FREQUENCY ANALYSIS
by
R.B. Randall, B. Tech., B.A.
Revision September 19873rd edition 1st print
ISBN 87 87355 07 8
Pantedn OonmereK.Lateon & Son AVS -OK-2600 GlostrupFOREWORD
In the years since the First and Second Editions of this book were published a
great many developments have occurred; in particular there has been a huge
swing from analog to digital analysis techniques, even for day-to-day analysis.
The main reason for this development has been the increase in power and
reduction in cost of FFT (Fast Fourier Transform) analyzers, but digital filters
have also increasingly taken over some of the application areas formerly
covered by analog filters.
At the same time it has been realised that the primary distinction is not
between digital and analog analysis, but rather between the FFT process, where
the signal is analyzed in blocks, and analysis using filters, where the signal is
Processed continuously. Digital filters are a special case of the latter, but
otherwise have very similar properties to analog filters. For this reason the
layout of the book has been modified considerably, with digital filters generally
being treated along with analog filters, and FFT techniques treated separately.
The basic theory has been expanded to include a discussion of the Hilbert
transform, and in particular how a model of frequency analysis, based on an
analytic signal, relates to the model promulgated earlier, which is based on a
sum of positive and negative frequency phasors. In particular in the case of
modulated signals (involving amplitude, phase and frequency modulation) the
approach using an analytic signal is useful, because its amplitude represents
the amplitude modulation signal (plus DC offset) while its phase represents the
phase (or frequency) modulation signal superimposed on the carrier compo-
nent.
The analysis of stationary signals, using filters and FFT techniques, respec-
tively, is treated in Chapters 3 and 4. Despite rearrangement, much of the
material is similar to that in the earlier editions, although the discussion of FFT
techniques has been expanded somewhat. At the same time, obsolete material
on “Time Compression” and “High Speed” analysis has been reduced consid-
erably, or dropped.Similarly, Chapter 5, on the analysis of short transients, contains much
material from the earlier editions, in particular with respect to analog analysis,
but the section on FFT techniques has been expanded considerably, in line with
the overwhelming preponderance of these methods today.
The subject of Chapter 6, the analysis of more slowly varying non-stationary
signals, was treated very briefly in the earlier editions, but the treatment here
has been expanded greatly. Only FFT analysis is covered, as this is by far the
most practical way of applying the “time windows” involved. This chapter also
contains new material on “order tracking” where the sampling frequency of the
FFT analyzer is tied directly to machine shaft speed, and the x-axis becomes
one of harmonic order, rather than frequency.
Chapters 7 and 8 also represent considerable expansions of material treated
briefly in the earlier editions. Chapter 7 covers dual channel FFT analysis, while
Chapter 8 covers Cepstrum analysis. In the intervening years there have been
considerable developments in both fields, and for this reason there is some-
what more emphasis on examples of application in these two chapters com-
pared with earlier chapters.
Overall, it is hoped that the new arrangement will provide more direct access
to the somewhat larger body of material.
R.B. RandallCONTENTS
INTRODUCTION .....
THEORETICAL FREQUENCY ANALYSIS ...
2.1. BASIC CONCEPTS
Complex Notation
Delta Functions ..
2.2. FOURIER ANALYSIS
Fourier Series
Fourier Transform
Sampled Time Functions
Discrete Fourier Transform
Spectral Density
Power vs. Energy .
2.3. BANDWIDTH
2.4, SIGNAL TYPES ..
Stationary Deterministic Signals
Stationary Random Signais
Pseudo-random Signals
Transient Signals ..
Non-Stationary Signals
2.5. CONVOLUTION ..
Convolution with a Delta Function
Convolution of Spectral Functions
Convolution Theorem ..
2.6. THE HILBERT TRANSFORM.
Analytic Signals
Modulation3. FILTER ANALYSIS OF STATIONARY SIGNALS ...
3.1.
3.2,
3.4.
3.5.
3.6.
. RECORDERS ......
FILTERS
Practical Filter Characteristic
Filter Response Time
DETECTORS .. 87
Squaring .. . 88
Averaging 90
Square Root Extraction . 97
Logarithmic Conversion - 99
Detector Response .. . 99
AC Recording
DC Recording
ANALOG ANALYZER TYPES
Discrete Stepped Filter Analyzers
Sweeping Filter Analyzers...
Real-time Parallel Analyzers
Time Compression Analyzers
DIGITAL FILTERING
The Z-transform
The General Multiple-Pole Filter...
The Briel & Kjzer Digital Frequency Analyzer Type 2131 .
1h2-octave Analysis
Filter Characteristics .
PRACTICAL ANALYSIS OF STATIONARY SIGNALS ..
Stepped Filter Analysis
Averaging Time T,
DC or AC Recording 127
Writing Speed 127
Paper Speed 127
Swept Filter Analysis . 130
Constant vs. Constant Percentage Bandwidth
Linear vs. Logarithmic Frequency Scale
Choice of Bandwidth
Averaging Time T,
Filter Dwell Time 7)
Writing Speed W
Paper Speed P ..
Summary ...
Calibration ..
Digital filter analysis
Linear vs. Exponential Averaging .
Averaging Time T, ...4. FAST FOURIER TRANSFORM (FFT) ..
4.1. THE FFT ALGORITHM
4.2. THE FFT FOR REAL-VALUED SER
4.3. LIMITATIONS OF FFT PROCESSING
Aliasing
Window Effects
Picket Fence Effect .
4.4. ZOOM FFT ..
Real-time Zoom ....
Non-destructive Zoom
Comparison of Zoom Techniques ..
4.5. PRACTICAL ANALYSIS OF STATIONARY SIGNALS
Analysis Parameters
Calibration and Sc:
Averaging
Real-Time Analysis .
4.6 HILBERT TRANSFORM TECHNIQUES ..
Amplitude Demodulation
Phase Demodulation ..
5. TRANSIENT ANALYSIS .
5.1. FFT TECHNIQUES
Short transients
Longer transients
5.2. FILTER ANALYSIS TECHNIQUES
Digital Filter Analysis of Transients
Sequential Filter Analysis...
Method A ..
Method B ..
Method C ..
Examples
6. NON-STATIONARY SIGNALS ..
6.1. SCAN ANALYSIS ..
Choice of analysis parameters .
Cyclic signals
Spectral Variations with Machine Speed
6.2. ORDER TRACKING ANALYSIS
Optimum Choice of Internal Filter
Zoom tracking ....7. DUAL CHANNEL ANALYSIS . caer
7.1. CROSS SPECTRUM ..
Definitions and Calculation Procedures
Applications .........
7.2. COHERENCE .....
Definition and Properties
Applications ...
7.3. FREQUENCY RESPON: ‘UNCTIONS ..
Definitions and Calculation Procedures
Effects of Noise on H, and Hp .
Effects of Leakage on H, and H, ...
Applications of Frequency Response Functions
7.4. CORRELATION FUNCTIONS ...
Definitions and Calculation Procedures ....
Properties and Applications of Correlation Functions
Autocorrelation Function...
Cross Correlation Function ..
7.5. IMPULSE RESPONSE FUNCTIONS ..
Definitions and Calculation Procedures
Properties and Applications ..
228
228
230
230
245
250
252
252
256
256
259
263
263
265
8. CEPSTRUM ANALYSIS
8.1. DEFINITIONS AND CALCULATION PROCEDURES .. 272
8.2. PROPERTIES OF THE CEPSTRUM 275
8.3. APPLICATIONS OF THE POWER CEPSTRUM 282
Echo Detection and Removal ... 282
283.
284
Properties of a reflecting surface
Speech Analysis .
Machine diagnostics .
8.4, PRACTICAL CONSIDERATIONS WITH THE POWER CEPSTRUM 291
8.5. THE COMPLEX CEPSTRUM . 297
8.6. APPLICATIONS OF THE COMPLEX CEPSTRUM .. 299
Echo Removal . 299
Deconvolution ... - 303
APPENDIX A — Mathematical Details ...
APPENDIX B — Mean Square Error for Narrow
Band Random Noise. 318
APPENDIX C — Response of Physical Systems
REFERENCES1. INTRODUCTION
The object of frequency analysis is to break down a complex signal into its
components at various frequencies, and it can be shown that this is possible for
all practical signals (Ref. 1.1). The word “components” can be interpreted in
several different ways, however, and it is important to clarify the differences
between them.
Mathematicians and theoretically inclined engineers tend to interpret “com-
ponents” as the results of a Fourier Analysis, while practical engineers often
think in terms of measurements made with filters tuned to different frequencies.
It is one of the purposes of this book to clarify the relationships between these
two approaches. Another is to give a simple pictorial interpretation of Fourier
analysis which greatly facilitates the understanding of fundamental relation-
ships, in particular the connection between mathematical theory and practical
analysis by both analog and digital means.
The approach taken is to consider frequency components as vectors (or
phasors) rotating in a complex plane rather than as sums of sines and/or
cosines. A typical (co-) sinusoidal component is represented initially as the sum
of two contrarotating vectors (representing positive and negative frequency,
respectively). This gives a meaning to the mathematically useful concept of
negative frequency, and a two-sided frequency domain gives a valuable sym-
metry with the time domain. It also allows for the simple pictorial representation
of Fourier analysis which is developed and used extensively in this book.
Later, in connection with a discussion of the Hilbert transform, an alternative
representation is developed, based on the projection on the real axis of a single
rotating vector. This gives a one-sided frequency spectrum, but results in a
complex time signal (analytic signal), whose real and imaginary components
are related by a Hilbert transform. This representation is very useful in the
analysis of modulated signals (amplitude, phase, and frequency modulation).
In both approaches a considerable simplification of the mathematics arises
from the fact that a vector (having two components e.g. amplitude and phase)
can be represented as a single complex variable and because differentiation
WWand integration of such complex variables are so simple. It is assumed that the
reader is familiar with the basic relationships between complex exponentials
and sinusoids, but a brief résumé of the most important relationships is given at
the start of Chapter 2.
Chapter 2 continues with a discussion of Fourier analysis, first the various
forms taken by the Fourier transform, then the important practical consider-
ation of bandwidth. This leads into a discussion of the different types of signal
encountered in practice, and how they appear in spectral and other representa-
tions. Next, the very important subject of convolution is treated in some detail.
It is shown how the output of a linear physical system is obtained by convolving
the input signal with the impulse response of the system and how this rather
licated operation transforms by the Fourier transform to a multiplication
(which in turn becomes an addition in the normal logarithmic representation of
spectra). Other applications of this so-called “Convolution Theorem” are also
given. The chapter closes with an introduction to the Hilbert transform, how it
can be calculated, and its application to the analysis of modulated signals.
Chapter 3 is entitled “Filter Analysis of Stationary Signals” and covers both
traditional analysis with analog filters and detectors, and analysis with digital
filters (and detectors) which have many similarities with their analog counter-
parts. The chapter starts with a discussion of the function and properties of the
basic elements of an analyzer based on filters, viz., the filter, the detector (for
measuring the power in the filter output signal) and the graphic recorder often
used to record the spectrum on paper. Any one of these elements can deter-
mine the speed with which the analysis can be made. The chapter goes on to
describe the various types of analyzers available, encompassing those based
on both analog and digital filters. Finally, the practical analysis of stationary
signals is discussed in detail, and formulae and procedures developed for the
optimum choice of analyzer and analysis parameters such as filter bandwidth,
averaging time, and recorder paper speed and writing speed. This is done for
stepped filter analysis (e.g. 1/s-octave and ‘/1-octave filters), then sweeping
narrow-band filters, and finally for real-time analysis using digital filters. Cali-
bration of the results is also clarified. These procedures allow determination of
results with a known degree of accuracy in the shortest possible time.
Chapter 4 introduces another digital analysis method, the so-called Fast
Fourier Transform (FFT), which since its introduction in 1965 has revolutionized
signal analysis. It is first shown how the FFT procedure greatly speeds up the
calculation of the DFT (Discrete Fourier Transform) introduced in Chapter 2,
and how the results obtained using the DFT differ from those obtained by direct
Fourier transform. Precautions have to be taken to avoid problems with the
three “pitfalls” of the DFT, viz., “aliasing”, “leakage”, and the “picket fence
effect”. The principles of “Zoom” analysis are explained and compared for the
two major techniques used to obtain increased resolution in a limited frequency
band. Finally, procedures are given for the practical analysis of stationary
12signals, including normal spectrum analysis, and analysis of demodulated sig-
nals obtained using Hilbert transform techniques. In Chapter 4 the discussion is
limited to simple spectrum analysis of a single channel signal, for direct com-
parison with the methods of the previous chapter. It is seen that the FFT is best
adapted to constant bandwidth analysis on a linear frequency scale.
Chapter 5 discusses the analysis of transient signals, firstly using FFT tech-
niques, and then using filters. Analysis using analog filters has now largely been
superseded by digital techniques (FFT and digital filters) but is included for the
sake of completeness. Even though many analyzers give results in RMS or
mean square (i.e. “power”) values, the conversion of the results to “energy
spectral density” (ESD), or “band energy \evel” is described for all analysis
techniques.
Chapter 6 covers the analysis of non-stationary signals such as speech, or
machine vibration signals obtained during run-ups and run-downs in speed. The
approach used employs time windows which are moved along the signal in
overlapping steps so that the signal in each window position is either quasi-
stationary, or an isolated event. The results are usually displayed in a 3-
dimensional spectral map, with time or shaft RPM as the third axis. For cyclic
signals, such as those from reciprocating machines, the third axis may be crank
angle (over 360° or 720°) and the possibility of averaging over several cycles is
explained. Order analysis (where the frequency axis is changed to one of
harmonic order) is also described. Because the analyzer sampling frequency is
then variable, being linked to shaft speed, problems may be encountered with
aliasing, and a range of measures to avoid this are described.
Chapter 7 is a brief introduction to multiple channel FFT analysis, where the
signals are analyzed two at a time. These techniques are usually employed to
determine the response properties of physical systems (e.g. structural, acousti-
cal, electrical systems) rather than from an interest in the signals themselves.
Dual channel analysis involves the calculation of a number of functions not
previously defined, viz., cross spectrum, coherence, and frequency response
functions, in the frequency domain, and autocorrelation, cross correlation, and
impulse response functions in the time domain. The definitions, properties,
calculation procedures and major applications of these functions are described
and illustrated with examples.
Chapter 8 is a discussion of Cepstrum analysis, which has been included
because it can readily be carried out in modern FFT analyzers, and because it
throws light on the material of earlier chapters, both with respect to signal
analysis, and system analysis. The cepstrum is the inverse Fourier transform of
a logarithmic spectrum and often extracts valuable information from the latter
in the same way that normal Fourier analysis extracts information from a time
signal. There are two types of cepstrum, the “power cepstrum” and the “com-
plex cepstrum”. The former is obtained from the power spectrum, while the
13latter is derived from a complex spectrum and involves both log amplitude and
phase information at each frequency. For this reason the process of calculating
the complex cepstrum is reversible back to a time signal. Some applications of
the cepstrum are based on its simple ability to detect periodic structures in the
spectrum (e.g. harmonics, sidebands and echoes); others are based on the fact
that source and transmission path effects, convolved together in the time
domain, are additive and often separable in the cepstrum, thus allowing decon-
volution by selective editing.
For the somewhat more novel topics of the later chapters, some examples of
applications have been given, but for simple frequency analysis not much has
been said about applications; it being assumed that a reader already had a
need for frequency analysis, and only required information on how to obtain the
results. Even so, it is perhaps desirable to give a brief survey of the major
reasons for the widespread use of frequency analysis.
Without a doubt the major reason is that many physical and biological
systems only respond to a limited range of frequencies. For example, the
human ear only perceives sound within the range 20 Hz — 20 kHz, while for
estimating the effects of “whole body vibration” the range is rather more
restricted (typically 1 — 80 Hz). Quite often physical systems, be they electrical
circuits or civil engineering structures, only have strong resonances or excita-
tions within a limited frequency range, and hence the study of their behaviour
when subjected to an external excitation can be restricted to this frequency
range, thus greatly simplifying the problem. As mentioned previously, the calcu-
lation of the response of a linear system reduces in the frequency domain to a
simple multiplication. An input at a given frequency gives an output only at that
same frequency, and thus frequency analysis permits treating each frequency
individually without regard to what is happening at other frequencies.
Sometimes frequency analysis is used to make a picture clearer. Quite often
the “noisiness” of a signal is contained mainly in its phase spectrum, the power
spectrum being relatively stable, and quite often it is the power spectrum which
gives most information with respect to “average“ effects.
The frequency content of a signal can often be used diagnostically; for
example in tracing the source of a noise problem to a particular machine, or of
a mechanical problem to a particular rotating component.
On the other hand it must be kept in mind that frequency analysis is not a
universal panacea. In enhancing periodic events it tends to smear out individual
‘ones; for example, one obt information about toothmeshing frequencies i
a gearbox but tends to lose information about individual gear teeth.