LabVIEW Signal Processing

Course Manual
LabVIEW Signal Processing Course Manual

Course Software Version 1.0

September 1997 Edition

Part Number 321569A-01

Copyright
Copyright 1997 by National Instruments Corporation, 6504 Bridge Point Parkway, Austin, TX 78730-5039.
(512) 794-0100. Under the copyright laws, this publication may not be reproduced or transmitted in any form, electronic or
mechanical, including photocopying, recording, storing in an information retrieval system, or translating, in whole or in part,
without the prior written consent of National Instruments Corporation.
Trademarks

TM

LabVIEW , BridgeVIEW , and The Software is the Instrument are trademarks of National Instruments Corporation.
Product and company names listed are trademarks or trade names of their respective companies.

Internet Support

[email protected]
E-mail: [email protected]
FTP Site: ftp.natinst.com
Web Address: http://www.natinst.com
Bulletin Board Support

BBS United States: (512) 794-5422

BBS United Kingdom: 01635 551422
BBS France: 01 48 65 15 59
Fax-on-Demand Support

(512) 418-1111
Telephone Support (U.S.)

Tel: (512) 795-8248

Fax: (512) 794-5678
International Offices

Australia 03 9879 5166, Austria 0662 45 79 90 0, Belgium 02 757 00 20, Canada (Ontario) 905 785 0085,
Canada (Qubec) 514 694 8521, Denmark 45 76 26 00, Finland 09 725 725 11, France 01 48 14 24 24,
Germany 089 741 31 30, Hong Kong 2645 3186, Israel 03 5734815, Italy 02 413091, Japan 03 5472 2970,
Korea 02 596 7456, Mexico 5 520 2635, Netherlands 0348 433466, Norway 32 84 84 00, Singapore 2265886,
Spain 91 640 0085, Sweden 08 730 49 70, Switzerland 056 200 51 51, Taiwan 02 377 1200,
U.K. 01635 523545
National Instruments Corporate Headquarters

6504 Bridge Point Parkway Austin, TX 78730-5039 Tel: (512) 794-0100

Table of Contents
Student Guide
A.
B.
C.
D.
E.
F.
G.

Self-Paced Use .......................................................................................................SG-2

Course Description ................................................................................................SG-4
Prerequisites ...........................................................................................................SG-4
Course Goals ..........................................................................................................SG-5
Course Non-Goals .................................................................................................SG-5
Course Map ............................................................................................................SG-6
Course Conventions ...............................................................................................SG-7

Lesson 1
Background
A. Sampling Signals ...................................................................................................1-2
B. Sampling Considerations .......................................................................................1-3
C. Why Do You Need Antialiasing Filters? ...............................................................1-6
D. Why Use Decibels? ................................................................................................1-7
E. Overview of the Advanced Analysis Library ........................................................1-11
Summary ......................................................................................................................1-14

Lesson 2
Signal Generation
A. Normalized Frequency ...........................................................................................2-2
B. Wave and Pattern VIs ............................................................................................2-7
Summary .......................................................................................................................2-16

Lesson 3
Signal Processing
A. The Discrete Fourier Transform (DFT) and the Fast Fourier Transform (FFT) ...3-2
B. Frequency Spacing and Symmetry of the DFT/FFT .............................................3-6
C. The Power Spectrum ..............................................................................................3-16
Summary ......................................................................................................................3-21

National Instruments Corporation

iii

LabVIEW Signal Processing Course Manual

Table of Contents

Lesson 4
Windowing
A. About Spectral Leakage and Smoothing Windows .............................................. 4-2
B. Windowing Applications ...................................................................................... 4-6
C. Characteristics of Different Types of Window Functions .................................... 4-7
Summary ...................................................................................................................... 4-22

Lesson 5
Measurement
A. The Measurement VIs ........................................................................................... 5-2
B. Calculating the Frequency Spectrum of a Signal .................................................. 5-3
C. Coherent Gain (CG) and Equivalent Noise Bandwidth (ENBW) ........................ 5-7
D. Harmonic Distortion ............................................................................................. 5-13
Summary ..................................................................................................................... 5-18

Lesson 6
Digital Filtering
A. What Is Filtering? ................................................................................................. 6-2
B. Ideal Filters ........................................................................................................... 6-3
C. Practical (Nonideal) Filters ................................................................................... 6-5
D. Advantages of Digital Filters over Analog Filters ................................................ 6-7
E. IIR and FIR Filters ................................................................................................ 6-8
F. Infinite Impulse Response Filters ......................................................................... 6-10
G. The Transient Response of IIR Filters .................................................................. 6-24
H. Finite Impulse Response Filters ............................................................................ 6-28
I. Digital Filter Design Toolkit ................................................................................. 6-32
Summary ..................................................................................................................... 6-33

Lesson 7
Curve Fitting
A. About Curve Fitting .............................................................................................. 7-2
B. General LS Linear Fit ........................................................................................... 7-12
C. Nonlinear Lev-Mar Fit .......................................................................................... 7-21
D. Fitting a Curve to Gaussian (Normal) Data Points ............................................... 7-25
Summary ..................................................................................................................... 7-31

Lesson 8
Linear Algebra
A. Linear Systems and Matrix Analysis .................................................................... 8-2
B. Basic Matrix Operations and Eigenvalues-Eigenvector Problems ....................... 8-18
C. Matrix Inverse and Solving Systems of Linear Equations ................................... 8-31
D. Matrix Factorization ............................................................................................. 8-36
Summary ..................................................................................................................... 8-42

LabVIEW Signal Processing Course Manual

iv

National Instruments Corporation

Table of Contents

Lesson 9
Probability and Statistics
A. Probability and Statistics ...................................................................................... 9-2
B. Statistics ................................................................................................................ 9-4
C. Probability ............................................................................................................. 9-16
Summary ...................................................................................................................... 9-23

Lesson 10
Digital Filter Design Toolkit
A. Review of Digital Filters ...................................................................................... 10-2
B. The Digital Filter Design Toolkit ........................................................................ 10-4
C. Designing IIR Filters ........................................................................................... 10-9
D. Designing FIR Filters ........................................................................................... 10-18
E. Analyzing your Filter Design .............................................................................. 10-29
F. Format of Filter Coefficient Text Files ................................................................ 10-38
Summary .................................................................................................................... 10-41

Lesson 11
G Math Toolkit
A. Organization of the G Math Toolkit .................................................................... 11-2
B. Parser VIs ............................................................................................................. 11-3
C. Solving Differential Equations ............................................................................ 11-14
Summary .................................................................................................................... 11-31

Lesson 12
Third-Octave Analyzers
A. Filter Banks .......................................................................................................... 12-2
B. Octave Analyzers ................................................................................................. 12-4
C. 1/3-Octave Analyzer or Third-Octave Analyzer .................................................. 12-6
D. Applications ......................................................................................................... 12-7
E. The Third-Octave Analyzer Toolkit .................................................................... 12-8
F. Calibration Using the Third-Octave Analyzer ..................................................... 12-19
Summary .................................................................................................................... 12-27

Lesson 13
Joint Time-Frequency Analysis
A. Why Do You Need Joint Time-Frequency Analysis? ......................................... 13-2
B. Joint Time-Frequency Analysis ........................................................................... 13-5
C. The JTFA Toolkit and Its Applications ............................................................... 13-14
Summary .................................................................................................................... 13-22

National Instruments Corporation

LabVIEW Signal Processing Course Manual

Table of Contents

Lesson 14
Wavelet and Filter Banks Designer
A. Signal Representation .......................................................................................... 14-2
B. The Wavelet Transform ....................................................................................... 14-9
C. Applications of Wavelets ..................................................................................... 14-12
D. Wavelets and Filter Banks ................................................................................... 14-13
E. Using the Wavelet and Filter Banks Designer ..................................................... 14-17
Summary .................................................................................................................... 14-26

Appendix A
Error Codes
Appendix B
Frequently Asked Questions
Appendix C
References
Appendix D
Glossary

LabVIEW Signal Processing Course Manual

vi

National Instruments Corporation

Student Guide
Introduction
Welcome to the LabVIEW Signal Processing Course. This course is
intended for practicing engineers and scientists who want to learn how
to use LabVIEW or BridgeVIEW to process and analyze digital signals
in real-world practical applications. The course focuses on the advanced
analysis library and the various signal processing toolkits that are
specifically devoted to designing digital filters, solving mathematics
problems, and analyzing nonstationary signals. In addition to teaching
you how to use the analysis VIs and toolkits, the course also covers the
basic fundamentals necessary for understanding and interpreting the
analysis results.
This student guide describes the course contents and suggests ways in
which you can most effectively use the course materials. The guide
discusses the following topics:
A. Self-Paced Use
B. Course Description
C. Prerequisites
D. Course Goals
E. Course Non-Goals
F. Course Map
G. Course Conventions

National Instruments Corporation

SG-1

LabVIEW Signal Processing Course Manual

Student Guide

A. Self-Paced Use
Thank you for purchasing the LabVIEW Signal Processing Course kit. You
should be able to begin developing your application soon after you have
worked through this manual. This course manual and accompanying
software are used in the two-day, hands-on LabVIEW Signal Processing
Course. Several exercises in this manual use the following National
Instruments hardware products:

AT-MIO-16E-2 data acquisition board

DAQ Signal Accessory

To get started, read the information on the next page regarding the
accompanying disks and then follow the instructions on the subsequent
pages for the computer platform you are using. If you have comments,
suggestions for improving this course, or are not satisfied with the material,
please contact:
LabVIEW Signal Processing Technical Support
6504 Bridge Point Parkway
Austin, TX 78730-5039
(512) 795-8248
[email protected]

Attending the Course

You can apply the full purchase price of this course kit toward the
corresponding course registration fee if you register within 90 days of
purchasing the kit. To register for a course or for course information, please
contact National Instruments.

North America
Telephone: (512) 794-0100
E-mail: [email protected] (information requests only)
24-hour automated retrieval of course outlines/latest course schedule
Fax on Demand: (800) 329-7177 or (512) 418-1111
World Wide Web: http://www.natinst.com/custed

Other Countries
Please contact your local National Instruments branch office (the phone
numbers are on the back cover).

LabVIEW Signal Processing Course Manual

SG-2

National Instruments Corporation

Student Guide

Course Disk
The following table lists the contents of the LabVIEW Signal Processing
Course disk. The course disk contains a zip file containing two VI libraries.
Filename

Note:

Description

Lvspc.zip

A compressed file containing the VIs used in the

course exercises (Lvspcex.llb) as well as the VIs
containing the solutions (Lvspcsol.llb).

unzip.exe

A utility to decompress lvspc.zip.

The solution VIs have the word Solution at the end of the VI name.

You Will Need the Following Equipment:

IBM PC AT or compatible.

LabVIEW or BridgeVIEW for Windows Full Development System,

ver 4.0 or later.

AT-MIO-16E-2 data acquisition board.

DAQ Signal Accessory.

OptionalA word processing application such as Write or Wordpad.

Installing the Course Software

1. Copy the files Lvspc.zip and unzip.exe from the PC disk
accompanying this manual to the Labview directory on your hard
disk.
2. Type in the following at the DOS prompt: unzip -d Lvspc.zip
<enter>. This extracts the VI libraries that contain the class VIs
(Lvspcex.llb) and the solution VIs (Lvspcsol.llb). In
addition, it also extracts a DFD folder and a TOA folder to be used
for the lessons on the Digital Filter Design Toolkit and the
Third-Octave Analyzer Toolkit, respectively.

National Instruments Corporation

SG-3

LabVIEW Signal Processing Course Manual

Student Guide

The course assumes the following directory structure:

Root Directory
C:

\Labview
Directory

\TOA
Directory

\DFD
Directory

Lvspcex.llb
Lvspcsol.llb
LabVIEW libraries

B. Course Description
The LabVIEW Signal Processing course teaches you how to implement and
use the VIs in the advanced analysis library and the signal processing
toolkits, and also how to interpret and understand the results of your
analysis. The course is divided into lessons, each covering a topic or a set of
topics. Each lesson consists of:
An introduction that describes the lessons purpose and what you
will learn.

A discussion of the topics.

A set of exercises to reinforce the topics presented in the discussion.
A set of additional exercises to be done if time permits.
A summary that outlines important concepts and skills taught in the
lesson.
Review questions to check for understanding.

Some of the topics have been simplified considerably to make the basic
concepts and ideas easier to understand. The simplification has been
done at the expense of mathematical detail while striving to provide
clear and precise concepts. For a more detailed presentation of such
topics, see the list of references at the end of this manual.

C. Prerequisites

Familiarity with the Windows operating system.

Familiarity with basic LabVIEW programming techniques.

Experience writing algorithms in the form of flowcharts or block

diagrams.

Previous exposure to digital signal processing, through either an

introductory course or work experience.

LabVIEW Signal Processing Course Manual

SG-4

National Instruments Corporation

Student Guide

D.Course Goals
This course teaches you to:

Become familiar with the analysis capabilities of LabVIEW and

BridgeVIEW.

Understand the basics of digital signal processing and analysis.

Choose intelligently between several options/methods that are

available for performing similar tasks (for example, choosing
between different types of windows, filter design methods, or
algorithms for curve fitting).

Implement the VIs from the analysis library in practical applications

for solving real-world problems.

Learn about the various specialized toolkits such as those available

for solving mathematical problems, analyzing nonstationary signals,
or designing digital filters.

E. Course Non-Goals
It is not the purpose of this course to do any of the following:

Teach LabVIEW or BridgeVIEW basics.

Teach programming theory.

Discuss every built-in LabVIEW Analysis VI.

Discuss each and every analysis algorithm.

Develop a complete application for any student in the class.

National Instruments Corporation

SG-5

LabVIEW Signal Processing Course Manual

Student Guide

F. Course Map
Day 1

Day 2

Lesson 1
Background

Lesson 6
Digital Filtering

Lesson 2
Signal Generation

Lesson 7
Curve Fitting

Lesson 3
Signal Processing

Lesson 8
Linear Algebra

Lesson 4
Windowing

Lesson 9
Probability and
Statistics

Lesson 5
Measurement

Lessons 10-14
Toolkits

LabVIEW Signal Processing Course Manual

SG-6

National Instruments Corporation

Student Guide

G.Course Conventions
The following conventions are used in this course manual:
bold

Text in bold refers to LabVIEW menus, menu

items, palettes, subpalettes, functions, and
VIs. For example, File.

italic

Text in italics is for emphasis, a

cross-reference, or an introduction to a key
concept.

Courier

Text in this font indicates drive names,

libraries, directories, pathnames, filenames,
and sections of programming code. Courier
also indicates information you must type. For
example, type Digital Indicator at the
prompt.

Courier italic

Text in this font denotes that you must supply

the appropriate words or values in the place of
these items.

Courier bold

Text in this font denotes a computer prompt.

<>

Angle brackets enclose the name of a key. For

example, <Enter>.

A hyphen between two or more key names

enclosed in angle brackets denotes that you
should simultaneously press the named keys.
For example, <Control-Alt-Delete>.

National Instruments Corporation

SG-7

LabVIEW Signal Processing Course Manual

Student Guide

Notes

LabVIEW Signal Processing Course Manual

SG-8

National Instruments Corporation

Lesson 1
Background
Introduction
A digital signal is one that can assume only a finite set of values in both the
dependent and independent variables. The independent variable is usually
time or space, and the dependent variable is usually amplitude.
Digital signals are everywhere in the world around us. Telephone
companies use digital signals to represent the human voice. Radio, TV,
and hi-fi sound systems are all gradually converting to the digital
domain because of its superior fidelity, noise reduction, and its signal
processing flexibility. Data is transmitted from satellites to earth ground
stations in digital form. NASAs pictures of distant planets and outer
space are often processed digitally to remove noise and to extract useful
information. Economic data, census results, and stock market prices are
all available in digital form. Because of the many advantages of digital
signal processing, analog signals are also converted to digital form
before they are processed with a computer. This lesson provides a
background in basic digital signal processing and an introduction to the
LabVIEW/BridgeVIEW Analysis Library, which consists of hundreds
of VIs for signal processing and analysis.

You Will Learn:

A. About the digital (sampled) representation of an analog signal.
B. About aliasing and how to prevent it.
C. About the need for antialiasing filters.
D. About why we use the decibel scale to display amplitudes.
E. About the contents of the LabVIEW/BridgeVIEW Analysis Library.

National Instruments Corporation

1-1

LabVIEW Signal Processing Course Manual

Lesson 1 Background

A. Sampling Signals
To use digital signal processing techniques, you must first convert an analog
signal into its digital representation. In practice, this is implemented by
using an analog-to-digital (A/D) converter. Consider an analog signal x(t)
that is sampled every t seconds. The time interval t is known as the
sampling interval or sampling period. Its reciprocal, 1/t, is known as the
sampling frequency, with units of samples/second. Each of the discrete
values of x(t) at t = 0, t, 2t, 3t, etc., is known as a sample. Thus, x(0),
x(t), x(2t), ...., are all samples. The signal x(t) can thus be represented by
the discrete set of samples
{x(0), x(t), x(2t), x(3t), , x(kt), }.
Figure 1-1 below shows an analog signal and its corresponding sampled
version. The sampling interval is t. Observe that the samples are
defined at discrete points in time.

10t
t

4t 7t

Figure 1-1. Analog Signal and Corresponding Sampled Version

In this course, the following notation represents the individual samples:

x[i] = x(it),

for i = 0, 1, 2,

If N samples are obtained from the signal x(t), then x(t) can be
represented by the sequence
X = {x[0], x[1], x[2], x[3], , x[N-1] }
This is known as the digital representation or the sampled version of
x(t). Note that the sequence X = {x[i]} is indexed on the integer variable
i, and does not contain any information about the sampling rate. So by
knowing just the values of the samples contained in X, you will have no
idea of what the sample rate is.

LabVIEW Signal Processing Course Manual

1-2

National Instruments Corporation

Lesson 1

Background

B. Sampling Considerations
A/D converters (ADCs) are an integral part of data acquisition (DAQ)
boards. One of the most important parameters of an analog input system is
the rate at which the DAQ board samples an incoming signal. The sampling
rate determines how often an analog-to-digital (A/D) conversion takes
place. A fast sampling rate acquires more points in a given time and can
form a better representation of the original signal than a slow sampling rate.
Sampling too slowly may result in a poor representation of your analog
signal. Figure 1-2 shows an adequately sampled signal, as well as the effects
of undersampling. The effect of undersampling is that the signal appears as
if it has a different frequency than it truly does. This misrepresentation of a
signal is called an alias.

Adequately Sampled

Aliased Due to Undersampling

Figure 1-2. Aliasing Effects of an Improper Sampling Rate

According to the Nyquist theorem, to avoid aliasing you must sample at

a rate greater than twice the maximum frequency component in the
signal you are acquiring. For a given sampling rate, the maximum
frequency that can be represented accurately, without aliasing, is known
as the Nyquist frequency. The Nyquist frequency is one half the
sampling frequency. Signals with frequency components above the
Nyquist frequency will appear aliased between DC and the Nyquist
frequency. The alias frequency is the absolute value of the difference
between the frequency of the input signal and the closest integer
multiple of the sampling rate. Figures 1-3 and 1-4 illustrate this
phenomenon. For example, assume fs, the sampling frequency, is
100 Hz. Also, assume the input signal contains the following

National Instruments Corporation

1-3

LabVIEW Signal Processing Course Manual

Lesson 1 Background

Magnitude

frequencies25 Hz, 70 Hz, 160 Hz, and 510 Hz. These frequencies are
shown in the following figure.

F1
25 Hz

F2
70 Hz

Frequency

s/2=50
Nyquist Frequency

F3
160 Hz

s=100
Sampling Frequency

F4
510 Hz

500

Figure 1-3. Actual Signal Frequency Components

In Figure 1-4, frequencies below the Nyquist frequency (fs/2=50 Hz) are
sampled correctly. Frequencies above the Nyquist frequency appear as
aliases. For example, F1 (25 Hz) appears at the correct frequency, but F2
(70 Hz), F3 (160 Hz), and F4 (510 Hz) have aliases at 30 Hz, 40 Hz, and
10 Hz, respectively. To calculate the alias frequency, use the following
equation:
Alias Freq. = ABS (Closest Integer Multiple of Sampling Freq. Input Freq.)
where ABS means the absolute value. For example,
Alias F2 = |100 - 70| = 30 Hz
Alias F3 = |(2)100 - 160| = 40 Hz

Magnitude

Alias F4 = |(5)100 - 510| = 10 Hz

Solid Arrows Actual Frequency
Dashed Arrows Alias

F2 alias
30 Hz
F4 alias
10 Hz

F1
25 Hz

Frequency

F3 alias
40 Hz

s/2=50
Nyquist Frequency

F2
70 Hz

F3
160 Hz

s=100
Sampling Frequency

F4
510 Hz

500

Figure 1-4. Signal Frequency Components and Aliases

LabVIEW Signal Processing Course Manual

1-4

National Instruments Corporation

Lesson 1

Background

A question often asked is, How fast should I sample? Your first
thought may be to sample at the maximum rate available on your DAQ
board. However, if you sample very fast over long periods of time, you
may not have enough memory or hard disk space to hold the data. Figure
1-5 shows the effects of various sampling rates. In case A, the sine wave
of frequency f is sampled at the same frequency f. The reconstructed
waveform appears as an alias at DC. However, if you increase the
sampling rate to 2f, the digitized waveform has the correct frequency
(same number of cycles), but appears as a triangle waveform. By
increasing the sampling rate to well above f, for example 5f, you can
more accurately reproduce the waveform. In case C, the sampling rate is
at 4f/3. Because in this case the Nyquist frequency is below f
(4f/3 * 1 2 = 2f/3), this sampling rate reproduces an alias waveform of
incorrect frequency and shape.

A
Sampled at s

B
Sampled at 2s

C
Sampled at 4s/3

Figure 1-5. Effects of Sampling at Different Rates

The Nyquist theorem gives you a starting point for the adequate
sampling rategreater than two times the highest frequency component
in the signal. Unfortunately, this rate is often inadequate for practical
purposes. Real-world signals often contain frequency components that
lie above the Nyquist frequency. These frequencies are erroneously
aliased and added to the components of the signal that are sampled
accurately, producing distorted sampled data. Therefore, for practical
purposes, sampling is usually done at several times the maximum
frequencyfive to 10 times is typical in industry.
Note:

Sampling should be done at least at the Nyquist frequency, but usually

much higher.

National Instruments Corporation

1-5

LabVIEW Signal Processing Course Manual

Lesson 1 Background

C. Why Do You Need Antialiasing Filters?

You have seen that the sampling rate should be at least twice the maximum
frequency of the signal that you are sampling. In other words, the maximum
frequency of the input signal should be less than or equal to half of the
sampling rate. But how do you ensure that this is definitely the case in
practice? Even if you are sure that the signal being measured has an upper
limit on its frequency, pickup from stray signals (such as the powerline
frequency or from local radio stations) could contain frequencies higher
than the Nyquist frequency. These frequencies may then alias into the
desired frequency range and thus give us erroneous results.
To be completely sure that the frequency content of the input signal is
limited, a lowpass filter (a filter that passes low frequencies but
attenuates the high frequencies) is added before the sampler and the
ADC. This filter is called an antialias filter because by attenuating the
higher frequencies (greater than Nyquist), it prevents the aliasing
components. Because at this stage (before the sampler and the ADC) you
are still in the analog world, the antialiasing filter is an analog filter.
An ideal antialias filter is as shown in figure (a) below.

f1

frequency

(a) ideal anti-alias filter

filter output

filter output

transition band

frequency
f1
f2
(b) practical anti-alias filter

An ideal anti-aliasing filter passes all the desired input frequencies

(below f1) and cuts off all the undesired frequencies (above f1). However,
such a filter is not physically realizable. In practice, filters look as shown
in figure (b) above. They pass all frequencies < f1, and cut-off all
frequencies > f2. The region between f1 and f2 is known as the transition
band, which contains a gradual attenuation of the input frequencies.
Although you want to pass only signals with frequencies < f1, those
signals in the transition band could still cause aliasing. Therefore, in
practice, the sampling frequency should be greater than two times the
highest frequency in the transition band. So, this turns out to be more
than two times the maximum input frequency (f1). That is one reason
why you may see that the sampling rate is more than twice the maximum
input frequency. We will see in a later lesson how the transition band of
the filter depends on the filter type being designed.

LabVIEW Signal Processing Course Manual

1-6

National Instruments Corporation

Lesson 1

Background

D. Why Use Decibels?

On some instruments, you will see the option of displaying the amplitude in
a linear or decibel (dB) scale. The linear scale shows the amplitudes as they
are, whereas the decibel scale is a transformation of the linear scale into a
logarithmic scale. You will now see why this transformation is necessary.
Suppose you want to display a signal with very large as well as very
small amplitudes. Assume you have a display of height 10 cm and will
use the entire height of the display for the largest amplitude. So, if the
largest amplitude in the signal is 100 V, a height of 1 cm of the display
corresponds to 10 V. If the smallest amplitude of the signal is
0.1 V, this corresponds to a height of only 0.1 mm. This will barely be
visible on the display!
To see all the amplitudes, from the largest to the smallest, you need to
change the amplitude scale. Alexander Graham Bell invented a unit, the
Bell, which is logarithmic, compressing large amplitudes and expanding
the small amplitudes. However, the Bell was too large of a unit, so
commonly the decibel (1/10th of a Bell) is used. The decibel (dB) is
defined as
one dB = 10 log10 (Power Ratio) = 20 log10 (Voltage Ratio)
The following table shows the relationship between the decibel and the
power and voltage ratios.
dB

Power Ratio

Voltage Ratio

+40

10000

100

+20

100

10

+6

+3

1.4

-3

1/2

1/1.4

-6

1/4

1/2

-20

1/100

1/10

-40

1/10000

1/100

Thus, you see that the dB scale is useful in compressing a wide range of
amplitudes into a small set of numbers. The decibel scale is often used in
sound and vibration measurements and in displaying frequency domain
information. You will now do an exercise that shows a signal in linear and
logarithmic scales.

National Instruments Corporation

1-7

LabVIEW Signal Processing Course Manual

Lesson 1 Background

Exercise 1-1
Objective: To build a VI that displays the signal amplitude in both linear and dB scales.
This VI will display the square of 100 data points on a waveform graph. The
fifth data point will create a spike. You will observe that the spike is visible
on the dB scale.

Front Panel

1.

Build a VI with the front panel shown above.

The Selector control (Controls List and Ring Enumerated
Type) has two options, Linear scale and Logarithm (dB) scale.

LabVIEW Signal Processing Course Manual

1-8

National Instruments Corporation

Lesson 1

Background

Block Diagram

2. Build the block diagram as shown above.

The For Loop (Functions Structures subpalette)
generates the square of 100 data points to be displayed on
the Waveform Graph (Controls Graph palette). (You
add +1 to the loop count to avoid taking the logarithm of
zero, which results in a value of -.) Thus, the values of the
data points range from 12 to 1002, giving a total range of 1
to 10,000. This corresponds to a ratio of 10,000 between
the largest (10,000) and the smallest (1) squared value.
The Replace Array Element function (Functions Array
subpalette) replaces the 5th data point, which has a value of
52 = 25, by 150, to create a spike at the fifth element. You
will see how the spike is barely noticeable on the linear
scale, but is easily distinguishable on the dB scale.
Depending on the selector control, the Case structure
(Functions Structures subpalette) either passes the data
directly (Linear scale) to the Waveform Graph or
calculates 20 times the logarithm to the base 10 (Logarithm
(dB) scale) of the data points and sends the result to the
Waveform Graph.
The Logarithm Base 10 function is found in the
Functions Numeric Logarithmic subpalette.

National Instruments Corporation

1-9

LabVIEW Signal Processing Course Manual

Lesson 1 Background

3. Select the Linear option from the Selector control, and run the VI.
Note that the spike at element 5 is barely visible.
4. Select the Logarithm (dB) option from the Selector control and run
the VI. Note that the spike at element 5 is very easily noticeable.
Note:

Observe the change in the y-axis scale as you switch between the Linear
and Logarithm (dB) options.
5. After you have finished, save the VI as dB_linear.vi in the
Lvspcex.llb library.

End of Exercise 1-1

LabVIEW Signal Processing Course Manual

1-10

National Instruments Corporation

Lesson 1

Background

E. Overview of the Advanced Analysis Library

Once the analog signal has been converted to digital form by the ADC and
is available in your computer as a digital signal (a set of samples), you will
usually want to process these samples in some way. The processing could
be to determine the characteristics of the system from which the samples
were obtained, to measure certain features of the signal, or to convert them
into a form suitable for human understanding, to name a few.
The LabVIEW/BridgeVIEW Analysis library contains VIs to perform
extensive numerical analysis, signal generation and signal processing,
curve fitting, measurement, and other analysis functions. The Analysis
Library, included in the LabVIEW/BridgeVIEW full development
system, is a key component in building a virtual instrumentation system.
Besides containing the analysis functionality found in many math
packages, it also features many unique signal processing and
measurement functions that are designed exclusively for the
instrumentation industry.
The analysis VIs are available in the Analysis subpalette of the
Functions palette in LabVIEW or BridgeVIEW.

There are 10 analysis VI libraries. The main categories are:

Signal Generation: VIs that generate digital patterns and
waveforms.

National Instruments Corporation

1-11

LabVIEW Signal Processing Course Manual

Lesson 1 Background

Digital Signal Processing: VIs that perform frequency

domain transformations, frequency domain analysis, time
domain analysis, and other transforms such as the Hartley and
Hilbert transforms.
Measurement: VIs that perform measurement-oriented
functions such as single-sided spectrums, scaled windowing,
and peak power and frequency estimation.
Filters: VIs that perform IIR, FIR, and nonlinear digital
filtering functions.
Windows: VIs that perform data windowing.

Curve Fitting: VIs that perform curve fitting functions and

interpolations.
Probability and Statistics: VIs that perform descriptive
statistics functions, such as identifying the mean or the
standard deviation of a set of data, as well as inferential
statistics functions for probability and analysis of variance
(ANOVA).
Linear Algebra: VIs that perform algebraic functions for real
and complex vectors and matrices.
Array Operations: VIs that perform common, one- and
two-dimensional numerical array operations, such as linear
evaluation and scaling.
Additional Numerical Methods: VIs that use numerical
methods to perform root-finding, numerical integration, and
peak detection.
In this course, you will learn how to design and use the VIs from the
analysis library to build a function generator and a simple, yet practical,
spectrum analyzer. You will also learn how to design and use digital
filters, the purpose of windowing, and the advantages of different types
of windows, how to perform simple curve-fitting tasks, and much more.
The exercises in this course require the LabVIEW/BridgeVIEW full
development system. For the more adventurous, an extensive set of
examples that demonstrate how to use the analysis VIs can be found in
the labview examples analysis folder.

LabVIEW Signal Processing Course Manual

1-12

National Instruments Corporation

Lesson 1

Background

In addition to the Analysis library, National Instruments also offers

many analysis add-ons that make LabVIEW or BridgeVIEW one of the
most powerful analysis software packages available. These add-ons
include the Joint Time-Frequency Analysis Toolkit, which includes the
National Instruments award-winning Gabor Spectrogram algorithm that
analyzes time-frequency features not easily obtained by conventional
Fourier analysis; the G Math Toolkit, which offers extended math
functionality like a formula parser, routines for optimization and solving
differential equations, numerous types of 2D and 3D plots, and more; the
Digital Filter Design Toolkit; and many others. These specialized
add-ons will also be discussed later in this course.

National Instruments Corporation

1-13

LabVIEW Signal Processing Course Manual

Lesson 1 Background

Summary

This lesson introduced the digital (sampled) representation of a

signal.

To convert an analog signal into a digital signal, the sampling

frequency (fs) should be at least twice the highest frequency
contained in the signal. If this is not the case, the frequencies in the
signal that are greater than the Nyquist frequency (fs/2) appear as
undesirable aliases.

You can use a lowpass filter before sampling the analog signal to
limit its frequency content to less than fs/2. Such a filter used to
prevent the effect of aliasing is known as an antialias filter.

You saw how to use a logarithmic scale (the decibel) to display a

large range of values. It does this by compressing large values and
expanding small ones.

This lesson also gave an overview of the LabVIEW/BridgeVIEW

Analysis Library and its contents.

Review Questions
1. Give some examples of digital signals in everyday life.
2. Given a set of sample values x = {x[i]} where i is an integer variable,
what is the sampling rate?
3. What is aliasing? How can it be avoided?
4. Given that the sampling frequency is 100 Hz, what is the alias
frequency (if any) for the following: 13 Hz, 25 Hz, 40 Hz, 75 Hz,
99 Hz, 101 Hz, 200 Hz, and 350 Hz?
5. Why do we use the decibel scale? In what applications is it normally
used?
6. Which of the following is possible using the analysis VIs?
a. Finding the mean or standard deviation of census data.
b. Designing a filter to remove noise from an electrocardiogram.
c. Detecting peaks in a blood pressure waveform to measure the
heart rate.
d. Interpolating between data points to plot the trajectory of an
object (for example, a comet or a cannonball).

LabVIEW Signal Processing Course Manual

1-14

National Instruments Corporation

Lesson 1

Background

Notes

National Instruments Corporation

1-15

LabVIEW Signal Processing Course Manual

Lesson 1 Background

Notes

LabVIEW Signal Processing Course Manual

1-16

National Instruments Corporation

Lesson 2
Signal Generation
Introduction
In this lesson, you will learn how to use the VIs in the analysis library to
generate many different types of signals. Some of the applications for signal
generation are:

Simulating signals to test your algorithm when real-world signals are

not available (for example, when you do not have a DAQ board for
obtaining real-world signals).

Generating signals to apply to a D/A converter (for example, in

control applications such as opening or closing a valve).

You Will Learn:

A. About the concept of normalized frequency.
B. About the difference between Wave and Pattern VIs (for example,
the Sine Wave VI and the Sine Pattern VI).
C. About how to build a simple function generator using the VIs in the
Signal Generation subpalette.

National Instruments Corporation

2-1

LabVIEW Signal Processing Course Manual

Lesson 2 Signal Generation

A. Normalized Frequency
In the analog world, a signal frequency is measured in Hz or cycles per
second. But the digital system often uses a digital frequency, which is
the ratio between the analog frequency and the sampling frequency:
digital frequency = analog frequency / sampling frequency
This digital frequency is known as the normalized frequency. Its units
are cycles/sample.
Some of the Signal Generation VIs use an input frequency control, f, that
is assumed to use normalized frequency units of cycles per sample. This
frequency ranges from 0.0 to 1.0, which corresponds to a real frequency
range of 0 to the sampling frequency fs. This frequency also wraps
around 1.0, so that a normalized frequency of 1.1 is equivalent to 0.1.
As an example, a signal that is sampled at the Nyquist rate (fs/2) means
that it is sampled twice per cycle (that is, two samples/cycle). This
will correspond to a normalized frequency of 1/2 cycles/sample =
0.5 cycles/sample. The reciprocal of the normalized frequency, 1/f, gives
you the number of times that the signal is sampled in one cycle.
When you use a VI that requires the normalized frequency as an input,
you must convert your frequency units to the normalized units of
cycles/sample. You must use these normalized units with the
following VIs.

Sine Wave

Square Wave

Sawtooth Wave

Triangle Wave

Arbitrary Wave

Chirp Pattern

LabVIEW Signal Processing Course Manual

2-2

National Instruments Corporation

Lesson 2

Signal Generation

If you are used to working in frequency units of cycles, you can convert
cycles to cycles/sample by dividing cycles by the number of samples
generated. The following illustration shows the Sine Wave VI, which is
being used to generate two cycles of a sine wave.

The following illustration shows the block diagram for converting

cycles to cycles/sample.

You need only divide the frequency (in cycles) by the number of
samples. In the above example, the frequency of 2 cycles is
divided by 50 samples, resulting in a normalized frequency of
f = 1/25 cycles/sample. This means that it takes 25 (the reciprocal of f)
samples to generate one cycle of the sine wave.
However, you may need to use frequency units of Hz (cycles/second). If
you need to convert from Hertz (or cycles/second) to cycles/sample,
divide your frequency in cycles/second by the sampling rate given in
samples/second.
cycles/second
cycles
------------------------------------- = ----------------samples/second
sample
National Instruments Corporation

2-3

LabVIEW Signal Processing Course Manual

Lesson 2 Signal Generation

The following illustration shows the Sine Wave VI used to generate a

60 Hz sine signal.

Below is a block diagram for generating a Hertz sine signal. You divide
the frequency of 60 Hz by the sampling rate of 1000 Hz to get the
normalized frequency of f = 0.06 cycles/sample. Therefore, it takes
almost 17 (1/0.06) samples to generate one cycle of the sine wave.

The signal generation VIs create many common signals required for
network analysis and simulation. You can also use the signal generation
VIs in conjunction with National Instruments hardware to generate
analog output signals.

LabVIEW Signal Processing Course Manual

2-4

National Instruments Corporation

Lesson 2

Signal Generation

Exercise 2-1
Objective: To understand the concept of normalized frequency.
1. Build the VI front panel and block diagram shown below.

Front Panel

Block Diagram

Sine Wave VI (Analysis Signal Generation subpalette).

National Instruments Corporation

2-5

LabVIEW Signal Processing Course Manual

Lesson 2 Signal Generation

2. Select a frequency of 2 cycles (frequency = 2 and f type = cycles)

and number of samples = 100. Run the VI. Note that the plot will
show 2 cycles. (The normalized frequency indicator tells you the
normalized frequency.)
3. Increase the number of samples to 150, 200, and 250. How many
cycles do you see?
4. Now keep the number of samples = 100. Increase the number of
cycles to 3, 4, and 5. How many cycles do you see?
Thus, when you choose the frequency in terms of cycles, you will see
that many cycles of the input waveform on the plot. Note that the
sampling rate is irrelevant in this case.
5. Change f type to Hz and sampling rate (Hz) to 1000.
6. Keeping the number of samples fixed at 100, change the frequency
to 10, 20, 30, and 40. How many cycles of the waveform do you see
on the plot for each case? Explain your observations.
7. Repeat the above step by keeping the frequency fixed at 10 and
change the number of samples to 100, 200, 300, and 400. How
many cycles of the waveform do you see on the plot for each case?
Explain your observations.
8. Keep the frequency fixed at 20 and the number of samples fixed at
200. Change the sampling rate (Hz) to 500, 1000, and 2000. Make
sure you understand the results.
9. Save the VI as Normalized Frequency.vi in the library
Lvspcex.llb.

End of Exercise 2-1

LabVIEW Signal Processing Course Manual

2-6

National Instruments Corporation

Lesson 2

Signal Generation

B. Wave and Pattern VIs

You will notice that the names of most of the signal generation VIs have the
word wave or pattern in them. There is a basic difference in the operation
of the two different types of VIs. It has to do with whether or not the VI can
keep track of the phase of the signal that it generates each time it is called.

Phase Control
The wave VIs have a phase in control where you can specify the initial
phase (in degrees) of the first sample of the generated waveform. They
also have a phase out indicator that specifies what the phase of the next
sample of the generated waveform is going to be. In addition, a reset
phase control decides whether or not the phase of the first sample
generated when the wave VI is called is the phase specified at the phase
in control, or whether it is the phase available at the phase out control
when the VI last executed. A TRUE value of reset phase sets the initial
phase to phase in, whereas a FALSE value sets it to the value of phase
out when the VI last executed.
The wave VIs are all reentrant (can keep track of phase internally) and
accept frequency in normalized units (cycles/sample). The only pattern
VI that presently uses normalized units is the Chirp Pattern VI. Setting
the reset phase Boolean to FALSE allows for continuous sampling
simulation.
Note:

Wave VIs are reentrant and accept the frequency input in terms of
normalized units.
In the next exercise, you will generate a sine wave using both the Sine
Wave VI and the Sine Pattern VI. You will see how in the Sine Wave
VI you have more control over the initial phase than in the Sine
Pattern VI.

National Instruments Corporation

2-7

LabVIEW Signal Processing Course Manual

Lesson 2 Signal Generation

Exercise 2-2
OBJECTIVE: To generate a sine wave of a particular frequency and see the effect of aliasing.

Front Panel

1. Open the Generate Sine VI from the library Lvspcex.llb.

2. The front panel contains controls for the number of sample points to
be generated, the amplitude, analog frequency, and initial phase (in
degrees) of the sine wave to be generated, and the frequency at which
this waveform is sampled.
3. Do not change the front panel default values. Switch to the block
diagram.

Block Diagram

4. Examine the block diagram.

Sine Wave VI (Analysis Signal Generation subpalette). In
this exercise, this VI generates 100 points of a 10 Hz sine
wave sampled at 100 Hz.

LabVIEW Signal Processing Course Manual

2-8

National Instruments Corporation

Lesson 2

Signal Generation

5. Notice in the block diagram that the signal frequency is divided by

the sampling frequency before it is connected to the Sine Wave VI.
This is because the Sine Wave VI requires the digital (normalized)
frequency of the signal.
6. Run the VI. With the default front panel values, a 10 Hz sine wave
should appear on the graph.

Sampling and Aliasing

7. Change the signal frequency on the front panel to 90 Hz and observe
the waveform. The resulting signal looks just like the 10 Hz
waveform.
As you saw in the previous lesson, this phenomenon is called
aliasing, which occurs only in the digital domain. The famous
Nyquist Sampling Theorem dictates that the highest representable
useful frequency is at most half of the sampling frequency. In our
case, the sampling frequency is 100 Hz, so the maximum
representable frequency is 50 Hz. If the input frequency is over
50 Hz, as in our case of 90 Hz, it will be aliased back to
((n*50) - 90) Hz > 0, which is (100-90) Hz, or 10 Hz. In other words,
this digital system with a sampling frequency of 100 Hz cannot
discriminate 10 Hz from 90 Hz, 20 Hz from 80 Hz, 51 Hz from
49 Hz, and so on.
The Importance of an Analog Antialiasing Filter
Therefore, in designing a digital system, you must make sure that any
frequencies over half of the sampling frequency do not enter the
system. Once they are in, there is no way to remove them! To prevent
aliasing, you typically use an analog antialiasing lowpass filter. So,
in this example, you can use an analog antialiasing filter to remove
any frequencies over 50 Hz. After the signal is filtered, you are
assured that whenever you see a 10 Hz signal with a 100 Hz sampling
frequency, it is 10 Hz and not 90 Hz.
8. When you are done, stop the VI by clicking on the STOP button.
Close the VI. Do not save any changes.

End of Exercise 2-2

National Instruments Corporation

2-9

LabVIEW Signal Processing Course Manual

Lesson 2 Signal Generation

Exercise 2-3
Objective:

To generate a sinusoidal waveform using both the Sine Wave VI and the Sine Pattern
VI and to understand the differences.
1. Build the VI front panel and block diagram shown below.

Front Panel

LabVIEW Signal Processing Course Manual

2-10

National Instruments Corporation

Lesson 2

Signal Generation

Block Diagram

Sine Pattern VI (Analysis Signal Generation palette).

Sine Wave VI (Analysis Signal Generation palette).

2. Set the controls to the following values:

cycles or freq:

2.00

sampling freq:

100

phase in:

0.00

reset phase:

OFF

Run the VI several times.

Observe that the Sine Wave plot changes each time you run the VI.
Because reset phase is set to OFF, the phase of the sine wave changes
with each call to the VI, being equal to the value of phase out during
the previous call. However, the Sine Pattern plot always remains the
same, showing 2 cycles of the sinusoidal waveform. The initial phase
of the Sine Pattern plot is equal to the value set in the phase in
control.
Note:

Phase in and phase out are specified in degrees.

3. Change phase in to 90 and run the VI several times. Just as before,
the Sine Wave plot changes each time you run the VI. However, the
Sine Pattern plot does not change, but the initial phase of the
sinusoidal pattern is 90 degreesthe same as that specified in the
phase in control.
4. With phase in still at 90, set reset phase to ON and run the VI several
times. The sinusoidal waveforms shown in both the Sine Wave and

National Instruments Corporation

2-11

LabVIEW Signal Processing Course Manual

Lesson 2 Signal Generation

Sine Pattern plots start at 90 degrees, but do not change with

successive calls to the VI.
5. Keeping reset phase as ON, run the VI several times for each of the
following values of phase in: 45, 180, 270, and 360. Note the initial
phase of the generated waveform each time that the VI is run.
6. When you have finished, save the VI as Wave and Pattern.vi in the
library Lvspcex.llb.

End of Exercise 2-3

LabVIEW Signal Processing Course Manual

2-12

National Instruments Corporation

Lesson 2

Signal Generation

Exercise 2-4 (Optional)

Objective: To build a simple function generator.
In this exercise, you will build a very simple function generator than can
generate the following waveforms:

Sine Wave

Square Wave

Triangle Wave

Sawtooth Wave

1. Build the VI front panel and block diagram shown below.

Front Panel

The Signal Source control selects the type of waveform that you
want to generate.
The square duty cycle control is used only for setting the duty cycle
of the square wave.
The samples control determines the number of samples in the plot.
Note that these are all wave VIs, and therefore they require the
frequency input to be the normalized frequency. So, you divide
frequency by the sampling rate and the result is the normalized
frequency wired to the f input of the VIs.

National Instruments Corporation

2-13

LabVIEW Signal Processing Course Manual

Lesson 2 Signal Generation

Block Diagram

Sine Wave VI (Analysis Signal Generation subpalette)

generates a sine wave of normalized frequency f.
Triangle Wave VI (Analysis Signal Generation subpalette)
generates a triangular wave of normalized frequency f.
Square Wave VI (Analysis Signal Generation subpalette)
generates a square wave of normalized frequency f with
specified duty cycle.
Sawtooth Wave VI (Analysis Signal Generation subpalette)
generates a sawtooth wave of normalized frequency f.
2. Select a sampling rate of 1000 Hz, amplitude = 1, samples = 100,
frequency = 10, reset phase = ON, and signal source = sine wave.
Note that because sampling rate = 1000 and frequency = 10 Hz,
every 100 samples corresponds to one cycle.
3. Run the VI and observe the resulting plot.

LabVIEW Signal Processing Course Manual

2-14

National Instruments Corporation

Lesson 2

Signal Generation

4. Change samples to 200, 300, and 400. How many cycles of the
waveform do you see? Explain why.
5. With samples set to 100, change reset phase to OFF. Do you notice
any difference in the plot?
6. Change frequency to 10.01 Hz. What happens? Why?
7. Change reset phase to ON. Now what happens? Explain why.
8. Repeat steps 4 7 for different waveforms selected in the Signal
Source control.
9. When you finish, save the VI as Function Generator.vi in the
Lvspcex.llb library.

End of Exercise 2-4

National Instruments Corporation

2-15

LabVIEW Signal Processing Course Manual

Lesson 2 Signal Generation

Summary
In this lesson, you learned:

About the normalized frequency (f) that has units of cycles/sample.

How to generate a sine wave of a particular frequency.

That the wave VIs can keep track of the phase of the generated
waveform.

How to build a simple function generator that can generate a sine,

square, triangular, and sawtooth wave.

Review Questions
1. Name two practical applications in which you would want to
generate signals.
2. What is the normalized signal frequency for the following?
a. sampling frequency = 100 Hz
number of samples = 200
signal frequency = 15 Hz
b. sampling frequency = 100 Hz
number of samples = 200
signal frequency = 15 cycles
3. What are two main differences between the Wave and Pattern VIs?
4. Which of the following VIs require a normalized frequency input?
a. Sine wave
b. Sine pattern
c. Chirp pattern
d. Square wave

LabVIEW Signal Processing Course Manual

2-16

National Instruments Corporation

Lesson 2

Signal Generation

Notes

National Instruments Corporation

2-17

LabVIEW Signal Processing Course Manual

Lesson 2 Signal Generation

Notes

LabVIEW Signal Processing Course Manual

2-18

National Instruments Corporation

Lesson 3
Signal Processing
Introduction
In this lesson, you will learn the basics of transforming a signal from the
time domain into the frequency domain.

You Will Learn:

A. About the discrete Fourier transform (DFT) and the fast Fourier
transform (FFT).
B. How to determine the frequency spacing between the samples of the
FFT (that is, the relationship between the sampling frequency fs, number
of samples N, and the frequency spacing f).
C. About the power spectrum and how it differs from both the DFT and
the FFT.
D. About how to interpret the information in the frequency domain for
the DFT/FFT and the power spectrum, for both even and odd N.

National Instruments Corporation

3-1

LabVIEW Signal Processing Course Manual

Lesson 3 Signal Processing

A. The Discrete Fourier Transform (DFT) and the Fast Fourier

Transform (FFT)
The samples of a signal obtained from a DAQ board constitute the time
domain representation of the signal. This representation gives the
amplitudes of the signal at the instants of time during which it had been
sampled. However, in many cases you want to know the frequency content
of a signal rather than the amplitudes of the individual samples. The
representation of a signal in terms of its individual frequency components is
known as the frequency domain representation of the signal. The frequency
domain representation could give more insight about the signal and the
system from which it was generated.
The algorithm used to transform samples of the data from the time
domain into the frequency domain is known as the discrete Fourier
transform or DFT. The DFT establishes the relationship between the
samples of a signal in the time domain and their representation in the
frequency domain. The DFT is widely used in the fields of spectral
analysis, applied mechanics, acoustics, medical imaging, numerical
analysis, instrumentation, and telecommunications.

DFT

frequency domain representation

time domain representation of x[n]

Suppose you have obtained N samples of a signal from a DAQ board. If

you apply the DFT to N samples of this time domain representation of
the signal, the result is also of length N samples, but the information it
contains is of the frequency domain representation. The relationship
between the N samples in the time domain and the N samples in the
frequency domain is explained below.

LabVIEW Signal Processing Course Manual

3-2

National Instruments Corporation

Lesson 3

Signal Processing

If the signal is sampled at a sampling rate of fs Hz, then the time interval
between the samples (that is, the sampling interval) is t, where
1
t = --fs

The sample signals are denoted by x[i], 0 i N-1 (that is, you have a
total of N samples). When the discrete Fourier transform, given by
N1

Xk =

x e

j2ik N

for k = 0, 1, 2, , N-1

(1)

i=0

is applied to these N samples, the resulting output (X[k], 0 k N-1) is the

frequency domain representation of x[i]. Note that both the time domain x
and the frequency domain X have a total of N samples. Analogous to the
time spacing of t between the samples of x in the time domain, you have a
frequency spacing of
f
1
f = ----s = ---------N
Nt

between the components of X in the frequency domain. f is also known as

the frequency resolution. To increase the frequency resolution (smaller f)
you must either increase the number of samples N (with fs constant) or
decrease the sampling frequency fs (with N constant).
In the following example, you will go through the mathematics of
equation (1) to calculate the DFT for a DC signal.

DFT Calculation Example

amplitude

In the next section, you will see the exact frequencies to which the N
samples of the DFT correspond. For the present discussion, assume that
X[0] corresponds to DC, or the average value, of the signal. To see the result
of calculating the DFT of a waveform with the use of equation (1), consider
a DC signal having a constant amplitude of +1 V. Four samples of this signal
are taken, as shown in the figure below.

+1 V

time
0

National Instruments Corporation

3-3

LabVIEW Signal Processing Course Manual

Lesson 3 Signal Processing

Each of the samples has a value +1, giving the time sequence
x[0] = x[1] = x[2] = x[3] = 1
Using equation (1) to calculate the DFT of this sequence and making use
of Eulers identity,
exp (j) = cos( ) - jsin( )
you get:
N1

X[0] =

x e

j2i0 N

= x[0] + x[1] + x[2] + x[3] = 4

i=0

X [ 1 ] = x [ 0 ] + x [ 1 ] cos --- j sin --- + x [ 2 ] ( cos ( ) j sin ( ) ) +

2
2

3
3
x [ 3 ] cos ------ j sin ------ = ( 1 j 1 + j ) = 0
2

2
X [ 2 ] = x [ 0 ] + x [ 1 ] ( cos ( ) j sin ( ) ) + x [ 2 ] ( cos ( 2 ) j sin ( 2 ) ) +
x [ 3 ] ( cos ( 3 ) j sin ( 3 ) ) = ( 1 1 + 1 1 ) = 0
3
3
X [ 3 ] = x [ 0 ] + x [ 1 ] cos ------ j sin ------ + x [ 2 ] ( cos ( 3 ) j sin ( 3 ) ) +
2

2
9
9
x [ 3 ] cos ------ j sin ------ = ( 1 j 1 j ) = 0

2
2

Therefore, except for the DC component, X[0], all the other values are
zero, which is as expected. However, the calculated value of X[0]
depends on the value of N (the number of samples). Because you had
N = 4, X[0] = 4. If N = 10, then you would have calculated X[0] = 10.
This dependency of X[.] on N also occurs for the other frequency
components. Thus, you usually divide the DFT output by N, so as to
obtain the correct magnitude of the frequency component.

Magnitude and Phase Information

You have seen that N samples of the input signal result in N samples of the
DFT. That is, the number of samples in both the time and frequency
representations is the same. From equation (1), you see that regardless of
whether the input signal x[i] is real or complex, X[k] is always complex
(although the imaginary part may be zero). Thus, because the DFT is
complex, it contains two pieces of informationthe amplitude and the
phase. It turns out that for real signals (x[i] real) such as those obtained from
the output of one channel of a DAQ board, the DFT is symmetric about the
index N/2 with the following properties:
| X[k] | = | X[N-k] |

and

phase ( X[k] ) = - phase( X[N-k] )

The terms used to describe this symmetry are that the magnitude of X[k]
is even symmetric, and phase(X[k]) is odd symmetric. An even
symmetric signal is one that is symmetric about the y-axis, whereas an

LabVIEW Signal Processing Course Manual

3-4

National Instruments Corporation

Lesson 3

Signal Processing

odd symmetric signal is symmetric about the origin. This is shown in the
following figures.

even symmetry

odd symmetry

The net effect of this symmetry is that there is repetition of information

contained in the N samples of the DFT. Because of this repetition of
information, only half of the samples of the DFT actually need to be
computed or displayed, as the other half can be obtained from this
repetition.
Note:

If the input signal is complex, the DFT will be nonsymmetric and you
cannot use this trick.

National Instruments Corporation

3-5

LabVIEW Signal Processing Course Manual

Lesson 3 Signal Processing

B. Frequency Spacing and Symmetry of the DFT/FFT

Because the sampling interval is t seconds, and if the first (k = 0) data
sample is assumed to be at 0 seconds, the kth (k > 0, k integer) data sample
is at kt seconds. Similarly, the frequency resolution being f
f
N

( f = ----s ) means that the kth sample of the DFT occurs at a frequency of
kf Hz. (Actually, as you will soon see, this is valid for only up to about
half the number of samples. The other half represent negative frequency
components.) Depending on whether the number of samples, N, is even or
odd, you can have a different interpretation of the frequency corresponding
to the kth sample of the DFT.

Even Number of Samples

For example, suppose N is even and let p = N
---- . The following table
2

shows the frequency to which each element of the complex output sequence
X corresponds.
Array Element

Corresponding Frequency

X[0]

DC component

X[1]

X[2]

2f

X[3]

3f

.
.
.

.
.
.

X[p-2]

(p-2)f

X[p-1]

(p-1)f

X[p]

pf (Nyquist frequency)

X[p+1]

- (p-1)f

X[p+2]

- (p-2)f

.
.
.

.
.
.

X[N-3]

- 3f

X[N-2]

- 2f

X[N-1]

- 1f

LabVIEW Signal Processing Course Manual

3-6

National Instruments Corporation

Lesson 3

Signal Processing

Note that the pth element, X[p], corresponds to the Nyquist frequency.
The negative entries in the second column beyond the Nyquist frequency
represent negative frequencies.
For example, if N = 8, p = N/2 = 4, then
X[0]

DC

X[1]

X[2]

2f

X[3]

3f

X[4]

4f (Nyquist freq)

X[5]

-3f

X[6]

-2f

X[7]

-f

Here, X[1] and X[7] will have the same magnitude, X[2] and X[6] will
have the same magnitude, and X[3] and X[5] will have the same
magnitude. The difference is that whereas X[1], X[2], and X[3]
correspond to positive frequency components, X[5], X[6], and X[7]
correspond to negative frequency components. Note that X[4] is at the
Nyquist frequency.
The following illustration represents this complex sequence for N = 8.

DC

positive
frequencies

negative
frequencies

Nyquist component

Such a representation, where you see both the positive and negative
frequencies, is known as the two-sided transform.

National Instruments Corporation

3-7

LabVIEW Signal Processing Course Manual

Lesson 3 Signal Processing

Odd Number of Samples

1
Now suppose that N is odd. Let p = N
------------- . The following table shows
2

the frequency to which each element of the complex output sequence X

corresponds.
Array Element

Corresponding Frequency

X[0]

DC component

X[1]

X[2]

2f

X[3]

3f

.
.
.

.
.
.

X[p-1]

(p-1)f

X[p]

pf

X[p+1]

-pf

X[p+2]

- (p-1)f

.
.
.

.
.
.

X[N-3]

- 3f

X[N-2]

- 2f

X[N-1]

- f

Note that when N is odd, N/2 is not an integer, and thus there is no
component at the Nyquist frequency.
For example, if N = 7, p = (N-1)/2 = (7-1)/2 = 3, and you have
X[0]

DC

X[1]

X[2]

2f

X[3]

3f

X[4]

-3f

X[5]

-2f

X[6]

-f

LabVIEW Signal Processing Course Manual

3-8

National Instruments Corporation

Lesson 3

Signal Processing

Now X[1] and X[6] have the same magnitude, X[2] and X[5] have the
same magnitude, and X[3] and X[4] have the same magnitude. However,
whereas X[1], X[2], and X[3] correspond to positive frequencies, X[4],
X[5], and X[6] correspond to negative frequencies. Because N is odd,
there is no component at the Nyquist frequency.
The following illustration represents the preceding table for N = 7.

DC

positive
frequencies

negative
frequencies

This is also a two-sided transform, because you have both the positive
and negative frequencies.

Fast Fourier Transforms

Direct implementation of the DFT (equation (1) on page 3-3) on N data
samples requires approximately N2 complex operations and is a
time-consuming process. However, when the size of the sequence is a power
of 2,
N = 2m for m = 1, 2, 3,
you can implement the computation of the DFT with approximately
N log (N) operations. This makes the calculation of the DFT much
2
faster, and DSP literature refers to these algorithms as fast Fourier
transforms (FFTs). The FFT is nothing but a fast algorithm for
calculating the DFT when the number of samples (N) is a power of 2.
The advantages of the FFT include speed and memory efficiency,
because the VI performs the transform in place. The size of the input
sequence, however, must be a power of 2. The DFT can efficiently
process any size sequence, but the DFT is slower than the FFT and uses
more memory, because it must store intermediate results during
processing.

National Instruments Corporation

3-9

LabVIEW Signal Processing Course Manual

Lesson 3 Signal Processing

Zero Padding
A technique employed to make the input sequence size equal to a power
of 2 is to add zeros to the end of the sequence so that the total number of
samples is equal to the next higher power of 2. For example, if you have
10 samples of a signal, you can add six zeros to make the total number of
samples equal to 16 (= 24a power of 2). This is shown below:

In addition to making the total number of samples a power of two so that

faster computation is made possible by using the FFT, zero padding also
helps in increasing the frequency resolution (recall that f = fs/N) by
increasing the number of samples, N.

FFT VIs in the Analysis Library

The analysis library contains two VIs that compute the FFT of a signal.
They are the Real FFT and Complex FFT.
The difference between the two VIs is that the Real FFT computes the
FFT of a real-valued signal, whereas the Complex FFT computes the
FFT of a complex-valued signal. However, keep in mind that the outputs
of both VIs are complex.
Most real-world signals are real valued, and hence you can use the Real
FFT for most applications. Of course, you could also use the Complex
FFT by setting the imaginary part of the signal to zero. An example of
an application where you could use the Complex FFT is when the signal
consists of both a real and imaginary component. Such a type of signal
occurs frequently in the field of telecommunications, where you
modulate a waveform by a complex exponential. The process of

LabVIEW Signal Processing Course Manual

3-10

National Instruments Corporation

Lesson 3

Signal Processing

modulation by a complex exponential results in a complex signal, as

shown below:
e-jt
x(t)

y(t) = x(t)cos(t) - jx(t) sin(t)

The block diagram below shows a simplified version of how you can
generate 10 cycles of a complex signal:

To next stage

National Instruments Corporation

3-11

LabVIEW Signal Processing Course Manual

Lesson 3 Signal Processing

Exercise 3-1
Objective:

To display the two-sided and the one-sided Fourier transform of a signal using the
Real FFT VI, and to observe the effect of aliasing in the frequency spectrum.
1. Build the VI front panel and block diagram as shown below.

Front Panel

LabVIEW Signal Processing Course Manual

3-12

National Instruments Corporation

Lesson 3

Signal Processing

Block Diagram

Array Size function (Functions Array subpalette) scales

the output of the FFT by the number of samples so as to
obtain the correct amplitude of the frequency components.
Sine Wave function (Functions Analysis Signal
Generation subpalette) generates a time domain sinusoidal
waveform.
Real FFT function (Functions Analysis Digital Signal
Processing subpalette) computes the FFT of the input data
samples. The output of the Real FFT function is divided by
the FFT size (number of data points) to obtain the correct
sample values.
Complex to Polar function (Functions Numeric
Complex subpalette) separates the complex output of the
FFT into its magnitude and phase parts. The phase
information is in units of radians. Here you are displaying
only the magnitude of the FFT.
The frequency spacing, f, is given by dividing the sampling freq
by the # of samples.
2. Select frequency (Hz) = 10, sampling freq = 100, and
# of samples = 100. Run the VI.
Notice the plots of the time waveform and the frequency spectrum.
Because sampling freq = # of samples = 100, you are in effect
sampling for 1 second. Thus, the number of cycles of the sine wave
you see in the time waveform is equal to the frequency(Hz) you
select. In this case, you will see 10 cycles. (If you change the
frequency (Hz) to 5, you will see five cycles.)

National Instruments Corporation

3-13

LabVIEW Signal Processing Course Manual

Lesson 3 Signal Processing

Two-Sided FFT
3. Examine the frequency spectrum (the Fourier transform). You will
notice two peaks, one at 10 Hz and the other at 90 Hz. The peak at
90 Hz is actually the negative frequency of 10 Hz. The plot you see
is known as the 2-sided FFT because it shows both the positive and
the negative frequencies.
4. Run the VI with frequency (Hz) = 10 and then with
frequency (Hz) = 20. For each case, note the shift in both peaks of
the spectrum.
Note:

Also observe the time domain plot for frequency (Hz) = 10 and 20. Which
one gives a better representation of the sine wave? Why?
5. Because fs = 100 Hz, you can accurately sample only signals having
a frequency < 50 Hz (Nyquist frequency = fs/2). Change frequency
(Hz) to 48 Hz. You should see the peaks at +- 48 Hz on the spectrum
plot.
6. Now change frequency (Hz) to 52 Hz. Is there any difference
between the result of step 5 and what you see on the plots now?
Because 52 > Nyquist, the frequency of 52 is aliased to
|100 - 52| = 48 Hz.
7. Change frequency (Hz) to 30 Hz and 70 Hz and run the VI. Is there
any difference between the two cases? Explain why.
8. Save this VI as FFT_2sided.vi in the library Lvspcex.llb.

One-Sided FFT
9. Modify the block diagram of the VI as shown in the following
diagram. You have seen that the FFT had repetition of information
because it contained information about both the positive and the
negative frequencies. This modification now shows only half the
FFT points (only the positive frequency components). This
representation is known as the 1-sided FFT. The 1-sided FFT shows
only the positive frequency components. Note that you need to
multiply the positive frequency components by two to obtain the
correct amplitude. The DC component, however, is left untouched.

LabVIEW Signal Processing Course Manual

3-14

National Instruments Corporation

Lesson 3

Signal Processing

Equal To O? function (Functions Comparison subpalette) tests

to see if the array index is equal zero. If so, it corresponds to the
D.C. component and should not be multiplied by two.
10. Run the VI with the following values: frequency (Hz) = 30,
sampling freq = 100, # of samples = 100.
11. Change the value of frequency (Hz) to 70 and run the VI. Do you
notice any difference between this and the result of step 9?
12. Save the VI as FFT_1sided.vi in the library Lvspcex.llb.

End of Exercise 3-1

National Instruments Corporation

3-15

LabVIEW Signal Processing Course Manual

Lesson 3 Signal Processing

C. The Power Spectrum

You have seen that the DFT (or FFT) of a real signal is a complex number,
having a real and an imaginary part. The power in each frequency
component represented by the DFT/FFT can be obtained by squaring the
magnitude of that frequency component. Thus, the power in the kth
frequency component (the kth element of the DFT/FFT) is given by |X[k]|2.
The plot showing the power in each of the frequency components is known
as the power spectrum. Because the DFT/FFT of a real signal is
symmetric, the power at a positive frequency of kf is the same as the power
at the corresponding negative frequency of -kf (DC and Nyquist
components not included). The total power in the DC and Nyquist
components are

X[0]

---and X N
2

, respectively.

Loss of Phase Information

Because the power is obtained by squaring the magnitude of the DFT/FFT,
the power spectrum is always real and all the phase information is lost. If
you want phase information, you must use the DFT/FFT, which gives you a
complex output.
You can use the power spectrum in applications where phase information is
not necessary (for example, to calculate the harmonic power in a signal).
You can apply a sinusoidal input to a nonlinear system and see the power in
the harmonics at the system output.

Frequency Spacing Between Samples

You can use the Power Spectrum VI in the Analysis Digital Signal
Processing subpalette to calculate the power spectrum of the time domain
data samples. Just like the DFT/FFT, the number of samples from the
Power Spectrum VI output is the same as the number of data samples
applied at the input. Also, the frequency spacing between the output samples
is f = fs/N.

LabVIEW Signal Processing Course Manual

3-16

National Instruments Corporation

Lesson 3

Signal Processing

In the following table, the power spectrum of a signal x[n] is represented

by Sxx. If N is even, let p = N
---- . The following table shows the format
2

of the output sequence Sxx corresponding to the power spectrum.

Array Element

Interpretation

Sxx[0]

Power in DC component

Sxx[1] = Sxx[N-1]

Power at frequency f

Sxx[2] = Sxx[N-2]

Power at frequency 2f

Sxx[3] = Sxx[N-3]

Power at frequency 3f

.
.
.

.
.
.

Sxx[p-2]= Sxx[N-(p-2)]

Power at frequency (p-2)f

Sxx[p-1] = Sxx[N-(p-1)]

Power at frequency (p-1)f

Sxx[p]

Power at Nyquist frequency

The following illustration represents the information in the preceding

table for a sine wave with amplitude = 2 Vpeak (Vpk), and N = 8.

DC

negative
frequencies

positive
frequencies

Nyquist component

The output units of the Power Spectrum VI are in Volts rms squared
(V2rms). So, if the peak amplitude (Vpk) of the input signal is 2 Vpk, its
2
rms value is Vrms = ------ =
2

National Instruments Corporation

3-17

2,

so V2rms = 2. This value is divided

LabVIEW Signal Processing Course Manual

Lesson 3 Signal Processing

equally between the positive and negative frequency components,

resulting in the plot shown above.
1
If N is odd, let p = N
------------- . The following table shows the format of
2

the output sequence Sxx corresponding to the power spectrum.

Array Element

Interpretation

Sxx[0]

Power in DC component

Sxx[1] = Sxx[N-1]

Power at frequency f

Sxx[2]= Sxx[N-2]

Power at frequency 2f

Sxx[3] = Sxx[N-3]

Power at frequency 3f

.
.
.

.
.
.

Sxx[p-2] = Sxx[N-(p-2)]

Power at frequency (p-2)f

Sxx[p-1] = Sxx[N-(p-1)]

Power at frequency (p-1)f

Sxx[p] = Sxx[p]

Power at frequency pf

The following illustration represents the information in the preceding

table for N = 7.

DC

LabVIEW Signal Processing Course Manual

positive
frequencies

3-18

negative
frequencies

National Instruments Corporation

Lesson 3

Signal Processing

Exercise 3-2
Objective: To observe the difference between the FFT and the power spectrum representations.
1. Open the FFT_1sided VI (from the library Lvspcex.llb) that
you built in the previous exercise. Modify the block diagram and
front panel as shown below.

Front Panel

Block Diagram

Sine Wave function (Functions Analysis Signal

Generation palette) generates a time domain sinusoidal
waveform.

National Instruments Corporation

3-19

LabVIEW Signal Processing Course Manual

Lesson 3 Signal Processing

Real FFT function (Functions Analysis Digital Signal

Processing subpalette) computes the FFT of the input data
samples.
Array Subset (Functions Array subpalette) returns a
portion of the array. Here you are selecting half the array.
Complex to Polar function (Functions Numeric
Complex subpalette) separates the complex output of the
FFT into its magnitude and phase parts. The phase
information is in radians. Here, you are displaying only the
magnitude of the FFT.
The power spectrum is obtained by squaring the magnitude of the FFT.
The division by 2 (1.414) makes the conversion from Vpk to Vrms.
Note:

You could also have wired the output of the Sine Wave VI directly to the
input of the Power Spectrum VI (Analysis Digital Signal Processing
subpalette). The output of the Power Spectrum VI would directly be the
power spectrum of the signal. However, in that case, the phase information
would be lost.
2. Enter the following values in the controls: amplitude = 1.414,
frequency = 20 Hz, sampling freq = 100, and # of samples = 100,
and run the VI. Do you notice any difference in the FFT and power
spectrum representations?
3. Change the amplitude to 1.00 and run the VI. What difference do you
notice in the FFT and power spectrum representations?
4. Save the VI as FFT and Power Spectrum.vi in the library
Lvspcex.llb.

End of Exercise 3-2

LabVIEW Signal Processing Course Manual

3-20

National Instruments Corporation

Lesson 3

Signal Processing

Summary

The time domain representation (sample values) of a signal can be

converted into the frequency domain representation using the
discrete Fourier transform (DFT).

Fast calculation of the DFT is possible by using an algorithm known

as the fast Fourier transform (FFT). You can use this algorithm when
the number of signal samples is a power of two.

The output of the conventional DFT/FFT is two-sided because it

contains information about both the positive and the negative
frequencies. This output can be converted into a one-sided DFT/FFT
by using only half the number of output points.

The frequency spacing between the samples of the DFT/FFT is

f = fs/N.

The power spectrum can be calculated from the DFT/FFT by

squaring the magnitude of the individual frequency components. The
Power Spectrum VI in the advanced analysis library does this
automatically for you. The Power Spectrum VI units of the output
are V2rms. However, the power spectrum does not provide any phase
information.

The DFT, FFT, and power spectrum are useful for measuring the
frequency content of stationary or transient signals. The FFT
provides the average frequency content of the signal over the entire
time that the signal was acquired. For this reason, you use the FFT
mostly for stationary signal analysis (when the signal is not
significantly changing in frequency content over the time that the
signal is acquired), or when you want only the average energy at each
frequency line.

For measuring frequency information that changes during the

acquisition, you should use the joint time-frequency analysis (JTFA)
toolkit or the wavelet and filter banks designer (WFBD) toolkit.
These toolkits are covered in later lessons.

National Instruments Corporation

3-21

LabVIEW Signal Processing Course Manual

Lesson 3 Signal Processing

Review Questions
1. Which of the following provides you with both the magnitude and
phase information?
a. FFT
b. Power spectrum
c. DFT
d. Time domain waveform
2. Which of the following are true?
a. The magnitude spectrum is always even symmetric.
b. The DFT is a fast algorithm for computing the FFT.
c. The frequency spacing is given by
fs
f = --------------------------------------------number of samples

where fs is the sampling frequency.

d. An even number of samples always results in a two-sided
transform.
3. If you have 1024 samples, how many times faster is the FFT as
compared to the DFT in calculating the Fourier transform?

LabVIEW Signal Processing Course Manual

3-22

National Instruments Corporation

Lesson 3

Signal Processing

Notes

National Instruments Corporation

3-23

LabVIEW Signal Processing Course Manual

Lesson 3 Signal Processing

Notes

LabVIEW Signal Processing Course Manual

3-24

National Instruments Corporation

Lesson 4
Windowing
Introduction
In this lesson, you will learn about windows and how they affect the spectral
characteristics of a signal.

You Will Learn:

A. About spectral leakage and smoothing windows.
B. About the difference (both time and frequency domains) between a
windowed and a nonwindowed signal.
C. About the differences between the various types of windows in the
Analysis library and their applications.
D. How to separate two sine waves of large amplitude difference but
with frequencies very close to each other.

National Instruments Corporation

4-1

LabVIEW Signal Processing Course Manual

Lesson 4 Windowing

A. About Spectral Leakage and Smoothing Windows

In practical applications, you can obtain only a finite number of samples of
the signal. When you use the DFT/FFT to find the frequency content of a
signal, it is inherently assumed that the data that you have is a single period
of a periodically repeating waveform. This is shown below in Figure 4-1.
The first period shown is the one sampled. The waveform corresponding to
this period is then repeated in time to produce the periodic waveform.

discontinuity

One period

time

Figure 4-1. Periodic Waveform Created from Sampled Period

As seen in the previous figure, because of the assumption of periodicity

of the waveform, discontinuities between successive periods will occur.
This happens when you sample a noninteger number of cycles. These
artificial discontinuities turn up as very high frequencies in the
spectrum of the signal, frequencies that were not present in the original
signal. These frequencies could be much higher than the Nyquist
frequency, and as you have seen before, will be aliased somewhere
between 0 and fs/2. The spectrum you get by using the DFT/FFT
therefore will not be the actual spectrum of the original signal, but will
be a smeared version. It appears as if the energy at one frequency has
leaked out into all the other frequencies. This phenomenon is known
as spectral leakage.
Figure 4-2 shows a sine wave and its corresponding Fourier transform.
The sampled time domain waveform is shown in Graph 1. Because the
Fourier transform assumes periodicity, you repeat this waveform in
time, and the periodic time waveform of the sine wave of Graph 1 is
shown in Graph 2. The corresponding spectral representation is shown
in Graph 3. Because the time record in Graph 2 is periodic, with no
discontinuities, its spectrum is a single line showing the frequency of the
sine wave. The reason that the waveform in Graph 2 does not have any
discontinuities is because you have sampled an integer number of cycles
(in this case, 1) of the time waveform.

LabVIEW Signal Processing Course Manual

4-2

National Instruments Corporation

Lesson 4

Windowing

Figure 4-2. Sine Wave and Corresponding Fourier Transform

In Figure 4-3, you see the spectral representation when you sample a
noninteger number of cycles of the time waveform (namely 1.25).
Graph1 now consists of 1.25 cycles of the sine wave. When you repeat
this periodically, the resulting waveform, as shown in Graph 2, consists
of discontinuities. The corresponding spectrum is shown in Graph 3.
Notice how the energy is now spread over a wide range of frequencies.
This smearing of the energy is spectral leakage. The energy has leaked
out of one of the FFT lines and smeared itself into all the other lines.

Figure 4-3. Spectral Representation When Sampling a Nonintegral Number of Samples

National Instruments Corporation

4-3

LabVIEW Signal Processing Course Manual

Lesson 4 Windowing

Leakage exists because of the finite time record of the input signal. To
overcome leakage, one solution is to take an infinite time record, from
- to +. Then the FFT would calculate one single line at the correct
frequency. Waiting for infinite time is, however, not possible in practice.
So, because you are limited to having a finite time record, another
technique, known as windowing, is used to reduce the spectral leakage.
The amount of spectral leakage depends on the amplitude of the
discontinuity. The larger the discontinuity, the more the leakage, and
vice versa. You can use windowing to reduce the amplitude of the
discontinuities at the boundaries of each period. It consists of
multiplying the time record by a finite length window whose amplitude
varies smoothly and gradually towards zero at the edges. This is shown
in Figure 4-4, where the original time signal is windowed using a
Hamming window. Notice that the time waveform of the windowed
signal gradually tapers to zero at the ends. Therefore, when performing
Fourier or spectral analysis on finite-length data, you can use windows
to minimize the transition edges of your sampled waveform. A
smoothing window function applied to the data before it is transformed
into the frequency domain minimizes spectral leakage.

LabVIEW Signal Processing Course Manual

4-4

National Instruments Corporation

Lesson 4

Windowing

Note that if the time record contains an integral number of cycles, as

shown in Figure 4-2, the assumption of periodicity does not result in any
discontinuities, and thus there is no spectral leakage. The problem arises
only when you have a nonintegral number of cycles.

Figure 4-4. Time Signal Windowed Using a Hamming Window

National Instruments Corporation

4-5

LabVIEW Signal Processing Course Manual

Lesson 4 Windowing

B. Windowing Applications
There are several reasons to use windowing. Some of these are:

To define the duration of the observation.

Reduction of spectral leakage.

Separation of a small amplitude signal from a larger amplitude signal

with frequencies very close to each other.

LabVIEW Signal Processing Course Manual

4-6

National Instruments Corporation

Lesson 4

Windowing

C. Characteristics of Different Types of Window Functions

Applying a window to (windowing) a signal in the time domain is
equivalent to multiplying the signal by the window function. Because
multiplication in the time domain is equivalent to convolution in the
frequency domain, the spectrum of the windowed signal is a convolution of
the spectrum of the original signal with the spectrum of the window. Thus,
windowing changes the shape of the signal in the time domain, as well as
affecting the spectrum that you see.
Many different types of windows are available in the LabVIEW/
BridgeVIEW analysis library. Depending on your application, one may
be more useful than the others. Some of these windows are:
1. Rectangular (None): The rectangular window has a value of one over
its time interval. Mathematically, it can be written as:
w[n] = 1.0

for n = 0, 1, 2........N-1

where N is the length of the window. Applying a rectangular window

is equivalent to not using any window. This is because the
rectangular function just truncates the signal to within a finite time
interval. The rectangular window has the highest amount of spectral
leakage. The rectangular window for N = 32 is shown below:

The rectangular window is useful for analyzing transients that have

a duration shorter than that of the window. It is also used in order
tracking, where the sampling frequency is adjusted depending on the
speed of the shaft of a machine. In this application, it detects the
main mode of vibration of the machine and its harmonics.
2. Exponential: The shape of this window is that of a decaying
exponential. It can be mathematically expressed as:
n
w[n] = ------------- ln ( f )
N 1

for n = 0, 1, 2.......N - 1

where f is the final value. The initial value of the window is one, and
it gradually decays towards zero. The final value of the exponential

National Instruments Corporation

4-7

LabVIEW Signal Processing Course Manual

Lesson 4 Windowing

can be adjusted to between 0 and 1. The exponential window for

N = 32, with the final value specified as 0.1, is shown below:

final
value

This window is useful in analyzing transients (signals that exist only

for a short time duration) whose duration is longer than the length of
the window. This window can be applied to signals that decay
exponentially, such as the response of structures with light damping
that are excited by an impact (for example, a hammer).
3. Hanning: This window has a shape similar to that of half a cycle of
a cosine wave. Its defining equation is
w[n] = 0.5 - 0.5cos(2n/N)

for n = 0, 1, 2, .....N-1

A Hanning window with N = 32 is shown below:

The Hanning window is useful for analyzing transients longer than

the time duration of the window, and also for general-purpose
applications.
4. Hamming: This window is a modified version of the Hanning
window. Its shape is also similar to that of a cosine wave. It can be
defined as
w[n] = 0.54 - 0.46cos(2n/N)

LabVIEW Signal Processing Course Manual

4-8

for n = 0, 1, 2, .....N-1

National Instruments Corporation

Lesson 4

Windowing

A Hamming window with N = 32 is shown below:

You see that the Hanning and Hamming windows are somewhat
similar. However, note that in the time domain, the Hamming
window does not get as close to zero near the edges as does the
Hanning window.
5. Kaiser-Bessel: This window is a flexible window whose shape the
user can modify by adjusting the parameter beta. Thus, depending on
your application, you can change the shape of the window to control
the amount of spectral leakage. The Kaiser-Bessel window for
different values of beta are shown below:

National Instruments Corporation

4-9

LabVIEW Signal Processing Course Manual

Lesson 4 Windowing

Note that for small values of beta, the shape is close to that of a
rectangular window. Actually, for beta = 0 .0, you do get a
rectangular window. As you increase beta, the window tapers off
more to the sides.
This window is good for detecting two signals of almost the same
frequency, but significantly different amplitudes.
6. Triangle: The shape of this window is that of a triangle. It is given by
w[n] = 1 - | (2n-N) / N | for n = 0, 1, 2, ..., n1
A triangle window for N = 32 is shown below:

LabVIEW Signal Processing Course Manual

4-10

National Instruments Corporation

Lesson 4

Windowing

What Type of Window Do I Use?

Now that you have seen several of the many different types of windows that
are available, you may ask, What type of window should I use? The
answer depends on the type of signal you have and what you are looking for.
Choosing the correct window requires some prior knowledge of the signal
that you are analyzing. In summary, the following table shows the different
types of signals and the appropriate windows that you can use with them.
Type of signal

Window

Transients whose duration is shorter than the length of the window

Rectangular

Transients whose duration is longer than the length of the window

Exponential, Hanning

General-purpose applications

Hanning

Order tracking

Rectangular

System analysis (frequency response measurements)

Hanning (for random excitation),

rectangular (for pseudorandom
excitation)

Separation of two tones with frequencies very close to each other, but
with widely differing amplitudes

Kaiser-Bessel

Separation of two tones with frequencies very close to each other, but
with almost equal amplitudes

Rectangular

In many cases, you may not have sufficient prior knowledge of the
signal, so you need to experiment with different windows to find the
best one.

National Instruments Corporation

4-11

LabVIEW Signal Processing Course Manual

Lesson 4 Windowing

The following table summarizes the different windows.

Window
Rectangular
(None)

Exponential

Equation

Shape

w [ n ] = 1.0

Detecting transients
whose duration is
shorter than the length
of the window; order
tracking; separating two
tones with frequencies
and amplitudes very
close to each other;
system response

n
w [ n ] = ------------- ln ( f )
N1
where f = final value

Hanning

2n
w [ n ] = 0.5 0.5 cos ----------
N

Hamming

2n
w [ n ] = 0.54 0.46 cos ----------
N

Kaiser-Bessel

I0 ( 1 a )
w [ n ] = -----------------------------I0 ( )

Triangle

N
w [ n ] = 1 2n
---------------N

Transients whose
duration is longer than
the length of the
window

General-purpose
applications; system
analysis; transients
whose duration is longer
than the length of the
window

Separation of two tones

with frequencies very
close to each other, but
with almost equal
amplitudes

LabVIEW Signal Processing Course Manual

Applications

4-12

National Instruments Corporation

Lesson 4

Windowing

Exercise 4-1
Objective: To see the effect of windowing on spectral leakage.
1. Open the Spectral Leakage VI from the library Lvspcex.llb.
The VI is running when it opens. Using this VI, you can see the effect
of windowing on spectral leakage.

Note:

The Spectral Leakage VI searches for the Nyquist Shift VI. The Nyquist
Shift VI is in LabVIEW Examples Analysis dspxmpl.llb.
You can see three plots on the front panel:
Graph 1 shows the time record of the signal that has been sampled.
Graph 2 shows the repeated time record (assuming periodicity).
Graph 3 shows the frequency spectrum (in dB). The white line
shows the spectrum without windowing and the yellow line shows
the spectrum by windowing using a Hanning window.
You can use the cycles dial to control the number of time domain
waveform cycles that have been sampled. The display below the
Cycles dial tells you the exact number of cycles (to two decimal
places). You can also type a specific value in this display.
2. First, you will see the effect of windowing when you sample an
integral number of cycles.
Set the Cycles dial to 1.0. (or type 1.0 in the display beneath it.)

National Instruments Corporation

4-13

LabVIEW Signal Processing Course Manual

Lesson 4 Windowing

As you can see in Graph 1, you have exactly one cycle of the time
waveform. Graph 2 shows the repeated time record. Notice the
absence of any discontinuities, and the peak corresponding to the
sine wave in the frequency domain (Graph 3). You see two peaks
because you have the two-sided spectrum.
Notice the spreading around the frequency components. This
spreading is the effect of using a window function. In this case, you
used the Hanning window. Different windows have different
amounts of spreading.
3. Now see what happens when you sample a nonintegral number of
cycles. Set the Cycles dial to 1.3 and observe the difference in the
plots in Graph 3. Experiment by changing the cycles dial and
observing the waveforms in graphs 2 and 3.
In the white plot corresponding to No Window, the energy in the
frequency of interest spreads out across the spectrum. Hence the
frequency of interest is sometimes not clearly distinguishable. In the
yellow plot corresponding to Hanning Window, the spectral leakage
across the spectrum is reduced and the energy is more concentrated
around the frequency of interest.
4. Stop the VI by pressing the STOP button.
5. Close the VI. Do not save any changes.

End of Exercise 4-1

LabVIEW Signal Processing Course Manual

4-14

National Instruments Corporation

Lesson 4

Windowing

Exercise 4-2 (Optional)

Objective:

To see the difference (both time and frequency domains) between a windowed and
nonwindowed signal.
1. Build the VI front panel and block diagram as shown below.

Front Panel

National Instruments Corporation

4-15

LabVIEW Signal Processing Course Manual

Lesson 4 Windowing

Block Diagram

The Sine Pattern VI (Functions Analysis Signal

Generation subpalette) generates a sine wave with the
number of cycles specified in the cycles control.
The time waveform of the sine wave is windowed using the
Hamming Window VI (Functions Analysis Windows
subpalette), and both the windowed and nonwindowed time
waveforms are displayed on the left two plots on the front
panel.
The Amplitude and Phase Spectrum VI (Functions
Analysis Measurement subpalette) obtains the amplitude
spectrum of the windowed and nonwindowed time
waveforms. These waveforms are displayed on the two plots
on the right side of the front panel.
2. Set cycles to 10 (an integral number) and run the VI. Note that the
spectrum of the windowed signal is broader (wider) than the
spectrum of the nonwindowed signal. But both the spectra are
concentrated near 10 on the x-axis.
3. Change cycles to 10.25 (a nonintegral number) and run the VI. Note
that the spectrum of the nonwindowed signal is now more spread out
than it was before. This is because now you have a noninteger
number of cycles, and when you repeat the waveform to make it
periodic, you get discontinuities. The spectrum of the windowed
signal is still concentrated, but that of the nonwindowed signal has
now smeared all over the frequency domain. (This is spectral
leakage.)

LabVIEW Signal Processing Course Manual

4-16

National Instruments Corporation

Lesson 4

Windowing

4. Change cycles to 10.5 and observe the frequency domain plots.

Spectral leakage of the original signal is clearly apparent.
5. When you finish, save the VI as Windowed and Unwindowed
Signal.vi in the library Lvspcex.llb.
6. Close the VI.

End of Exercise 4-2

National Instruments Corporation

4-17

LabVIEW Signal Processing Course Manual

Lesson 4 Windowing

Exercise 4-3
Objective:

To learn about the different windows in the Analysis library.

1. Open the Window Plots VI from the library Lvxpcex.llb. It is
running when it opens.

The topmost plot shows you the shapes (in the time domain) of six
different types of windows in the Analysis library. They are all
shown on the same plot for comparison purposes.
The bottom three plots show the effect of multiplying a time domain
signal (a sine wave) by the window. The left plot shows the original
time signal, the middle plot shows the shape of the window being
applied, and the right plot shows the resulting signal.
In the Window Selector control, you can select one of the six
different types of windows.
The final value (exponential) control specifies the value to which
the exponential window should decay. This value is normally
between 0 and 1.
The value in the beta (Kaiser-Bessel) control can be adjusted to
change the shape of the Kaiser-Bessel window. The higher the value
of beta, the lower the spectral leakage, and vice versa.
2. Select different windows in the Window Selector control and
observe their shapes.
LabVIEW Signal Processing Course Manual

4-18

National Instruments Corporation

Lesson 4

Windowing

In particular, select the Rectangular (None) window. What

difference do you see in the first and the third plots shown on the
bottom? Explain.
Select the Exponential window. Observe the shape when you change
the value in the final value (exponential) control. What happens as
the final value increases? Decreases? Is equal to 1.0?
Select the Kaiser-Bessel window. Observe the shapes when you
change the value of beta between -10 and +10. What happens when
beta = 0.0?
3. When you finish, stop the VI by pressing the STOP button.
4. Close the VI. Do not save any changes.

End of Exercise 4-3

National Instruments Corporation

4-19

LabVIEW Signal Processing Course Manual

Lesson 4 Windowing

Exercise 4-4
Objective:

To use windows to separate two sine waves of almost the same frequency, but
widely differing amplitudes.
In this exercise, two sine waves of different amplitudes are summed
together and then transformed into the frequency domain. Sine wave 1
has a much smaller amplitude than sine wave 2. Without windowing, it
is not possible to distinguish between the two sine waves in the
frequency domain. With an appropriate choice of a window, you can
clearly separate the peaks in the frequency domain corresponding to the
two sine waves. The frequency domain plot shows the results so that you
can compare the effect of different window functions.
1. Open and run the Window Comparison VI from the library
Lvspcex.llb.

The frequency of each sine wave is adjustable with either the

knob or digital controls.

The amplitude of each sine wave is adjustable with the digital

controls.

You can select a different window function from the window 1

and window 2 controls.

2. Using the digital controls, set the amplitude of Sine Wave 1 as 0.001,
and that of Sine Wave 2 as 1.000. With the knob controls, set the

LabVIEW Signal Processing Course Manual

4-20

National Instruments Corporation

Lesson 4

Windowing

frequency of Sine Wave 1 to near 70, and that of Sine Wave 2 to near
60. In effect, you are adjusting the frequency of Sine Wave 2 using
the knob control, so that the smaller amplitude is nearer the larger
amplitude in the frequency domain plot.
3. Notice in the graph that when the frequency of the smaller amplitude
signal (Sine Wave 1) is closer to that of the larger amplitude signal
(Sine Wave 2), the peak corresponding to the smaller signal is not
detected. Applying a window function is the only way to detect the
smaller signal. The discontinuity is what causes the spectrum to
spread out. Signals at smaller amplitudes are lost in the sidelobes of
the larger amplitude signal.
4. Compare different window functions by choosing another window
from the window 1 and window 2 controls. Which one(s) can
distinguish between the two frequency components?
5. When you are done, stop the VI by clicking on the STOP button.
Close the VI. Do not save any changes.

End of Exercise 4-4

National Instruments Corporation

4-21

LabVIEW Signal Processing Course Manual

Lesson 4 Windowing

Summary

The DFT/FFT assumes that the finite time waveform that you obtain
is one period of a periodic signal that exists for all time. This
assumption of periodicity could result in discontinuities in the
periodic signal and gives rise to a phenomenon known as spectral
leakage, whereby the energy at a particular frequency leaks
throughout the spectrum.

To reduce the spectral leakage, the finite time waveform is

multiplied by a window function.

Windows can be used to separate two sine waves that have widely
different amplitudes, but are very close in frequency.

Review Questions
1. Why does spectral leakage occur?
2. Name four applications of using windows.
3. Which window(s) would you use for the following?
a. Separation of two tones with frequencies very close to each other.
b. System analysis.
c. Detecting the modes of vibration of a machine.

LabVIEW Signal Processing Course Manual

4-22

National Instruments Corporation

Lesson 4

Windowing

Notes

National Instruments Corporation

4-23

LabVIEW Signal Processing Course Manual

Lesson 4 Windowing

Notes

LabVIEW Signal Processing Course Manual

4-24

National Instruments Corporation

Lesson 5
Measurement
Introduction
You will now learn about some of the VIs that are already available in the
analysis library to perform various signal processing tasks. These VIs are
collectively referred to as measurement VIs.

You Will Learn:

A. About the measurement VIs and how they can perform various signal
processing operations.
B. How to calculate the frequency (amplitude and phase) spectrum of a
time domain signal, with the appropriate units.
C. About the coherent gain (CG) and equivalent noise bandwidth
(ENBW) window constants.
D. How to determine the total harmonic distortion present in a signal.

National Instruments Corporation

5-1

LabVIEW Signal Processing Course Manual

Lesson 5 Measurement

A. The Measurement VIs

The measurement VIs perform specific measurement tasks such as:

Calculating the total harmonic distortion present in a signal.

Determining the impulse response, or transfer function, of a system.

Estimating pulse parameters such as the rise time, overshoot, and

so on.

Computing the amplitude and phase spectrum of a signal.

Calculating the AC and DC components of a signal.

In the past, these computations have traditionally been performed by

benchtop instruments. The measurement VIs make these measurements
possible in the G programming language on your desktop computer.
They are built on top of the digital signal processing VIs and have the
following characteristics:

The input time-domain signal is assumed to be real valued.

Outputs are in magnitude and phase, scaled, and in the appropriate

units, ready for immediate graphing.

The spectrums calculated are single-sided and range from DC to

Nyquist (sampling frequency/2).

Wherever appropriate, corrections are automatically applied for the

windows being used. The windows are scaled so that each window
gives the same peak spectrum amplitude result within its amplitude
accuracy constraints.

In general, you can directly connect the inputs of the measurement VIs
to the output of data acquisition VIs. The outputs of the measurement
VIs can be connected to graphs for an appropriate visual display.

DAQ VIs

Measurement VIs

Graphing and
Plotting VIs

Several measurement VIs perform commonly used time

domain-to-frequency domain transformations such as calculation of the
amplitude and phase spectrum, the power spectrum, the network transfer
function, and so on. Other measurement VIs interact with VIs that
perform such functions as scaled time domain windowing and power and
frequency estimation.

LabVIEW Signal Processing Course Manual

5-2

National Instruments Corporation

Lesson 5

Measurement

B. Calculating the Frequency Spectrum of a Signal

In many applications, knowing the frequency content of a signal provides
insight into the system that generated the signal. The information thus
obtained can be used in the design of bridges, for calibration purposes, for
estimating the amount of noise and vibration generated by parts of
machines, and so on. The next exercise demonstrates how to use the
Amplitude and Phase Spectrum VI to identify two frequency
components.

National Instruments Corporation

5-3

LabVIEW Signal Processing Course Manual

Lesson 5 Measurement

Exercise 5-1
OBJECTIVE: To compute the frequency spectrum of a signal.

Front Panel

1. Open the Compute Frequency Spectrum VI found in the

Lvspcex.llb library. Two sine waves of frequencies 2 Hz and
10 Hz are superimposed. The 10 Hz sine wave has an amplitude of
2 V, and the 2 Hz sine wave has an amplitude of 1 V. The sampling
frequency is 100 Hz and 200 points of data are generated.
2. Switch to the block diagram.

LabVIEW Signal Processing Course Manual

5-4

National Instruments Corporation

Lesson 5

Measurement

Block Diagram

3. Examine the block diagram.

The Amplitude and Phase Spectrum VI (Analysis
Measurement subpalette) calculates the amplitude spectrum
and the phase spectrum of a time domain signal. The
connections to this VI are shown below.

The input time domain signal is applied at the Signal (V) control. The
magnitude and phase of the input signal spectrum are available at the
Amp Spectrum Mag (Vrms) and Amp Spectrum Phase (radians)
outputs, respectively.
Note:

The initial phase input to the Sine Wave VI is specified in degrees.

Note:

If the units of the input time domain signal are in volts peak (Vp), the units
of the magnitude of the amplitude spectrum is in volts rms (Vrms). The
relationship between the units is Vrms = Vp/ 2 = 0.707*Vp.
4. Run the VI.
The graph should display two peaks, one at 2 Hz and the other at
10 Hz. The amplitude of the 2 Hz sine wave is 0.717 V, and that of
the 10 Hz waveform is 1.414 V, which are the rms values for sine
waveforms of amplitudes 1 and 2 V, respectively. (The RMS of a
sine waveform = 0.707*peak amplitude.)

National Instruments Corporation

5-5

LabVIEW Signal Processing Course Manual

Lesson 5 Measurement

5. Change the phase of the sine waves by adjusting the Initial Phase 1
and Initial Phase 2 controls, and run the VI.
Do you notice any change in the time waveform? The spectrum?
6. Make the parameters of both the sine waves equal. That is, set
amplitude 1 = amplitude 2 = 2, frequency 1 = frequency 2 = 10,
Initial Phase 1 = Initial Phase 2 = 0, and sampling frequency = 200,
and run the VI.
Is the amplitude of the peak in the power spectrum the value that you
would expect?
7. When you are done, close the VI. Do not save any changes.

End of Exercise 5-1

LabVIEW Signal Processing Course Manual

5-6

National Instruments Corporation

Lesson 5

Measurement

C. Coherent Gain (CG) and Equivalent Noise Bandwidth (ENBW)

You saw in Lesson 4 that many different types of windows can be applied
to a signal. These windows have different shapes, and they affect the signal
in different ways. It is important to have some method to compare the effect
that different windows have on the signal. Two parameters that are useful in
comparing various types of windows are the coherent gain (CG) and the
equivalent noise bandwidth (ENBW).

Coherent Gain
The coherent gain of a window is the zero frequency gain (or the dc gain)
of the window. It is calculated by normalizing the maximum amplitude of
the window to one, and then summing the values of the window amplitudes
over the duration of the window. The result is then divided by the length of
the window (that is, the number of samples).
For example, consider the rectangular window shown below with
amplitude equal to A and nine samples:
A

0
You first normalize (divide) all the heights by A to get the maximum
height equal to one:
1

0
Then you add all the heights to get nine (nine lines each with a height
equal to one). This sum is then divided by the number of samples (nine)
to get a value of one. Thus, the CG of the rectangular window is equal
to one. Mathematically, the CG is given by
N1

1
CG = ---N

w[n]
n=0

where N is the total number of samples over the duration of the window
and w[n] are the normalized amplitudes of the samples.

National Instruments Corporation

5-7

LabVIEW Signal Processing Course Manual

Lesson 5 Measurement

For comparison purposes, the table below shows the CG of several

commonly used windows. Note that the rectangular (uniform) window
has the highest CG, whereas the CG of other windows is lower than that
of the rectangular window. Can you explain why this is so?
Window

CG

Uniform (None)

1.00

Hamming

0.54

Hanning

0.5

Triangle

0.5

Exact Blackman

0.46

Blackman

0.42

Blackman-Harris

0.42

4 Term B-Harris

0.36

Flat Top

0.28

7 Term B-Harris

0.27

Equivalent Noise Bandwidth (ENBW)

You can use the ENBW to compare frequency responses of different shapes.
An ideal frequency response is supposed to be rectangular in shape (see the
lesson on digital filters). However, in practice, the frequency response
differs from the ideal. Because different windows have different shapes of
their frequency response, they will pass different amounts of noise power.
The ENBW for a particular window is equal to the width of a frequency
response having an ideal rectangular shape that will pass the same amount
of noise power as the frequency response of that window.

LabVIEW Signal Processing Course Manual

5-8

National Instruments Corporation

Lesson 5

Measurement

Suppose that the solid line in the figure below shows the frequency
response of a window, and the dashed line shows the ideal rectangular
response. The responses are first adjusted to have a gain of unity at zero
frequency (DC).

amplitude

ideal
rectangular
response
frequency
response
of window

ENBW

frequency

To calculate the ENBW, the width of the rectangular response is

adjusted so that it has the same area as that of the nonideal response of
the window. This width is then equal to the ENBW for that window. It
is found that the ENBW is the smallest for the uniform window, and
larger for the other windows. The table below shows the ENBW for
different windows, relative to the ENBW for the uniform window.
Window

National Instruments Corporation

ENBW

Uniform (None)

1.00

Hamming

1.36

Hanning

1.5

Triangle

1.33

Exact Blackman

1.57

Blackman

1.73

Blackman-Harris

1.71

4 Term B-Harris

2.00

Flat Top

2.97

7 Term B-Harris

2.63

5-9

LabVIEW Signal Processing Course Manual

Lesson 5 Measurement

Exercise 5-2
Objective: To learn about the Coherent Gain and Equivalent Noise Bandwidth properties of
windows.
This exercise will familiarize you with different window shapes and their
CG and ENBW. The values of the CG and ENBW are used in other
measurement VIs such as the Power and Frequency Estimate VI and
Spectrum Unit Conversion VI.
1. Open the Time Domain Windows VI found in the library
Lvspcex.llb.
2. Open and examine the block diagram.

The VI that windows the input signal (a sine wave generated by the
Sine Wave VI) and gives as the output the resulting windowed
waveform is the Scaled Time Domain Window VI. The
connections to this VI are shown below.

The input time waveform is applied at the Waveform control, and the
window selection is done by the window control. The VI also outputs the
CG and the ENBW of the selected window at the window constants
terminal.
The resulting output waveform is automatically scaled so that when you
compute the amplitude or power spectrum of the windowed waveform,
all the windows will give the same value.

LabVIEW Signal Processing Course Manual

5-10

National Instruments Corporation

Lesson 5

Measurement

You will pass the sine wave through two of these VIs and compare the
resulting windowed waveforms.
3. Switch to the front panel.

The Window 1 and Window 2 controls select the two types of windows
that you want to apply to your signal (the sine wave).
The number of samples and the frequency of the sine wave are controlled
by the # samples and frequency controls, respectively. The sampling
frequency is adjusted by the fs control.
The topmost plot shows you the original time domain waveform
(without windowing). The lower two plots show you the signal after
application of the two windows specified in the Window 1 and
Window 2 controls.
4. Select Window 1 as None (Uniform) and Window 2 as Hanning and
run the VI. Leave all the controls at their default values.
Compare the waveforms in Plot 1 and Plot 2 and observe that using
the Uniform window is equivalent to not using any window.

National Instruments Corporation

5-11

LabVIEW Signal Processing Course Manual

Lesson 5 Measurement

Observe from Plot 2 and Plot 3 the difference in the windowed time
domain waveform due to the application of the uniform and Hanning
windows, respectively.
Note the differences in the CG and ENBW of the two windows.
5. The shapes of the Hamming and Hanning windows are very close to
each other. Choose these windows in the Window 1 and Window 2
controls. Run the VI and compare the waveforms in Plot 2 and
Plot 3.
Can you notice any difference? Which one is wider? In particular,
compare the values of the ENBW and the CG.
6. As mentioned before, the CG was the same as the DC gain. Choose
different windows and run the VI. Observe that multiplying the
maximum amplitude of the windowed signal by the CG of the
window gives unity.
7. When you finish, close the VI. Do not save any changes.

End of Exercise 5-2

LabVIEW Signal Processing Course Manual

5-12

National Instruments Corporation

Lesson 5

Measurement

D.Harmonic Distortion
When a signal, x(t), of a particular frequency (for example, f1) is passed
through a nonlinear system, the output of the system consists of not only the
input frequency (f1), but also its harmonics (f2 = 2*f1, f3 = 3*f1, f4 = 4*f1,
and so on). The number of harmonics, and their corresponding amplitudes,
that are generated depends on the degree of nonlinearity of the system. In
general, the more the nonlinearity, the higher the harmonics, and vice versa.
f1

nonlinear system

f1, 2f1, 3f1, 4f1, ...

An example of a nonlinear system is a system where the output y(t) is the

cube of the input signal x(t).
cos(t)

y(t) = f(x) = x3(t)

cos3(t)

So, if the input is

x(t) = cos(t),
the output is
x3(t) = 0.5*cos(t) + 0.25*[ cos(t) + cos(3t) ]
Therefore, the output contains not only the input fundamental frequency
of , but also the third harmonic of 3.

Total Harmonic Distortion

To determine the amount of nonlinear distortion that a system introduces,
you need to measure the amplitudes of the harmonics that were introduced
by the system relative to the amplitude of the fundamental. Harmonic
distortion is a relative measure of the amplitudes of the harmonics as
compared to the amplitude of the fundamental. If the amplitude of the
fundamental is A1, and the amplitudes of the harmonics are A2 (second
harmonic), A3 (third harmonic), A4 (fourth harmonic), ...AN (Nth
harmonic), the total harmonic distortion (THD) is given by
2

A 1 + A 2 + A 3 + ...A N
THD = ------------------------------------------------------------A1

and the percentage total harmonic distortion (% THD) is

2

100 A 1 + A 2 + A 3 + ...A N
%THD = ---------------------------------------------------------------------------A1

National Instruments Corporation

5-13

LabVIEW Signal Processing Course Manual

Lesson 5 Measurement

In the next exercise, you will generate a sine wave and pass it through a
nonlinear system. The block diagram of the nonlinear system is shown
below:

Verify from the block diagram that if the input is x(t) = cos(t), the
output is
y(t) = cos(t) + 0.5cos2(t) + 0.1n(t)
= cos(t) + [1 + cos(2t)]/4 + 0.1n(t)
= 0.25 + cos(t) + 0.25cos(2t) + 0.1n(t)
Therefore, this nonlinear system generates an additional DC component
as well as the second harmonic of the fundamental.

Using the Harmonic Analyzer VI

You can use the Harmonic Analyzer VI to calculate the %THD present in
the signal at the output of the nonlinear system. It finds the fundamental and
harmonic components (their amplitudes and corresponding frequencies)
present in the power spectrum applied at its input, and calculates the
percentage of total harmonic distortion (%THD) and the percentage of total
harmonic distortion plus noise (%THD + Noise). The connections to the
Harmonic Analyzer VI are shown below:

To use this VI, you need to give it the power spectrum of the signal
whose THD you want it to calculate. Thus, in this example, you need to
make the following connections:

LabVIEW Signal Processing Course Manual

5-14

National Instruments Corporation

Lesson 5

Measurement

The Scaled Time Domain Window VI applies a window to the output

y(t) of the nonlinear system (Your System). This is then passed on to the
Auto Power Spectrum VI, which sends the power spectrum of y(t) to
the Harmonic Analyzer VI, which then calculates the amplitudes and
frequencies of the harmonics, the THD, and the %THD.
You can specify the number of harmonics you want the VI to find in the
# harmonics control. Their amplitudes and corresponding frequencies
are returned in the Harmonic Amplitudes and Harmonic Frequencies
array indicators.
Note:

The number specified in the # harmonics control includes the

fundamental. So, if you enter a value of 2 in the # harmonics control, it
means to find the fundamental (say, of freq f1) and the second harmonic
(of frequency f2 = 2*f1). If you enter a value of N, the VI will find the
fundamental and the corresponding (N-1) harmonics.
The following are explanations of some of the other controls:
fundamental frequency is an estimate of the frequency of the
fundamental component. If left as zero (the default), the VI uses the
frequency of the non-DC component with the highest amplitude as the
fundamental frequency.
window is the type of window you applied to your original time signal.
It is the window that you select in the Scaled Time Domain Window
VI. For an accurate estimation of the THD, it is recommended that you
select a window function. The default is the uniform window.
sampling rate is the input sampling frequency in Hz.
The % THD + Noise output requires some further explanation. The
calculations for % THD + Noise are almost similar to that for % THD,
except that the noise power is also added to that of the harmonics. It is
given by
sum(APS)
%THD + Noise = 100 ---------------------------A1

where sum(APS) is the sum of the Auto Power Spectrum elements minus
the elements near DC and near the index of the fundamental frequency.
National Instruments Corporation

5-15

LabVIEW Signal Processing Course Manual

Lesson 5 Measurement

Exercise 5-3
Objective: To use the Harmonic Analyzer VI for harmonic distortion calculations.
1. Open the THD Example VI from the Lvspcex.llb library.
2. Switch to the block diagram.

Some of this will already be familiar to you. Your System is the nonlinear
system that you saw previously. Its output is windowed, and the power
spectrum calculated and given to the Harmonic Analyzer VI.
The Sine Wave VI generates a fundamental of frequency specified in the
fundamental frequency control.
The output of the Harmonic Analyzer VI is in Vrms (if the input from the
Auto Power Spectrum is in V2rms). This output is then squared to convert
it to V2rms.
3. Switch to the front panel.
At the bottom, you see a plot of the power spectrum of the output of
the nonlinear system. On the top right side are the array indicators
for the frequencies and amplitudes of the fundamental and its
harmonics. The size of the array depends on the value entered in the
# harmonics control.

LabVIEW Signal Processing Course Manual

5-16

National Instruments Corporation

Lesson 5

Measurement

4. Change the fundamental frequency to 1000, # harmonics to 2, and

run the VI several times. Each time, note the values in the output
indicators (Harmonic Frequencies, Harmonic Amplitudes, % THD,
and % THD + Noise).
Why do you get different values each time you run the VI?
Which of the values, % THD or % THD + Noise, is larger? Can you
explain why?
5. Run the VI with different selections of the window control and
observe the peaks in the power spectrum.
Which window gives the narrowest peaks? The widest? Can you
explain why?
Hint: See the values of the ENBW for each window in the table on
page 9.
6. Change the fundamental frequency to 3000 and run the VI.
Why do you get an error?
Hint: Consider the relationship between the Nyquist frequency and
the frequency of the harmonic(s).
7. When you finish, close the VI.

End of Exercise 5-3

National Instruments Corporation

5-17

LabVIEW Signal Processing Course Manual

Lesson 5 Measurement

Summary

The ready-made VIs to perform common measurements are available

in the Analysis Measurements subpalette.

Some of these measurement tasks include calculating the amplitude

and phase spectrum of a signal and the amount of harmonic
distortion. Other VIs calculate properties of a system such as its
transfer function, its impulse response, the cross power spectrum
between the input and output signals, and so on.

Because a real-world signal is time-limited by a window function,

the study of certain properties of these window functions is an
important consideration in interpreting the results of your
measurements.

The CG is a measure of the DC gain of the window, whereas the

ENBW is a measure of the amount of noise power that a window
introduces into a measurement.

Review Questions
1. Why is it important to know the coherent gain of a window?
2. Which Measurement VIs calculate both the amplitude and phase
spectrum of the input waveform?
3. Name some applications where you would use the Measurement VIs.
4. When a Measurement VI calculates the spectrum of a signal, is it a
one-sided or a two-sided spectrum?
5. What is the peak value corresponding to 2Vrms?
6. Which window has the highest coherent gain? Why?
7. What is the difference between %THD and %THD + Noise? Which
is larger?

LabVIEW Signal Processing Course Manual

5-18

National Instruments Corporation

Lesson 5

Measurement

Notes

National Instruments Corporation

5-19

LabVIEW Signal Processing Course Manual

Lesson 5 Measurement

Notes

LabVIEW Signal Processing Course Manual

5-20

National Instruments Corporation

Lesson 6
Digital Filtering
Introduction
In this lesson, you will learn about the characteristics of different types of
digital filters and how to use them in practical filtering applications.

You Will Learn:

A. What is filtering and why it is needed (applications).
B. About the frequency response characteristics of different types of
ideal filterslowpass, highpass, bandpass, bandstop.
C. About the differences between practical (nonideal) filters and ideal
filters.
D. About the advantages of digital filters over analog filters.
E. About the differences between IIR and FIR filters.
F. About the characteristics of different types of IIR filters.
G. About the transient response of IIR filters.
H. About the characteristics of FIR filters.

National Instruments Corporation

6-1

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

A. What Is Filtering?
Filtering is the process by which the frequency content of a signal is altered.
It is one of the most commonly used signal processing techniques. Common
everyday examples of filtering are the bass and treble controls on your
stereo system. The bass control alters the low-frequency content of a signal,
and the treble control alters the high-frequency content. By varying these
controls, you are actually filtering the audio signal. Some other applications
where filtering is useful are removing noise and performing decimation
(lowpass filtering the signal and reducing the sample rate).

LabVIEW Signal Processing Course Manual

6-2

National Instruments Corporation

Lesson 6

Digital Filtering

B. Ideal Filters
Filters alter or remove unwanted frequencies. Depending on the frequency
range that they either pass or attenuate, they can be classified into the
following types:

A lowpass filter passes low frequencies, but attenuates high

frequencies.

A highpass filter passes high frequencies, but attenuates low

frequencies.

A bandpass filter passes a certain band of frequencies.

A bandstop filter attenuates a certain band of frequencies.

frequency

fc
Lowpass

frequency

fc
Highpass

amplitude

amplitude

amplitude

amplitude

The ideal frequency response of these filters is shown below:

fc1 fc2
Bandpass

frequency

fc1 fc2
Bandstop

frequency

You see that the lowpass filter passes all frequencies below fc, whereas the
highpass filter passes all frequencies above fc. The bandpass filter passes all
frequencies between fc1 and fc2, whereas the bandstop filter attenuates all
frequencies between fc1 and fc2. The frequency points fc, fc1 and fc2 are
known as the cutoff frequencies of the filter. When designing filters, you
need to specify these cut-off frequencies.

fc
Lowpass

freq

National Instruments Corporation

SB

PB

fc
Highpass

freq

SB

PB

SB

fc1 fc2
Bandpass

6-3

amplitude

SB

amplitude

PB

amplitude

amplitude

The frequency range that is passed through the filter is known as the
passband (PB) of the filter. An ideal filter has a gain of one (0 dB) in the
passband so that the amplitude of the signal neither increases nor
decreases. The stopband (SB) corresponds to that range of frequencies
that do not pass through the filter at all and are rejected (attenuated). The
passband and the stopband for the different types of filters are shown
below:

freq

PB

SB

PB

fc1 fc2
Bandstop

freq

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

Note that whereas the lowpass and highpass filters have one passband and
one stopband, the bandpass filter has one passband, but two stopbands, and
the bandstop filter has two passbands, but one stopband.

How Filters Affect Signal Frequency Content

Suppose you have a signal containing frequencies of 10 Hz, 30 Hz, and
50 Hz. This signal is passed through a lowpass, highpass, bandpass, and
bandstop filter. The lowpass and highpass filters have a cutoff frequency of
20 Hz, and the bandpass and bandstop filters have cutoff frequencies of
20 Hz and 40 Hz. The output of the filter in each case is shown below:

10

10

30 50

30 50

freq

freq

20
Lowpass

20
Highpass
30 50

20
10

10

LabVIEW Signal Processing Course Manual

30 50

30 50

freq

freq

6-4

freq

10

freq

40

Bandpass

freq

30

20 40
Bandstop
10

50

freq

National Instruments Corporation

Lesson 6

Digital Filtering

C. Practical (Nonideal) Filters

The Transition Band
Ideally, a filter should have a unit gain (0 dB) in the passband, and a gain of
zero (- dB) in the stopband. However, in a real implementation, not all of
these criteria can be fulfilled. In practice, there is always a finite transition
region between the passband and the stopband. In this region, the gain of the
filter changes gradually from 1 (0 dB) in the passband to 0 (-) in the
stopband. The following diagrams show the passband, the stopband, and the
transition region (TR) for the different types of nonideal filters. Note that the
passband is now the frequency range within which the gain of the filter
varies from 0 dB to -3 dB. Although the -3 dB range is most commonly
used, depending on the application, other values (-0.5 dB, -1 dB, etc.) may
also be considered.

PB

TR

SB

SB

TR

PB

0dB
-3dB

0dB
-3dB

fc
Lowpass

SB

TR

PB

fc
Highpass

TR

SB
PB

0dB
-3dB

TR

SB

TR

PB

0dB
-3dB
fc1
fc2
Bandpass

fc1

fc2
Bandstop

Passband Ripple and Stopband Attenuation

In many applications, it is okay to allow the gain in the passband to vary
slightly from unity. This variation in the passband is called the passband
ripple and is the difference between the actual gain and the desired gain of
unity. The stopband attenuation, in practice, cannot be infinite, and you
must specify a value with which you are satisfied. Both the passband ripple
and the stopband attenuation are measured in decibels or dB, defined by:
dB = 20*log10(Ao(f)/Ai(f))
where log10 denotes the logarithm to the base 10, and Ai(f) and Ao(f) are
the amplitudes of a particular frequency f before and after the filtering,
respectively.

National Instruments Corporation

6-5

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

For example, for -0.02 dB passband ripple, the formula gives:

-0.02 = 20*log10(Ao(f)/Ai(f))
Ao(f)/Ai(f) = 10-0.001 = 0.9977
which shows that the ratio of input and output amplitudes is close to
unity.
If you have -60 dB attenuation in the stopband, you have
-60 = 20*log10(Ao(f)/Ai(f))
Ao(f)/Ai(f) = 10-3 = 0.001

amplitude

which means the output amplitude is 1/1000 of the input amplitude. The
following figure, though not drawn to scale, illustrates this concept.

0 dB

1.0
0.9977

-0.02 dB

passband
ripple
stopband
attenuation

0.001

-60 dB
frequency

Note:

Attenuation is usually expressed in decibels without the word minus, but

a negative dB value is normally assumed.

LabVIEW Signal Processing Course Manual

6-6

National Instruments Corporation

Lesson 6

Digital Filtering

D.Advantages of Digital Filters over Analog Filters

An analog filter has an analog signal at both its input and its output. Both
the input, x(t), and output, y(t), are functions of a continuous variable t and
can take on an infinite number of values. Analog filter design is about 50
years older than digital filter design. Thus, analog filter design books
featuring simple, well- tested filter designs exist and can be found
extensively in the literature. However, this type of filter design is often
reserved for specialists because it requires advanced mathematical
knowledge and understanding of the processes involved in the system
affecting the filter.
Modern sampling and digital signal processing tools have made it
possible to replace analog filters with digital filters in applications that
require flexibility and programmability. These applications include
audio, telecommunications, geophysics, and medical monitoring. The
advantages of digital filters over analog filters are:

They are software programmable, and so are easy to build and test.

They require only the arithmetic operations of multiplication and

addition/subtraction and so are easier to implement.

They are stable (do not change with time nor temperature) and
predictable.

They do not drift with temperature or humidity or require precision

components.

They have a superior performance-to-cost ratio.

They do not suffer from manufacturing variations or aging.

National Instruments Corporation

6-7

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

E. IIR and FIR Filters

1.0

Filter

amplitude

amplitude

IMPULSE

IMPULSE
RESPONSE

time

Fourier
Transform

amplitude

Another method of classification of filters is based on their impulse

response. But what is an impulse response? The response of a filter to an
input that is an impulse (x[0] = 1 and x[i] = 0 for all i 0) is called the
impulse response of the filter (see figure below). The Fourier transform
of the impulse response is known as the frequency response of the filter.
The frequency response of a filter tells you what the output of the filter
is going to be at different frequencies. In other words, it tells you the
gain of the filter at different frequencies. For an ideal filter, the gain
should be 1 in the passband and 0 in the stopband. So, all frequencies in
the passband are passed as is to the output, but there is no output for
frequencies in the stopband.

FREQUENCY
RESPONSE

time

frequency

If the impulse response of the filter falls to zero after a finite amount of time,
it is known as a finite impulse response (FIR) filter. However, if the
impulse response exists indefinitely, it is known as an infinite impulse
response (IIR) filter. Whether the impulse response is finite or not (that is,
whether the filter is FIR or IIR) depends on how the output is calculated.
The basic difference between FIR and IIR filters is that for FIR filters,
the output depends only on the current and past input values, whereas for
IIR filters, the output depends not only on the current and past input
values, but also on the past output values.
As an example, consider a cash register at a supermarket. Let x[k] be the
cost of the kth item that a customer buys, where 1 k N, and N is the
total number of items. The cash register adds the cost of each item to
produce a running total. This running total y[k] , up to the kth item,
is given by
y[k] = x[k] + x[k-1] + x[k-2] + x[k-3] + .....+ x[1]

(1a)

Thus, the total for N items is y[N]. Because y[k] is the total up to the kth
item, and y[k-1] is the total up to the (k-1)st item, you can rewrite
equation (1a) as
y[k] = y[k-1] + x[k]

LabVIEW Signal Processing Course Manual

(1b)

6-8

National Instruments Corporation

Lesson 6

Digital Filtering

If you add a sales tax of 8.25%, equations (1a) and (1b) can be
rewritten as
y[k] = 1.0825x[k] + 1.0825x[k-1] + 1.0825 x[k-2] + 1.0825x[k-3] +
... + 1.0825x[1]
(2a)
y[k] = y[k-1] + 1.0825x[k]

(2b)

Note that both equations (2a) and (2b) are identical in describing the
behavior of the cash register. The difference is that whereas (2a) is
implemented only in terms of the inputs, (2b) is implemented in
terms of both the input and the output. Equation (2a) is known as the
nonrecursive, or FIR, implementation. Equation (2b) is known as the
recursive, or IIR, implementation.

Filter Coefficients
In equation (2a), the multiplying constant for each term is 1.0825. In
equation (2b), the multiplying constants are 1 (for y[k-1]) and 1.0825 (for
x[k]). These multiplying constants are known as the coefficients of the
filter. For an IIR filter, the coefficients multiplying the inputs are known as
the forward coefficients, and those multiplying the outputs are known as
the reverse coefficients.
Equations of the form 1a, 1b, 2a, or 2b that describe the operation of the
filter are known as difference equations.

Advantages and Disadvantages of FIR and IIR Filters

Comparing IIR and FIR filters, the advantage of digital IIR filters over finite
impulse response (FIR) filters is that IIR filters usually require fewer
coefficients to perform similar filtering operations. Thus, IIR filters execute
much faster and do not require extra memory, because they execute in place.
The disadvantage of IIR filters is that the phase response is nonlinear. If
the application does not require phase information, such as simple signal
monitoring, IIR filters may be appropriate. You should use FIR filters
for those applications requiring linear phase responses. The recursive
nature of IIR filters makes them more difficult to design and implement.

National Instruments Corporation

6-9

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

F. Infinite Impulse Response Filters

You have seen that infinite impulse response filters (IIR) are digital filters
whose output is calculated by adding a weighted sum of past output values
with a weighted sum of past and current input values. Denoting the input
values by x[.] and the output values by y[.], the general difference equation
characterizing IIR filters is
a 0 y [ i ] + a 1 y [ i 1 ] + a 2 y [ i 2 ] + ... + a Ny 1 y [ i ( N y 1 ) ] =
b 0 x [ i ] + b 1 x [ i 1 ] + b 2 x [ i 2 ] + ... + b Nx 1 x [ i ( N x 1 ) ]
a 0 y [ i ] = a 1 y [ i 1 ] a 2 y [ i 2 ] + ... a Ny 1 y [ i ( N y 1 ) ] +
b 0 x [ i ] + b 1 x [ i 1 ] + b 2 x [ i 2 ] + ... + b Nx 1 x [ i ( N x 1 ) ]

1
y [ i ] = -----
a0

Ny 1

Nx 1

a [ j ]y [ i j ] +

j=1

b [ k ]x [ i k ]

k=0

(3)

where Nx is the number of forward coefficients (b[k]) and Ny is the number

of reverse coefficients (a[j]). The output sample at the present sample
index i is the sum of scaled present and past inputs (x[i] and x[i-k] when
j 0) and scaled past outputs (y[i-j]). Usually, Nx is equal to Ny and this
value is known as the order of the filter.
Note:

In all of the IIR filters implemented in LabVIEW/BridgeVIEW, the

coefficient a0 is 1.

Practical IIR Filters

A lower order reduces arithmetic operations and therefore reduces
computation error. A problem with higher order filtering is that you quickly
run into precision errors with orders much greater than 20-30. This is the
main reason for the cascade implementations over the direct form.
Refer to the Analysis VI Reference Manual for more details on cascade
form implementations. It is recommended that the orders of 1-20 are
reasonable, with 30 being an upper limit. A higher order also means more
filter coefficients and hence longer processing time.
The impulse response of the filter described by equation (3) is of infinite
length for nonzero coefficients. In practical filter applications, however,
the impulse response of stable IIR filters decays to near zero in a finite
number of samples.
In practice, the frequency response of filters differs from that of ideal
filters. Depending on the shape of the frequency response, the IIR filters
can be further classified into

Butterworth filters

Chebyshev filters

LabVIEW Signal Processing Course Manual

6-10

National Instruments Corporation

Lesson 6

Chebyshev II or inverse Chebyshev filters

Elliptic or Cauer filters

Digital Filtering

The characteristics of each filter type are described below.

Butterworth Filters

normalized amplitude

A Butterworth filter has no ripples in either the passband or the stopband.

Due to the lack of ripples, it is also known as the maximally flat filter. Its
frequency response is characterized by a smooth response at all frequencies.
The following illustration shows the response of a lowpass Butterworth
filter of different ordersthe x-axis scaling is in terms of f / fNyquist,
whereas the y-axis is scaled so that the gain in the passband is unity.

f / fNyquist
The region where the output of the filter is equal to 1 (or very close to 1)
is the passband of the filter. The region where the output is 0 (or very
close to 0) is the stopband. The region in between the passband and the
stopband where the output gradually changes from 1 to 0 is the transition
region.
The advantage of Butterworth filters is a smooth, monotonically
decreasing frequency response in the transition region. As seen from the
figure, the higher the filter order, the steeper the transition region.

Chebyshev Filters
The frequency response of Butterworth filters is not always a good
approximation of the ideal filter response because of the slow rolloff
between the passband (the portion of interest in the spectrum) and the
stopband (the unwanted portion of the spectrum). On the other hand,
Chebyshev filters have a smaller transition region than a Butterworth filter
of the same order. However, this is achieved at the expense of ripples in the
passband. Using LabVIEW or BridgeVIEW, you can specify the maximum

National Instruments Corporation

6-11

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

amount of ripple (in dB) in the passband for a Chebyshev filter. The
frequency response characteristics of Chebyshev filters have an equiripple
(ripples all have the same magnitude) magnitude response in the passband,
monotonically decreasing magnitude response in the stopband, and a
sharper rolloff in the transition region as compared to Butterworth filters of
the same order.

amplitude (dB)

The following graph shows the response of a lowpass Chebyshev filter

of different orders. In this case, the y-axis scaling is in decibels. Once
again, note that the steepness of the transition region increases with
increasing order. Also, the number of ripples in the passband increases
with increasing order.

f / fNyquist
The advantage of Chebyshev filters over Butterworth filters is the
sharper transition between the passband and the stopband with a
lower-order filter. As mentioned before, this produces smaller absolute
errors and higher execution speeds.

Chebyshev II or Inverse Chebyshev Filters

Chebyshev II, also known as inverse Chebyshev or Type II Chebyshev
filters, are similar to Chebyshev filters, except that Chebyshev II filters have
ripples in the stopband (as opposed to the passband), and are maximally flat
in the passband (as opposed to the stopband). For Chebyshev II filters, you
can specify the amount of attenuation (in dB) in the stopband. The
frequency response characteristics of Chebyshev II filters are equiripple
magnitude response in the stopband, monotonically decreasing magnitude
response in the passband, and a rolloff sharper than Butterworth filters of the
same order. The following graph plots the response of a lowpass Chebyshev
II filter of different orders.

LabVIEW Signal Processing Course Manual

6-12

National Instruments Corporation

Digital Filtering

amplitude (dB)

Lesson 6

f / fNyquist
The advantage of Chebyshev II filters over Butterworth filters is that
Chebyshev II filters give a sharper transition between the passband and the
stopband with a lower order filter. This difference corresponds to a smaller
absolute error and higher execution speed. One advantage of Chebyshev II
filters over regular Chebyshev filters is that Chebyshev II filters have the
ripples in the stopband instead of the passband.

Elliptic Filters
You saw that Chebyshev (type I or II) filters have a sharper transition region
than a Butterworth filter of the same order. This is because they allowed
ripples in the passband (type I) or the stopband (type II). Elliptic filters
distribute the ripples over both the passband as well as the stopband.
Equiripples in the passband and the stopband characterize the magnitude
response of elliptic filters. Therefore, compared with the same order
Butterworth or Chebyshev filters, the elliptic design provides the sharpest
transition between the passband and the stopband. For this reason, elliptic
filters are quite popular in applications where short transition bands are
required and where ripples can be tolerated. The following graph plots the
response of a lowpass elliptic filter of different orders. The x-axis scaling is
in terms of f / fNyquist, whereas the y-axis is scaled so that the gain in the
passband is unity.

National Instruments Corporation

6-13

LabVIEW Signal Processing Course Manual

normalized amplitude

Lesson 6 Digital Filtering

f / fNyquist

Notice the sharp transition edge for even low-order elliptic filters. For
elliptic filters, you can specify the amount of ripple (in dB) in the
passband as well as the attenuation (in dB) in the stopband.

IIR Filter Comparison

A comparison of the lowpass frequency responses for the four different IIR
filter designs, all having the same order (five), is shown in the figure below.
The Elliptic filter has the narrowest transition region, whereas the
Butterworth filter has the widest.

f/fNyquist

LabVIEW Signal Processing Course Manual

6-14

National Instruments Corporation

Lesson 6

Digital Filtering

The following table compares the filter types.

IIR Filter
Design

Response
Characteristics

Butterworth

No ripples

Chebyshev

Ripples in PB

Inverse
Chebyshev

Ripples in SB

Elliptic

Ripples in PB and SB

Width of Transition Region

for a Fixed Order

Order Required for

Given Filter Specifications

Widest

Highest

Narrowest

Lowest

The LabVIEW and BridgeVIEW digital filter VIs handle all the design
issues, computations, memory management, and actual data filtering
internally, and are transparent to the user. You do not need to be an
expert in digital filters or digital filter theory to process the data. All you
need to do is to specify the control parameters such as the filter order,
cutoff frequencies, amount of ripple, and stopband attenuation.

How Do I Decide which Filter to Use?

Now that you have seen the different types of filters and their
characteristics, the question arises as to which filter design is best suited for
your application. In general, some of the factors affecting the choice of a
suitable filter are whether you require linear phase, whether you can tolerate
ripples, and whether a narrow transition band is required. The following
flowchart is expected to serve as a guideline for selecting the correct filter.
Keep in mind that in practice, you may need to experiment with several
different options before finally finding the best one.

National Instruments Corporation

6-15

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

linear phase?

FIR Filter

N
ripple OK?

narrow
transition band?

Y
Elliptic

high order
Butterworth

Y
narrowest possible transition region?

low order
Butterworth

ripple
in PB?

Inv. Chebyshev

ripple
in SB?

Chebyshev

arbitrary
filter specs?

Elliptic

Y
FIR

LabVIEW Signal Processing Course Manual

6-16

National Instruments Corporation

Lesson 6

Digital Filtering

Exercise 6-1
Objective:

To filter data samples that consist of both high-frequency noise and a sinusoidal
signal.
In this exercise, you combine a sine wave generated by the Sine
Pattern VI with high-frequency noise. (The high-frequency noise is
obtained by highpass filtering uniform white noise with a
Butterworth filter.) The combined signal is then lowpass filtered by
another Butterworth filter to extract the sine wave.

1. Open a new VI and build the front panel as shown above.

a. Select a Digital Control from the Numeric palette and label it
Frequency.
b. Select Vertical Slide from the Numeric palette and label it
Cut-Off Frequency.
c. Select another Vertical Slide from the Numeric palette and label
it Filter Order.
d. Select a Waveform Graph from the Graph palette for
displaying the noisy signal, and another Waveform Graph for
displaying the original signal.

National Instruments Corporation

6-17

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

2. Build the block diagram as shown below.

Sine Pattern VI (Functions Analysis Signal Generation

subpalette) generates a sine wave of the desired frequency.
Uniform White Noise VI (Functions Analysis Signal
Generation subpalette) generates uniform white noise that is
added to the sinusoidal signal.
Butterworth Filter VI (Functions Analysis Filters
subpalette) highpass filters the noise.
Note that you are generating 10 cycles of the sine wave, and there are
1000 samples. Also, the sampling frequency to the Butterworth
Filter VI on the right side is specified as 1000 Hz. Thus, effectively
you are generating a 10 Hz signal.
3. Switch back to the front panel. Select a Frequency of 10 Hz, a
Cut-Off Frequency of 25 Hz, and a Filter Order of 5. Run the VI.
4. Reduce the Filter Order to 4, 3, and 2, and observe the difference in
the filtered signal. Explain what happens as you lower the filter
order.
In particular, observe the filtered waveform. At the beginning, there
is a flat region. The length of this region depends on the order of
the filter. Section G discusses this further.
5. When you finish, save the VI as Extract the Sine Wave.vi in the
Lvspcex.llb library.
6. Close the VI.

End of Exercise 6-1

LabVIEW Signal Processing Course Manual

6-18

National Instruments Corporation

Lesson 6

Digital Filtering

Exercise 6-2
Objective: To compare the frequency response characteristics of various IIR filters.
1. Open the IIR Filter Design VI from the Lvspcex.llb library.

The front panel offers the choice of different types of filters.

The Filter Design control selects one of the four different designs of
filters: Butterworth, Chebyshev, Chebyshev Type II, or Elliptic.
The Filter Type control selects one of four different types of filters:
highpass, lowpass, bandpass, or bandstop.
The Display control selects the display of the magnitude response
(magnitude of the frequency response) to be either linear or
logarithmic.

National Instruments Corporation

6-19

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

2. The block diagram is shown below.

As seen in the block diagram, note that the frequency response is

obtained by applying an impulse at the input of a filter and
calculating the Fourier transform of the output.
The Impulse Pattern VI (Functions Analysis Signal
Generation subpalette) generates an impulse that is given to
the selected filter. The number of sample points is equal to
1024.
The Real FFT VI (Functions Analysis Digital Signal
Processing subpalette) computes the Fourier transform of the
output of the filter.
The Array Subset VI (Functions Array subpalette) selects
513 FFT points out of 1024, so as to generate a one-sided
spectrum.
The Complex to Polar VI (Functions Numeric Complex
subpalette) converts the complex output of the Real FFT VI
to its polar (magnitude and phase) representation. The
magnitude can then be plotted in either the linear or dB
scales.

LabVIEW Signal Processing Course Manual

6-20

National Instruments Corporation

Lesson 6

Digital Filtering

Normally, the phase shown is limited to between - and +.

So, even if the phase lies outside the range - and + , it is
wrapped around to lie between these values. The Unwrap
Phase VI (Functions Analysis Digital Signal Processing
subpalette) is used to unwrap the phase to its true value,
even if its absolute value exceeds .
3. Select a Filter Design of Butterworth, Filter Type = Lowpass,
Ripple = 10, Attenuation = 40, Order = 4, Display = Logarithmic,
sampling rate = 1000, Lower Cut-Off Frequency = 100, and Higher
Cut-Off Frequency = 300. Run the VI.
4. Increase the filter Order to 5, 10, 15, and 20, and note the difference
in the magnitude and phase responses. In particular, what changes do
you notice in the transition region?
5. Keep the filter Order fixed at 5, and change the Filter Design to
select different IIR filters. Note the changes in the magnitude and
phase plots. For a given filter order, which of the four different filter
designs has the smallest transition region?
6. When you finish, stop the VI by clicking on the STOP button in the
lower right corner.
7. Close the VI. Do not save any changes.

End of Exercise 6-2

National Instruments Corporation

6-21

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

Exercise 6-3 (Optional)

OBJECTIVE: To use a digital filter to remove unwanted frequencies.
In this exercise, you will add two sine waves of different frequencies and
then filter the resulting waveform using a Butterworth lowpass filter to
obtain only one of the sine waves.
1. Open the Low Pass Filter VI from the library Lvspcex.llb. This
VI shows how to design a lowpass Butterworth filter to remove a
10 Hz signal from a 2 Hz signal.

Front Panel

The number of samples to be generated and the sampling frequency

is controlled by the samples and sampling frequency controls,
respectively.
The amplitude and frequency of the two sine waves can be controlled
by the amplitude 1, amplitude 2, frequency 1, and frequency 2
controls on the front panel.
The cutoff freq:fl control controls the cutoff frequency of the
lowpass filter, whose order is adjusted by the order control.

LabVIEW Signal Processing Course Manual

6-22

National Instruments Corporation

Lesson 6

Digital Filtering

2. Examine the block diagram.

Block Diagram

Sine Wave VI generates the two sine waves.

Amplitude and Phase Spectrum VI determines the
amplitude and phase spectrum of the output of the filtered
signal. In this exercise, you are interested in only the
amplitude spectrum.
The reciprocal of the sampling frequency gives the time interval, t,
between samples.
Remember that the frequency spacing, f, is obtained by dividing the
sampling frequency by the number of samples.
The two sine waves being combined together have frequencies of
10 Hz and 2 Hz. To separate them, you should set the cut off
frequency of the lowpass filter to somewhere between these two
values.
3. Keeping the cutoff freq: fl control at 7 Hz, run the VI. Observe that
the amplitude of the 2 Hz signal is much larger than that of the 10 Hz
signal.
4. Reduce the filter order to 5 and run the VI. Repeat with an order
of 3. What do you notice about the spectrum amplitudes?
5. Increase the order to 12 and run the VI. Observe the spectrum
amplitudes. Explain what happens.
6. When you finish, close the VI. Do not save any changes.

End of Exercise 6-3

National Instruments Corporation

6-23

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

G.The Transient Response of IIR Filters

You have seen that the output of a general IIR filter is given by
y [ i ] = a 1 y [ i 1 ] a 2 y [ i 2 ] ... a Ny 1 y [ i ( N y 1 ) ] + b 0 x [ i ]
+ b 1 x [ i 1 ] + b 2 [ i 2 ] + ... + b Nx 1 x [ i ( N x 1 ) ]

(4)

where Nx is the number of forward coefficients, Ny is the number of

reverse coefficients, and a0 is assumed to be equal to 1. Consider a
second-order filter where Nx = Ny = 2. The corresponding difference
equation is:
y [ i ] = a1 y [ i 1 ] a2 y [ i 2 ] + b0 x [ i ] + b1 x [ i 1 ] + b2 [ i 2 ]

(5)

To calculate the current output (at the ith instant) of the filter, you need
to know the past two outputs (at the (i-1)st and the (i-2)nd time instants)
as well as the current input (at the i th time instant) and past two inputs
(at the (i-1)st and (i-2)nd time instants).
Now suppose that you have just started the filtering process by taking the
first sample of the input data. However, at this time instant you do not
yet have the previous inputs (x[i-1] and x[i-2]) or previous outputs
(y[i-1] and y[i-2]). So, by default, these values are assumed to be zero.
When you get the second data sample, you already have the previous
input (x[i-1]) and the previous output (y[i-1]) that you calculated from
the first sample, but not yet x[i-2] and y[i-2]. Again, by default, these are
assumed to be zero. It is only after we start processing the third input
data sample that all the terms on the right hand side of equation (5) above
now have the previously calculated values. Thus, there is a certain
amount of delay before which there are calculated values for all the
terms on the RHS of the difference equation describing the filter. The
output of the filter during this time interval is a transient and is known
as the transient response. For lowpass and highpass filters implemented
in the LabVIEW/BridgeVIEW Analysis library, the duration of the
transient response, or delay, is equal to the order of the filter. For
bandpass and bandstop filters, this delay is 2*order.
IIR filters in the analysis library contain the following properties.

Negative indices resulting from equation (4) are assumed to be zero

the first time you call the VI.

Because the initial filter state is assumed to be zero (negative

indices), a transient proportional to the filter order occurs before the
filter reaches a steady state. The duration of the transient response,
or delay, for lowpass and highpass filters is equal to the order of the
filter:
delay = order

LabVIEW Signal Processing Course Manual

6-24

National Instruments Corporation

Lesson 6

Digital Filtering

The duration of the transient response for bandpass and bandstop

filters is twice the filter order:
delay = 2 * order

So, each time one of the filter VIs is called, this transient appears at the
output. You can eliminate this transient response on successive calls by
enabling the state memory of the VI. To enable state memory, set the
init/cont control of the VI to TRUE (continuous filtering). You will see
how to do this in a later exercise.

Original Signal
Filtered Signal

The number of elements in the filtered sequence equals the number

of elements in the input sequence.

The filter retains the internal filter state values when the filtering
completes.

National Instruments Corporation

6-25

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

Exercise 6-4
Objective: To see the difference in the filter response with and without enabling state memory.
1. Open the Low Pass Filter VI from Lvspcex.llb.

2. Run the VI with the values shown on the front panel above, but with
order = 7. (Do not worry about the init/cont (init: F) control for
now.)
Observe the upper Sine Waveform graph, which shows two plots.
The white dashed plot is the combined signal, whereas the green full
line plot is the filtered signal.
3. Change the filter order to 10 and run the VI. Observe the first few
values of the plot corresponding to the filtered signal.
4. Change the filter order to 15, 20, and 25, and run the VI. Each time,
observe the transient that occurs at the beginning of the plot
corresponding to the filtered signal. This transient can be removed
after the first call to the VI by enabling its state memory. This is done
by setting the init/cont control of the VI to TRUE. Setting this
control to TRUE is equivalent to continuous filtering, and except for
the first call to the VI, each successive call will not have the
transient.

LabVIEW Signal Processing Course Manual

6-26

National Instruments Corporation

Lesson 6

Digital Filtering

5. Now connect a Boolean control as shown in the diagram to the

init/cont input of the Butterworth Filter VI.

6. With the init/cont(init:F) control set to OFF, and order = 15, run the
VI several times. Note the presence of the transient at the beginning
of the filtered signal each time the VI runs.
7. Now set the init/cont(init:F) control to ON, and run the VI several
times. Observe that the transient is present only the first time that the
VI is run. On successive calls to the VI, the transient no longer exists.
8. When you finish, save the VI.

End of Exercise 6-4

National Instruments Corporation

6-27

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

H.Finite Impulse Response Filters

You have seen that the output of an IIR filter depends on the previous
outputs as well as the current and previous inputs. Because of this
dependency on the previous outputs, the IIR filter has infinite memory,
resulting in an impulse response of infinite length.
On the other hand, the output of an FIR filter depends only on the current
and past inputs. Because it does not depend on the past outputs, its
impulse response decays to zero in a finite amount of time. The output
of a general FIR filter is given by
y[i] = b0x[i] + b1x[i-1] + b2x[i-2] + ... + bNx[i-(M-1)]
where M is the number of taps of the filter and b0, b1, ... bM-1, are its
coefficients. FIR filters have some important characteristics:
1. They can achieve linear phase response, and hence they can pass a
signal without phase distortion.
2. They are always stable. During filter design or development, you do
not need to worry about stability concerns.
3. FIR filters are simpler and easier to implement.
The following graphs plot a typical magnitude and phase response of
FIR filters versus normalized frequency. The discontinuities in the phase
response arise from the discontinuities introduced when you compute
the magnitude response using the absolute value. Notice that the
discontinuities in phase are on the order of . The phase, however, is
clearly linear.
Magnitude(dB)
(dB)
Magnitude

1.20
1.00
0.80
0.60
0.40
0.20

f (Hz)

0.00
0.00

0.10

0.20

0.30

0.40

0.50

Phase(radians)
(radians)
Phase

0.0
-2.0
-4.0
-6.0
-8.0
-10.0

f (Hz)
-12.0
0.00

LabVIEW Signal Processing Course Manual

0.10

6-28

0.20

0.30

0.40

0.50

National Instruments Corporation

Lesson 6

Digital Filtering

Exercise 6-5
Objective: To see the frequency response characteristics of FIR filters.
In this exercise, you will see the magnitude and phase response
characteristics of FIR filters. You will also see the effect of using different
windows on the filter response characteristics.
1. Open the FIR Windowed Filter Design VI from Lvspcex.llb.

The Filter Type control specifies the type of FIR filterlowpass,

highpass, bandpass, or bandstop.
The taps control specifies the number of filter coefficients. Note that
taps must be greater than 0. The higher the number of taps, the steeper
the transition between the passband and the stopband.
The window control selects among nine different types of windows to be
applied to the input signal before it is filtered.
The Display control selects the display units on the Magnitude plot to be
linear or logarithmic.
The sampling frequency: fs, low cutoff frequency: fl, and high cutoff
frequency: fh controls specify the desired response characteristics of the
filter. The high cutoff frequency: fh is required only when Filter Type
is chosen as bandpass or bandstop.

National Instruments Corporation

6-29

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

The Magnitude and Phase plots show you the magnitude and phase
response of the filter. Note that because the filter under consideration is
an FIR filter, it is expected to have a linear phase response.
2. Observe the block diagram.

Impulse Pattern VI (Analysis Signal Generation subpalette)

generates an impulse to be applied to the input of the FIR filter.
The response of the filter to the impulse will give us its impulse
response. Finding the FFT of the impulse response gives the
magnitude and phase response.
FIR Windowed Filter VI (Analysis Filters subpalette) is an
FIR filter that filters the input data sequence (in this case an
impulse) using the set of windowed FIR filter coefficients
specified by the sampling frequency: fs, low cutoff frequency:
fl, high cutoff frequency: fh, and taps controls. This filter also
windows the input signal with the type of window selected in the
window control.
Zero Padder VI (Analysis Digital Signal Processing
subpalette) resizes the input sequence to the next higher power
of two by adding zeros to the end of the sequence. By converting
the total number of samples to a power of two, the calculation of
the Fourier transform of the impulse response of the FIR filter
can be done faster by the Real FFT VI.
The default values have been chosen so that you will see the response
characteristics of a lowpass FIR filter with cut-off frequency of 15 Hz.

LabVIEW Signal Processing Course Manual

6-30

National Instruments Corporation

Lesson 6

Digital Filtering

You are not using any window on the input signal (that is, you are using
a rectangular window) that is sampled at 100 Hz.
3. Run the VI with the default values. Observe the linear phase
response in the passband.
4. Select different types of windows using the window control and
observe both the magnitude and phase responses. Notice how the
choice of a window affects both the responses. The phase in all cases
will be linear.
The taps control affects the width of the transition region.
5. Observe the Magnitude response plot with the Filter Type control set
to lowpass filter, the low cutoff frequency: fl set to 15 Hz, sampling
frequency: fs set to 100 Hz, window set to Kaiser-Bessel, and taps
set to 33.
6. Change taps to 55 and observe how the transition region becomes
narrower.
7. Change taps to 10 and observe the increase in the width of the
transition region.
8. You can experiment with different values of the other controls.
When you are done, stop the VI by clicking on the STOP button in
the lower right corner.
9. Close the VI without saving any changes.

End of Exercise 6-5

National Instruments Corporation

6-31

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

I. Digital Filter Design Toolkit

While it is quite easy and straightforward to design a digital filter in
LabVIEW or BridgeVIEW using the analysis VIs, it does require a basic
understanding of digital signal theory. For the best tool to learn more about
digital filter design, National Instruments offers an add-on toolkit called the
Digital Filter Design (DFD) Toolkit. This toolkit was developed using G
and has an excellent interactive graphical user interface. By drawing lines
in the frequency domain, all types of digital filters can be easily designed.
The designed filter coefficients can then be integrated into LabVIEW,
BridgeVIEW, LabWindows/CVI, and other programming environments.
You will learn more about the DFD Toolkit in Lesson 10.

LabVIEW Signal Processing Course Manual

6-32

National Instruments Corporation

Lesson 6

Digital Filtering

Summary

You have seen from the frequency response characteristics that

practical filters differ from ideal filters.

For practical filters, the gain in the passband may not always be
equal to 1, the attenuation in the stopband may not always be -, and
there exists a transition region of finite width.

The width of the transition region depends on the filter order, and the
width decreases with increasing order.

The output of FIR filters depends only on the current and past input
values, whereas the output of IIR filters depends on the current and
past input values as well as the past output values.

IIR filters can be classified according to the presence of ripples in the

passband and/or the stopband.

Because of the dependence of its output on past outputs, a transient

appears at the output of an IIR filter each time the VI is called. This
transient can be eliminated after the first call to the VI by setting its
init/cont control to a TRUE value.

Review Questions
1. Name three practical examples where filtering is used.
2. If the stopband attenuation of a filter is -80 dB, what is the output of
the filter to an input signal of 5 V peak if the frequency of the signal
lies in the stopband of the filter?
3. Which filter would you use for the following applications?
a. When you want the narrowest possible transition region with the
smallest order.
b. When you cannot tolerate any ripples in either the passband or
the stopband.
c. When linear phase is important.
4. Why does a transient initially appear at the output of a filter? How
can it be eliminated on successive calls to a VI?

National Instruments Corporation

6-33

LabVIEW Signal Processing Course Manual

Lesson 6 Digital Filtering

Notes

LabVIEW Signal Processing Course Manual

6-34

National Instruments Corporation

Lesson 7
Curve Fitting
Introduction
In this lesson, you will learn about the Analysis VIs that are used to fit
curves to data points.

You Will Learn:

A. About curve fitting and its applications.
B. About the General Least Squares Linear Fit VI.
C. About the Nonlinear Levenberg-Marquardt Fit VI.
D. About fitting a curve to Gaussian (Normal) data points.

National Instruments Corporation

7-1

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

A. About Curve Fitting

In the digital domain, a data set can be represented by two input sequences,
Y Values and X Values. A sample or point in the data set can be written as
(x[i], y[i])
where x[i] is the ith element of the sequence X Values, and y[i] is the ith
element of the sequence Y Values. Each y[i] is related to the
corresponding x[i]. You are interested in finding the relationship
between the y[i] and the x[i] in the digital domain, and expressing it in
the form of an equation in the analog domain.
Curve fitting acts as a bridge between the digital and analog worlds.
Using curve fitting, digital data can be represented with a continuous
model having a certain set of parameters. The basic idea is to extract a
set of curve parameters or coefficients from the data set to obtain a
functional description of the data set. This functional description
consists of the set of parameters a0, a1, ..., ak that best matches the
experimental model from which the data samples x[i] and y[i] were
obtained. Once you obtain the functional model, you can use it to
estimate missing data points, interpolate data, or extrapolate data.
Note:

Note that y[i] is a function of both the parameters ak, as well as the data
x[i]. The following discussion refers to the terms linear and nonlinear
often. These terms refer to the relationship between y and a, and not y
and x.
The Analysis library offers both linear and nonlinear curve fitting
algorithms. The different types of curve fitting in LabVIEW are outlined
below:

Linear Fitfits experimental data to a straight line of the form

y = mx + c.
y[i]=a0+a1*x[i]

Exponential Fitfits data to an exponential curve of the form

y = aexp(bx)
y[i] = a0*exp(a1*x[i])

General Polynomial Fitfits data to a polynomial function of the

form y = a + bx + cx2 + ...
y[i] = a0+a1*x[i]+a2*x[i]2...
but with selectable algorithms for better precision and accuracy.

LabVIEW Signal Processing Course Manual

7-2

National Instruments Corporation

Lesson 7

Curve Fitting

General Linear Fitfits data to

y[i] = a0+a1*f1(x[i])+a2*f2(x[i]) + ...
where y[i] is a linear combination of the parameters a0, a1, a2.... The
general linear fit also features selectable algorithms for better
precision and accuracy. For example, y = a0 + a1*sin(x) is a linear fit
because y has a linear relationship with parameters a0 and a1.
Polynomial fits are always linear fits for the same reason. But special
algorithms can be designed for the polynomial fit to speed up the
fitting processing and improve accuracy.

Nonlinear Levenberg-Marquardt Fitfits data to

y[i] = f(x[i], a0, a1, a2...)
where a0, a1, a2... are the parameters. This method is the most general
method and does not require y to have a linear relationship with a0,
a1, a2.... It can be used to fit linear or nonlinear curves, but is almost
always used to fit a nonlinear curve, because the general linear fit
method is better suited to linear curve fitting. The LevenbergMarquardt method does not always guarantee a correct result, so it is
absolutely necessary to verify the results.

Mean Squared Error

The algorithm used to fit a curve to a particular data set is known as the
Least Squares method. Let the observed data set be denoted by y(x), and let
f(x,a) be the functional description of the data set where a is the set of curve
coefficients that best describes the curve. The error e(a) between the
observed values, and its functional description, is defined as
e(a) = [f(x,a) - y(x)]
For example, let a be the vector a = {a , a }. The functional description
0
1
of a line is
f(x,a) = a + a x.
0

The least squares algorithm estimates the values for a from the values of
y[i] and x[i]. After you have the values for a, you can obtain an estimate
of the observed data set for any value of x using the functional
description f(x,a). The curve fitting VIs automatically set up and solve
the necessary equations and return the set of coefficients that best
describes your data set. You can thus concentrate on the functional
description of your data and not worry about the methods used for
solving for a.
For each of the observed data points x[i], the differences between the
polynomial value f(x[i],a) and the original data y(x[i]) are called the
residuals and are given by
ei(a) = fi(x[i],a) - yi(x[i])
National Instruments Corporation

7-3

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

The mean square error (MSE) is a relative measure of these residuals

between the expected curve values calculated by the functional
description f(x[i],a), and the actual observed values of y[i], and is
given by
N1

1
MSE = ---N

N1

1
( f i y i ) = ---N
2

i=0

(e (a))

i=0

where f is the sequence representing the fitted values, Y is the sequence

representing the observed values, and N is the number of sample points
observed. The smaller the MSE, the better is the fit between the
functional description and the observed y(x).
In general, for each predefined type of curve fit, there are two types of
VIs, unless otherwise specified. One type returns only the coefficients.
The other type returns the coefficients, the corresponding expected or
fitted curve, and the MSE.

Applications of Curve Fitting

The practical applications of curve fitting are numerous. Some of them
are listed below.

Removal of measurement noise.

Filling in missing data points (for example, if one or more

measurements were missed or improperly recorded).

Interpolation (estimation of data between data points) (for example,

if the time between measurements is not small enough).

Extrapolation (estimation of data beyond data points) (for example,

if you are looking for data values before or after the measurements
were taken).

Differentiation of digital data. (For example, if you need to find the

derivative of the data points. The discrete data can be modeled by a
polynomial, and the resulting polynomial equation can be
differentiated.)

Integration of digital data (for example, to find the area under a curve
when you have only the discrete points of the curve).

To obtain the trajectory of an object based on discrete measurements

of its velocity (first derivative) or acceleration (second derivative).

LabVIEW Signal Processing Course Manual

7-4

National Instruments Corporation

Lesson 7

Curve Fitting

Exercise 7-1
OBJECTIVE: To perform a linear curve fit on experimental data.

Front Panel

1. Open the Linear Curve Fit VI from the library Lvspcex.llb.

This example assumes that you have collected 10 pairs of
experimental data t and y, and have reason to believe there may be a
linear relationship between each pair.
2. On the front panel, the input data control on the left shows the values
of the data points t[i] and y[i]. After calculating the equation for the
linear fit, the calculated values of y[i] for the measured values of t[i]
are shown on the fitted data indicator on the right. The VI also gives
the values of the parameters a0 and a1 (a and b indicators on the front
panel) and the resulting MSE.
3. Switch to the block diagram.

National Instruments Corporation

7-5

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

Block Diagram

4. Examine the block diagram.

Linear Fit VI (Analysis Curve Fitting subpalette). In this
exercise, this VI fits the data to a line and finds the
coefficients a and b such that y[i] = a + b*t[i], as well as the
mean squared error between the data and the linear fit.
5. The input data is in a two-dimensional array, which is a common
format when the data is collected from DAQ hardware. You use the
Index Array VI to obtain two one-dimensional arrays, y[i] and t[i].
6. The MSE is the mean squared error. A smaller error indicates a
better fit.
7. Run the VI. The graph should display the original data, as well as the
linear fit.
8. Change the values of y[i] in the input data control. Observe the
corresponding plots and the effect on the MSE and slope when there
are points that dont fit well.
9. When you finish, stop the VI by clicking on the STOP button.
10. Close the VI. Do not save any changes.

End of Exercise 7-1

LabVIEW Signal Processing Course Manual

7-6

National Instruments Corporation

Lesson 7

Curve Fitting

Exercise 7-2
OBJECTIVE: To perform a polynomial curve fit on experimental data.

Front Panel

1. Open the Polynomial Fit VI from the library Lvspcex.llb. This

example assumes that the experimental input data has a polynomial
relationship where
y[i] = a0 + a1*t[i] + a2*t[i]2 ...
2. When the polynomial order is 1, there are two coefficients (a0 and a1)
and the result is a linear fit as in exercise 7-1. However, when the
order is 2, it is a second order polynomial fit with three coefficients.
The polynomial coefficients are stored in an array a[i]. You can use
the polynomial order control to choose the order of the polynomial.
3. Switch to the block diagram.

National Instruments Corporation

7-7

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

Block Diagram

4. Examine the block diagram.

General Polynomial Fit VI (Analysis Curve Fitting
subpalette). In this exercise, this VI fits the data to a
second-order polynomial curve and returns the fit data, the
coefficients, and the mean squared error between the data
and the polynomial fit.
5. You use the polynomial fit to obtain the fitting coefficients a0, a1, a2,
etc. In general, you want to use the lowest order possible to fit the
polynomial.
6. Run the VI. The original as well as the fitted data should appear on
the graph.
7. Change the values of y[i] in the input data control and the order of
the polynomial in the polynomial order control, and observe the
changes in the plots and the MSE.
8. When you finish, stop the VI by clicking on the STOP button.
9. Close the VI. Do not save any changes.

End of Exercise 7-2

LabVIEW Signal Processing Course Manual

7-8

National Instruments Corporation

Lesson 7

Curve Fitting

Exercise 7-3
OBJECTIVE: To use and compare the Linear, Exponential and Polynomial Curve Fit VIs to obtain
the set of least square coefficients that best represent a set of data points.
1. Open the Regressions Demo VI from the library Lvspcex.llb.
The front panel and block diagram are already built for you.

Front Panel

This VI generates noisy data samples that are approximately linear,

exponential, or polynomial. It then uses the corresponding analysis
curve fitting VIs to determine the parameters of the curve that best fits
those data points. (At this stage, you do not need to worry about how the
noisy data samples are generated.) You can control the noise amplitude
with the Noise Level control on the front panel.

National Instruments Corporation

7-9

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

Block Diagram

2. Select Linear in the Algorithm Selector control, and set the Noise
Level control to about 0.1. Run the VI. Note the spread of the data
points and the fitted curve (straight line).
3. Experiment with different values of Order and Noise Level. What do
you notice? How does the mse change?
4. Change the Algorithm Selector to Exponential and run the VI.
Experiment with different values of Order and Noise Level. What do
you notice?
5. Change the Algorithm Selector to Polynomial and run the VI.
Experiment with different values of Order and Noise Level. What do
you notice?

LabVIEW Signal Processing Course Manual

7-10

National Instruments Corporation

Lesson 7

Curve Fitting

6. In particular, with the Algorithm Selector control set to Polynomial,

change the Order to 0 and run the VI. Then change it to 1 and run
the VI. Explain your observations.
7. Depending on your observations in steps 2, 3, 4, and 5, for which of
the algorithms (Linear, Exponential, Polynomial) is the Order
control the most effective? Why?
8. Close the VI. Do not save any changes.

End of Exercise 7-3

National Instruments Corporation

7-11

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

B. General LS Linear Fit

The Linear Fit VI calculates the coefficients a0 and a1 that best fits the
experimental data (x[i] and y[i]) to a straight line model given by
y[i] = a0 + a1*x[i]
Here, y[i] is a linear combination of the coefficients a0 and a1. You can
extend this concept further so that the multiplier for a1 is some function
of x. For example:
y[i] = a0 + a1*sin(x[i]) or
y[i] = a0 + a1*x[i]2 or
y[i] = a0 + a1*cos(x[i]2)
where is the angular frequency. In each of these cases, y[i] is a linear
combination of the coefficients a0 and a1. This is the basic idea behind
the General LS Linear Fit VI, where the y[i] can be linear
combinations of several coefficients, each of which may be multiplied
by some function of the x[i]. Therefore, you can use it to calculate
coefficients of the functional models that can be represented as linear
combinations of the coefficients, such as
y = a0 + a1*sin(x) or
y = a0 + a1*x2 + a2*cos(x2)
y = a0 + a1*(3sin(x)) + a2*x3 + a3 / x + ...
In each case, note that y is a linear function of the coefficients (although
it may be a nonlinear function of x).
You will now see how to use the General LS Linear Fit VI to find the
best linear fit to a set of data points. The inputs and outputs of the
General LS Linear Fit VI are shown below.

The data that you collect (x[i] and y[i]) is to be given to the inputs H and
Y Values. The Covariance output is the matrix of covariances between
the coefficients ak, where cij is the covariance between ai and aj, and ckk
is the variance of ak. At this stage, you need not be concerned about the
inputs Standard Deviation, covariance selector, and algorithm. For

LabVIEW Signal Processing Course Manual

7-12

National Instruments Corporation

Lesson 7

Curve Fitting

now, you will just use their default values. You can refer to the LabVIEW
Analysis VI Reference Manual for more details on these inputs.
The matrix H is known as the observation matrix and will be explained
in more detail later. Y Values is the set of observed data points y[i]. For
example, suppose you have collected samples (Y Values) from a
transducer and you want to solve for the coefficients of the model:
y = a o + a 1 sin ( x ) + a 2 cos ( x ) + a 3 x

You see that the multiplier for each aj (0 j 3) is a different function.

For example, a0 is multiplied by 1, a1 is multiplied by sin(x), a2 is
multiplied by cos(x), and so on. To build H, you set each column of H
to the independent functions evaluated at each x value, x[i]. Assuming
there are 100 x values, H would be:

H =

1 sin ( x 0 ) cos ( x 0 ) x 0

1 sin ( x 1 ) cos ( x 1 ) x 1

1 sin ( x 2 ) cos ( x 2 ) x 2

1 sin ( x 99 ) cos ( x 99 ) x 99

If you have N data points and k coefficients (a0, a1, ....ak-1) for which to
solve, H will be an N-by-k matrix with N rows and k columns. Thus, the
number of rows of H is equal to the number of elements in Y Values,
whereas the number of columns of H is equal to the number of
coefficients for which you are trying to solve.
In practice, H is not available and must be built. Given that you have the
N independent X Values and observed Y Values, the following block
diagram demonstrates how to build H and use the General LS Linear
Fit VI.

National Instruments Corporation

7-13

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

Exercise 7-4
OBJECTIVE: To learn how to set up the input parameters and use the General LS Linear Fit VI.
This exercise demonstrates how to use the General LS Linear Fit VI to
obtain the set of least square coefficients a and the fitted values, and also
how to set up the input parameters to the VI.
The purpose is to find the set of least square coefficients a that best
represent the set of data points (x[i], y[i]). As an example, suppose you
have a physical process that generates data using the relationship
y = 2h 0 ( x ) + 3h 1 ( x ) + 4h 2 ( x ) + noise

(1)

where
2
h 0 ( x ) = sin ( x ),
h 1 ( x ) = cos ( x ),
1
h 2 ( x ) = ------------ , and
x+1

noise is a random value. Also, assume you have some idea of the general
form of the relationship between x and y, but are not quite sure of the
coefficient values. So, you may think that the relationship between x and
y is of the form
y = a0 f0 ( x ) + a1 f1 ( x ) + a2 f2 ( x ) + a3 f3 ( x ) + a4 f4 ( x )

(2)

where
f o ( x ) = 1.0,
2
f 1 ( x ) = sin ( x ),
f 2 ( x ) = 3 cos ( x ),
1
f 3 ( x ) = ------------ ,
x+1
f4 ( x ) = x

Equations (1) and (2) respectively correspond to the actual physical

process and to your guess of this process. The coefficients you choose in
your guess may be close to the actual values, or may be far away from
them. Your objective now is to accurately determine the coefficients a.

Building the Observation Matrix

To obtain the coefficients a, you must supply the set of (x[i], y[i]) points in
the arrays H and Y Values (where the matrix H is a 2D array) to the General
LS Linear Fit VI. The x[i] and y[i] points are the values observed in your

LabVIEW Signal Processing Course Manual

7-14

National Instruments Corporation

Lesson 7

Curve Fitting

experiment. A simple way to build the matrix H is to use the Formula Node
as shown in the following block diagram.

You can easily edit the formula node to change, add, or delete functions.
At this point, you have all the necessary inputs to use the General LS
Linear Fit VI to solve for a. To obtain equation (1) from equation (2),
you need to multiply f0(x) by 0.0, f1(x) by 2.0, f2(x) by 1.0, f3(x) by 4.0
and f4(x) by 0.0. Thus, looking at equations (1) and (2), note that the
expected set of coefficients are
a = { 0.0, 2.0, 1.0, 4.0, 0.0 }

The block diagram below demonstrates how to set up the General LS

Linear Fit VI to obtain the coefficients and a new set of y values.

The subVI labeled Data Create generates the X and Y arrays.

You can replace this icon with one that actually collects the data
in your experiments. The icon labeled H(X,i) generates the 2D
matrix H.
The last portion of the block diagram overlays the original and the
estimated data points and produces a visual record of the General LS

National Instruments Corporation

7-15

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

Linear Fit. Executing the General LS Linear Fit VI with the values of
X, Y, and H returns the following set of coefficients.

The resulting equation is thus

y = 0.0298(1) + 2.1670sin(x2) + 1.0301(3cos(x)) + 3.9226/(x+1) +
0.00(x4)
= 0.0298 + 2.1670sin(x2) + 1.0301(3cos(x)) + 3.9226/(x+1)
The following graph displays the results.

You will now see the VI in which this particular example has been
implemented.
1. Open the General LS Fit Example VI from the library
Lvspcex.llb.
2. Examine the block diagram and the front panel.
noise amplitude: can change the amplitude of the noise added to the
data points. The larger this value, the more the spread of the data
points.
NumData: the number of data points that you want to generate.
algorithm: provides a choice of six different algorithms to obtain the
set of coefficients and the fitted values. In this particular example,
there is no significant difference among different algorithms. You
can select different algorithms from the front panel to see the results.
In some cases, different algorithms may have significant differences,
depending on your observed data set.

LabVIEW Signal Processing Course Manual

7-16

National Instruments Corporation

Lesson 7

Curve Fitting

MSE: gives the mean squared error. The smaller the MSE, the better
the fit.
error: gives the error code in case of any errors. If error code = 0, it
indicates no error. For a list of error codes, see the appendix.
Coefficients: the calculated values of the coefficients (a0, a1, a2, a3,
and a4) of the model.
3. Run the VI with progressively larger values of the noise amplitude.
What happens to the observed data plotted on the graph? What about
the MSE?
4. For a fixed value of noise amplitude, run the VI by choosing
different algorithms from the algorithm control. Do you find that any
one algorithm is better than the other? Which one gives you the
lowest MSE?
5. When you finish, close the VI. Do not save any changes.

End of Exercise 7-4

National Instruments Corporation

7-17

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

Exercise 7-5
OBJECTIVE: To predict production costs using the General LS Linear Fit VI.
The VIs that you have seen so far have been used to fit a curve to a function
of only one variable. In this exercise, you will use the General LS Linear
Fit VI to fit a curve to a multivariable function. In particular, the function
will have two variables, X1 and X2. You can, however, generalize it to
functions of three or more variables.
Suppose you are the manager of a bakery and want to estimate the total
cost (in dollars) of a production of baked scones using the quantity
produced, X1, and the price of one pound of flour, X2. To keep things
simple, the following five data points form this sample data table.
Cost (dollars)

Quantity

Flour Price

X1

X2

$150

295

3.00

$75

100

3.20

$120

200

3.10

$300

700

2.80

$50

60

2.50

You want to estimate the coefficients to the equation:

Y=a0 + a1X1 + a2X2.
You can use the General LS Linear Fit VI. It must have the inputs H,
the observation matrix, and Y Values, a vector of values of the LHS of the
above equation. Each column of H is the observed data for each of the
independent variables associated with the coefficients a0, a1, and a2.
Note that the first column is one because the coefficient a0 is not
associated with any independent variable. Thus, H should be filled in as:
1
1
H = 1
1
1

295
100
200
700
60

3.00
3.20
3.10
2.80
2.50

In LabVIEW (or BridgeVIEW), the observed data would normally

appear in three arrays (Y, X1, and X2). The following block diagram
demonstrates how to build H.

LabVIEW Signal Processing Course Manual

7-18

National Instruments Corporation

Lesson 7

Curve Fitting

Block Diagram

1. Open the Predicting Cost VI from the library Lvspcex.llb. The

values of Y, X1, and X2 have already been entered into the
corresponding controls on the front panel.
2. Examine the block diagram. Be sure you understand how to build the
matrix H. Most of the rest of the block diagram is used to build the
string that displays the equation of the functional model. You do not
need to worry about the rest of the diagram for now.
3. Switch to the front panel and run the VI. Check to see whether the
matrix H was created correctly.

National Instruments Corporation

7-19

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

Front Panel

After running the Predicting Cost VI, the following coefficients are
obtained.

and the resulting equation for the total cost of scone production is
therefore:
Y=-20.34 + 0.38X1 + 19.05X2
4. Experiment with different values of X2 (flour price).
5. When you finish, close the VI. Do not save any changes.

End of Exercise 7-5

LabVIEW Signal Processing Course Manual

7-20

National Instruments Corporation

Lesson 7

Curve Fitting

C. Nonlinear Lev-Mar Fit

So far, you have seen VIs that are used when there is a linear relationship
between y and the coefficients a0, a1, a2, .... However, when a nonlinear
relationship exists, you can use the Nonlinear Lev-Mar Fit VI to determine
the coefficients. This VI uses the Levenberg-Marquardt method, which is
very robust, to find the coefficients A = {a0, a1, a2, ..., ak} of the nonlinear
relationship between A and y[i]. The VI assumes that you have prior
knowledge of the nonlinear relationship between the x and y coordinates.
As a preliminary step, you need to specify the nonlinear function
in the Formula Node on the block diagram of one of the subVIs
of the Nonlinear Lev-Mar Fit VI. This particular subVI is the
Target Fnc and Deriv NonLin VI. You can access the Target
Fnc and Deriv NonLin VI by selecting it from the menu that
appears when you select Project This VIs SubVIs.
Note:

When using the Nonlinear Lev-Mar Fit VI, you also need to specify the
nonlinear function in the Formula Node on the block diagram of the
Target Fnc and Deriv NonLin VI.
The connections to the Nonlinear Lev-Mar Fit VI are shown below:

X and Y are the input data points x[i] and y[i].

Initial Guess Coefficients is your initial guess as to what the
coefficient values are. The coefficients are those used in the formula
that you entered in the Formula Node of the Target Fnc and Deriv
NonLin VI. Using the Nonlinear Lev-Mar Fit VI successfully
sometimes depends on how close your initial guess coefficients are
to the actual solution. Therefore, it is always worth taking the time
and effort to obtain a good initial guess to the solution from any
available resource.
For now, you can leave the other inputs to their default values. For
more information on these inputs, see the LabVIEW (or
BridgeVIEW) Analysis VI Reference Manual.
Best Fit Coefficients: the values of the coefficients (a0, a1, ...) that
best fit the model of the experimental data.

National Instruments Corporation

7-21

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

Exercise 7-6
OBJECTIVE: To create a general exponential signal a*exp(b*x) + c + noise. Then, use the
Nonlinear Lev-Mar Fit VI to fit the data and get the best guess coefficients a, b,
and c of the general exponential signal.
In this exercise, you will see how to use the Nonlinear Lev-Mar Fit VI to
determine the coefficients a, b, and c, of a nonlinear function given by
a*exp(b*x) + c.
1. Open the Nonlinear Lev-Mar Exponential Fit VI from the library
Lvspcex.llb. The front panel is shown below.

Front Panel

The a, b, and c controls determine the actual values of the

coefficients a, b and c. The Initial Coefficients control is your
educated guess as to the actual values of a, b and c. Finally, the Best
Guess Coef indicator gives you the values of a, b and c calculated by
the Nonlinear Lev-Mar Fit VI. To simulate a more practical
example, you also add noise to this equation, thus making it of the
form:
a*exp(b*x) + c + noise
The noise level control adjusts the noise level. Note that the actual
values of a, b, and c being chosen are +1.0, -0.1 and 2.0. In the Initial
Coefficients control, the default guess for these is a = 2.0, b = 0, and
c = 4.0.

LabVIEW Signal Processing Course Manual

7-22

National Instruments Corporation

Lesson 7

Curve Fitting

2. Examine the block diagram.

Block Diagram

The data samples of the exponential function are simulated

using the Exponential VI (Numeric Logarithmic subpalette)
and uniform white noise is added to the samples with the help of
the Uniform White Noise VI (Analysis Signal Generation
subpalette).
3. From the Project menu, select Unopened SubVIs Target Fnc and
Deriv NonLin VI. The front panel of the Target Fnc and Deriv
NonLin VI opens, as shown below.

National Instruments Corporation

7-23

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

4. Switch to the block diagram.

Observe the formula node at the bottom. It has the form of the
function whose parameters (a0, a1, and a2) you are trying to
evaluate.
5. Close the front panel and the block diagram of the Target Fnc and
Deriv NonLin VI.
6. Run the NonLinear Lev-Mar Exponential Fit VI. Note that the
values of the coefficients returned in Best Guess Coef are very close
to the actual values entered in the Initial Coefficients control. Also
note the value of the mse.
7. Increase the noise level from 0.1 to 0.5. What happens to the mse and
the coefficient values in Best Guess Coef? Why?
8. Change the noise level back to 0.1 and the Initial Coefficients to 5.0,
-2.0, and 10.0, and run the VI. Note the values returned in the Best
Guess Coef and the mse indicators.
9. With the noise level still at 0.1, change your guess of the Initial
Coefficients to 5.0, 8.0, and 10.0, and run the VI. This time, your
guess is further away than the one you chose in step 4. Note the error!
This goes to show how important it is to have a reasonably educated
guess for the coefficients.
10. When you finish, close the VI. Do not save any changes.

End of Exercise 7-6

LabVIEW Signal Processing Course Manual

7-24

National Instruments Corporation

Lesson 7

Curve Fitting

D.Fitting a Curve to Gaussian (Normal) Data Points

Real-world data is very often Gaussian distributed. That means that many of
the data points lie close to a particular value, known as the mean, and the
number of data points is smaller as you get further away from the mean. The
mathematical description of a Gaussian (also known as a Normal)
distribution is:
1 x 2
1
f ( x ) = -------------- exp --- ------------
2
2

(3)

where is the mean and is the standard deviation. The following

figure shows the Gaussian distribution with = 0 and = 0.5, 1.0,
and 2.0.

As seen from the figure, the curve is bell shaped and is symmetric about the
mean, . The peak of the curve occurs at . The standard deviation, ,
determines the spread of the curve around the mean. The smaller the value
of , the more concentrated the curve around the mean, the higher the peak
at the mean, and the steeper the descents on both sides.
If you have data that is Normally distributed, you will find that the
standard deviation is an important parameter in determining the limits
within which a certain percentage of your data values are expected to
occur. For example,
1. About two-thirds of the values will lie between - and +.
2. About 95% of the values will lie between -2 and +2.

National Instruments Corporation

7-25

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

3. About 99.75% of the values will lie between -3 and +3. Thus,
you see that almost all the data values lie between -3 and +3. This
is illustrated in the figures below:
68%

16%

16%

+
95.5%

2.25%

2.25%

+ 2

99.76%

0.12%

0.12%

+ 3

Notice that the two parameters that completely describe Gaussian data
are the mean and the standard deviation of the data. If you believe your
data has a Gaussian distribution, you could determine its mean and
standard deviation. This has numerous applications, such as determining
whether:
The dimensions of products being manufactured (for example,
thickness of plates) are to within specified limits.
The values of components (for example, resistance of resistors) are
to within a specified tolerance.
LabVIEW Signal Processing Course Manual

7-26

National Instruments Corporation

Lesson 7

Curve Fitting

Exercise 7-7
Objective:

To fit a curve to noisy Gaussian data.

1. Open the Normal (Gaussian) Fit VI from the library
Lvspcex.llb.

National Instruments Corporation

7-27

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

2. Switch to the block diagram.

Block Diagram

Observe that the parameters a and b in the Nonlinear Lev-Mar

Exponential Fit VI of the previous exercise have been replaced by mean
and sigma. Note that the controls for the parameter c and the number of
points have been removed. The For Loop generates the range of the data to
lie between -5.0 and +5.0. Most of the additions on the block diagram have
to do with implementing equation (3 ) on page 7-24. Uniform white noise is
then added to the Gaussian data generated.
3. Select Project Unopened SubVIs Target Fnc & Deriv
Nonlin VI.

LabVIEW Signal Processing Course Manual

7-28

National Instruments Corporation

Lesson 7

Curve Fitting

4. Modify the block diagram of the Target Fnc & Deriv Nonlin VI as
shown below.

Block Diagram

5. Go back to the Normal (Gaussian) Fit VI and enter the following

values in the controls on the front panel:
sigma

1.0

mean

0.0

noise level

0.1

6. Clear the Initial Coefficients array by popping up on it with the right

mouse button and selecting Data Operations Empty Array. Set
the values in Initial Coefficients to 2.0 and 2.0. These are your
guesses of sigma and mean, the actual values of which you chose as
1.0 and 0.0 in the corresponding controls.
7. Run the VI several times and observe the values of sigma and mean
in the Best Guess Coeff indicator. In the plot, you can also see the
spread of the data and the fitted Gaussian curve.
8. Change the value of sigma to 1.0 and run the VI several times. Each
time observe the fitted curve, the mse, and the values of the Best
Guess Coeff.
9. Change Initial Coefficients [0] (the guess for sigma) to 50.0 and run
the VI several times. Does the VI obtain a good estimate of sigma in
the array Best Guess Coeff?
10. Change Initial Coefficients [0] (the guess for sigma) to 500.0 and
again run the VI several times. Now how accurate is the estimate?

National Instruments Corporation

7-29

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

11. When you finish, save and close the VI.

In this exercise and the previous one, you had to open the Target Fnc
and Deriv NonLin VI and enter equations in the formula node. This
requires a certain amount of work, as well as an understanding of the
formula node, on your part. A much simpler and easier method is to have
the flexibility of entering the formulas directly on the front panel,
without accessing the block diagram. As you will see in a later lesson,
this is possible by using the G Math Toolkit, which is a mathematics
package (written entirely in G) that is extremely useful for solving
differential equations, optimization, and a wide variety of other common
math problems.

End of Exercise 7-7

LabVIEW Signal Processing Course Manual

7-30

National Instruments Corporation

Lesson 7

Curve Fitting

Summary

Curve fitting is useful in fitting an equation to a set of data points.

Some of the practical applications of curve fitting are to remove
measurement noise, to fill in missing data points, interpolation,
extrapolation, and integration and differentiation of digital data.

In particular, you have learned how to use the Linear, Exponential,

Polynomial, General Linear LS Fit, and Nonlinear Lev-Mar Fit
VIs to perform several different types of linear and nonlinear fits.

The MSE is a useful criterion in determining the fit accuracy.

Review Questions
1. Name five applications of curve fitting.
2. Which curve fitting VI would you use to determine the parameters
(denoted by the ai...i integer) of the following models?
a. y = a0exp(a1x)
b. y = a0exp(a1x) + a2
c. y = a1x + a2x2
a
x+1

3
d. y = a1sin(x) + a2cos(x) = -----------

e. t = a1y + a2x2 where x and y are different variables

National Instruments Corporation

7-31

LabVIEW Signal Processing Course Manual

Lesson 7 Curve Fitting

Notes

LabVIEW Signal Processing Course Manual

7-32

National Instruments Corporation

Lesson 8
Linear Algebra
Introduction
In this lesson, you will learn the basic theory behind the Linear Algebra VIs
in the Analysis library and how these VIs can be used in different
applications. Matrix computations, such as matrix-matrix multiplication
and many others discussed throughout this lesson, form a significant
component of linear algebra and are very important in analysis. The VIs
discussed in this lesson form the basis of different algorithms used in many
DSP, control, and measurement applications. Therefore, it is important that
you completely understand the theory and different VIs discussed in this
lesson.

You Will Learn:

A. About linear systems and matrix analysis.
B. About basic matrix operations and eigenvalue-eigenvector
problems.
C. About the inverse of a matrix and solving systems of linear
equations.
D. About matrix factorization.

National Instruments Corporation

8-1

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

A. Linear Systems and Matrix Analysis

Systems of linear algebraic equations arise in many applications that
involve scientific computations such as signal processing, computational
fluid dynamics, and others. Such systems may occur naturally or may be the
result of approximating differential equations by algebraic equations.

Types of Matrices
Whatever the application, it is always necessary to find an accurate solution
for the system of equations in a very efficient way. In matrix-vector
notation, such a system of linear algebraic equations has the form Ax = b,
where A is an n n matrix, b is a given vector consisting of n elements, and
x is the unknown solution vector to be determined. A matrix is represented
by a 2D array of elements. These elements may be real numbers, complex
numbers, functions, or operators. The matrix A shown below is an array of
m rows and n columns with m n elements.

A =

a 0, 0

a 0, 1

a 0, n 1

a 1, 0

a 1, 1

a 1, n 1

a m 1, 0 a m 1, 1 a m 1, n 1

Here, ai,j denotes the (i,j)th element located in the ith row and the jth
column. In general, such a matrix is called a rectangular matrix. When
m = n , so that the number of rows is equal to the number of columns, it
is called a square matrix. An m 1 matrix (m rows and one column) is
called a column vector. A row vector is a 1 n matrix (1 row and n
columns). If all the elements other than the diagonal elements are zero
(that is, ai,j = 0, i j ), such a matrix is called a diagonal matrix. For
example,
4 0 0
A = 0 5 0
0 0 9

is a diagonal matrix. A diagonal matrix with all the diagonal elements

equal to one is called an identity matrix, also known as unit matrix. If all
the elements below the main diagonal are zero, then the matrix is known
as an upper triangular matrix. On the other hand, if all the elements
above the main diagonal are zero, then the matrix is known as a lower
triangular matrix. When all the elements are real numbers, the matrix is
referred to as a real matrix. On the other hand, when at least one of the
elements of the matrix is a complex number, the matrix is referred to as
a complex matrix. To make things simpler to understand, you will work
mainly with real matrices in this lesson. However, for the adventurous,
there are also some exercises involving complex matrices.

LabVIEW Signal Processing Course Manual

8-2

National Instruments Corporation

Lesson 8

Linear Algebra

Determinant of a Matrix
One of the most important attributes of a matrix is its determinant. In
the simplest case, the determinant of a 2 x 2 matrix A = a b is given
c d

by ad bc . The determinant of a square matrix is formed by taking the

determinant of its elements. For example, if
2 5 3
A = 6 1 7
1 6 9

the determinant of A, denoted by A , is

A

= 253

= 2 1 7 5 6 7 + 3 6 1
6 1 7
1 9
1 6
69
1 6 9

= 2 ( 33 ) 5 ( 47 ) + 3 ( 35 )

=-196
The determinant tells many important properties of the matrix. For
example, if the determinant of the matrix is zero, then the matrix is
singular. In other words, the above matrix (with nonzero determinant) is
nonsingular. You will revisit the concept of singularity later in section
C, when the lesson discusses the solution of linear equations and matrix
inverses.

Transpose of a Matrix
The transpose of a real matrix is formed by interchanging its rows and
columns. If the matrix B represents the transpose of A, denoted by AT, then
bj,i=ai,j . For the matrix A defined above,
2 6 1
T
B = A = 5 1 6
3 7 9

In case of complex matrices, complex conjugate transposition is defined.

If the matrix D represents the complex conjugate transpose1 of a
complex matrix C, then
D = C d i, j = c j, i
H

That is, the matrix D is obtained by replacing every element in C by its

complex conjugate and then interchanging the rows and columns of the
resulting matrix.

1. Complex Conjugate: if a = x + iy, complex conjugate a* = x - iy

National Instruments Corporation

8-3

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

A real matrix is called a symmetric matrix if the transpose of the matrix

is equal to the matrix itself. The example matrix A is not a symmetric
matrix. If a complex matrix C satisfies the relation C = CH, then C is
called a Hermitian matrix.

Obtaining One Vector as a Linear Combination of Other Vectors

(Linear Independence)
A set of vectors x1, x2, ...., xn is said to be linearly dependent if and only if
there exist scalars 1, 2, ..., n, not all zero, such that
1 x1 + 2 x2 + + n xn = 0

In simpler terms, if one of the vectors can be written in terms of a linear

combination of the others, then the vectors are said to be linearly
dependent.
If the only set of i for which the above equation holds is 1 = 0 ,
2 = 0 , ..., n = 0 , the set of vectors x1, x2, ...., xn is said to be linearly
independent. So, in this case, none of the vectors can be written in terms
of a linear combination of the others. Given any set of vectors, the above
equation always holds for 1 = 0 , 2 = 0 , ..., n = 0 . Therefore, to
show the linear independence of the set, you must show that 1 = 0 ,
2 = 0 , ..., n = 0 is the only set of i for which the above equation
holds.
For example, first consider the vectors
x = 1
2

y = 3
4

Notice that 1 = 0 and 2 = 0 are the only values for which the relation
1 x + 2 y = 0 holds true. Hence, these two vectors are linearly
independent of each other. Now, consider vectors
x = 1
2

y = 2
4

Notice that, if 1 = 2 and 2 = 1 , then 1 x + 2 y = 0 . Therefore, these

two vectors are linearly dependent on each other. You must completely
understand this definition of linear independence of vectors to fully
appreciate the concept of the rank of the matrix as discussed next.

How Can You Determine Linear Independence? (Matrix Rank)

The rank of a matrix A, denoted by (A), is the maximum number of linearly
independent columns in A. If you look at the example matrix A, you will find
that all the columns of A are linearly independent of each other. That is,
none of the columns can be obtained by forming a linear combination of the

LabVIEW Signal Processing Course Manual

8-4

National Instruments Corporation

Lesson 8

Linear Algebra

other columns. Hence, the rank of the matrix is 3. Consider one more
example matrix, B, where
0 1 1
B = 1 2 3
2 0 2

This matrix has only two linearly independent columns, because the
third column of B is linearly dependent on the first two columns. Hence,
the rank of this matrix is 2. It can be shown that the number of linearly
independent columns of a matrix is equal to the number of independent
rows. So, the rank can never be greater than the smaller dimension of the
matrix. Consequently, if A is an n m matrix, then
( A ) min ( n, m )

where min denotes the minimum of the two numbers. In matrix theory,
the rank of a square matrix pertains to the highest order nonsingular
matrix that can be formed from it. Remember from the earlier discussion
that a matrix is singular if its determinant is zero. So, the rank pertains
to the highest order matrix that you can obtain whose determinant is not
zero. For example, consider a 4 x 4 matrix
1
B = 0
1
1

2
1
0
1

3
1
1
0

4
0
2
2

For this matrix, det ( B ) = 0 , but

12 3
0 1 1
10 1

= 1

Hence, the rank of B is 3. A square matrix has full rank if and only if its
determinant is different from zero. Matrix B is not a full-rank matrix.

Magnitude (Norms) of Matrices

You must develop a notion of the magnitude of vectors and matrices to
measure errors and sensitivity in solving a linear system of equations. As an
example, these linear systems can be obtained from applications in control
systems and computational fluid dynamics. In two dimensions, for example,
you cannot compare two vectors x = x1 x2 and y = y1 y2 , because you
might have x1 > y1 but x2 < y2 . A vector norm is a way to assign a scalar
quantity to these vectors so that they can be compared with each other. It is
similar to the concept of magnitude, modulus, or absolute value for scalar
numbers.

National Instruments Corporation

8-5

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

There are several ways to compute the norm of a matrix. These include
the 2-norm (Euclidean norm), the 1-norm, the Frobenius norm (F-norm),
and the Infinity norm (inf-norm). Each norm has its own physical
interpretation. Consider a unit ball containing the origin. The Euclidean
norm of a vector is simply the factor by which the ball must be expanded
or shrunk to encompass the given vector exactly. This is shown in the
figures below:

22

22

1
2

Figure 1a

Figure 1b

Figure 1c

Figure 1a shows a unit ball of radius = 1 unit. Figure 1b shows a vector

of length 2 2 + 2 2 = 8 = 2 2 . As shown in figure 1c, the unit ball must
be expanded by a factor of 2 2 before it can exactly encompass the
given vector. Hence, the Euclidean norm of the vector is 2 2 .
The norm of a matrix is defined in terms of an underlying vector norm.
It is the maximum relative stretching that the matrix does to any vector.
With the vector 2-norm, the unit ball expands by a factor equal to the
norm. On the other hand, with the matrix 2-norm, the unit ball may
become an ellipsoidal (ellipse in 3D), with some axes longer than others.
The longest axis determines the norm of the matrix.

LabVIEW Signal Processing Course Manual

8-6

National Instruments Corporation

Lesson 8

Linear Algebra

Some matrix norms are much easier to compute than others. The 1-norm
is obtained by finding the sum of the absolute value of all the elements
in each column of the matrix. The largest of these sums is called the
1-norm. In mathematical terms, the 1-norm is simply the maximum
absolute column sum of the matrix.
n1

= max j

i, j

i=0

For example, A = 1 3
2 4

then A 1 = max ( 3, 7 ) = 7 . The inf-norm of a matrix is the maximum

absolute row sum of the matrix
n1

= max i

i, j

j=0

In this case, you add the magnitudes of all elements in each row of the
matrix. The maximum value that you get is called the inf-norm. For the
above example matrix, A = max ( 4, 6 ) = 6 .
The 2-norm is the most difficult to compute because it is given by the
largest singular value of the matrix. Singular values are discussed in
Section D, and Exercise 8-9 verifies the validity of the above statement.

Determining Singularity (Condition Number)

Whereas the norm of the matrix provides a way to measure the magnitude
of the matrix, the condition number of a matrix is a measure of how close
the matrix is to being singular. The condition number of a square
nonsingular matrix is defined as
cond ( A ) = A

1
p

where p can be one of the four norm types discussed above. For example,
to find the condition number of a matrix A, you can find the 2-norm of
A, the 2-norm of the inverse1 of the matrix A, denoted by A-1, and then
multiply them together. As mentioned earlier, the 2-norm is difficult to
calculate on paper. You can use the Matrix Norm VI from the Analysis
library to compute the 2-norm. For example,
1
A = 1 2 , A = 2 1 , A
3 4
1.5 0.5

= 5.4650, A

1
2

= 2.7325, cond ( A ) = 14.9331

1. The inverse of a square matrix A is a square matrix B such that AB=I, where I is the identity matrix.
Matrix inverses and their applications are described in more detail later in section C.

National Instruments Corporation

8-7

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

The condition number can vary between 1 and infinity. A matrix with a
large condition number is nearly singular, while a matrix with a
condition number close to 1 is far from being singular. The matrix A
above is nonsingular. However, consider the matrix B =

1 0.99
1.99 2

The condition number of this matrix is 47168, and hence the matrix is
close to being singular. As you might recall, a matrix is singular if its
determinant is equal to zero. However, the determinant is not a good
indicator for assessing how close a matrix is to being singular. For the
matrix B above, the determinant (0.0299) is nonzero; however, the large
condition number indicates that the matrix is close to being singular.
Remember that the condition number of a matrix is always greater than
or equal to one; the latter being true for identity and permutation
matrices1. The condition number is a very useful quantity in assessing
the accuracy of solutions to linear systems.
In this section, you have become familiar with some basic notation and
fundamental matrix concepts such as determinant of a matrix and its
rank. The following exercise should help you further understand these
terms, which will be used frequently throughout the rest of the lesson.

1. A permutation matrix is an identity matrix with some rows and columns exchanged.

LabVIEW Signal Processing Course Manual

8-8

National Instruments Corporation

Lesson 8

Linear Algebra

Exercise 8-1
Objective:

To construct matrices, compute condition number, rank, and determinant, and

observe how the condition number affects the accuracy of the solution.
In this exercise, you will complete a VI that constructs four different
matrices of size 9 9 . You will determine the condition number, rank, and
determinant of these matrices. You will then solve a curve fitting problem
and observe how the condition number of these matrices affects the
accuracy of the final solution.
1. Open the Construct Matrices VI from Lvspcex.llb.

2. The array population (nine elements) contains the population data

for the United States for the years 1900 to 1980, at intervals of 10
years. This data has been plotted for you in the graph labeled
population graph. To find the population in any one of the
intermediate years, you need to interpolate between these nine data
points. This interpolation can be achieved by first fitting a curve to
these data points and then using the curve to obtain the intermediate
values. This curve is represented by a unique polynomial of degree
eight that interpolates these nine data points, but that polynomial can
be represented in many different ways. Consider the following four
ways to represent the individual terms in the polynomial:
(i) b j ( t ) = t j
(ii) b j ( t ) = ( t 1900 )j
(iii) b j ( t ) = ( t 1940 )j
(iv) b j ( t ) = ( ( t 1940 ) 40 )j

National Instruments Corporation

8-9

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

For example, the actual polynomial in case (i) is given by

y = a0 + a1t + a2t2 + ...
where the ai are the curve parameters to be determined.
To determine these parameters, it is necessary to solve a linear system
of equations va = y, where v is a matrix, a = [a0, a1, ...], and y is the
population vector.
For each of these four representations, you can generate the matrix V,
where the (i,j)th element of this matrix is given by
v i, j = b j ( t [ i ] )

Such a matrix V is referred to as the Vandermonde matrix. For example,

if you are using the polynomial representation (i) above, the
Vandermonde matrix will look like

V =

n1

n1

t0

t0 t0

t1

t1 t1

2
tn 1 tn 1

n1

tn 1

where ti (0 i n-1) is the ith element of the vector year.

LabVIEW Signal Processing Course Manual

8-10

National Instruments Corporation

Lesson 8

Linear Algebra

Block Diagram

3. Open the block diagram for this VI and complete it as shown above.
Matrix Condition Number VI (Analysis Linear Algebra
Advanced Linear Algebra subpalette). In this exercise, this
function computes the matrix condition number.
Determinant VI (Analysis Linear Algebra subpalette). In this
exercise, this function computes the matrix determinant.
Matrix Rank VI (Analysis Linear Algebra Advanced
Linear Algebra subpalette). In this exercise, this function
computes the matrix rank.
Bundle function (Cluster subpalette) In this exercise, this
function assembles the population array and the year array to
plot the population array against the year array.

National Instruments Corporation

8-11

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

4. The subdiagrams in Case t^j, (t-1900)^j, (t-1940)^j, and

((t-1940)/40)^j construct the matrices corresponding to the
polynomial functions (i), (ii), (iii), and (iv), respectively. The
subdiagrams in the first three cases, as shown above, are already

LabVIEW Signal Processing Course Manual

8-12

National Instruments Corporation

Lesson 8

Linear Algebra

built for you. You need only complete the subdiagram in Case
((t-1940)/40)^j.

5. The Matrix Condition Number VI has an input called norm type.

Earlier in the section, you looked at different types of norms. This VI
can compute the condition number using four different norm types.
In this exercise, you will set norm type = 0, which is the 2-norm.

6. The Determinant VI has an input called matrix type. In this

exercise, you will set matrix type = 0 (general matrix).

7. The Matrix Rank VI has an input called tolerance. Leave this

terminal unconnected, using the default value for this exercise.
8. Return to the front panel and run the VI. Choose different
polynomial functions using the polynomial function selector control.
9. Look at the condition numbers of the four matrices. Which is very
close to being singular?
10. Save and close this VI.

National Instruments Corporation

8-13

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

11. You will now use the matrices (computed in the Construct Matrices
VI) to calculate the population for any year between 1900 and 1980.
To do so, open the Compute Population VI from Lvspcex.llb.
This VI computes a polynomial to interpolate the data values to the
population data. It then computes the population for a specified year
using the Horners nested evaluation scheme1.
12. Return to the front panel for the VI opened in step 10. Set the
choose year control to 1950. Using each of the four polynomial
functions, run the VI to compute the population for this year. The red
dot on the population graph shows the population value for the year
chosen. Which of these values is closest to the true value of
151,325,798, according to the 1950 census?
13. Save the VI and close it.

End of Exercise 8-1

1. For more information on this method, refer to Scientific Computing, An Introduction Survey by M.T.
Heath, McGraw-Hill, 1997.

LabVIEW Signal Processing Course Manual

8-14

National Instruments Corporation

Lesson 8

Linear Algebra

Exercise 8-2
Objective: To study special matrices.
In this exercise, you will learn to use the Create Special Matrix VI in
the Analysis library. Examine the different types of special matrices that
this VI creates. Note that this VI also generates the Vandermonde matrix
used in the previous exercise.

Front Panel

1. Build the front panel as shown above. You can resize the
two-dimensional matrices D, U, and L to see all the elements in the
matrix.

National Instruments Corporation

8-15

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

Block Diagram

2. Build the block diagram as shown above.

Create Special Matrix VI (Analysis Linear
Algebra Advanced Linear Algebra subpalette). In this
exercise, this function creates special matrices. You will
complete it in steps 4, 5, and 6.

3. Construct Diagonal Matrix D using the Create Special Matrix VI

from the Analysis Linear Algebra Advanced Linear Algebra
subpalette. The matrix size input determines the dimension size of
the output Special Matrix. Choose matrix size = 6.
4. The matrix type input determines the type of matrix that is generated
at the output Special Matrix. Choose matrix type = Diagonal.
5. Enter the diagonal elements 1, 2, 3, 4, 5, and 6 in the Diagonal
Elements control on the front panel. Connect this control to the Input
Vector1 terminal. The Input Vector1 terminal is the input to
construct a special matrix depending on the matrix type chosen.
6. Return to the front panel and run the VI.
7. Compute the determinant of this matrix as you did in the earlier
exercise. Do you find anything interesting about this determinant

LabVIEW Signal Processing Course Manual

8-16

National Instruments Corporation

Lesson 8

Linear Algebra

value? If yes, what is it? If no, take a close look at the diagonal
elements of the matrix D.
8. Select the Upper Lower Symmetric Matrix VI from
Lvspcex.llb. Set Select Matrix Type = 0 (upper triangular
matrix). Set matrix Size = 6. Wire the matrix D to the input matrix
terminal. The output matrix U is an upper triangular matrix.

9. Compute the determinant of this matrix. Do you find anything

interesting about this determinant value? If yes, what is it?
10. Choose matrix type = 1 (lower triangular matrix). Set matrix Size
equal to 6. Wire the matrix D to the input matrix terminal. The
output matrix L is a lower triangular matrix. Repeat step 6.
11. Save the VI as Special Matrix.vi and close it.
Note:

The determinant for all the three matrices is equal to the product of the
diagonal elements of the matrices.

End of Exercise 8-2

National Instruments Corporation

8-17

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

B. Basic Matrix Operations and Eigenvalues-Eigenvector

Problems
In this section, consider some very basic matrix operations. Two
matrices, A and B, are said to be equal if they have the same number of
rows and columns and their corresponding elements are all equal.
Multiplication of a matrix A by a scalar is equal to multiplication of
all its elements by the scalar. That is,
C = A c i, j = a i, j

For example,
2 1 2 = 2 4
3 4
6 8

Two (or more) matrices can be added or subtracted if and only if they
have the same number of rows and columns. If both matrices A and B
have m rows and n columns, then their sum C is an m-by-n matrix
defined as C = A B , where c i, j = a i, j bi, j . For example,
1 2 + 2 4 = 3 6
3 4
5 1
8 5

For multiplication of two matrices, the number of columns of the first

matrix must be equal to the number of rows of the second matrix. If
matrix A has m rows and n columns and matrix B has n rows and p
columns, then their product C is an m-by-p matrix defined as C = AB ,
where
n1

c i, j =

i , k b k, j

k=0

For example,
1 2 2 4 = 12 6
3 4
5 1
26 16

LabVIEW Signal Processing Course Manual

8-18

National Instruments Corporation

Lesson 8

Linear Algebra

So, you multiply the elements of the first row of A by the corresponding
elements of the first column of B and add all the results to get the
elements in the first row and first column of C. Similarly, to calculate
the element in the ith row and the jth column of C, multiply the elements
in the ith row of A by the corresponding elements in the jth column of C,
and then add them all. This is shown pictorially as:

R1

Cm

R1 C1

R1 Cm

Rn C1

Rn Cm

C1

Rn

Matrix multiplication, in general, is not commutative. That is, AB BA .

Also, remember that multiplication of a matrix by an identity matrix
results in the original matrix.

Dot Product and Outer Product

If X represents a vector and Y represents another vector, then the dot
product of these two vectors is obtained by multiplying the corresponding
elements of each vector and adding the results. This is denoted by
n1

XY =

x y

i i

i=0

where n is the number of elements in X and Y. Note that both vectors

must have the same number of elements. The dot product is a scalar
quantity, and has many practical applications.

National Instruments Corporation

8-19

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

For example, consider the vectors a = 2i + 4j and b = 2i + j in a

two-dimensional rectangular coordinate system.

a=2i+4j

= 36.86
b=2i+j

The dot product of these two vectors is given by

d = 2 2 = (2 2) + (4 1) = 8
4
1

The angle between these two vectors is given by

ab
8
= inv cos ------------ = inv cos ------ = 36.86 o ,
10
a b

where |a| denotes the magnitude of a.

Force a

Body

As a second application, consider a body on which a constant force a

acts. The work W done by a in displacing the body is defined as the
product of |d| and the component of a in the direction of displacement d.
That is,
W = a d cos = a d

On the other hand, the outer product of these two vectors is a matrix. The
(i,j)th element of this matrix is obtained using the formula
a i, j = x i y j

For example, 1 3 = 3 4
2

LabVIEW Signal Processing Course Manual

6 8

8-20

National Instruments Corporation

Lesson 8

Linear Algebra

Eigenvalues and Eigenvectors

To understand eigenvalues and eigenvectors, start with the classical
definition. Given an n n matrix A, the problem is to find a scalar and a
nonzero vector x such that
Ax = x

Such a scalar is called an eigenvalue, and x is a corresponding

eigenvector.
Calculating the eigenvalues and eigenvectors are fundamental principles
of linear algebra and allow you to solve many problems such as systems
of differential equations when you understand what they represent.
Consider an eigenvector x of a matrix A as a nonzero vector that does not
rotate when x is multiplied by A (except perhaps to point in precisely the
opposite direction). x may change length or reverse its direction, but it
will not turn sideways. In other words, there is some scalar constant
such that the above equation holds true. The value is an eigenvalue
of A.
Consider the following example. One of the eigenvectors of the
matrix A, where A = 2 3 , is x = 0.62 . Multiplying the matrix A and
3 5

1.00

the vector x simply causes the vector x to be expanded by a factor of

6.85. Hence, the value 6.85 is one of the eigenvalues of the matrix A. For
any constant , the vector x is also an eigenvector with eigenvalue ,
because
A ( x ) = Ax = x

In other words, an eigenvector of a matrix determines a direction in

which the matrix expands or shrinks any vector lying in that direction by
a scalar multiple, and the expansion or contraction factor is given by the
corresponding eigenvalue. A generalized eigenvalue problem is to find
a scalar and a nonzero vector x such that
Ax = Bx

where B is another n n matrix.

The following are some important properties of eigenvalues and
eigenvectors:

The eigenvalues of a matrix are not necessarily all distinct. In other

words, a matrix can have multiple eigenvalues.

All the eigenvalues of a real matrix need not be real. However,

complex eigenvalues of a real matrix must occur in complex
conjugate pairs.

National Instruments Corporation

8-21

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

The eigenvalues of a diagonal matrix are its diagonal entries, and the
eigenvectors are the corresponding columns of an identity matrix of
the same dimension.

A real symmetric matrix always has real eigenvalues and

eigenvectors.

As discussed earlier, eigenvectors can be scaled arbitrarily.

There are many practical applications in the field of science and

engineering for an eigenvalue problem. For example, the stability of a
structure and its natural modes and frequencies of vibration are
determined by the eigenvalues and eigenvectors of an appropriate
matrix. Eigenvalues are also very useful in analyzing numerical
methods, such as convergence analysis of iterative methods for solving
systems of algebraic equations, and the stability analysis of methods for
solving systems of differential equations.
The LabVIEW/BridgeVIEW EigenValues and Vectors VI is shown
below. The Input Matrix is an N-by-N real square matrix. Matrix type
determines the type of the input matrix. Matrix type could be 0,
indicating a general matrix1, or 1, indicating a symmetric matrix. A
symmetric matrix always has real eigenvalues and eigenvectors.

Output option determines what needs to be computed. Output

option = 0 indicates that only the eigenvalues need to be computed.
Output option = 1 indicates that both the eigenvalues and the
eigenvectors should be computed. It is computationally very expensive
to compute both the eigenvalues and the eigenvectors. So, it is important
that you use the output option control in the EigenValues and Vectors
VI very carefully. Depending on your particular application, you might
just want to compute the eigenvalues or both the eigenvalues and the
eigenvectors. Also, a symmetric matrix needs less computation than an
unsymmetric matrix. So, choose the matrix type control carefully.
In this section, you learned about some basic matrix operations and the
eigenvalues-eigenvectors problem. The next example introduces some
VIs in the Analysis library that perform these operations.

1. General Matrix: A matrix with no special property such as symmetry or triangular structure.

LabVIEW Signal Processing Course Manual

8-22

National Instruments Corporation

Lesson 8

Linear Algebra

Exercise 8-3
Objective: To learn basic matrix manipulations
You will build a VI that will help you further understand the basic matrix
manipulations discussed in the first part of the previous section. You will
also learn some very interesting matrix properties.

1. Build the front panel as shown above. You can resize the
two-dimensional matrices A, B, and C to see all the elements in the
matrix.
2. In this exercise, you will experiment with three different types of
matrices, namely upper triangular, lower triangular, and symmetric
matrix. You will use the Upper Lower Symmetric Matrix VI that
you used in the previous exercise.

National Instruments Corporation

8-23

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

3. Build the block diagram as shown above.

Create Special Matrix VI (Analysis Linear Algebra
Advanced Linear Algebra subpalette). In this exercise, this
function creates special matrices.
AxB VI (Analysis Linear Algebra subpalette) In this exercise,
this function multiplies two matrices).
4. Create a diagonal matrix using the Create Special Matrix VI. Select
the Upper Lower Symmetric Matrix VI from the library
Lvspcex.llb. This VI will create the appropriate type of matrix
depending on the Matrix Type control.
5. Select the A x B VI from the Analysis Linear Algebra subpalette.
You will use this VI to multiply matrix A and matrix B and the result
of this multiplication is stored in matrix C.
6. Return to the front panel. Choose Matrix A Size = 6 and Matrix B
Size = 6. Set the Diagonal Elements A and Diagonal Elements B
controls for both the matrices to some value (similar to what you did
in the previous exercise). The size of this array is equal to the matrix
size.
7. Choose Matrix Type = 0 (Upper Triangular Matrix). Run the VI.
Observe the structure of the upper triangular matrices A and B.
Matrix C is the product of these two matrices. Do you find anything
interesting about the structure of this matrix?
8. Now choose Matrix Type = 1 (Lower Triangular Matrix). Run the
VI. Observe the structure of the two lower triangular matrices A

LabVIEW Signal Processing Course Manual

8-24

National Instruments Corporation

Lesson 8

Linear Algebra

and B. Matrix C is the product of these two matrices. Do you find

anything interesting about the structure of this matrix?
9. Now choose Matrix Type = 2 (Symmetric Matrix). Run the VI.
Observe the structure of the two symmetric matrices A and B.
Matrix C is the product of these two matrices. Do you find anything
interesting about the structure of this matrix?
10. Change the Matrix A Size to 5. Keep the Matrix B Size at 6. Run the
VI. Why did you get an error1?
11. Save the VI as Matrix Multiplication.vi and close it.

End of Exercise 8-3

1. Description of Error Code -20039: The number of columns in the first matrix is not equal to the number
of rows in the second matrix or vector.

National Instruments Corporation

8-25

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

Exercise 8-4
Objective: To study positive definite matrices.
In many applications, it is advantageous to determine if your matrix is
positive definite. This is because if your matrix is indeed positive
definite, you can save a significant amount of computation time when
using VIs to compute determinants, solve linear systems of equations, or
compute the inverse of a matrix. All these VIs let you choose the matrix
type, and properly identifying the matrix can significantly improve
performance.
In this exercise, you will learn about complex positive definite matrices.
A complex matrix is positive definite if and only if it is Hermitian; that
is, A = A H , and the quadratic form X H AX > 0 for all nonzero vectors X.

Front Panel

1. Build the matrix A and the rest of the front panel as shown in the front
panel above.
2. Switch to the block diagram and use the Test Complex Positive
Definite VI to check if this matrix is positive definite.
Test Complex Positive Definite VI (Analysis Linear
Algebra Complex Linear Algebra Advanced Complex
Linear Algebra subpalette). Use this function to check if the
input matrix A is positive definite.
3. Enter the vector X shown below in the array of controls x.
x =

LabVIEW Signal Processing Course Manual

2.34 + 9.8i
1.23 + 4.5i
3.45 4.56i

8-26

National Instruments Corporation

Lesson 8

Linear Algebra

4. Enter the complex conjugate CX of X in the array of controls cx.

(This part has not been completed in the front panel above.)
Hint: Complex Conjugate of a+ib is a-ib
Complex A x Vector VI (Analysis Linear
Algebra Complex Linear Algebra subpalette). This
function multiplies a complex input matrix and a complex
input vector.
5. Switch to the block diagram. Compute the complex matrix vector
multiplication Y = AX .
Hint: Use the Complex A x Vector VI
Complex Dot Product VI (Analysis Linear
Algebra Complex Linear Algebra subpalette) Use this
function to compute the dot product of two complex vectors.
6. Compute the complex dot product of the vectors CX and Y. That is,
Z = CX Y .
Hint: Use the Complex Dot Product VI.
7. You have now computed the product Z = X H AX . What is the value of
Z? Does this result verify the above definition of complex positive
definite matrices?
8. Compare your block diagram with the diagram shown below.

9. Save the VI as Positive Definite Matrix.vi and close it.

End of Exercise 8-4

National Instruments Corporation

8-27

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

Exercise 8-5
Objective: To compute the eigenvalues and eigenvectors of a real matrix.
As explained earlier, the eigenvalues-eigenvector problem is widely
used in a number of different practical applications. For example, in the
design of control systems, the eigenvalues help determine whether the
system is stable or unstable. If all the eigenvalues have nonpositive real
parts, the system is stable. However, if any of the eigenvalues have a
positive real part, it means that the system is unstable. If the system is
unstable, you can design a feedback system to obtain the desired
eigenvalues and ensure stability of the overall system.
In this exercise, you will use the EigenValues and Vectors VI to
compute all the eigenvalues and eigenvectors of a real matrix. You will
also learn an alternative definition of eigenvalues and numerically verify
this definition.

1. Build the front panel as shown above. You can resize the controls to
view all the elements in the matrices A and EigenVectors and the
array EigenValues. The size of the matrix A is 10 10 .
Eigenvalues is a one-dimensional complex array of size 10
containing all the computed eigenvalues of the input matrix.
Eigenvectors is a 10 10 complex matrix containing all the
computed eigenvectors of the input matrix. The ith column of
Eigenvectors is the eigenvector corresponding to the ith component
of the Eigenvalues vector.

LabVIEW Signal Processing Course Manual

8-28

National Instruments Corporation

Lesson 8

Linear Algebra

2. Build the block diagram as shown below.

EigenValues and Vectors VI (Analysis Linear Algebra
subpalette) In this exercise, you will use this VI to compute the
eigenvalues and the eigenvectors of the input matrix.

3. Matrix A is a real matrix consisting of randomly generated numbers.

Use the EigenValues and Vectors VI from the Analysis Linear
Algebra subpalette to compute both the eigenvalues and the
eigenvectors of this matrix. Remember to set the output option
control to 1. Choose matrix type = General.

4. Earlier in this section, you looked at the classical definition of

eigenvalues and eigenvectors. A different and widely used definition
of eigenvalues is as follows. The eigenvalues of A are the values
such that det ( A I ) = 0 , where det stands for the determinant of
the matrix. In this exercise, you will numerically verify the
validity of this definition. Select the Check Definition VI from
Lvspcex.llb. The round LED on the front panel will glow green
if the definition is true and turn to red if the definition fails.

National Instruments Corporation

8-29

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

5. Run the VI several times. You will notice that the eigenvalues can be
real or complex numbers although the matrix is purely real. The same
holds true for the eigenvectors also. Furthermore, you will notice
that complex eigenvalues of a real matrix always occur in complex
conjugate pairs (that is, if + i is an eigenvalue of a real matrix, so
is i ).
6. Did the definition stated above in step 3 always hold true?
7. Save the VI as My EigenValues and Vectors.vi and close it.

End of Exercise 8-5

LabVIEW Signal Processing Course Manual

8-30

National Instruments Corporation

Lesson 8

Linear Algebra

C. Matrix Inverse and Solving Systems of Linear Equations

The inverse, denoted by A 1 , of a square matrix A is a square matrix such
that
1

A A = AA

= I

where I is the identity matrix. The inverse of a matrix exists if and only
if the determinant of the matrix is not zero (that is, it is nonsingular). In
general, you can find the inverse of only a square matrix. You can,
however, compute the pseudoinverse of a rectangular matrix, as
discussed later in section D.

Solutions of Systems of Linear Equations

In matrix-vector notation, a system of linear equations has the form Ax = b ,
where A is a n n matrix and b is a given n-vector. The aim is to determine
x, the unknown solution n-vector. There are two important questions to be
asked about the existence of such a solution. Does such a solution exist, and
if it does is it unique? The answer to both of these questions lies in
determining the singularity or nonsingularity of the matrix A.
As discussed earlier, a matrix is said to be singular if it has any one of
the following equivalent properties:

The inverse of the matrix does not exist.

The determinant of the matrix is zero.

The rows (or columns) of A are linearly dependent.

Az = 0

for some vector z 0 .

Otherwise, the matrix is nonsingular. If the matrix is nonsingular, its

inverse A 1 exists, and the system Ax = b has a unique solution: x = A 1 b
regardless of the value for b. On the other hand, if the matrix is singular,
then the number of solutions is determined by the right-side vector b. If
A is singular and Ax = b , then A ( x + z ) = b for any scalar , where the
vector z is as in the last definition above. Thus, if a singular system has
a solution, then the solution cannot be unique.
It is not a good idea to explicitly compute the inverse of a matrix,
because such a computation is prone to numerical inaccuracies.
Therefore, it is not a good strategy to solve a linear system of equations
by multiplying the inverse of the matrix A by the known right-side
vector. The general strategy to solve such a system of equations is to
transform the original system into one whose solution is the same as that
of the original system, but is easier to compute. One way to do so is to

National Instruments Corporation

8-31

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

use the Gaussian Elimination1 technique. The three basic steps involved
in the Gaussian Elimination technique are as follows. First, express the
matrix A as a product A = LU where L is a unit lower triangular matrix
and U is an upper triangular matrix. Such a factorization is known as LU
factorization. Given this, the linear system Ax = b can be expressed as
LUx = b . Such a system can then be solved by first solving the lower
triangular system Ly = b for y by forward-substitution. This is the
second step in the Gaussian Elimination technique. For example, if
L = a 0
b c

y = p
q

b = r
s

s bp )
then p = --r- ,q = (------------------ . The first element of y can be easily determined
a

due to the lower triangular nature of the matrix L. Then you can use this
value to compute the remaining elements of the unknown vector
sequentially. Hence, the name forward-substitution. The final step
involves solving the upper triangular system Ux = y by
back-substitution. For example, if
U = a b
0 c

x = m
n

y = p
q

p bn )
then n = q---, m = (------------------- . In this case, this last element of x can be easily
c

determined and then used to determine the other elements sequentially.

Hence, the name back-substitution. So far, this lesson has discussed the
case of square matrices. Because a nonsquare matrix is necessarily
singular, the system of equations must have either no solution or a
nonunique solution. In such a situation, you usually find a unique
solution x that satisfies the linear system in an approximate sense.
The Analysis library includes VIs for computing the inverse of a matrix,
computing LU decomposition of a matrix, and solving a system of linear
equations. It is important to identify the input matrix properly, as it helps
avoid unnecessary computations, which in turn helps to minimize
numerical inaccuracies. The four possible matrix types are general
matrices, positive definite matrices2, and lower and upper triangular
matrices. If the input matrix is square, but does not have a full rank (a
rank-deficient matrix), then the VI finds the least square solution x. The
least square solution is the one that minimizes the norm of Ax b . The
same holds true also for nonsquare matrices.

1. For more information on Gaussian Elimination, see Matrix Computations by G.H. Golub and C.F. Van
Loan. The John Hopkins University Press, Baltimore, 1989.
T

2. A real matrix is positive definite if and only if it is symmetric and the quadratic form X AX > 0 for all
nonzero vectors X.

LabVIEW Signal Processing Course Manual

8-32

National Instruments Corporation

Lesson 8

Linear Algebra

Exercise 8-6
Objective: To compute the inverse of a matrix.
You will build a VI that computes the inverse of a matrix A. Further, you
will compute a matrix B, which is similar to matrix A. A matrix B is similar
to a matrix A if there is a nonsingular matrix T such that B = T1 AT so that
A and B have the same eigenvalues. You will verify this definition of similar
matrices.

Front Panel

1. Build the front panel as shown above. Matrix A is a 2 2 real matrix.

Matrix T is a 2 2 nonsingular matrix that will be used to construct
the similar matrix B.

National Instruments Corporation

8-33

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

Block Diagram

2. Construct the block diagram as shown above.

Inverse Matrix VI (Analysis Linear Algebra subpalette). In
this exercise, this function computes the inverse of the input
matrix A.
AxB VI (Analysis Linear Algebra subpalette). In this
exercise, this function multiplies two two-dimensional input
matrices.
EigenValues and Vectors VI (Analysis Linear Algebra
subpalette). In this exercise, this VI computes the eigenvalues
and eigenvectors of the input matrix.
3. Return to the front panel and run the VI. Check if the eigenvalues of
A and the similar matrix B are the same.
4. Save the VI as Matrix Inverse.vi and close it.

End of Exercise 8-6

LabVIEW Signal Processing Course Manual

8-34

National Instruments Corporation

Lesson 8

Linear Algebra

Exercise 8-7
Objective: To solve a system of linear equations.
Many practical applications require you to solve a system of linear
equations. A very important area of application is related to military
defense. This includes analysis of electromagnetic scattering and radiation
from large targets, performance analysis of large radomes, and design of
aerospace vehicles having low radar cross sections (the stealth technology).
A second area of application is in the design and modeling of wireless
communication systems such as hand-held cellular phones. This list of
applications goes on and on, and therefore it is very important for you to
properly understand how to use the VIs in the Analysis library to solve a
linear system of equations.

1. Use the Solve Linear Equations VI in the Analysis Linear

Algebra subpalette to solve the system of equations Ax = b where
the Input Matrix A and the Known Vector b are
A =

2
2 4 2
4 9 3 , b = 8
10
2 1 7

Choose matrix type equal to general.

2. Use A x Vector.vi to multiply the matrix A and the vector x (output
of the above operation) and check if the result is equal to the vector
b above.
3. Save the VI as Linear System.vi and close it.

End of Exercise 8-7

National Instruments Corporation

8-35

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

D.Matrix Factorization
The previous section discussed how a linear system of equations can be
transformed into a system whose solution is simpler to compute. The basic
idea was to factorize the input matrix into the multiplication of several,
simpler matrices. You looked at one such technique, the LU decomposition
technique, in which you factorized the input matrix as a product of upper
and lower triangular matrices. Other commonly used factorization methods
are Cholesky, QR, and the Singular Value Decomposition (SVD). You can
use these factorization methods to solve many matrix problems, such as
solving linear system of equations, inverting a matrix, and finding the
determinant of a matrix.
If the input matrix A is symmetric and positive definite, then an LU
factorization can be computed such that A = U T U , where U is an upper
triangular matrix. This is called Cholesky factorization. This method
requires only about half the work and half the storage compared to LU
factorization of a general matrix by Gaussian elimination. As you saw
earlier in Exercise 1-4, it is easy to determine if a matrix is positive
definite by using the Test Positive Definite VI in the Analysis library.
A matrix Q is orthogonal if its columns are orthonormal. That is, if
T
Q Q = I , the identity matrix. QR factorization technique factors a matrix
as the product of an orthogonal matrix Q and an upper triangular matrix
R. That is, A = QR . QR factorization is useful for both square and
rectangular matrices. A number of algorithms are possible for QR
factorization, such as the Householder transformation, the Givens
transformation and the fast Givens transformation.
The singular value decomposition (SVD) method decomposes a matrix
into the product of three matrices: A = USV T . U and V are orthogonal
matrices. S is a diagonal matrix whose diagonal values are called the
singular values of A. The singular values of A are the nonnegative square
roots of the eigenvalues of A T A , and the columns of U and V, which are
called left and right singular vectors, are orthonormal eigenvectors of
T
T
AA and A A , respectively. SVD is useful for solving analysis problems
such as computing the rank, norm, condition number, and pseudoinverse
of matrices. The following section discusses this last application.

Pseudoinverse
The pseudoinverse1 of a scalar is defined as 1 if 0 , and zero
otherwise. You can now define the pseudoinverse of a diagonal matrix by
transposing the matrix and then taking the scalar pseudoinverse of each

1. In case of scalars, pseudoinverse is the same as the inverse.

LabVIEW Signal Processing Course Manual

8-36

National Instruments Corporation

Lesson 8

Linear Algebra

entry. Then the pseudoinverse of a general real m n matrix A, denoted by

A , is given by
A = VS U

Note that the pseudoinverse exists regardless of whether the matrix is

square or rectangular. If A is square and nonsingular, the pseudoinverse
is the same as the usual matrix inverse. The Analysis library includes a
VI for computing the pseudoinverse of real and complex matrices.

National Instruments Corporation

8-37

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

Exercise 8-8
Objective: To compute Cholesky decomposition and QR decomposition.
Exercise 1-4 discussed complex positive definite matrices. You verified that
the matrix A
A =

139
91.30 + 47.06i 67.64 + 62.35i
91.30 47.06i
152.41
94.32 + 47.52i
67.64 62.35i 94.32 47.52i
262.00

is a positive definite matrix. In this exercise, compute the Cholesky

decomposition of this matrix. Also, compute the QR decomposition of a
matrix B.

Front Panel

1. Build the front panel as shown above. You can resize the array of
controls to see all the elements of the matrix.
2. The matrix U is an upper triangular matrix that is the result of the
Cholesky Decomposition of the matrix A.
3. The matrix B is a rectangular matrix. That is, the number of rows m
is different from the number of columns n. The result of the QR
factorization is an m m orthogonal matrix Q and an upper triangular
matrix R of size m n .
4. In this exercise, you will also verify the definition of orthogonal
matrices, Q T Q = I . The Result matrix contains the product of the
transpose of the orthogonal matrix and the orthogonal matrix itself.

LabVIEW Signal Processing Course Manual

8-38

National Instruments Corporation

Lesson 8

Linear Algebra

Block Diagram

5. Build the block diagram as shown above

Complex Cholesky Factorization VI (Analysis Linear
Algebra Complex Linear Algebra Advanced Complex
Linear Algebra subpalette). In this exercise, this function
computes the Cholesky decomposition of the positive definite
input complex matrix.
QR Factorization VI (Analysis Linear Algebra Advanced
Linear Algebra subpalette). In this exercise, this function
computes the QR factorization of the input real matrix.
6. Select the Complex Cholesky Factorization VI from the
Analysis Linear Algebra Complex Linear Algebra
Advanced Complex Linear Algebra subpalette. The output
terminal of this VI is connected to matrix U.
7. Generate a random matrix B. Select the QR Factorization VI from
the Analysis Linear Algebra Advanced Linear Algebra
subpalette. Connect the matrix B as the input matrix for QR
factorization. Connect the Q and R outputs of this VI to matrix Q and
matrix R, respectively.
8. Compute the transpose of the matrix Q and then multiply this
transpose with the original matrix Q. The result of this operation is
connected to the result matrix.

National Instruments Corporation

8-39

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

9. Choose a value for the number of rows and the number of columns.
Run the VI. Notice the upper triangular structure of the output of the
Cholesky decomposition and the R output of the QR factorization.
Notice the structure of the orthogonal matrix Q. You can run the VI
a number of times to generate different matrices and check if this
definition is true.
10. Save the VI as QR Factor.vi and close it.

End of Exercise 8-8

LabVIEW Signal Processing Course Manual

8-40

National Instruments Corporation

Lesson 8

Linear Algebra

Exercise 8-9 (Optional)

Objective: To compute singular value decomposition.
In this exercise, you will compute the singular value decomposition of the
following matrix.
1
A = 4
7
10

2
5
8
11

3
6
9
12

Use the SVD Factorization VI from the Analysis Linear Algebra

Advanced Linear Algebra subpalette. Compute the rank of the matrix
A using the Matrix Rank VI (Analysis Linear Algebra Advanced
Linear Algebra subpalette). Do you see an interesting relation between
the rank of this matrix and the number of nonzero singular values? (The
singular values are stored in the one-dimensional array S.) Now compute
the 2-norm of this matrix using the Matrix Norm VI (Analysis Linear
Algebra Advanced Linear Algebra subpalette). Do you see an
interesting relation between this number and the largest singular value
of the matrix?
Note:

The rank of a matrix is equal to the number of nonzero singular values,

which in this example is equal to 2. Also, as discussed at the beginning of
this lesson, the 2-norm of a matrix is equal to its largest singular value.

End of Exercise 8-9

National Instruments Corporation

8-41

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

Summary

A matrix can be considered as a two-dimensional array of m rows

and n columns. Determinant, rank, and condition number are some
important attributes of a matrix.

The condition number of a matrix affects the accuracy of the final

solution.

The determinant of a diagonal matrix, an upper triangular matrix, or

a lower triangular matrix is the product of its diagonal elements.

Two matrices can be multiplied only if the number of columns of the

first matrix is equal to the number of rows in the second matrix.

An eigenvector of a matrix is a nonzero vector that does not rotate

when the matrix is applied to it. Similar matrices have the same
eigenvalues.

The existence of a unique solution for a system of equations depends

on whether the matrix is singular or nonsingular.

Review Questions

0 1 1

1. For the matrix given by A = 1 2 3 , calculate its rank, determinant,

1-norm, and inf-norm.
2 0 1
Which LabVIEW/BridgeVIEW VI(s) could you use to check your
answer?
2. The condition number of a matrix B is 15.3, and that of a matrix C is
30,532. Which of these matrices, B or C, is closer to being singular?
3. Which of the following is true, and which is false?
a. The eigenvalues of a real matrix are always real.
b. The rank of an m x n matrix could at most be equal to the larger
of m or n (m n).
4. For the two vectors given by x = [1,2] and y = [3,4], calculate:
a. Their dot product
b. Their outer product
c. The angle between the two vectors
Which LabVIEW/BridgeVIEW VI(s) could you choose to check
your answer?
5. Which VI could you use to check if a matrix is positive definite?
Why might you want to make such a check?
6. Why is matrix factorization important? Which are the matrix
factorization VIs available in LabVIEW/BridgeVIEW?

LabVIEW Signal Processing Course Manual

8-42

National Instruments Corporation

Lesson 8

Linear Algebra

Notes

National Instruments Corporation

8-43

LabVIEW Signal Processing Course Manual

Lesson 8 Linear Algebra

Notes

LabVIEW Signal Processing Course Manual

8-44

National Instruments Corporation

Lesson 9
Probability and Statistics
Introduction
In this lesson, you will learn some of the fundamental concepts in
probability and statistics such as mean or average, variance, histogram, and
others. The lesson describes different LabVIEW/BridgeVIEW VIs that
compute these quantities and show how they can be used in different
applications.

You Will Learn:

A. What the terms probability and statistics mean, and how are they
relevant in different areas of applications.
B. About the most commonly used concepts in statistics and how to use
the Statistics VIs.
C. About the most commonly used concepts in probability and how to
use the Probability VIs.

National Instruments Corporation

9-1

LabVIEW Signal Processing Course Manual

Lesson 9 Probability and Statistics

A. Probability and Statistics

We live in an information age in which facts and figures form an important
part of life. Statements such as There is a 60% chance of thunderstorms,
Joe was ranked among the top five in the class, Michael Jordan has an
average of 30 points this season, and so on are common. These statements
give a lot of information, but we seldom think about how this information
was obtained. Was there a lot of data involved in obtaining this information?
If there was, how did someone condense it to single numbers such as 60%
chance and average of 30 points or terms such as top five. The answer to
all these questions brings up the very interesting field of statistics.
First, consider how information (data) is generated. Consider the
statistics of part of the 1997 basketball season. Michael Jordan of the
Chicago Bulls played 51 games, scoring a total of 1568 points. This
includes the 45 points he posted, including the game-winning buzzer
three-pointer, in a 103-100 victory over the Charlotte Hornets; his 36
points in an 88-84 victory over the Portland Trail Blazers; a season high
of 51 points in an 88-87 victory over the New York Nicks; 45 points,
seven rebounds, five assists, and three steals in a 102-97 victory over the
Cleveland Cavaliers; and his 40 points, six rebounds, and six assists in a
107-104 victory over the Milwaukee Bucks. The point is not that Jordan
is a great player, but that a single player can generate lots of data in a
single season. The question is, how do you condense all this data so that
it brings out all the essential information and is yet easy to remember?
This is where the term statistics comes into the picture.
To condense all the data, single numbers must make it more intelligible
and help draw useful inferences. For example, consider the number of
points that Jordan scored in different games. It is difficult to remember
how many points he scored in each game. But if you divide the total
number of points that Jordan scored (1568) by the number of games he
has played (51), you have a single number of 30.7 and can call it points
per game average.
Suppose you want to rate Jordans free throw shooting skills. It might be
difficult to do so by looking at his performance in each game. However,
you can divide the number of free throws he has scored in all the games
by the total number of free throws he was awarded. This shows he has a
free throw percentage of 84.4%. You can obtain this number for all the
NBA players and then rank them. Thus, you can condense the
information for all the players into single numbers representing free
throw percentage, points per game, and three-point average. Based on
this information, you can rank players in different categories. You can
further weight these different numbers and come up with a single
number for each player. These single numbers can then help in judging

LabVIEW Signal Processing Course Manual

9-2

National Instruments Corporation

Lesson 9 Probability and Statistics

the Most Valuable Player (MVP) for the season. Thus, in a broad sense,
the term statistics implies different ways to summarize data to derive
useful and important information from it.
The next question is, what is probability? You have looked at ways to
summarize lots of data into single numbers. These numbers then help
draw conclusions for the present. For example, looking at Jordans
statistics for the 1996 season helped the NBA officials elect him the
MVP for that season. It also helped people to infer that he is one of the
best players in the game. But can you say anything about the future? Can
you measure the degree of accuracy in the inference and use it for
making future decisions? The answer lies in the theory of probability.
Whereas, in laymens terms, one would say that it is probable that
Jordan will continue to be the best in the years to come, you can use
different concepts in the field of probability, as discussed later in this
lesson, to make more quantitative statements.
In a completely different scenario, there may be certain experiments
whose outcomes cannot be predetermined, but certain outcomes may be
more probable. This once again leads to the notion of probability. For
example, if you flip an unbiased coin in the air, what is the chance that
it will land heads up? The chance or probability is 50 %. That means, if
you repeatedly flip the coin, half the time it will land heads up. Does this
mean that 10 tosses will result in exactly five heads? Will 100 tosses
result in exactly 50 heads? Probably not. But in the long run, the
probability will work out to be 0.5.
To summarize, whereas statistics allows you to summarize data and
draw conclusions for the present, probability allows you to measure the
degree of accuracy in those conclusions and use them for the future.

National Instruments Corporation

9-3

LabVIEW Signal Processing Course Manual

Lesson 9 Probability and Statistics

B. Statistics
In this section, you will look at different concepts and terms commonly used
in statistics and see how to use the Analysis VIs in different applications.

Mean
Consider a data set X consisting of n samples x 0 , x 1 , x 2 , x 3 , , x n 1 . The
mean value (a.k.a. average) is denoted by x and is defined by the formula
1
x = --- ( x 0 + x 1 + x 2 + x 3 + + x n 1 )
n

In other words, it is the sum of all the sample values divided by the
number of samples. As you saw in the Michael Jordan example above,
the data set consisted of 51 samples. Each sample was equal to the
number of points that Jordan scored in each game. The total of all these
points was 1568, divided by the number of samples (51) to get a mean
or average value of 30.7.
The input-output connections for the Mean VI are shown below.

Median
Let S = {s 0, s 1, s 2, , s n 1} represent the sorted sequence of the data set X.
The sequence can be sorted either in the ascending order or in
descending order. The median of the sequence is denoted by x median and
is obtained by the formula
si
x median =
0.5 ( s k 1 + s k )

n is odd
n is even

n
1
where i = n----------- and k = --- .
2

In words, the median of a data sequence is the midpoint value in the

sorted version of that sequence. For example, consider the sequence
{5, 4, 3, 2, 1} consisting of five (odd number) samples. This sequence is
already sorted in the descending order. In this case, the median is the
midpoint value, 3. Consider a different sequence { 1, 2, 3, 4 } consisting of
four (even number) samples. This sequence is already sorted in the
ascending order. In this case, there are two midpoint values, 2 and 3. As
per the formula above, the median is equal to 0.5 ( 2 + 3 ) = 2.5 . If a

LabVIEW Signal Processing Course Manual

9-4

National Instruments Corporation

Lesson 9 Probability and Statistics

student X scored 4.5 points on a test and another student Y scored 1 point
on the same test, the median is a very useful quantity for making
qualitative statements such as X lies in the top half of the class or Y
lies in the bottom half of the class.
The input-output connections for the Median VI are shown below.

Sample Variance
The sample variance of the data set X consisting of n samples is denoted by
2
s and is defined by the formula
2
2
2
1
2
s = ------------ [ ( x 1 x ) + ( x 2 x ) + + ( x n x ) ]
n1

where x denotes the mean of the data set. Hence, the sample variance is
equal to the sum of the squares of the deviations of the sample values
from the mean divided by n-1.
Note:

The above formula does not apply for n = 1. However, it does not mean
anything to compute the sample variance if there is only one sample in the
data set.
The input-output connections for the Sample Variance VI are shown
below.

In other words, the sample variance measures the spread or dispersion of

the sample values. If the data set consists of the scores of a player from
different games, the sample variance can be used as a measure of the
consistency of the player. It is always positive, except when all the
sample values are equal to each other and in turn equal to the mean.
There is one more type of variance called population variance. The
formula to compute population variance is similar to the one above to
compute sample variance, except for the (n-1) in the denominator
replaced by n.

National Instruments Corporation

9-5

LabVIEW Signal Processing Course Manual

Lesson 9 Probability and Statistics

The input-output connections for the Variance VI are shown below.

The Sample Variance VI computes sample variance, whereas the

Variance VI computes the population variance. Whereas statisticians
and mathematicians prefer to use the latter, engineers prefer to use the
former. It really does not matter for large values of n , say n 30 .
Note:

Use the proper type of VI suited for your application.

Standard Deviation
The positive square root of the sample variance s 2 is denoted by s and is
called the standard deviation of the sample.
The input-output connections for the Standard Deviation VI are shown
below.

Mode
The mode of a sample is a sample value that occurs most frequently in the
sample. For example, if the input sequence X is
X = { 0, 1, 3, 3, 4, 4, 4, 5, 5, 7 }

then the mode of X is 4, because that is the value that most often occurs
in X.
The input-output connections for the Mode VI are shown below.

LabVIEW Signal Processing Course Manual

9-6

National Instruments Corporation

Lesson 9 Probability and Statistics

Moment About Mean

If X represents the input sequence with n number of elements in it, and x is
the mean of this sequence, then the mth-order moment can be calculated
using the formula
n1

1
= --n

(x x)

i=0

In other words, the moment about mean is a measure of the deviation of

the elements in the sequence from the mean. Note that for m = 2 , the
moment about mean is equal to the population variance.
The input-output connections for the Moment About Mean VI are
shown below.

Histogram
So far, this lesson has discussed different ways to extract important features
of a data set. The data is usually stored in a table format, which many people
find difficult to grasp. It is generally useful to display the data in some form.
The visual display of data helps us gain insights into the data. Histogram is
one such graphical method for displaying data and summarizing key
information. Consider a data sequence X = { 0, 1, 3, 3, 4, 4, 4, 5, 5, 8 } . Divide
the total range of values into 8 intervals. These intervals are 0-1, 1-2,
2-3, ..., 7-8. The histogram for the sequence X then plots the number of data
samples that lie in that interval, not including the upper boundary.
3

The figure above shows that one data sample lies in the range 0-1 and 1-2,
respectively. However, there is no sample in the interval 2-3. Similarly, two

National Instruments Corporation

9-7

LabVIEW Signal Processing Course Manual

Lesson 9 Probability and Statistics

samples lie in the interval 3-4, and three samples lie in the range 4-5.
Examine the data sequence X above and be sure you understand this
concept.
There are different ways to compute data for a histogram. Next you will
see how it is done in the Histogram VI using the sequence X.

As shown above, the inputs to this VI are the input sequence X and the
number of intervals m. The VI obtains Histogram:h(x) as follows. It
scans X to determine the range of values in it. Then the VI establishes
the interval width, x , according to the specified value of m
max min
x = -------------------------m

where max is the maximum value found in X, min is the minimum value
found in X, and m is the specified number of intervals.
0
Let m = 8 . Then x = 8----------- = 1
8

Let represent the output sequence X Values. The histogram is a

function of X. This VI evaluates the elements of using
i = min + 0.5x + ix

for

i = 0, 1, 2, , m 1

For this example, 0 = 0.5, 1 = 1.5, , 7 = 7.5 .

The VI then defines the ith interval to be in the range of values from
i 0.5x up to but not including i + 0.5x ,
i = [ ( i 0.5x ), ( i + 0.5x ) ), for

i = 0, 1, 2, , m 1

and defines the function y i ( x ) = 1 for x belonging to i and zero

elsewhere. The function has unity value if the value of x falls within the
specified interval, not including the boundary. Otherwise, it is zero.
Notice that the interval is centered about i and its width is x . If a value
is equal to max, it is counted as belonging to the last interval.
For our example, 0 = [ 0, 1 ), 1 = [ 1, 2 ), , 7 = [ 7, 8 ) and as an example
y 0 ( 0 ) = 1 and y 0 ( 1 ) = y 0 ( 3 ) = y 0 ( 4 ) = y 0 ( 5 ) = y 0 ( 8 ) = 0 .
Finally, the VI evaluates the histogram sequence H using
n1

hi =

y (x )
i

i = 0, 1, 2, , m 1

for

j=0

LabVIEW Signal Processing Course Manual

9-8

National Instruments Corporation

Lesson 9 Probability and Statistics

where h i represents the elements of the output sequence Histogram: h(X)

and n is the number of elements in the input sequence X. For this
example, h 0 = 1, h 4 = 3, , h 7 = 1 .
The Analysis library also has a General Histogram VI that is more
advanced than the Histogram VI. Please refer to the LabVIEW Analysis
VI Reference Manual for detailed information.

Mean Square Error (MSE)

If X and Y represent two input sequences, the mean square error is the
average of the sum of the square of the difference between the
corresponding elements of the two input sequences. The following formula
is used to find the mse.
n1

1
mse = --n

(x y )
i

i=0

where n is the number of data points.

S1

y1

S2

y2

Consider a digital signal x fed to a system, S1. The output of this system
is y1. Now you acquire a new system, S2, which is theoretically known
to generate the same result as S1 but has two times faster response time.
Before replacing the old system, you want to be absolutely sure that the
output response of both the systems is the same. If the sequences y1 and
y2 are very large, it is difficult to compare each element in the
sequences. In such a scenario, you can use the MSE VI to calculate the
mean square error (mse) of the two sequences y1 and y2. If the mse is
smaller than an acceptable tolerance, the system S1 can be reliably
replaced by the new system S2.
The input-output connections for the MSE VI are shown below.

National Instruments Corporation

9-9

LabVIEW Signal Processing Course Manual

Lesson 9 Probability and Statistics

Root Mean Square (RMS)

The root mean square x of a sequence X is the positive square root of the
mean of the square of the input sequence. In other words, you can square the
input sequence, take the mean of this new squared sequence, and then take
the square root of this quantity. The formula used to compute the rms
value is
n1

x =

1
--n

2
i

i=0

where n is the number of elements in X.

RMS is a widely used quantity in the case of analog signals. For a sine
voltage waveform, if Vp is the peak amplitude of the signal, then the root
V

mean square voltage Vrms is given by ------p- . The following figure shows a
2

voltage waveform of peak amplitude = 2 V and the RMS value of

2 1.41 V computed using the RMS VI from the Analysis library.

The input-output connections for the RMS VI are shown below.

LabVIEW Signal Processing Course Manual

9-10

National Instruments Corporation

Lesson 9 Probability and Statistics

Exercise 9-1
Objective:

To use different Statistics VIs in an example using Michael Jordans basketball

scores.
In this exercise, you will learn how to use different Statistics VIs in an
interesting example. The data set consists of the number of points scored by
Michael Jordan of the Chicago Bulls in each of the 51 games he played
during part of the 1997 NBA season. You will use some of the concepts
discussed in the previous section to decipher this information and create
single numbers that are easy to remember and yet reveal all the information
that the entire data set provides.

Front Panel

1. Open the Statistics VI from Lvspcex.llb.

2. Open the front panel as shown above. The Points control is the data
set consisting of the number of points scored by Jordan in each game
this season. The number of games he has played this season to date
is 51.
The Total Number of Points digital indicator is the sum of all the
elements in the given data set.
The Points Per Game digital indicator is the average of Jordans
scores this season.
The Sample Variance, Median, and Mode digital indicators are the
sample variance, median, and mode of the data set.

National Instruments Corporation

9-11

LabVIEW Signal Processing Course Manual

Lesson 9 Probability and Statistics

Histogram of Points is an XY graph that shows the distribution of

points in different intervals. The number of intervals is set using the
Number of intervals digital control. Points in Each Game is a
waveform graph that plots the number of points scored in each game.

Block Diagram

3. Build the diagram shown above:

Sample Variance VI (Analysis Probability and Statistics
subpalette). In this exercise, this function computes the sample
variance of the data set Points.
Mean VI (Analysis Probability and Statistics subpalette). In
this exercise, this function computes the mean value (average)
of the data set Points.
Histogram VI (Analysis Probability and Statistics
subpalette). In this exercise, this function computes the
histogram of the data set Points.

LabVIEW Signal Processing Course Manual

9-12

National Instruments Corporation

Lesson 9 Probability and Statistics

Median VI (Analysis Probability and Statistics subpalette).

In this exercise, this function computes the median of the data
set Points.
Mode VI (Analysis Probability and Statistics subpalette). In
this exercise, this function computes the mode of the data set
Points.
Bundle function (Cluster subpalette) In this exercise, this
function assembles the outputs of the Post Processing VI
(explained later) to plot on the XY graph Histogram of Points.
4. Follow the instructions in steps 4 through 9 and build the block
diagram as shown above.
5. You will compute the Points Per Game (Jordans average) by using
the Mean VI in the Analysis library. Connect the Points control to
the input terminal X and the output mean to the Points Per Game
indicator.

6. You will compute the Sample Variance using the Sample Variance
VI in the Analysis library. Once again, connect the Points control to
the input terminal X and the output sample variance to the Sample
Variance indicator.

Note:

This VI also computes the mean of the data set.

National Instruments Corporation

9-13

LabVIEW Signal Processing Course Manual

Lesson 9 Probability and Statistics

7. You will compute the Median using the Median VI in the Analysis
library. Connect the Points control to the input terminal X and the
output median to the Median indicator.

8. You will compute the Mode using the Mode VI in the Analysis
library. Connect the Points control to the input terminal X. Choose
the number of intervals equal to 3. Connect the output mode to the
Mode indicator.

9. You first will compute the data for the histogram using the
Histogram VI. Connect the control Points to the input terminal X
and set the number of intervals equal to 3. This VI generates
histogram values and X values, which are the midpoints of the
different intervals as discussed above. You can plot the histogram by
using the X values for the X axis and the histogram values for the Y
axis. If you are interested, try doing this and observe the histogram.

10. Generally, you may want to view the histogram in a different way.
Select the PostProcessing VI from Lvspcex.llb. Connect the
Points control to the Points input terminal, the histogram output of
the Histogram VI to the Histogram input terminal, and set the
number of intervals equal to 3 (same value you chose earlier for the
Histogram VI). This VI generates the data for plotting the histogram
in a better way. The X axis values are stored in the Boundaries
output terminal, and the Y axis values are stored in the PlotValues

LabVIEW Signal Processing Course Manual

9-14

National Instruments Corporation

Lesson 9 Probability and Statistics

output terminal. You can use the output of this VI for plotting the
histogram on the Histogram of Points XY graph.

11. Return to the front panel. Set the number of intervals equal to 3 and
run the VI. Study the different output values. See how the histogram
(on the left) provides more information than just plotting the points
in each game (on the right).
12. Save the VI and close it.

End of Exercise 9-1

National Instruments Corporation

9-15

LabVIEW Signal Processing Course Manual

Lesson 9 Probability and Statistics

C. Probability
In any random experiment, there is always a chance that a particular event
will or will not occur. A number between 0 and 1 is assigned to measure this
chance, or probability, that a particular event occurs. If you are absolutely
sure that the event will occur, its probability is 100% or 1.0, but if you are
sure that the event will not occur, its probability is 0.
Consider a simple example. If you roll a single unbiased die, there are
six possible events that can occureither a 1, 2, 3, 4, 5, or 6 can result.
What is the probability that a 2 will result? This probability is one in six,
or 0.16666. You can define probability in simple terms as: The
probability that an event A will occur is the ratio of the number of
outcomes favorable to A to the total number of equally likely outcomes.

Random Variables
Many experiments generate outcomes that you can interpret in terms of real
numbers. Some examples are the number of cars passing a stop sign during
a day, number of voters favoring candidate A, and number of accidents at a
particular intersection. The values of the numerical outcomes of this
experiment can change from experiment to experiment and are called
random variables. Random variables can be discrete (if they can take on
only a finite number of possible values) or continuous. As an example of the
latter, weights of patients coming into a clinic may be anywhere from, say,
80 to 300 pounds. Such random variables can take on any value in an
interval of real numbers. Given such a situation, suppose you want to find
the probability of encountering a patient weighing exactly 172.39 pounds.
You will see how to calculate this probability next using an example.
Consider an experiment to measure the life lengths x of 50 batteries of a
certain type. These batteries are selected from a larger population of
such batteries. The histogram for observed data is shown below.

histogram

LabVIEW Signal Processing Course Manual

9-16

life length in hundreds of hours

National Instruments Corporation

Lesson 9 Probability and Statistics

This figure shows that most of the life lengths are between zero and 100
hours, and the histogram values drop off smoothly as you look at larger
life lengths.
You can approximate the histogram shown above by an exponentially
decaying curve. You could take this function as a mathematical model
for the behavior of the data sample. If you want to know the probability
that a randomly selected battery will last longer than 400 hours, this
value can be approximated by the area under the curve to the right of the
value 4. Such a function that models the histogram of the random
variable is called the probability density function.
To summarize all the information above in terms of a definition, a
random variable X is said to be continuous if it can take on the infinite
number of possible values associated with intervals of real numbers, and
there is a function f(x), called the probability density function, such that
1. f ( x ) 0

for all x

2.
3.

f ( x ) dx = 1

P( a X b) =

f ( x ) dx
a

Notice from equation (3) above that for a specific value of the
a

continuous random variable, that is for X=a, P ( X = a ) = f ( x ) dx = 0 . It

a

should not be surprising that you assign a probability of zero to any

specific value, because there are an infinite number of possible values
that the random variable can take. Therefore, the chance that it will take
on a specific value X = a is extremely small.
The previous example used the exponential function model for the
probability density function. There are a number of different choices for
this function. One of these is the Normal Distribution, discussed below.

National Instruments Corporation

9-17

LabVIEW Signal Processing Course Manual

Lesson 9 Probability and Statistics

Normal Distribution
The normal distribution is one of the most widely used continuous
probability distributions. This distribution function has a symmetric bell

shape, as shown above. The curve is centered at the mean value x = 0 , and
its spread is measured by the variance s 2 = 1 . These two parameters
completely determine the shape and location of the normal density function,
whose functional form is given by
1 ( x x )2 ( 2s2 )
f ( x ) = ------------- e
s 2

Suppose a random variable Z has a normal distribution with mean equal

to zero and variance equal to one. This random variable is said to have
standard normal distribution.
The Normal Distribution VI computes the one-sided probability, p, of
the normally distributed random variable x.
p = Prob ( X x )

where X is a standard normal distribution with the mean value equal to

zero and variance equal to one, p is the probability and x is the value.

Suppose you conduct an experiment in which you measure the heights

of adult males. You conduct this experiment on 1000 randomly chosen
men and obtain a data set S. The histogram distribution has many
measurements clumped closely about a mean height, with relatively few
very short and very tall males in the population. Therefore, the
histogram can be closely approximated by a normal distribution. Now
suppose that, among a different set of 1000 randomly chosen males, you

LabVIEW Signal Processing Course Manual

9-18

National Instruments Corporation

Lesson 9 Probability and Statistics

want to find the probability that the height of a male is greater than or
equal to 170 cm. You can use the Normal Distribution VI to find this
probability. Set the input x = 170 . Thus, the choice of the probability
density function is fundamental to obtaining a correct probability value.

The Inverse Normal Distribution VI performs exactly the opposite

function as the Normal Distribution VI. Given a probability p, it finds
the values x that have the chance of lying in a normally distributed
sample. For example, you might want to find the heights that have a 60%
chance of lying in a randomly chosen data set.
As mentioned earlier, there are different choices for the probability
density function. The well-known and widely used ones are the
Chi-Square distribution, the F distribution, and the T-distribution1. The
Analysis library includes VIs that compute the one-sided probability for
these different types of distributions. In addition, it also has VIs that
perform the inverse operation.

1. Interested readers should refer to the latest edition of the Schaums Outline Series on Theory and Problems of Probability and Statistics by Murray Spiegel, McGraw-Hill, Inc., 1975, for detailed discussion on
these three types of distributions.

National Instruments Corporation

9-19

LabVIEW Signal Processing Course Manual

Lesson 9 Probability and Statistics

Exercise 9-2
Objective: To understand key probability concepts.
In this exercise, you will first generate a data sample with standard normal
distribution and then use the Normal Distribution VI to check the
probability of a random variable x.

Front Panel

1. Build the front panel as shown above. NoisePlot is a waveform

graph, whereas NoiseHistogram is an XY graph.

LabVIEW Signal Processing Course Manual

9-20

National Instruments Corporation

Lesson 9 Probability and Statistics

Block Diagram

2. Build the block diagram as shown above. The Gaussian White Noise
VI generates a Gaussian-distributed pattern with mean value equal
to 0 and standard deviation set by the user using the input standard
deviation. Samples is the number of samples of the Gaussian noise
pattern. Seed is the seed value used to generate the random noise.
Gaussian White Noise VI (Analysis Signal Generation
subpalette). In this exercise, this function generates a Gaussian
white noise pattern.
Histogram VI (Analysis Probability and Statistics
subpalette). In this exercise, this function computes the
histogram of the Gaussian noise pattern.
Normal Distribution VI (Analysis Probability and
Statistics subpalette). In this exercise, this function computes
the one-sided probability of the normally distributed random
variable Random Variable.
Connect the Gaussian noise pattern to the waveform graph Noise
Plot.

3. You will compute the histogram of the Gaussian noise pattern using
the Histogram VI used in the previous exercise.

National Instruments Corporation

9-21

LabVIEW Signal Processing Course Manual

Lesson 9 Probability and Statistics

4. As discussed earlier, do some postprocessing to plot the histogram in

a different way. Select the PostProcessing VI from Lvspcex.llb.
5. Bundle the output of this VI and connect it to the Noise Histogram.
6. Select the Normal Distribution VI. Connect the Random Variable
control to the input terminal and connect the output to the probability
indicator.
7. Return to the front panel. Set the Number of Samples control to
2048, Standard Deviation to 1, Seed to 2 and Number of intervals to
10. Run the VI.
8. You will see the Gaussian white noise on the Noise Plot graph. It is
difficult to tell much from this plot. However, the histogram plot for
the same noise pattern provides a lot of information. It shows that
most of the samples are centered around the mean value of zero.
From this histogram, you can approximate this noise pattern by a
Normal Distribution function (Gaussian distribution). Because the
mean value is zero and you set the standard deviation equal to one,
the probability density function is actually a standard normal
distribution.
Note:

It is very important that you carefully choose the proper type of distribution
function to approximate your data. In this example, you actually plotted
the histogram to make this decision. Many times, you can make an
intelligent decision based solely on prior knowledge of the behavior and
characteristics of the data sample.
9. Return to the front panel and enter a value for Random Variable.
This VI will compute the one-sided probability of this normally
distributed random variable. Remember, you have assumed that the
variable is normally distributed by looking at the histogram.
10. Save the VI as Probability.vi and close it.

End of Exercise 9-2

LabVIEW Signal Processing Course Manual

9-22

National Instruments Corporation

Lesson 9 Probability and Statistics

Summary

Different concepts in statistics and probability help decipher

information and data to make intelligent decisions.

Mean, median, sample variance, and mode are some of the statistics
techniques to help in making inferences from a sample to a
population.

Histograms are widely used as a simple but informative method of

data display.

Using the theory of probability, you can make inferences from a

sample to a population and then measure the degree of accuracy in
those inferences.

Review Questions
1. What is the difference between probability and statistics? Which
VI(s) would you use in each case?
2. What is the difference between:
a. Mode and median?
b. Sample variance and population variance?
Which VI(s) would you use in each case?
3. Name some real-world practical applications of using:
a. Histograms
b. The Gaussian probability density function
4. What is the difference between the Normal Distribution VI and the
Inv Normal Distribution VI?

National Instruments Corporation

9-23

LabVIEW Signal Processing Course Manual

Lesson 9 Probability and Statistics

Notes

LabVIEW Signal Processing Course Manual

9-24

National Instruments Corporation

Lesson 10
Digital Filter Design Toolkit
Introduction
In this lesson, you will learn how to use the Digital Filter Design (DFD)
Toolkit to design FIR and IIR filters to meet required specifications. You
will also see how to use the DFD toolkit to analyze your filter design in
terms of its frequency response, impulse and step responses, and its
pole-zero plot.

You Will Learn:

A. About digital filters.
B. About the Digital Filter Design Toolkit.
C. About designing IIR filters.
D. About designing FIR filters.
E. How to use the Digital Filter Design Toolkit to analyze your filter
design.

National Instruments Corporation

10-1

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

A. Review of Digital Filters

In a previous lesson, you learned about the basic theory behind the operation
of digital filters. Below is a brief review of some of the important
information you need to know to use the Digital Filter Design (DFD)
Toolkit.

Filtering
Filtering is one of the most common signal processing techniques and is the
process by which the frequency content of a signal is altered. Some of the
practical applications of filtering are in the bass and treble controls of your
stereo to adjust the frequency response, in the tuning circuits of your radio
and television receivers to select a particular channel, in telephone handsets
to limit the frequency content of the sound signals to 3 KHz, and many
others in the audio, telecommunications, geophysics, and medical fields.

Why Digital Filters?

Until the advent of the computer age, filtering was in analog form using
resistors, inductors, and capacitors. Both the input and the output of the filter
were analog signals. Designing analog filters is a specialized task requiring
a good mathematical background and proper understanding of the filtering
process. However, with the widespread use of computers, digital
representation and processing of signals gained immense popularity due to
the numerous advantages that digital signals have over their analog
counterparts. Because of this, analog filters have gradually been replaced by
digital filters. The advantages of digital filters over analog filters are:

They are software programmable, and so are easy to build and test.

They require only the arithmetic operations of multiplication and

addition/subtraction and so are easier to implement.

They are stable (do not change with time nor temperature) and
predictable.

They do not drift with temperature or humidity or require precision

components.

They have a superior performance-to-cost ratio.

They do not suffer from manufacturing variations or aging.

Filter Response Characteristics

The range of frequencies that a filter passes through it is known as the
passband, whereas the range of frequencies that are attenuated is known as
the stopband. Between the passband and the stopband is a transition region
where the gain falls from one (that is, 0 dB in the passband) to zero or a very
small value (in the stopband). The passband, stopband, and the transition
region for a lowpass filter are shown below.

LabVIEW Signal Processing Course Manual

10-2

National Instruments Corporation

Lesson 10

passband

transition
region

Digital Filter Design Toolkit

stopband

passband ripple

stopband
attenuation

cutoff frequency

The figure above also shows the passband ripple, the stopband
attenuation, and the cut-off frequency, three specifications that are needed
in designing digital filters. The passband ripple (in dB) is the maximum
deviation in the passband from 0 dB, whereas the stopband attenuation
is the minimum attenuation (in dB) in the stopband. In the DFD toolkit,
the passband ripple is also referred to as the passband response.

National Instruments Corporation

10-3

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

B. The Digital Filter Design Toolkit

Lesson 6 introduced the Digital Filter VIs available in the
LabVIEW/BridgeVIEW Advanced Analysis library. Using these VIs is one
way to design your digital filters. However, National Instruments also
provides the Digital Filter Design (DFD) toolkit, which is a complete filter
design and analysis tool you can use to design digital filters to meet your
precise filter specifications. You can graphically design your IIR and FIR
filters, interactively review filter responses, save your filter design work,
and load your design work from previous sessions. If you have a National
Instruments data acquisition (DAQ) device, you can perform real-world
filter testing from within the DFD application. You can view the time
waveforms or the spectra of both the input signal and the filtered output
signal while simultaneously redesigning your digital filters.
After you design your digital filter, you can save the filter coefficients
to a file on your drive. The filter coefficient files can then be loaded for
later implementation by LabVIEW, BridgeVIEW, LabWindows/CVI, or
any other application. The following diagram shows you the conceptual
overview of the DFD Toolkit.

Digital Filter
Design
Application

Save

Load

Filter
Specification
Files

Save

Filter Coefficient Files

LabVIEW

HiQ

LabWindows/CVI

LabVIEW Signal Processing Course Manual

Data Acquisition
and Filtering

10-4

Windows DLL

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

Main Menu
When you launch the DFD application, you get the following panel, which
is referred to as the Main Menu:

Design Options
From the Main Menu, you can choose any of the following four methods of
designing digital filters:

Classical IIR Designfor designing IIR filters by specifying the

frequency response characteristics.

Classical FIR Designfor designing FIR filters by specifying the

frequency response characteristics.

Pole-Zero Placementfor designing either IIR or FIR filters by

adjusting the location of the poles and zeros (in the z-plane) of the
filter transfer function.

Arbitrary FIR Designfor designing FIR filters by specifying the

gain of the filter at selected (two or more) frequencies.

If you double-click on one of the four design selections (or single-click

on a selection and then click on the Open button) in the Main Menu, the
DFD application loads and runs the selected design panel, in which you
can design your filter.

Loading Previously Saved Specifications

You can also load a previously designed filter specification file directly
from the Main Menu by clicking on the Load Filter Spec button. You will
then be prompted to select the filter specification file that you saved during
a previous design work.

National Instruments Corporation

10-5

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

Customizing the DFD Application

By clicking on the Preferences... button in the Main Menu, you can edit
your DFD application preferences for future design sessions.

The selections in the window above tell the DFD application to preload
one or more of the filter design panels into memory when it is started.
Preloading filter designs increases the time taken for the Main Menu to
open. However, when you select a particular design panel from the Main
Menu, the corresponding design panel opens almost immediately. If you
have limited amount of memory on your computer, you may want to
reconsider how many (if any) of the design panels you preload into
memory.

Quitting the DFD Application

Choose the Quit button to exit the DFD application.

Manipulating the Graphical Display

Each design panel has a graphical display showing you the frequency
response of the filter that you are designing. The graphical displays provide
you with considerable flexibility in adjusting the magnitude response that
you are seeing. Before you move on to the next section (Designing IIR
Filters), you should become familiar with the display options.

Panning and Zooming Options

The graph palette has controls for panning (scrolling the display area of a
graph) and for zooming in and out of sections of the graph. The graph palette

LabVIEW Signal Processing Course Manual

10-6

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

is included with many DFD graphs. A graph with its accompanying graph
palette is shown below.

Graph

Palette

If you press the x autoscale button, shown at left, the DFD application
autoscales the X data of the graph. If you press the y autoscale button,
shown at left, the DFD application autoscales the Y data of the graph. If
you want the graph to autoscale either of the scales continuously, click
on the lock switch, shown at the left, to lock autoscaling on.
The scale format buttons, shown left, give you run-time control over the
format of the X and Y scale markers, respectively.
You use the remaining three buttons to control the operation mode for
the graph.
Normally, you are in standard operate mode, indicated by the plus or
crosshatch. In operate mode, you can click in the graph to move cursors
around.
The panning tool switches to a mode in which you can scroll the visible
data by clicking and dragging sections of the graph.
The zoom tool zooms in on a section of the graph by dragging a selection
rectangle around that section. If you click on the zoom tool, you get a
pop-up menu you can use to choose some other methods of zooming.
This menu is shown below.

National Instruments Corporation

10-7

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

A description of each of these options follows.

Zoom by rectangle.
Zoom by rectangle, with zooming restricted to x data (the y scale
remains unchanged.
Zoom by rectangle, with zooming restricted to y data (the x scale
remains unchanged).
Undo last zoom. Resets the graph to its previous setting.
Zoom in about a point. If you hold down the mouse on a specific point,
the graph continuously zooms in until you release the mouse button.
Zoom out about a point. If you hold down the mouse on a specific point,
the graph continuously zooms out until you release the mouse button.
Note:

For the last two modes, you can zoom in and zoom out about a point.
Shift-clicking zooms in the other direction.

Graph Cursors
Below are illustrations of a waveform graph showing two cursors and the
cursor movement control.

cursor movement
control

You can move a cursor on a graph or chart by dragging it with the

Operating tool, or by using the cursor movement control. Clicking the
arrows on the cursor movement control causes all cursors selected to
move in the specified direction. You select cursors by moving them on
the graph with the Operating tool.

LabVIEW Signal Processing Course Manual

10-8

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

C. Designing IIR Filters

You have seen that IIR filters are digital filters whose impulse response is
infinitely long. In practice, the impulse response decays to a very small
value in a finite amount of time. The output of an IIR filter depends on both
the previous and past inputs and the past outputs. For calculating the current
output of the IIR filter, let (Ny - 1) be the number of past outputs and let
(Nx - 1) be the number of past inputs. Denoting the inputs to the filter as x[.]
and the outputs as y[.], the equation for the output of an IIR filter can be
written as:
a 0 y [ i ] = a 1 y [ i 1 ] a 2 y [ i 2 ] + ... a Ny 1 y [ i ( N y 1 ) ] +
b 0 x [ i ] + b 1 x [ i 1 ] + b 2 x [ i 2 ] + ... + b Nx 1 x [ i ( N x 1 ) ]

1
y [ i ] = ------
a0

Ny 1

Nx 1

a [ j ]y [ i j ] +

j=1

b [ k ]x [ i k ]

k=0

(1)

In the above equation, the b[k] are known as the forward coefficients and
the a[j] are known as the reverse coefficients. The output sample at the
present sample index i is the sum of scaled present and past inputs (x[i]
and x[i-k] when k 0) and scaled past outputs (y[i-j]). Usually Nx is equal
to Ny, and this value is known as the order of the filter.

Implementation of IIR filters

IIR filters implemented in the form given by equation (1) are known as
direct form IIR filters. Direct form implementations are usually sensitive to
errors due to the number of bits used to represent the values of the
coefficients (quantization error) and to the precision used in performing the
computations. In addition, a filter designed to be stable can become unstable
when the number of coefficients (that is, the order of the filter) is increased.
A less sensitive implementation is obtained by breaking up the higher
order direct form implementation into an implementation that has
several cascaded filter stages, but where each filter in the cascade is of
a lower order, as shown in the figure below:
x[i]

stage 1

y[i]

(a) Direct Form (of higher order)

x[i]

stage 1

stage 2

stage Ns

y[i]

(b) Cascaded Direct Form Filter Stages (each filter is of lower order than the filter in (a))

National Instruments Corporation

10-9

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

Typically, each lower order filter stage in the cascade form is a

second-order stage. Each second-order stage can be implemented in the
direct form, where you must maintain two past inputs (x[i-1] and x[i-2])
and two past outputs (y[i-1] and y[i-2]). The output of the filter is
calculated using the equation
y[i] = b0x[i] + b1x[i-1] + b2x[i-2] - a1y[i-1] - a2y[i-2]

(2)

The implementation can also be done in the more efficient direct form II,
where you maintain two internal states (s[i-1] and s[i-2]). The output of
the filter is then calculated as follows:
s[i] = x[i] - a1s[i-1] - a2s[i-2]

(3)

y[i] = b0s[i] + b1s[i-1] + b2s[i-2]

The direct form II is a more efficient structure because it uses less
memory. It needs to store only two past internal states (s[i-1] and s[i-2]),
whereas the direct form structure needs to store four past values (x[i-1],
x[i-2], y[i-1] and y[i-2]).

IIR Filter Designs

Depending on whether the ripple in the filters frequency response lies
in the passband and/or the stopband, IIR filters are classified as follows:

Butterworth: no ripple in either the passband nor the stopband.

Chebyshev: ripples only in the passband

Inverse Chebyshev: ripples only in the stopband.

Elliptic: ripples in both the passband and the stopband.

The advantage of using Butterworth filters is for applications where you

want a smooth filter response and no ripples. However, a higher order
Butterworth filter (as compared to Chebyshev, Inverse Chebyshev, or
Elliptic) is generally required for the same filter specifications. This
increases the processing time for Butterworth filters.
The advantage of Chebyshev and Inverse Chebyshev filters over
Butterworth filters is their sharper transition band for the same order
filter. On deciding which of these two types of filters to use, the
advantage of Inverse Chebyshev filters over Chebyshev filters is that
they distribute the ripples in the stopband instead of in the passband.
Because elliptic filters distribute the ripples in both the passband and the
stopband, they can usually be implemented with the smallest order for
the same filter specifications. Hence, they have faster execution speeds
than either of the other filters.

LabVIEW Signal Processing Course Manual

10-10

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

Applications of IIR Filters

The advantage of IIR filters over FIR filters is that IIR filters usually require
fewer filter coefficients to perform similar filtering operations. Thus, they
execute much faster and do not require extra memory, because they execute
in place.
The disadvantage of IIR filters is that they have nonlinear phase
characteristics. Hence, if your application requires a linear phase
response, then you should use an FIR filter instead. However, for
applications where phase information is not necessary, such as in simple
signal monitoring, then IIR filters can be used. Thus, the bandpass filters
in real time octave analyzers are commonly IIR filters, because of their
faster speed and also because it is necessary to determine the distribution
of sound power over several frequency bands, but there is no need to
determine the phase of the signal. The applications of such octave
analyzers where phase information is not important are vibration tests of
aircraft and submarines, testing of appliances, etc.

National Instruments Corporation

10-11

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

Exercise 10-1
Objective: To design an IIR bandpass filter for use in an octave analyzer.
1. Launch the Digital Filter Design Toolkit application. The Main
Menu panel opens as shown.

2. In the Main Menu, select Classical IIR Design and click on the Open
button. The design panel of the Classical IIR Design opens as shown.

On the left, a plot shows the Magnitude vs. Frequency response

characteristic of the filter you design. The specifications for your

LabVIEW Signal Processing Course Manual

10-12

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

filter can be entered in the text entry portion at the upper right side
of the design panel.

The passband response is the minimum gain allowed in the

passband. This is represented by the horizontal blue cursor line in the
Magnitude vs. Frequency plot. With the reference at 0 dB, it is also
the same as the passband ripple.
The passband frequencies determine the frequency edges of the
passband. For lowpass and highpass filters, you have only one
frequency edge. For bandpass and bandstop filters, you will have
two. These frequencies are represented by the vertical blue lines in
the Magnitude vs. Frequency plot.
The stopband attenuation is the minimum attenuation in the
stopband. The horizontal red cursor line represents this attenuation
in the Magnitude vs. Frequency plot.
The stopband frequencies determine the frequency edges of the
stopband. For lowpass and highpass filters, you have only one
frequency edge. For bandpass and bandstop filters, you will have
two. These frequencies are represented by the vertical red lines in the
Magnitude vs. Frequency plot.
The sampling rate control specifies the sampling rate in samples per
second (Hz).
The type control specifies one of the four classical filter types:

Lowpass

Highpass

Bandpass

Bandstop

The design control specifies one of the four classical filter design
algorithms:

National Instruments Corporation

Butterworth

Chebyshev

10-13

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

Inverse Chebyshev

Elliptic

Below the text entry portion is an indicator showing the order of the
IIR filter.

The DFD application automatically estimates the filter order to be

the lowest possible order that meets or exceeds the desired filter
specifications.
At the bottom left of the Classical IIR Design panel is the message
window where error messages are displayed.

You will use the Classical IIR Design panel to design an IIR bandpass
filter that can be used in an octave analyzer. Octave analyzers (see the
lesson on the Third-Octave Analyzer Toolkit) are used in applications
where you need to determine how the signal power is distributed over a
particular frequency range. These applications include the fields of
architectural acoustics, noise and vibration tests in aircraft and
submarines, testing of household appliances, etc.
An octave analyzer uses bandpass filters to separate the signal power
into several frequency bands. The American National Standards Institute
requires that these filters adhere to certain specifications. Some
specifications for one of these filters are:
fp1 = 890.90 Hz
fp2 = 1122.46 Hz
maximum passband ripple 50 millibels
fs1 = 120.48
fs2 = 8300
stopband attenuation 65 dB
Because the purpose of the bandpass filter is to determine the level of
sound power in a particular frequency band, and the phase information
in the signal is not being used, it is not necessary for the filter to be linear
phase. Hence, you can choose an IIR filter for this application. You will
use the DFD toolkit to design the IIR filter to meet these specifications.

LabVIEW Signal Processing Course Manual

10-14

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

3. Change the type control in the text entry box to bandpass.

fp1
fp2

fs2
fs1

The explanation of the controls in the text entry box is the same as
before, except note that because you have selected a bandpass filter,
there are two controls for the passband frequencies and the stopband
frequencies. They are denoted by fp1, fp2, fs1, and fs2, as shown in the
following figure, and bear the following relationship:
fs1 < fp1 < fp2 < fs2

fs1

fp1

fp2

fs2

passband and stopband frequencies for a bandpass filter

4. Looking at the specifications in step 2, enter the following values in
the controls in the text entry box
passband response

-0.5

passband frequencies

890.90 and 1122.46

stopband attenuation

-65

stopband frequencies

120.48 and 8300

sampling rate

25600

type

bandpass

design

elliptic

Note that 50 millibels =

50
-----------1000

50
- deciBels = 0.5 dB. That is
Bels = -------100

why you entered the passband response as -0.5 dB.

National Instruments Corporation

10-15

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

tracking square cursor

5. On the Magnitude Vs. Frequency graph is a tracking square cursor
that you can move around. The frequency and the corresponding
magnitude of the point where the cursor is placed, is displayed on the
frequency and magnitude indicators below the graph. Move the
square cursor to the passband region and verify that the attenuation
in the passband is never below -0.5 dB, as was specified in the
passband response control.
6. The default filter design is Elliptic. Change the filter design one by
one to Butterworth, Chebyshev, and Inverse Chebyshev. Note the
order of the filter. For the same filter order, which of the four filter
designs has the sharpest transition region?
7. You can save the specifications of the filter in a file for later use.
From the DFD menu, select Save Spec... When asked for the name
of the file in which to save the filter specifications, type in
bandpass.iir. In Exercise 10-3, you will load this file and
analyze the characteristics of the filter that you have just designed.

LabVIEW Signal Processing Course Manual

10-16

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

8. You can also save the designed filter coefficients in a file for later
use with the DFD toolkit, or with other programs. From the DFD
menu, select Save Coeff.... When asked for the name of the file in
which to save the filter coefficients, type bpiir.txt. Save the file
as a text file. The appendix gives the format of the text file in which
the coefficients are stored.

9. Now that you have saved the filter specifications, and the filter
coefficients, you can close the application. Select File Close to
close the Classical IIR Design panel. Then select Quit in the Main
Menu to exit the DFD Toolkit.

End of Exercise 10-1

National Instruments Corporation

10-17

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

D.Designing FIR Filters

As opposed to IIR filters, whose output depends on both its inputs and
outputs, the output of an FIR filter depends only on its inputs. Because the
current output is independent of past outputs, its impulse response is of
finite length. The output of a general FIR filter is given by
y[i] = b0x[i] + b1x[i-1] + b2x[i-2] + ........ + bNx[i-N]

(4)

where N is the order of the filter and b0, b1, .... bN, are its coefficients.
FIR filters have certain advantages as compared to IIR filters.

They can achieve linear phase response, and hence they can pass a
signal without phase distortion.

They are always stable. During filter design or development, you do

not need to worry about stability concerns.

FIR filters are simpler and easier to implement.

Applications of FIR filters

Many applications require the filters to be linear phase. In this case, you
should use FIR filters. However, FIR filters generally need to be of a
higher order than IIR filters, to achieve the same magnitude response
characteristics. So, if linear phase is not necessary, but speed is an important
consideration, you can use IIR filters instead.
In applications where a signal needs to be reconstructed after it has been
split up into several frequency bands, it is important that the filter is
linear phase. In the reconstruction process, the loss of phase could result
in the reconstructed signal being quite different from the original one.
An example of such an application is in the Wavelet and Filter Banks
Design toolkit, where you reconstruct the original signal (either a 1D
waveform or a 2D image) from the wavelet coefficients. The filters
used are FIR filters. The applications are in areas where phase
information is an important consideration, such as noise removal, data
compression, etc.

LabVIEW Signal Processing Course Manual

10-18

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

Classical FIR Design and Arbitrary FIR Design

Using the DFD Toolkit, you can design FIR filters in the Classical FIR
Design panel (shown below) which is very similar to the Classical IIR
Design panel that you have seen.

The panel includes a graphical interface with the Magnitude vs.

Frequency graph and cursors on the left side, and a text-based interface
with digital controls on the right side. The differences are the absence of
the design control in the text entry box and the addition of the minimize
filter order control below the text entry box.

This button controls whether the DFD application minimizes the

estimated filter order. If this button is OFF, the DFD application uses a
fast formula to estimate the filter order to meet or exceed the desired
filter specifications. If this button is ON, the DFD application iteratively
adjusts the filter order until it finds the minimum order that meets or
exceeds the filter specifications.
The FIR filters that are designed use the Parks-McClellan equiripple FIR
filter design algorithm and include the lowpass, highpass, bandpass, and
bandstop types. The Parks-McClellan algorithm minimizes the
difference between the desired and actual filter response across the
entire frequency range.

National Instruments Corporation

10-19

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

You can also design FIR filters with an arbitrary frequency response by
selecting Arbitrary FIR Design in the Main Menu.

Arbitrary means that you can specify exactly what the magnitude of the
filter response should be at specific frequencies. In the next exercise, you
will design an FIR filter by specifying an arbitrary frequency response.

LabVIEW Signal Processing Course Manual

10-20

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

Exercise 10-2
Objective:

To design an FIR filter which filters the data according to A-weighting.

The human sense of hearing responds differently to different frequencies
and does not perceive sound equally. Certain filters are used to filter the
sound applied at their input such that they mimic the human hearing
response to audio signals. An application of this is in third-octave analyzers,
where, to mimic the response of the human ear, the analyzer output is
weighted according to the table shown below:
Frequency (Hz)

Weighting (dB)

Frequency (Hz)

Weighting (dB)

10

-70.4

500

-3.2

12.5

-63.4

630

-1.9

16

-56.7

800

-0.8

20

-50.5

1000

25

-44.7

1250

+0.6

31.5

-39.4

1600

+1.0

40

-34.6

2000

+1.2

50

-30.3

2500

+1.3

63

-26.2

3150

+1.2

80

-22.5

4000

+1.0

100

-19.1

5000

+0.5

125

-16.1

6300

-0.1

160

-13.4

8000

-1.1

200

-10.9

10000

-2.5

250

-8.6

12500

-4.3

315

-6.6

400

-4.8

This type of weighting is known as A-weighting.

National Instruments Corporation

10-21

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

1. Launch the DFD Toolkit.

2. Select Arbitrary FIR Design in the Main Menu. You will access
the following design panel:

The panel includes a graphical interface with the Magnitude vs.

Frequency graph on the left side and a text-based interface with digital
controls on the right side. In the array on the right hand side, you can
enter or modify the array magnitude response points (frequency and
magnitude). From these points, the DFD application forms a desired
magnitude response that covers the entire frequency range from 0.0 to
half the sampling rate. The DFD application then takes this desired
response, along with the filter order, and uses the Parks-McClellan
algorithm to design an optimal equiripple FIR filter.

LabVIEW Signal Processing Course Manual

10-22

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

The graph below plots the desired and actual magnitude response of the
designed FIR filter.

Figure 10-1. Desired and Actual Magnitude Response

The y-axis is in linear or decibel units, depending on how you set the
button in the upper left corner of the graph. The x-axis is in Hertz. The
full scale ranges from 0.0 to Nyquist (sampling rate/2).
dB

The dB button controls the display units (linear or decibel) of all the
magnitude controls and displays. These controls and displays include the
Magnitude vs. Frequency graph (y-axis) and the magnitudes in the array
of frequency-magnitude points.
The following array is the array of frequency-magnitude points the DFD
application uses to construct the desired filter magnitude response. The
DFD application forms the desired filter response by interpolating
between these points.

National Instruments Corporation

10-23

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

The frequency of each point is in Hertz and the magnitude is in linear or

decibel units of gain, depending on the setting of the button in the upper
left corner of the Arbitrary Magnitude Response graph.
You can select points in this array by clicking in the circle to the right of
each point. You can then delete the selected points by clicking on the
delete button, or move them by clicking on the desired direction
diamond in the lower right corner of the Arbitrary Magnitude Response
graph.
The # points control specifies the total number of frequency-magnitude
points the DFD application uses to create the desired filter magnitude
response.
# points
10

Reducing this number deletes points from the end of the

frequency-magnitude array, while increasing this number, inserts the
additional number of points to the right of the selected point.
If you want to select more than one frequency-magnitude point on the
response graph, you should set the multiple selection button to ON.

LabVIEW Signal Processing Course Manual

10-24

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

Clicking on a point you already selected removes that point from the
selection list.

The interpolation control selects the type of interpolation the DFD

application uses to generate the desired response from the array of
frequency-magnitude points.
linear interpolation

Choose linear interpolation to create flat filters (lowpass, highpass,

bandpass, and bandstop). Choose spline interpolation to create
smoothly-varying filters.

ins

To insert a frequency-magnitude point between the selected point and

the next point, click on the ins button.
If the selected point is the last point in the frequency-magnitude array,
the DFD application inserts the new point between the last two points of
the array.
The DFD application inserts new points at halfway along the line
connecting the two outer points.

del

To delete the selected frequency-magnitude points, click on the del

button. The DFD application deletes all selected points.
These points are the selected frequency-magnitude points. You can
select points on the Arbitrary Magnitude Response graph by clicking on
the point, or directly from the frequency-magnitude array shown at the
right by clicking on the circle to the right of each point.

National Instruments Corporation

10-25

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

The filter order control specifies the total number of coefficients in the
digital FIR filter.

The ripple indicator displays the largest absolute error (linear) between
the desired and actual filter responses.
ripple 1.3709E-2

The message window displays errors that occurred during the FIR design
procedure.
message

The locked frequencies box allows you to lock the present frequency
values of the frequency-magnitude points. If you click in this box, you
can alter only the magnitude or y-value of the frequency-magnitude
points.

The uniform spacing box is used to space the frequency values of the
frequency-magnitude points. If you click in this box, the DFD
application spaces the frequency-magnitude points uniformly from 0.0
to sampling rate/2, inclusive.
Clicking in the sort by frequency box tells the DFD application to sort
the frequency-magnitude points in both the response graph and the array
according to ascending frequency. The value of each frequencymagnitude point remains unchanged; however, the point order may
change.
Clicking on import from file enables you to import frequency-magnitude
points from a text file.

LabVIEW Signal Processing Course Manual

10-26

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

The sampling rate control specifies the sampling rate in samples per
second (Hertz).
sampling rate

8000.00

3. The filter specifications shown in the table on page 10-21 have been
saved in the Dfd\Aweight.fir file. Load these specifications by
selecting DFD Menu Load Spec... and choosing the
Aweight.fir file.
4. You could also have entered these values directly on the front panel
and then saved them by selecting Save Specs... from the DFD Menu.

When prompted for a file name, type Aweight.fir.

5. With the multiple selection control set to OFF, move some of the
points on the Arbitrary Magnitude Response graph. To do this,
move your cursor close to a point till the cursor changes shape, as
shown below:

Hold down the left mouse button and move the point. Observe how
the response of the filter changes as the point is moved.
6. Choose locked frequencies and try to do the same as in the previous
step. Now you should be able to change the magnitude of the selected
point, but not its frequency.

National Instruments Corporation

10-27

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

7. Change the methods of interpolation between linear interpolation and

spline interpolation. Observe the difference in the shape of the filter
response. Linear interpolation is used to create flat filters, such as
lowpass, highpass, bandpass, and bandstop filters. Spline
interpolation is used to create smoothly-varying filters.
8. Close the Arbitrary FIR Design panel by selecting Close from the
File menu.
9. Quit the application by selecting Quit from the Main Menu.

End of Exercise 10-2

LabVIEW Signal Processing Course Manual

10-28

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

E. Analyzing your Filter Design

After designing your IIR or FIR filter, you can analyze your design in
several ways. For example, you can see the effect that the filter has on the
amplitude and phase of input signals at different frequencies by observing
its magnitude and phase responses. Several types of analysis methods are
available in the Digital Filter Design Toolkit. These are explained below.
You will first look at the response of the filter to special kinds of input
signals.

Impulse Response
The output of the filter when the input is an impulse is known as the impulse
response of the filter. An impulse in the digital world has an amplitude of 1
at index 0 and an amplitude of 0 for all other indices. An impulse and the
impulse response are shown in the figures below.

The impulse response has a very special meaning in the case of FIR
filters. The impulse response of an FIR filter gives the coefficients of
that filter. Thus, the impulse response is a useful method for determining
the coefficients of an FIR filter. Furthermore, the number of nonzero
terms in the impulse response gives the number of coefficients in the
filter. (For an IIR filter, the relationship is much more complicated, and
the above discussion does not apply.)
Another use of the impulse response is that, for both FIR and IIR filters,
the output of the filter is given by the convolution of the input signal and
the impulse response of the filter.

Step Response
The output of the filter when the input is a unit step is known as the step
response of the filter. A unit step in the digital world has an amplitude of 0

National Instruments Corporation

10-29

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

for all negative indices, and an amplitude of 1 at index zero and for all
positive indices. A unit step and the step response are shown in the figures
below.

The step response is important if you will use the filter in a control
system. You can then see how the parameters of the control system (such
as the rise time, overshoot, etc.) are affected by the filter. The step
response also shows you how long the filter will take to respond to a
sudden change in the input.

Frequency Response (Magnitude Response and Phase Response)

The frequency response in useful in that it shows the effect that the filter has
on the amplitude and phase of input signals at different frequencies.
Because the filter can affect both the magnitude and the phase of the input
signal, the frequency response consists of two partsthe magnitude
response and the phase response. An example of these responses is shown
below. The x-axis units are normalized in terms of the sampling frequency.
Magnitude(dB)
(dB)
Magnitude

1.20
1.00
0.80
0.60
0.40
0.20

f (Hz)

0.00
0.00

LabVIEW Signal Processing Course Manual

0.10

10-30

0.20

0.30

0.40

0.50

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

Phase(radians)
(radians)
Phase

0.0
-2.0
-4.0
-6.0
-8.0
-10.0

f (Hz)
-12.0
0.00

0.10

0.20

0.30

0.40

0.50

Not only does the frequency response enlighten you as to the effect that
the filter has on signals of specific frequencies, it also allows you to
determine what happens to arbitrary signals as they pass through the
filter. Because most signals can be expressed as a sum of exponentials
(sines and cosines), you can break a signal down into its individual
components and determine the effect of the filter on those components.

The Z-Domain: Transfer Function H(z) and the Pole-Zero Plot

The transfer function, H(z), of a digital filter can be expressed as a ratio of
polynomials,
N(z)
H ( z ) = ----------D(z)

where N(z) is a numerator polynomial, and D(z) is the denominator

polynomial. (For an FIR filter, D(z) = 1.) H(z) is also known as the
z-transform of the filter.
The values of z at which N(z) is equal to zero are known as the zeros of
the filter, because for these values, H(z) is also equal to zero. The values
of z at which D(z) is equal to zero are known as the poles of the filter,
because at these values, H(z) is equal to infinity. A plot of the poles and
zeros of the filter is known as the pole-zero plot. Because z is a complex
number, the pole-zero plot is shown in terms of the real part of z on the
x-axis and the imaginary part of z on the y-axis.
The pole-zero plot is useful in determining the stability of the filter. As
long as all the poles of the filter have a magnitude less than one, the filter
is stable. If any of the poles of the filter have a magnitude greater than
one, the filter will be unstable. That means that the output of the filter
will continue to grow indefinitely even if the input is no longer applied.
The values of z for which its magnitude is equal to one is drawn on the
pole-zero plot as a circle having its center at the origin and having radius
equal to one. Thus, so long as the poles of the filter lie inside the circle,
the filter will be stable. If even one of the poles lie outside the circle, the
filter will be unstable.

National Instruments Corporation

10-31

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

The Digital Filter Design Toolkit gives you both the transfer function of
the filter and its corresponding pole-zero plot. The figures below show
the pole-zero plots for both a stable and an unstable filter. Each o depicts
a zero and each x depicts a pole.

1.0

1.0
x
x

o
0.0

0.0
o
o
x

-1.0
-1.0

0.0

x
1.0

stable filter

LabVIEW Signal Processing Course Manual

-1.0
-1.0

0.0

1.0

unstable filter

10-32

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

Exercise 10-3
Objective:

To analyze the design of an IIR filter.

You will load the filter specifications you saved in Exercise 10-1 in the file
bandpass.iir and see its impulse and step responses, and the pole-zero
plot of its transfer function.
1. Open the DFD application.
2. Choose the Classical IIR Design panel.
3. When the Classical IIR Design panel opens, select Load Spec... from
the DFD Menu.

When prompted for the filename, select bandpass.iir.

4. From the DFD menu, select Analysis.

The Analysis of Filter Design window opens, as shown below.

National Instruments Corporation

10-33

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

In this panel, you can view the filter magnitude response, phase
response, impulse response, step response, and pole-zero plot of the
filter you designed in the first exercise. You can also view and print
full-screen plots of each response. From the full-screen views, you can
save the analysis results to text files.
If you select DFD Menu Analysis from a filter design panel, the
Analysis of Filter Design panel uses that particular filter design to
compute the various filter responses. You can also analyze any of the
four filter designs from the Design Analyzed ring selector; the Analysis
of Filter Design panel uses the filter parameters from the selected filter
design.
The DFD Menu can be used to load filter designs from previous work,
open the DAQ and Filter panel, go to the selected filter design panel, or
return to the Filter Design Main Menu.

LabVIEW Signal Processing Course Manual

10-34

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

The Design Analyzed control selects which filter control to analyze. If

you continue to modify the same filter design that is presently being
analyzed, the DFD will recompute all filter responses.

Analysis Displays

zoom
box

Each of the five filter plots has a zoom box in the upper right corner.
Clicking in this box brings up a full-screen version of that plot. In the
full-screen versions of these plots, you can change the units from linear
to decibel (magnitude response), from radians to degrees (phase
response) or from seconds to samples (impulse and step responses).
From each full-screen view, you can save the response data to text files.

Magnitude Response
5. The magnitude response is the magnitude of the filters response
H(f) as frequency varies from zero to half the sampling rate. Look at
the magnitude response of the designed filter. You can see that it is
indeed a bandpass filter.

Phase Response
6. The phase response is the phase of the filters response H(f) as
frequency varies from zero to the sampling rate. The following figure
illustrates the phase response of the selected filter design. Note that
the phase is displayed in radians. You can obtain a display in degrees
by clicking on the zoom box of the Phase Response plot. A new
window will appear, which will give you the option of choosing
between the appropriate display units.

National Instruments Corporation

10-35

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

Impulse Response
7. The impulse response of a digital filter is the filters output when the
input is a unit sample sequence (1, 0, 0, ...). The input before the
unity sample is also zero. The following figure shows the impulse
response of the selected filter design.

Observe that although it is an IIR filter, the impulse response decays

toward zero after a finite amount of time.

Step Response
8. The step response of a digital filter is the filters output when the
input is a unit step sequence (1, 1, 1, ...). The input samples before
the step sequence are defined as zero. The following figure shows the
step response of the designed filter.

Z-Plane Plot
9. The following figure illustrates the z-plane plot of the filter poles and
zeros.

Each pole is represented by a red x. Each zero is represented by a

blue o. Compare this with the Magnitude Response of the filter and
observe that the zeros are at 0 and , which correspond to frequencies
0 and Nyquist (fs/2). The location of the poles is about 1000 Hz,
which is why there is a peak in the magnitude response around
1000 Hz.

LabVIEW Signal Processing Course Manual

10-36

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

H(z) for IIR Filters

10. H(z) is the z-transform of the designed digital filter.

Previously, you saw that the IIR filters are implemented as cascaded
second-order stages. For an IIR filter, H(z) can be represented by a
product of fractions of second-order z polynomials.
Ns

H(z) =

Nk ( z )

------------Dk ( z )
k=1

Nk(z) = numerator for stage k

Dk(z) = denominator for stage k
Ns = number of second-order stages
You can view the N(z) and D(z) polynomials for other stages by
incrementing the index shown in the upper left corner of the H(z)
display.
11. Close the panel by selecting Close from the File menu. Quit the DFD
application.

End of Exercise 10-3

National Instruments Corporation

10-37

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

F. Format of Filter Coefficient Text Files

When you save your filter coefficients to a text file, the DFD application
generates a readable text file containing all the information you need to
implement the designed FIR or IIR digital filter. This section details the
format for both FIR and IIR filter coefficient files.

FIR Coefficient File Format

The following table gives an example FIR coefficient text file and
description:
coefficient file example

description

FIR Filter Coefficients

type of file

Sampling Rate

sampling rate label

8.000000E+3

sampling rate in Hz

filter order label

22

filter order

h[0..21]

coefficients label

6.350871E-3

1st coefficient, h[0]

-8.833535E-3

2nd coefficient, h[1]

-2.847674E-2

4.626607E-2

4.103986E-2

-1.114579E-1
-1.412791E-2
1.810791E-1
-5.984635E-2
-2.002337E-1
1.516199E-1
1.516199E-1
-2.002337E-1
-5.984635E-2
1.810791E-1

LabVIEW Signal Processing Course Manual

10-38

National Instruments Corporation

Lesson 10

coefficient file example

Digital Filter Design Toolkit

description

-1.412791E-2
-1.114579E-1
4.103986E-2
4.626607E-2

-2.847674E-2

-8.833535E-3

6.350871E-3

last coefficient, h[N-1]

You can implement the FIR filter using equation (4) directly.

IIR Coefficient File Format

IIR coefficient files are slightly more complex than FIR coefficient files. IIR
filters are usually described by two sets of coefficients, a and b coefficients.
There are a total of M * S a coefficients and (M+1)*S b coefficients, where
M is the stage order (usually 2) and S is the number of stages. An IIR filter
with three second-order stages has two a coefficients per stage for a total of
six a coefficients, and three b coefficients per stage for a total of nine b
coefficients.
The following table gives an example IIR coefficient text file and
description:
coefficient file example

National Instruments Corporation

description

IIR Filter Coefficients

coefficient type

Sampling Rate

sampling rate label

8.000000E+3

sampling rate in Hz

Stage Order

stage order label

order of each stage

Number of Stages

number of stages label

number of stages

a Coefficients

a coefficients label

number of a coefficients

3.801467E-1

a1 for stage 1

10-39

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

coefficient file example

description

8.754090E-1

a2 for stage 1

-1.021050E-1

a1 for stage 2

9.492741E-1

a2 for stage 2

8.460304E-1

a1 for stage 3

9.540986E-1

a2 for stage 3

b Coefficients

b coefficients label

number of b coefficients

1.514603E-2

b0 for stage 1

0.000000E+0

b1 for stage 1

1.514603E-2

b2 for stage 1

1.000000E+0

b0 for stage 2

6.618322E-1

b1 for stage 2

1.000000E+0

b2 for stage 2

1.000000E+0

b0 for stage 3

1.276187E+0

b1 for stage 3

1.000000E+0

b2 for stage 3

You can implement the IIR filter in cascade stages by using equation (2)
(maintaining two past inputs and two past outputs for each stage), or by
using the direct form II equations (maintaining two past internal states),
as in equation (3).

LabVIEW Signal Processing Course Manual

10-40

National Instruments Corporation

Lesson 10

Digital Filter Design Toolkit

Summary
You have learned that FIR filters are used for applications where you need
a linear phase response, such as for applications that require reconstructing
the original waveform after filtering, noise removal, and data compression.
For applications where phase is not an important consideration (such as for
simple signal monitoring) and where faster speeds are necessary, you can
use IIR filters.
You saw that the DFD toolkit allows you to interactively design both
FIR and IIR filters. The design could be done either by specifying the
filter parameters (classical IIR design and classical FIR design),
deciding the location of the poles and zeros in the z-plane (pole-zero
placement), or arbitrarily specifying the magnitude response
characteristics (arbitrary FIR design) of the filter.
After the filter has been designed, you can analyze your filter in terms
of its magnitude and phase responses, impulse and step responses, and
the pole-zero plot. You can also save the filter coefficients for use in
other applications.

National Instruments Corporation

10-41

LabVIEW Signal Processing Course Manual

Lesson 10

Digital Filter Design Toolkit

Notes

LabVIEW Signal Processing Course Manual

10-42

National Instruments Corporation

Lesson 11
G Math Toolkit
Introduction
The G Math Toolkit offers a new paradigm for mathematics, numerical
recipes in G, with hundreds of math VIs for solving differential equations,
optimization, root finding, and so on. All VIs in the G Math Toolkit are
written in G, so you can quickly modify them for your custom applications.
A main feature of the G Math Toolkit is that it adds to LabVIEW and
BridgeVIEW the ability to enter complex formulas directly onto the front
panel of a VI.
This toolkit is intended for use by scientists, engineers, and
mathematicians, and by anyone needing to solve mathematical problems
in a simple, quick, and efficient manner. It can also be used as an
educational aid by those interested in expanding their knowledge of
mathematics.

You Will Learn:

A. About the organization of the G Math Toolkit.
B. About the different types of parser VIs and how to use them to parse
formulas entered directly on the front panel.
C. About solving differential equations using the differential
equation VIs.

National Instruments Corporation

11-1

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

A. Organization of the G Math Toolkit

The G Math Toolkit consists of nine libraries, each specifically suited to
solve problems in a particular area of mathematics. These libraries are:
Parser.llb: Consists of the VIs that act as an interface
between the end user and the programming system. These VIs
parse the user-given formula and convert it to a form that can be
used for evaluating the results.
Visualiz.llb: These are the Data Visualization VIs for
plotting and visualizing data in both 2D and 3D. They include
advanced methods such as animation, contour plots, and
surface cuts.
Ode.llb: The VIs in this library solve ordinary differential
equations VIs, both numerically and symbolically.
Zero.llb: Used for finding the zeros of 1D or nD, linear or
nonlinear functions (or system of functions).
Opti.llb: The optimization VIs that determine local minima
and maxima of real 1D or nD functions. You can choose between
optimization algorithms based on derivatives of the function and
others working without these derivatives.
1Dexplo.llb: Contains VIs that allow the study of real-valued
1D functions, with and without additional parameters, given in
symbolic form.
2Dexplo.llb: A collection of VIs that deliver information
about 2D functions given in symbolic form, where
parameterization is allowed. Extrema (minima and maxima) and
partial derivatives can be numerically calculated.
Function.llb: These VIs evaluate some common
mathematical functions.
Trans.llb: A group of VIs that implement some transforms
commonly used in mathematics and signal processing.
The G Math libraries consist of more than 100 VIs you can use for
solving your mathematics problems. In the next section, you will
concentrate on learning more about the parser VIs and use some of them
to build a simple arbitrary waveform generator. The importance of the
VIs in the parser library is in enabling users to enter formulas directly
on the front panel. Thus, the user can enter a formula on the front
panel and the arbitrary waveform generator will generate and plot the
corresponding signal.

LabVIEW Signal Processing Course Manual

11-2

National Instruments Corporation

Lesson 11

G Math Toolkit

B. Parser VIs
The VIs in the Parser library act as the interface between the user and the
VIs in the other libraries. The formulas entered on the front panel can have
any number of variables. The formulas are first parsed to determine the
variables and the values to be assigned to them. They are then evaluated to
a number by substituting numeric values for the variables.

Direct and Indirect Forms

Because there are basically two steps involved in this process (parsing and
then evaluation), there are two forms of parser VIsthe direct form and the
indirect form. In the direct form, both the parsing and evaluation are done
in the same VI. In the indirect form, the parsing and evaluation are done in
separate VIs.
As an example, the direct version of the Eval Formula Node VI can be
represented as in the following block diagram.

On the other hand, the indirect forms split the VI in two subVIs, as
shown in the following illustration. You can use the indirect form in
larger applications, where a two-step process (parsing and then
evaluating) is more efficient.

National Instruments Corporation

11-3

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

The following table summarizes the different types of parser VIs.

Direct Form (Parsing and evaluation done in one VI)

Example

No Variables

Eval Formula String

Evaluates RHS of a formula

without variables.

sin(1.2) + 5

Single Variable

Eval Single-Variable
Scalar

Evaluates RHS of a formula of

one variable at one specified
point.

cos(x) at x = 3.142

Eval Single-Variable
Array

Evaluates RHS of a formula of

one variable at several specified
points.

cos(x) at
x = 0, 0.1, 0.2, ...

Eval Formula Node

Evaluates both sides of

formula(s) with several variables.

x = a + b at a = 1, b = 2

Eval Multi-Variable
Scalar

Evaluates RHS of a formula with

several variables at one specified
point.

sin(x) + cos(y) + z
at x = 1.0, y = 2.5,
z = 3.1

Eval Multi-Variable
Array

Evaluates RHS of a formula with

several variables at several
specified points.

sin(x) + cos(y) + z
at x = 1, y = 2, z = 3 and
x = 3, y = 10, z = 1 and
x = 0.1, y = -2, z = 0.0

Many Variables

Indirect Form (Parsing and evaluation in separate VIs)

Many Variables (The
combination evaluates
the RHS of a formula of
several variables at the
specified point.)

Parse Formula String

Parses the RHS of a formula to

determine input variables and the
operations performed on them.

Eval Parsed Formula

String

Evaluates the parsed formula

with the specified values for the
input variables.

Many Variables (The

combination evaluates
both sides of formula(s)
of several variables at
the specified point.)

Parse Formula Node

Parses both sides of the

formula(s) to determine the input
and output variables and the
operations to be performed on
them.

Eval Parsed Formula

Node

Evaluates the parsed formula(s)

with the specified values for the
input variables.

Substitute Variables

Substitutes specified formulas for

variables in the main formula.

x = sin(z) + 7 * y

x=a+b
y=a*b

Others

Note:

The front panel of each parser VI has an example that shows how to enter
values in the control inputs.

LabVIEW Signal Processing Course Manual

11-4

National Instruments Corporation

Lesson 11

G Math Toolkit

The direct form of parser VIs are more widely used than the indirect
form. The following flowchart guides you through the selection of a
specific parser VI of the direct form.

Parsing and Evaluation

in the same VI?

Select one of
the indirect form
Parser VIs

one

Evaluation
at single or
multiple points?

single

Eval Single-Variable Scalar

How many
variables?

multiple

many

Evaluate both
sides of formula?

Y
Eval
Formula Node

Eval Single-Variable Array

N

single

Evaluation
at single or
multiple points?

Eval Multi-Variable Scalar

multiple

Eval Multi-Variable Array

Comparison with Formula Node

A parser VI scans an input string and interprets this string as a collection of
formulas. Then, the parser VI replaces the formulas with numeric
calculations and outputs the results.The parser VI routines deal only with
real numbers. There are some differences between the parser in the G Math
Toolkit and the Formula Node found in the original LabVIEW (or
BridgeVIEW) package. The following table outlines these differences.
Meaning

Variables

Formula Node

No restrictions

Parser VI Routines
(G Math Toolkit)
Only a, a0, ..., a9, ...
z, z0, ..., z9, are valid

National Instruments Corporation

11-5

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

Meaning

Formula Node

Parser VI Routines
(G Math Toolkit)

Binary functions

Max, min, mod, rem

Not available

More complex math

functions

Not available

Gamma, ci, si, spike, step,

square

Assignment

Not available

Logical, conditional,
inequality, equality

?:, ||, &&, !=, ==, <, >, <=,

>=

Not available

pi

pi ( 1 ) = ,pi ( 2 ) = 2

The precedence of operators is the same for the G Math Toolkit parser
VIs as those of the formula nodes in LabVIEW and BridgeVIEW.

Error Structure
The parser VIs use the following error handling structure. This structure
consists of a Boolean status button, a signed 32-bit integer numeric code
indicator, and a string source indicator. These error handler components are
explained below:
status is TRUE if an error occurred. If status is TRUE, this VI does not
perform any operations.
code is the error code number identifying the error.
source explains the error in more detail.
The default status of the error in structure is FALSE (no error),
indicated by an error code of 0.

With this structure, an application can programmatically decide the

accuracy of formulas and control the data flow in case of errors. The
application uses the source field of the error handling structure as a
storage for a wrong formula input. This field displays limited error

LabVIEW Signal Processing Course Manual

11-6

National Instruments Corporation

Lesson 11

G Math Toolkit

descriptions if an error is detected in your program. See the Error Codes

appendix for the error codes and the messages of the parser VI routines.

Functions Available for Use with Parser VIs

Most of the functions that you can use in the formula node can also be used
in the parser VIs. However, there are some differences. For a complete list
of functions that you can use with the parser VIs, refer to the appendix.
The parser VIs are extremely powerful and can be used in a wide variety
of applications. The first few exercises in this lesson cover building an
arbitrary waveform generator that uses several of the parser routines and
to understand their functionality.

National Instruments Corporation

11-7

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

Exercise 11-1
Objective:

To build an arbitrary waveform generator using the Eval Single-Variable Array VI.
In this exercise, you will build an arbitrary waveform generator using the
parser VIs. You will enter the formula for the waveform on the front panel
and see the output waveform on a graph indicator.

Selection of Direct or Indirect Form Parser VI

First, you must decide whether to use the direct or the indirect form of the
parser VIs. Because you are interested only in displaying the result of the
evaluation on a graph, and are not concerned about the results of parsing,
you will use the direct form of the parser VIs.

Selection of Parser VI
Looking at the table on page 11-4, depending on the number of variables in
your formula, you need to choose the appropriate VI from several available
choices of VIs of the direct form. Because you will plot a 1D function, you
need a VI that can handle functions of at least one variable. And because you
want to evaluate that function at more than one point, choose the Eval
Single-Variable Array VI (G Math Parser subpalette). The inputs and
outputs of this VI are shown below.

formula: a control for specifying your mathematical formula consisting

of one variable.
X Values: values of the variable at which the formula is evaluated.
Y Values: numeric results of the evaluation of the formula in formula at the
values specified in X Values.
You need to specify at least the inputs formula and X Values. For the
formula input, you will enter a control on the front panel. On this
control, you can type in your input formula. For the X Values, you will
build a For Loop on the block diagram that will generate 1000 points at
which to evaluate the equation you will type in the formula control. The
corresponding front panel and block diagram are shown below. Note that
the For Loop generates X Values from 0 to 10 in increments of 0.01.

LabVIEW Signal Processing Course Manual

11-8

National Instruments Corporation

Lesson 11

G Math Toolkit

1. Build the VI front panel and block diagram shown below.

Front Panel

The formula string control is where you enter your formula.

The Y Values indicator shows the result of the calculation of the formula
at specified discrete points.
The Waveform graph shows you a plot of the function you typed in the
formula control.

Block Diagram

Eval Single-Variable Array VI (G Math Parser subpalette)

evaluates a function of a single variable at multiple points.
The For Loop generates the discrete set of points at which the formula
in the formula control will be evaluated. These points range from 0 to
9.99 in increments of 0.01.
2. After building the above VI, return to the front panel.
3. Type in the equation sin(x) in the formula control and run the VI. See
the result displayed on the graph indicator.

National Instruments Corporation

11-9

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

Note that sin(x) has only one variable, called x. You could also call
the variable y or z or anything else. The appendix lists the rules for
variable names.
4. Change the equation to sin(x) + cos(x/2) and run the VI.
5. Change the equation to step(y-1) + sinc(y-2) + sin(y*10) and run the
VI. Note that now you have changed the variable to y.
6. Type in any other equation in the formula control and run the VI. The
appendix contains a list of available functions you can use in the
equation.
Note:

If you get an error, refer to the appendix for a list of error codes.
7. Save the VI as EvalSVA.vi in the library Lvspcex.llb.
8. Close the VI.
You have now seen how easy it is to build the waveform generator where
the user can specify the formula for an arbitrary waveform on the front
panel. In the next exercise, you will modify this waveform generator to
have the functionality of easily specifying even more complex
waveforms.

End of Exercise 11-1

LabVIEW Signal Processing Course Manual

11-10

National Instruments Corporation

Lesson 11

G Math Toolkit

Exercise 11-2
Objective:

To generate even more complex waveforms by incorporating the Substitute

Variables VI in the function generator.
Another useful VI in the parser library is the Substitute Variables VI (G
Math Parser subpalette). The connections to this VI are shown in the
figure below. It is used to substitute formulas for parameters that are already
defined in formulas.

original formula: the main formula you type in.

Substitution Rules: specify the substitutions to be made for the parameters
in original formula.
formula after substitution: the resulting formula after the parameter
substitutions specified in Substitution Rules.
You will now modify the waveform generator you built in the previous
exercise to use the Substitute Variables VI. Suppose you want to
generate a generic waveform that is of the form sin(A) + cos(B), where
A and B can themselves be functions like sin(x), cos(x), square(x),
sinc(x), ln(x), etc. But you want to use different functions for A and B
each time you run the VI. This is accomplished as shown in the front
panel and block diagrams below.

National Instruments Corporation

11-11

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

Substitute Variables VI (G Math Parser subpalette)

substitutes for parameters defined in formulas. The substitution
is done according to the rules specified in the Substitution
Rules cluster array.
Eval Single-Variable Array VI (G Math Parser subpalette)
evaluates a function of one variable at specified multiple points.
The For Loop generates a thousand points, ranging from 0 to 9.99 in
steps of 0.01, at which to evaluate the user-specified formula.

The original formula string control is where you enter your formula of
one variable. The formula may have one or more parameters.
Each element (cluster) in the Substitution Rules array specifies a
parameter and its corresponding substitution.
Y Values is an array digital indicator that shows the result of calculating
the formula at specified points.
1. Build the VI as shown in the figures above. The formula after
substitution output of the Substitute Variables VI goes to the
formula input of the Eval Single-Variable Array VI.
Hint: Build the block diagram first, pop up to create the controls and
indicators, and then add the waveform graph.
2. In the original formula control, type in the formula sin(A) + cos(B).
This will be the generic formula. In the Substitution Rules control,
you will provide the user with the capability of specifying the
functions to be substituted for A and B.

LabVIEW Signal Processing Course Manual

11-12

National Instruments Corporation

Lesson 11

G Math Toolkit

3. In element 0 of the Substitution Rules control, type in

parameter name:

parameter content:

sin(x)

In element 1, type in
parameter name:

parameter content:

cos(x)

Note that this is the same as if you had typed sin(sin(x)) + cos(cos(x))
in the formula control in Exercise 11-1.
4. Run the VI and see the waveform on the graph display.
5. Change the values in the Substitution Rules control to
element 0
element 1

parameter name:

parameter content:

sin(x)

parameter name:

parameter content:

square(x)

Run the VI. The resulting equation is the same as if you had typed
sin(sin(x)) + cos(square(x)) in the formula control in Exercise 1-1.
6. Now change the control to
element 0
element 1

parameter name:

parameter content:

step(x-5)

parameter name:

parameter content:

square(x)

Run the VI.

With this simple VI that you have built, you now can generate any
type of waveform that you so desire, with the functions given in the
appendix.
7. Save the VI as ESVA_SV.vi in the library Lvspcex.llb.
8. Close the VI.

End of Exercise 11-2

National Instruments Corporation

11-13

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

C. Solving Differential Equations

You can use G Math to solve both ordinary or partial, linear or nonlinear,
and either first or higher order differential equations. G Math has seven VIs
for solving various types of differential equations. These VIs are listed
below, classified according to the order of the differential equation they can
solve.
Note:

The order of a differential equation is the order of the highest derivative in

the differential equation.
The first five VIs can solve a set (one or more) of first-order differential
equations. The next two VIs are for solving higher order differential
equations. Note that you can convert a higher order differential equation
into a set of first-order differential equations (later in this lesson, you will
see an example of how to do this).
VIs for solving a set (one or more) of first-order differential equations:
ODE Cash Karp 5th order for solving differential equations
using the Cash Karp method.
ODE Euler Method for solving differential equations using the
Euler method.
ODE Runge Kutta 4th order for solving differential equations
using the Runge Kutta method.
ODE Linear System Numeric for numerical solution of a linear
system of differential equations.
ODE Linear System Symbolic for symbolic solution of a linear
system of differential equations.
Out of these, the first three are for solving nonhomogeneous
(right side 0) differential equations, whereas the last two are for
homogeneous (right side = 0) differential equations.
VIs for solving higher order differential equations:
ODE Linear nth order Numeric for numeric solution of a linear
system of nth order differential equations.
ODE Linear nth order Symbolic for symbolic solution of a
linear system of nth order differential equations.
Both these VIs are for solving homogeneous differential equations.

LabVIEW Signal Processing Course Manual

11-14

National Instruments Corporation

Lesson 11

G Math Toolkit

Solving Nonhomogeneous Differential Equations

From the previous discussion, you see that there are three VIs available for
solving nonhomogeneous differential equations. They are

ODE Cash Karp 5th order VI

ODE Euler Method VI

ODE Runge Kutta 4th order VI

Each VI employs a different method for solving the differential

equations. Each method uses a parameter known as the step size
(denoted by h) that determines the spacing between points at which the
solution is evaluated. This step size is a constant for the last two VIs
(which employ the Euler and Runge-Kutta methods), whereas it is
variable (it automatically adapts itself to the solution) for the first VI
(which employs the Cash Karp method). The Euler method is the
simplest method, but the Runge-Kutta method gives a more accurate
solution.
The question arises as to which of these three VIs you should choose for
your application. The general guidelines are:

Select the ODE Euler Method VI for very simple ODEs.

For all other cases, choose the ODE Runge-Kutta 4th Order VI or
the ODE Cash Karp 5th Order VI.

If you need equidistant points (that is, constant hfor example,

for robot control applications), choose the ODE Runge-Kutta
4th Order VI.

If you are interested in a global solution and fast computation,

choose the ODE Cash Karp 5th Order VI.

A General Class of Second-Order Differential Equations

As an example, a general class of second-order differential equations is
described by the following initial value problem
2

dy
d y
a -------- + b ------ + cy = g ( t )
dt
dt

(1)

dy ( 0 ) dy 0
- = -------with y ( 0 ) = y 0 and ------------dt

dt

where y(0) is the value of y at t = 0, dy(0)/dt is the value of dy/dt at y=0,

and g(t) is known as the forcing function. y(0) and dy(0)/dt are known as
the initial conditions (ICs). Note that because the right side of equation
(1) is not equal to zero, it is a nonhomogeneous differential equation. In
the absence of a forcing function (g(t) = 0), it would be a homogeneous
differential equation.

National Instruments Corporation

11-15

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

Some of the practical applications of this class of differential equations

are for modeling:
1. The motion of a mass on a vibrating spring
2

md y cdy
------------ + --------- + ky = F ( t )
2
dt
dt

where m is the mass, c is the damping coefficient, k is the spring

constant, and F(t) is the applied force.
2. Flow of an electrical current in a series circuit
2

dQ Q dE ( t )
d Q
- + R ------- + ---- = ------------L --------2
dt C
dt
dt

where R, L and C are the resistance, inductance, and capacitance,

respectively, in the circuit, Q is the charge flowing in the series
circuit, and E(t) is the applied voltage.
3. The motion of an oscillating pendulum
2

d c d g
--------2 + ------ ------ + --- sin ( ) = 0
ml dt l
dt

where m is the mass of the pendulum, l is the length of the rod, g is

the acceleration due to gravity, and is the angle between the rod and
a vertical line passing through the point where the rod is fixed (that
is, the equilibrium position).
To solve equations of the type of eqn (1) using the ODE Euler Method
VI, the ODE Runge-Kutta 4th Order VI, or the ODE Cash Karp 5th
Order VI, you first need to convert them into a set of first-order
differential equations. This is achieved by making the substitution
x1 = y and x2 = dy/dt . Substituting these in equation (1), you get
dx 2
a -------- + bx 2 + cx 1 = g ( t )
dx
dx 2 g ( t ) cx 1 bx 2
-------- = -------------------------------------dt
a

Thus, now you have converted the second-order differential equation

given by (1) to the following equivalent set of first order differential
equations
dx 1
-------- = x 2
dt

(2)

dx 2 g ( t ) cx 1 bx 2
-------- = -------------------------------------dt
a

(3)

You will use G Math to solve this set of equations in the next exercise.

LabVIEW Signal Processing Course Manual

11-16

National Instruments Corporation

Lesson 11

G Math Toolkit

Exercise 11-3
Objective:

To build a general G Math VI for applications whose solutions are second-order

differential equations.
In this exercise, you will build a VI that will solve the general second-order
differential equation of the type given in equation (1). This equation is also
given below:
2

d y bdy
a -------2- + --------- + cy = g ( t )
dt
dt
dy ( 0 ) dy 0
- = -------with y ( 0 ) = y 0 and ------------dt

dt

You will see how incorporation of the Substitute Variables VI enables

you to solve problems related to a wide variety of applications.
Because you want to solve the above equation, which is a
nonhomogeneous equation, you are faced with three choices of VIs:

ODE Cash Karp 5th Order

ODE Euler Method

ODE Runge Kutta 4th Order

Because this is a simple second-order differential equation, you can

actually select any of these VIs. Choose the ODE Euler Method VI,
which has the following inputs and outputs:

X: an array of strings listing the dependent variables

time start: the point in time at which to start the calculations
time end: the point in time at which to end the calculations
h: time increment (step rate) at which to perform the calculations
X0: the initial conditions (I.C.). There is a one-to-one relationship
between the components of X and that of X0. That is, the first value
in X0 is the I.C. of the first variable listed in X, the second value in
X0 is the I.C. of the second variable listed in X, and so on.
time: a string that defines what the independent variable is, usually
time (t).

National Instruments Corporation

11-17

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

F(X,t): the right sides of the set of first-order differential equations.

There is a one-to-one relationship between the elements of F(X,t)
and X. This will be explained in more detail later.
Note:

F(X,t) requires its input as a set of first-order differential equations.

Therefore, to solve equation (1) with this VI, you need to use the equivalent
form given by equations (2) and (3).
ticks: is the time in milliseconds required for the calculations
Times: the time instants at which the solution of the differential
equations are evaluated. For the ODE Euler Method VI and the
ODE Runge Kutta 4th Order VI, these instants begin at time start
with increments of h, until time end.
XValues: contains the solution for each of the dependent variables
of the differential equations at each time instant in Times. The first
column is the solution for the first variable in X, the second column
is the solution for the second variable in X, and so on.
error: contains an error code in case of any error. Refer to the
appendix for the list of errors and the corresponding error codes.
1. Build the VI whose front panel and block diagram are shown below.
Hint: An easy way to build the VI is to build the block diagram first, pop
up on the terminals of the ODE Euler Method VI, and select Create
Control for the inputs and Create Indicator for the outputs. The
controls and indicators automatically appear in the correct form on the
front panel. Finally, you can add the Waveform graph.

LabVIEW Signal Processing Course Manual

11-18

National Instruments Corporation

Lesson 11

G Math Toolkit

Front Panel

The time start and the time end controls contain the starting and
ending time instants.
The differential equations are typed in the F(x,t) control.
The dependent variables are entered into the X control.
The independent variable is entered into the time control.
X0 contains the initial values (initial conditions) of the dependent
variables.
h contains the step size.
Substitution Rules contains the parameters and the corresponding
substitution.
Note:

There is a one-to-one correspondence between the elements of X, X0, and

F(x,t). The first element in X0 corresponds to the initial value of the first
element in X, the second element in X0 corresponds to the initial value of
the second element in X, and so on. Also, the first element in F(x,t)
corresponds to the derivative of the first element in X, the second element
in F(x,t) corresponds to the derivative of the first element in X, and so on.

National Instruments Corporation

11-19

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

Block Diagram

ODE Euler Method VI (G Math Ordinary Differential

Equations) solves a set of differential equations using the Euler
method.
Substitute Variables VI (G Math Parser) substitutes the
specified values for the corresponding parameters.
2. Enter the right side of equations (2) and (3), respectively, in the first
two elements of F(X,t). Thus, you will enter x2 and (g - b*x2 -c*x1)/a.
Equation (2) in the second element of F(X,t) contains the variables
a, b, c, and g. The values of these variables will change depending
on the application. Hence, in the block diagram, you use the
Substitute Variables VI to assign different values to these variables.
You can enter these values in the Substitution Rules control on the
front panel.
3. X contains the names of the dependent variables, which in this case
are entered as x1 and x2.
4. time contains the name of the independent variable, which in this
case is t.
X0 will contain the initial conditions for x1 and x2. The first value
entered in X0 will correspond to the first variable entered in X. The
second value entered in X0 will correspond to the second variable
entered in X. The values of the ICs will vary depending on the
application.

LabVIEW Signal Processing Course Manual

11-20

National Instruments Corporation

Lesson 11

G Math Toolkit

5. Save the VI as ODE_2nd.vi in the library Lvspcex.llb.

Now you have a VI ready for solving any second-order
nonhomogeneous differential equation. All that remains is to enter
the corresponding values of a, b, c, and g in the Substitution Rules
control, and the initial conditions of the variables X0 control. Then
you can specify the time interval (between time start and time end,
in increments of h) for which you want the solution and run the VI.
Out pops the answer. You will continue this procedure in the next
exercise.

End of Exercise 11-3

National Instruments Corporation

11-21

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

Exercise 11-4
Objective: To design the suspension system of an automobile.
You will now use the VI you built in the previous exercise to solve a
typical mechanical modeling problem. In particular, you will design the
suspension system of an automobile to have a suitable response when the
automobile goes over a pothole, a speed bump, and so on.
You can model the suspension system by the nonhomogeneous
differential equation given in equation (1) (reproduced below),
2

d y bdy
a -------2- + --------- + cy = g ( t )
dt
dt
dy ( 0 ) dy 0
- = -------with y ( 0 ) = y 0 and ------------dt

dt

where
y(t) is the position of the automobile. y(t) = 0 indicates that the
vehicle is in a balanced or a stable position.
dy/dt is the up and down speed of the automobile.
c is the coil constant of the spring used in the suspension system. It
is proportional to the displacement of the spring from its equilibrium
position. A larger value of c indicates a strong spring coil and thus a
tough suspension system. c is known as the spring constant.
b is a resistance coefficient that is proportional to the up and down
speed of the automobile. A larger value of b indicates a stronger
resistance to the up and down movement of the automobile, and thus
a smoother suspension system. b is known as the shock constant.
a is the weight of the vehicle.
g(t) is the external force that causes the suspension system of the
automobile to deviate from its equilibrium position. Some possible
causes may include a speed bump, a pothole, or a stone in the middle
of the road.
1. Using the VI developed in the previous exercise, enter the following
values in the controls on the front panel:
time start: 0
time end: 100
h: 0.1
Substitution Rules
parameter name:

parameter content:

1000

125

1000

X0:

LabVIEW Signal Processing Course Manual

11-22

National Instruments Corporation

Lesson 11

G Math Toolkit

The initial conditions of 0 specified in X0 indicate that the vehicle is

initially in a stable equilibrium position. g = 0, means that there is no
disturbing force.
2. Run the VI and note the waveforms on the XY graph. What do you
see? Explain the results.
3. Now assume that after driving for 10 seconds, the driver of the
automobile drove on to the sidewalk. This action can be modeled by
substituting a step function for g. In the Substitution Rules control,
change the value of g from 0 to 100*step(t-10) and run the VI.
Notice that until t = 10, the values of x and dx/dt are equal to zero
because no force (g) has been applied until then. At t = 0, a step
function of magnitude 100 is applied and both the position and the
up and down speed of the automobile change. However, as time
passes, the oscillations subside and both values tend to settle down
to an equilibrium position. You can verify this by entering 400 in the
time end control and running the VI.
4. A speed bump can be approximately modeled by the following value
for g:
10*sin(t)*(step(t) - step(t-3.142))
The multiplying constant (in this case 10) controls the height of the
speed bump, while the remaining term models one bump.
How the above formula corresponds to a speed bump is shown in the
figure below. The upper plot is a plot of one cycle of sin(t). The
middle plot is a plot of step(t) - step(t-3.142), whereas the lower plot
is the multiplication of the top two plots resulting in sin(t)*(step(t) step((t-3.142)).

National Instruments Corporation

11-23

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

Enter the above value of g and run the VI.

What do you see? Explain the results.
5. A pothole can be modeled in a similar fashion, but with the opposite
sign: -10*sin(t)*(step(t) - step(t-3.142))
Enter the above value of g and run the VI. What change do you notice
as compared with the results of step (4)?
6. Experiment with the value of c so that the oscillations reduce to zero
in less than 80 seconds.
Hint: reduce the value of c to a lower value (for example, 250)
7. Keeping the value of c at 1000, change the value of b so that the
oscillations reduce to zero in less than 80 seconds.
Hint: increase the value of b to a higher value (for example, 175)
Thus, you see that by proper selection of b and/or c, you can design the
suspension system of an automobile to behave in a certain way (that is,
to have a certain response). The selection of b and/or c could correspond
to a particular choice of material to be used in building the suspension
system.
8. Close the VI when you are finished.

End of Exercise 11-4

LabVIEW Signal Processing Course Manual

11-24

National Instruments Corporation

Lesson 11

G Math Toolkit

Exercise 11-5
Objective: To see some of the capabilities of the G Math Toolkit for 1D functions.
1. Launch the 1D Explorer Example VI from the
Labview/Examples/Gmath/Math.llb folder. This VI shows
a good overview of some of the mathematical functionalities of G
Math.

2. The default formula in the formula control is

sinc(c)+sin(2*c)+sin(3*c)+sin(2*c*c). Choose the following
options by holding down <Shift> and clicking the left mouse button
on the following selections

Modified Function Graph

Integration Graph

Roots

Maxima

3. Run the VI and click the Start button on the bottom right of the
panel. Plots of the above selections are displayed. Note that you can
enter the start and end points directly on the front panel through the
start and end controls. Also, note that LabVIEW returned the graphs
quickly because LabVIEW runs compiled. (Other math packages are
not compiled.)

National Instruments Corporation

11-25

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

4. Line up the cursor on a maxima and see its value.

5. Enter a new formula sin(exp(x)) in the formula control and click on
the Start button.
6. Zoom in on a maxima using the graph zoom option.
7. Experiment by typing in other formulas in the formula control.
8. When you finish, you can stop the VI by clicking on the STOP
button at the bottom right of the panel.

End of Exercise 11-5

LabVIEW Signal Processing Course Manual

11-26

National Instruments Corporation

Lesson 11

G Math Toolkit

Exercise 11-6 (Optional)

Objective:

To simulate the tank flow problem.

This exercise is a practical application that combines entering formulas on
the front panel, solving an ordinary differential equation, and visualizing a
process to simulate a tank inflow and outflow process.
Consider a cylindrical tank of constant cross section A cm2. Water is
pumped into the tank from the top at a constant rate fi(t) cm3/s. Water
flows out of the bottom of the tank by a valve of area a(t) cm2. Note that
both the input and output flow rates are functions of time. You will
observe how the height of the water, h(t), in the tank varies with time.
The solution to this problem is a first-order differential equation
given by
dh ( t )
a ( t ) fi ( t )
------------- = 2gh ( t ) --------- + --------dt
A
A

where g is the acceleration due to gravity, equal to 980 cm/s2.

The input flow rate to the tank, as well as the area of the outflow valve,
can be modeled as equations on the front panel. Select the following
equations for these values:

National Instruments Corporation

Input flow, fi(t):

340*square(t)

Area of valve, a(t), for the outflow:

sin(t) + 1

11-27

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

1. Open the Process Control Explorer Example VI, in the

Labview/Examples/Gmath/Misc.llb folder. The
explanation of the front panel is as already described above.
Note:

In this example, all variables in the differential equation can be controlled

from the front panel.
2. Run the VI and select the Calculation button. This button will flash
while the calculation is in progress.
3. Once the calculation is over, select the Simulation button. The
simulated system is then graphed.
4. Change the input flow rate and area of tank to:
fi(t), input flow rate: 340*sin(t)
a(t), area of valve:

0.01

5. Select the Calculation button and then the Simulation button. Notice
that the tank is now gradually emptying.
6. Select the STOP button.

End of Exercise 11-6

LabVIEW Signal Processing Course Manual

11-28

National Instruments Corporation

Lesson 11

G Math Toolkit

Exercise 11-7 (Optional)

Objective:

To see an example of the G Math Toolkit for data visualization.

1. In the Labview/Examples/Gmath/Graphics.llb folder,

open and run the Representation Function Graph 2D VI.
Note:

The default function value is sin(3*x)*cos(5*y). This is a function of two

variables, x and y. The range of values to be plotted is indicated on the Start
and End controls on the front panel. The default values are 0.0 and 1.0 for
both the x and y variables.
When the interactive subVI comes up, you can look at the plot from
different viewpoints by controlling the values of psi, phi, and r with
the help of the slider controls. Psi and phi control the angles with
respect to the xz and xy planes, respectively, where r controls the

National Instruments Corporation

11-29

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

distance from the origin. How these controls correspond to different

viewpoints is shown in the figure below:

z
observer

phi
x
psi

2. Change the psi value on the slider control. Note what happens.
3. Change the phi value on the slider control. Note what happens.
4. To return to the main VI, select the stop button on the front panel.
5. When you are done, close the VI.

End of Exercise 11-7

LabVIEW Signal Processing Course Manual

11-30

National Instruments Corporation

Lesson 11

G Math Toolkit

Summary
In this lesson, you saw the basic components of the G Math Toolkit and how
to use it for solving problems in engineering, math, and education. In
particular, you built a simple arbitrary waveform generator using the VIs in
the parser library, and you also designed the suspension system of an
automobile by making use of the differential equation VIs. Finally, you
experimented with other VIs that showed some of the more advanced
capabilities of G Math.

National Instruments Corporation

11-31

LabVIEW Signal Processing Course Manual

Lesson 11

G Math Toolkit

Notes

LabVIEW Signal Processing Course Manual

11-32

National Instruments Corporation

Lesson 12
Third-Octave Analyzers
Introduction
A third-octave analyzer measures the spectral energy contained in a
specific set of third-octave bands and provides a standardized
narrowband spectral analysis. It consists of a set of bandpass filters
connected in parallel. It is widely used in the field of acoustics and audio
signal processing to measure sound or acoustic intensity in various
frequency bands. Its applications lie in the fields of vibration tests of
machines, architectural acoustics, power measurements, etc. In this
lesson, you will learn about the specifics of the bandpass filters used in
third-octave analyzers, their applications, and how to use the
Third-Octave Analyzer Toolkit.

You Will Learn:

A. About filter banks.
B. About how an octave analyzer is a special type of filter bank.
C. About third-octave analyzers.
D. About the applications of octave and third-octave analyzers.
E. About the features of the Third-Octave Analyzer Toolkit.

National Instruments Corporation

12-1

LabVIEW Signal Processing Course Manual

Lesson 12

Third-Octave Analyzers

A. Filter Banks
A filter bank is a group of bandpass filters connected in parallel. Each filter
is tuned to a different frequency range. A simple filter bank consisting of
five filters is shown in the figure below. The center frequencies of each
bandpass filter are denoted by f1, f2, f3, f4, and f5, whereas their bandwidths
are denoted by B1, B2, B3, B4, and B5, respectively. There is a special name
given to the center frequencies. For the mth filter, let the center frequency be
fm. Then fm is known as the center frequency, the midband frequency, or
the geometric mean frequency.1 Note that in the figure, the bandwidths of
all the filters are the same. That is, B1 = B2 = B3 = B4 = B5.

B1

B2

B3

B4

B5

f1

f2

f3

f4

f5

frequency

One of the properties of bandpass filters is their quality factor, Q. The

quality factor is a measure of how selective the filter is in passing
frequencies around the center frequency and attenuating unwanted
frequencies. It is defined as the ratio of the center frequency of the filter
to its bandwidth.
Q = fm/Bm
Thus, for a fixed center frequency, the larger the bandwidth, the smaller
the quality factor, and vice versa. In the figure above, because the
bandwidths of all the filters are the same, but the center frequencies are
different, the quality factor is different for each filter.

1. Note that the actual term used depends on the relationship between f m and the upper and lower passband
frequency (fU and fL, respectively). This will be discussed in more detail later.

LabVIEW Signal Processing Course Manual

12-2

National Instruments Corporation

Lesson 12

Third-Octave Analyzers

A practical application of a bandpass filter is in a radio or television

receiver. Radio and television channels all occupy different frequency
ranges. The antenna picks up signals of all the frequencies of all the
channels. However, inside the radio or television receiver, the signal is
passed through a bandpass filter used to select a particular channel (the
one that you want to hear or see) and reject the other channels. When you
change the tuning knob on your radio, or the channel selector knob on
your TV, you are actually changing the center frequency of this
bandpass filter. It is desirable for the bandpass filter to pass the selected
channel and reject the unwanted channels. Hence, the bandpass filter
needs to have a high Q so that you do not hear (or see) two or more
channels at the same time.

National Instruments Corporation

12-3

LabVIEW Signal Processing Course Manual

Lesson 12

Third-Octave Analyzers

B. Octave Analyzers
Now consider the filter bank shown in the figure below. It consists of four
bandpass filters each with a different center frequency and different
bandwidths. The center frequencies are denoted by f1, f2, f3, and f4, whereas
the bandwidths are denoted by B1, B2, B3, and B4. The filters with higher
center frequencies have larger bandwidths, and filters with lower center
frequencies have lower bandwidths.
The special feature of this filter bank is that the center frequency of any
of the filters is exactly twice (one octave higher) that of the center
frequency of the filter just below it and exactly 1/2 (one octave lower)
that of the center frequency of the filter just above it.
f

m
That is, ---------- = 2

fm 1

and

fm
1
---------- = --fm + 1
2

B1

B2

B3

B4

f1

f2

f3

f4

frequency

This same relationship holds true for the bandwidths of any two adjacent
filtersthe bandwidth of any filter is exactly twice that of the bandwidth
of the filter just below it and exactly 1/2 that of the bandwidth of the
filter just above it.
B
Bm 1

m
That is, ------------ = 2

and

Bm
1
------------ = --Bm + 1
2

Note that the quality factor = Q = fm/Bm is now the same for all filters,
because both fm and the bandwidth of adjacent filters increase (or
decrease) by the same amount. Therefore, these bandpass filters are
known as constant Q bandpass filters.
Such a filter bank, where the ratios of center frequencies and bandwidths
of adjacent filters are powers of 2 (21 = one octave), is known as an

LabVIEW Signal Processing Course Manual

12-4

National Instruments Corporation

Lesson 12

Third-Octave Analyzers

octave analyzer. Both the center frequencies and bandwidths increase

logarithmically, and all the bandpass filters have the same value of Q.
The bandwidth of each filter covers a frequency range of one octave.
For each bandpass filter, let fm denote the center frequency, and fU and
fL denote the upper and lower passband frequencies, respectively. This
is shown in the figure below.

fL

fm

fU

The relationship between them is that fm is the geometric mean of fU and

fL
fm =

fu fL

and hence fm is also known as the geometric mean frequency.

National Instruments Corporation

12-5

LabVIEW Signal Processing Course Manual

Lesson 12

Third-Octave Analyzers

C. 1/3-Octave Analyzer or Third-Octave Analyzer

Now you can probably make a good guess as to what is a 1/3-octave
analyzer, also known as a third-octave analyzer (TOA). It is very similar to
an octave analyzer but with a different ratio of center frequencies and
bandwidths of adjacent filters. It consists of a set of bandpass filters
connected in parallel, where the ratio of center frequencies and bandwidths
of adjacent filters is 21/3.
fm
13
---------- = 2
fm 1

and

fm
1 3
---------- = 2
fm + 1

Bm
13
------------ = 2
Bm 1

and

Bm
1 3
------------ = 2
Bm + 1

Because the ratio is now 21/3 (approximately = 1.26) it requires three of

these filters to cover a frequency range of one octave. The net effect of
this is that three filters of a 1/3-octave analyzer cover the same
frequency range as one filter of an octave analyzer. This is shown in the
figure below for one filter of the octave analyzer.
Three filters of a 1/3 octave analyzer (dotted lines)
One filter of an octave analyzer
(solid line)

2f
one octave

Note:

freq

In general, you can have a 1/n-octave analyzer where the ratio of center
frequencies and bandwidths of adjacent bandpass filters is 21/n. In
practice, the most popular ones are 21, 21/3, and 21/12. These correspond to
ratios with numeric values of 2, 1.26, and 1.06, respectively.
For 1/3-octave analyzers also, there is f m =
geometric mean frequency.

fU fL

where fm is called the

The filters in both octave and third-octave analyzers are also known as
constant percentage bandwidth filters. That is, the bandwidth of the
filters are a constant percentage of the center frequency. For an octave
filter, the ratio of f2/f1 = 2, and its bandwidth is 1/ 2 = 71% of the center
frequency. For a 1/3-octave filter, the ratio f2/f1 = 3 2 , and its bandwidth
is 21/6 - 2-1/6 = 23% of the center frequency.

LabVIEW Signal Processing Course Manual

12-6

National Instruments Corporation

Lesson 12

Third-Octave Analyzers

D.Applications
The applications of 1/n-octave analyzers are mainly in the fields of acoustics
and audio signal processing. Basically, you want to measure the sound or
acoustic intensity in various frequency bands. A specific example is in the
aviation industry, where measurement of noise emissions from an aircraft
are necessary for it to get airworthiness certification. Federal aviation
regulations require the use of real-time 1/3-octave analyzers for data
analysis. Manufacturers of appliances and office equipment also tend to
specify sound power levels in their product documentation so that
customers can compare their products and choose the quieter one. By
measuring sound power levels in a room or other enclosure such as a concert
hall, acoustic engineers determine the proper material and design for the
enclosure.
Other applications are in

Vibration tests in aircraft and submarines

Sound power determination

Acoustic intensity measurement

Audio equalization

Architectural acoustics

Appliance testing

The information obtained from a TOA can be used to reduce vibrations

in aircraft and submarines, to reduce noise in aircraft, submarines, and
household appliances, and to make sound (for example, music) more
pleasant to hear (for example, equalizers and architectural acoustics).

National Instruments Corporation

12-7

LabVIEW Signal Processing Course Manual

Lesson 12

Third-Octave Analyzers

E. The Third-Octave Analyzer Toolkit

You saw that the applications of 1/n-octave analyzers are in those fields
where there is a need to measure sound power in different frequency bands.
To compare different measurements with each other, measurements are
performed according to certain standards. The American National
Standards Institute (ANSI) has specified the standards for the design of
octave and third-octave filters. This information is found in the ANSI S1.11
(1986) specifications for octave-band and fractional-octave-band
analog and digital filters. The National Instruments Third-Octave
Analyzer Toolkit conforms to these specifications. The center frequencies
of the BPFs of the third-octave analyzer, as given in the ANSI S1.11 (1986)
specifications, are shown in the following table. Also shown is the
frequency range corresponding to the passband of the filters. There
are a total of 30 specified center frequencies spanning a range from 25 Hz
20 KHz. However, in the National Instruments 1/3-octave analyzer, an
additional BPF with a center frequency of 10 Hz has been added to the lower
frequency range, giving a total of 31 BPFs.
Center freq. (Hz)

Passband (Hz)

Center freq. (Hz)

Passband (Hz)

Center freq. (Hz)

Passband (Hz)

25

22.4 - 28.2

400

355 - 447

6300

5620 - 7080

31.5

28.2 - 35.5

500

447 - 562

8000

7080 - 8910

40

33.5 - 44.7

630

562 - 708

10000

8910 - 11200

50

44.7 - 56.2

800

708 - 891

12500

11200 - 14100

63

56.2 - 70.8

1000

891 - 1120

16000

14100 - 17800

80

70.8 - 89.1

1250

1120 - 1410

20000

17800 - 22400

100

89.1 - 112

1600

1410 - 1780

125

112 - 141

2000

1780 - 2240

160

141 - 178

2500

2240 - 2820

200

178 - 224

3150

2820 - 3550

250

224 - 282

4000

3550 - 4470

315

282 - 355

5000

4470 - 5620

LabVIEW Signal Processing Course Manual

12-8

National Instruments Corporation

Lesson 12

Third-Octave Analyzers

The Third-Octave Analyzer Toolkit can be used as a stand-alone

application, an add-on toolkit for LabVIEW or BridgeVIEW, or with
LabWindows/CVI. It can analyze the frequency of the signal in 31
frequency bands.
The figure below shows the front panel of the TOA stand-alone
application. It can analyze up to four channels of data. Depending on the
number of channels you choose, you could get either one, two, three, or
four graphs on the front panel. In this particular figure, you see four
graphs. The x-axis of each graph is the frequency scale and the y-axis of
each graph is the power amplitude (in decibel) scale. Each graph also has
a cursor that you can move to see the power value in each frequency
band. This cursor is shown by the vertical line with an x through it.
The value of the center frequency (in hertz) in the frequency band and
the corresponding power value (in decibel) is displayed at the top of each
graph.

On the front panel above, Acquire retrieves and analyzes a block of data.
The analyzer does not start acquiring data until you click on this button.
The data can be acquired in single or continuous mode. The mode can be
chosen by clicking on the up or down arrows on the control to the left of
the Acquire button. When you select single, every time that you click the
Acquire button, the analyzer acquires and analyzes a new block of data
and displays the result on the corresponding graph. When you choose
continuous, the analyzer starts to acquire and analyze data when you
click the Acquire button. After it has displayed the data on the
corresponding graph, it acquires the next block of data and analyzes and
displays it. This continues until you click on the Stop Acquire button.

National Instruments Corporation

12-9

LabVIEW Signal Processing Course Manual

Lesson 12

Third-Octave Analyzers

(The Acquire button becomes the Stop Acquire button during an

acquisition.)
Amplitude Table shows a table with all 31 bands of output power values
for each channel.
Save saves the 31 bands of power values in a spreadsheet format where
each column represents 31 bands of the power value for each channel.
This button saves the power amplitude as well as some status
information, such as channel numbers, window type, average type, and
weighting value. We will discuss this status information later.
Quit stops the analyzer.
Setup opens the Setup panel shown below. The setup panel selects the
parameters that you want to customize for your application. It appears
when you first run the TOA. However, you can always access it at any
time by clicking on the Setup button.

device specifies the identification number for your DAQ board.

In Windows, you can assign this number to your device through
software. For the Macintosh, the device parameter indicates the physical
slot number into which you insert your device. You can find this number
by opening NI-DAQ in the control panels folder.

LabVIEW Signal Processing Course Manual

12-10

National Instruments Corporation

Lesson 12

Third-Octave Analyzers

sampling rate determines the rate at which the signal is sampled, and
thus the frequency range being analyzed (this is because as you have
already seen, the sampling frequency must be at least twice the
maximum frequency of the signal). Three choices of sampling rates are
available: 12.8 KHz, 25.6 KHz, and 51.2 KHz. The corresponding
frequency ranges being analyzed are: 5 Hz - 5 KHz, 10 Hz - 10 KHz, and
20 Hz - 20 KHz. These sampling rates were chosen because they are
software selectable with the National Instruments dynamic signal
acquisition boards (AT-A2150, NB-A2150F, and AT-DSP2200), which
you can use with the TOA. The sample rate of 51.2 KHz conforms to the
Extended range ANSI specifications.
Channel # indicates the channels from which to acquire the data and
display it. Up to four channels can be chosen (and the channel number
can also be the same in every control). Each channel has a box with a
cross in it. If you do not need to use a channel, click inside the box until
the cross disappears. This will disable all the parameters associated with
that particular channel.
Window Type selects one of four commonly used windows
(Rectangular, Hamming, Hanning, and Blackman) for each channel. The
Window Type parameter defaults to the Hanning window.
Next, you will complete an exercise which uses the third-octave analyzer
and highlights more of its features.

National Instruments Corporation

12-11

LabVIEW Signal Processing Course Manual

Lesson 12

Third-Octave Analyzers

Exercise 12-1
Objective:

To analyze a signal consisting of a sine wave with harmonics and added white
noise.

Note:

This exercise requires a modified library of the actual third-octave

analyzer toolkit. If the file TOA\Octave.llb does not exist on your hard
drive, you cannot do this exercise. In that case, you may read this exercise
to become familiar with the front panel controls and indicators, and then
move on to the next section.
1. Open the Third-Octave Analyzer VI from the folder
Toa\Octave.llb. You will see the following front panel:

Front Panel

You see four graphs on the screen. This is because you can analyze
signals coming from up to four channels. The number of graphs
displayed on the screen depends on the number of channels you are
analyzing.
The two controls (Frequency and Amplitude) on the right side of the
front panel have been added only for the purpose of this exercise.
They do not exist in the actual Third-Octave Analyzer Toolkit. They
adjust the amplitude and frequency of a sine wave (with harmonics)
that has been corrupted by white noise. On the front panel of the
TOA, use the Frequency dial to control the frequency of the
fundamental component of the sine wave between 0 - 10 KHz. Use
the Amplitude slider control to adjust the amplitude of the
fundamental between 0 - 20.

LabVIEW Signal Processing Course Manual

12-12

National Instruments Corporation

Lesson 12

Third-Octave Analyzers

Although you do not need to be concerned about this for now, for
those who are interested, the block diagram of the VI (Signal.vi) used
to generate this simulated signal is shown below:

The above VI also adds to the fundamental its 3rd and 7th harmonics
of amplitudes 1/5th and 1/15th, respectively. The addition of
harmonics is simulated by the lower two Sine Wave VIs.

Setup Window
2. With the Amplitude and Frequency controls both set to 0.00, run
the VI.

National Instruments Corporation

12-13

LabVIEW Signal Processing Course Manual

Lesson 12

Third-Octave Analyzers

A new window called Setup appears.

The Setup window gives you control over the number of channels to
sample, the sampling rate, FFT size, and much more. For now, you
will leave the following controls at their default values:
device

sampling rate

25.6 K Hz

data blocks to average

FFT size

512

Internal Data Averaging

no averaging

3. In the Setup window, click on the white box to the left of the labels
ChB, ChC, and ChD. When you click on the white boxes, the
crosses in them disappear, and the selections for those channels are
grayed out.

LabVIEW Signal Processing Course Manual

12-14

National Instruments Corporation

Lesson 12

Third-Octave Analyzers

4. Now, because only the white box to the left of ChA has a cross in it,
you will use only ChA to acquire the signal.
The device control corresponds to the number of the DAQ board in
your computer. You can obtain the correct device number for your
DAQ board by using the NI-DAQ Configuration Utility. For the
purpose of this exercise, leave device set to 1.
The Channel # control determines the particular channel on the DAQ
board from which you will acquire the sampled signal. (A DAQ
board usually has more than one channeltypically 8 or 16.) Leave
this set to 0.
The Window Type control determines the type of window you want
to apply to the data samples.
5. Click on the Done button. The Setup window will disappear, and you
will return to the main panel of the TOA. You can bring up the Setup
window at any time to change the configuration parameters by
clicking on the Setup button on the bottom left of the main panel of
the TOA. Note that only one graph is displayed, because you had
selected only 1 channel in the Setup window.

Single and Continuous Acquisition

6. The Single button (just next to the Setup button) is a menu ring
control that determines how the samples of the data are collected.
When set to Single mode, only one block of data is acquired when

National Instruments Corporation

12-15

LabVIEW Signal Processing Course Manual

Lesson 12

Third-Octave Analyzers

the Acquire button is pressed. However, when set to Continuous

mode, blocks of data are acquired one after the other when the
Acquire button is pressed. While in the continuous mode of
operation, the Acquire button changes to the Stop Acquire button.
To stop the continuous data acquisition, you must press the Stop
Acquire button.
7. With the menu ring control on Single, click the Acquire button.
After a short time interval during which the data is obtained and the
calculations for the power output are being done, a histogram
appears on the graph. This is the histogram of white noise only (there
is no signal because the amplitude of the sine wave has been set to
0.0 on the slider control). Each rectangle of the histogram gives the
power in the frequency band of the corresponding BPF. Note that the
power increases as you go toward higher frequencies. Why do you
think that is so?
Hint: White noise has a flat frequency spectrum (that is, the same
average power at all frequencies). The filters used in the TOA have
increasing bandwidths as the center frequency increases.
8. Click on the Acquire button again. Once again, after the data is
obtained and the calculations are over, the display changes. The data
is obtained, the calculations done, and the display updated each time
you click on the Acquire button. However, if you do not click the
Acquire button, nothing happens. This is the operation of the Single
acquisition mode.
9. Click on the arrows of the Single button and change it to Continuous
mode. Click on the Acquire button to start the acquisition.
Now the operation of obtaining data, calculating the power outputs
of the BPFs, and updating the data is done continuously. Note that
the Acquire button has changed to the Stop Acquire button. You
must click on the Stop Acquire button if you want to stop the
continuous data acquisition.

Changing the FFT Size

10. You will notice a time delay between successive updating of the
display. This is because the default size of the FFT has been set to
512 data points. It takes a certain amount of time to acquire the data
points, perform the necessary calculations, and display the results.
To decrease this time, you can choose a smaller FFT size.
Click on the Setup button. The Setup window comes up. Set FFT
size to 256 and click on the Done button. Now notice that the display
is much faster.

LabVIEW Signal Processing Course Manual

12-16

National Instruments Corporation

Lesson 12

Third-Octave Analyzers

Adding a Signal to the White Noise

11. Set the Frequency control to 100 Hz and the Amplitude control to
10. The display will show you three peaks. The highest peak is from
the output of the BPF that covers the frequency range in which the
fundamental sine wave of frequency 100 Hz falls. The other two
peaks are from the outputs of the BPFs whose frequency bands
include the 3rd and 7th harmonics (that is, 300 Hz and 700 Hz).

Cursor Control
12. Move the cursor around the display. On the top left, you can see the
power output (in dB) and the corresponding center frequency (in Hz)
of the BPF corresponding to the rectangle on which you place the
cursor.

The Amplitude Table

13. Another way to see the power in each of the different frequency
bands is to click on the Amplitude Table button. A new window
comes up with a table showing the center frequencies and the
corresponding power output (in dB) of each of the BPFs. To close
this window, click on the Ok button.

Multiple Channels and Weighting

14. You will now acquire and display data from more than one channel.
Click on the Setup button to bring up the Setup window. In the
Setup window, click on the white box to the left of the label ChB and
change its Channel # to 0 (the same as for ChA). Change the
Weighting of ChB to A-weighting.
You have just sampled the same signal on both channels A and B.
However, the signal obtained on ChA is displayed with no
weighting, whereas the signal obtained on ChB is displayed with
A-weighting. Weighting is done because the human ear responds
differently to different frequencies. By using A-weighting, the
analyzer mimics the human hearing response to audio signals.
Internally, the analyzer accomplishes this by adding a specified
weighting value (in dB) to the calculated power before displaying it.
Other weighting selections that are available are none (no weighting)
and custom (user defined).

National Instruments Corporation

12-17

LabVIEW Signal Processing Course Manual

Lesson 12

Third-Octave Analyzers

15. In the Setup window, click on the View Weighting bar to see a table
showing the weighting values for each frequency band for each
channel. Click on the OK button to close the window showing the
table.

16. In the Setup window, click on the Done button.

17. Now you are back to the main front panel of the TOA. The top graph
shows the display from ChA (with no weighting) and the bottom
graph shows the display from ChB (with A-weighting). You can thus
compare the outputs from different channels.
18. Click on the Stop Acquire button to stop the data acquisition.
19. Click on the Quit button to stop the VI.
20. Close the VI. Do not save any changes.

End of Exercise 12-1

LabVIEW Signal Processing Course Manual

12-18

National Instruments Corporation

Lesson 12

Third-Octave Analyzers

F. Calibration Using the Third-Octave Analyzer

You can also use the TOA in applications involving calibration. For this
purpose, the first step is to apply to the TOA a reference signal from a highly
accurate source. The result obtained is then saved (by clicking on the Save
button) as a reference for future use. This previously recorded signal can be
loaded later for use as a reference by clicking on the Reference button. You
can compare this reference signal with the current signal you are obtaining,
because the analyzer plots both signals on the same graph.
The following three exercises outline the steps on how to use the TOA for
calibration. In the next exercise, you will generate a reference signal. In the
exercise following that, you will save the results of third-octave analysis on
this reference signal. In the final exercise, you will compare these results
with those obtained from another source.
The reference signal is obtained from the function generator on a DAQ
Signal Accessory. Some of the connections on the box are as shown below.

DAQ Signal Accessory Box

A B

Frequency
Range

Analog
Out

DAQ
Signal Accessory

Ch 0

Analog
In

Frequency
Adjust

Function
Generator

You will use this box to generate a sinusoidal signal of frequency 200 Hz
and peak amplitude of 1 V. This will be your reference signal.

National Instruments Corporation

12-19

LabVIEW Signal Processing Course Manual

Lesson 12

Third-Octave Analyzers

Exercise 12-2
Objective:

To generate a sinusoidal reference signal of frequency 200 Hz and peak amplitude

of 1.0 V.
1. Make the following connections between the DAQ Signal Accessory
and data acquisition board on your computer.
Connect the sine wave output of the Function Generator to Analog In
channel 1. Thus the sine wave will be available at channel # 1 of the
DAQ board.
Set the Frequency Range slider control to 100Hz-10kHz.
Connect the DAQ Signal Accessory cable connection through a
ribbon cable to the DAQ board on your computer.

ribbon cable

A B

Frequency
Range

Frequency
Adjust

13 kHz 1 MHz
1 kHz 100 kHz
100 Hz 10 kHz
Analog
Out

DAQ
Signal Accessory

Ch 0

Analog
In

Function
Generator

wire
connection

2. Bring up LabVIEW (or BridgeVIEW) and load the Examples

Analysis measure daqmeas.llb Simple Spectrum Analyzer VI.
You will use this VI to measure the exact amplitude and frequency

LabVIEW Signal Processing Course Manual

12-20

National Instruments Corporation

Lesson 12

Third-Octave Analyzers

of the sinusoidal signal from the Function Generator output of the

DAQ Signal Accessory.

3. Make the following changes to the default settings on the Simple

Spectrum Analyzer VI:
Enter a value of 1 in the channel(0) string control.
Change number of samples to 1000.
Select None (Uniform) window in the Window control.
Leave all the other settings at their default values.
Note:

With a sampling rate of 20000 and number of samples selected as 1000, the
frequency spacing is 20000/1000 = 20 Hz. The FFT lines will thus fall on
frequency values of 0, 20, 40, 60, and so on. The signal of 200 Hz will
therefore fall exactly on the eleventh FFT line.
4. Run the VI. In the Est Power Peak and the Est Frequency Peak
indicators on the top, you will see the estimate of the power and
frequency of the highest peak in the spectrum.
5. Adjust the Frequency Adjust control on the DAQ Signal Accessory
to get an Est Frequency Peak of 200 Hz (or as close to it as possible).
(You will need to run the VI each time after you change the
Frequency Adjust control.)
6. Once you obtain a frequency of 200 Hz, note the value in the Est
Power Peak indicator. This value is in units of V2rms. You can

National Instruments Corporation

12-21

LabVIEW Signal Processing Course Manual

Lesson 12

Third-Octave Analyzers

convert this to a peak voltage (Vpeak) reading by using the following

formula:
V peak =

2 V rms

You should obtain a value near 1.0 V.

7. Make a note of the frequency and the corresponding voltage Vpeak. Be
sure not to change the settings on the DAQ Signal Accessory.
8. When you are done, close the VI. Do not save any changes.

End of Exercise 12-2

You will use the sinusoidal signal of the frequency and amplitude that
you have noted in the previous exercise, as the reference signal against
which we can compare other signals. This reference signal has a
frequency of 200 Hz and a peak amplitude of 1.0 V.
Note:

The method you have used in the previous exercise to generate the
reference signal is just to illustrate the principles involved in calibration of
the TOA. In practice, the reference signal should be obtained from a more
accurate and reliable source.

LabVIEW Signal Processing Course Manual

12-22

National Instruments Corporation

Lesson 12

Third-Octave Analyzers

Exercise 12-3
Objective:

To save the results obtained from the TOA for a standard reference signal.
In this exercise, you will use the same connections (between the DAQ
Signal Accessory box and your computer) as those you had before in the
previous exercise.
1. Run the Third-Octave Analyzer from the Start Programs menu.
2. In the Setup panel, make sure only ChA is selected. This is done by
clicking in the white box to the left of ChB, ChC, and ChD, so that
the white box does not have a cross in it, and also that the selections
for these channels are grayed out.
3. Change the Channel # setting to 1. Leave the other settings at their
default values as shown:
device:

sampling rate:

51.2KHz

data blocks to average: 1

Window Type: Hanning

Average Type:

linear

Weighting:

no weighting

FFT size:

512

Internal Data
Averaging:

no
averaging

4. When you are finished, click on the Done button at the bottom right.
5. On the TOA front panel, press the Acquire button. The VI has now
obtained samples of the sinusoidal signal of frequency 200 Hz and

National Instruments Corporation

12-23

LabVIEW Signal Processing Course Manual

Lesson 12

Third-Octave Analyzers

peak amplitude of 1 V and is showing the results of third-octave

analysis on this signal.

Move the cursor so that it aligns with the band corresponding to the
largest peak in the display. The frequency reading on the top of the
display should read 200 Hz. The amplitude reading should be near
-3.0 dB (which corresponds to a peak amplitude of 1.0 V).
You will now save these results as a reference that can be compared
with other signals.
6. Click on the Save button. When prompted for the filename, type in
standard.ref. You could use any filename, but it is
recommended that you use one with a .ref extension if you plan to
use it as a reference file.
The file standard.ref now contains the results of third-octave
analysis on the reference signal.
7. Exit the VI by clicking on the Quit button at the bottom right.

End of Exercise 12-3

LabVIEW Signal Processing Course Manual

12-24

National Instruments Corporation

Lesson 12

Third-Octave Analyzers

Exercise 12-4
Objective:

To compare the results of third-octave analysis on a given signal with that obtained
from a standard reference signal.
In the previous exercise, you had saved the results of third-octave
analysis on our standard reference signal. In this exercise, you will
compare these results with those obtained from another source.
Exchange your DAQ Signal Accessory with someone else in the class
and make the connections as you had done in the previous two exercises.
Assume that your previous DAQ Signal Accessory was well calibrated
and of high enough quality to generate the standard reference signal.
You will now compare those results with those obtained using the other
DAQ Signal Accessory.

Note:

Do not change the settings on the other DAQ Signal Accessory.

1. Open the Third-Octave Analyzer from the Start Programs menu.
2. In the Setup panel, make the selection as you had done in the
previous exercise. That is, select only ChA, with a corresponding
Channel # of 1. The other controls have the following values:
device:

sampling rate:

51.2KHz

data blocks to average: 1

Window Type: Hanning

Average Type:

linear

Weighting:

no weighting

FFT size:

512

Internal Data
Averaging:

no averaging

3. After verifying the above settings, click on the Done button.

4. On the front panel, click on the Reference button. A new window
comes up from which you can select the file standard.ref that
you had saved in the previous exercise.
This file contains the results of third-octave analysis on a reference
signal of 200 Hz having a peak amplitude of 1 V. The results stored
in the file will be plotted in white.
5. Click on the Acquire button to obtain a single block of data. The
results will be overlaid on the same plot as that of the reference
signal.

National Instruments Corporation

12-25

LabVIEW Signal Processing Course Manual

Lesson 12

Third-Octave Analyzers

6. Move the cursor around the plot. For the band on which the cursor is
placed, the top left of the display shows you the center frequency, the
dB levels of the two plots, and the difference in dB levels of the two
plots. The figure below shows the display for the band with the
200 Hz center frequency.

center frequency
current power value
reference power value

difference of current power value

with reference power value

7. To calibrate the new DAQ Signal Accessory that you obtained, you
need to adjust the amplitude and frequency on the new box so that
the difference in power reading in the band with a center frequency
of 200 Hz is zero. That will mean that both the DAQ Signal
Accessory boxes are producing signals with exactly the same
frequency and amplitude.
8.

When you finish, exit the VI by clicking on the Quit button.

End of Exercise 12-4

LabVIEW Signal Processing Course Manual

12-26

National Instruments Corporation

Lesson 12

Third-Octave Analyzers

Summary
In this lesson, you learned the basic theory behind the operation of octave
analyzers. You have seen that they consist of a set of bandpass filters. The
ratio of the center frequencies and bandwidths of adjacent bandpass filters
is equal to 2 for an octave analyzer and 21/3 (approximately 1.26) for a TOA.
That means that as you go higher in frequency, the center frequency and
bandwidth of one BPF is twice that of the adjacent lower frequency BPF for
the octave analyzer (or 1.26 times that of the adjacent lower frequency BPF
for a third-octave analyzer).
You used the TOA on a signal consisting of a sine wave (and its
harmonics) with added white noise. Specifically, you learned:

How to select the number of channels.

The difference between single and continuous modes of operation.

How to change the FFT size.

About weighting the power output of the BPFs.

How to see the power output and center frequency values by using
the cursor or the Amplitude Table control.

Finally, you saw how to use the TOA to compare the power levels
between two signals, one of which is a reference signal.

National Instruments Corporation

12-27

LabVIEW Signal Processing Course Manual

Lesson 12

Third-Octave Analyzers

Notes

LabVIEW Signal Processing Course Manual

12-28

National Instruments Corporation

Lesson 13
Joint Time-Frequency
Analysis
Introduction
In this lesson, you will learn the basic theory behind joint time-frequency
analysis (JTFA) and how to use the JTFA Toolkit in a variety of
applications.

You Will Learn:

A. Why you need joint time-frequency analysis.
B. About joint time-frequency analysis.
C. About the JTFA Toolkit and its applications.

National Instruments Corporation

13-1

LabVIEW Signal Processing Course Manual

Lesson 13

Joint Time-Frequency Analysis

A. Why Do You Need Joint Time-Frequency Analysis?

A signal can be represented in a number of different ways. You want the
representation that most accentuates the qualities of the signal in which you
are interested. You may be very familiar with the representation of a signal
as a function of time. This representation shows how the signal magnitude
changes over time. Alternately, you can also represent the signal as a
function of frequency by performing a Fourier transform. The figure below
shows the time representation of a sine wave and its Fourier transform.

Traditionally, signals are studied as a function of time or frequency, but

not both. However, a number of signals encountered in real-world
applications have time-dependent frequency representations. One such
example is signals representing the tones of music that vary with time.
Therefore, in many practical applications, it is very useful to
characterize the signal in time and frequency domains simultaneously.

LabVIEW Signal Processing Course Manual

13-2

National Instruments Corporation

Lesson 13

Joint Time-Frequency Analysis

The figure above shows the time-frequency plot of a bird sound. The plot
on the right is the standard Fourier spectrum. From this spectrum alone,
you cannot tell how the frequencies have changed over time. Beneath the
time-frequency plot is a plot of the time waveform of the bird sound.
This shows you only how the sound level changes as a function of time.
The advantage of having a time frequency plot is that not only can you
tell what the range of frequencies were, but also how these frequencies
changed as a function of time. For this bird sound, you can see that at the
beginning, the bird was making a sound at higher frequency, which then
changed as a function of time, as indicated on the graph. Furthermore, in
the time-frequency plot, not only can you see how the frequency
changed in time, but you can see the intensity of the frequency, indicated
by the relative brightness levels of the plot.

National Instruments Corporation

13-3

LabVIEW Signal Processing Course Manual

Joint Time-Frequency Analysis

frequency

Lesson 13

time

The figure above shows graphs of a speech signal. The plot at the bottom
of the figure is the time waveform of such a signal. This plot shows how
the magnitude of this signal changes as a function of time. The plot in
the upper right corner is the frequency representation of the same signal.
This representation reveals four prominent frequency tones in the
spectrum. However, from the spectrum alone, you cannot tell how these
frequencies evolve over time. This is where joint time-frequency
analysis comes into the picture. The 2-D contour plot in the figure above
is the result of such an analysis. This plot is the time-dependent
spectrum, a function of both time and frequency, and reveals how
different frequency tones evolved as a function of time. From this, not
only can you see how the frequency changed, but you can also see the
intensity of the frequencies, as shown by the relative brightness levels of
the plot. Consequently, by using JTFA, you can better understand the
mechanism of human speech.
From these two examples, you have seen that time-frequency analysis
offers a better understanding of the nature of the signal. Several other
applications in this lesson illustrate this point.

LabVIEW Signal Processing Course Manual

13-4

National Instruments Corporation

Lesson 13

Joint Time-Frequency Analysis

B. Joint Time-Frequency Analysis

This section briefly introduces different techniques for joint time-frequency
analysis and outlines the algorithms associated with each technique. Refer
to Joint Time-Frequency Analysis: Methods and Applications by Shie
Qian and Dapang Chen, Prentice Hall, 1996 for a detailed discussion of
these algorithms.
There are two types of joint time-frequency representations, namely the
linear and the bilinear (quadratic). All linear transformations are
achieved by comparing the signal to be analyzed with a set of prudently
selected elementary functions, known as the analysis function. As an
example, consider the Fourier transform F ( ) of a signal f(t).

F ( ) f ( t )e

jt

In this case, e jt is the analysis function. The elementary functions

selected in the inverse transformation are called synthesis functions. For
example, in the inverse Fourier transform,

f ( t ) F ( )e

jt

jt

is the synthesis function. The short-time Fourier transform (STFT)

and the wavelet transform are two widely used methods to obtain a linear
representation. Wavelets are discussed in detail in the next lesson.
e

Short-Time Fourier Transform (STFT)

Similar to the example seen above, for the continuous-time STFT, the
analysis function and the synthesis function have the same form. In this
case, the inverse problemthat is, recovering the original time functions
is very simple.
For digital signal processing applications, it is necessary to extend the
STFT framework to discrete-time signals. For a given discrete
signal s [ i ] , the following equation defines the STFT for 0 k L--- :
--L- 1

mk
STFT [ i, k ] =
s [ i m ]g [ m ]W L

m = L--
2

where g[m] is an analysis signal.

National Instruments Corporation

13-5

LabVIEW Signal Processing Course Manual

Lesson 13

Joint Time-Frequency Analysis

t0

t18
t11

t4

time-slice 1

time-slice 2

time-slice 3

FFT

FFT

FFT

Square
Magnitude

Square
Magnitude

Square
Magnitude

time

t18
t11

t4

time

The STFT is also called a sliding-window fast Fourier transform (FFT)

because a window function breaks the signal into several time slices.
The FFT computes the frequency spectrum for each slice of windowed
data, and then you take the square magnitude of each FFT. The result of
this operation for each time slice is associated with the time index in the
middle of that particular slice of windowed data. For example, the result
of the operation for time-slice1 is associated with the time index t4, and
that for time-slice3 is associated with the index t18. To establish a
complete three-dimensional spectrogram, you can slide the window to
the right one or more points at a time and compute a new spectrogram.
Consider one classical example. A signal known as the frequency hopper
signal is widely used in digital communication:

b0
Digital
Transmitter

input stream

b1

Assume that two bits are being transmitted simultaneously over two
transmission lines. Let the frequency corresponding to the combination

LabVIEW Signal Processing Course Manual

13-6

National Instruments Corporation

Lesson 13

Joint Time-Frequency Analysis

00 be f1, the frequency corresponding to 01 be f2, to 10 be f3, and to 11

be f3. Then, depending on which combination is being transmitted, the
signal hops between these four frequencies, hence the name frequency
hopper.
The figure below shows the time waveform of a frequency hopper
signal, the corresponding spectrum as calculated by FFT, and the STFT
distribution of this signal. From the time waveform, you cannot see any
frequency information. From the spectrum, you can see that there are
four distinct frequency components but cannot tell when those
components occurred in time. The STFT representation provides both
the time and frequency information in the same plot. From this
spectrogram, not only can you distinctly see the four frequency levels,
but you can also see when these frequencies changed in time.

Wigner-Ville Distribution
There are different ways to obtain bilinear1 joint time-frequency
representations. The first is the Wigner-Ville distribution. This technique is
very simple and it better characterizes the signals time-dependent spectra
than the STFT spectrogram. Also, this distribution possesses many
properties useful for signal analysis.

1. Called bilinear because the signal x(t) appears twice in the equations for each representation. For more information,
refer to Joint Time-Frequency Analysis: Methods and Applications by Shie Qian and Dapang Chen, Prentice Hall,
NJ, 1996.

National Instruments Corporation

13-7

LabVIEW Signal Processing Course Manual

Lesson 13

Joint Time-Frequency Analysis

The figure above shows the time waveform of a frequency hopper signal,
the corresponding spectrum as calculated by FFT, and the Wigner-Ville
distribution of this signal. From the time waveform, you cannot see the
frequency information. From the spectrum, you can see four distinct
frequency components but cannot tell when those components occurred
in time. The Wigner-Ville distribution provides both the time and
frequency information in the same plot. However, as you can see from
the plot, the main deficiency of this distribution is the cross-term
interference. This cross-term interference occurs at frequencies midway
between the two main frequencies. If there are four main frequencies (as
in the example above), there are six terms due to cross interference.
However, a good characteristic of these cross-term interferences is that
they highly oscillate between positive and negative values. These
positive and negative parts cancel out each other on averaging. To do so,
you can apply a low-pass filter to retain the low frequency components
and remove the high frequency parts. Because the discarded
high-frequency parts have small averages, the lowpass filtered
Wigner-Ville distribution presumably preserves the useful properties
with reduced cross-term interference.
There are two types of lowpass filters, namely the linear type and the
nonlinear type. There are different types of linear filters. It is interesting to
note that all these linear representations can be written in a general form that
was introduced by L. Cohen1.

1. Interested readers should refer to Generalized Phase-Space Distribution Functions, L. Cohen, in the
Journal of Mathematical Physics, vol. 7, pp. 781-806, 1966.

LabVIEW Signal Processing Course Manual

13-8

National Instruments Corporation

Lesson 13

Joint Time-Frequency Analysis

Cohens Class
Linear filters fall into a general category known as Cohens class of filters.
With no filtering, the Cohens class distribution is the same as the
Wigner-Ville Distribution. In the JTFA Toolkit, two of the most commonly
used linear filters are implemented below. They are the cone-shaped
distribution and the Choi-Williams distribution.

Cone-Shaped Distribution
frequency

time

When the lowpass filter has a cone shape as shown above, the Cohens
class distribution is known as the cone-shaped distribution. The figure
below shows the cone-shaped distribution of the frequency hopper
signal.

National Instruments Corporation

13-9

LabVIEW Signal Processing Course Manual

Lesson 13

Joint Time-Frequency Analysis

Choi-Williams Distribution
frequency

time

When the lowpass filter has an exponential shape, as shown above, the
Cohens class distribution is known as the Choi-Williams distribution
(CWD). The figure below shows the Choi-Williams distribution of the
frequency hopper signal.

LabVIEW Signal Processing Course Manual

13-10

National Instruments Corporation

Lesson 13

Joint Time-Frequency Analysis

Both these linear filters give reduced cross-term interference at the

expense of poor resolution. When you select one of these two algorithms
using the JTFA selector control on the toolkit front panel (discussed in
detail in the next section), you are asked to set the value of a parameter.
This parameter allows you to balance the cross-term interference and
resolution. The bigger this parameter is, the less the interference;
however , the lower interference is at the cost of smeared resolution.
Start with a value of 0.5 for the parameter. The desired value for this
parameter is application dependent; however, it is always desirable to
choose this value so that less interference occurs.

Gabor Spectrogram
This section discusses the nonlinear type of low-pass filters described by the
time-frequency distribution series, also known as the Gabor Spectrogram1.
The advantage of the nonlinear type of filters is that it gives both good
resolution and small cross-term interference. The Gabor spectrogram
decomposes the Wigner-Ville distribution as DC plus a group of oscillated
time-frequency functions.

1. The Gabor spectrogram was invented by researchers at National Instruments and is a National
Instruments patent. It has won several technological awards. For more information on how to construct
such a filter, refer to S.Qian and D.Chen, Discrete Gabor Transform, IEEE Transactions on Signal
Processing, vol. 41, no. 7, July 1993, pp. 2429-2439

National Instruments Corporation

13-11

LabVIEW Signal Processing Course Manual

Lesson 13

Joint Time-Frequency Analysis

When you select the Gabor spectrogram using the JTFA selector on the
toolkit front panel, you are asked to choose the order of the filter. The
lower order Gabor spectrogram has less cross-term interference but
lower resolution. The higher order Gabor spectrogram has better
resolution but more cross-term interference. For order = 0, the Gabor
spectrogram is similar to the STFT when using Gaussian window
function. As the order gets larger, the Gabor spectrogram converges to
the Wigner-Ville distribution. However, it is computationally more
expensive. The best choice is usually order three to four. In this case, the
Gabor spectrogram not only has better resolution than the STFT, but also
possesses much less cross-term interference than the Choi-Williams and
Wigner-Ville distributions. You can choose the Gabor basis to be
wideband, mediumband or narrowband. The matching indicator tells
you how closely the basis function approximates your input signal. The
larger the value, the closer the match. The figure on the following page
shows the third-order Gabor Spectrogram of the Frequency Hopper
signal.

LabVIEW Signal Processing Course Manual

13-12

National Instruments Corporation

Lesson 13

Joint Time-Frequency Analysis

To summarize, you have examined different algorithms commonly used

for obtaining a joint time-frequency representation. There has been
minimal mathematical description of these algorithms. However, if you
have understood the discussion above, you are ready to use the JTFA
toolkit for different practical applications, as you will see in the next
section.

National Instruments Corporation

13-13

LabVIEW Signal Processing Course Manual

Lesson 13

Joint Time-Frequency Analysis

C. The JTFA Toolkit and Its Applications

With the JTFA Toolkit, you can use any of the aforementioned algorithms
to analyze stored data files and view the resulting spectrogram on an
intensity plot. It is a stand-alone application that you can use in real-world
applications such as radar and economic data analysis, where you need to
analyze a signal in the time and frequency domain simultaneously. The list
of such applications is endless. This section discusses simple applications.
These applications may seem rather specific, but the ideas and
methodologies behind them are of general interest to all JTFA users. The
figure below shows the front panel of the JTFA analyzer.

Acquiring Data
When you start the JTFA application, it prompts you to adjust the
parameters and then press Process. You first read in the signal to be
analyzed. This signal can be stored in an existing file, which must be either
a datalog format file (an internal file format used by LabVIEW) or an ASCII
text file. If your file is text, JTFA converts it to datalog format when you
click on the Read file button. You must convert the file to datalog, but you
can save the converted file under a new name to preserve your original file.
After conversion, the JTFA analyzer opens your file as a datalog file. On the
front panel, you see two more buttons labeled Set DAQ and Read DAQ. If
your computer is equipped with a National Instruments DAQ board, click

LabVIEW Signal Processing Course Manual

13-14

National Instruments Corporation

Lesson 13

Joint Time-Frequency Analysis

on the Read DAQ button; the JTFA analyzer starts collecting data from the
DAQ board and stores it a specified data file. When you first click on the
Read DAQ button, it prompts you to set the DAQ parameters. The
application then uses these parameters until you exit. You can use the Set
DAQ button to change the DAQ parameters. After you select the file to
analyze, the toolkit reads a block of data or frame of samples from the file.

Analyzing Data
You can select one of the different algorithms discussed in the previous
section to analyze your data using the JTFA selector control. Each control
has a different set of parameters. The default is set to the short-time Fourier
transform (STFT) algorithm. In this case, the parameters to be selected are
the window selector control and the window length control. The other five
choices are the Gabor Spectrogram, the Choi-Williams distribution, the
cone-shaped distribution, the Wigner-Ville distribution, and the Adaptive
Spectrogram1. The application then analyzes the samples and displays the
resultant spectrogram using an intensity plot. It also displays the traditional
spectrum to the right and the time-waveform below the two-dimensional
spectrogram.
The parameters start at and block length control the time range of the
analyzed signal. While the former determines the start time, the latter
determines the length of one frame of the analyzed signal. If the length
of the signal is less than start at + block length, the JTFA analyzer
processes whatever is available. Because of memory limitations, if the
length of the signal is too long, JTFA displays the longest time duration
allowed in a status indicator dialog box and stops processing.
The parameters start freq. and bandwidth control the frequency range
of the analyzed signal. start freq determines the lower boundary of the
frequency. The selection of the bandwidth is limited to 2-k * Nyquist
frequency, 0 k 5 . If start freq. + bandwidth is greater than the
Nyquist frequency, the JTFA ignores the setting of start freq. and
automatically sets the lower boundary of the frequency to zero (default
value).
By clicking on the subband button, you can select the preemphasis
control. You can use the preemphasis filter to reduce the influence of the
DC component and enhance the high-frequency component. The degree
of preemphasis is controlled by the preemphasis parameter. When this
value is equal to zero (default), there is no preemphasis. When this value
is equal to one, JTFA completely removes the DC component, and the

1. For a detailed discussion on the Adaptive Spectrogram, refer to Joint Time-Frequency Analysis: Methods and Applications, by Shie Qian and Dapang Chen, Prentice Hall, NJ, 1996.

National Instruments Corporation

13-15

LabVIEW Signal Processing Course Manual

Lesson 13

Joint Time-Frequency Analysis

frequency components in the vicinity of the Nyquist frequency are

approximately doubled.
On a final note, if you change the JTFA selector, subband, or any of
the parameters, you must recompute the spectrogram to see the effect of
the changes. To recompute the spectrogram, click on the Process button.
The JTFA computes the spectrogram using the values in the controls at
the time you click on the Process button.

Post Analysis of Data

After you have analyzed the data using the JTFA toolkit, you might want to
save the data plots to a file for later use. You can save the time waveform,
spectrum, and spectrogram as spreadsheet files. Select the plots that you
want to save (using the selector to the right of the Save button) and then
click on the Save button. You can then process the next frame of samples by
clicking on the Next button. After you finish analyzing one file, you may
close the current file and open a new one. To open a new file, click on the
Read file button and indicate the file that you want to open. The program
closes the original file automatically.
The following exercises will help you further understand how to use the
JTFA Toolkit.

LabVIEW Signal Processing Course Manual

13-16

National Instruments Corporation

Lesson 13

Joint Time-Frequency Analysis

Exercise 13-1
Objective:

To use the JTFA Toolkit to analyze a doppler signal

In this exercise, you will familiarize yourself with different controls in the JTFA
Toolkit. You will analyze a doppler signal and see the advantages of a JTFA
spectrogram over a normal frequency spectrum.

1. Open the front panel of the JTFA Toolkit as shown above.

2. You will analyze data in the file c:\jtfa\data\doppler.log.
Click on the Read file button and choose this data file. The name of
the data file can now be seen in the current file box.
3. Adjust some of the parameters. Change the bandwidth to 1/4 band.
You will experiment with all the different algorithms that can be
selected using the JTFA selector. First, select the STFT algorithm.
You will use the Hamming window and window length = 64. Use the
default values for all the other parameters.
4. After setting all parameters, click on the Process button.
Note:

Remember that the application will not start computing until you click on
the Process button.
5. After the computation is complete, three waveforms will appear on
the JTFA window. You will now analyze these waveforms. The time

National Instruments Corporation

13-17

LabVIEW Signal Processing Course Manual

Lesson 13

Joint Time-Frequency Analysis

waveform of the doppler signal can be seen in the bottom left part of
the figure. The standard Fourier spectrum can be seen on the top right
part. From this plot, you can see the range of frequencies present in
the spectrum. But the plot does not tell how these frequencies
evolved as a function of time. The larger plot in the upper left corner
is the time-dependent spectrum, a function of both time and
frequency. From it, not only can you see how the frequency changed
with time, but you also can see the intensity of the frequencies as
shown by the relative brightness levels of the plot.
6. You can now select different algorithms in the JTFA selector control
and see how the time-frequency spectrogram looks in each case.

End of Exercise 13-1

LabVIEW Signal Processing Course Manual

13-18

National Instruments Corporation

Lesson 13

Joint Time-Frequency Analysis

Exercise 13-2
Objective:

To study the application of detecting impulse signals in low-orbit satellites

using JTFA.
In this exercise, you will examine a very interesting application for joint
time-frequency analysis. This application involves the detection of impulse
signals by low-orbit satellites.
ionosphere
satellite
Earth

test
site

non-linear
chirp

Consider an impulse signal generated at some location on Earth. Such an

impulse may be caused by nuclear weapons testing, and hence the
detection and estimation of this signal is an important national security
issue. But the detection of this signal is not very easy. As this signal
passes through the dispersive media surrounding the Earth (for example,
the ionosphere), the signal becomes a nonlinear chirp signal.
Furthermore, it is severely corrupted by random noise from the
ionosphere, and hence the detection of this signal via standard Fourier
transform techniques is not possible. However, as you will see in this
example, JTFA helps to detect this signal properly.

National Instruments Corporation

13-19

LabVIEW Signal Processing Course Manual

Lesson 13

Joint Time-Frequency Analysis

1. Open the front panel of the JTFA Toolkit as shown above, if it is not
already open from the previous exercise.
2. You will analyze data in the file c:\jtfa\data\impulse.log.
Click on the Read file button and choose this data file. The name of
the data file can now be seen in the current file box.
3. Adjust some of the parameters. Change the panel color to gray using
the front panel switch. You will experiment with all the available
algorithms using the JTFA selector. First, select the Gabor
spectrogram. Use the default values for all the other parameters.
4. After setting all parameters, click on the Process button. Remember
that the application will not start computing until you have done this.
5. After the computation is complete, three waveforms will appear on
the JTFA front panel. The time waveform of the ionized impulse
signal can be seen in the bottom left part of the figure. As explained
earlier, after passing through dispersive media, such as the
ionosphere, the impulse signal becomes a nonlinear chirp signal with
added random noise. Because of the low signal-to-noise ratio, you
can hardly see this chirp signal in the time waveform. It is
completely hidden by the noise signal. The standard Fourier
spectrum can be seen on the top right part. From this plot, you can
see the range of frequencies that are present in the spectrum. The

LabVIEW Signal Processing Course Manual

13-20

National Instruments Corporation

Lesson 13

Joint Time-Frequency Analysis

larger plot in the upper left corner is the time-dependent spectrum, a

function of both time and frequency. From this plot, you can
immediately identify the presence of the chirp-type signal arching
across the joint time-frequency domain. Using such a representation,
you can do extensive postprocessing using the Analysis Library VIs
and mask the desired signal, as shown in figure below. You can then
apply the inverse transformation to recover the original time
waveform. The figure below compares the noisy signal and the
reconstructed signal.

6. You can now select different algorithms and see how the
time-frequency spectrogram looks in each case.
7. Exit the JTFA analyzer by pressing the Quit button.

End of Exercise 13-2

National Instruments Corporation

13-21

LabVIEW Signal Processing Course Manual

Lesson 13

Joint Time-Frequency Analysis

Summary:

The need to analyze a signal in time and frequency domains

simultaneously has led to the popularity of joint time-frequency
analysis.

The JTFA toolkit is an excellent tool to learn this technique and use
it in many real-world applications.

Short time-frequency transform, Gabor spectrogram, Wigner-Ville

distribution, Choi-Williams distribution, Cone-Shaped distribution,
and adaptive spectrogram are some of the algorithms used for joint
time-frequency analysis. The choice of algorithm depends on your
application.

Some of the interesting applications for JTFA are in radar, medical

imaging, and economic data analysis. In Exercise 13-2, you and
scientists at the Los Alamos National Laboratory made substantial
progress in detecting a radio frequency (RF) nonlinear chirp-type
signal in a noisy environment.

LabVIEW Signal Processing Course Manual

13-22

National Instruments Corporation

Lesson 13

Joint Time-Frequency Analysis

Notes

National Instruments Corporation

13-23

LabVIEW Signal Processing Course Manual

Lesson 13

Joint Time-Frequency Analysis

Notes

LabVIEW Signal Processing Course Manual

13-24

National Instruments Corporation

Lesson 14
Wavelet and Filter Banks
Designer
Introduction
Wavelets are being applied in a diverse range of applications for the analysis
of nonstationary signals. These applications include removing noise from
signals, detecting abrupt discontinuities, and compressing large amounts of
data. The design of wavelets is closely related to the design of filters. The
Wavelet and Filter Banks Designer provides you with the flexibility to
interactively design the filters, and hence wavelets, for your specific
application.

You Will Learn:

A. About signal representation through transforms: the Fourier
transform and the short-time Fourier transform (STFT).
B. About the wavelet transform.
C. About the applications where wavelet analysis is useful.
D. About the relationship between wavelets and filter banks.
E. How to use the Wavelet and Filter Banks Designer (WFBD) Toolkit
on both a 1D signal (an electrocardiogram) and a 2D signal (an
image of a fingerprint).

National Instruments Corporation

14-1

LabVIEW Signal Processing Course Manual

Lesson 14

Wavelet and Filter Banks Designer

A. Signal Representation
Before learning about the wavelet transform and its applications, you first
need to understand some basic concepts from linear algebra. In particular,
you need to be familiar with inner products and basis vectors.

Inner Product
The inner product of two vectors is a mathematical operation that
determines the similarity of these vectors with respect to each other. For
example, given the vectors x = (x1, x2), and y = (y1, y2), their inner
product is
<x,y> = x1*y1 + x2*y2
The inner product determines the similarity by calculating the projection
(or component) of one vector in the direction of the other. For example,
if you have two vectors, say v and x, the projection (denoted by p) of v
on x is given by
<v,x>
p = -------------x

where ||x|| is the magnitude of the vector x. If x is a unit vector,

||x|| = 1, and the projection of v on x is simply
p = <v,x>
The magnitude of the projection |p| is the length of the orthogonal
projection of v on a straight line having the same direction as the unit
vector x. The projection p itself could be either positive, negative, or
zero. This is illustrated in the figure below:
v
v

x
p=0

p>0

x
p<0

The larger the value of |p|, the more similar the vectors are to one
another, and vice versa. When the inner product is equal to zero, the two
vectors are perpendicular to each other. Another way to say this is that
the two vectors are orthogonal when their inner product is equal to zero.

Basis Vectors
Consider a vector in 2D. Any such vector can be written as the weighted
sum of two other unit vectors, i = (1,0) and j = (0,1). For example, the vector

LabVIEW Signal Processing Course Manual

14-2

National Instruments Corporation

Lesson 14

Wavelet and Filter Banks Designer

v = (4,1) can be written as v = 4*i + 1*j. This is shown graphically in the

figure below:

j
0

The vectors i and j are known as the basis vectors in 2D. Every other
vector in 2D can be written as a combination of these two basis vectors.
An important property of basis vectors is that they are perpendicular
(orthogonal) to each other. That corresponds to their inner product being
equal to zero. Also, their magnitudes are equal to one. For the basis
vectors i and j in 2D,
<i, j> = < (1,0), (0,1) > = 1*0 + 0*1 = 0
In other words, this means that they are linearly independent (see Lesson
8, Linear Algebra) and you cannot express one of them in terms of the
other.
For the vector v = (4,1), you can take its inner product with respect to the
unit vectors i and j:
<v,i> = 4*1 + 1*0 = 4 and
<v,j> = 4*0 + 1*1 = 1
This gives the projection of v on the unit vectors. Because <v,i> is
greater than <v,j>, you can say that the vector v is more similar to i than
to j. This is readily apparent from the figure above, where v is more
horizontal than vertical. (Note that because i and j are unit vectors, their
magnitude is equal to 1.0. Thus, in the above equations, we did not
specifically normalize by the magnitude.)
If you know the values of the inner product of v with the basis vectors i
and j, you can obtain v from the basis vectors by multiplying them with
the values of the corresponding inner products and adding up the results.
For the example given,
4*i + 1*j = 4*(1,0) + 1*(0,1) = (4,0) + (0,1) = (4,1) = v.
The important points to remember from the above discussion are:

National Instruments Corporation

Basis vectors are chosen to be orthogonal to each other.

14-3

LabVIEW Signal Processing Course Manual

Lesson 14

Wavelet and Filter Banks Designer

Any vector v can be broken down as the sum of weighted basis

vectors. This is achieved by taking the inner product of the vector
with each of the basis vectors.

The larger the value of the inner product, the greater the similarity of
the given vector with respect to that particular basis vector.

You can reconstruct the vector v by knowing the values of its inner
product with each of the basis vectors. The reconstruction is
achieved by multiplying the inner product with the corresponding
basis vector and adding the results.

Fourier Transform
Just as you can break up any vector in 2D in terms of a set of basis
vectors i and j, you can also break up a time signal s(t) in terms of a set
of basis functions. A well known popular example of basis functions in
signal analysis are the sines and cosines of the Fourier transform. These
sines and cosines are harmonically related to each other, which means
that their frequencies are all integer multiples of a fundamental
frequency. In addition, the sines and cosines are orthogonal (if you take
their inner product, the result is zero).
When you take the Fourier transform of a time signal s(t), you are
actually taking the inner product of the signal with each of the basis
functions. So you actually are seeing the similarity of the signal to each
of the sines and cosines of different frequencies. If you get a large value
of the inner product with a basis function at a particular frequency, it
means that quite a bit of that frequency is present in the signal. If the
inner product happens to be zero, it means that frequency is not present
in s(t).
Although the Fourier transform is a useful tool in obtaining information
about the frequency content of a signal, its disadvantage is that it is
unable to tell us when in time a particular frequency occurs. For
example, consider the two chirp signals shown in the top two plots of the
figure on the next page. The frequency of the top signal is increasing,
whereas that of the bottom signal is decreasing. However, both have
exactly the same magnitude of the Fourier transform, as shown in the
lowermost figure.

LabVIEW Signal Processing Course Manual

14-4

National Instruments Corporation

Lesson 14

Wavelet and Filter Banks Designer

Looking at the Fourier transform of a signal, you cannot tell how its
frequency changes with time. All that you observe is the frequency
content of the sampled signal in the given time interval. The drawback
of this form of traditional Fourier transform analysis is that it is only
useful for the analysis of stationary signals (that is, signals whose
frequency content does not vary with time).

National Instruments Corporation

14-5

LabVIEW Signal Processing Course Manual

Lesson 14

Wavelet and Filter Banks Designer

Windowed Fourier Transform or Short-Time Fourier Transform (STFT)

One method of obtaining time information (in addition to the frequency
information) with the Fourier transform is to break up the signal into small
time intervals before taking the Fourier transform. For example, suppose
you have 1000 samples of a signal collected over a period of 10 seconds (so
the sampling frequency is 100 Hz). You could take the Fourier transform of
the entire 1000 samples, as shown below:

1000 samples of a signal

0
0

200
2

400
4

600
6

800
8

1000
10

time
(secs)

Fourier
Transform

signal spectrum

amplitude

frequency

The result is a two-dimensional spectrum where the x-axis is the

frequency and the y-axis is the amplitude. The frequency resolution in
this case is f = fs/1000, where fs is the sampling frequency. For the
same sampling rate, to increase the frequency resolution (that is, reduce
f ), you need to increase the number of samples to more than 1000.

LabVIEW Signal Processing Course Manual

14-6

National Instruments Corporation

Lesson 14

Wavelet and Filter Banks Designer

However, you can also divide the signal into five equal time intervals of
two seconds each, and analyze the frequency content of each interval
separately. In each of these time intervals, you will have 200 samples of
the signal. You can then take the Fourier transform of successive
2-second time intervals, as shown below.

1000 samples of a signal

0
0

200
2

400
4

600
6

800
8

200 samples

200 samples

200 samples

200 samples

200 samples

Fourier
Transform

Fourier
Transform

Fourier
Transform

Fourier
Transform

Fourier
Transform

1000
10

time
(secs)

freq
color or grayscale
corresponds to
amplitude

10

time

spectra centered at 1, 3, 5, 7, and 9 seconds

The result is a 3D spectrum, where the x-axis represents time, the y-axis
represents frequency, and the z-axis (out of the page) represents the
amplitude of the frequencies during the time for which the transform was
taken. The z-axis can be in color or gray scale to represent the amplitude
values. This method of dividing the signal into smaller time intervals and
then taking the Fourier transform of successive intervals is known as the
short-time Fourier transform (STFT).
Note that each spectrum is obtained over a time interval of 2 seconds.
So, you know that any frequency component indicated in the spectrum
occurred somewhere within those 2 seconds. Thus this method provides
you with better time resolution than the first case, where you had an
interval of 10 seconds, somewhere during which a frequency shown in
the spectrum could have occurred. However, note that the frequency

National Instruments Corporation

14-7

LabVIEW Signal Processing Course Manual

Lesson 14

Wavelet and Filter Banks Designer

resolution is now f = fs/200, and is now five times larger than in the first
case.
Good time resolution is necessary for the rapidly changing
(high-frequency) parts of a signal, whereas good frequency resolution is
necessary for the smoothly varying (low frequency) parts of a signal. To
improve the frequency resolution, you can (for a given sampling
frequency) increase the size of the time interval during which you obtain
the samples, but then the time resolution is affected. If you improve the
time resolution by decreasing the time interval, the frequency resolution
is affected, because now you have fewer samples, so f is larger. Hence,
there is a trade-off between the time resolution and the frequency
resolution when using the STFT.

LabVIEW Signal Processing Course Manual

14-8

National Instruments Corporation

Lesson 14

Wavelet and Filter Banks Designer

B. The Wavelet Transform

The wavelet transform is another form of analysis where you can measure
the similarity of a signal to a set of basis functions. For the Fourier
transform, these basis functions were limited to being sines and cosines of
different frequencies. Also, they existed for all time. However, for the
wavelet transform, there are many possibilities of different shapes of basis
functions. In addition, the functions are time limited (they are nonzero only
within a finite time interval) and are usually irregular, with an average value
of zero. These basis functions used in the wavelet transform are called
wavelets. An example of a basis function for the Fourier transform and the
wavelet transform are as shown:
Basis function
for
Fourier
transform

Basis function
for
wavelet
transform

Using the wavelet basis functions, you can obtain both good time
resolution and good frequency resolution. You first select (or design) a
prototype wavelet known as a mother wavelet. This wavelet is then
either stretched or compressed in time to obtain other wavelets for
the basis. The wavelet obtained by stretching has a longer time
duration than the mother wavelet and is good for extracting the
low-frequency information of the signal. The wavelet obtained by
compressing has a shorter time duration than the mother wavelet and
is good for extracting the high-frequency information of the signal.
Thus, it is useful for temporal analysis. The scale change (compressing
or stretching) is done in powers of two. The figure below shows a mother
wavelet and two other wavelets obtained from it:

National Instruments Corporation

14-9

LabVIEW Signal Processing Course Manual

Lesson 14

Wavelet and Filter Banks Designer

mother wavelet

compress to 1/2

stretch by 2

An important step in wavelet analysis is the choice of wavelets for the

basis functions. As discussed, you want to see the similarity of the signal
under consideration to the basis functions. So depending on the signal
(that is, application), a particular shape of a wavelet may be a better
representation for the signal than another shape. Unlike the Fourier
transform, where the shapes of the basis functions are restricted to being
sines and cosines, in the wavelet transform you can have a large number
of possibilities. However, some restrictions do apply (such as the basis
functions being orthogonal to each other), but these are handled
automatically by the Wavelet and Filter Banks Designer (WFBD)
toolkit, and you, as the user, need not worry about them. The figure
below shows examples of some other wavelets.

LabVIEW Signal Processing Course Manual

14-10

National Instruments Corporation

Lesson 14

Wavelet and Filter Banks Designer

There are several wavelet packages available on the market today.

Whereas other wavelet packages have a fixed selection of wavelets from
which you can choose, the National instruments WFBD lets you design
your own wavelets to best suit your particular application. By
experimenting with the parameters on the Design Panel of the WFBD,
you can design different wavelets. Unlike the Fourier transform,
wavelets are specifically suited for the analysis of nonstationary signals
(signals whose frequency contents vary with time).
The National Instruments WFBD is an interactive design package. It
enables you to design wavelets for both 1D signals and 2D images. The
signal or image can be loaded from a file or acquired by using DAQ or
IMAQ hardware.

Comparison of the Wavelet Transform and Fourier Transform

There are several important differences between the wavelet and Fourier
transforms. The following table summarizes these differences.

Fourier Transform

National Instruments Corporation

Wavelet Transform

Basis functions

Sines and cosines

Wavelets

Assumption

Basis functions exist for all time

Basis functions are time

limited

Type of signal

Stationary

Nonstationary

Analysis

Same resolution at all

frequencies

Multiresolution

14-11

LabVIEW Signal Processing Course Manual

Lesson 14

Wavelet and Filter Banks Designer

C. Applications of Wavelets
The wavelet transform is useful for the analysis of nonstationary signals,
whose frequency contents vary with time. Examples of nonstationary
signals are biomedical (for example, electrocardiogram), music, turbulence,
seismic, sound, and vibration data that change slowly or abruptly. For
signals like these, information about when the changes occur can often be
very important.
You have seen that you can use the wavelet transform to have both good
time and good frequency resolution simultaneously. Wavelets with a
short time duration are good for extracting high-frequency information
(such as discontinuities and abrupt changes) from a signal, whereas
wavelets with a longer time duration are useful for low-frequency
analysis. Thus, it is possible to see both the forest and the trees.
Wavelets have been used in diverse fields covering a wide variety of
applications. Some of the practical uses of wavelets are:

Removing noise from a signal.

Feature extraction for use in pattern recognition and classification.

Detection of discontinuities.

Data compression (for example, images) that can be used to speed

image processing, implement faster modems, speed transfers on the
Internet, video conferencing, satellite image transmission, and
telecommunications.

Wavelets vs. Joint Time-Frequency Analysis

Both the WFBD and the JTFA toolkit are useful in analyzing nonstationary
signals. But when would you choose one over the other? The answer lies in
what you hope to achieve. The general rule is that if you are interested in
applications that require reconstruction, you should use wavelet analysis.
Examples of such applications are noise removal and data compression. In
noise removal, you want to reconstruct the original signal from a signal
containing both additive noise and the original signal. In data compression,
you want to reconstruct the original signal from a compressed version of the
signal. However, if you are more interested in obtaining insight into the
frequency contents of a signal, or in understanding the physical nature of the
process that generated the signal, the JTFA toolkit may be better for such
cases.

LabVIEW Signal Processing Course Manual

14-12

National Instruments Corporation

Lesson 14

Wavelet and Filter Banks Designer

D.Wavelets and Filter Banks

It turns out that the design of wavelets is closely related to the design of filter
banks. But what is a filter bank? It consists of a set of filters that breaks a
signal into different frequency ranges. Consider two filters: a lowpass filter,
G0, and a highpass filter, G1. You will use these filters to split a signal into
different frequency bands, as shown in the figure below:

10 kHz

G0

s[n]
G0

G1

5 kHz
2

s1[n]

s2[n]

s3[n]

2
G1

20 kHz

10 kHz

The input signal, s[n], consists of frequencies in the range of 0 - 20 KHz.

It is first lowpass filtered by G0 and highpass filtered by G1 around the
frequency of 10 KHz. The output of the highpass filter is s3[n] and
occupies a frequency range of 10 KHz - 20 KHz. The output of the
lowpass filter is in the range 0 - 10 kHz and is once again filtered, this
time around 5 KHz, by both a highpass filter and a lowpass filter, to
give the signals s2[n] and s3[n]. S2[n] occupies a frequency range of
5 KHz - 10 KHz, and s1[n] occupies a range of 0 - 5 KHz. In the figure,
the downward pointing arrow with a 2 next to it shows the process of
downsampling (decimation) by a factor of two. Because the frequency
span of the filtered signal has been reduced by half, you can safely
decimate and use only half the number of samples.
So, now the original signal s[n] has been broken down into three signals,
each occupying different frequency ranges shown below:

National Instruments Corporation

14-13

LabVIEW Signal Processing Course Manual

Lesson 14

Wavelet and Filter Banks Designer

s(t) 0 20 kHz

s1[n] 0 5 kHz

s2[n] 5 10 kHz

s3[n] 10 20 kHz

In practice, you do not need to stop after only two stages of filtering. You
could continue to split the lowpass filtered signal into even more
frequency bands, as shown below:

s[n]

G1

G1

G0

G0

G1

G0

G1

G0

Note:

Although the process of decimation has not been explicitly shown, it is

assumed.
The design of the filters is such that the outputs of each highpass filter are
approximations of the wavelet transform. The wavelet obtained from the
highpass filter corresponding to the lowest frequency band has the longest
time duration and could be the mother wavelet. It is thus able to extract
low-frequency information from the signal. As you go toward the higher
frequencies, the impulse response (that is, wavelets) from the highpass
filters are the compressed versions of the mother wavelet and have
correspondingly shorter time durations. Thus, they are more suitable for
extracting the high-frequency information (for example, discontinuities)
from the signal. All these wavelets together form the set of basis functions
to represent a signal.
The process of highpass filtering is equivalent to taking the inner
product of the signal with respect to the wavelet basis functions. The

LabVIEW Signal Processing Course Manual

14-14

National Instruments Corporation

Lesson 14

Wavelet and Filter Banks Designer

output of each highpass filter is the value of the inner product of the
signal with one of the wavelets. This value is the wavelet coefficient and
indicates the similarity of the signal with respect to the corresponding
wavelet.

s[n]

G1

wavelet coefficient

G0

G1

wavelet coefficient

G0

G1

G0

wavelet coefficient
G1

G0

wavelet coefficient
(impulse response
could be
mother wavelet)

If the filters that make up the filter bank satisfy certain conditions, it is
possible to reconstruct the original signal s[n] from s1[n], s2[n],
and s3[n].
s1[n] 0 - 5 KHz

s2[n] 5 - 10 KHz

s3[n] 10 - 20 KHz

s[n] 0 - 20 KHz

This can be achieved by using another set of lowpass filters, H0, and
highpass filter, H1, and adding their outputs together, as shown below:

National Instruments Corporation

14-15

LabVIEW Signal Processing Course Manual

Lesson 14

Wavelet and Filter Banks Designer

5 kHz
s1[n]

10 kHz
H0

20 kHz

H0
+
s2[n]

2
+

s[n]

H1
s3[n]

2
H1
10 kHz

This process of filtering is now equivalent to reconstructing the original

signal from the wavelet basis functions. (The upward pointing arrow
with a 2 next to it shows the process of upsampling by a factor of two.)
A special characteristic about these filter banks as compared to the
traditional filter banks is that the outputs of these filter banks can be used
to reconstruct the original signal s(t). The filter banks are said to have
perfect reconstruction. From the traditional filter banks, it is not always
possible to reconstruct the original signal. For reconstruction, the filters
being used need to satisfy certain conditions, and thus a special
relationship exists between the filters G0, G1, H0, and H1. Because of
this relationship, it becomes necessary only to design the lowpass filters,
G0 and H0. Knowing G0 and H0, the highpass filters G1 and H1 are
automatically designed by the software.
In the next section, you will see how designing wavelets with the help of
the WFBD involves designing the lowpass filters G0 and H0.

LabVIEW Signal Processing Course Manual

14-16

National Instruments Corporation

Lesson 14

Wavelet and Filter Banks Designer

E. Using the Wavelet and Filter Banks Designer

Using the WFBD package, there are three important steps involved in the
design of wavelets. Because of the relationship between wavelets and filter
banks, these steps basically boil down to selecting and designing the filters
for the filter banks.

Selecting the Type of Filter Bank

The first step consists of selecting a particular type of filter bank. There are
two types of filter banks that you can choose from, orthogonal and
biorthogonal. The orthogonal filter bank is a special case of a biorthogonal
filter bank. The main difference between the orthogonal and biorthogonal
filter banks is that the biorthogonal filters can be linear phase, but the
orthogonal filters cannot. So, if your application requires linear phase, then
you should not use the orthogonal filter banks. (Linear phase is important
for images, because the eye is very sensitive to changes in phase.)

Selecting the Type of Filter

The next step deals with designing the lowpass filters G0 and H0, and has
to do with designing a filter P0 which is the cascade of the two filters G0
and H0. The filter P0 is illustrated as shown:

P0

G0

H0

The final step involves splitting the filter P0 into G0 and H0. These three
steps are outlined below:
Step 1: Select the type of filter bank. The available choices are:

Orthogonal

Biorthogonal

Step 2: Select the type of filter for P0. The available choices are:

Maxflat

Positive equiripple

General equiripple

Any of the three types of filters mentioned in step 2 can be used with
biorthogonal filter banks. However, with orthogonal filter banks, you
can have only maxflat or positive equiripple filters, but not general
equiripple.

National Instruments Corporation

14-17

LabVIEW Signal Processing Course Manual

Lesson 14

Wavelet and Filter Banks Designer

There is no hard and fast rule, or even general guidelines, as to which

type of filter you should choose for any specific application. You will
need to experiment with the different choices to select a suitable one.
Step 3: Select the type of lowpass filter G0 and separate P0 into G0 and
H0. This step consists of assigning the zeros of P0 to the lowpass filters
G0 and H0. The user decides which zeros to assign to G0 and which to
assign to H0.

Which Type of Filter or Filter Bank Is a Good Choice?

Different selections in steps 1, 2, and 3 will result in different wavelets.
Although some guidelines have been given as to which choice is preferable
over the other, in general there is no hard and fast rule for a particular choice
of filter or filter bank. The choice of a suitable wavelet for a particular
application is usually done by a trial and error process. Thus, you will need
to experiment with the different combinations of available choices to obtain
the best wavelet for your particular application.

LabVIEW Signal Processing Course Manual

14-18

National Instruments Corporation

Lesson 14

Wavelet and Filter Banks Designer

The Design Panel

When you bring up the WFBD application from the Programs menu, you
get the following front panel.

Step 1:
select type of
filter banks

frequency
response of
lowpass filter G0
and highpass
filter G1

Step 2:
select type
of filter P0
(P0 is the
cascade of
the lowpass
filters G0 and H0)

distribution of zeros
of lowpass filters
G0 and H0
x: G0
o: H0

Step 3:
select type of
lowpass filter G0

An explanation of the plots and controls is provided in the above figure.

The figure also shows you the sequence of steps that you need to follow
while designing your wavelet.
In the following exercises, you will see how this panel is used for
analyzing an ECG signal (a 1D signal) and for data compression of
fingerprints (a 2D image).

National Instruments Corporation

14-19

LabVIEW Signal Processing Course Manual

Lesson 14

Wavelet and Filter Banks Designer

Exercise 14-1
Objective:

To design a wavelet that captures the important features of an ECG signal, but with
less than half the number of data points.
1. Launch the WFBD from the Programs menu.
2. The Design Panel appears. It gives you the power of designing
arbitrary wavelets for your particular application. The Design Panel
is shown below:

You have seen that the wavelet transform can be implemented as a

bank of filters that decompose a signal into multiple frequency
bands. It separates and retains the signal features in one or a few of
these subbands. Thus, one of the biggest advantages of using the
wavelet transform is that signal features can be easily extracted.
Because of their ability to effectively extract signal features, wavelet
transforms find many applications in the fields of data compression,
pattern recognition, edge detection, echo detection, speech
recognition, and texture analysis.

LabVIEW Signal Processing Course Manual

14-20

National Instruments Corporation

Lesson 14

Wavelet and Filter Banks Designer

The key to creating a successful wavelet application is to select an

appropriate wavelet, which is equivalent to selecting a good set of
filters in the filter bank. You saw that there are three major steps
involved in the design of the filters and filter banks. These steps are
repeated below.
Step 1: Select the type of filter bank. The available choices are
orthogonal or biorthogonal.
Step 2: Select the type of filter for P0. The available choices are
maxflat, positive equiripple, or general equiripple. Remember that
the general equiripple option is not available with orthogonal filter
banks.
Step 3: Select the type of lowpass filter G0 and separate P0 into G0
and H0. The separation is achieved by deciding which zeros to assign
to G0 and which to assign to H0.
The wavelet that you design will depend on the choices that you
select in steps 1, 2, and 3. Thus, you can experiment with the
different combinations to obtain the best wavelet for your particular
application.
3. In the Design Panel, select Menu 1D Data Test. A new window
(1D_Test) will appear. In this window, you will load a 1D signal (for
this example, you will use an ECG signal) having 3600 data points
(samples) and see how it can be represented with a smaller number
of samples.
4. In the new window, select Data Read from File txt file. Choose
the file Ecg.txt.
5. You will now see four plots on the screen. In the topmost plot, you
see the original ECG signal. It consists of 3600 data points
(samples). You want to preserve the main features of this signal by
using a minimum number of data points.
Note that the original ECG signal has been decomposed into three
parts by several stages of lowpass and highpass filtering. This is
shown in the plots labeled path1, path 2, and path 3. A 0 indicates a
stage of lowpass filtering, and a 1 indicates a stage of highpass
filtering. Referring back to the explanation in section D, the figures
in the plots labeled path1, path 2, and path 3 correspond to s3[n],
s2[n], and s1[n], respectively. The lowermost plot, obtained after two
stages of lowpass filtering, contains the main features of the original
ECG, but has only about 900 data points. Its size is therefore about
one-fourth of that of the original data. The other two plots basically
consist of only higher frequency noise and can be ignored.

National Instruments Corporation

14-21

LabVIEW Signal Processing Course Manual

Lesson 14

Wavelet and Filter Banks Designer

Therefore, the plot in path 3 can be used to represent the original

ECG, but with fewer data points.

6. Switch to the Design Panel. Experiment with the controls on the

design panel so as to change the wavelet being used, and note the
corresponding changes in the plots on the 1D_Test window.
7. In particular, change the type of filter from maxflat to positive
equiripple. Note that some features of the original signal appear in
the noise channels also. Now the plot labeled path 2 also contains
some feature of the original ECG. Thus, this is not a good design for
the wavelet, because now to represent the original signal, you cannot
use only the 900 sample points of the lowermost plot, because it does
not contain all the information of the original ECG signal. Some
of this information is also present in the plot labeled path 2.
8. To see the wavelet, select Menu Wavelet and Filters in the Design
Panel. For more details, refer to the WFBD reference manual.

End of Exercise 14-1

LabVIEW Signal Processing Course Manual

14-22

National Instruments Corporation

Lesson 14

Wavelet and Filter Banks Designer

Exercise 14-2
Objective:

To design a wavelet for reconstructing a fingerprint.

1. Open the WFBD Toolkit from the Programs menu, if it is not already
open from the previous exercise.
2. In the Design Panel, select MENU 2D Data Test. A new window
will appear with the title 2D_Test. In this window, you can load
two-dimensional signals (for example, images) on which to perform
wavelet analysis.
3. From the control in the lower right corner of the 2D_Test window,
select Data text and choose the file Finger.txt.
You will see two fingerprints on the upper half of the 2D_Test
window. On the top left figure, you will see an original fingerprint
consisting of 150x150 pixels. Law enforcement agencies have a need
to keep millions of fingerprints on file. This can obviously occupy a
lot of disk space. In addition, it would take an extremely long time to
transmit so much data over the network. So, wavelets can be used to
reduce the amount of data to represent the fingerprint. By an
appropriate choice of wavelets, you can reduce the number of data
points by over 50 percent.
The figure on the top right shows the image that has been
reconstructed by using the wavelet transform. The ratio of the
number of data points used for reconstruction to the number of data
points of the original image is shown in the Remaining Data control.
In the example shown, only 50 percent of the total number of points
(150x150) are used to reconstruct the original fingerprint. Note that
there is hardly any difference between the original image and the
reconstructed image.

National Instruments Corporation

14-23

LabVIEW Signal Processing Course Manual

Lesson 14

Wavelet and Filter Banks Designer

The lowermost figures show the images obtained after the process of
lowpass and highpass filtering (refer to section D). To recapitulate,
if the input image is s[n], then the resulting images after filtering are
s1[n], s2[n], s3[n], and s4[n], as shown below.
G1

s4[n]

G0

s3[n]

G1

s2[n]

G0

s1[n]

G1

s[n]

G0

Thus, s1[n] is obtained after two stages of lowpass filtering, s4[n] is

obtained after two stages of highpass filtering, and so on. If H
denotes the result of highpass filtering, and L denotes a result of

LabVIEW Signal Processing Course Manual

14-24

National Instruments Corporation

Lesson 14

Wavelet and Filter Banks Designer

lowpass filtering, then their relationship with the images on the

figure of the bottom half of the 2D_Test window is shown below:

LL
s1[n]

LH
s2[n]

HL
s3[n]

HH
s4[n]

4. Increase Data Used from 50 percent to 75 percent. Note that there is

not much improvement in the reconstructed image.
5. Decrease Data Used to 25 percent. Is there a difference between the
original and the reconstructed image? Experiment on the design
panel to get a good reconstructed image with Data Used set at
25 percent.
6. When you are done, exit the WFBD by selecting Menu Quit in the
Design Panel.

End of Exercise 14-2

National Instruments Corporation

14-25

LabVIEW Signal Processing Course Manual

Lesson 14

Wavelet and Filter Banks Designer

Summary
You have seen that the basis functions for the wavelet transform are known
as wavelets and are compressed and stretched versions of one another, with
a finite time duration. Wavelets with a short time duration are good for
high-frequency analysis, whereas those with a longer time duration are good
for low-frequency analysis. Using wavelets, it is possible to simultaneously
get good time as well as frequency resolution. Depending on your
application, a particular wavelet may be more suitable than another wavelet.
Hence, designing an appropriate wavelet is an important part of wavelet
analysis. The design of wavelets is closely related to the design of filter
banks. Using the Wavelet and Filter Banks Designer, you can interactively
design the filters in the filter bank and simultaneously monitor the effect
they have on the signal. Because of their ability for effective feature
extraction, wavelets are widely used for data compression.

LabVIEW Signal Processing Course Manual

14-26

National Instruments Corporation

Lesson 14

Wavelet and Filter Banks Designer

Notes

National Instruments Corporation

14-27

LabVIEW Signal Processing Course Manual

Lesson 14

Wavelet and Filter Banks Designer

Notes

LabVIEW Signal Processing Course Manual

14-28

National Instruments Corporation

Appendix A
Error Codes
Analysis Error Codes
Table A-1. Analysis Error Codes
Code

Name

Description

NoErr

No error; the call was successful.

20001

OutOfMemErr

There is not enough memory left to perform

the specified routine.

20002

EqSamplesErr

The input sequences must be the same size.

20003

SamplesGTZeroErr

The number of samples must be greater than

zero.

20004

SamplesGEZeroErr

The number of samples must be greater than

or equal to zero.

20005

SamplesGEOneErr

The number of samples must be greater than

or equal to one.

20006

SamplesGETwoErr

The number of samples must be greater than

or equal to two.

20007

SamplesGEThreeErr

The number of samples must be greater than

or equal to three.

20008

ArraySizeErr

The input arrays do not contain the correct

number of data values for this VI.

20009

PowerOfTwoErr

The size of the input array must be a power

of two:
size = 2m, 0 < m < 23.

20010

MaxXformSizeErr

The maximum transform size has been

exceeded.

20011

DutyCycleErr

The duty cycle must meet the condition:

0 duty cycle 100.

National Instruments Corporation

A-1

LabVIEW Signal Processing Course Manual

Appendix A

Error Codes

Table A-1. Analysis Error Codes (Continued)

Code

Name

Description

20012

CyclesErr

The number of cycles must be greater than

zero and less than or equal to the number of
samples.

20013

WidthLTSamplesErr

The width must meet the condition:

0 < width < samples.

20014

DelayWidthErr

The delay must meet the condition:

0 (delay + width) < samples.

20015

DtGEZeroErr

dt must be greater than or equal to zero.

20016

DtGTZeroErr

dt must be greater than zero.

20017

IndexLTSamplesErr

The index must meet the condition:

0 index < samples.

20018

IndexLengthErr

The index must meet the condition:

0 (index + length) < samples.

20019

UpperGELowerErr

The upper value must be greater than or

equal to the lower value.

20020

NyquistErr

The cutoff frequency, fc, must meet the

condition:
fs
0 f c --2

20021

OrderGTZeroErr

The order must be greater than zero.

20022

DecFactErr

The decimating factor must meet the

condition: 0 < decimating samples.

20023

BandSpecErr

The band specifications must meet the

condition:
f
s

0 f flow f high --2

20024

RippleGTZeroErr

The ripple amplitude must be greater than

zero.

20025

AttenGTZeroErr

The attenuation must be greater than zero.

20026

WidthGTZeroErr

The width must be greater than zero.

20027

FinalGTZeroErr

The final value must be greater than zero.

20028

AttenGTRippleErr

The attenuation must be greater than the

ripple amplitude.

20029

StepSizeErr

The step-size, , must meet the condition:

0 0.1.

LabVIEW Signal Processing Course Manual

A-2

National Instruments Corporation

Appendix A

Error Codes

Table A-1. Analysis Error Codes (Continued)

Code

Name

Description

20030

LeakErr

The leakage coefficient must meet the

condition:
0 leak .

20031

EqRplDesignErr

The filter cannot be designed with the

specified input values.

20032

RankErr

The rank of the filter must meet the

condition:
1 (2 rank + 1) size.

20033

EvenSizeErr

The number of coefficients must be odd for

this filter.

20034

OddSizeErr

The number of coefficients must be even for

this filter.

20035

StdDevErr

The standard deviation must be greater than

zero for normalization.

20036

MixedSignErr

The elements of the Y Values array must be

nonzero and either all positive or all
negative.

20037

SizeGTOrderErr

The number of data points in the Y Values

array must be greater than two.

20038

IntervalsErr

The number of intervals must be greater than

zero.

20039

MatrixMulErr

The number of columns in the first matrix is

not equal to the number of rows in the
second matrix or vector.

20040

SquareMatrixErr

The input matrix must be a square matrix.

20041

SingularMatrixErr

The system of equations cannot be solved

because the input matrix is singular.

20042

LevelsErr

The number of levels is out of range.

20043

FactorErr

The level of factors is out of range for some

data.

20044

ObservationsErr

Zero observations were made at some level

of a factor.

20045

DataErr

The total number of data points must be

equal to the product of the levels for each
factor and the observations per cell.

National Instruments Corporation

A-3

LabVIEW Signal Processing Course Manual

Appendix A

Error Codes

Table A-1. Analysis Error Codes (Continued)

Code

Name

Description

20046

OverflowErr

There is an overflow in the calculated

F-value.

20047

BalanceErr

The data is unbalanced. All cells must

contain the same number of observations.

20048

ModelErr

The Random Effect model was requested

when the Fixed Effect model was required.

20049

DistinctErr

The x values must be distinct.

20050

PoleErr

The interpolating function has a pole at the

requested value.

20051

ColumnErr

All values in the first column in the X matrix

must be one.

20052

FreedomErr

The degrees of freedom must be one or more.

20053

ProbabilityErr

The probability must be between zero and

one.

20054

InvProbErr

The probability must be greater than or equal

to zero and less than one.

20055

CategoryErr

The number of categories or samples must be

greater than one.

20056

TableErr

The contingency table must not contain a

negative number.

20061

InvSelectionErr

One of the input selections is invalid.

20062

MaxIterErr

The maximum iterations have been

exceeded.

20063

PolyErr

The polynomial coefficients are invalid.

20064

InitStateErr

This VI has not been initialized correctly.

20065

ZeroVectorErr

The vector cannot be zero.

LabVIEW Signal Processing Course Manual

A-4

National Instruments Corporation

Appendix A

Error Codes

G Math Toolkit Error Codes

Table A-2.
Error Code
Number
0

National Instruments Corporation

G Math Toolkit Error Codes

Error Code Description

No error

23001

Syntax error of parser

23002

Discrepancy between function, variables and

coordinates

23003

Number of contours out of range

23004

Number of color palettes out of range

23005

Negative distance

23006

Not a valid path

23007

Not a graphs file

23008

Wrong input, Euler method

23009

Wrong input, Runge Kutta method

23010

Wrong input, Cash Karp method

23011

Nonpositive step rate

23012

Nonpositive accuracy

23013

Matrix vector conflict

23014

A and X0 have different dimensions

23015

Empty X0

23016

Singular eigenvector matrix

23017

Multiple roots

23018

Left point is a root

23019

Right point is a root

23020

Left point greater than right point

23021

Both function values have the same sign

23022

Nonpositive accuracy or nonpositive delta x(h)

23023

Wrong dimension of start

23024

No root found

A-5

LabVIEW Signal Processing Course Manual

Appendix A

Error Codes

Table A-2.

G Math Toolkit Error Codes (Continued)

Error Code
Number

LabVIEW Signal Processing Course Manual

Error Code Description

23025

Nonvalid triplet (a,b,c)

23026

No optimum found

23027

Not exactly one variable

23028

Wrong model equation

23029

Levenberg Marquardt has failed

23030

m >= n >= 0 is violated or the matrix of derivatives has

the wrong dimension

23031

No valid point

23032

Maximum does not exist

23033

Vectors have different dimensions or empty vectors

23034

Ill conditioned system

23035

Nonpositive number

23036

Different parameters

23037

Not exactly two functions

23038

No variables in expression

23039

Parameter problem

23040

Derivative out of range

23041

Not exactly two variables

23042

Negative argument

23043

Argument out of range (0,1]

23044

Argument out of range [0,1]

23045

n<k

23046

Empty array

23047

Argument out of range [0,100]

23048

Invalid time increment

23049

Invalid window length

23050

Signal length not a multiple of number

A-6

National Instruments Corporation

Appendix A

Table A-2.

Error Codes

G Math Toolkit Error Codes (Continued)

Error Code
Number

Error Code Description

23051

Signal length not a power of two

23052

Signal length not a prime and 5

23053

Signal length not a power of two and 4

23054

Nonunique variables

Functions for Use with G Math Toolkit Parser VIs

Table A-3. Functions for Use with G Math Toolkit Parser VIs
Function

Corresponding G Math
Function name

Description

abs(x)

Absolute Value

Returns the absolute value of x.

acos(x)

Inverse Cosine

Computes the inverse cosine of x.

acosh(x)

Inverse Hyperbolic Cosine

Computes the inverse hyperbolic

cosine of x in radians.

asin(x)

Inverse Sine

Computes the inverse sine of x in

radians.

asinh(x)

Inverse Hyperbolic Sine

computes the inverse hyperbolic

sine of x in radians.

atan(x)

Inverse Tangent

Computes the inverse tangent of

x in radians.

atanh(x)

Inverse Hyperbolic Tangent

Computes the inverse hyperbolic

tangent of x in radians.

ci(x)

Cosine Integral

Computes the cosine integral of

x where x is any real number.

ceil(x)

Round to +Infinity

Rounds x to the next higher

integer (smallest integer x.)

cos(x)

Cosine

Computes the cosine of x in

radians.

cosh(x)

Hyperbolic Cosine

Computes the hyperbolic cosine

of x in radians.

cot(x)

Cotangent

Computes the cotangent of x in

radians (1/tan(x)).

csc(x)

Cosecant

Computes the cosecant of x in

radians (1/sin(x)).

National Instruments Corporation

A-7

LabVIEW Signal Processing Course Manual

Appendix A

Error Codes

Table A-3. Functions for Use with G Math Toolkit Parser VIs (Continued)
Function

Corresponding G Math
Function name

Description

exp(x)

Exponential

Computes the value of e raised to

the power x.

expm1(x)

Exponential(Arg)1

Computes the value of e raised to

x
the power of x 1 ( e 1 )

floor(x)

Round to Infinity

Truncates x to the next lower

integer (Largest integer x)

gamma(x)

Gamma Function

( n + 1 ) = n! for all natural

numbers n.

getexp(x)

Mantissa and exponent

Returns the exponent of x.

getman(x)

Mantissa and exponent

Returns the mantissa of x.

int(x)

Round to nearest integer

Rounds its argument to the nearest

even integer.

intrz

Round toward zero

Rounds x to the nearest integer

between x and zero.

ln(x)

Natural Logarithm

Computes the natural logarithm

of x (to the base e).

lnpl(x)

Natural Logarithm
(Arg + 1)

Computes the natural logarithm

of (x + 1).

log(x)

Logarithm Base 10

Computes the logarithm of x

(to the base 10).

log2(x)

Logarithm Base 2

Computes the logarithm of x

(to the base 2).

pi(x)

Represents the value

= 3.14159...

pi(x) = x*
pi(1) =
pi(2.4) = 2.4*

rand( )

Random Number (01)

Produces a floating-point number

between 0 and 1.

sec(x)

Secant

Computes the secant of x

(1/cos(x)).

si(x)

Sine Integral

Computes the sine integral of x

where x is any real number.

sign(x)

Sign

Returns 1 if x is greater than 0.

Returns 0 if x is equal to 0.
Returns 1 if x is less than 0.

LabVIEW Signal Processing Course Manual

A-8

National Instruments Corporation

Appendix A

Error Codes

Table A-3. Functions for Use with G Math Toolkit Parser VIs (Continued)
Function

Corresponding G Math
Function name

Description

sin(x)

Sine

Computes the sine of x in radians.

sinc(x)

Sinc

Computes the sine of x divided

by x in radians (sin(x)/x).

sinh(x)

Hyperbolic Sine

Computes the hyperbolic sine

of x in radians.

spike(x)

Spike function

spike(x) returns:
1 if 0 x 1
0 for any other value of x.

sqrt(x)

Square Root

Computes the square root of x.

square(x)

Square function

square (x) returns:

1 if 2n x ( 2n + 1 )
0 if 2n + 1 x ( 2n + 2 )
where x is any real number and
n is any integer.

step(x)

Step function

step(x) returns:
0 if x < 0
1 if any other condition obtains.

tan(x)

Tangent

Computes the tangent of x in

radians.

tanh(x)

Hyperbolic Tangent

Computes the hyperbolic tangent

of x in radians.

G Math Toolkit Parser Error Codes

Table A-4.
Error
Code
Number

National Instruments Corporation

G Math Toolkit Parser Error Codes

Error Code Description

Error Example

No error

sin(x)

Bracket problem at the

beginning

1+x)

Incomplete function
expression

sin(x)+

Bracket problem

()

Bracket problem at the end

(1+x

A-9

LabVIEW Signal Processing Course Manual

Appendix A

Error Codes

Table A-4.

LabVIEW Signal Processing Course Manual

G Math Toolkit Parser Error Codes (Continued)

Wrong decimal point

1,2 (US)

Wrong number format

1e3 instead of 1E3

Wrong function call

sin()

Not a valid function

sins(x)

Incomplete expression

x+

10

Wrong variable name

a11

11

Wrong letter

sin(X)

12

Too many decimal points

1.23.45

21

Contains more than one

variable

1+x+y4

22

Inconsistency in variables
or numbers

Depends on application

23

Contains variables

Depends on application

24

Variables output problem

Depends on application

A-10

National Instruments Corporation

Appendix B
Frequently Asked Questions
Background
1. Can the antialias filter be a hybrid filter?
When going from the analog to the digital domain, the antialias
filter must be an analog filter. In general, when going from a domain
P to a domain Q, the filter should be in domain P to avoid the
problem of aliasing in the domain Q. Thus, when sampling to go
from an analog to a digital signal, the filter should be an analog filter.
However, when performing decimation to reduce the sampling rate,
you are going from the digital domain (with a higher sampling rate)
to the digital domain (with a lower sampling rate). In this case, the
filter before the decimation is performed will be a digital filter.

Signal Generation
1. Why are there Pattern VIs, when the Wave VIs will do all that the
pattern VIs can and more?
The historical reason is that Pattern VIs were created first. Also, the
Pattern VIs are meant to be used to generate a block of data at a time,
whereas the Wave VIs can be called iteratively in a loop.
2. Why do I need normalized frequency? Why not just use the
common frequency unit of Hertz in the input controls and have the
VI automatically convert it internally into normalized frequency if
necessary?
Because some people think in terms of Hertz (cycles per second) and
some think in terms of cycles. Normalized frequency handles both.
3. What does it mean that the Wave VIs are reentrant?
You can have several of them on the same block diagram, and each
one will have its own code. They will all work independently of each
other.

National Instruments Corporation

B-1

LabVIEW Signal Processing Course Manual

Appendix B

Frequently Asked Questions

Signal Processing
1. What are the limits/problems of zero padding? Does it introduce
harmonics? Does it affect the frequency content of a signal in any
way?
Zero padding performs interpolation in the frequency domain. No, it
does not introduce harmonics. In addition to improving (that is,
lowering) the frequency resolution, faster algorithms (using the FFT
instead of the DFT) are possible.
2. What is the physical meaning of the cross power spectrum?
It is a measure of the similarity of two signals in the frequency
domain. It is the frequency-domain equivalent of the
cross-covariance function. It shows the joint presence of energy in
the two signals.

Windows
1. Why do different toolkits provide only some specific window
choices? Why not all window choices?
One reason is that the windows chosen are the most common
windows for that particular application. The other is that several of
the windows give very similar results, so there is no point in
including all of them.

Digital Filters
1. What is meant by linear phase?
Linear phase in digital filters means that the phase distortion is
nothing more than a digital delay. All input samples will be shifted
by some constant number of samples, so this phase change can be
easily fixed and/or modeled.
Nonlinear phase means that the individual sine waves that make up
the input signal get shifted in time by different amounts. This sort of
phase change is very difficult to work around. Some signals (like the
square wave) are very sensitive to this sort of phase distortion.
2. What does it mean when the IIR filters execute in place?
In place means that the input array space (memory locations) are
being reused as the output array space. In place usually implies lower
demands on memory.
3. Do the FIR filters have only zeros?
FIR filters do have zeros, but they also have poles at the origin. If you
take the Z-transform of an FIR filter, and rewrite it as the product of
factored terms, you will find that in addition to zeros, they also have
poles at the origin.

LabVIEW Signal Processing Course Manual

B-2

National Instruments Corporation

Appendix B

Frequently Asked Questions

Curve Fitting
1. What is the limit in LabVIEW and BridgeVIEW on the number of
parameters used for curve fitting?
There is no limit. However, keep in mind that the memory
requirements, as well as the time required to find the solution, will
increase with increasing number of parameters.
2. Some curve fitting algorithms allow you to give weight to certain
data points. Can I do that in the VIs from the advanced analysis
library of LabVIEW and BridgeVIEW?
No, it is not possible.
3. In performing a fit, is there any rule of thumb as to how many data
points to use?
Normally, you need at least one more than the number of parameters
for which you are trying to solve. But there is no such rule as to
whether you should use at least five times more, 10 times more, etc.
As an example, in performing a polynomial fit, the number of data
points to be used to obtain a good fit may be correlated to how
close the data is to the underlying polynomial (that is, how much
noise). It depends.
4. In the Levenberg-Marquardt VI, why is the derivative information
needed?
The information is needed to calculate the Jacobian, which is needed
in the algorithm to solve for the coefficients that you are trying to
determine. See the Analysis VI Reference Manual for details.

Digital Filter Design Toolkit

1. Is a demo version of the toolkit available?
Yes. In addition to being a separate toolkit, the DFD also ships with
the Signal Processing Suite (SPS). The entire SPS is available in
demo form also.
2. Is there a UNIX version of the DFD Toolkit?
No, it has not been released on UNIX.
3. In the Arbitrary FIR Design panel, if I select import from file and
then cancel the operation, I get an error, Error 43 occurred....
Why is that?
This is due to a bug that will be fixed in the next release.

Wavelets
1. When I do the decimation (by 2), what options do I have? Can I
throw away every even sample, every odd sample, the first half, the
second half, etc?
National Instruments Corporation

B-3

LabVIEW Signal Processing Course Manual

Appendix B

Frequently Asked Questions

As long as you filter the data before performing the decimation, it

does not matter whether you throw away every odd sample or every
even sample. However, you must be consistent. You cannot throw
away the odd samples at the beginning and then the even samples
later, or vice versa. Similarly, you cannot throw away every seventh
sample in the beginning, and then every fourth sample later. It is not
proper to remove the first half samples, or the next half samples.
Normally, just the odd or even samples are removed.
2. When I do the upsampling (by 2), what options do I have?
In upsampling by 2, you need to insert zeros so as to double the
number of samples. The zeros to be inserted could be in either the
odd samples position or the even samples position. Depending on the
manner in which downsampling was done, you would insert zeros in
the same place. So, if you removed the odd-numbered samples
during downsampling, you would insert them in the odd-numbered
places during upsampling. However, if you removed the
even-numbered samples during downsampling, you would insert
them in the even-numbered places.
3. In Wavelets 1D Data Test ekg.txt save the 2nd plot, if I see the
text file, all the numbers are zero because they were very small and
were not saved with a sufficiently high precision. So, how can I
change the format of saving the text file?
If you are using the wavelets executable, you cannot change the
format in which the text file is being saved. However, if you are
using the LabVIEW wavelet VI libraries, you can go into the block
diagram to change the format.

LabVIEW Signal Processing Course Manual

B-4

National Instruments Corporation

Appendix C
References
The following documents contain more detailed information about the
topics discussed in this course.

National Instruments Manuals

Digital Filter Design Toolkit Reference Manual

G Math Toolkit Reference Manual

JTFA Reference Manual

LabVIEW Analysis VI Reference Manual

LabVIEW Data Acquisition Course Manual

Third-Octave Analysis Toolkit Reference Manual

Wavelet and Filter Banks Design Reference Manual

National Instruments Application Notes

Application Note #023Digital Signal Processing Fundamentals

Application Note #041The Fundamentals of FFT-Based Signal

Analysis and Measurement in LabVIEW and LabWindows

Application Note #040Fast Fourier Transforms and Power

Spectra in LabVIEW

Application Note #091G MathA New Paradigm for

Mathematics

Application Note #097Designing Filters Using the Digital Filter

Design Toolkit

Advanced Engineering Mathematics, Erwin Kreyszig, Fifth Edition,

Wiley Eastern Limited, 1983

ANSI S1.11-1986 Specification for Octave-Band and

Fractional-Octave-Band Analog and Digital Filters

Other Documents

National Instruments Corporation

C-1

LabVIEW Signal Processing Course Manual

Appendix C

References

Designing Digital Filters, Charles S. Williams, Prentice-Hall, 1986

Digital Filters, R.W.Hamming, Third Edition, Prentice-Hall, 1989

Discrete-Time Signal Processing, Alan V. Oppenheim and Ronald

W. Schafer, Prentice Hall, 1989

The FFT: Fundamentals and Concepts, Robert W. Ramirez,

Prentice-Hall, 1985

Joint Time-Frequency Analysis: Methods and Applications, Shie

Qian and Dapang Chen, Prentice-Hall, 1996

Multirate Systems and Filter Banks, P.P.Vaidyanathan,

Prentice-Hall, 1993

Numerical Recipes, Press and Flannery and Teukolsky and

Vetterling, Cambridge University Press, 1986

On the Use of Windows for Harmonic Analysis with the Discrete

Fourier Transform, Frederic J. Harris, Proceedings of the IEEE,
Vol. 676, No. 1, January 1978, pp. 51-84.

Some Windows with Very Good Sidelobe Behaviour, Albert H.

Nuttall, IEEE Transactions on Acoustics, Speech, and Signal
Processing, Vol. ASSP-29, No. 1, February 1981, pp. 84-91.

Windows to FFT Analysis, Svend Gade and Henrik Herlufsen,

Sound and Vibration, March 1988, pp. 14-22.

LabVIEW Signal Processing Course Manual

C-2

National Instruments Corporation

Appendix D
Glossary

Prefix

Meaning

Value

M-

mega-

106

K-

kilo-

103

m-

milli-

10-3

micro-

10-6

n-

nano-

10-9

Numbers/Symbols
1D

One-dimensional.

2D

Two-dimensional.

Infinity.

Pi.

Delta. Difference. x denotes the value by which x changes from

one index to the next.

A
AC

Alternating current.

ADC

Analog to digital convertor, the hardware that converts an analog

signal into a digital signal.

analog signal

A signal whose values are continuous and are defined at each and
every instant of time.

National Instruments Corporation

D-1

LabVIEW Signal Processing Course Manual

Appendix D Glossary

ANSI

American National Standards Institute.

antialias filter

An analog lowpass filter used to limit the frequency of the analog

signal, before sampling, to less than the Nyquist frequency. This
prevents aliasing in the digital domain.

array

Ordered, indexed set of data elements of the same type.

B
bandpass filter

A filter that passes signals within a certain band of frequencies.

bandstop filter

A filter that attenuates signals within a certain band of frequencies.

BPF

Bandpass filter.

BSF

Bandstop filter.

Butterworth filter

A filter characterized by a smooth response at all frequencies, and

the absence of ripples in both the passband and the stopband. Due
to the absence of ripples, it is also known as a maximally flat filter.

C
Chebyshev filter

A filter characterized by ripples in the passband, but a smooth

response in the stopband.

Chebyshev II filter

See Inverse Chebyshev Filter.

coherent gain (CG)

The coherent gain is the zero frequency gain (or DC gain) of a

window. The rectangular (or uniform) window has the highest CG.

complex conjugate
transpose

A matrix operation consisting of taking the complex conjugate of

each element of the matrix and then transposing the resulting
matrix.

complex matrix

A matrix with at least one element that is a complex number.

condition number

A measure of how close the matrix is to being singular. The

condition number of a square nonsingular matrix is defined as
cond ( A ) = A

1
p

where p corresponds to the pth norm of the matrix. The condition

number can vary between 1 and infinity. A matrix with a large
condition number is nearly singular, while a matrix with a
condition number close to 1 is far from being singular.

LabVIEW Signal Processing Course Manual

D-2

National Instruments Corporation

Appendix D

curve fitting

Glossary

Technique for extracting a set of curve parameters or coefficients

from a data set to obtain a functional description of the data set.

D
DAQ

Data acquisition.

data acquisition

Process of acquiring data, typically from A/D or digital input

plug-in boards.

dB

Decibels.

DC

Direct current.

decibel

A logarithmic scale used to compress large amplitudes and expand

small amplitudes. It is given by
one dB = 10 log10 (Power Ratio) = 20 log10 (Voltage Ratio)

DFT

Discrete Fourier transform, the algorithm used to transform

samples of the data from the time domain into the frequency
domain.

difference equations

Equations that describe the operation of a system (for example, a

filter) in the discrete time domain.

digital signal

A signal that can take on only specific amplitude values that are
defined at discrete points in time.

discrete-time signal

A signal whose values are continuous in amplitude, but which is

defined only at discrete points in time.

DSP

Digital signal processing.

E
elliptic filter

A filter characterized by ripples in both the passband and the

stopband.

equivalent noise
bandwidth (ENBW)

The equivalent noise bandwidth of a window is the width of an

ideal rectangular response that will pass the same amount of noise
power as the frequency response of the window. The rectangular
(or uniform) window has the smallest ENBW.

National Instruments Corporation

D-3

LabVIEW Signal Processing Course Manual

Appendix D Glossary

F
FFT

Fast Fourier transform. A fast method for calculating the discrete

Fourier transform. It is used when the number of samples is a power
of two.

filter bank

A set of filters connected in parallel. Each filter may be tuned to a

different frequency range. The filters in the filter bank may or may
not have the same bandwidth.

FIR filter

Finite impulse response filter. A type of filter whose output

depends only on the current and past inputs. It is also known as a
nonrecursive filter.

forward coefficients

Forward coefficients are those that multiply inputs.

frequency response

The Fourier transform of the impulse response. It consists of two

parts, the magnitude response and the phase response. The
magnitude response is a plot of the magnitude of the frequency
response at different frequencies, whereas the phase response is a
plot of the phase of the frequency response at different frequencies.
(see Impulse Response)

G
Gaussian probability
density function

A density function that is completely characterized by its mean and

standard deviation and is given by
1 x 2
1
f ( x ) = -------------- exp --- ------------
2
2

where is the mean and is the standard deviation. It is also

known as the normal probability density function.

H
harmonic distortion

The distortion inherent in a nonlinear system that results in

generation of frequencies at its output that are harmonics of the
input frequency. The more the degree of nonlinearity of the system,
the higher the frequencies of the harmonics.

Hermitian matrix

A complex matrix whose complex conjugate transpose is equal to

the matrix itself. See also complex conjugate transpose.

highpass filter

A filter that passes frequencies above a certain cut-off frequency.

It passes high frequencies, but attenuates low frequencies.

LabVIEW Signal Processing Course Manual

D-4

National Instruments Corporation

Appendix D

HPF

Highpass filter.

Hz

Hertz, or cycles per second.

Glossary

I
IEEE

Institute of Electrical and Electronics Engineers.

IIR filter

Infinite impulse response filter. A type of filter whose output

depends not only on the current and past inputs, but also on past
outputs. It is also known as a recursive filter.

IMAQ

Image acquisition.

impulse

A signal that has a value of one at a particular time instant and zero
everywhere else.

impulse response

The response of a system to an input that is an impulse.

inf

Digital display value for a floating-point representation of infinity.

in place

In place means that the input array space (memory locations) are
being reused as the output array space. In place usually implies
lower demands on memory.

inverse Chebyshev filter A filter characterized by a smooth response in the passband, but
with ripples in the stopband.
I/O

Input/output. The transfer of data to or from a computer system

involving communications channels, operator input devices, and/or
data acquisition and control interfaces.

J
joint time-frequency
analysis (JTFA)

A method of analysis that simultaneously provides both time and

frequency information. It shows how the frequency spectrum of a
signal varies with time.

K
Levenberg-Marquardt

National Instruments Corporation

A general curve fitting algorithm used to estimate the coefficients

of a curve to fit a set of samples. It can be used for both linear and
nonlinear relationships, but is almost always used to fit a nonlinear
curve. This is because the general linear fit method is better suited
to linear curve fitting. The Levenberg-Marquardt method does not

D-5

LabVIEW Signal Processing Course Manual

Appendix D Glossary

always guarantee a correct result, so it is absolutely necessary to

verify the results.

L
linear phase

Linear phase in digital filters means that the phase distortion is

nothing more than a digital delay. All input samples will be shifted
by some constant number of samples, so this phase change can be
easily fixed and/or modeled. Nonlinear phase means that the
individual sine waves that make up the input signal get shifted in
time by different amounts. This sort of phase change is very
difficult to work around. Some signals (like the square wave) are
very sensitive to this sort of phase distortion.

lower triangular matrix

A matrix whose elements above the main diagonal are all zero.

lowpass filter

A filter that passes frequencies below a certain cut-off frequency.

It passes low frequencies, but attenuates high frequencies.

LPF

Lowpass filter.

LU decomposition

A method that factors a matrix as a product of an upper and a lower

triangular matrix.

M
matrix

Two-dimensional array.

MB

Megabytes of memory.

mother wavelet

A prototype wavelet, which is compressed or expanded in time to

derive other wavelets.

MSE

Mean squared error. The MSE is a relative measure of the residuals

between the expected curve values and the actual observed values.

N
NaN

Digital display value for a floating-point representation of not a

number, typically the result of an undefined operation, such as
log(-1).

norm

The norm of a vector or matrix is a measure of the magnitude of the

vector or matrix. There are different ways to compute the norm of
a matrix. These include the 2-norm (Euclidean norm), the 1-norm,
the Frobenius norm (F-norm), and the Infinity norm (inf-norm).

LabVIEW Signal Processing Course Manual

D-6

National Instruments Corporation

Appendix D

Glossary

Each norm has its own physical interpretation. The

LabVIEW/BridgeVIEW Matrix Norm VI can be used to compute
the norm of a matrix.
normal probability
density function

See Gaussian probability density function.

nonlinear phase

See linear phase.

nonrecursive filter

See FIR filter.

nonsingular matrix

Matrix in which no row or column is a linear combination of any

other row or column, respectively.

normalized frequency

A frequency (in Hertz) that is specified as a ratio with respect to the

sampling frequency (in samples/second). Its units are in
cycles/sample. However, if the frequency is given in terms of
cycles, then it is divided by the number of samples to convert it to
the normalized frequency.

Nyquist frequency (fN)

Half the sampling frequency, fN = fs/2, where fs is the sampling

frequency.

Nyquist theorem

A theorem stating that to recover an analog signal from its samples,

the sampling frequency should be at least twice the highest
frequency in the signal.

O
observation matrix (H)

A matrix used as an input to the General LS Linear Fit VI. If there

are N data points, and k coefficients (a0, a1, ...ak-1) for which to
solve, H will be an N-by-k matrix with N rows and k columns. Thus,
the number of rows of H is equal to the number of data points,
whereas the number of columns of H is equal to the number of
coefficients for which we are trying to solve.

octave

A doubling in frequency.

one-sided transform

A representation consisting of only the positive frequency (and

DC) components.

P
passband

National Instruments Corporation

The range of frequencies that are passed by a filter with a gain of

almost one (0 dB).

D-7

LabVIEW Signal Processing Course Manual

Appendix D Glossary

passband ripple

The amount of variation of the passband gain from unity (0 dB). It

is usually specified in dB. It is given by
A (f)
Ai ( f )

o
ripple (dB) = 20log 10 ------------

where Ai(f) and Ao(f) are the amplitudes of a particular frequency

before and after the filtering, respectively.
pattern VIs

These are signal generation VIs that do not keep track of the phase
of the signal that they generate each time they are called. The Chirp
Pattern VI is the only pattern VI that requires its frequency input
in terms of normalized frequency.

pole-zero plot

A plot showing the positions of the poles and zeros of a system. The
pole-zero plot is useful in determining the stability of the system.

power spectrum

The power spectrum of a signal gives you the power in each of its
frequency components. It can be calculated by squaring the
magnitude of the Fourier transform (DFT or FFT) of the signal.

Q
quality factor (Q)

A measure of how selective a bandpass filter is in passing frequencies

around the center frequency and attenuating unwanted frequencies. It
is defined as the ratio of the center frequency of the filter to its
bandwidth.
Q = fm/Bm
where fm is the center frequency, and Bm is the bandwidth of the
filter. Thus, for a fixed center frequency, the larger the bandwidth
the smaller the quality factor, and vice versa.

R
rank

The rank of a matrix A, denoted by (A), is the maximum number

of linearly independent columns in A. The number of linearly
independent columns of a matrix is equal to the number of
independent rows. So, the rank can never be greater than the
smaller dimension of the matrix. Consequently, if A is an n m
matrix, then
( A ) min ( n, m )

where min denotes the minimum of the two numbers. The rank of a
square matrix pertains to the highest order nonsingular matrix that
can be formed from it. So, the rank pertains to the highest order
matrix that we can obtain whose determinant is not zero. A square

LabVIEW Signal Processing Course Manual

D-8

National Instruments Corporation

Appendix D

Glossary

matrix is said to have full rank if and only if its determinant is

different from zero.
recursive filter

See IIR filter.

regression analysis

See curve fitting.

reverse coefficients

The reverse coefficients are those that multiply the outputs.

RHS

Right hand side.

ripple

A measure of the deviation of a filter from the ideal filter

specifications.

RMS

Root mean square.

S
sampling frequency

The number of samples acquired per second. Its units are

samples/second.

short-time Fourier
transform (STFT)

The term for taking a Fourier transform of shorter time intervals

of samples of a signal, rather on of the entire set of samples. Also
known as the windowed Fourier transform.

singular value
decomposition (SVD)

A method that decomposes a matrix into the product of three

matrices A = USV T , where U and V are orthogonal matrices, and
S is a diagonal matrix. SVD is useful for solving analysis problems
such as computing the rank, norm, condition number, and
pseudoinverse of matrices.

spectral leakage

A phenomenon where it appears as if energy has leaked out from

one frequency into another. It occurs because of the discontinuities
introduced when the sampled waveform is repeated periodically in
time. The larger the discontinuity, the more the leakage. Leakage
can be reduced by reducing the amplitude of the discontinuities.
The reduction is achieved by use of multiplying the time domain
waveform by a window function. Note that if there are an integer
number of cycles in the sampled waveform, there is no leakage.

step response

The response of a system to a step input.

stopband

The range of frequencies that are attenuated by the filter.

National Instruments Corporation

D-9

LabVIEW Signal Processing Course Manual

Appendix D Glossary

stopband attenuation

The amount of attenuation in the stopband of a filter, usually

specified in dB. It is given by
Ao ( f )
A(dB) = 20log 10 -----------Ai ( f )

where Ai(f) and Ao(f) are the amplitudes of a particular frequency

before and after the filtering, respectively.
SVD

Singular value decomposition.

symmetric matrix

A matrix whose transpose is equal to the matrix itself.

T
total harmonic distortion

A relative measure of the amplitudes of the fundamental to the

amplitudes of the harmonics. If the amplitude of the fundamental is
A1, and the amplitudes of the harmonics are A2 (2nd harmonic), A3
(3rd harmonic), A4 (4th harmonic), ...AN (Nth harmonic), then the
total harmonic distortion (THD) is given by
2

THD =

A 1 + A 2 + A 3 + ... + A N
------------------------------------------------------------------A1

When the THD is expressed as a percentage, it is known as the

percentage total harmonic distortion (%THD) and is given by
100 A

+A

+A

+ ... + A

N
1
2
3
% THD = ----------------------------------------------------------------------------------

A1

transient response

The transient that initially appears at the output of a filter when the
VI is run. The time duration of the transient depends on the order
of the filter. It can be eliminated by setting the init/cont control of
the VI to TRUE.

transition region

The region between the passband and the stopband where the gain
of the filter varies from one (0 dB) or almost one (in the passband)
to a very small value (in the stopband).

transpose

A matrix operation that consists of interchanging the rows and

columns of a matrix.

two-sided transform

A representation consisting of both the positive and negative (and

DC) frequency components.

U
upper triangular matrix

A matrix whose elements below the main diagonal are all zero.

LabVIEW Signal Processing Course Manual

D-10

National Instruments Corporation

Appendix D

Glossary

V
vector

One-dimensional array.

W
wavelets

The time-limited basis functions of the wavelet transform.

wavelet transform

The transform whose basis functions are time limited and are
known as wavelets. Using the wavelet transform, it is possible to
have both good time and good frequency resolution
simultaneously. The selection of a suitable wavelet is an important
consideration in the use of the wavelet transform.

wave VIs

Signal generation VIs that keep track of the phase of the signal that
they generate each time they are called. They require their
frequency input to be in terms of normalized frequency.

Wigner-Ville distribution A method of joint time-frequency analysis. A drawback of this

method is the presence of unwanted cross-terms between actual
signal frequency components. These cross-terms can be removed
by averaging.
window

A smoothing function applied to a time domain waveform, before

it is transformed into the frequency domain, so as to minimize
spectral leakage.

windowed Fourier
transform

See short-time Fourier transform (STFT).

Z
zero padding

National Instruments Corporation

Addition of zeros to the end of a sequence so that the total number

of samples is equal to the next higher power of two. When zero
padding is applied to a set of samples in the time domain, faster
computation is possible by using the FFT instead of the DFT. In
addition, the frequency resolution (f) is improved (made smaller)
because f = fs/N, where fs is the sampling frequency and N is the
total number of samples.

D-11

LabVIEW Signal Processing Course Manual

Appendix D Glossary

Notes

LabVIEW Signal Processing Course Manual

D-12

National Instruments Corporation