EE 215 Semester Project

Signal Analysis using Discrete Fourier Transforms

Department of Electrical and Computer Engineering

Missouri University of Science and Technology

13 Electrical Engineering 215 Section 1 A

Abstract

The purpose of this project is to bridge the relationship between the Continuous

time and the Discrete time domains. The relationship will be shown through the use of

Discrete Fourier analysis. The essential idea of Fourier analysis is the use of Fourier

Transforms to convert from the time domain signal to its frequency domain equivalent.

In this project the Transforms to be used are the DTFT, and the DFT. Using MATLAB as

the tool to compute the transforms, relationships between the transforms will be realized

and ultimately a connection between the Discrete time domain and the Continuous time

domain.

Electrical Engineering 215 Section 1A

2
 Introduction 4

Background and Theory 4

Procedure 7

Results & Analysis 9

Conclusions 12

Appendices 13

Appendix A (Figures) 13

Appendix B (MATLAB code) 24

References 39

Electrical Engineering 215 Section 1A

3
Introduction
Two of the most rapidly evolving and advancing areas of Electrical Engineering

are the areas of signal processing and control systems. These areas both rely heavily

on the ability to analyze signals and systems. While all ‘real’ physical signals and

systems are functions of time, it is often helpful to consider them as functions of

frequency. When analyzing a signal, the calculations will generally become more

simplified in when analyzed with respect to the frequency. That said however, typically

the transformation equation is more complicated and is usually computed using a

computer based program.

Background and Theory

The first step in analyzing a signal with respect to frequency is transforming the

time domain signal into the frequency domain. Such a transformation can be completed

using several different Fourier transforms, however the transform chosen depends on

properties of the time domain signal, and the desired frequency domain signal.

In this project the signal analysis was done on the computer using MATLAB.

Because of the limitations of MATLAB and the computer based analysis in general,

signals cannot be considered an infinitely long continuous time signal, but instead as

finite amounts of data points that were inherently ‘sampled’ from the real continuous

time signal. The sampled signals are called Discrete time signals. Due to the discrete

time nature of the signal, the Fourier Transforms used in the analysis processes are

Electrical Engineering 215 Section 1A

4
modified to fit such a type of signal. The transforms used in this project for these

Discrete signals were the “Discrete Time Fourier Transform” (DTFT) and the “Discrete

Fourier Transform” (DFT). As the names would imply, the transforms are extremely

similar in definition.

The definition of the Discrete Time Fourier Transform (DTFT) is shown in Figure

1 of Appendix A. The DTFT is essentially a transform that is only realizable for discrete

signals with an ‘infinite’ number of samples. When brought into the frequency domain,

the DTFT result will be an inherently ‘infinite’ amount of complex discrete frequency

values representing the amplitude and phase of the signal at the respected frequencies.

Due the resulting amount of infinite discrete frequencies the DTFT result can be

considered in continuous frequency, that is being defined at all frequencies. Also

because it is physically impossible to have an infinite number of samples, the DTFT is

generally regarded as a theoretical transform.

The Discrete Fourier Transform (DFT) on the other hand, is more practical in

nature than the DTFT. The definition of the DFT is shown in Figure 2 of Appendix A.

The essential difference between the DFT and the DTFT coincides with the fact that the

DFT is defined over a finite length of time and consists of a finite amount of equally

spaced values. Due to this fact, the DFT is more practical to compute on a computer

than the DTFT. As one may realize, the larger the number of values, or samples,

included in the DFT computation, the closer the result of the DFT becomes to the values

given by the DTFT.

Electrical Engineering 215 Section 1A
5
 The computation required for the definition of the DFT is considerably complex,

and lengthy. While a computer can solve this much quicker than by hand, the

computation time by the computer will increase exponentially as the number of samples

is increased. It is for this reason that a faster computer algorithm to compute the DFT

was created. This algorithm is called the “Fast Fourier Transform”. The actual details of

this algorithm are beyond the scope of this report. However it can be said that the FFT

algorithm reduces computation time by removing N2 complex multiplications and N(N-1)

complex additions. These reductions in computation are noticeable even with a small

amount of samples.

Considering the importance in the number of discrete samples taken from a

Continuous time function to obtain an accurate waveform, a sampling criterion has been

formed to provide a rule of thumb for sampling frequencies in periodic functions. The

name of this criterion is the Nyquist Criterion. Simply put the Nyquist Criterion states

that for a Continuous time periodic signal to be accurately represented as a Discrete

time signal, the sampling frequency must be greater than two time the fundamental

frequency of the signal.

Electrical Engineering 215 Section 1A

6
Procedure

Through preforming the DTFT, DFT, and the FFT on two different continuous

time functions the relationship between the discrete and continuous time domains will

be validated. Two functions were tested in this project. These signals were the cosine

function and the rectangle function shown in Figure 3 of Appendix A. Furthermore, as it

was mentioned previously, the transforms were computed using MATLAB.

The first step was converting the Continuous time signal into the Discrete time

signal. This process was simplified immensely through MATLAB. The MATLAB codes

for both of the functions are shown in Appendix B. Several sampling frequencies were

chosen below, above and at the Nyquist criterion for the cosine function. Because the

rectangle function is not periodic, a sampling frequency of 10 was chosen and

increased accordingly. The time step was quickly found to be the reciprocal of the

sampling frequency. Using this, the complete time-frame and time steps of the Discrete

time signal were declared and the function was computed for its respected time-frame

and time step.

With the newly found discrete time version of the signal found, the next logical

step was to compute the Discrete Time Fourier Transform (DTFT). Using the definition

shown previously in Figure 1 of Appendix A, the summation was implemented into

MATLAB as a set of two nested ‘for’ loops. From this calculation the resulting frequency

Electrical Engineering 215 Section 1A

7
domain signal had both an amplitude and phase associated with its respected

frequencies. The phase portion of the signal will be ignored throughout the remainder of

the project. By computing the frequency for one period and then scaling the axis and

amplitudes appropriately, the results are represented appropriately and are computed

much faster.

In order to verify the results of the computed DTFT, a theoretical or ‘modified’

version of the DTFT will be used. Computing the ‘modified’ version of the DTFT involves

convoluting the analytically calculated or ‘theoretical’ DTFT of the original discrete signal

with the DTFT of a windowing function that represents the desired width. The result

from this convolution should be nearly identical to the computed DTFT just before. This

cross-referenced DTFT function will be referred to as the ‘modified’ DTFT. For this

project the windowing function was chosen to be a rectangular function.

Following the completion of the DTFT, the DFT will be computed using the

original definition shown in Figure 2 of Appendix A. The computation involved with the

DFT is nearly identical to the DTFT. For this reason, the resulting frequency domain

signal will be found by using a set of two nested ‘for’ loops. Once again the key

difference between the DFT and the DTFT is that the DFT is in discrete frequency, while

the DTFT is in continuous frequency. Hence the visible differences in the graphs.

The last portion of the procedure involves computing the DFT, not by using the

definition, but by the Fast Fourier Transform (FFT) algorithm. The results of this
Electrical Engineering 215 Section 1A
8
transform will be similar to the results of the DFT from the definition. An apparent

difference however is the method of implementation. Due to the way the FFT algorithm

is built into MATLAB, a function called ‘fftshift’ is needed in conjunction with the ‘fft’

command. By using the ‘fftshift’ command, the resulting frequency signal is shifted in

such a way that the zero frequency component of the FFT array is moved to the middle.

By having the zero frequency component in the center of the array, the plots are a more

accurate representation of the resulting frequency signal.

Results & Analysis

The results from the procedure were, for the most part, accurate with the

theoretical background. Starting with the cosine function, it became apparent that the

higher the sampling frequency was above the Nyquist criterion the closer the discrete

time function became to the real continuous time signal. With a lower sampling

frequency, the discrete time signal was not a good representation of the Continuous

time function. The results of the conversion for the cosine function into Discrete time is

shown in Figure 4 of Appendix A.

Calculating the DTFT was the second step in the procedure. From the theory, the

output for the cosine function should have a peak amplitude value equal to the absolute

value of the maximum value of the frequency calculated for. In this case that value was

0.5. Furthermore, these peak values should occur at the fundamental frequency of the

Electrical Engineering 215 Section 1A

9
periodic function. This value was 1250 for the cosine function. The results of this

calculation are shown in Figure 5 of Appendix A. Next, the modified DTFT was

calculated to confirm the previous results by using the analytical derived results

convoluted with a rectangular window function. The results of the modified DTFT are

shown in Figure 6 of Appendix A.

Following the results for both of the DTFT calculations, the DFT of the cosine

was found in the same manner. That is, gradually increasing the sampling frequency

above the Nyquist criterion until the result matched theoretical results. The results of the

DFT calculations using the DFT definition are shown in Figure 7 of Appendix A. To

validate these results the same calculation was preformed again, only using the FFT

algorithm. Doing so confirmed that the resulting signal found from the DFT definition,

was indeed correct. The results from the FFT calculation are shown in Figure 8 of

Appendix A.

After the Fourier analysis of the cosine function, the rectangular function from

Figure 3 of Appendix A was analyzed in a similar manner. The main difference with the

rectangular function is that it is not a periodic signal. As such, the Nyquist criterion was

not applicable. Therefore the sampling frequency was chosen and increased until

waveform converged to a general form or appeared similar to the theoretical results.

After converting the Continuous time signal into the Discrete time version, the resulting

waveform is shown in Figure 9 of Appendix A.

Electrical Engineering 215 Section 1A

10
 Once the Discrete time version of the signal had been found, the next step was

to calculate the DTFT by using the definition. In the same manner that was done with

the cosine function, the DTFT on the rectangle function was realized and found to result

in the graphs shown in Figure 10 of appendix A. In cross-referencing the results found

from the DTFT the modified DTFT calculations were also computed. A similar

rectangular windowing function was used in this calculation as well. The waveforms

from this computation are shown in Figure 11 of Appendix A.

The DFT calculations were also found to match the results of the DTFT for the

rectangular function. The results are shown in Figure 12 of Appendix A. Furthermore the

FFT matched the DFT results. The FFT output is shown in Figure 13 of Appendix A

Electrical Engineering 215 Section 1A

11
Conclusions

Following the completion of the project, it was clear that as the sampling

frequency increased, the more accurate the discrete time signal and frequency

transforms represented their continuous time counterpart. Furthermore it became

obvious that the resulting waveforms were more accurate once the sampling frequency

surpassed the Nyquist criterion. Given a sufficient amount of samples the DTFT and

DFT computations will converge to the same waveform. It is this waveform that, given a

sufficiently large amount of samples, will become the same waveform that would result

from computations in the Continuous time domain.

Having completed the project, relationships between the Fourier transforms, and

sampling frequencies have been found. It is the relationship between these two that

determine the accuracy, computation time, and similarity to the continuous time domain

function. It is the knowledge of this relationship in Fourier analysis that will aid in future

electrical engineering endeavors such as Signal Processing and Control system

involving computer based analysis.

13 Electrical Engineering 215 Section 1 A

Appendices
Appendix A (Figures)

{ } ( ) ∑

Figure 1. Definition of the Discrete Time Fourier Transform (DTFT)

{ } ∑

where N is the number of samples taken

Figure 2. Definition of the Discrete Fourier Transform (DFT)

( ) ( ) and ( ) ( ⁄ )

Figure 3. The functions to be analyzed

13 Electrical Engineering 215 Section 1 A

 Figure 4. Discrete time version of the Cosine function

Electrical Engineering 215 Section 1A

14
 Figure 5. DTFT of the Cosine function

Electrical Engineering 215 Section 1A

15
 Figure 6. Modified DTFT of the Cosine function

Electrical Engineering 215 Section 1A

16
 Figure 7. DFT of the Cosine function

Electrical Engineering 215 Section 1A

17
 Figure 8. FFT of the Cosine function

Electrical Engineering 215 Section 1A

18
 Figure 9. Discrete time version of the rectangular function

Electrical Engineering 215 Section 1A

19
 Figure 10. DTFT of the rectangular function

Electrical Engineering 215 Section 1A

20
 Figure 11. Modified DTFT of the rectangular function

Electrical Engineering 215 Section 1A

21
 Figure 12. DFT of the rectangular function

Electrical Engineering 215 Section 1A

22
 Figure 13. FFT of the rectangular function

Electrical Engineering 215 Section 1A

23
 Appendix B (MATLAB code)

Electrical Engineering 215 Section 1A

24