(Download PDF) A History of Cepstrum Analysis and Its Application To Mechanical Problems Robert B Randall Full Chapter PDF

A history of cepstrum analysis and its
application to mechanical problems

Robert B. Randall
Visit to download the full and correct content document:
https://ebookmass.com/product/a-history-of-cepstrum-analysis-and-its-application-to-
mechanical-problems-robert-b-randall/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
Spatiotemporal Analysis of Air Pollution and Its

Application in Public Health Lixin Li
https://ebookmass.com/product/spatiotemporal-analysis-of-air-
pollution-and-its-application-in-public-health-lixin-li/
Freedom of Speech and Expression: Its History, Its

Value, Its Good Use, and Its Misuse Richard Sorabji
https://ebookmass.com/product/freedom-of-speech-and-expression-
its-history-its-value-its-good-use-and-its-misuse-richard-
sorabji/
Overdoing Democracy: Why We Must Put Politics In Its

Place Robert B. Talisse
https://ebookmass.com/product/overdoing-democracy-why-we-must-
put-politics-in-its-place-robert-b-talisse/
The Aqueous Chemistry of Polonium and the Practical

Application of its Thermochemistry 1st Edition Susan A.
Brown
https://ebookmass.com/product/the-aqueous-chemistry-of-polonium-
and-the-practical-application-of-its-thermochemistry-1st-edition-
susan-a-brown/
Modeling and analysis of modern fluid problems Zhang
https://ebookmass.com/product/modeling-and-analysis-of-modern-
fluid-problems-zhang/
Truth to Power: A History of the U.S. National

Intelligence Council Robert Hutchings
https://ebookmass.com/product/truth-to-power-a-history-of-the-u-
s-national-intelligence-council-robert-hutchings/
Groundwater Sustainability: Conception, Development,

and Application Robert E. Mace
https://ebookmass.com/product/groundwater-sustainability-
conception-development-and-application-robert-e-mace/
The History of Texas 6th Edition Robert A. Calvert
https://ebookmass.com/product/the-history-of-texas-6th-edition-
robert-a-calvert/
You: A Natural History William B. Irvine
https://ebookmass.com/product/you-a-natural-history-william-b-
irvine/
Mechanical Systems and Signal Processing 97 (2017) 3–19
Contents lists available at ScienceDirect
Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/ymssp
A history of cepstrum analysis and its application to mechanical

problems
Robert B. Randall
School of Mechanical and Manufacturing Engineering, University of New South Wales, Sydney 2052, Australia
a r t i c l e i n f o a b s t r a c t
Article history: It is not widely realised that the first paper on cepstrum analysis was published two years
Received 9 August 2016 before the FFT algorithm, despite having Tukey as a common author, and its definition was
Received in revised form 13 December 2016 such that it was not reversible even to the log spectrum. After publication of the FFT in
Accepted 20 December 2016
1965, the cepstrum was redefined so as to be reversible to the log spectrum, and shortly
Available online 30 December 2016
afterwards Oppenheim and Schafer defined the ‘‘complex cepstrum”, which was reversible
to the time domain. They also derived the analytical form of the complex cepstrum of a
Keywords:
transfer function in terms of its poles and zeros. The cepstrum had been used in speech
Cepstrum analysis
History
analysis for determining voice pitch (by accurately measuring the harmonic spacing),
Gear diagnostics but also for separating the formants (transfer function of the vocal tract) from voiced
Machine diagnostics and unvoiced sources, and this led quite early to similar applications in mechanics. The first
Operational modal analysis was to gear diagnostics (Randall), where the cepstrum greatly simplified the interpretation
of the sideband families associated with local faults in gears, and the second was to extrac-
tion of diesel engine cylinder pressure signals from acoustic response measurements (Lyon
and Ordubadi). Later Polydoros defined the differential cepstrum, which had an analytical
form similar to the impulse response function, and Gao and Randall used this and the com-
plex cepstrum in the application of cepstrum analysis to modal analysis of mechanical
structures. Antoni proposed the mean differential cepstrum, which gave a smoothed result.
The cepstrum can be applied to MIMO systems if at least one SIMO response can be sepa-
rated, and a number of blind source separation techniques have been proposed for this.
Most recently it has been shown that even though it is not possible to apply the complex
cepstrum to stationary signals, it is possible to use the real cepstrum to edit their (log)
amplitude spectrum, and combine this with the original phase to obtain edited time sig-
nals. This has already been used for a wide range of mechanical applications. A very pow-
erful processing tool is an exponential ‘‘lifter” (window) applied to the cepstrum, which is
shown to extract the modal part of the response (with a small extra damping of each mode
corresponding to the window). This can then be used to repress or enhance the modal
information in the response according to the application.
Ó 2016 Elsevier Ltd. All rights reserved.
1. Introduction
The first paper on cepstrum analysis [1] defined it as ‘‘the power spectrum of the logarithm of the power spectrum”. The
original application was to the detection of echoes in seismic signals, where it was shown to be greatly superior to the auto-
correlation function, because it was insensitive to the colour of the signal. This purely diagnostic application did not require
E-mail address: b.randall@unsw.edu.au
http://dx.doi.org/10.1016/j.ymssp.2016.12.026
0888-3270/Ó 2016 Elsevier Ltd. All rights reserved.
4 R.B. Randall / Mechanical Systems and Signal Processing 97 (2017) 3–19
returning to the log spectrum, and the reason for the definition was presumably that software for producing power spectra
was readily available. Even though [1] had one of the authors of the FFT algorithm (John Tukey) as a co-author it was pub-
lished two years before the FFT algorithm [2] and thus the potential of the latter was not yet realised. Tukey himself in [1]
writes ‘‘Although spectral techniques, involving second-degree operations, are now quite familiar, the apparently simpler
first-degree Fourier techniques are less well known” (he then refers to a paper of which he is the sole author, which confirms
that he made the remark). Thus, the original cepstrum was not reversible even to the log spectrum, and so even though the
word and concept of a ‘‘lifter” (a filter in the cepstrum) was defined in the original paper, this was enacted as a convolutive
filter applied to the log spectrum rather than a window in the cepstrum. The diagnostic application was also satisfactory for
determining the voice pitch of voiced speech [3], so speech analysis was one of the earliest applications, much of the devel-
opment being done at Bell Telephone Labs.
The early history of the development of the cepstrum was recently published by the two pioneers Oppenheim and Schafer
[4], but their paper primarily covers applications in speech analysis, communications, seismology and geophysics, and no
mechanical applications are given. They describe how Oppenheim’s PhD dissertation at MIT [5] introduced the concept of
‘‘homomorphic systems”, where nonlinear relationships could be converted into linear ones to allow linear filtering in the
transform domain. Typical examples were conversion of multiplication into addition by the log operation, and conversion
of convolution by first applying the Fourier transform (to convert the convolution to a multiplication) followed by the log-
arithmic conversion. It was a small further step to perform a (linear) inverse Fourier transform on the log spectrum to obtain
a new type of cepstrum, and this was the basis of Schafer’s PhD dissertation at MIT [6]. By retaining the phase in all oper-
ations, the ‘‘complex cepstrum” was defined as the inverse Fourier transform of the complex logarithm of the complex spec-
trum, and this was thus reversible to the time domain. Oppenheim and Schafer collaborated on a book [7], in which a
number of further developments and applications of the cepstrum were reported, in particular the analytical form of the cep-
strum for sampled signals, where the Fourier transform could be replaced by the z-transform.
2. Basic relationships and definitions
2.1. Formulations
The original definition of the (power) cepstrum was:
C p ðsÞ ¼ jIflogðF xx ðf ÞÞgj2 ð1Þ
where Fxx(f) is a power spectrum, which can be an averaged power spectrum or the amplitude squared spectrum of a single
record.
The definition of the complex cepstrum is:
C c ðsÞ ¼ I1 flogðFðf ÞÞg ¼ I1 flnðAðf ÞÞ þ j/ðf Þg ð2Þ

where
Fðf Þ ¼ Iff ðtÞg ¼ Aðf Þej/ðf Þ ð3Þ

in terms of the amplitude and phase of the spectrum. It is worth noting that the ‘‘complex cepstrum” is real, despite its name,
because the log amplitude of the spectrum is even, and the phase spectrum is odd.
The new power cepstrum is given by:
C p ðsÞ ¼ I1 flogðF xx ðf ÞÞg ð4Þ
which for the spectrum of a single record (as in (3)) can be expressed as:
C p ðsÞ ¼ I1 f2 lnðAðf ÞÞg ð5Þ

The so-called real cepstrum is obtained by setting the phase to zero in Eq. (2):
C r ðsÞ ¼ I1 flnðAðf ÞÞg ð6Þ

which is seen to be simply a scaled version of (5).
It should be noted that the complex cepstrum requires the phase /(f) to be unwrapped to a continuous function of fre-
quency, and this places a limit on its application. It cannot be used for example with stationary signals, made up of discrete
frequency components (where the phase is undefined at intermediate frequencies) and stationary random components
(where the phase is random). This point is taken up later. It can only be used with well-behaved functions such as impulse
responses, where the phase is well-defined (and often related to the log amplitude). The unwrapping can usually be accom-
plished using an algorithm such as UNWRAP in MatlabÒ, although the latter may give errors where the slope of the phase
function is steep. This problem can usually be resolved by repeatedly interpolating the spectra by factors of 2 until the same
result is obtained with successive steps. The penultimate step can then be used. Errors in unwrapping can also occur when
the amplitude is very small, so that noise affects the phase estimates.
R.B. Randall / Mechanical Systems and Signal Processing 97 (2017) 3–19 5
As mentioned above, Oppenheim and Schafer in [7] derived the analytical formulation of the cepstrum for transfer func-
tions, from digitised signals, expressed using the z-transform in terms of their poles and zeros in the z-plane. Thus, for a gen-
eral impulse response function:
X
2N X
2N
hðnÞ ¼ Ai esi nDt ¼ Ai zni ð7Þ
i¼1 i¼1
where the Ai are residues, si and zi poles in the s-plane and z-plane, respectively, and Dt the time sample spacing, the transfer
function in the z-plane is given by:
QM i Q o
B ð1 ai z1 Þ Mi¼1 ð1 bi zÞ
HðzÞ ¼ QNi¼1 Q No
ð8Þ
i¼1 ð1 c i z Þ i¼1 ð1 di zÞ
i 1
where the ci and ai are poles and zeros inside the unit circle in the z-plane, respectively, and the di and bi are the (reciprocals
of the) poles and zeros outside the unit circle, respectively (so that jai j; jbi j; jci j; jdi j < 1). Oppenheim and Schafer [7] show that
for this transfer function H(z), the cepstrum can be represented as:
C h ðnÞ ¼ lnðBÞ; n¼0

X an X cn
C h ðnÞ ¼ i
n
þ i
n
; n>0
i i ð9Þ
X bn X dn
C h ðnÞ ¼ i
n
i
n
; n<0
i i
Here, n is time sample number, known as ‘‘quefrency” (see Section 2.2). It will be seen that the poles and zeros inside the
unit circle, which define the minimum phase part of the transfer function, have a causal cepstrum, with positive quefrency
components only, while the poles and zeros outside the unit circle, defining the maximum phase part, form the cepstrum at
negative quefrency. Oppenheim and Schafer thus were able to demonstrate that the real and imaginary parts of the Fourier
transform of the cepstrum of a minimum phase function (the log amplitude and phase spectra, respectively) are Hilbert
transforms of each other. Hence, if it is known that a transfer function is minimum phase (often the case for many simple
mechanical structures), it is not necessary to measure or unwrap the phase, as it can be calculated from the log amplitude.
In fact, the easiest way to do this is to form the complex cepstrum (one-sided) from the real cepstrum (real and even) by
doubling positive quefrency components and setting negative quefrency components to zero. The log amplitude and phase
spectra can then be obtained by a forward transform of the complex cepstrum.
One further property of the cepstrum, already mentioned as a homomorphic property, was developed quite early, and
that is that forcing function and transfer function effects are additive in the cepstrum. For SIMO (single input, multiple out-
put) systems, any output signal is the convolution of the input signal with the impulse response function of the transmission
path, and this convolutive relationship is converted to an addition by the cepstrum operation as follows:
If yðtÞ ¼ f ðtÞ hðtÞ ð10Þ
then Yðf Þ ¼ Fðf Þ Hðf Þ ð11Þ
logðYðf ÞÞ ¼ logðFðf ÞÞ þ logðHðf ÞÞ ð12Þ
and
I1 flogðYðf ÞÞg ¼ I1 flogðFðf ÞÞg þ I1 flogðHðf ÞÞg ð13Þ
Note that this only applies for SIMO systems, since for MIMO (multiple input, multiple output) systems each response
signal is a sum of convolutions. This point is taken up later.
This relationship was used in speech analysis, for example to separate the speech source (voiced or unvoiced) from the
transmission effect of the vocal tract (whose resonances are called ‘‘formants”), and in [7] a vocoder is proposed where the
speech information can be considerably compressed by transmitting only the low quefrency part of the cepstrum (giving the
formants) and an indicator of whether the speech segment is voiced or unvoiced, and if so the voice pitch. To reconstruct the
speech, the formants were used to generate a filter, which was excited either by a noise source (unvoiced) or a periodic pulse
generator with frequency the same as the voice pitch.
2.2. Terminology
In the original paper [1], the authors coined the word ‘‘cepstrum” by reversing the first syllable of ‘‘spectrum”, the justi-
fication being that it was a ‘‘spectrum of a spectrum”. Similarly, the word ‘‘quefrency” was obtained from ‘‘frequency”, and
the authors also suggested a number of others, including:
rahmonic from harmonic

lifter from filter
gamnitude from magnitude
saphe from phase
darius from radius
dedomulation from demodulation
Of these, cepstrum, quefrency, rahmonic and lifter are useful in clarifying that the operations or features refer to the cep-
strum, rather than the spectrum or time signal, and are still regularly used in the literature, as well as in this paper. Note
that the autocorrelation function is also a ‘‘spectrum of a spectrum”, so the distinctive feature of the cepstrum is the log con-
version of the spectrum.
3. Early applications in mechanics and acoustics
3.1. Diagnostics of families of harmonics or sidebands
The author became aware of the possibility of using the cepstrum to detect and quantify families of periodically spaced
spectral components by reading [3] and realised that this would not only detect families of harmonics, but also equally
spaced modulation sidebands. Local faults in gears give an impulsive modulation of the gearmesh signals (both amplitude
and frequency modulation) resulting in large numbers of sidebands spaced at the speed of the gear on which the local fault
is located. The majority of such sidebands are only visible on a spectrum with a log amplitude scale, and so the cepstrum is
an ideal way to collect the (average) information of a large number of such sidebands into a relatively small number of rah-
monics in the cepstrum, of which the first contains most of the information. He had even experimented with calculating the
cepstrum using FFT methods before joining the company Brüel and Kjær in Copenhagen in 1971. A Fortran version of the FFT
algorithm was obtained from [8]. However, even though B&K had a fully designed mini-computer-based FFT analyser in
1971, this was never released onto the market, and so other methods of obtaining the cepstrum were sought. In 1973 an
Application Note No. 13-150 entitled ‘‘Cepstrum Analysis and Gearbox Fault Diagnosis” [9] was published by B&K, in which
six methods were described for calculating the (power) cepstrum, primarily for the purpose of diagnosing gear faults. Several
were variations on the theme of using a heterodyne analyser for performing the ‘‘frequency” analysis on a linear frequency
scale with constant bandwidth, first to obtain the original power spectrum on a log amplitude scale, and then the second
analysis of the log spectrum to the cepstrum on a linear scale. In between, the log spectrum was stored in a medium, which
could either be a circulating digital memory (a Digital Event Recorder type 7502), or a tape loop on an FM tape recorder. In
both cases the stored spectrum was played back at a much higher rate than recorded. The initial analysis could either be
generated directly from a continuous recording of the signal on a tape recorder, or sometimes from a recirculated section
of signal recorded in a second Digital Event Recorder, possibly using an analogue ‘‘Gauss Impulse Multiplier” type 5623,
to apply a Gaussian, rather than Hanning, weighting to the selected and periodically repeated signal section.
A result using a Digital Event Recorder for intermediate storage is shown in Fig. 1, with both frequency analyses done by a
heterodyne analyser. In the log spectrum on the left, a large number of sidebands are visible around the first three harmonics
of gearmesh frequency, and the cepstrum on the right shows that these are spaced at the high speed shaft speed of 85 Hz.
The machine analysed was a gearbox between a gas turbine at 85 Hz and a generator at 50 Hz, and the cepstrum shows
immediately that the fault is on the gear rotating at turbine speed. The component was much smaller after repair. It should
be noted that the frequency estimation (actually the reciprocal of the equivalent quefrency) is very accurate because it is the
average of the sideband spacing over the whole spectrum.
2×TM
TM = Toothmesh 3×TM
TM
Frequency (Hz)
Fig. 1. Log spectrum and cepstrum for a gear fault.

Even though the B&K FFT analyser was not released, the designer, Arne Grøndahl, wrote driver routines to allow the
assembler language FFT algorithm to be called from a Fortran program, and it was possible to perform cepstrum calculations,
even the complex cepstrum, with fully FFT-based methods, as early as 1975, and the author published further results of gear
diagnostics at that time [10].
At about the same time, Georges Sapy, of Electricité de France (EDF) found that the cepstrum could be used to detect dam-
aged or missing blades in a steam turbine [11]. The lack of guidance of the steam by the missing or faulty blades gave rise to a
series of impulsive interactions with each stator blade, so that a casing mounted accelerometer would detect the pulses from
the nearest section at a rate corresponding to the rotor speed, 50 Hz. This then gave rise to an increase in intermediate fre-
quency harmonics in comparison to the normal condition, as shown in Fig. 2.
Fig. 3 shows that the power cepstrum (original definition) is an efficient way of detecting and quantifying the pattern of
additional harmonics in the intermediate frequency range (750–5000 Hz).
Fig. 2. Log spectra (a) without blade fault and (b) with blade fault.
Fig. 3. Cepstra (a) without blade fault and (b) with blade fault.
3.2. Separation of source and transmission path
One of the first groups to apply the cepstrum to the separation of forcing function and transfer function in mechanical
systems was Professor Richard Lyon of MIT, and co-workers. With his doctoral student Afarin Ordubadi he showed how
the complex cepstrum could be used to extract the cylinder pressure (source) signal from measured acoustic responses
[12]. Fig. 4 shows the estimated pressure signals from two cylinders of a diesel engine from this paper.
In 1984, the author proposed using the cepstrum for separating the gearmesh forcing function (single input) from the
structural transfer function in a gearbox [13]. This was considered important because for example an increase in the second
harmonic of the gearmesh frequency might be due to tooth wear (normally in two patches on each tooth, one on each side of
the pitchline, where there is a rolling rather than sliding action). However, it might be because a natural frequency of the
structure or internals has reduced so as to coincide with the harmonic in question because of a developing crack, and this
situation might have a very different prognosis. Fig. 5 shows an example [14] based on this idea, where the periodic forcing
function of the gearmesh was removed in the cepstrum using a comb lifter adjusted to the gearmesh quefrency, leaving a
cepstrum dominated by the transfer function to the measurement point. The edited cepstrum was transformed back to
the log spectrum, which shows that the latter was little changed for two measurements on the same gearbox, where one
had the presence of cracked teeth. Even though a modal analysis was not carried out, it is obvious that the natural frequen-
cies did not change appreciably.
Fig. 4. Estimated cylinder pressure using the complex cepstrum [12].
Fig. 5. Use of liftering in the cepstrum to remove the forcing function component from the cepstra for a gear with and without cracked teeth, leaving the
part dominated by the structural transfer functions [14].
4. Later developments up to the present
4.1. Applications of the real cepstrum in machine diagnostics
From 1975 to 1980, the author made a number of further developments to the application of the real cepstrum to gear
diagnostics, presented at a number of conferences, and which culminated in the publication of a second Application Note
with the same title as [9], published by B&K in 1980, but also in the journal Maintenance Management International [15].
In this, a number of practical considerations in the calculation and interpretation of the cepstrum are discussed, of which
perhaps the most important is the recommendation to use the amplitude of the ‘‘analytic cepstrum” in place of the normal
definition of the real or power cepstrum, in particular when it is based on a section of the spectrum not starting from zero
frequency (eg a zoom spectrum) or when sidebands are not simultaneously a harmonic series passing through zero fre-
quency (as can happen with planetary gears because the sun speed is not a subharmonic of the mesh frequency). In this case,
the rahmonics are not necessarily positive peaks, but the quefrency corresponding to a sideband spacing can actually be at a
zero crossing between a positive and negative peak. Fig. 6 illustrates this for two slightly displaced zoom spectra of a gearbox
signal.
The analytic cepstrum is simply calculated using Hilbert transform theory by setting the negative frequency part of the
log spectrum to zero, and doubling the positive frequency part (in dB). This means that the real and imaginary parts of the
resulting analytic cepstrum are Hilbert transforms, but the magnitude or envelope always has its peak at the quefrency cor-
responding to the sideband spacing. Note that the analytic cepstrum (as opposed to the ‘‘complex cepstrum”) is genuinely
complex.
Ref. [16] now contains all this information plus advice on how to optimise the use of the cepstrum by editing of the log
spectrum and of the cepstrum itself to modify the log spectrum.
In 1991, Capdessus and Sidahmed showed how the cepstrum was very useful for separating the effects of two gears with
only a small difference in the numbers of teeth. As mentioned above, the harmonics of gearmesh frequency are surrounded
by sidebands spaced at the two shaft speeds, but when the ratio is close to unity it is difficult to separate the sidebands in the
spectrum. In [17] a case is presented where the numbers of teeth on the meshing gears are 20 and 21, respectively, and the
two components in the cepstrum were quite distinct, allowing diagnosis of which gear developed a defect.
From 1999 to 2004, Mohamed El Badaoui and co-workers introduced a number of new aspects into the use of cepstrum
analysis for gear diagnostics. One was presented in [18] and considerably expanded in [19], and is based on the fact that
under fairly broad conditions there is a tendency for the sum of the first rahmonic peaks, corresponding to each of the
Fig. 6. Advantage of the analytic cepstrum for zoom spectra.

two gears in a meshing pair, to remain constant, so that if one increases as the result of a local fault, the other decreases
correspondingly. An example from [19] is given in Fig. 7 where A1 and A2 represent the two first rahmonics. The authors
defined a diagnostic indicator d, which is the ratio of specially normalised versions of A1 and A2 which is relatively insen-
sitive to signal/noise ratio, even though this changes the absolute magnitude of the numerator and denominator terms. This
is seen to give a very good trend parameter for a developing fault (in fact the same data as treated in [17]).
The other major contribution was based on the fact that the cepstrum of a signal with an inverted echo is entirely neg-
ative, so that its integral over quefrency becomes markedly more negative when such an inverted echo is in the analysed
section of a time record. The ‘‘moving cepstrum integral” (MCI) [20] was based on this and was applied to the detection
of spalled gear teeth, the rationale being that the signal on exiting a spall would be an inverted version of the signal on entry,
thus giving an inverted echo. The principle is illustrated in Fig. 8a and b shows an example of its application from [20].
Endo in [21] showed that there were inverted echoes in the signals from tooth cracks as well, so the MCI could not be used
alone to distinguish them.
4.2. Applications of the cepstrum in modal analysis
Most of the following is discussed in more detail in [22] and newer developments will be reported in a forthcoming over-
view of this topic. The topic is thus treated here rather summarily, and this paper concentrates on more general mechanical
applications, in particular in machine diagnostics and condition monitoring.
Fig. 7. Application of the diagnostic indicator d from [19].
Fig. 8. (a) Principle of Moving Cepstrum Integral and (b) example of application from [20].
4.2.1. The differential cepstrum

One further major theoretical development with the cepstrum was the proposal in1981 by Polydoros and Fam of the dif-
ferential cepstrum [23]. This was originally defined as the inverse z-transform of the derivative of the log spectrum. Thus:

d H0 ðzÞ
C d ðsÞ ¼ Z 1 ½log HðzÞ ¼ Z 1 ð14Þ
dz HðzÞ
Because taking the derivative in one domain corresponds to multiplication by the independent variable in the other
domain, the differential cepstrum can be written in terms of the poles and zeros in the z-plane (cf Eq. (9)) as:
X X
C hd ðnÞ ¼ ani þ cni ; n>0
Xi n Xi
n
ð15Þ
C hd ðnÞ ¼ bi di ; n < 0
i i
Note that in [23] these equations are given for quefrency displaced by one sample (0 ? 1) but the zero origin has been
maintained here to emphasize the similarity with the complex cepstrum. It will be recognised that the form of the differen-
tial cepstrum (sums of complex exponentials) is the same as that of the impulse response function, except that there are
similar terms for both poles and zeros (the latter with negative sign) and all ‘‘residues” are equal to unity.
4.2.2. Modal analysis using the cepstrum of response signals

In 1996, Gao and Randall [24,25] extracted the modal properties (poles and zeros) of a transfer function from measured
response signals, in a case where the forcing function was confined to the very low quefrency region of the response cep-
strum. This applies wherever the excitation is impulsive with a relatively smooth and flat spectrum over the frequency range
of interest, eg a hammer blow, or in fact broadband random noise in cases where the smoothed power spectrum of the
response can be used, eg for minimum phase systems where the phase does not have to be measured.
This is illustrated in Fig. 9, where it can be seen that the regenerated cepstrum cannot be distinguished from the mea-
sured one in the high quefrency region dominated by the transfer function (which was the section curve-fitted for the poles
and zeros).
It should be noted that the force spectrum was quite flat, but fell off by about 15 dB over the frequency range up to
3.2 kHz, though this did not affect the curve-fitting of the poles and zeros. Thus the cepstrum method does not demand that
the excitation is white, as do some other methods of operational modal analysis.
Fig. 9a shows the response autospectrum at the driving point (one end) of a free-free beam excited by a hammer blow.
Noise in the response is negligible. Fig. 9b shows the cepstrum derived directly from this, and the cepstrum of the transfer
function obtained from it by curve-fitting equations of the form of (9) for the poles and zeros inside the unit circle using a
least squares optimisation (Levenberg-Marquardt) method [24]. The cepstrum of the transfer function was then regenerated
using Eq. (9). Minimum phase properties were assumed for this simple structure. Equally good results were obtained by
curve-fitting the differential cepstrum, using both the Levenberg-Marquardt method and the ITD (Ibrahim Time Domain)
method, originally developed for extracting impulse responses from free decay vibration signals.
When an attempt was made to regenerate the frequency response functions (FRFs) from the curve-fitted poles and zeros,
it was realised that there are two pieces of information lacking to fully restore them:
Force cepstrum concentrated here

Dominated by transfer function
(a) (b)
Fig. 9. Extraction of the complex cepstrum of a transfer function from the response autospectrum of a beam excited by a hammer blow (a) driving point
response autospectrum and (b) measured and regenerated cepstra [24].
1. An overall scaling factor that is part of the zero quefrency value of the cepstrum which is not curve-fitted.
2. An equalisation curve due to the lack of residual information from out-of-band modes that are excluded from the mea-
sured cepstrum.
In [25] it is described how this information can be obtained by other means from earlier or similar FRFs, for example a
finite element (FE) model of the structure. Neither of the two above factors is sensitive to small changes in the exact pole
and zero positions, and so it is possible to track slow changes in the dynamic properties (such as from a developing crack)
or update the FE model to have the correct (measured) in-band poles and zeros. Examples are given in [25].
In [25] the way of generating the equalisation curves was using ‘‘phantom zeros” in-band to compensate for the missing
out-of-band poles and zeros. This was effectively the same as done for the established Rational Fraction Polynomial method.
Even though pole-zero and pole-residue models are identical for systems with a finite number of DOFs, in practice models
must be truncated. For pole-residue models, this means that the residues are correct but the zeros wrong, but for pole-zero
models the zeros are correct but the residues wrong. Almost no compensation is required for driving point measurements,
because there are equal numbers of poles and zeros truncated in the out-of-band region, and these tend to cancel in the in-
band region. However, the fewer zeros there are in the transfer function, the greater the imbalance that has to be compen-
sated for with the equalisation curve.
Later papers, culminating in [26], have developed improved methods for determining the equalisation curve, including
scaling. With respect to scaling, the method in [26] tends to give the correct result for free-free systems if the inertial prop-
erties of the rigid body test object are correct, since these determine the zero frequency values of the FRFs. This will be the
case for example where the reference FRF is an actual measurement and the updating is to follow a change in stiffness (eg a
developing crack). It is also relatively simple to arrange for an FE model to have the correct rigid body inertial properties.
For constrained test objects, it is a little more difficult, because the zero frequency values are then determined by the stiff-
ness properties, and these are likely to be affected by poor modelling of joints, etc. However, a relatively simple updating of
the FE model to match the frequencies of the first one or two elastic modes will usually ensure that the zero frequency value
is correct, even if higher frequency modes are not correct because of local stiffness anomalies.
4.2.3. The mean differential cepstrum

In 2000 [27] Antoni et al. proposed the ‘‘mean differential cepstrum” (MDC), which can be calculated by a formula similar
to Eq. (14), to which it reverts for a single realization:
Fig. 10. Non-minimum phase FRF recovered using the MDC (from [27]).

E½H0 ðf ÞH ðf Þ
C hmd ðsÞ ¼ I1 ð16Þ
E½Hðf ÞH ðf Þ
This permits averaging over a number of realizations, for example of the response of a system excited by a burst random
sequence. Even mixed phase FRFs can be obtained using the MDC, even in the presence of extraneous noise in the signals. A
typical result is shown in Fig. 10.
4.2.4. MIMO systems

As mentioned above, the cepstrum can only separate forcing and transfer functions for SIMO systems, so in the more gen-
eral MIMO case, it is first necessary to convert it into a sum of SIMOs, or at least extract one SIMO response from the total.
This is the subject of the topic ‘‘blind source separation” (BSS), and so in principle any BSS method could be used to extract
responses to a single excitation. Ref. [28] is a special issue of Mechanical Systems and Signal Processing devoted to the appli-
cations of BSS to mechanical systems.
One application of particular interest, not discussed there, is the extraction of the cepstrum of the transmission path for a
single cyclostationary source with a particular cyclic frequency from a MIMO response signal [29]. The intended application
was to diesel railcars where the engine combustion pulses would constitute such a source, since the combustion signal is a
typical second order cyclostationary (CS2) signal, with a broadband carrier, amplitude modulated by a deterministic signal at
the firing frequency. In [29] it is shown that in the spectral correlation diagram for such a CS2 signal, the cyclic spectral den-
sity at this cyclic frequency contains transfer function information only for this particular source, and thus constitutes a
method to extract SIMO information in the presence of multiple other sources. The cyclic spectral density can be curve-
fitted for this modal information, as for any other single source response. This was demonstrated using cepstral curve-
fitting in [29].
5. Editing time signals using the real cepstrum
A new development, published after both [22] and [16], introduces the possibility of editing time signals using the real
cepstrum. It was previously thought that this could only be done using the complex cepstrum, but as pointed out above this
is not possible for stationary signals and so a large number of applications are excluded. It was realised [30] that there are
many situations where editing could be carried out by modification of the amplitude only, which could be achieved using the
real cepstrum. The modified amplitude spectrum can then be combined with the original phase spectrum of each record to
generate an edited time record. A case in point is where discrete frequencies are to be removed. Whole families of uniformly
spaced harmonics or sidebands can be removed by removing a small number of rahmonics in the cepstrum. Removing a dis-
crete frequency really means setting the value at that frequency to the expected value of the noise, of which the best guide
are the frequency components on either side of the discrete frequency, but usually at a much lower level. Setting the value of
the discrete rahmonics to zero automatically smooths over the amplitude of the log spectrum in the vicinity of the corre-
sponding harmonics and/or sidebands (since notches in the log spectrum would equally give non-zero cepstrum components
as for peaks) so at least the amplitude estimate would be correct. The phase in the modified spectrum at the position of the
removed components would still be the same as that for the original discrete frequencies, and thus in general incorrect, but it
should be kept in mind that these are typically spaced by 20 lines or more, and thus often negligible once their amplitude is
reduced to that of the noise components, which in any case have random phase. The principle of the procedure is shown in
Fig. 11. Schematic diagram of the cepstral method for removing selected components such as families of spectral harmonics and/or sidebands from time
signals.
Fig. 11, where the original phase of each record is retained for combination with the modified amplitude obtained by liftering
the real cepstrum.
Fig. 12 shows a typical application to signals from a gearbox with a faulty bearing. It shows the original time signal, and
compares it with the result of removing all harmonics of the gear signal by two methods (the gear ratio is 1:1 so only one set
of harmonics had to be removed). The first method, Fig. 12b, removed the Time Synchronous Average (TSA), while the cep-
strum method results are shown in Fig. 12c. Even though the result of the cepstral method is slightly noisier than for the TSA,
the subsequent envelope analysis of the edited signals for the bearing diagnosis was just as successful.
Note that this cepstral method can also remove families of equally spaced sidebands, even where these are not simulta-
neously harmonics, something which cannot be done by TSA.
5.1. Application to operational modal analysis
A frequent problem with operational modal analysis (OMA) is the presence of disturbing components in the signal, for
example discrete frequencies. These are often treated by the OMA software as modes with very low damping. Families of
harmonics and sidebands can be individually removed, as just described, but another alternative that is often possible is
to apply a shortpass lifter to the cepstrum to greatly enhance the modal information with respect to everything else. By com-
paring Eqs. (9) and (15) it is seen that both contain damped exponentials, with the frequencies of the modes (and similar
terms for the zeros), but the cepstrum is additionally weighted by 1/n, making it shorter than the equivalent impulse
response. Thus the information about all modes is concentrated at low quefrency. Applying an exponential window (short-
pass lifter) to the cepstrum removes almost all extraneous information (including discrete frequencies) at higher quefren-
cies, and the only effect on the modal information is to add a precisely known amount of extra damping to each mode.
This is the same in principle as using an exponential window in impact tests to cause impulse responses to die out before
the end of the record, but can in this case be applied to stationary signals, rather than just transients starting at time zero.
Fig. 13 shows an example where this has been applied to a signal from a high speed gas turbine, whose spectrum is dom-
inated by multiple families of discrete harmonics. Fig. 13a is the cepstrum of the original log spectrum Fig. 13d, and Fig. 13b
is an exponential lowpass lifter chosen to have a time constant of about 0.0025 s, this being approximately equivalent to the
damping of the lowest frequency mode (the lowest frequency peak in Fig. 13e). Thus, after applying this lifter, the damping
of this lowest mode would be approximately doubled, but since the added bandwidth is constant it would have progressively
less effect on high frequency modes. The damping of all modes can in any case be compensated.
It will be seen that this almost blind operation has resulted in processed data almost completely dominated by the modal
information (as characterised by the log spectrum in Fig. 13e). Even though only the spectrum is shown, time records are
available which could be processed by standard OMA software. Unfortunately, a modal analysis could not be carried out
in this case as only one signal was available.
20
0
(a)
-20
-40
0 1 2 3 4
Acceleration (m/s 2 )
20
0 (b)
-20
0 1 2 3 4
20
0 (c)
-20
0 1 2 3 4
Shaft rotation
Fig. 12. Time domain signals for gearbox test rig (a) Raw signal, (b) residual signal (after removing the synchronous average) and (c) residual signal after
editing the Cepstrum to remove the shaft rahmonics.
4
0.06 10
0.04 (a) (d)
0.02
3
0 10
-0.02
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
2
10
1
(b) 10
1
0.5
4
10
0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 (e)
3
10
0.06
0.04 (c) 2
0.02 10
0
-0.02 1
10
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0 10 20 30 40
Quefrency (s) Frequency (kHz)
Fig. 13. Use of an exponential lowpass lifter to enhance modal information (a) Full cepstrum (of (d)), (b) exponential lifter, (c) liftered cepstrum, (d) original
spectrum and (e) liftered spectrum.
Examples of application of these procedures to OMA are given in Refs. [31,32].
5.2. Application to diagnostics of variable speed machines
It has recently been realised that the ability of an exponential lifter to isolate the modal properties of the signal transfer
path are very valuable in the diagnostics of variable speed machines, such as wind turbines, which typically vary in speed by
±30%. In particular the signals from gears and rolling element bearings vary in completely different ways under speed vari-
ation, something that was disguised by constant speed operation. A recent paper [33] demonstrates this for gears and bear-
ings, showing that for gears, the effects of dominant forcing frequencies passing through resonances impedes the diagnostics,
whereas the opposite is the case for bearings where resonances carry the fault information. In both cases the signals should
be processed by an exponential shortpass lifter prior to the order tracking used to compensate for the speed variation. In the
case of gears, the modal information is removed before order tracking, while for bearings it can be retained and the deter-
ministic forcing frequencies removed since they often mask the bearing signals. In any case, a number of bearing diagnostic
techniques make use of the resonance information, and should be applied before order tracking. Examples of each are given
in [33] and a short résumé is repeated here.
Fig. 14. (From [33]) (a) Original spectrum (Signal 1-2) and result of exponential lifter (red) and (b) spectrum with modal information removed. (For
interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 14a shows the spectrum of a signal (No. 1-2) from a speedup gearbox driving a pump, where the speed varied around
22 Hz by about ±15%. It also shows the result (in terms of spectral amplitude) of applying an exponential shortpass lifter with
time constant 5 ms (3 dB bandwidth 64 Hz). Fig. 14b shows the result of subtracting this modal information in the cepstrum,
leaving primarily the spectrum of the forcing functions from the gears.
Fig. 15 compares the order tracked spectra for these two cases and also for another signal (No. 2-1) where the speed var-
ied around 15 Hz by about ±15%. It is seen that without the pre-processing, the order spectra are quite different, with the
harmonics of the gearmesh frequency (GM) being greatly affected by the fixed resonances at different speeds, whereas
the liftered signals have quite similar order spectra. It would be much easier with the cepstrally processed signals, for exam-
ple, to detect an increase in the second harmonic of the forcing function as a result of uniform tooth wear.
Fig. 15. (From [33]) Effect of liftering on order tracked spectra (a, b) Signal 1-2, (c, d) Signal 2-1, (a, c) Original and (b, d) Liftered.
Fig. 16. (From [33]) Signal envelopes for different methods (Row 1) Original (Row 2) SK filtered (Row 3) MED (Row 4) Exponential liftered (Column 1)
A - 25% (Column 2) A - end (Column 3) C - 25% (Column 4) C – end.
Fig. 17. (From [33]) Envelope spectra at different positions along record; Left: 25% (half speed), Right: near end (a) Point A, (b) Point C, (c) C after MED and
(d) C After exponential liftering.
Fig. 16 shows the effect of three pre-processing methods applied before order tracking to two signals from a bearing test
machine subject to a linear run-up from 0 to 40 Hz shaft speed over 40 s. The bearing had a localised outer race fault aligned
with the load zone, and measurements were made at two measurement points, Point A on the bearing housing (in the load
direction) and Point C at a remote point on the bedplate.
The three pre-processing methods were:
(1) Finding an optimum passband to maximise the spectral kurtosis (SK) of the filtered signal, using the fast kurtogram
[34]. This usually finds a resonance or band of resonances which give the most impulsive responses (IRs) excited by
the bearing fault, and has to be done before order tracking.
(2) Minimum entropy deconvolution (MED), which generates an inverse filter to remove the smearing effect of the trans-
fer path from the original impulses, and attempt to retrieve the original impulses, which by their nature will be shorter
than the IRs [35]. Since the IRs are constant in time, this also has to be done before order tracking.
(3) Exponential lifter to generate the modal information and remove deterministic forcing functions (as in Fig. 14a) and
generate time signals using this amplitude information and the original phase. These should be dominated by the
bearing fault information carried by those resonances.
Envelopes were generated of the original signal and the three processed signals using standard Hilbert transform meth-
ods, and these are compared in Fig. 16. They were made at one quarter full speed (25% of the record) and at full speed (end of
record). Note that the original IRs, and the results of pre-processing, have the same length, independent of their spacing,
which reduces with increasing speed.
At Point A, the original IRs are just separated, even at full speed, but the other three methods result in much shorter
pulses. The shortest pulses are perhaps given by the cepstral method, although with a little more noise than the SK and
MED methods. At Point C, the original IRs are much longer, and no longer fully separated at full speed. The best result is given
by the SK method, so it is likely the kurtosis was maximised by a high frequency resonance with corresponding short IR. The
MED and cepstral methods do not perform as well at this measurement point, most likely because the transfer function
(measured on the baseplate) contained many resonances over a very wide frequency range, not all of which would have car-
ried the bearing fault information, and making it difficult to find an inverse filter with limited order. On the other hand this
would not be typical of most measurements in practice, which are made on or close to the bearing being monitored.
Even so, the results of envelope analysis, carried out at half full speed and full speed, for Point A and Point C are shown in
Fig. 17. These include the MED and cepstral methods, and show that the envelope analysis is very robust and allows an easy
diagnosis of outer race fault (order 3.98) in all cases. There is seen to be a component at shaft speed, and also sidebands with
this spacing, more prominent at high speed, so this is likely due to residual unbalance adding to the static load.
It is interesting that the cepstral method shows the least effect of speed variation, as predicted by the theory.
6. Conclusion
Few people realise that the cepstrum is older than the FFT algorithm, or that its original definition made it less flexible as
a result. Although originally applied to problems in seismology and speech analysis, similar background situations meant
that it was applied quite early to mechanical problems as well. There are two main areas of applicability, which have been
considerably developed over the years, one in detecting, classifying and perhaps removing families of harmonics and side-
bands in vibration and acoustic signals, and the other in blind separation of source and transfer function effects, at least for
SIMO systems. The latter leads naturally to applications in operational modal analysis, and in machine diagnostics under
variable speed conditions. It is thought likely that as new methods are developed for blind separation of sources (or at least
the responses to the different sources in the general case of convolutive mixtures), that the possibilities given by the cep-
strum to separate different types of sources, as well as forcing function from transfer function for each path, will become
even more extensive. Despite its long history, the author believes that the cepstrum still has not been fully exploited, and
it is hoped that this paper may stimulate further interest in its possibilities.
Acknowledgments
The contributions of all my co-authors are gratefully acknowledged.
References
[1] B.P. Bogert, M.J.R. Healy, J.W. Tukey, The quefrency alanysis of time series for echoes: cepstrum, pseudo-autocovariance, cross-cepstrum, and saphe
cracking, in: M. Rosenblatt (Ed.), Proc. of the Symp. on Time Series Analysis, Wiley, NY, 1963, pp. 209–243.
[2] J.W. Cooley, J.W. Tukey, An algorithm for the machine calculation of complex fourier series, Math. Comp. 19 (90) (1965) 297–301.
[3] A.M. Noll, Cepstrum pitch determination, J.A.S.A. 41 (2) (1967) 293–309.
[4] A.V. Oppenheim, R.W. Schafer, From frequency to quefrency: a history of the cepstrum, IEEE Signal Process. Mag. (2004), pp. 95–99, 106 (September).
[5] A.V. Oppenheim, Superposition in a class of nonlinear systems Ph.D. dissertation, MIT, 1964 (May).
[6] R.W. Schafer, Echo removal by discrete generalized linear filtering Ph.D. dissertation, MIT, 1968 (January).
[7] A.V. Oppenheim, R.W. Schafer, Digital Signal Processing, Prentice-Hall, Englewood Cliffs, NJ, 1975.
[8] E.A. Robinson, Multi-Channel Time Series Analysis With Digital Computer Programs, Holden-Day, San Francisco, 1967.
[9] R.B. Randall, Cepstrum Analysis and Gearbox Fault Diagnosis, Brüel and Kjær Application Note No. 13-150, Copenhagen, 1973.
[10] R.B. Randall, Gearbox Fault Diagnosis using Cepstrum Analysis, Proc. IVth World Congress on the Theory of Machines and Mechanisms, Newcastle UK,
IMechE, vol. 1, 1975, pp. 169–174.
[11] G. Sapy, Une Application du Traitement Numérique des Signaux au Diagnostic Vibratoire de Panne : La Détection des Ruptures d’Aubes Mobiles de
Turbines, Tome XX, Automatisme, 1975, pp. 392–399.
[12] R.H. Lyon, A. Ordubadi, Use of cepstra in acoustic signal analysis, J. Mech. Des. 104 (1982) 303–306.
[13] R.B. Randall, Separating excitation and structural response effects in gearboxes, in: Third Int. Conf. on Vib. in Rotating Machines, IMechE, York, 1984,
pp. 101–107.
[14] R.B. Randall, Advanced Machine Diagnostics, The Shock and Vibration Digest, vol. 29(1), Springer, Berlin, 1997, pp. 6–30.
[15] R.B. Randall, Cepstrum Analysis and Gearbox Fault Diagnosis, Maintenance Management International, Elsevier, 1982/1983, pp. 183–208.
[16] R.B. Randall, Vibration-Based Condition Monitoring: Industrial, Aerospace and Automotive Applications, Wiley, Chichester, UK, 2011 (January).
[17] C. Capdessus, M. Sidahmed, Analyse des vibrations d’un engrenage: cepstre, corrélation, spectre, Traitement du Signal 8 (5) (1991) 365–372.
[18] M. El Badaoui, F. Guillet, J. Danière, Contribution du Cepstre d’Énergie au Diagnostic de Réducteur Complexe à Engrenage, Revue Française de
Mécanique, RFM, No. 1999-1, 1991, pp. 4–7.
[19] M. El Badaoui, F. Guillet, J. Danière, New applications of the real cepstrum to gear signals, including definition of a robust fault indicator, Mech. Syst.
Signal Process. 18 (5) (2004) 1031–1046.
[20] M. El Badaoui, J. Antoni, F. Guillet, J. Danière, P. Velex, Use of the moving cepstrum integral to detect and localise tooth spalls in gears, Mech. Syst.
Signal Process. (2001).
[21] H. Endo, R.B. Randall, C. Gosselin, Differential diagnosis of spalls vs. cracks in the gear tooth fillet region, J. Fail. Anal. Prev. 4 (5) (2004) 57–65.
[22] R.B. Randall, Cepstral Methods of Operational Modal Analysis, Chapter 24 in Encyclopedia of Structural Health Monitoring, Wiley, 2009.
[23] A. Polydoros, A.T. Fam, The differential cepstrum: definitions and properties, in: Proc. IEEE Int. Symp. Circuits Systems, 1981, pp. 77–80.
[24] Y. Gao, R.B. Randall, Determination of frequency response functions from response measurements. Part I: Extraction of poles and zeros from response
cepstra, Mech. Syst. Signal Process. 10 (3) (1996) 293–317.
[25] Y. Gao, R.B. Randall, Determination of frequency response functions from response measurements. Part II: Regeneration of frequency response
functions from poles and zeros, Mech. Syst. Signal Process. 10 (3) (1996) 319–340.
[26] W.A. Smith, R.B. Randall, Cepstrum-based operational modal analysis revisited: a discussion on pole–zero models and the regeneration of frequency
response functions, Mech. Syst. Signal Process. 79 (2016) (2016) 30–46.
[27] J. Antoni, F. Guillet, J. Danière, Identification of non-minimum phase transfer functions from output-only measurements, ISMA25 Conference, in:
ISMA25 Conference, KUL, Leuven Belgium, 2000.
[28] J. Antoni, S. Braun (Eds.), Special Issue: Blind Source Separation. Mechanical Systems and Signal Processing, vol. 19(6), November, 2005.
[29] D. Hanson, R.B. Randall, J. Antoni, D.J. Thompson, T.P. Waters, R.A.J. Ford, Cyclostationarity and the cepstrum for operational modal analysis of mimo
systems—Part I: Modal parameter identification, Mech. Syst. Signal Process. 21 (6) (2007) 2441–2458.
[30] N. Sawalhi, R.B. Randall, Editing Time Signals using the Real Cepstrum, in: MFPT Conference, Virginia Beach, May, 2011.
[31] R.B. Randall, B. Peeters, J. Antoni, S. Manzato, New cepstral methods of signal pre-processing for operational modal analysis, in: ISMA2012, Leuven,
Belgium, 2012, pp. 755–764.
[32] R.B. Randall, M.D. Coats, W.A. Smith, Repressing the effects of variable speed harmonic orders in operational modal analysis, Mech. Syst. Signal Process.
79 (2016) 3–15.
[33] R.B. Randall, W.A. Smith, New cepstral methods for the diagnosis of gear and bearing faults under variable speed conditions, in: ICSV23 Conference,
Athens, July, 2016.
[34] J. Antoni, Fast computation of the kurtogram for the detection of transient faults, Mech. Syst. Signal Process. 21 (2007) 108–124.
[35] N. Sawalhi, R.B. Randall, H. Endo, The enhancement of fault detection and diagnosis in rolling element bearings using minimum entropy deconvolution
combined with spectral kurtosis, Mech. Syst. Signal Process. 21 (2007) 2616–2633.
Another random document with
no related content on Scribd:
“Are you troubled with the fear that your hearers lack the
intelligence to appreciate the fine points of your analyses?
Let such fear vanish, for there can be no lack of
understanding with these hearers. Some of them are men
of experience in campaigns; others are in the habit of
instructing themselves from books, and have come to the
performance each furnished with a scroll with which to
freshen his memory, while each also is fully armed with
mother-wit. Have no fear therefore. They will have full
understanding of all that you may wish to discuss before
them.”
Müller proceeds to make an analysis of the purport of the
references in this passage, pointing out that the experience of old
campaigners would help them to the appreciation of the robust and
stirring compositions of Æschylus, while the scholarly habits of the
lovers of books would keep them in close sympathy with the complex
intellectual problems considered by Euripides.
The sharper edge of the comparison is directed against Euripides,
who is always referred to by Aristophanes as a book-worm. Müller
further contends that the references to each hearer being “provided
with his little book” (or book of the play) must be understood as
merely a piece of humorous exaggeration, as during the last years of
the Peloponnesian war, when the resources of Athens had been
seriously diminished, when poverty was general, and men’s minds
were agitated with the excitement of the campaign, few people could
have had the money for the buying, or the leisure for the reading, of
books.
Athenæus concludes, from a fragment of the comedy writer Alexis
(a contemporary of Alexander), that it was not until the time of
Alexander that the reading of books played any important part in the
intellectual life of the Greeks.[73] In the comedy of Prodicus, entitled
The Choice of Hercules, portions of which have been preserved in
the Memorabilia of Xenophon, Linus, the instructor of Hercules, is
represented as directing his pupil to select for his reading one out of
a number of books which are lying before him. Among the authors
whose works are specified in the list are Orpheus, Hesiod, Homer,
Chœrilus, and Epicharmus. (The last named is the first Greek writer
of comedy of whom we have any trustworthy account. His first work
was produced about 500 b.c.)[74] Hercules, passing by the poetry,
seizes a volume on cookery, the work of an actor named Simos, who
was also famous as a cook.[75]
Artemon, a grammarian of Cassandria in Macedonia, who wrote
shortly after the death of Aristotle and who made a collection of the
letters of Aristotle, published a dissertation on the collecting and the
use of books, which gives ground for the impression that in his time
there was already in Macedonia or Northern Greece a circle of
bibliophilists, ready to give attention to the counsels of this
forerunner of Dibdin, and possibly able also to pay for the books.
A piece of evidence against the contention that the price of books
was high in the time of Plato, is supplied, according to certain
commentators, by Plato himself. From a paragraph in the Apology
Boeckh[76] understands that some kind of book-trade must have
been carried on in the orchestra of the theatre (during the time, of
course, when no performance was going on), and that the writings of
Anaxagoras were offered for sale for one drachme; and
Buchsenschutz[77] takes the same view of Plato’s reference. The
words used by Plato are put into the mouth of Socrates, who is
represented as contending; first, that the opinions for the utterance
of which he has been charged with heresy or impiety, are in
substance the same as those already given to the world by
Anaxagoras and others; second, that these views have been so
widely published that they have become public property, for the
quoting of which no single person can properly be held responsible;
and thirdly, that they can be obtained in the theatre for a drachme.
The particular writings of Anaxagoras to which Socrates here refers,
contain his theories concerning the nature of the sun, the moon, the
earth, and the creating power of divinity. Schmitz is, however,
inclined to believe not that the books containing these doctrines
could be purchased in the theatre, but that the theories of
Anaxagoras were at the time freely quoted in the popular dramas
(such as those of Euripides), and that it was in listening to these
plays in the theatre that the public could without difficulty obtain a
knowledge of the new views.[78]
The usual price of admission to the Athenian theatres was, in the
time of Pericles, two oboli, or about six cents, but on special
holidays, when the performance continued three days, this price was
often raised to a drachme, or eighteen cents.[79] In the absence of
any other references to this supposed practice of turning orchestra
stalls into book-stalls, the weight of probability appears to favor the
conclusions of Schmitz rather than those of Boeckh.
Schmitz admits that it is not practicable to find in the existing
dramas of Euripides examples of such presentation of the
Anaxagorian theories of the universe, but he points out that a large
portion of the writings of this author was undoubtedly lost in the
destruction of the great war, and that this same war prevented any
wide distribution of the authenticated copies, although many of the
tragedies were so popular that the songs from them were sung
throughout the land. By the end of the war the fame of the tragedies
had reached Sicily, although very few of the manuscripts could yet
have got across the sea. After the defeat of the Athenians before
Syracuse, some of those who had been captured or who, escaping
from the Syracusans, had wandered over the island, found a
temporary livelihood or even purchased their freedom by reciting the
plays of Euripides, and on their return to Athens they took occasion
to express to the poet their gratitude for the timely service rendered
by his genius.[80]
To the coast cities of Asia Minor, as well as throughout the Greek
colonies of the Mediterranean, had come the fame of the new
tragedian, although here also copies of the plays themselves appear
to have been very scarce. Plutarch relates[80] that the inhabitants of
Caunus (a city of Caria), when besought for shelter by an Athenian
vessel chased by pirates, wanted first to know whether the
Athenians could recite for them the songs of Euripides.[81] It is to be
hoped that the Caunusians did not insist upon being paid in
advance, and upon having the recitations made before they
permitted the hard-pressed vessel to gain the shelter of the harbor.
In all places and among all classes where Greek was the language,
the songs of Euripides appear to have secured an immediate
popularity, while by the scholars also was given an appreciation no
less cordial. Both Plato[82] and Aristotle[83] ranked Euripides above
Sophocles and Æschylus.
Alexander the Great entertained the guests at his banquets by
reciting long passages from Euripides.[84] Throughout Greece these
tragedies appear for many years to have been the compositions
most frequently selected for public readings. Lucian relates[85] that
the Cynic Demetrius, who lived in Corinth in the first century, and
whom Seneca refers to as a new friend, heard an “uneducated man”
read before an audience The Bacchantes of Euripides. As the reader
came to the lines in which the messenger announces the “terrible
deed” of Agave and the fearful fate of Pentheus, Demetrius snatched
the book from his hands with the words: “It is better for poor
Pentheus to be murdered by me than by you.” The point of interest
for Lucian (who wrote about 150 a.d.) was the play on the term
“murdered,” and for us the example of the practice, in the first
century, of the public reading of standard literature, so general that
an audience (rather than not to hear the composition) would listen
even to an “ignorant reader.”
Returning to the question of the distribution and price of books, we
find a reference by Xenophon[86] to some “chests full of valuable
books” having been saved “with other costly articles” from the cargo
of an Athenian vessel shipwrecked at Salmydessus, a city on the
Euxine.
This appears to be the earliest reference on record to any sending
of supplies of books from Greece to the colonies, but even here
there is no evidence that the volumes were forwarded by dealers,
and it is probable that the “chests” contained the private library of
some wealthy Athenian collector who had migrated to Pontus. There
is no question, however, but that in the time of Xenophon (445-355
b.c.) Athens was the centre not only of the literary activity of Greece,
but of any book-trade that existed.
It seems evident that in Greece, as later in Rome, the earliest
booksellers were the scribes, who with their own labor had prepared
the parchment or papyrus scrolls which constituted their stock in
trade.
The next step in the development of the business was a very
natural one, namely, the introduction of the capitalist, who, instead of
working with his own hands, employed a staff of copyists and sold
the products of their labor. It is only surprising that the continued high
price paid for fair copies of noted works and the steady demand for
such copies, should not have tempted dealers more rapidly into the
business. The principal obstacle was for many years the difficulty of
securing a sufficiency of skilled copyists the accuracy of whose work
could be trusted. According to Schmitz, there is no mention of the
appearance of booksellers in Athens earlier than the fifth century
b.c.
The Athenian comedy, which touched with its keen raillery every
phase of life, whether public or private, did not overlook this new
mode of occupation. The references are as a rule not complimentary,
but, as the comedians spared nothing in their mockery, the fact need
not stand to the discredit of the first booksellers. Possibly the earliest
mention of the trade is by Aristomenes, who, in a comedy entitled
The Deceivers (performed about 470 b.c.), speaks of a “Dealer in
Books.” Cratinus, in his play The Mechanics (written about 450 b.c.),
mentions a copyist (βιβλιογράφος)[87]; Theopompus, writing about
330 b.c., uses the term “bookseller”[88] (βιβλιοπώλης); Nicophon
gives a list of “men who support themselves with the labor of their
hands” (χειρογάστορες), and in this list groups the bibliopoles in with
the dealers in fish, fruit, figs, leather, meal, and household
utensils.[89] It would seem as if in this instance the term βιβλιοπώλης
must have been used as synonymous with or at least as including
βιβλιογράφος, the scribe and the seller of the manuscripts being one
and the same person. Antiphanes, born in Rhodes b.c. 408, who is
credited by Suidas with having written over three hundred dramas,
which were very popular in Athens, refers to “book-copyists,” and
also to books which had been “sewed and glued.”[90] The comic
writer, Plato, who was a contemporary of Socrates, makes first
mention of “written leaves,” i. e., papyrus. The term used by him,
χάρται, was, according to Birt, when standing alone, more usually
applied to leaves of papyrus prepared for writing, but still blank;
χάρται γεγραμμένοι standing for the inscribed leaves.
We may conclude from Nicophon’s having included the booksellers
in his list of traders that they had their shops or stalls on the market-
place. Eupolis also speaks of the “place where books are sold,” (οὗ
τὰ βιβλία ὤνια),[91] and it appears therefore that as early as 430 b.c.
a special place in the market must have been reserved for the book-
trade—an Athenian Paternoster Row, or, more nearly perhaps, a
Quai Voltaire. It was, however, not until the time of Alexander the
Great that the business of making and selling books—that is,
attested copies of the works of popular writers—appears to have
developed into importance.
Until the business of book-making had become systematized, the
admirers of a poet or philosopher were obliged to supply themselves
with his works through their own handiwork, unless they were
fortunate enough to possess slaves educated as scribes. This test of
the reader’s admiration was assuredly rather a severe one. It is
certain that the number of disciples of modern authors would be
enormously limited if, as a first condition for the enjoyment of their
writings, the would-be readers were under the necessity of
transcribing the copies with their own hands. Imagine the extent of
the task for the admirers of Clarissa Harlowe, or for those who
absorbed their history through the ninety odd romances of G. P. R.
James!
As the supply of educated slaves increased, there was, of course,
less need for individual scholars to devote their own handiwork to
copying of manuscripts for their libraries. It was cheaper to employ
the labor of slaves, and to use their own time for more important
work. The names of some of the slaves who did good service as
scribes have been preserved in history. Mention has already been
made of Cephisophon, the slave, secretary, and personal friend of
Euripides. One of Plato’s dialogues is distinguished by the name of
Phædon of Elis, who had been sold as a slave in his youth and had
been employed as a scribe. The attention of Socrates was attracted
by his capable work, and he persuaded Crito to purchase his
freedom.[92]
The poet Philoxenus of Cythera was sold as a slave to
Melanippides (the younger), whom he served as a scribe, and whose
poetry he was said to have surpassed with his own productions.
There are many similar instances both of slaves who succeeded in
securing an education and in doing noteworthy literary work, and of
men of education who had, through the fortunes of war or through
the loss of their property, fallen into the position of slaves, and who
were then utilized by their masters for literary work.
There is also evidence that the state caused intelligent slaves to be
instructed in writing in order to be able to use them for work on the
public records or as clerks for the officials.[93]
It is to be borne in mind that the (to us) extraordinary extent to
which the Greeks were able to develop their power of memorizing
enabled them often to trust to their memory where modern students
would be helpless without the written (or the printed) word. “My
father,” says Niceratus in The Banquet of Xenophon, “compelled me
to learn by heart all the poetry of Homer, and I could repeat without
break the entire Iliad and Odyssey.”[94] The boys in school were
given as their daily task the memorizing of the works of the poets,
and what was begun under compulsion appears to have been
continued in later life as a pleasure.
Such an exceptional development of the power of memory, making
of it almost a distinct faculty from that which the present generation
knows under the name, may properly be credited with some
influence upon the slowness of the growth among the ancients of
any idea of property in an intellectual production. As long as men
could carry their libraries in their heads, and when they desired to
entertain themselves with a work of literature, needed only to think it
to themselves (or even to recite it to themselves) instead of being
under the necessity of reading it to themselves, they could hardly
have the feeling that comes to the modern reader (if he be a
conscientious person) of an indebtedness to the author, an
indebtedness which is in large part connected with the actual use of
the copy of the work. In the early Greek community, a very few
copies (or even a single copy) of a great poem were sufficient in a
short space of time to place the work of the poet in the minds of all
the active-minded citizens, such men as would to-day be frequenters
of the bookstores. In the Homeric times it proved, in fact, to be
possible to permeate a community with the inspiration of the national
epics without the aid of any written copies whatever. For the service
rendered by these early bards, the community might, and very
possibly did, feel under an obligation of some kind, but the individual
reciter who had absorbed the poems into the possession of his
memory, and the readers to whom he transmitted the enjoyment of
these poems, could not have suggested to them any such feeling of
personal obligation to the poet as is experienced by the reader of to-
day who is called upon to buy from the author, through the publisher,
the text of any work of which he desires the enjoyment. The Greek of
these earlier times needed no texts and dreamed of no bookseller.
He inherited from his ancestors the poetry of the preceding
generation with the same sense of natural right as that with which he
took possession of his ancestral acres; and he absorbed into his
memory for his daily enjoyment the poetry of his own day with the
same freedom and almost the same unconsciousness as that with
which he took into his lungs the air about him. In this way the
literature with which he had to do became really a part of himself,
and he may be said to have become possessed of it in a way which
would hardly be possible for one who was simply a reader of books.
It is not easy to realize how much we have lost in these days of
printed books in losing this magnificent power of memorizing our
literature and carrying it about with us, instead of going to our
libraries for it and taking it in by scraps. How much more to us, for
instance, would Shakespeare’s plays stand for, if they could be
stored in our heads ready for use when wanted, instead of being
available, as at present, only in the occasional reading circle, or the
still less frequent Shakespearian revival.
An author who seems to have taken exceptional pains to secure a
circulation for his productions was Demosthenes, but it is to be borne
in mind that his interest as a politician, or perhaps it is fairer to say
as a statesman, desiring to arouse public opinion in behalf of his
policy, was probably even keener than his ambition as an author
hoping for a popular appreciation of his eloquence. Whatever the
motive or combination of motives, it appears that after the delivery of
an oration he would act as his own reporter, writing out revised
copies and distributing the same among his friends for
distribution.[95] He had a special interest in securing a wide popular
circulation for his speeches in the matter of the guardianship, and for
those against Æschines and in behalf of Phormion, and the copies of
these,[96] prepared by his own hand or under his orders, certainly
came into the hands of many readers. Copies of the speeches made
by Demosthenes against Philip must have been brought to the latter
by some of the orator’s opponents. Such at least is the interpretation
given by Schmitz to the well known exclamation of Philip: “If I had
heard him speak these words, I should myself have been compelled
to lead the campaign against Philip.”[97]
An early reference to the practice of making publication of a book in
any formal manner (as distinguished from the permission accorded
to friends to make transcripts for their own use) is given by Isocrates,
writing about 400 b.c. He speaks of hesitating to publish his
Panathenaicus (φανερὰν ποιῆσαι, διαδιδόναι). He began the work,
says Birt, when he was already ninety-four, was obliged to leave it on
account of illness, but took it up again three years later, and it was
then that (conscientious author as he was) he hesitated to give the
volume to the public, because some friend to whom he had read it
was not fully in accord with its conclusions.[98]
The development of the trade of making and selling books came
but slowly, but received no little impetus through the taste for
literature implanted by Aristotle in his royal pupil Alexander. The
latter appears to have given frequent commissions to his friend
Harpalus for the purchase of books. From the mention by
Plutarch[99] it has been thought Harpalus must have been sent from
Asia with instructions to procure for Alexander a long series of works
whose titles are given. Schmitz points out, however, that Alexander
could hardly have been in a position during his Asiatic campaigns
and journeyings to collect a library, and these commissions to
Harpalus must have been made at an earlier date, before Alexander
had left Macedonia and while the “friend of his youth” was sojourning
in Athens.
The one point that is clear and that is of interest to us in this
connection is that, at about 330 b.c., Harpalus was able to purchase
in Athens, which was already referred to as the centre of the book-
trade of Greece, “many tragedies of Euripides, Æschylus, and
Sophocles, dithyrambic poems by Telestes and Philoxenus, the
historical writings of Philistus of Syracuse, together with a number of
rare works.” From Athens also, at about the same time, Mnaseas,
the father of Zeno, brought to his son, in the course of “various
business journeys,” copies of all the “published writings of
Socrates.”[100] There is also a reference in Dionysius of
Halicarnassus[101] to the many volumes of Isocrates which had been
published (literally “placed among the people”) by the Athenian
booksellers. Schmitz speaks of the great impetus given to the
production of books, that is, to the reproduction of copies of the
works of the writers accepted as standard, by the literary taste and
ambition of many of the successors of Alexander, notably the
Ptolemies in Alexandria and the Attali of Pergamum. He mentions
further that as one result of the greater and more rapid production of
manuscripts there was a considerable deterioration in the quality and
standard of accuracy of the copies. The complaints of readers and
collectors concerning the errors and omissions in the manuscripts
begin from this time to be very frequent. It would, in fact, have been
very surprising if the larger portion of the manuscripts that came into
the market had not been more or less imperfect. As soon as their
production became a matter of trade instead of, as at first, a labor of
love on the part of scholars, the work of copying came into the hands
of scribes working for pay, or of slaves, and partly from lack of
literary interest, partly also doubtless from pure ignorance, the many
opportunities for blunders appear to have been taken full advantage
of. Fortunately it was only the readers who suffered, and the authors,
long since dead, were spared the misery of knowing how grievously
their productions were mutilated. Different sets of copyists naturally
came to have varying reputations for accurate or inaccurate
manuscripts. Diogenes Laërtius[102] speaks of skilled scribes sent
from Pella by Antigonus Gonatas to Zeno, the Stoic, to be employed
in making trustworthy transcripts of that philosopher’s works, for
which the Macedonian king had a great admiration. Diogenes tells us
further that when Zeno, who came from Citium in Cyprus, first
arrived in Athens, he had suffered shipwreck and had lost near the
Piræus, just as he was reaching his journey’s end, both his vessel
and the Phœnician wares which constituted its cargo. Discouraged
by his misfortune, he strolled gloomily along the avenue from the
harbor (“by the dark rows of the olive trees”) toward the city in which
he was now a poverty-stricken stranger. As he reached the market-
place and passed a bookseller’s shop, he heard the bookseller read
aloud. He stopped to listen, and there came to him words of good
counsel from the Memoirs of Xenophon. “Cultivate a cheerful
endurance of trouble and an earnest striving after knowledge, for
these are the conditions of a useful and happy life.” Cheered by this
hopeful counsel, Zeno entered the bookseller’s shop and inquired
where he should find the teachers from whom he could learn such
wise philosophy. In reply, the bookseller, evidently well informed as
to the literary life of his city, pointed out the cynic Crates who
happened to be passing at the moment.[103]
The intellectual life of Athens, which a century before had centred
about the dramatic poets, appears at this time to have been
principally devoted to the study of philosophy. Among the other
noteworthy changes that had been brought about during the hundred
odd years since the death of Euripides, was the evolution of the
bookseller or publisher who had now evidently become a permanent
institution, and whose shop is recognized as a centre of literary
information.
We can imagine some European student landing, two thousand
years later, in Boston and applying, with an inquiry similar to that put
by Zeno, at the corner shop of Ticknor & Fields. How easy would
have been the answer if at the moment had passed along
Washington Street the slender figure of Emerson!
The question has been raised whether the passage from Diogenes,
above quoted, might not indicate that booksellers or others, owning
manuscript copies of popular works, made a regular business of
reading aloud to hearers paying for the privilege. Such a practice
would apparently have fitted in very well with the customs of the
time, and would have met the needs of many of the poorer students
for whom the purchase of manuscripts was still difficult. It would also
have formed a very natural sequence to the long-standing custom of
the recital from memory of the works of the old poets. While it seems
very possible from the conditions that public readers found
occupation in this way, there is no trustworthy evidence to such
effect.
While Zeno was teaching in Athens, a certain Callinus appears to
have won distinction among the scribes of Athens for the accuracy
and beauty of his manuscripts. The Peripatetic philosopher Lycon,
who died about 250 b.c., bequeathed to his slave Chares such of his
writings as had already been “published,” while the unpublished
works were left to Kallinus “in order that accurate transcripts of the
same might be prepared for publication.”[104]
As the rivalry which continued for some time between the Ptolemies
and the Attali in the collecting of libraries caused the price of books
in Athens to remain high, a further result was the establishing of
other centres of book-production, of which for a long time the island
of Rhodes was the most important. By about 250 b.c., the literary
activity of the Alexandrian scholars, encouraged by Ptolemy
Philadelphus, to whom the founding of the great library was probably
due, caused Alexandria to become one of the great book-marts of
the world.
After the first conquest of Greece by the Romans had been
practically completed by the capture of Corinth in 146 b.c., there
appears to have been a revival in Athens of the trade in books,
owing to the increased demand from the scholars of Rome, where
Greek was accepted as the language of refined literature and where
Greek authors were diligently studied. Lucullus is said by
Plutarch[105] to have brought from Rome (about 66 b.c.) many books
gathered as booty from the cities of Asia Minor, and many more
which he had purchased in Athens, together with a great collection of
statues and paintings.
The great hall or library in which his collections were stored became
the resort of the scholarly and cultivated society of the city, and its
treasures of art and literature were, according to Plutarch, freely
placed at the disposal of any visitors fitted to appreciate them. Sulla,
without claiming to be a scholar, was also a collector of Greek books.
He secured in Athens the great library of Apellicon of Teos, which
included the writings of Aristotle and of Theophrastus. Apellicon, who
died in Athens in the year 84 b.c., had a mania for collecting books,
and was reputed to be by no means scrupulous as to the means by
which he acquired them. If he saw a rare work which he could not
purchase, he would, if possible, steal it; and once he was near losing
his life in Athens in being detected in such a theft. His Aristotle
manuscripts, which were said to be the work of the philosopher’s
own hand, had been found in a cave at Troas where they had
suffered greatly from worms and dampness.[106] After the
manuscripts reached Rome they were transcribed by Tyrannion the
grammarian. He sent copies to Andronicus of Rhodes, which
became the basis of that philosopher’s edition of Aristotle’s
works.[107] Pomponius Atticus utilized his sojourn in Athens (in 83
b.c.) not only to familiarize himself with the great works of Greek
literature, but to cause to be made a number of copies of some of
the more popular of these, which copies he afterwards sold in Rome
“to great advantage.”[108]
There is a reference in Pliny to a miniature copy of the Iliad
prepared about this time, which was so diminutive that it could be
contained in a nutshell. He speaks of it as Ilias in nuce. Pliny refers
to Cicero as his authority for the existence of this manuscript, in
which he is interested principally as an evidence of the possibilities
of human eyesight. Its interest in connection with our subject is of
course as an example of the perfection which had been attained in
the first century before Christ in the art of book production.[109]
Notwithstanding the stimulus given to the production of manuscripts
by the increasing demand for these in Italy, books continued to be
dear, even through the greater part of the first century. The men of
Ephesus who were induced under the teachings of Paul to burn their
books concerning “curious arts” counted the price of them and found
it to be fifty thousand pieces of silver.
The history of Greek literature presents few other instances of the
destruction of books, whether for the sake of conscience or for the
good of the community, or under the authority of the state. There are,
however, occasional references to the exercise on the part of the
rulers of a supervision of the literature of the people on the ground of
protecting their morals or religion. Probably the earliest instances in
history of the prosecution of a book on the ground of its pernicious
doctrines is that of the confiscation, in Athens, of the writings of
Protagoras, which were in 411 b.c. condemned as heretical.
All owners of copies of the condemned writings were warned by
heralds to deliver the same at the Agora, and search was made
among the private houses of those believed to be interested in the
heretical doctrines. The copies secured were then burned in the
Agora. Diogenes Laërtius, by whom the incident is narrated, goes on
to say that the destruction was by no means complete, even of the
copies in Athens, while no copies outside of Athens were
affected.[110] The attempt to suppress the doctrines of the
philosopher by means of putting his books on an index expurgatorius
was probably as little successful as were similar attempts with the
doctrines of other “heretics” in later centuries.
The fact that high prices could be depended upon for copies of
standard works ought to have insured a fair measure of accuracy in
the manuscript. Complaints, however, appear repeatedly in the
writing of the time (i. e., the century before and that succeeding the
birth of Christ) of the bad work furnished by the scribes. Much of the
copying appears to have been done in haste, and with bad or
careless penmanship, so that words of similar sound were
interchanged and whole lines omitted or misplaced, and the
difficulties of obtaining trustworthy texts of the works of older writers
were enormously and needlessly increased. In order to enable a
number of copyists to work together from one text, it appears that the
original manuscript was often read aloud, the work of the scribes
being thus done by ear. This would account for the interchanging of
words resembling each other in sound.
Strabo, writing shortly before the birth of Christ, refers to an
example of this unsatisfactory kind of bookmaking.
The grammarian Tyrannion, in publishing in company with certain
Roman booksellers his edition of the writings of Aristotle, confided
the work to scribes, whose copies were never even compared with
the original manuscript. And, says Strabo, editions of other important
classics, offered for sale in Alexandria and Rome, had been
prepared with no more care.[111] The reputation of the manuscripts
transcribed at this period in Athens appears to have been but little
better. The making, that is to say the duplication and publishing of
books, had come to be a trade, and a trade of considerable
importance, but the men who first engaged in it appear to have had
little professional or literary standard, and not to have realized that
profits could be secured from quality of work as well as from quantity,
and that for a publisher a reputation for accurate and trustworthy
editions could itself be made valuable capital.
The publishers of Greece appear to have been characterized by
modesty, for not one of those who did their work at the time of the
greatest prosperity of the book-trade in Greece has left his name on
record for posterity. The days were still to come when every book
would bear its imprint bringing into lasting association the name of its
publisher with that of the author. The Greek publishers appear not to
have assumed, like the later Tonson, an ownership in their poets, nor
do we, on the other hand, find in the utterances of the poets any
expressions corresponding to the famous “My Murray” of Lord Byron.
Curtius speaks of a reference in an inscription to the “Ptolemy” or
“Ptolemaic” bookstore, but the name of the bookseller is not given. It
is only later, when the Greek book-trade was in its decline, that we
come across the names of two dealers in books, Callinus and
Atticus. They are mentioned as famous during the lifetime of Lucian
(about 120 to 200 a.d.), the former for the beauty and the latter for
the accuracy of his manuscripts. It is an interesting coincidence that
this Callinus, noted for the beauty of his texts, bears the same name
as the scribe commended three centuries before by Zeno for the
beauty and accuracy of his manuscripts. Their copies were much
prized and brought high prices, not in Athens only, but in scholarly
circles elsewhere. It is evident that each of these booksellers began
business as a scribe, selling only the work produced by his own
hands, but that as their orders increased it became necessary for
them to employ a number of copyists, whose script, receiving a
personal supervision and doubtless a careful collation with the
original texts, could be guaranteed as up to the standard of their own
handiwork. Of the other booksellers who were in Athens in his time
Lucian speaks very contemptuously. “Look,” he says, “at these so-
called booksellers, these peddlers! They are people of no scholarly
attainments or personal cultivation; they have no literary judgment,
and no knowledge how to distinguish the good and valuable from the
bad and worthless.”[112] Lucian had evidently a high standard of
what a publisher ought to be.
Some of these Athenian booksellers whom Lucian thus berates for
stupidity, appear also to have borne a poor reputation for honesty.
Among other misdeeds charged against them was one, the ethics of
which might have belonged to a much later period of bookmaking. In
order to give to modern manuscripts the appearance of age, and to
secure for them a high price as rare antiquities, they would bury
them in heaps of grain until the color had changed and they had
become tattered and worm-eaten. Lucian also satirizes the ambition
of certain wealthy and ignorant individuals to keep pace with the
literary fashion of the time, and to secure a repute for learning by
paying high prices for great collections of costly books, which, when
purchased, gave enjoyment “to none but the moths and the
mice.”[113] It was partly due to the competition of wealthy collectors
of this kind that, notwithstanding the great increase in the production
of copies, the price of books remained high, much to the detriment of
all impecunious students.
The beauty of the calligraphy of the manuscripts of Callinus is
known to us only through Lucian, but there are several writers who
bear testimony to the accuracy of the transcripts prepared by his
rival Atticus, who must, by the way, not be confused with the Roman
Atticus, the friend of Cicero. Harpocration of Alexandria, known
principally as the author of one of the first Greek dictionaries, makes
several references to the authority of the Atticus editions of the
speeches of Æschines and Demosthenes. The famous Codex
Parisinus of Demosthenes is believed by Sauppe to be based upon
the excellent textual authority of a manuscript of Atticus, and Sauppe
further contends that, if Atticus did not work from an absolute
original, he must have had before him a very well authenticated
copy. In the fragment of a work by Galen (who wrote in Rome about
165 a.d.) upon certain passages in the Timæus of Plato which had to
do with medicine, Galen makes Atticus his authority for the passages
quoted by him, as if we were indebted to this bookseller for the text
of the Timæus that has been preserved.[114]
From the time of Lucian the interest in books steadily increased,
book-collecting became fashionable, especially in Rome, and
bibliophiles and bibliomaniacs were gradually evolved. At this time
the beautifully written and carefully collated manuscripts which
emanated from Athens bore a high reputation as compared with the
much cheaper but less attractive and less trustworthy copies, which
were produced in Alexandria and in Rome. In the book-shops of
these two cities, during the first two centuries, a swifter and less
accurate system of transcribing appears to have prevailed, the work
being largely done by slaves or by scribes who did not have accurate
knowledge of the literature on which they were engaged, while the
necessity of a careful collating of each copy with the original appears
frequently to have been overlooked. Origen, writing about 190 a.d.,
speaks of confiding his works to the “swift writers of Alexandria” in
order to secure for them a speedy and a wide circulation. He was
looking for no other return for his labors than a large circle of
readers, and a large influence for his teachings, and the proceeds of
the sales of these “swiftly written copies” were in all probability
entirely appropriated by the booksellers who owned or who
employed the scribes.
After the conquest of Greece by the Romans the centre of book
production passed from Athens first to Alexandria and later to Rome.
For centuries to come, however, the book production of the world
was chiefly concerned with the works of Greek authors, and the
literary activity of successive generations drew its inspirations from
Greek sources; and the writers of Greece, whose brilliant labors
brought no remuneration for the laborers, gave to their country and
to the world a body of literature which at least in one sense of the
term can properly be called a magnificent literary property.
CHAPTER III.
Alexandria.
D URING the middle of the third century before Christ, the centre
of literary activity was transferred from Athens to Alexandria,
which became, under Ptolemy Philadelphus, and for more than three
centuries remained, the great book-producing mart of the world. The
literature of Alexandria was not, like that of Athens, and later that of
Rome, something of slow growth and gradual development; the
literary ambition and the resources of the second Ptolemy proved
sufficient to bring together in a few years’ time a great body of writers
and students and to place at their disposal the largest collection of
books known to antiquity.
The most important step in the undertaking of securing for the royal
young city of the Nile the literary leadership of the world was the
establishment of the great Museum, which appears to have
comprised in one organization a great lending and reference library,
a series of art collections, a group of colleges endowed for research
(of the type of “All Souls” at Oxford), a university of instruction, and
an academy with functions like those of the Paris Academy,
assuming authority to fix a standard of language and of literary
expression, and possibly even to decide concerning the relative rank
of writers. The Museum (whose name is of course evidence of its
Greek origin and character) is said to date from the year 290 b.c., in
which case the founding of it must be credited to Ptolemy Soter, the
father of Philadelphus, but its full organization and effective work
certainly belonged to the reign of the latter.
Schools of instruction and courses of lectures had, as we have
seen, existed at Athens for a century or more, and Athens had also
possessed as early as 300 b.c., at least one public library.
Alexandria, however, presents the first example of a university
established on a state foundation, and offering to literary and

(Download PDF) A History of Cepstrum Analysis and Its Application To Mechanical Problems Robert B Randall Full Chapter PDF

Uploaded by

Copyright:

Available Formats

You might also like

(Download PDF) A History of Cepstrum Analysis and Its Application To Mechanical Problems Robert B Randall Full Chapter PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Download PDF) A History of Cepstrum Analysis and Its Application To Mechanical Problems Robert B Randall Full Chapter PDF

Uploaded by

Copyright:

Available Formats

A history of cepstrum analysis and its

application to mechanical problems

Spatiotemporal Analysis of Air Pollution and Its

Freedom of Speech and Expression: Its History, Its

Overdoing Democracy: Why We Must Put Politics In Its

The Aqueous Chemistry of Polonium and the Practical

Truth to Power: A History of the U.S. National

Groundwater Sustainability: Conception, Development,

The History of Texas 6th Edition Robert A. Calvert

You: A Natural History William B. Irvine

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

A history of cepstrum analysis and its application to mechanical

E-mail address: b.randall@unsw.edu.au

2. Basic relationships and definitions

The original definition of the (power) cepstrum was:

C p ðsÞ ¼ jIflogðF xx ðf ÞÞgj2 ð1Þ

C c ðsÞ ¼ I1 flogðFðf ÞÞg ¼ I1 flnðAðf ÞÞ þ j/ðf Þg ð2Þ

Fðf Þ ¼ Iff ðtÞg ¼ Aðf Þej/ðf Þ ð3Þ

C p ðsÞ ¼ I1 flogðF xx ðf ÞÞg ð4Þ

C p ðsÞ ¼ I1 f2 lnðAðf ÞÞg ð5Þ

C r ðsÞ ¼ I1 flnðAðf ÞÞg ð6Þ

C h ðnÞ ¼ lnðBÞ; n¼0

If yðtÞ ¼ f ðtÞ hðtÞ ð10Þ

then Yðf Þ ¼ Fðf Þ Hðf Þ ð11Þ

logðYðf ÞÞ ¼ logðFðf ÞÞ þ logðHðf ÞÞ ð12Þ

rahmonic from harmonic

3. Early applications in mechanics and acoustics

3.1. Diagnostics of families of harmonics or sidebands

Fig. 1. Log spectrum and cepstrum for a gear fault.

3.2. Separation of source and transmission path

Fig. 4. Estimated cylinder pressure using the complex cepstrum [12].

4. Later developments up to the present

4.1. Applications of the real cepstrum in machine diagnostics

Fig. 6. Advantage of the analytic cepstrum for zoom spectra.

4.2. Applications of the cepstrum in modal analysis

Fig. 7. Application of the diagnostic indicator d from [19].

4.2.1. The differential cepstrum

4.2.2. Modal analysis using the cepstrum of response signals

Force cepstrum concentrated here

4.2.3. The mean differential cepstrum

4.2.4. MIMO systems

5. Editing time signals using the real cepstrum

5.1. Application to operational modal analysis

Examples of application of these procedures to OMA are given in Refs. [31,32].

5.2. Application to diagnostics of variable speed machines

The contributions of all my co-authors are gratefully acknowledged.

You might also like