MATLAB SIMPLIFIED

Practical Programming and Signal Processing for Scientists

Copyright 2013 David Mann, Loggerhead Instruments

 Table of Contents
Getting Started........................................................... 5
MATLAB .................................................................... 6
Launch MATLAB ........................................................ 6
Matrices ................................................................... 12
A First Program ........................................................ 14
Sources for MATLAB Help ....................................... 15
MATLAB Help File ....................................................... 15
Mathworks Online........................................................ 17
Learning to Program ................................................ 18
MATLAB Scripts (m-files) ......................................... 18
MATLAB Path .......................................................... 19
Fundamental Programming Constructs .................... 21
For Loops .................................................................... 21
If .................................................................................. 22
Relational and Logic Operators ................................... 23
Basic File Input ........................................................ 29
Matrix Manipulation .................................................. 40
Introduction to Time Series Analysis ........................ 52
Analog and Digital Signals ....................................... 52
Sampling Rate ......................................................... 53
Aliasing........................................................................ 54
Avoiding Aliasing ......................................................... 56
Nyquist Theorem ............................................................... 56
Anti-Aliasing Filters ........................................................... 56
Sampling Resolution ................................................ 57
Bits .............................................................................. 57
 Digital Resolution ......................................................... 59
Measurement Resolution ............................................. 59
Decimal Numbers ........................................................ 60
Reading Data Files with fread() ............................... 60
File Reading Example: ...................................................... 61
MATLAB Functions.................................................. 63
RMS ............................................................................. 63
Using RMS Example .................................................... 65
Fourier Transform .................................................... 67
Measuring Amplitudes of Specific Frequencies ........... 75
Determining Peak Amplitude and Frequency............... 75
Additional Details ......................................................... 76
Digital Filters............................................................ 77
Finite Impulse Response (FIR) Filters ..................... 78
What about the end bits? ............................................. 81
Visualizing the frequency response of the filter............ 81
More Reason to ♥ MATLAB! ........................................ 83
Infinite Impulse Response (IIR) Filters ..................... 84
Try This ........................................................................ 87
How Do IIR Filters Work? ............................................ 88
Automatic Signal Detection...................................... 90
Energy Detectors ..................................................... 90
Energy Detection in the Time Domain ......................... 90
Energy Detector Example ........................................ 91
Step 1. Filter the signal .................................................... 92
Step 2. Rectify Signal ........................................................ 93
Step 3. Envelope Signal ................................................... 94
Step 4. Threshold Signal ................................................... 94
find ..................................................................................... 96
Image Processing .................................................... 97
Basic MATLAB Image Functions ................................. 97
Creating and plotting an image ................................ 98
Plotting the Matrix as an Image ......................................... 99
Colormap ......................................................................... 100
Try This ...................................................................... 105
Loading and Manipulating an Image ...................... 105
Thresholding Images ............................................. 107
 1
Chapter

Getting Started

T
his introduction to programming and signal processing with
MATLAB is aimed towards teaching basic programming skills to
scientists who have never programmed before. While computers
increasingly allow collections of large datasets, software for
analyzing these datasets typically lags behind. Scientists are then left with
two choices. The first is to try to get by with pre-packaged software. These
software packages are often less than ideal for the task at hand, and often
lead to a computer aided ‘brute-force’ data analysis. While this is sometimes
adequate, if any of the parameters used in the data analysis change, it often
has to be repeated again by brute force. The second option is to learn how
to program with a software package geared towards scientists and engineers,
such as MATLAB. While this may involve a fair amount of pain at the
beginning, being a proficient programmer allows you the flexibility to ask
questions unbounded by what software packages currently exist. A major
advantage of writing your own software is that if you decide to change
parameters in the analysis, it is little effort to make a small change and run
the software again to process the data. OK—so there is a third
option…paying someone to program for you. But, programming is fun, so
why pay someone else to do it?

It is impossible to learn how to program just by reading a text. You must

write code. This text includes many example programs that you should try.
Don’t be lazy. Type them in. Make typos and see what kinds of errors you
get. And, don’t be limited by the techniques shown here. Don’t be afraid to
tweak the code to do something new. Programming involves reworking

5
code until you get it right. Do not get discouraged. It is a rewarding
experience to break through the bugs to get something that works.

MATLAB
MATLAB was chosen as the programming environment for this text,
because it is easily available (although, relatively expensive) and has a
tremendous number of pre-programmed routines available to the scientific
programmer. There are other languages that you could use to do the same
things, such as C or the open source language Python. You will find that
once you learn to code in MATLAB, it is not too difficult to migrate to
another language. They all share the same basic constructs.

Launch MATLAB
The examples in this text will use MATLAB 7. While earlier versions of
MATLAB will work just as well, MATLAB 7 has vastly improved graphics
editing capabilities. The basic functionality of MATLAB can be extended
with various toolboxes. Some chapters will use the filter functions of the
signal processing toolbox. Otherwise, the basic MATLAB program will be
sufficient for the examples in this text. A student version of MATLAB is
available.

Launch MATLAB by double-clicking on the MATLAB icon. The

Command Window is where you type programming commands (Fig. 1).

6
Figure 1-1. MATLAB interface.

Type

>>x=10

in the workspace and press ENTER.

MATLAB tells you x is now equal to 10 and a new variable ‘x’ has appeared
in the pane on the upper left called the workspace (Figure 1-2).

7
Figure 1-2. Screenshot showing creation of a variable in MATLAB.

Programming involves variables just as in algebra. You can perform

mathematical operations on these variables just as with algebra.

For example, type

>>y=x/20

A new variable ‘y’ is created in the workspace, and its value is calculated as
0.5.

8
Figure 1-3. Creation of a second variable in the MATLAB workspace by dividing x by
20.

Array Editor Window

Double click on the ‘x’ variable in the workspace, and you will see a
spreadsheet window pop-up showing you the value of ‘x’. Close this
window.

Figure 1-4. MATLAB array editor. Double-click on a variable in the workspace to

open the values in the array editor.

TIP: Suppressing Command Window Output with ;

9
If a line does not end with a semicolon ‘;’, its output will be printed on the
MATLAB screen (as in Figures 1-2 and 1-3). Many times you do not need
to see the output of every variable. Use a semicolon ‘;’ to suppress the
output.

>> y=x/20;

10
Table 1-1 MATLAB Basic Math Functions

Function Command

Addition +

Subtraction -

Multiplication *

Division /

Power ^

Square Root sqrt

Log base 10 log10

Exponential exp

Natural logarithm log

Trigonometric functions sin, cos, tan, asin,

acos, atan

Average mean

Standard deviation std

11
Matrices
Setting variables to single values is great, but if you have a lot of data to
analyze it will not get you far. Fortunately, MATLAB allows you to create
and perform math on arrays or matrices of numbers. In fact, that is why it is
called MATLAB, for ‘MATrix LABoratory’.

Generating Matrices with Square Brackets [ ]

Type:

>> z=[1:10];

Note that nothing seems to have happened. The semicolon suppressed the
output to the screen. But, notice that a new variable ‘z’ has appeared in the
workspace.

Figure 1-5. Creation of a matrix with values from 1 to 10 in the variable ‘z’.

Double click on the ‘z’ in the Workspace to open it in the array editor. You
will now see that ‘z’ has values ranging from 1 to 10 increasing by 1 (Figure
12
1-6). Also notice that the Size of ‘z’ listed in the workspace is 1x10. That
tells you there is 1 row and 10 columns.

Figure 1-6. Variable ‘z’ opened in array editor.

Say we want these values to go from 10 to 100. We can simply multiply ‘z’
by 10.

Type:

>>z=z*10

This is an important trick (it saves you from coming up with a lot of variable
names). In plain English it means, ‘the new values of z will equal the old
values of z times 10’. The new values of z will be shown as:

z =

10 20 30 40 50 60 70 80 90 100

This quick introduction to matrices is enough to get you started writing a

program. We will come back to matrices and matrix manipulation in later
chapters. Matrices are one of the unique features of MATLAB over
traditional programming languages such as ‘C’, which does not have even
simple math functions for these types of arrays.
13
Note: From now on the examples will just show what you type in
the workspace, rather than MATLAB screen shots.

A First Program
Most (if not all) programming texts start off with a simple program called
‘Hello World’ that prints the text ‘Hello World’ to the screen. Well, this one
won’t. What you really want to do as a scientist is manipulate, analyze, and
plot data. So, our first program will be a simple program to create and plot
two variables against one another. We’ll call this the scientific equivalent of
‘Hello World’.

Example 1-1
>> x=[0:10];
>> y=[0:2:20];
>> plot(x,y)

14
Figure 1-7. Basic MATLAB plot.

Program Description

>> x=[0:10];
This line creates a matrix in variable ‘x’ from 0 to 10 counting by 1.

>> y=[0:2:20];
This line creates a matrix in variable ‘y’ from 0 to 20 counting by 2. If you
look in the Workspace window in MATLAB you will see both ‘x’ and ‘y’ are
matrices of size 1x11 (i.e. 1 row and 11 columns).

>> plot(x,y)
Plot is one of the most fundamental MATLAB commands. There are many
variations of plot. This one plots two variables against each other. The first
variable in the list is plotted on the x-axis and the second variable is plotted
on the y-axis. It is important that each variable have the same number of
values. Once you type plot, a new window is created and in this case titled
‘Figure No. 1’. You can edit the plot using the menus in the Figure.

TIP: Save time retyping long commands

↑ Scroll to previously typed commands with the Up Arrow.

Sources for MATLAB Help

You will always encounter a point in programming when you get stuck.
Really stuck.

MATLAB Help File

The first thing to do is to check the MATLAB help file. It has complete
descriptions of all MATLAB functions with some examples. The most
complete version of MATLAB help is accessed from the Help menu in
MATLAB (Fig. 1-8). It is not a bad idea to just go through the entire help
file to see what functions MATLAB has available.

15
Figure 1-8. MATLAB help. Use it.

You can also access basic help descriptions from the MATLAB command
line.

Type:

>> help plot

You will see a detailed (but not complete) text help for the MATLAB plot
command.

Try This

Using the MATLAB help try plotting the same data from Example 1-1 with
red dots (without the line) so that the output looks like Figure 1-9.

16
Figure 1-9. Variation on MATLAB plot.

Mathworks Online
The MATLAB help file is also available online (www.mathworks.com). The
Mathworks also posts m-files written by other users of MATLAB. These
are often quite good and can save a good deal of time programming basic
functions. So, if you think MATLAB should have a command for what you
are trying to do, but doesn’t, check their online database of programs.

17
 2
Chapter

Learning to Program
This chapter introduces two basic programming constructs: loops and
conditional statements. The majority of programs can be written with these
programming constructs. Up until now we have just run through a set of
statements that mover from one to the next. Loops are used to cycle
through blocks of code to repeatedly perform the same set of operations.
Conditional statements are used to test for certain conditions, such as
equality of a variable with a value, and are used in program flow control.

MATLAB Scripts (m-files)

MATLAB Scripts are known as m-files and have the extension *.m (where *
represents the script name). M-files are simply text files containing
programming statements. The advantage of m-files is that a series of
programming statements can be grouped together, and then run with a
single command in the MATLAB command window.

To create an m-file, click on the File menu, select New and then M-file.

Type in the commands used for Example 1-1 (see Figure 2.1).

Save file as example1.m (Click on the File menu, then click on Save. Be sure
to remember where you saved this file.).

The first two lines that start with a percentage sign (%) are comments. Any
line that starts with a % is ignored by MATLAB. Comments are important
18
for documenting your m-files, especially when they get complicated. They
allow other people to figure out what your code does, and remind you when
you go back to an m-file that you wrote months ago.

Figure 2-1. An example m-file using the commands from Example 1-1.

MATLAB Path
Before you can run the m-file, you need to make sure that MATLAB knows
where to find the file you just created.

The MATLAB path is simply the list of directories where MATLAB ‘looks’
for your m-files. It starts at the top of the list and works its way down. So, if
you save a new m-file to a folder that is not in the path, and then try to run it
from the command line, you will get an undefined function error.

A new m-file called example1 was saved to a new folder in My Documents.

When you go to run it, you get this error.

>> example1

??? Undefined function or variable 'example1'.

19
To add the correct path, click on the File menu and select Set Path (Figure
2-2).

Click Add Folder, then search for the folder where you saved example1.m.
Click OK.

Click the Save button in the Set Path dialog box, then click the Close button.

To run example1.m simply type the m-file name in the MATLAB command
window:

>> example1

Figure 2-2. MATLAB Set Path dialog box that allows the addition of directories for
MATLAB to search for m-files.

20
TIP: M-file naming

Only use alphanumeric characters in creating m-file names. Do not use

spaces, dashes, or other special characters. Do not use the names of already
existing m-files, including all MATLAB functions like mean, std, sin. If
different m-files exist with the same name, MATLAB will run the one it
finds first as it goes through the path.

Fundamental Programming Constructs

The two basic programming constructs we will learn about are loops and
conditional statements. There are several types of loop statements, one of
the most powerful is the ‘for’ loop, which we will learn about first.

For all of these examples, type the code below into an m-file and then run it.

For Loops
The easiest way to learn about for loops is to run one.

Example 2-2
% example2: a basic for loop
for x=1:3 % loop x from 1 to 3 counting by 1
y=x*2 % do some math; y is x times 2
end % end of for loop

When you run this example, you get the following output.

>> example2

y = 2

y = 4

y = 6

How does a ‘for’ loop work?

21
The program ran through the loop between the ‘for’ and the ‘end’ statement
three times. On the first pass into the for loop, x=1, so y=2. When the
‘end’ statement was reached, the program execution returned to the ‘for’ line
and incremented x, so that now x=2. Then on the next line y=4. Again, the
‘end’ is reached and the program loops back to the ‘for’ statement and
increments x by 1. Now, x=3. The program continues to the next line, and
calculates y=6. Again, the ‘end’ is reached. This time the program looks at
the ‘for’ line and sees that x has reached the maximum value of 3, and the
program ends.

Just like the MATLAB colon operator, the ‘for’ loop can step using different
step sizes. Try the following code example:

Example 2-3
for x=10:-2:2 % loop x from 10 to 2 stepping by -2
y=x*2 % do some math;
end % end of for loop>> x=[0:10];

If
The most important control statement is the ‘if’ statement. ‘If’ statements
are used to check whether a value satisfies a certain condition, and if it does
it executes a block of code. Like loops there are several versions of the
‘if…then’ statement. Here we will learn the most basic version.

Example 2-4
for x=1:5 %loop from 1 to 10
if (x>4) % check if x is >4
x % show the x value to the screen
disp('X is greater than 4');
end % end of if block
end % end of for loop

Example 2-4 is a simple (and somewhat useless) example that sets up a for
loop where x will step from 1 to 5. The ‘if’ statement checks whether x>4.
If x is less than or equal to 4, as it is on the first four steps, then the program
skips to the ‘end’ following the ‘if’. If x is greater than 4, as it is when x=5,
the lines of code after the ‘if’ statement are executed. The first just displays
the value of x to the command window. The ‘disp’ command displays the
text between the single quotes to the command window.
22
TIP: Indentation

Notice in Example 2-4 how the program is indented after the ‘for’ and ‘if’
statements. This is done so that it is clear where the ‘if’ and ‘for’ code blocks
are. Since each has an ‘end’, it is easy to see which statements are within the
‘if’ statement and which are within the ‘for’ loop.

Relational and Logic Operators

Note that all sorts of logic operators can be used within an ‘if’ statement
(Table 2-1). Note that to test equality, you need a double equal sign (e.g. if
(x==2)). This is because if a single equal sign was used, x would be reset to
the value 2.

Table 2-1. Relational operators that can be used with ‘if’ statements.

Operation Symbol

Greater than >

Less than <

Equal ==

Not equal ~=

Greater than or equal >=

Less than or equal <=

Logic operators can be used to test multiple conditions within an ‘if’

statement (Table 2-2).

23
Table 2-2. Logical operators that can be used to test multiple conditions in
an if statement.

Operation Symbol

And &

Or |

Example 2-5 demonstrates the use of the double equal sign and the ‘or’
logical operator to test whether a value is odd or even.

The ‘else’ statement is useful when you want to run a set of commands
when the ‘if’ statement fails.

Example 2-5
for x=1:5 % loop from 1 to 5
x % display x value
if((x==1)|(x==3)|(x==5)) % test whether x is odd
disp('x is odd'); % if x is odd, say x is odd
else % otherwise, if x is not odd
disp('x is even'); % say x is even
end % end of if
end % end of for loop

Now you know most of the basic tools of programming to actually get
something accomplished.

Let’s look at a more useful program that uses the looping functions. Say we
have a time series of temperature measured every hour for one month, and
we want to calculate the minimum, maximum, and average temperature for
each day.

First we need to get the time series into MATLAB (the time series has been
supplied as temp.txt). To do this we will use the MATLAB importing
features.

Click on the File menu and select Import Data.

24
Navigate to temp.txt and click on the file. This will open the MATLAB
import wizard, which will load the data file into a variable called temp. Now,
let’s write an m-file to step through the temp matrix to calculate the
minimum, maximum, and average for each day.

Example 2-6
% tempanalysis
% m-file to calculate daily minimum, maximum, and average
% temperatures.
% This program assumes that temperature data have been
% loaded into the Workspace and given the variable name
% temp.
% The temp data have been collected hourly for one month

nsamples=length(temp); %number of temp samples in temp

counter=0; %initialize counter to 0

for x=1:24:nsamples-23 %loop through samples stepping by 24

%(24 h in a day)
counter=counter+1; %increment counter
minT(counter)=min(temp(x:x+23)); %get minimum value
%over 24 hours x:x+23
maxT(counter)=max(temp(x:x+23)); % get max value
avgT(counter)=mean(temp(x:x+23)); % calc average
end

plot(minT,'blue'); % plot minimum daily values in blue

hold on % hold on will overlay next plot
plot(maxT,'red'); % plot max daily values in red
plot(avgT,'green'); % plot average daily values in green

Program Description

nsamples=length(temp); %number of temp samples in temp

The ‘length’ function returns how many data points are in a given matrix if it
is one-dimensional (has 1 row or 1 column). For matrices that are two-
dimensional (has multiple rows and columns), ‘length’ returns the number of
row. The purpose of this function is to determine how much data we have
to process. This lets you feed the same program with datasets with different
amounts of data without reprogramming.
25
counter=0; %initialize counter to 0
A counter is a variable that gets incremented in a program, and is usually
used to keep track of the number of times something has been done. Here
we initialize a counter, which we have called ‘counter’, to zero. It will later
be used to provide and index into the matrices holding the minimum,
maximum, and average temperatures.

for x=1:24:nsamples-23 %loop through samples stepping by 24

%(24 h in a day)
This is our main loop to go through the dataset. We are incrementing by 24
because there are 24 data points (hourly) for each day in our dataset. The
loop continues to the number of samples – 23 points. We will see why you
want the -23 points in a moment. You can think of this loop as providing
an index into the matrix of hourly temperature readings. On the first pass
through x=1. On the second pass through x=25. In the temperature matrix
(called ‘temp’) x=1 will be used point to the first row of data; x=25 will be
used to point to the 25th row, which corresponds to the second day.

counter=counter+1; %increment counter

Here we increment the counter by 1. Because all MATLAB arrays are
indexed starting at 1 (and not 0, as in many other programming languages)
we place this counter line here so the counter=1 the first time through.

minT(counter)=min(temp(x:x+23)); %get minimum value

%over 24 hours x:x+23
The min function returns the minimum value over the range of values
indicated inside the parentheses. Since we only want to determine the min
for 1 day, the argument for the min function is temp(x::x+23). So, on the
first pass through the for loop, x=1, and the min function will read as
min(temp(1:24)). On the second pass through the for loop, x=25, so the
min function will read as min(temp(25:48)). You can see then how the for
loop stepping by 24 allows the min function to be calculated on one day’s
worth of data each time through the loop.

The value returned by the min function is saved in a matrix called minT.
The (counter) part allows us to save into specific columns in the matrix. The
first time through the loop, counter=1, so the value is saved in minT(1); i.e.
the first column. On the second time through the loop, counter=2, so the
value is saved in minT(2); i.e. the second column.

26
 maxT(counter)=max(temp(x:x+23)); % get max value
avgT(counter)=mean(temp(x:x+23)); % calc average
The max function calculates the maximum value, just as the min function
calculated the minimum value. The mean function calculates the average.

end
This is the end of the for loop. You can see that there are four functions in
this for loop. The first increments the counter, and the other three calculate
the minimum, maximum, and average values. The for loop continues until
x>=nsamples-23. Why do we have the -23 in here? Think about what
would happen in the min function. If on the last pass through x=nsamples,
the min function would read min(temp(nsamples:nsamples+23)). If the
temp matrix only has a total of nsamples, you can’t ask for data beyond that;
nsamples+23 would not be in the matrix.

Try running the program without the -23 in the for loop, and you get my
favorite MATLAB error:
??? Index exceeds matrix dimensions.

Error in ==> C:\tempanalysis.m

On line 11 ==> minT(counter)=min(temp(x:x+23)); %get

minimum value over 24 hours from x:x+23

This error can be fairly common in complicated programs. It is OK to

make errors, but it is also important to know where to look to fix them.
MATLAB tells you what line to look in for the error. As you can see it tells
use the problem is with the ‘min’ function. But, as you know, the fix is not
in the ‘min’ function. You need to fix the ‘for’ loop so that x doesn’t get so
big that x+23 will be beyond the end of the matrix. The point is, many
times ‘errors’ in the logic of the program will make themselves known at
other places in the program. Be prepared to think through what could be
causing the problem.

plot(minT,'blue'); % plot minimum daily values in blue

The ‘plot’ function plots the values of minT in blue.

hold on % hold on will overlay next plot

27
We want to plot the maximum and mean values on the same plot as well.
We use ‘hold on’ to tell MATLAB not to erase the figure when new plot
commands are issued. Without the ‘hold on’ command, the following ‘plot’
commands would overwrite previous ‘plot’ commands.

plot(maxT,'red'); % plot max daily values in red

plot(avgT,'green'); % plot average daily values in green
These are the last two plot commands that plot the maximum daily values in
red, and the average daily values in green. After the last line of the program,
it simply ends (you don’t have to write anything special to end the program),
and control returns to the MATLAB command window.

28
 3
Chapter

Basic File Input

Up until now we have used the Import Wizard to load data from files into
matrices. There are MATLAB commands to read files into MATLAB
matrices that can be used within programs. This is helpful for when you
have multiple data files to process. This chapter will use a basic MATLAB
command called importdata()to read in text files.

Create a basic text file with the following numbers in it:

Save it as test.txt in the C: directory.

Now try to import the data.

In the command window type:

>> test=importdata('test.txt')

29
What you will likely get is the following error:

??? Error using ==> importdata at 114

Unable to open file.

This is because MATLAB can’t find your file. MATLAB looks in the
current directory for the text file, and no further. It does not use the
MATLAB path to look for data files, just m-files.

Now, if you specify the directory in the call to importdata, it works.

>> test=importdata('c:\test.txt')

Note that any text characters you pass to MATLAB functions need to be
surrounded by single quotes as above. Otherwise, MATLAB will try to
interpret the text as another variable in the workspace.

Changing Directories
Another way to handle this is to change the current directory to the directory
that has your data in it prior to running the importdata command.

There are two ways to do this. The first is to use the Browse button on the
MATLAB toolbar, and navigate to the directory of interest.

30
The second way to do this is to use the change directory command, cd.

For example, type:

>> cd('c:\')

The current MATLAB directory will be changed to the c: directory.

Now, you can import the test.txt file without specifying the path.

>> test=importdata('test.txt')

Header Lines
Many files you will want to import will have header lines containing text.
Since MATLAB arrays typically hold only one kind of data (e.g. numbers or
text), you need to be able to deal with these when you import data.

If you look at the Help file for importdata, you will see several extended
forms of this command. One form is able to deal with header lines.

A = importdata(filename,delimiter,headerline)

delimiter is the character used to separate columns.

31
headerline is how many rows of text that serves as a header or labels.

Create new text file, this time with a header line. Let’s say you have
maximum temperatures for six days in a month. There are two columns,
one for day, and one for temperature. Make sure the values are separated by
spaces in the text file. Save this in the ‘c:’ directory as temperature.txt.

Day Temperature
10 22
11 23
12 24
13 21
14 20
15 25

To open temperature.txt into a variable, type:

>> temperature=importdata('c:\temperature.txt','
',1)

We are loading the values into a matrix called temperature. The delimiter is
set to a space (' '), because that is the character separating columns
(commas and tabs are also commonly used as delimiters). Headerline is set
to 1, because the first row contains the header text data. MATLAB returns
the following:

temperature =

data: [6x2 double]

textdata: {'Day' 'Temperature'}

colheaders: {'Day' 'Temperature'}

The matrix temperature that has been created is actually a MATLAB

structure, which itself contains three matrices. The first is data, which is
the day and temperature data from the first two columns. The matrices
textdata and colheaders contain the header text information (i.e.

32
the column headers). Structures are useful ways of organizing data of
different types, like character text and numbers.

MATLAB Structures
In the example above temperature is a structure containing three matrices.
The syntax for accessing particular matrices within a structure is as follows:
structure.matrix, where structure is the structure name, and matrix is the name of
the matrix within the structure you want to access.

For example, if we want to look at the temperature data in the Command

Window, we would type:

>>temperature.data

MATLAB would return

ans =

10 22

11 23

12 24

13 21

14 20

15 25

If we wanted to know what the column headers were, we would type:

>> temperature.colheaders

MATLAB would return:

ans =

'Day' 'Temperature'
33
If we wanted to plot the date versus the temperature, we can address each
part of the structure separately.

>>
plot(temperature.data(:,1),temperature.data(:,2)
);

Then to label the plot we could use the data from

temperature.colheaders.

>> xlabel(temperature.colheaders(1));

>> ylabel(temperature.colheaders(2));

We get this plot, which has the appropriate label for each axis.

25

24.5

24

23.5

23
Temperature

22.5

22

21.5

21

20.5

20
10 10.5 11 11.5 12 12.5 13 13.5 14 14.5 15
Day

34
File Open Example
For Example 3-1 we will build on Example 2-6 and modify it so that it can
automatically load and process multiple files within a directory.

We will make three text files with temperature data in them as before (called
temperature1.txt, temperature2.txt, and temperature3.txt). Place these all in
one folder called Example31 (with no other text files).

temperature1.txt

Day Temperature
10 22
11 23
12 24
13 21
14 20
15 25

temperature2.txt

Day Temperature
10 25
11 26
12 22
13 19
14 20
15 31

temperature3.txt

Day Temperature
10 12
11 13
12 14
13 11
14 10
15 15

35
Example 3-1

% Example 3.1
% M-file to calculate minimum, maximum, & average temperatures
% This program assumes there are multiple .txt files with
% temperature data in the same folder.

path='c:\Example31'; % this is the directory where we have the

% temperature files
cd(path); % change directory to directory with temperature data
files=dir('*.txt'); % get all .txt filenames in the directory
nfiles=length(files); % number of files to process

%loop through files

for x=1:nfiles

% Load data from current file

% Note that files is a structure.
% Use files.name to get filenames.

temperature=importdata(files(x).name,' ',1);

%get minimum value from 2nd column

minT(x)=min(temperature.data(:,2));

% get max value

maxT(x)=max(temperature.data(:,2));

% calculate average
avgT(x)=mean(temperature.data(:,2));

end

Program Description

path='c:\Example31'; % this is the directory where we have the

% temperature files
cd(path); % change directory to directory with temperature data
A variable called path is created and hard-coded with the name of the
directory that contains the temperature.txt files, and then cd is used to
change the current directory to that directory. It is useful to give variable
names even when hard-coding (meaning typing in a name or value that
36
could change in the future), because it allows you to easily modify the
program at a later date. It is also useful to place these at the top of the m-
file where you can easily find them..

files=dir('*.txt'); % get all .txt filenames in the directory

The dir command is used to find all file names with the extension .txt.
The asterisk (*) serves as a wildcard, meaning that any filename that ends in
.txt will be found. The files are returned in a structure that is named files. If
we examine this structure after running the program, by typing into the
Command Window:

>> files

We get the following:

files =

3x1 struct array with fields:

name
date
bytes
isdir
datenum

This tells us files is a 3x1 structure array. The 3 is because we put three .txt
files in the directory. Here we are just concerned with the file names. To
look at the file names we can type into the MATLAB command window:

>> files.name

We get the following returned:

ans =
temperature1.txt
ans =
temperature2.txt
ans =
temperature3.txt

37
These are the names of all of the *.txt files in the current directory. You
need to be careful when using this command, because the filenames are
retrieved in alphabetical order. Don’t count on them being in chronological
order.

nfiles=length(files); % number of files to process

The length command is used to determine the number of files found by
the dir command. This will be used in a for loop to individually load and
analyze each of the files.

for x=1:nfiles
The for statement will increment from 1 to nfiles. In this case nfiles=3,
because we placed three .txt files in the directory.

temperature=importdata(files(x).name,' ',1);
The importdata command is used as before. This time we need to use
the variable x to get the file name from the files structure. The first time
through the loop, x=1, so files(1).name will be ‘temperature1.txt’. The ' '
tells importdata that the text files are space-delimited. The 1 means that
the first row is a header. The file is then loaded into the structure called
temperature

minT(x)=min(temperature.data(:,2));
maxT(x)=max(temperature.data(:,2));
avgT(x)=mean(temperature.data(:,2));
The structure temperature holds both the header text as well as the date and
temperature data, as we saw above. The min function finds the minimum
value from the second column of temperature.data, which contains the
temperature readings. The max function finds the maximum value, and the
mean function calculates the average.

The values are stored in new matrices called minT, maxT, and avgT, which are
incremented with the x. So, minT(1) holds the minimum value of the second
column in the file temperature1.txt. avgT(2) holds the mean value of the
second column in the file temperature2.txt.

end

38
The end statement is the end of the for loop. When the end is reached,
the program execution returns to the for line, and increments the value of
x. This way each file is opened in turn and analyzed.

Next Steps
You can see the power of this simple routine. It can go through a directory
with thousands of files, determine their names, open them, and process
them. Simply by changing the path variable setting, any folder could be
analyzed.

There were undoubtedly bits of the examples here that did not make sense.
The structure returned by dir was not fully explained. Check out the
MATLAB Help file to learn what data the additional members of the
structure hold. Explore these in the command window. These other
structure members can come in handy when sorting through large numbers
of data files.

There are also many other ways to import data into MATLAB. We will
explore some of these in later chapters, but it is also a good idea to read this
section of the MATL

39
 4
Chapter

Matrix Manipulation
Up until now we have just created matrices by either using MATLAB’s
colon operator or by reading in text files. There are many times you will
want to add or delete rows or columns. This chapter covers basic matrix
manipulation in MATLAB as well as some techniques for using data
returned from one function to index into a matrix.

Deleting Rows
Let’s create a simple [5x3] (5 row by 3 column matrix).

a=[1 2 3; 4 5 6; 7 8 9; 10 11 12; 13 14 15];

Here we have directly entered the data manually into an array. Values in
columns are separated by spaces (you could also use commas), and rows are
separated by semicolons. If we look at the value in the array editor we see
our five rows by three columns.

40
To delete the second row, type:

>> a(2,:)=[];

Here we use the Null matrix operator (closed square brackets) to delete the
part of the matrix specified by (2,:), which means the 2nd row and all of the
columns. In the Array Editor window you can see that the second row has
been deleted.

Deleting Columns
Deleting columns is similar to deleting rows.

Start with the same array as before:

>>a=[1 2 3; 4 5 6; 7 8 9; 10 11 12; 13 14 15];

Now, let’s delete the third column. Type

>> a(:,3)=[];

This tells MATLAB to delete ([]) from the matrix a, all of the rows (:) from
column 3. The data in the array editor now look like:

If you get real angry at MATLAB, you can delete an entire matrix.

41
>> a=[];

This is actually useful when you want to make sure a matrix is empty before
filling it with data again. Say you had loaded a matrix with 10 rows of data.
Then you start processing another dataset and load only 5 rows of data into
the same matrix. You will still have the last 5 rows from the first set of data
you loaded. So, you need to clear the matrix before loading the second set.

Concatenating Matrices
Concatenating means combining. Say you have to matrices that you want to
combine into one. Perhaps they were data files from different years that had
the same format, and you want to put them all together.

First let’s create two matrices

>> a=[1:10];

>> b=[20:2:30];

Matrix a.

Matrix b.

42
Matrix a goes from 1 to 10, counting by 1. Matrix b goes from 20 to 30,
counting by 2. So, matrix a has 10 values, and matrix b has 6 values.

Let’s concatenate them into a new matrix (c), which should end up with 16
values.

>> c=[a b];

Now, if you look at c in the workspace, you will see it is 1 row by 16

columns.

Matrix c, formed by concatenating matrix a and b.

Again, we have used the square brackets to manipulate matrices. This time
we told MATLAB to put a and b into the same matrix. The space means to
put them in the same row.

You could also use the function horzcat to accomplish the same thing.

>> c=horzcat(a,b);

If we wanted them in separate columns, we would use the semicolon

operator (or the function vertcat).

Try concatenating them by putting them into different columns:

>> c=[a;b];

??? Error using ==> vertcat

CAT arguments dimensions are not consistent.

43
You get an error saying that the dimensions are not consistent. What
happened here was that you were trying to make a new matrix where the
first row was a and the second row was b. The problem is that a had 10
values, but b only had 6 values. Standard MATLAB arrays do not allow
missing values (unlike most spreadsheet programs).

Transpose Operator (‘)

Many times you will want to transpose rows and columns. Since many
functions by default operate on columns of data, it is often useful to be able
to transpose rows and columns if you import data that are organized into
rows.

For example, let’s create a [2x10] matrix that we can flip using some of the
techniques we just learned.

>> a=[1:10];

>> b=[21:30];

>> c=vertcat(a,b);

So, matrix a is 1 row x 10 columns and counts from 1 to 10. Matrix b is also
1x10, and counts from 21 to 30. We use the vertcat function to create
matrix c, which is now 2 rows x 10 columns. Below you can see these values
in the Array Editor window.

To transpose this array. type:

>> d=c';

Now d is 10 rows by 2 columns.

44
Flipping (flipud and fliplr)
Another useful technique is flipping a matrix. MATLAB provides flipud
(flip up-down) to flip a matrix along rows, and fliplr (flip left-right) to
flip a matrix along columns.

For example, if we take c from above, and use fliplr, we get:

>> e=fliplr(c);

Now, instead of starting at 1, the matrix starts at 10 and counts down across
columns.

Sorting (sort and sortrows)

MATLAB has a basic function for sorting data. For example, say we create
a matrix of random numbers ranging from 0 to 10.

>> r_unsorted=rand(10,1)*10;

We use rand to create 10 rows by 1 column of random numbers. The rand

function returns values between 0 and 1, so to get the value to range
between 0 and 10, we multiply the matrix by 10.

45
List of 10 random numbers between 0 and 10.

If we want to sort these numbers in ascending order (smallest to largest), we

type:

>> r_sorted=sort(r_unsorted);

By default, the sort command sorts in ascending order. Notice how we

are using matrix names (like r_sorted) that are more descriptive of the data
they contain.

You can see the MATLAB help file for other versions of sort (like
descending). Often you have a two-dimensional matrix that you want to
sort based on the values in one column, but keeping the rows together. The
MATLAB function sortrows does this.

Create a 5x4 matrix of random numbers from 0 to 100.

>> r_unsorted=rand(5,4)*100;

46
Now, let’s sort by rows in ascending order on the first column.

>> row_sorted=sortrows(r_unsorted);

Matrix Math (not really)

You will often want to perform math on matrices where you add, multiply,
or divide on a cell-by-cell basis. Since MATLAB was originally written to
perform true matrix math (in the linear algebra sense), the basic operators
for multiplying (*), dividing (/), power (^), and others perform linear algebra
on the matrices. When you want to perform a cell-by-cell mathematical
operation, you typically need to put a period in front of the operator.

For example, create two arrays:

>> a=[1:5];

>> b=[2:2:10];

So, a = [1, 2, 3, 4, 5]; b=[2, 4, 6, 8, 10]. To multiply these together, cell-by-

cell:

>> c=a.*b;

Now c=[2, 8, 18, 32, 50];

47
If you tried using just * (without the period), you would get an error in this
case.

File Open with Concatenation Example

For Example 4-1 we will build on Example 3-1 and modify it so that it
concatenates the temperature values together.

Remember, these are the data files we had for Example 3-1.

temperature1.txt

Day Temperature
10 22
11 23
12 24
13 21
14 20
15 25

temperature2.txt

Day Temperature
10 25
11 26
12 22
13 19
14 20
15 31

temperature3.txt

Day Temperature
10 12
11 13
12 14
13 11
14 10

48
15 15

Example 4-1

% Example 4.1

% This program assumes that there are multiple

% .txt files with temperature data in the same folder.
% This m-file will open them and concatenate
% them, and then calculate the min, max, and
% mean temp of the entire matrix

path='c:\Example31'; % directory with temperature files

cd(path); % change directory to directory with temp data
files=dir('*.txt'); %get all .txt filenames in the directory
nfiles=length(files); %number of files to process

all_temp=[]; %creates matrix with no values

%loop through files
for x=1:nfiles
% Load data from current file
% Note that files is a structure.
% Use files.name to get filenames.
temperature=importdata(files(x).name,' ',1);

%concatenate 2nd column onto all_temp

all_temp=[all_temp temperature.data(:,2)];
end
%get minimum value from 2nd column
minT=min(all_temp);
% get max value
maxT=max(all_temp);
% calculate average
avgT=mean(all_temp);

Program Description

path='c:\Example31'; % directory with temperature files

cd(path); % change directory to directory with temp data
files=dir('*.txt'); %get all .txt filenames in the directory
nfiles=length(files); %number of files to process
The first four lines are the same as Example 3-1. They change the directory
to the directory containing the temperature files, load the file names into the
files structure, and uses length to determine how many files there are.
49
all_temp=[]; %creates matrix with no values
This command creates a new empty matrix called all_temp. We need to do
this because later on we concatenate to this matrix within the loop. If the
matrix did not exist the first time through the for loop, there would be an
error.

%loop through files

for x=1:nfiles
This is the same for loop we had in Example 3-1. It says loop through the
files from 1 to nfiles.

temperature=importdata(files(x).name,' ',1);
This is the same line as in Example 3-1. It imports the data from the file
where the filename is in the files structure. The second argument indicates
the file is space-delimited, and the third argument indicates that there is 1
header line.

%concatenate 2nd column onto all_temp

all_temp=[all_temp temperature.data(:,2)];
This is the important line. We are collecting each column of the temperature
data in a matrix called all_temp. On the first time though, all_temp is initially
an empty matrix. We want only the temperature data (not the date data in
the first column), so we concatenate the second column from the temperature
structure (temperature.data(:,2)) onto the existing all_temp
matrix. On the second time through the all_temp matrix has one column of
data, to which we concatenate the temperature data from the next file, and
so on.

end
This is the end of the for loop. Notice that we have not done any of our
minimum, maximum, or mean calculations inside the loop. That is because
we are loading all of the temperature values into one matrix first.

%get minimum value from 2nd column

minT=min(all_temp);
% get max value
maxT=max(all_temp);
% calculate average
avgT=mean(all_temp);

50
After the loop all of the temperature values are in the all_temp matrix, so we
can use the simple min, max, and mean functions to calculate the values
for each column. Compare this to what we did in Example 3-1. There are
many ways to achieve the same end in programming, and usually it doesn’t
matter which route you take (as long as you get there).

51
 5
Chapter

Introduction to Time Series

Analysis
This chapter introduces the basics of sampling analog signals and concludes
with the introduction of MATLAB functions. Even if you know basic
sampling theory, read the end of the chapter where MATLAB functions are
described, because functions are valuable for expanding the basic MATLAB
toolboxes with custom functions.

Analog and Digital Signals

We live in an analog world. That means that everything changes
continuously. The air temperature changes continuously. The tides change
continuously. Time moves continuously. Computers are digital—meaning
that they operate on discrete data points. Therefore, to use computers to
process data, the analog signals must be converted to discrete digital
samples. The digital sampling of any process is simply a means of
converting an analog (continuous) signal into a series of discrete data points.
This chapter introduces time series analysis by discussing the fundamentals
of digital sampling.

The two fundamental aspects of digitization are the sample rate and the
sampling resolution. They are important to understand, because it is
important to understand the limitations and potential pitfalls of digital
sampling.
52
Sampling Rate
This plot shows a continuous tidal signal, which cycles once per day (Fig. 5-
1).

0.5
Tidal Height (m)

-0.5

-1
0 5 10 15 20 25 30 35 40 45 50
Time (h)

Figure 5-1. Time series showing tidal height continuously. This is an analog
signal.

Tides can be measured sampled by recording water depth every hour (Fig. 5-
2). In this example, the sample rate is 1/hour. The sample period is
1/sample rate; and is thus, 1 hour. This digital signal captures tidal variation.

0.5
Tidal Height (m)

-0.5

-1
0 5 10 15 20 25 30 35 40 45 50
Time (h)

Figure 5-2. Digital sampling of the analog signal from Fig 5-1. Each dot
shows a digital sample measured hourly.

53
Aliasing
Natural signals are never so clean. There may be variation in water height
due to waves, which vary on a shorter time scale than the tide we are
interested in measuring. If we ignore this we run the risk of collecting
erroneous data.

Figure 5-3 shows a more realistic dataset for what water depth might look
like when there are waves.

1.5
Tidal Height (m)

0.5

-0.5

-1
0 10 20 30 40 50
Time (h)

Figure 5-3. Tidal time series with variation in height caused by waves.

If we collect data once every hour, we end up collecting erroneous data with
respect to the tidal signal we are characterizing (Fig. 5-4).

54
 2

Tidal Height (m) 1.5

0.5

-0.5

-1
0 10 20 30 40 50
Time (h)

Figure 5-4. Digital sampling of the time series from Fig. 5-3 with digital
samples measured every hour.

This is more obvious if we connect the sampled data points with a line.
There are oscillations in the measured time signal that were not present in
the original signal. This type of undersampling is called aliasing. Aliasing is
bad.

1.5
Tidal Height (m)

0.5

-0.5

-1
0 10 20 30 40 50
Time (h)

Figure 5-5. Aliasing as a result of sampling an analog signal that varies faster
than the digital sampling rate.

55
Avoiding Aliasing
Nyquist Theorem
It is usually (but not always) easy to avoid aliasing. The first rule of thumb
(the Nyquist Theorem) is that you need to sample the data at least twice as
fast as the highest frequency in the system. For example, if you know that
the highest frequency of change in the system is once per hour, you need to
sample at least twice per hour. Indeed, you are better off sampling at an
even higher rate to give yourself some room for error.

Put another way, the sampling period needs to be at least half of the shortest
period of change in the data. So, if you know that the shortest period of
change of the signal is 1 day, you need to sample at least every 0.5 days.
Anti-Aliasing Filters
The second way to avoid aliasing is to use an anti-aliasing filter on the signal
from the transducer before you digitize the signal. A low-pass filter will let
frequencies lower than the cut-off frequency through the filter. If you have
a low-pass filter with a cut-off of 10 Hz (10 cycles per second), you need
only sample the filtered signal at 20 Hz (20 samples per second). In our tide
example, if we had an analog pressure transducer reporting the water depth,
we could (in theory) use a low-pass filter to filter out the wave ‘noise’ from
the tidal signal before digitizing.

A second way to do this is to sample the water depth at a high enough rate
to capture the variation due to waves (say once per second), and then
average the resulting data to yield the tide signal. Simple averaging is a
method of low-pass filtering. We will cover filters in more detail in later
chapters, as there are more sophisticated techniques that yield filters with
more desirable properties.

There are times when it is not possible to use an anti-aliasing filter, such as
when you are sampling populations of animals and it is impossible to do it
continuously. Consider monitoring changes in zooplankton populations at
one location over time with a plankton net tow. Each plankton net tow
might last 10 minutes with 20 minutes in between tows to rinse and prepare
the samples. In this case, there is no way to know if the plankton population
signal is being aliased. You would have to develop a new sampling protocol
to be sure to avoid aliasing. Such a protocol might involve some sampling at
high rates to understand the high frequency variability. In other cases, you
may know something fundamental about the system that limits the rate of
variability. For instance, if you are studying nesting birds, and count the
56
number of eggs laid, you may know that the biology of bird reproduction
does not allow more than one clutch of eggs to be laid per day. This
fundamental knowledge can be used to constrain sampling rates.

Sampling Resolution
Now that you know how often to sample, you need to also consider the
resolution with which you record the measurements. Up until now we have
ignored how numbers are stored on computers. There are several
computer data formats, which are directly tied to digital sampling
considerations and to other issues such as file formats.

Bits
Computers store information as bits whose value can be either 0 or 1. This
is a binary code—meaning that each digit can have only two values. The
reason for using a binary code is that data can be transmitted using switches
(a signal is either on or off). Binary coding is a base-2 numbering system.
To code numbers that are higher than one, you must string bits together.
Table 6-1 illustrates how 8 bits can be used to represent values from 0-255 (a
total of 256 values).

57
Table 5-1. Binary numbering system using 8 bits to represent an integer
value.

Binary Number Integer

(8 bits or 1 Byte) Value

00000000 0
00000001 1
00000010 2
00000011 3
00000100 4
00000101 5
... ...
11111111 255

58
Digital Resolution
If you sample a signal with 8-bit resolution, it means that 8-bits are used to
encode the value at each time point the signal is sampled (Figure 5-6). This
gives you 256 different values that can be used.

Time Value
Point

1 0

2 39

3 75

4 103

5 121

6 128

7 121

Figure 5-6. 8-bit encoding of an analog signal. At each time point the level
is measured and encoded as an 8-bit number.

While this looks good, you can see that there will be values that can not be
encoded with 8-bits. For instance, say your measurement from Figure 6-6 at
time point 2 was 39.2. There is no way to encode this as an 8-bit integer.
So, it will be rounded to 39. This rounding is called quantization error, and
it raises two issues: how to increase measurement resolution and how to
represent decimal values.

Measurement Resolution
Since the number of values that can be coded in binary is 2bits, to increase the
resolution increase the number of bits (Table 5-2). For example, a 12-bit
number can encode 212, or 4096, different values.

59
Table 5-2. Relationship between number of bits and the number of values
that can be represented.

Bit-Resolution Number of Values (2bits)

8 256
12 4096
16 65536
32 4295x106

Decimal Numbers
Decimal numbers can be represented in different ways. For instance some
bits can be used to represent the integer part of a decimal number, and the
remaining bits can be used to represent the fractional part of a decimal
number. A more commonly used format is floating point, which is the
default MATLAB data type. Double precision floating point uses 64 bits to
encode a number, with 15 decimal digits and an exponent. Double-
precision floating point numbers will not code more than 15 decimal places.
However, most data acquired from instruments is saved at a lower
resolution, like 16-bit integers. When it is imported into MATLAB, it is
usually encoded as double precision data. This essentially decreases the
amount of rounding error that occurs in subsequent calculations (at the
expense of memory).

Reading Data Files with fread()

We previously used the importdata() command to load text files. This
works with files of known data formats (like text and wav files). Text files
are an inefficient way of storing data, because the data values are actually
stored as text characters, and each of those characters is coded by the ASCII
character code. A more efficient format is binary, where data are stored as
the binary 0’s and 1’s that make up the data. Thus, in an 8-bit binary file
each value is represented by 8-bits, and these values are simply strung
together. It is crucial to know the bit-resolution in opening a binary file.
For instance, if data are saved as 8-bit values, but they are read as 16-bit
values, the numbers will be incorrectly interpreted, as in the example below.
In this example, the spaces are used for clarity; there are no spaces in a
binary data file.

60
8-bit Binary Data: 00000001 00000001 00000001
00000001

Integer Value: 1 1 1 1

If the same data file is read in as 16-bit binary data, the values are
misinterpreted.

16-bit Binary Data 0000000100000001

0000000100000001

Integer Value 257 257

MATLAB has the flexibility to load data files with arbitrary data formats.
For instance, if you need to load a 16-bit binary file you can use MATLAB’s
basic file reading functions.

Reading files is a three-step process.

1. Open file with fopen(). The returns what is known as a file ID

(FID) that is used in subsequent function calls. The purpose of the
FID is to allow multiple files to be open at the same time.

2. Read data with fread().

3. Close file with fclose().

File Reading Example:

We have a data file called tides.I16 in the c: directory, which has tidal data
encoded as a 16-bit integer binary file. The following code will load this file
into a variable called tides.

% 16-bit file read example

%name of file
filename='c:\tides.I16';

%Step 1: Open file for reading as binary

61
fid=fopen(filename);

%Step 2: Read data as 16-bit integer

tides=fread(fid,'int16');

%Step 3: Close file

fclose(fid);

Table 5-3. Common MATLAB data types for use with fread.

Data Type MATLAB fread argument

8-bit integer ‘int8’

16-bit integer ‘int16’

32-bit integer ‘int32’

32-bit floating point ‘float32’

64-bit floating point ‘float64’ or ‘double’

62
MATLAB Functions
RMS
One way to characterize signal levels of oscillating signals is to calculate the
RMS level. The reason is if a signal oscillates around a zero value, the
average value of the signal will be zero, no matter what the amplitude. The
RMS technique overcomes this problem by squaring all of the values before
taking the average.

RMS stands for root-mean-square. Simply put, first you square the data (to
change negative values to positive values), then you take the mean of those
squared values, and finally take the square root of the result to get back to
the original units. There is no MATLAB function to calculate rms levels, so
let’s write a simple one that we can use in the future.

This will be a MATLAB function, rather than just a plain m-file. A

MATLAB function is an m-file which returns a value or values to the calling
function. You have been using MATLAB functions all along in your m-
files, such as min, max, and mean. Every time you call one of these
functions you pass it the data to work on and the function returns the
calculated values. One important thing to know about functions is that they
have their own memory space that is separate from the MATLAB
workspace memory space. This means that the function can not ‘see’ any of
the variables in the workspace. That is why the data that will be operated on
needs to be passed in to the function.

Type in the following rms function and save it in a new folder called
MyToolbox as rms.m. The function file name has to be the same name as the
function. This is where you will save all of the MATLAB functions you
create, so that you can easily find them and use them. Remember to add
MyToolbox to the MATLAB Path (see Chapter 1).

function h=rms(data)
% h=rms(data)
% calculates root-mean-square
% of input matrix data

% Square the data using cell-by-cell multiplication

datasquared=data.*data;

63
% Calculate the mean of the squared data
mean_ds=mean(datasquared);

% Calculate the square-root of the mean

% h is what will be returned to the
% calling function
h=sqrt(mean_ds);

Program Description

function h=rms(data)
% h=rms(data)
% calculates root-mean-square
% of input matrix data

The first line of a MATLAB function starts with the word function followed
by a space and then the function declaration. In this case the function
declaration is h=rms(data). This determines what you will type from
the MATLAB command line or other m-files to use the rms function. The
matrix data will be passed from the calling function (i.e. the command line or
other m-file). The matrix h will be returned by the rms function, so we
want to make sure that we put the final result into h.

The three commented lines following the function declaration are important.
They will be the lines returned to the screen when you type:

>> help rms

MATLAB returns:

h=rms(data)
calculates root-mean-square
of input matrix data

These are useful in that they describe the syntax for how to use the rms
command, and what the function does.

64
% Square the data using cell-by-cell multiplication
datasquared=data.*data;
First we square the data by multiplying the data matrix by itself. Notice that
the .* function is used, because we want to multiply cell-by-cell.

% Calculate the mean of the squared data

mean_ds=mean(datasquared);
The mean of the squared data is calculated. Remember this will operate on
columns of data, so if a two-dimensional matrix is passed to the rms
function it will operate on columns.

% Calculate the square-root of the mean

% h is what will be returned to the
% calling function
h=sqrt(mean_ds);
The square root of the mean is then calculated and the result stored in h.
This is what is returned to the calling function.

Using RMS Example

Let’s use the new rms function to calculate the rms of the tidal data we
opened earlier. Simply add one line to the end of the file.

% 16-bit file read example with rms calculation

%name of file
filename='c:\tides.I16';

%Step 1: Open file for reading as binary

fid=fopen(filename);

%Step 2: Read data as 16-bit integer

tides=fread(fid,'int16');

%Step 3: Close file

fclose(fid);

tideRMS=rms(tides);

Program Description

65
tideRMS=rms(tides);
The rms function is used to calculate the rms of tides. The data we pass to
the rms function is the tides matrix. We store the resulting rms value in the
matrix tideRMS. Notice that in the rms function, we used the matrix h to
store the rms value. The calling function does not need to use the same
argument name or returned matrix name as the rms function.

Look in the MATLAB workspace. Notice that there are no matrices called
data, data_squared, mean_ds, or h. The matrices created in the rms function
are only in memory temporarily; they are not saved in the MATLAB
workspace.

66
 6
Chapter

Fourier Transform
This chapter introduces one of the most powerful (and old) techniques for
analyzing the periodicity of signals in a time series. We will not go through a
derivation of the Fourier Transform, but instead present an intuitive
description of what it is and how it is used in MATLAB. Other texts (e.g.
Digital Signal Processing for Scientists and Engineers) may be consulted for
mathematical details of how the Fourier Transform works.

The Fourier transform converts a signal from the time domain into a signal
in the frequency domain. When we use the word ‘domain’ we are referring
to the units on the x-axis. The term ‘transform’ means that you can convert
in both directions (Fig. 7-1).

Time Fourier Transform Frequency

Domain Domain
Amp Amp
Inverse Fourier
Transform
Time (s) Frequency (Hz)

Figure 6-1. Schematic of process of Fourier Transform converting from time domain into
67
frequency domain, and the reverse process, the Inverse Fourier Transform. Each plot has
the signal amplitude on the y-axis.

The Fourier Transform is one of the most commonly used tools for
analyzing signals. The idea behind it is that any signal can be broken down
into the sum of sine waves with different amplitudes and phases.

To understand the Fourier Transform we will create some sine wave signals
in MATLAB and then analyze them with the Fast Fourier Transform (FFT).
The FFT is a mathematically efficient way of calculating the Fourier
Transform.

1. To create a sine wave we first need to decide on our sample rate. Let’s
create a 1 s long signal with a 100 Hz sample rate. Note that we subtract off
one speriod from the dur when we create the signal, since we start counting
from 0.

srate = 100; % 100 Hz sample rate

speriod = 1/srate; % sample period (s)
dur=1; % duration in seconds
t=[0:speriod:dur-speriod];% Create a signal from 0 to 1 s.
The sample period is 0.01 seconds.

2. Now let’s make a sine wave with a frequency of 2 Hz (i..e. 2 cycles per
second). Note that since t is a matrix, we use ‘.*’ in the equation with t.

freq=2; % frequency of sine wave in Hz

s=sin(2*pi*freq.*t); % calculate sine wave.
plot(t,s); % plot using t as the time base

In the plot see that we have exactly two sine wave cycles in one second.

68
 1

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 6-2. Sine wave that is 1 s in duration with a frequency of 2 Hz.

3. Now, let’s look at this signal with the FFT. First, we use the fft function
to calculate the Fourier Transform.
sfft=fft(s); % calculate Fast Fourier Transform

The first thing to note is that the values are returned as a complex number
which contains both the frequency components and the phase components.
For now we are just going to look at the frequency components by using the
abs (absolute value) function to return the magnitude of the FFT.

sfft_mag=abs(sfft); % take abs to return the magnitude

of the frequency components
plot(sfft_mag); % plot fft result

If you examine the sfft_mag matrix in the workspace, you will see it has the
same number of points (100) as the original time series. The plot shows two
peaks. If we zoom in on the first peak we see that it is plotted at index 3.

69
The second peak is the mirror image of the first peak rotated around the
Nyquist frequency. Since our sample rate is 100 Hz, the Nyquist frequency
would be at 50 Hz. Since the first index (or bin) in the FFT corresponds to
a frequency of 0 Hz, the index at 3 corresponds to 2 Hz.

60

50

40

30

20

10

0
0 20 40 60 80 100 120

Figure 6-3. Plot of magnitude of FFT of 2 Hz signal. Note that the x-axis values
correspond to the indices of the sfft_mag matrix, they do not correspond to frequency.

70
 50

45

40

35

30

25

20

15

10

0
1 2 3 4 5 6 7

Figure 6-4. Zoomed in view of Figure 7-3 showing peak at index 3.

Examine the magnitude of the peak at index 3. Notice that it is 50. If you
look back at Figure 6-2, you will see that the peak amplitude of the sine wave
was 1, not 50. Because of the way MATLAB returns the FFT, it needs to be
scaled by the number of points used in the FFT.

fftpts=length(s);
hpts=fftpts/2; % half of the number points in FFT
sfft_mag_scaled=sfft_mag/hpts; % scale FFT result by
half of the number of
points used in the FFT
plot(sfft_mag_scaled); % plot scaled fft result

Now when we plot the results, the y-axis is properly scaled to 1 (Fig. 6-5).

71
 1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 20 40 60 80 100 120

Figure 6-5. Amplitude-scaled FFT plot.

Figure 6-5 is now properly scaled for the amplitude, but we still do not have
a proper frequency scale. So, we need to create one. First, we need to know
the frequency resolution (or binwidth) we obtained from the FFT.

frequency resolution=sample rate/points in FFT

In our example:

binwidth = srate/fftpts; % 100 Hz sample rate/ 100 points

in FFT

So, our binwidth (or frequency resolution) is 1 Hz. Since the first bin in the
FFT result is 0 Hz, we create a matrix for the frequency axis starting at 0.
The resulting plot, now has both axes correctly scaled (Figure 6-6).

f=[0:binwidth:srate-binwidth]; % frequency scale goes

72
 from 0 to sample rate.
Since we are counting from zero,
we subtract off one binwidth to
get the correct number of points
plot(f,sfft_mag_scaled); % plot fft with frequency axis
xlabel(‘Frequency (Hz)’); % label x-axis
ylabel(‘Amplitude’); % label y-axis

0.9

0.8

0.7

0.6
Amplitude

0.5

0.4

0.3

0.2

0.1

0
0 10 20 30 40 50 60 70 80 90 100
Freuqency (Hz)

Figure 6-6. Scaled FFT of 2 Hz sine wave.

Since, the mirror image contains unnecessary information, we typically just

plot the first half of the data (Fig. 6-7). We already calculated hpts, which tells
us half of the number of points in our FFT, so we can use this to plot the
first half of each matrix.

%plot first half of data

plot(f(1:hpts),sfft_mag_scaled(1:hpts));

xlabel('Frequency (Hz)');
ylabel('Amplitude');

73
 1

0.9

0.8

0.7

0.6
Amplitude

0.5

0.4

0.3

0.2

0.1

0
0 5 10 15 20 25 30 35 40 45 50
Frequency (Hz)

Figure 6-7. First half of FFT magnitude results plotted, showing peak at 2
Hz.

74
Measuring Amplitudes of Specific Frequencies
The frequency resolution of the FFT is determined by the number of points
in the FFT.

As we have seen, the width of each bin is:

binwidth=sample rate/FFT points

Example,

If a 1000-point FFT was performed on a signal sampled at 8000 Hz, then

each bin in the FFT would represent (8000/1000)=8 Hz.

If you expect a signal at 800 Hz, and you want to measure its magnitude,
then you would look in the 101st bin. This is because 800 Hz/8 Hz
binwidth=100. Since 0 Hz is the first bin, 8 Hz the second, 16 Hz the 3rd...
you would need to look in bin 101 (not bin 100).

Determining Peak Amplitude and Frequency

To find the peak amplitude use the max function so that it returns both the
index of the peak and the magnitude of the peak.

Using our 2 Hz sine wave example:

[magnitude, index]=max(sfft_mag_scaled(1:hpts))

Returns:

magnitude = 1

index = 3

So, to calculate the frequency of the peak signal, we need to look up the
index in our frequency scale variable.

maxfreq=f(index)

Happily, MATLAB returns 2 (Hz).

75
Additional Details
The number of points used in an FFT can be specified in the fft function.

If the number of points in the signal is longer than fftpts, MATLAB will
truncate the signal and only use the first fftpts.

If the signal is shorter than fftpts, MATLAB will zero-pad the signal before
calculating the FFT.

signal_fft=fft(signal, fftpts); %calculate FFT using fftpts

Additional Exercises

1. Create a 1 second long sine wave with a 1000 Hz sample rate and a
frequency of 10 Hz and amplitude of 1.

2. Create a 1 second long sine wave with a 1000 Hz sample rate and a
frequency of 100 Hz and amplitude of 1.

3. Add signals from part 1 and part 2 together.

4. Perform FFT on signal from part 3 and plot with appropriate axes.

5. Add random noise to the signal from part 3.

6. Perform FFT on signal from part 5, and plot with appropriate axes.

76
 7
Chapter

Digital Filters

Filters are used to remove unwanted frequencies from a signal. In the

previous chapter we saw how the Fourier Transform could be used to
analyze the sinusoidal frequencies that compose signals. In many cases there
may be frequencies in our signal that we want to filter from the signal, such
as we may have electrical noise at 60 Hz (or 50 Hz depending on where you
live) contaminating the signal. Filters can be used to remove these unwanted
signals. Digital filtering is a vast subject, and like the Fourier Transform, it is
the subject of many signal processing textbooks. The purpose of this
chapter is to introduce you to two basic filter types and how to use them in
MATLAB.

The two filter types we will discuss are named the Finite Impulse Response
(FIR) filter and the Infinite Impulse Response (IIR) filter. While their names
may sound complicated, the implementation of these filters is
straightforward and involves nothing more than multiplication and addition.

77
Finite Impulse Response (FIR) Filters
Let’s start with an FIR filter that you are probably already familiar with, but
didn’t know it had such a fancy name: the moving average. In a moving
average a certain number of points in the time series are averaged together
over a sliding window to create a smoothed time series.

For example, say we have the following data (Table 7-1), and we wish to
calculate a 3-point moving average. We add the first three rows (4+2+1) to
get 7, which we divide by 3 to calculate the average. Then we slide the
window down one place, and add the second through fourth rows (2+1+5)
to get 8, which we divide by 3 to calculate the average.

Table 7-1. Three-point moving average.

Original 3-Point Sum 3-Point

Time Series Moving
Average

2 7 7/3=2.333

1 8 8/3=2.667

5 9 9/3=3

3 13 13/3=4.333

5 9 9/3=3

1 8 8/3=2.667

2 9 9/3=3

78
This is the general scheme of an FIR filter. The only difference is that in
FIR filter implementations we first multiply by a fraction (i.e. divide) and
then sum the results. For example, for a three-point moving average we
would multiply each value by 0.3333333 first and then sum the results. You
can see that this is mathematically the same as adding first and then dividing
by the number of values added together.

For our first FIR filter in MATLAB we will implement a 5-point moving
average, so we will multiply each value by 0.2 (=1/5) and then add the
results together. We will apply this filter to a random noise sequence, so we
can see the affect of the moving average.

The first step is to create an array with the factors we are going to use in the
moving average. These are referred to as filter coefficients . They are also
referred to as taps . We are going to use the MATLAB (and standard)
nomenclature and call these ‘b’. We then use the filter function to filter a
random time series called raw_data.

npts=1000;
b=[.2 .2 .2 .2 .2]; % create filter coefficients for 5-
point moving average
x=(rand(npts,1)*2)-1; % raw data from -1 to +1
filtered_data=filter(b,1,x); % filter using 'b'
coefficients
subplot(2,1,1); % 1st subplot
plot(x); % plot raw data
title('Raw Data');

subplot(2,1,2); % 2nd subplot

plot(filtered_data); %plot filtered data
title('Filtered Data');
xlabel('Time');

Program Description

b=[.2 .2 .2 .2 .2]; % create filter coefficients for 5-

point moving average
This creates an array with our filter coefficients. For a moving average the
same factor is repeated.

79
filtered_data=filter(b,1,x);
When using the filter function for FIR filters, the second argument is always
1. The reason for this will become clear later when we discuss IIR filters.
We pass it the filter coefficients ‘b’ and the data to be filtered ‘x’.

subplot(2,1,1); % 1st subplot

This example also demonstrates the use of subplot to create two plots in one
figure. The first two arguments are used to define the number of rows and
columns of plots there will be. The last number refers to which plot is the
active plot. In this example we are creating two rows by 1 column.

One can see how the moving average removes the peaks, and smooths, the
original time series (Figure 7-1).

Raw Data
1

0.5

-0.5

-1
0 100 200 300 400 500 600 700 800 900 1000

Filtered Data
1

0.5

-0.5

-1
0 100 200 300 400 500 600 700 800 900 1000
Time

80
Figure 7-1. Smoothing a random time series with a 5-point moving average. The moving
average removes the peakiness of the raw data.

What about the end bits?

You may have noticed in Table 7-1 that we have problems at the beginning
and ends of the time series where we do not have enough points to average
together. MATLAB deals with this by starting with the first value in time
series and multiplying it by the first filter coefficient to arrive at the first
value. The second value is the sum of the first value in the time series
multiplied by the first filter coefficient and the second value in the time series
multiplied by the second filter coefficient. Thus, in our 5-point moving
average example: filtered_data(2)= x(1)*0.2 + x(2)*0.2.

In general for our five-point moving average example:

filtered_data(n) = b(1)*x(n) + b(2)*x(n-1) + b(3)*x(n-2) + b(4)*x(n-3) +

b(5)*x(n-4)

You need to pay attention to this for large filters (filters with a lot of
coefficients), because it results in a time shift in the filtered signal.

Visualizing the frequency response of the filter

While you can generally see the result of the filter in Figure 7-1, it is difficult
to tell what frequencies were filtered out. The easiest way to visualize this is
to perform a Fourier Transform.

% Perform FFT on original and filtered signal

fftpts=npts; % number points in FFT
hpts=fftpts/2; % half number of FFT points

x_fft=abs(fft(x))/hpts; %scaled FFT of original signal

filtered_fft=abs(fft(filtered_data))/hpts; %scaled FFT of
filtered signal
subplot(2,1,1) %1st subplot
plot(x_fft(1:hpts)); %plot first half of data points
title('Raw Data');

subplot(2,1,2) %2nd subplot

plot(filtered_fft(1:hpts));%plot first half data points
title('Filtered Data');
xlabel('Frequency');

81
Program Description

fftpts=npts; % number points in FFT

hpts=fftpts/2; % half number of FFT points
Here we set up variables for the FFT.

x_fft=abs(fft(x))/hpts; %scaled FFT of original signal

filtered_fft=abs(fft(filtered_data))/hpts; %scaled FFT of
filtered signal
We have combined the FFT and associated commands from the previous
chapter into one line. First the FFT is calculated, then the absolute value is
taken to return the frequency magnitudes (and discarding the phase), then
the result is scaled by hpts.

The resulting plots (Figure 7-2) shows that the original signal has energy at
all frequencies. The signal filtered with the 5-point moving average shows a
reduction in energy at the higher frequencies. This makes intuitive sense,
because the filters smooths out the sharp excursions. This is a type of low-
pass filter, because it allows low-frequencies to pass through.

82
 Raw Data
0.08

0.06

0.04

0.02

0
0 50 100 150 200 250 300 350 400 450 500

Filtered Data
0.08

0.06

0.04

0.02

0
0 50 100 150 200 250 300 350 400 450 500
Frequency

Figure 7-2. Frequency domain plots from FFT of time series from Figure 7-1. The top
plot shows the original noise signal had energy at all frequencies. The bottom plot shows
that the 5-point moving average acts as a low-pass filter (and filters off high frequencies and
lets low frequencies through the filter).

More Reason to ♥ MATLAB !

Try experimenting with changing the number of points in the time series
and changing the number filter coefficients in the moving average.

Hint: an easier way to create an arbitrary number of filter coefficients for a

moving average filter is to use the ones function.

This creates 20 identical coefficients:

ntaps=20;
83
b=ones(1,ntaps)/ntaps; % create filter coefficients for
ntap-point moving average

Infinite Impulse Response (IIR) Filters

Infinite Impulse Response (IIR) filters are similar to FIR filters, but they also
include feedback. One of the advantages of IIR filters over FIR filters is that
they can achieve a much sharper filter with fewer filter coefficients. So, they
are more computationally efficient. They have some drawbacks as well—
because of the feedback IIR filters can become unstable and spiral out of
control (referred to as ‘blowing up’) for certain coefficients and inputs.

There are many types of IIR filters, usually named for some fantastic
mathematician. None are named for a biologist. They include Butterworth,
Chebyshev, Bessel, and elliptic. For this section we will just demonstrate use
of the Butterworth filter (mostly because we like the MATLAB name for it:
butter).

Follow me blindly here. We’ll first implement a Butterworth filter, look at

the result and compare it to what we got with the moving average filter.
Once we see what we get, we’ll see what computations are involved. You’ll
thank me later.

Let’s start with a noise signal and assume it is sampled at 1000 Hz and is
2000 points long (i.e. 2 seconds in duration).

% Butterworth IIR Filter

srate=1000; % sample rate

npts=2000; %npts in signal

Nyquist=srate/2; %Nyquist frequency

lpf=300; %low-pass frequency
order=5; %filter order

t=[0:npts-1]/srate; %time scale for plot

x=(rand(npts,1)*2)-1; % raw data from -1 to +1

84
[b,a]=butter(order,lpf/Nyquist); %create filter
coefficients
filtered_data=filter(b,a,x); % filter using 'b' and 'a'
coefficients

% Calculate FFT
fftpts=npts; % number points in FFT
hpts=fftpts/2; % half number of FFT points
binwidth=srate/fftpts;
f=[0:binwidth:srate-binwidth];

x_fft=abs(fft(x))/hpts; %scaled FFT of original signal

filtered_fft=abs(fft(filtered_data))/hpts; %scaled FFT of
filtered signal
subplot(2,2,1)
plot(t,x);
title('Raw Time Series');

subplot(2,2,3)
plot(t,filtered_data);
title('Filtered Time Series');
xlabel('Time (s)');

subplot(2,2,2)
plot(f(1:hpts),x_fft(1:hpts));
title('Raw FFT');

subplot(2,2,4)
plot(f(1:hpts),filtered_fft(1:hpts));
title('Filtered FFT');
xlabel('Frequency (Hz)');

Program Description

Nyquist=srate/2; %Nyquist frequency

lpf=300; %low-pass frequency
order=4; %filter order
This chunk of code creates all of the variables that we will use to calculate
the Butterworth filter coefficients. The low-pass corner frequency we will
use is set to 300 Hz. We will create a 4th order Butterworth filter. The
higher the filter order, the steeper the cutoff (and the higher the likelihood it
will become unstable). The 5-point moving average we ran before is a 4th

85
order FIR filter (the filter order is 1 less than the number of ‘b’ filter
coefficients).

[b,a]=butter(order,lpf/Nyquist); %create filter

coefficients
The butter function is used to calculate the filter coefficients. Note that this
returns both ‘b’ and ‘a’ coefficients. The ‘b’ coefficients are what we used
before in the moving average filter, and are known as the feed-forward
coefficients. The ‘a’ coefficients are the feedback coefficients. Both are used
in the filter.

If we look at the ‘b’ coefficients, we see that they are not all the same value:

0.167179268608490

0.668717074433960

1.00307561165094

0.668717074433960

0.167179268608490

filtered_data=filter(b,a,x); % filter using 'b' and 'a'

coefficients
The Butterworth filter uses the same filter function that we used for the FIR
filter. The only difference is in the coefficients used. The rest of the code is
the same as before, except this time we use appropriate time and frequency
axes, and create a 2x2 set of plots in one figure.

86
 Raw Time Series Raw FFT
1 0.08

0.5 0.06

0 0.04

-0.5 0.02

-1 0
0 0.5 1 1.5 2 0 200 400 600

Filtered Time Series Filtered FFT

2 0.08

1 0.06

0 0.04

-1 0.02

-2 0
0 0.5 1 1.5 2 0 200 400 600
Time (s) Frequency (Hz)

Figure 7-3. Raw time series (top) filtered with a 4th order Butterworth filter with a 300
Hz corner frequency (bottom). The time series are shown on the left and the frequency
domain plots are shown on the right.

Try This
1. Try changing the filter order to see the effect on the sharpness of the
cutoff.

2. Compare the sharpness of the Butterworth filter to the moving average

filter. Notice how the Butterworth filter does not show any ‘humps’ at
higher frequencies as you see in the moving average filter.

FDAtool: MATLAB has a Filter Design and Analysis Toolbox that can be
used to create and visualize filters. The ‘b’ and ‘a’ coefficients can then be
exported for use in an m-file. Just type:

87
>FDAtool

Figure 7-4. MATLAB’s FDAtool showing the calculation of a 5th order Butterworth
high-pass filter.

How Do IIR Filters Work?

In the MATLAB help file for filter, you will find a block diagram showing
how the filter function works (Figure 7-5). The input signal is x(m). It is
multiplied by each of the ‘b’ filter coefficients. In our 5-point moving
average example, each of the ‘b’ coefficients was 0.2. The output of the filter
function (essentially one big multiply and add function) is y(m). Thus, each
value of x(m) is multiplied by the appropriate filter coefficient and the
outputs are summed (indicated by summation: ∑).

88
In the IIR filter there is also feedback in which the ‘a’ feedback coefficients
are multiplied by the previous outputs and subtracted from the feedforward
results. Thus, each filtered output value of y can be calculated as:
y(n)=b(1)*x(n)+b(2)*x(n-1)+ ... + (feed-forward)
b(nb+1)*x(n-nb)

- a(2)*y(n-1) - ... - feedback

a(na+1)*y(n-na)

Figure 7-5. MATLAB’s filter function block diagram.

89
 8
Chapter

Automatic Signal Detection

This chapter will introduce basic techniques for automatically detecting
signals in noise. The focus will be on practical techniques, rather than on
signal detection theory. Performing signal detection is a lot like solving a
puzzle. Many times you can see the signal you are trying to detect in a figure.
In these cases, your brain is performing signal detection. While your brain is
not perfect, it is pretty good, and trying to figure out what your brain is
doing to find the signal can be a good first step towards developing a
program to perform the same functions.

We will focus on detecting signals along a single dimension (e.g. in a time

series or in the frequency domain). Similar techniques can be applied to
images, which are just two-dimensional signals.

Energy Detectors
The first type of signal detection we will implement is the energy detector.
Energy detectors use a threshold to determine when the signal of interest
goes above a certain level.

Energy Detection in the Time Domain

Energy detection in the time domain consists of the following steps:

Step 1: Filter signal to remove frequencies outside the range of the

frequencies contained in the signal of interest.

90
Step 2: Rectify signal.

Step 3: Filter rectified signal.

Step 4: Perform threshold detection.

Energy Detector Example

To show how this works by example we will create a sine wave embedded in
noise. The goal is to determine when the sine wave occurs in the noise
signal. This chunk of code creates a noise signal and embeds a short
duration sine wave in it at a random location (see top plot in Figure 8-1).

% Create Noise Signal and embed sine wave in it at a

random location

npts=5000; % # points in signal

srate=1000; % sample rate (Hz)
dur=npts/srate; % signal duration in sec

amp_n=3; % noise amplitude

amp_t=1; % sine wave amplitude
freq=100; % sine wave frequency
sinepts=400; % # points in sine wave

t=[0:npts-1]/srate; % signal time base

sine_t=[0:sinepts-1]/srate; % sine wave time base

noise=(rand(1,npts)-.5)*amp_n; %noise signal

sinewave=amp_t*(sin(2*pi*freq*sine_t)); % sine wave

% random location of sinewave

sinepos=floor(rand(1,1)*(npts-sinepts))+1;

signal=noise; % make signal equal to noise

endsine=sinepos+sinepts-1; %calc index end of sine wave

% add sinewave to signal starting at index sinepos

signal(sinepos:endsine)=signal(sinepos:endsine)+sinewave;

% Plot signal
subplot(5,1,1)

91
plot(t,signal);
hold on
%plot red dot on location of sine wave start
plot(sinepos/srate,0,'.r');
hold off
title('Original Signal');

Original Signal
5
0
-5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Step 1. Filter Signal
2
0
-2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Step 2. Rectify Signal
2
1
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Step 3. Envelope Signal
1
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Step 4. Threshold Signal
1
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s)

Figure 8-1. Original signal with sine wave embedded in a noise signal. The lower four
plots show the sequential signal processing steps used to detect the signal in the time domain.

Step 1. Filter the signal

The first issue when it comes to filtering the signal is what filter frequencies
do you use? In many situations you will know the frequency of the signal of
interest fairly well. For example, if you are designing an echosounder, you
will know the frequency of the signal that you broadcast from the
transducer, so you will know what filter frequencies to use to detect the
echoes. Or, you might be trying to detect the vocalization of an animal in a
noisy situation. If you have several recordings of the animals in quiet
92
situations (i.e. where you have a high signal to noise ratio), you can
determine what frequencies make up most of the energy in the sounds
produced by that species, and choose your filter corner frequencies based on
that.

In this case we will filter the signal with a 10 Hz wide bandpass filter
centered at the frequency of the sine wave. See the second plot in Figure 9-1
to see the result of the filtering in the time domain. Note that in each plot
we have plotted a red dot at the start time of the sine wave.

% Step 1: Filter signal

hbandwidth=5; %half bandwidth
nyquist=srate/2;
ntaps=200;
hpf=(freq-hbandwidth)/nyquist;
lpf=(freq+hbandwidth)/nyquist;
[b,a]=fir1(ntaps, [hpf lpf]); %calc filter coefficients
signal_f=filter(b,a,signal); %filter signal

subplot(5,1,2)
plot(t,signal_f); %plot filtered signal
hold on
%plot red dot on location of sine wave
plot(sinepos/srate,0,'.r');
hold off
title('Step 1. Filter Signal');

Step 2. Rectify Signal

The purpose of rectifying is to make all of the values in the signal positive.
So, we simply take the absolute value of the signal. See the third plot in
Figure 8-1 to see the effect of rectifying the signal.

% Step 2: Rectify signal

signal_r=abs(signal_f); %abs value of filtered signal

subplot(5,1,3)
plot(t,signal_r); %plot filtered signal
hold on
%plot red dot on location of sine wave
plot(sinepos/srate,0,'.r');
hold off
title('Step 2. Rectify Signal');
93
Step 3. Envelope Signal
We want to detect when the tone starts and when the tone stops. We do
not want to detect each small peak corresponding to the tone, so we will
low-pass filter the signal to extract the envelope (or the edges of the signal
that you might trace). Note that there are other techniques for doing this.
In this example we use a 6-pole Butterworth filter with a corner frequency at
10 Hz. This captures the envelope of the signal (Figure 8-1, fourth plot).
Notice that the filters are producing delays in the timing of the sine wave
(the signal is after the location of the red dot).

% Step 3: Low-pass filter rectified signal

lpf=10/nyquist; % low-pass filter corner frequency
npoles=6;
[b,a]=butter(npoles, lpf); % Butterworth filter coeff
signal_env=filter(b,a,signal_r); %Filter signal

subplot(5,1,4)
plot(t,signal_env); %plot filtered signal
hold on
%plot red dot on location of sine wave
plot(sinepos/srate,0,'.r');
hold off
title('Step 3. Envelope Signal');

Step 4. Threshold Signal

In this simple energy detector we will look for the times when the signal
goes above a threshold level. By filtering, rectifying, and enveloping our
signal we have created a time-domain signal that approximately follows the
envelope of the sine wave. To perform a threshold detection we set up a
variable called threshold and compare the enveloped time series to this
threshold.

In this case we set the threshold value to half of the amplitude of the original
sine wave. The result is either 0 or 1. It is 1 when the signal_env time series is
greater than the threshold value. This has been plotted on top of the
enveloped signal, and appears as a rectangular shape where the enveloped
time series exceeds the threshold. This is the gated signal. You can see that
there is a delay between the onset of this detected signal and the original
signal.
94
% Step 4: Threshold signal
threshold=amp_t/2;
gated=(signal_env>threshold);

subplot(5,1,5)
plot(t,signal_env); %plot filtered signal
hold on
%plot red dot on location of sine wave
plot(sinepos/srate,0,'.r');
plot(t, gated, 'r'); % plot threshold signal
hold off
title('Step 4. Threshold Signal');
xlabel('Time (s)');

diff (detecting signal onset)

Now we need to determine the time when the gated signal first increases to
1. The diff function performs a derivative-like process in which a new value is
calculated by subtracting the previous value from the current value. The time
series that results has one less point than the original time series, because
there is no point to subtract from the first point in the time series. For our
gated time series, this results in a 1 at the time of signal onset and a -1 at the
time the signal goes back below the threshold (Figure 8-2).

d_gated=diff(gated);
plot(d_gated);
1

0.5

-0.5

-1
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000

Figure 8-2. Differenced time series in which a value of +1 occurs at the sine wave onset
and a value of -1 occurs at the end of the sine wave.

95
find
The find function can be used to find the indices of values equal to +1.

find(d_gated==1)
This returns 2689, which is the index in the differenced time series equal to
1. To find the time corresponding to the onset, we need to use the index+1:
2690.

t(2690)=2.689

So, we detect the signal at 2.689 seconds. In this case, we know the signal
started at sinepos, which is index 2496. This corresponds to 2.495 s. Thus,
there is approximately a 200 ms delay in detection of the signal. The source
of the delay is from each of the filters and from the thresholding, which
defined the start as the time the enveloped signal exceeded half of the sine
wave amplitude. You can see that if the threshold was lowered, the delay
would be reduced. The downside of this is that it increases the probability
of falsely detecting background noise.

Try This

1. Use find to determine when the sine wave ended.

96
 9
Chapter

Image Processing
Images are composed of pixels (picture elements). This makes them easy to
store in MATLAB, because a two-dimensional matrix can be used to map a
color or intensity to each pixel in an image. The purpose of this chapter is to
introduce you to the basic MATLAB functions for importing images and
plotting them in figures. MATLAB also produces a more advanced image
processing toolbox, which has advanced functions for image anlaysis,
including Fourier analysis, edge finding, and filtering. MATLAB also
produces a mapping toolbox (maps are also images) that has useful features
for including maps in MATLAB figures.

Note that images can also be constructed from vectors, which are simply the
instructions for how to draw lines and polygons to recreate an image. We
will restrict our discussion to pixel-based images.

Basic MATLAB Image Functions

imread: image read is used to load images into MATLAB matrices.
MATLAB supports common image formats including BMP, TIF, GIF,
JPG, and PNG.

imwrite: image write will write a MATLAB matrix to a graphics files.

image: image is used to use data in a matrix to create an image in a figure.

imagesc: image scale is similar to image, but it allows you to scale the intensity
of the pixel range.

97
colormap: color maps are used by indexed images, where the matrix value
points into another colormap matrix where each index specifies the pixel
color (e.g. three values for RGB indexed images).

colorbar: colorbar creates a scale bar for the image

Creating and plotting an image

In this example we will create a matrix that will be used to create an image.
Let’s start with something easy and make a series of bars in an image. We
will make the image with 20 rows and 10 columns by first creating one
column of ones and a column of zeros, and then repeatedly concatenating
that five times.

rows=20; % variable for # rows

stripes=5;% variable for # stripes

% make one column of 1's and an adjacent column of 0's

x=horzcat(ones(rows,1),zeros(rows,1));

y=[]; %initialize and clear y matrix

for n=1:stripes
y=horzcat(y,x);% concatenate x onto y
end

clim=[0 1]; % color limits for imagesc

imagesc(y, clim); % plot scaled image
colormap(gray); % use gray scale color map

Click on the y matrix in the workspace view, and you will see a series of
columns of alternating between 1 and 0 (Fig. 9-1). You can imagine this as
our image. These values will directly map to pixels in the image, with the
value of 1 being one color and the value of 0 being another color.

98
Figure 9-1. Array Editor view of the y matrix showing alternating columns of 1’s and
0’s. This matrix will be used to create the image plot.
Plotting the Matrix as an Image
imagesc is used to plot the y matrix as an image.

clim=[0 1]; % color limits for imagesc

imagesc(y, clim); % plot scaled image
colormap(gray); % use gray scale color map

The clim color limit sets the range of values that we want to use to scale the
color in the image. Since we want a black and white image, with 1 for white
and 0 for black, we set the color limits to [0 1]. Once imagesc has created the

99
figure with the image, we set the colormap to gray. This results in a figure
with alternating white and black bars (Fig. 9-2).

10

12

14

16

18

20
1 2 3 4 5 6 7 8 9 10

Figure 9-2. Figure created using imagesc of matrix from figure 10-1. The rows are plotted
starting from the top of the figure.
Colormap
To understand what colormap is doing, click on the Edit menu of the figure
and select Colormap...

This will bring up the colormap editor (Fig. 9-3).

100
Figure 9-3. Colormap Editor showing currently selected gray colormap. Note that the
Color data min value is set to 0 (this is what makes 0 values black in the image), and the
Color data max value is set to 1 (this is what makes 1 values white in the image).

101
Try changing to a different colormap by clicking on the Tools menu and
selecting Autumn from the Standard Colormaps dropdown menu. You will
see that the figure now alternates between yellow and red bars. This is
because values of 0 in this colormap correspond to red, and values of 1
correspond to yellow.

Now, let’s create our own colormap.

1. Select the White colormap, from the Standard Colormaps

dropdown menu.

2. Double click on the bar on the far right below the colormap (Fig. 9-
4).

Double-click on the
bar below the
colormap to change
the color.

Figure 9-4. Colormap Editor dialog box, used to change colormaps.

3. This opens the Select Marker Color dialog box (Fig. 9-5). Now
select blue and click OK

102
 Figure 9-5. Dialog box to select marker color.

4. Now, you will see the the Colormap change (Fig. 9-6) as well as your
original figure (Fig. 9-7).

Figure 9-6. The Colormap Editor maximum value is now blue, and the intermediate
values grade evenly between white and blue.

103
Figure 9-7. Figure with the newly created color map. With the new colormap, values of 1
correspond to blue, and values of 0 correspond to white.

5. In MATLAB, type colorbar. This will create a colorbar scale on the

figure, showing which values correspond to which colors (Fig 9-8).

Figure 9-8. Figure with colorbar indicating the scale corresponding to the image colors.
104
 Try This
Modify the program to produce a gradation of bars ranging from 0 to 1 (e.g.
bars with values of 0, .1, .2, ..., 1), rather than alternating bars of 0 and 1 (Fig.
9-9).

10

12

14

16

18

20
1 2 3 4 5 6 7 8 9 10

Figure 9-9. Example modified to produce a gradation of bars with the gray
colormap.

Loading and Manipulating an Image

Load the file reef.tif which is a TIFF image of a coral reef near Tahiti.

>> r=imread('reef.tif','tif');

Note that in the workspace, the matrix r is 3-dimensional. It is 478x481x3.

The third dimension corresponds to the RGB values for the image. These
are the Red, Green, and Blue colors used to create the pixels. Also note that

105
the data type is uint8. This means an unsigned 8-bit integer is used to
represent the color of each pixel.

Let’s take a look at the values for the first pixel.

>> r(1,1,1:3)

ans(:,:,1) =
66
ans(:,:,2) =
153
ans(:,:,3) =
174

Remember each pixel is specified by three colors. The first pixel is made up
of Red: 66, Green 153, and Blue 174. Since they are 8-bit values they can
have 256 different values (28=256). The values range from 0 to 255. The
‘unsigned’ part of uint8, means that there are no negative values. Using
uint8 saves a lot of memory over MATLAB’s standard double data type.

Now let’s create our image.

>> image(r)

The image function decodes the RGB scale to create the color image (Fig. 9-
10).

106
 50

100

150

200

250

300

350

400

450

50 100 150 200 250 300 350 400 450

Figure 9-10. TIF image of coral reefs plotted using image.

Thresholding Images
To give an example of basic image processing, let’s write a script to calculate
the total area covered by reefs. The idea is that we will threshold the image
so that all pixels that are blue will be set to 0, and all pixels that are not blue
(brown where there are corals) will be set to 1. Then we just need to add all
the ones together.

We will only work with the blue layer (layer 3) for this. We will load the blue
layer into a new matrix called rb.

r=imread('reef.tif','tif'); %load image

107
rb=r(:,:,3); %just grab 3rd layer (blue)

imagesc(rb);
colormap(gray);

This results in an image where the light areas (areas with high values) are
areas with a lot of blue (Fig. 9-11). Areas with little blue, come out as dark
(areas with low values).

50

100

150

200

250

300

350

400

450

50 100 150 200 250 300 350 400 450

Figure 9-11. Blue layer of image from figure 10-10 plotted as a scaled image.

Now we threshold the image to turn it into a black and white image. The
goal is to have areas with corals white, and areas of sand result in black. In
this code, a new matrix called rbt is created which has values of 1 where rb
was less than 130, and values of 0 where rb was greater than or equal to 130.
The threshold value of 130 was determined by trial-and-error, for this
example. This code yields the black and white image shown in figure 9-12.

108
rbt=rb<130; %threshold dark values
imagesc(rbt)
colormap(gray)

Compare figure 9-11 and figure 9-12. This simple routine does a fairly good
job of finding areas with reefs. But, you can see in the upper right where
areas of sand have been ‘identified’ as reef (white), and areas in the left of the
image where reefs have been identified as ‘sand’ (black). We could achieve a
better discrimination if we used all color channels.

50

100

150

200

250

300

350

400

450

50 100 150 200 250 300 350 400 450

Figure 9-12. Thresholded image of coral reefs. Areas with reefs are in white (values of 1)
and areas of sand are black (values of 0).

To calculate the percent of the image that is composed of reefs, simply add
up all of the values in the matrix to get the number of white pixels (which

109
have the value 1). Then, divide this by the total number of pixels in the
image.

We have to use sum twice because the first sum returns the sum of each
column. The second sum then adds the column totals together. From this
we find that 0.1329 of the image is composed of reefs.

% calculate sum of all white points (=1)

reef=sum(sum(rbt)); %reef pts
totalpts=prod(size(rbt)); %#pts in image
percentreef=reef/totalpts

110
Formal Mathematical Description
Discrete Fourier Transform (DFT)

N-1
X(m)=∑x(n)[cos(2πnm/N)-isin(2πnm/N)]
n=0

where,
X(m)=the mth DFT output component, i.e. X(0), X(1), X(2)…X(N-1)
m=the index of the DFT output in the frequency domain
m=0,1,2,3,…,N-1

x(n)=the sequence of input samples, x(0), x(1), x(2), etc.

n=the time-domain index of the input samples, n=0,1,2,3,…,N-1
i= sqrt(-1)
N=the number of samples of the input sequence and the number of frequency points in the DFT output

For example, when N=4, n and m both go from 0 to 3.

The first point in the FFT corresponds to:

X(0)=x(0)cos(2π·0·0/4)-ix(0)sin(2π·0·0/4)
+ x(1)cos(2π·1·0/4)-ix(1)sin(2π·1·0/4)
+ x(2)cos(2π·2·0/4)-ix(2)sin(2π·2·0/4)
+ x(3)cos(2π·3·0/4)-ix(3)sin(2π·3·0/4)

To convert the complex number to something that makes sense (i.e. as the magnitude and phase of different
frequency components), you perform the following:

the magnitude of X(m) is:

Xmag(m)= │X(m)│=√(Xreal(m)2+Ximag(m)2)

the phase angle of X(m) is:

XØ=tan-1(Ximag(m)/ Xreal(m))

Fast Fourier Transform (FFT)

The FFT is essentially a mathematical trick to make calculation of the Fourier transform faster.

Windowing
Signals are often windowed before performing the FFT to minimize leakage between frequency bins.

Rectangular Window: Essentially ‘no window’.

Hanning Window: Reduces side-lobe leakage, but also reduces main signal.

111
112