = 4. The DFT is given by
N 1

X(k) =

x(n) e
= x(n) e

j k 2 n / N

j k 2 n / 4

0 k N1
, 0k3

n=0
3

= x(n) e

j k 2 n / 4

k = 0, 1, 2, 3

n=0

x(n) e

j k 2 n / 4

, k = 0, 1, 2, 3

n=0

k=0

X(0) =

x(n) e

j 0.2 n / 4

n=0

x(n) .1 = x(n)
n=0

n=0

= x(0) + x(1) + x(2) + x(3) = 1 + 0 + 0 + 0 = 1

k=1

X(1) = x(n) e

j1 . 2 n / 4

n=0

k=2

x(n) e

j 2 . 2 n / 4

n=0

k=3

x(n) e
n=0

= x(0) e

=1.1=1

x(n) e

jn

j 0

= x(0) e

=1.1=1

n=0

X(3) =

j 0

j n /2

n=0

X(2) =

x(n) e

j 3 . 2 n / 4

x(n) e

j3 n/2

= x(0) e

j 0

=1.1=1

n=0

The DFT is X(k) = {1, 1, 1, 1} and contains all (four) frequency components. In this example
X(k) is real-valued.

4-point DFT of {1, 0, 0, 0}

00

0.5

1.5

2.5