Time and Frequency Analysis of Discrete-Time Signals

4.1 Brief Theory of Discrete-Time Fourier Transform, Discrete

Fourier Transform, and Fast Fourier Transform
The Z-transform is an effective tool in linking the time and frequency
domains of a discrete-time signal x(n).
However, in order to specify practical properties of discrete-time systems,
such as low-pass filtering or high-pass filtering, it is necessary to
transform the complex z-plane to the real-frequency, , axis. !pecifically,
the region of the complex z-plane that is used in this transformation is the
unit circle, specified "y the region
The resulting transform is the Discrete-Time Fourier Transform (DTFT),
#ue to the need for a more applica"le and easily computa"le transform,
the Discrete Fourier Transform was introduced, which is very
homogeneous in "oth forward (time to frequency) and inverse (frequency
to time) formulations. The crowning moment in the evolution of #!$
came when the Fast Fourier Transform(FFT) was discovered "y %ooley
and Tukey in &'().
The **T, which is essentially a very fast algorithm to compute the #*T,
makes it possi"le to achieve real-time audio and video processing.
4.1.1 Discrete-Time Fourier Transform
The frequency response of the system is very important in defining the
practical property of the system, such as low-pass or high-pass filtering. +t
can "e o"tained "y considering the system function H(z) on the unit
circle. !imilarly, for a discrete-time sequence x(n), we can define the Z
-transform X (z) on the unit circle as follows,
TABL !.1
#T*T Theorems
The function X (e
j
) or X () is also called the Discrete-Time Fourier
Transform
(DTFT) of the discrete-time signal x(n). The inverse #T*T is defined "y
the following integral,
Properties of Discrete-Time Fourier Transform
- concise list of #T*T properties is given in Ta"le ..&.
Analog frequency and digital frequency
The fundamental relation "etween the analog frequency, / , and the
digital frequency, , is given "y the following relation,
or alternately,
where T is the sampling period, in sec., and fs =&/T is the sampling
frequency in H0.
1ote, however, the following interesting points,
2 The unit of / is radian3sec., whereas the unit of is 4ust radians.
2 The analog frequency, / , represents the actual !ysical frequency of
t!e "asic analog signal , for example, an audio signal (5 to 6 kH0) or a
video signal (5 to 6 7H0). The digital frequency, , is the transformed
frequency from 8quation ...a or 8quation ..." and can "e considered
as a mathematical frequency, corresponding to the digital signal.
"a#
"$#
F%&'( !.1
(a) -nalog frequency response and (") digital frequency response.
Analog frequency response and digital frequency response
9ne of the most important differences "etween discrete-time systems and
analog systems is that discrete-time systems have a periodic frequency
response, H(e
j
), while analog systems have a nonperiodic *ourier
transform H(j/)# *igure ..& illustrates this difference in "etween H(j/)
and H(e
j
).
4.1.) Discrete Fourier Transform
The #iscrete *ourier Transform (#*T) is a practical extension of the
#T*T, which is discrete in "oth time and the frequency domains#The
#T*T X() is a periodic function with period :; radians. This property is
used to the divide the frequency interval (5, :;) into $ points, to yield the
#*T of the discrete-time sequence x(n), 5 < n < $= & as follows,
TABL !.)
#*T Theorems
The +nverse #iscrete *ourier Transform (+#*T) is given "y the following
equation,
*ro+erties of the DFT
- concise list of #*T transform properties is given in Ta"le ..:. !ome of
the key features and practical advantages of the #*T are as follows,
2 The #*T maintains the time sequence x(n) and the frequency sequence
X(%) as finite vectors having the same length $. -dditionally, as seen
from 8quation ..6 and 8quation ..), the #*T and +#*T are "oth finite
sums, which makes it very convenient to program these equations on
computers and microprocessors.
2 T!e time-frequency relation is a very important relation in practical
#*T applications. The index n corresponds to the time value t = n>t,
sec., where >t is the sampling time interval. The index % corresponds
to the frequency value = %>, radians, where > is the #*T output
frequency interval. Then, for a given $-point #*T, the time frequency
relation is given "y
2 The concept of time s!ift in the #*T is defined circularly, the sequence
x(n)& 5 < n< $ = & is represented at $ equally spaced points around a
circle as shown in figure ..:a, for 1 ?@. Then, a circular shift,
represented as x((n = ))
@
), for example, is implemented "y moving the
entire sequence x(n) counter-clockwise "y five points, as illustrated in
*igure ..:". Hence, the sequence x(n) = Ax(5) x(&) x(:) x(.) x(6) x()) x(()
x(B)C, and the shifted sequence x((n = ))@) = Ax(.) x(6) x()) x(() x(B) x(5)
x(&) x(:)C#
F%&'( !.)
(a) !equence x(n) and (") circularly shifted sequence x((n = ))
@
)#
,ircular ,on-olution
-n $-point circular convolution of two sequences x(n) and !(n) is defined
as,
y(n) ?
$ote: The sequences x"n#, h"n#, and y"n# ha-e the same -ector length
of N.
.am+le
#etermine the circular convolution of the two @-point discrete-time
sequences, x&(n) and x:(n), given "y
Solution
The @-point circular convolution is given "y
"a# "$#
F%&'( !.!
(a) !equence x(m) and (") reflected sequence x(('m)@)#
%ircular convolution can "e carried out either "y analytic tec!niques,
such as the sliding tape method, or "y comuter tec!niques, such as
7-TD-E. Fe will discuss "oth approaches "elow.
The sliding tae met!od can "e done "y hand calculation, if the num"er
of points in the #*T, $& is quite small# The procedure is as follows,
2 Frite the sequences x&(m), x:(m)& and x:(('m)@) as shown "elow.
The sequence x
:
('m) is o"tained from the sequence x
:
(m) "y writing the
first element in the vector x
:
(m), then starting with the last element in
x
:
(m) and continuing "ac%(ards. Then, the dot product of the vectors
x&(m) and x
:
(('m)@) gives the convolution output x(5).
!imilarly, the next term in the sequence, x
:
((& = m)
@
), is o"tained "y
shifting x
&
('m) "y one step to the right, and "ac% again to t!e "eginning
of t!e )ector. The dot product of the vectors x
&
(m) and x
:
((& ' m)@) gives
the convolution output x(&).
2 -lternately, one could arrange the vector elements x&(m) and x:(m) in
$ = @ equally spaced points around a circle, as shown in *igure ...a. The
vector x (('m) ) is o"tained "y reflecting the vector elements of x
:
(m)
a"out the hori0ontal axis as shown in *igure ...". The vector x
:
((& ' m)
@
)
is o"tained "y shifting the elements of the vector x
:
(m) "y one position
counter-clockwise around the circle. Hence, the out- put vector is x(n) ?
A: : . 6 ) 6 . :C.
Gsing the comuter met!od, the circular convolution of the two
sequences, x
&
(n) and x
:
(n), can also "e o"tained "y using the convolution
property of the #*T, which is listed as $roperty : in Ta"le ..: a"ove.
This method consists of three steps.
2 Ste+ 1/ 9"tain the @-point #*Ts of the sequences x
&
(n) and x
:
(n),
2 Ste+ )/ 7ultiply the two sequences X&(%) and X:(%),
2 Ste+ !/ 9"tain the @-point +#*T of the sequence X(%)& to yield the
final output x(n),
7-TD-E program to implement the procedure a"ove is given
"elow,
$ote* 7-TD-E automatically utili0es a radix-: **T if 1 is a power of :.
+f 1 is not a power of :, then it reverts to a non-radix-: process.
Fast Fourier Transform
The fast *ourier transform (**T) is an efficient algorithm that is used for
converting a time-domain signal into an equivalent frequency-domain
signal, "ased on the discrete *ourier transform (#*T).
0.1 %1T(2D',T%21
The discrete *ourier transform converts a time-domain sequence into an
equivalent frequency-domain sequence. The inverse discrete *ourier
transform performs the reverse operation and converts a frequency-
domain sequence into an equivalent time-domain sequence. The fast
*ourier transform (**T) is a very efficient algorithm technique "ased on
the discrete *ourier transform, "ut with fewer computations required. The
**T is one of the most commonly used operations in digital signal
processing to provide a frequency spectrum analysis A&=(C. Two different
procedures are introduced to compute an **T, the decimation-in-
frequency and the decimation-in-time. !everal variants of the **T have
"een used, such as the Finograd transform, the discrete cosine transform
(#%T) , and the discrete Hartley transform . $rograms "ased on the
#%T, *HT, and the **T are availa"le in .
3.) D4L2*51T 2F T6 FFT AL&2(%T65 7%T6 (AD%8-)
The **T reduces considera"ly the computational requirements of the
discrete *ourier transform (#*T). The #*T of a discrete-time signal
x(nT) is
where the sampling period T is implied in x(n) and $ is the frame length.
The constants + are referred to as twiddle constants or factors, which
represent the phase, or
and is a function of the length $. 8quation ((.&) can "e written for % ? 5,
&, . . . , $ = &, as
This represents a matrix of $ H $ terms, since X(%) needs to "e calculated
for $ values of %. !ince ((..) is an equation in terms of a complex
exponential, for each specific % there are approximately $ complex
additions and $ complex multiplications. Hence, the computational
requirements of the #*T can "e very intensive, especially for large values
of $.
The **T algorithm takes advantage of the periodicity and
symmetry of the twiddle constants to reduce the computational
requirements of the **T. *rom the periodicity of +
and, from the symmetry of +
*igure (.& illustrates the properties of the twiddle constants + for $ ? @.
*or example, let % ? :, and note that from ((.6), +
&5
? +
:
, and from ((.)),
+
(
? =+
:
.
F%&'( 3.1 $eriodicity and symmetry of twiddle constant +.
*or a radix-: ("ase :), the **T decomposes an $-point #*T into two
($3:)-point or smaller #*TIs. 8ach ($3:)-point #*T is further
decomposed into two ($36)-point #*TIs, and so on. The last
decomposition consists of ($3:) two point #*TIs. The smallest transform
is determined "y the radix of the **T. *or a radix-: **T, $ must "e a
power or "ase of two, and the smallest transform or the last
decomposition is the two-point #*T. *or a radix-6, the last
decomposition is a four-point #*T.
0.! D,%5AT%21-%1-F(9'1,: FFT AL&2(%T65
7%T6 (AD%8-)
Det a time-domain input sequence x(n) "e separated into two halves,
Eecause (=&)
%
? & for even % and =& for odd %, ((.&5) can "e separated for
even and odd %, or
8quations ((.&.) and ((.&6) can "e more clearly written as two ($3:)-
point #*TIs, or
*igure (.: shows the decomposition of an $-point #*T into two ($3:)-
point #*TIs, for $ ? @. -s a result of the decomposition process, the XIs
in *igure (.: are even in the upper half and they are odd in the lower half.
The decomposition process can now "e repeated such that each of the
($3:)-point #*TIs is further decomposed into two ($36)-point #*TIs, as
shown in *igure (.., again using $ ? @ to illustrate.
The upper section of the output sequence in *igure (.: yields the
sequence X(5) and X(6) in *igure (.., ordered as even. X(:) and X(()
from *igure (.. represent the odd values. !imilarly, the lower section of
the output sequence in *igure (.: yields X(&) and X()), ordered as the
even values, and X(.) and X(B) as the odd values. This scram"ling is due
to the decomposition process. The final order of the output sequence
X(5), X(6), . . . in *igure (.. is shown to "e scram"led.
The output needs to "e resequenced or reordered.
The output sequence X(%) represents the #*T of the time sequence x(n).
This is the last decomposition, since we have now a set of ($3:) two-
point
#*TIs, the lowest decomposition for a radix-:. *or the two-point #*T,
X(%) in ((.&) can "e written as
9r
since +& ? e=j:J3: ? =&. 8quations ((.:5) and ((.:&) can "e represented
"y the flow graph in *igure (.6, usually referred to as a "utterfly. The
final flow graph of an eight-point **T algorithm is shown in *igure (.).
This algorithm is referred as decimation-in-frequency (#+*) "ecause the
output sequence X(%) is decomposed (decimated) into smaller
su"sequences, and this process continues through , stages or iterations,
where $ ? :,. The output X(%) is complex with "oth real and imaginary
components, and the **T algorithm can accommodate either complex or
real input values.
The **T is not an approximation of the #*T. +t yields the same result as
the #*T with less computations required. This reduction "ecomes more
and moreimportant with higher-order **T.
There are other **T structures that have "een used to illustrate the **T.
-n alternative flow graph to the one shown in *igure (.) can "e o"tained
with ordered output and scram"led input.
-n eight-point **T is illustrated through an exercise as well as through a
programming example. Fe will see that flow graphs for higher-order
**T (larger $) can readily "e o"tained.
F%&'( 3.) #ecomposition of $-point #*T into two ($3:)-point
#*TIs, for $ ? @.
F%&'( 3.! #ecomposition of two ($3:)-point #*TIs into four ($36)-
point #*TIs, for $ ? @.
.ercise 3.1 ight-*oint FFT 'sing Decimation-in-Frequency
Det the input x(n) represent a rectangular waveform, or x(5) ? x(&) ? x(:)
? x(.)? &, and x(6) ? x()) ? x(() ? x(B) ? 5. The eight-point **T flow
graph in *igure (.) can "e used to find the output sequence X(%), % ? 5,
&, . . . , B. Fith $ ? @, four twiddle constants need to "e calculated, or
F%&'( 3.0 8ight-point **T flow graph using decimation-in-frequency.
The intermediate output sequence can "e found after each stage.
1. At stage 1/
where x
,
(5), x
,
(&), . . . , x
,
(B) represent the intermediate output sequence
afterthe first iteration that "ecomes the input to the second stage.
). At stage )/
The resulting intermediate, second-stage output sequence x
,,
(5), x
,,
(&), . . .
X
,,
(B) "ecomes the input sequence to the third stage.
!. At stage !/
Fe now use the notation of XIs to represent the final output sequence.
The values X(5), X(&), . . . , X(B) form the scram"led output sequence. and
plot the output magnitude.
3.4 D,%5AT%21-%1-T%5 FFT AL&2(%T65 7%T6 (AD%8-)
Fhereas the decimation-in-frequency (#+*) process decomposes an
output sequence into smaller su"sequences, the decimation-in-time (#+T)
is another process that decomposes the input sequence into smaller
su"sequences. Det the input sequence "e decomposed into an even
sequence and an odd sequence, or
and
x(&), x(.), x()), . . . , x(:n K &)
Fe can apply ((.&) to these two sequences to o"tain
which represents two ($3:)-point #*TIs. Det
Then X(%) in ((.:.) can "e written as
*igure (.@ shows the decomposition of an eight-point #*T into two four-
point #*TIs with the decimation-in-time procedure. This decomposition
or decimation process is repeated so that each four-point #*T is further
decomposed into
F%&'( 3.; #ecomposition of eight-point #*T into two four-point
#*TIs using #+T