J. Szőke
O N -L IN E MEASUREMENTS A N D
COMPUTERIZED DATA PR O C ESSIN G O F SPECTRA
S ^ o a n ^ a A ia n S & c a d e m i^ o f c S c ie n c e j
CENTRAL
RESEARCH
INSTITUTE FOR
PHYSICS
BUDAPEST
KFKI-72-5
J.
S Z Ő K E
COMPUTERIZED MEASUREMENTS AND DATA PROCESSING
ON-LINE MEASUREMENTS AND COMPUTERIZED DATA PROCESSING
OF SPECTRA
ABSTRACT
A description is given of a computerised measuring centre for opti
cal spectroscopy, established at the Central Research Institute for Physics,
Budapest, with which measured data from a set of instruments covering the
entire spectrum can be collected and prepared on-line for evalution on a
large computer. The steps employed in spectral data processing and exper
iences and conclusions gained by the author in applying these techniques are
dealt with in some detail.
KIVONAT
A szerző ismerteti a Központi Fizikai Kutató Intézetben létrehozott
optikai spektroszkópiai komputeres mérőközpontot, amely a teljes spektrumot
átfogó műszerpark digitális mérési adatait on-line módon gyűjti és előkészí
ti nagy-komputeres adatkiértékelésre.
A közlemény ismerteti a szinképi adatfeldolgozás menetét, a szerző
által vizsgált módszerek alkalmazása kapcsán nyert tapasztalatokat és meg
állapításokat .
РЕЗКИЕ
Описывается оптическо-спектроскопический измерительный центр, осно
вывающийся на использовании ЭВМ, созданный в Центральном институте физиче
ских исследований. Измерительный центр собирает и подготавливает к обработке
на большой ЭВ М цифровые данные, полученные с помощью приборов для измерения
спектра. Излагаются ход обработки спектроскопических данных, опыт, приобре
тенный автором в связи с применением использованных им методов.
INTRODUCTION
Modern research work concerned with materials requires an ever in
creasing amount of possibly precise information on mulecular structures and
physical parameters. Spectroscopic methods are especially valuable in this
field, because spectra represent all the interactions between the structural
elements of matter and the electromagnetic field. The complete spectroscopic
description of any material - as can be seen in Fig. 1 - requires a number
Fig. 1
Scheme of the computerized spectroscopic work
2
of spectra made under different conditions of electric and magnetic fields,
medium, temperature, etc. and with the use of several types of measurement
/absorption, reflection, emission and scattering/ covering the whole optical
spectrum. In the evaluation work it is necessary to study the correlations
between the different experimental spectra and those between the experimental
and theoretical spectroscopic results. But the spectra themselves, and the
problems of their interpretation are complex in nature, thus their evalua
tion demands much time and extensive human intervention. The only realistic
approach that offers the possibility of the easy typifying of the results
and yet remains sufficiently flexible in its operation is the use of computer
methods, despite the hard work that must be put into selecting the best pro
cedures and determining the limits of their application. /Fig. 1J
In this paper the results of the systematic work carried out in this
field at the Central Research Institute for Physics, Budapest are described.
I. COMPUTERIZED SPECTROSCOPIC LABORATORY
A really effective optical spectroscopic laboratory must have four
types of instruments. With refeißnce to the instrument set built at our
laboratory, these are as follows:
1/ Vacuum UV spectrometer with concave grating, applicable in the range from
80000 to 40000 cm-1.
2/ Several UV and IR single-beam spectrometers with 1 m plane-grating mono
chromators as well as other similar constructions with different detectors
and grating sets. The gratings are interchangeable. The range of application
is from 50000 to 300 cm 1 .
The IR spectrometers are suitable for measuring transmission, reflec
tion, and molecular emission spectra. The sample is held in a termaperaturecontrolled helium cryostat.
The UV spectrometers can be used for measuring all types of electronic
transitions: transmission, reflection, steady-state luminescence and excita
tion spectra, and singlet-singlet and singlet-triplet lifetimes, with or
without polarization. Measurements can be performed at any temperature between
room and helium temperature.
3 / Interferometers for IR and FIR regions
MicheIson-type interferometers. An IR version is under development.
The FIR region is used from 600 to lO cm ^. The sample is held in a helium
cryostat.
3
4/ Scattering photometer for Raman spectroscopy
A photometer of this type is now being developed in our laboratory.
Spectrometers of the above types have been in use for a considerable time,
but recently we had to modify the conventional instruments for new tasks.
The following principles were adopted in our constructions.
All instruments must work
applies to
- fast
- at high scanning speed
the schanning mechanism
- periodically
the optical units
- in a single-beam mode operation
- automatically
- with a high stability in time
the whole instrument
The measurements are performed using a photon counter and/or a lock-in
amplifier with four parallel gain outputs in decimal steps /1.10, 100, lOOO/.
The digital resolution in all outputs is 12 bits.
The analogue or digital results are transferred through a 60 m cable
to the computer center /See Fig. 2/ and then are fed, either directly, or
Fig.2
Organization of the on-line computerized spectroscopic laboratory
4
after analogue-to-digital conversion, through a computer-controlled data
channel multiplexer into the computer memory. The start and sampling signals
are generated by the spectrometer. The analogue-to-digital conversion and
the computer storage are automatized by a timer. After the measuring process
the data are recorded on magnetic tape. All computer programs are stored in
a disc memory from which the necessary parts can be retrieved.
In the computer center there is a small computer /ТРА-1001/ with its
different input and output peripherials. An analyser system which can be used
with a fast timer and serves as another facility for high-speed measurements
is also used in the center. Its maximum sampling speed is 5 psec. Lifetimes
in the nsec range can be measured by a Berlman-type instrument [l] modified
for the needs of our measuring system. /Fig. 3/
TRIGGSR
SIGNAL
Fig. 3
Arrangement for lifetime measurement in the nsec region
The working cycle of our computerized spectroscopic equipment
is the following: /Fig. 4/
a/ The computer gives an analogue signal to the temperature
controller.
b / If necessary, the gain is manually controlled with the aid
of an oscilloscope.
5
Fig. 4
Outline of the on-line measurement
с/ The amplification is stored in the computer memory.
d / The maximum signal is set manually.
e/ The data are fed into the memory /maximum 3000 data/.
The noise level is determined and the necessary number of
measuring cycles is calculated.
f/ The measuring cycle is started manually. The timer transmits
the sampling signals only after the synchronizing signal has
been received.
>
g/ The measuring cycle is carried out as follows: In the first
cycle, on storing instruction, the computer examirtes all the
channels with maximum information content and feeds this content
and the channel number, into the memory. During and after the
measuring cycle the stored spectrum can be seen on parallel
display in the spectroscopic and computer laboratories.
h/ The second cycle begins on the appearence of a new start pulse.
The results are read by the computer point by point from the
6
channel chosen in the first cycle and a weighted average is
calculated from the stored and new values and also stored.
When the prescribed number of cycles is exhausted, the measure
ment is finished and the computer waits for further manual
instructions.
j/ If the visually inspected display figure is acceptable, the
data are recorded on magnetic tape.
The magnetic tape recorder consists of a commercial tape recorder
and an interface from which the data can be passed through a satellite small
computer /ТРА 1001/ into an ICT-1905 computer.
Before computer evaluation all the stored spectra are plotted by
an X,Y~ray recorder directly from the magnetic tape.
II. DATA PROCESSING
The digital spectra are evaluated in a big computer by subroutines
developed specially for data processing and are called the standard library
routines. The organization is suitable for the compilation of simple MAIN
programs.
1 / Correlations and calibrations
In the first step the data of the measured spectra must be correct
ed and calibrated to eliminate instrumental distortions appearing in the
numerical values of x- and y-coordinates.
a/ *lcoordinates
The exact values of the x-coordinate for every data can be
determined only by a spectral lamp calibration procedure. The calibration
curve is calculated by polynomial approximation.
b/ ^coordinates
Distortions caused by the instrument characteristics can be
eliminated by using special test material /e.g. quinine sulphate in
fluorescence spectroscopy/. In these cases reported data are usually applied.
The deviation of our measurements from the reported data gives a correction
array. Errors in the reported data do not cause any problem in the evalua
tion since the digital results can be corrected again at any time.
7
с/ £dgustment_gf_regions
This is important for electronic absorption spectroscopy where
the complete spectrum consists of several fragments. In our experience only
the Lambert law is applicable, even though it involves heavy technical
problems. The Beer law is valid only for low concentrations.
2 / Transformations
First step in the evaluation work is to transform the data into a
suitable form for the estimation. This procedure is extended to both the
x- and y-coordinates.
a/ ^coordinates
A number of variables are used to express the energy of the
radiation field /erg, cal/mol, v, v*, fresnel, eV etc/ or the length of
waves /А, nm, ym, my, etc/. The transformation processes are available in
the library as standard functions.
Ь/ YIcoordinates
The values of the y-coordiante give the intensity of the lightmatter interactions. Basically two types of spectra can be distinguished
from this point of view: relative and absolute spectra. In the first case,
it is necessary to calculate the ratio of two single-beam spectra. Since
the intensity values vary in a wide range, a logarithmic representation is
very often used. This form of spectrum is applied for storing, too.
с/ ZQterferograms
In contrast to conventional spectrometers, interferometers
do not contain a dispersive element, so the detector can not discriminate
between different radiation frequencies. The interferogram produced by
the usually applied Michelson-type interformeters transforms the polychromatic
signal as a whole. The intensity data can be described by the simple cosine
transform
+ 00
I(x)
j I(v) cos (2tt
x
v)dv
— 00
where I(x) is the intensity of the output signal as a function of the
mirror displacement x /integrated intensity at path difference x/ and I(v)
is the intensity value at frequency v, which can be obtained by the Fourier
transform
8
т\
IO)
I (x ) cos (2тг.x. v)dx
— оо
This simple procedure, in general, gives a very good spectrum, however,
it is difficult to construct interferometers which are perfectly compensated
over the wide frequency range, and this limitation leads to the appearance
of sine components in the interferogram. The general description of an interferogram is given by the following complex Fourier transform pair:
+oo
I (x ) = \ I(v ) exp [2?t.i .x .v] dv
I.(v) =
I I (x) exp [-2n .i .x .v] dx
—oo
Since the interferogram is an experimental result, the second
transform has to be calculated. Richards [2] suggests the following cosine
algorithm for computer evaluation
I (vi) = Fo + 2
N
£ Fn c°s(n.v..Tr/N )
n=l
where N is the path difference and N is the number of experimental points
in the frequency region studied. A is the average value of the last 10 points
and Fq is the intensity at zero path difference while
is the experi
mental intensity value of the interferogram after apodization /in Richard's
procedure
Fn = l(xn - A )/.
In Fig. 5
the interferogram and the high resolution transform of
the spectrum of uranyl-tris-carbonate can be seen.
3/ Smoothing
This procedure is important from the point of view of dataprocessing, since the mathematical approximations used are very sensitive
to the noise superimposed on the information. A number of methods of eliminat
ing noise from spectra are known. The most frequently used of these ares
9
.Ш 11М И М Н М 1Ш М *Д |ш 111ш и ш ш |Н 11М 11М 11111111Ш 11М ш ш тти 1||||ш ш 1Ш 111М 1тм 111ш н 1111Ш 1111Н 111||11Н 11111ш м ||
t n i i i m l l l l t i 1i I l l t l n i li l i i i ii í 1 11II i 1 1i т ■I ti I l i i n 111U 1il >t i i i I ■l l l i i i i t i I i : >i 1i П i ■ti >11 111 I И; t ii i ii l i i . l il iltlill
Fig. 5
Interferogram /upper curve/ and high resolution transformed
spectrum/lower curve/ of uranyl-triscarbonate fluorescence
spectrum
a / Fourier_Transform
In principle this method may be regarded as giving an exact
solution to the problem, but a practical difficulty restricts its usage.
This can be illustrated by Figures 6 and 7. In Fig. 6 the original spectrum
of Fig. 5 has been superimposed with 1 per cent of noise /in normal distribu
tion/ and in Fig. 7 with 5 per cent of noise. The Fourier transforms of the
two spectra demonstrate well that the reliability of the information rapidly
decreases with increasing noise /see Fig. 5/. With 1 per cent noise the
10
"limit of noise" can be easily evaluated but in the case of 5 per cent
noise the information and the noise are not separable, so the method is un
suitable .
b
íiiiiiiiiiH iM iiiiH illliiiiiiiiiiH iiiiiiiiiiiitiiiiH iiiiiiiiiiiiiiiH iiiiH iilU H H iiiH H iiiitié iiiiiiiH iiiiiiiiiiii
Fig. 6
Noise elimination by Fourier transform.
a: Noisy curve. Relative value of noise: - 0.5 per cent,
b: Fourier transform of the noisy curve,
c: Retransformed spectrum.
- l i
fe
iiiiiihmiiiiiiiliiiliitimniiHiimimiiiiiiiiiimimiiimiiiiiiiiniim tHiiitiiim iiiiiiiiiinim iiiiiiiiiiiiiiiiiiiiiiHiiiuiii
Fig. 7
Noise elimination by Fourier transform.
a: Noisy curve. Relative value of noise: ± 2.5 per cent,
b: Fourier transform of the noisy curve,
c: Retransformed curve.
b / Convolutions
The most advanced procedure is based on the least square method
developed by Savitzky and Golay [3]. This can be used successfully if the
12
experimental data are equidistant. The mathematical formalism is
Y
#
j
+m
.
■ i=-m
Л
ci
where Y j represents the data of the smoothed curve, С± and I. are the
elements of the data points of the convolution function, and of original
data array, respectively.
This procedure can be applied cyclically until the required extent
of smoothing is obtained. The numerical values of the convolution function
were given by Savitzky and Golay.
In our work convolution smoothing is also used in two other forms:
five point quadratic and nine point quartic smoothing.
с/ iDt§EE2 i§tion
A special application of the interpolation procedure for smoothing
arises when the experimental data are not equidistant, which is the case
usually encountered in spectroscopic practice. Smoothing has to be applied to
the experimental points before their multiplication since the last procedure
lends to the noise the character of information.
This procedure can also be used cyclically. The number of cycles
can be determined by a logical condition or can be calculated from the
experimental data. In the 5th step of the working cycle /see before/ the
value of the noise and the cycle number /п/ of the measurement have to be
determined. During the measurement the noise is reduced by l//n. The remain
ing noise can be eliminated in the smoothing step by two factors: the length
/к/ of the convolution function and the cycle number /Nc / of smoothing. Both
factors decrease the noise according to the formula
z/zQ = l/(Nc .k) 2
where
z
and
z
are the initial and final values of the noise, respectively,
о
If the permitted value of the noise /^ / is known, the number of smoothing
cycles can be calculated from the relation:
4/ Interpolation, multiplication and data selection
The smoothed data are usually not equidistant and the values of
x-coordinates are not rounded off. In order to calculate the proper values of
13
a spectrum at correct coordinates, Lagrange's interpolation formula is
used in the approximation to 5th degree. This method is applicable to
multiplication of the experimental data as well, but in this case the
spectrum must be noiseless.
Data selection is important for storage. It is preferable that
the spectrum should contain the lowest number of points wich carry the
complete information. The most effective factor to shorten a spectrum is
to omit the x-coordinates, but this is possible only if the data array is
equidistant. In this case the first x-coordinate and the interval between
the coordinates must be known.
The interpolation formula is a very effective tool for data selec
tion on the condition that the interpolation procedure can reproduce the
original spectrum to the desired accuracy /less than 0.1 per cent/. The
procedure reduces the data array by one-half in each cycle.
5 / Qualitative analysis of compounds
This type of work is extensively used in infrared spectroscopy.
For these purpose the spectral data must be normalized. Here the most
important parameters of the experimental spectra are the frequency and the
intensity of peaks, the latter is usually given by a relative value /weak,
medium, strong, etc./. For the determination of these parameters the first
step in the evaluation is the so-called "peak-find" operation based on the
Savitzky-Golay method [з]. The second step is a logical examination of the
experimental data, using the conclusions from experimental spectroscopy.
This computer procedure is similar in structure to the conventional
method but uses a huge mass of conclusions for the comparison and all
statements are numerically evaluated.
6 / Fine structure of spectra
Mathematically defined, a spectrum consists of a number of
components describing allowed and partly forbidden transitions. Their
intensities vary in a wide range. Some bands overlap each other completely.
The aim of spectroscopic work is to determine the parameters of the com
ponent band and to assign them to vibrations. However, the blurring of the
fine structure as a result of the very complexity of the spectrum presents
a considerable problem. The only workable analytical approach to the intense
components of the spectrum is to take the low intensity components to be
perturbations of the band shape. But this means that, if the Gaussian
14
distribution is accepted as the true shape of the component bands, the
perturbation causes a Lorentzian-type distortion, and since the evaluation
methods are not exact, this distortion can make the problem unsolvable.
Nevertheless, in our experience the Gaussian approximation is effective
if the perturbing bands are of low intensity. The analysis gives better
results when a Voight approximation is used, where the band shape is
described as a Gauss-Cauchy product function /see. in paragraph 6c./
v.
l
F(v.)
In this case the mathematical procedure is more complicated because the
approximation involves four parameters as compared with three of the
Gaussian approximation. The time and memory required for the calculation
increases exponentially with the number of parameters.
In our experience the spectrum components in electronic spectra
can be adequately described by the Gaussian approximation. Using Voight
approximation, the Gauss-Cauchy ratio can be 5:1 in a favourable case,
but if the resolution of the fine structure is poor, it may be as small as
1:1. On the other hand, the experience in infrared spectroscopy is that
the Cauchy approximation is usually dominant.
[4]
After the general considerations we examine the steps of the
spectrum analysis individually.
a / Determination_of_the_genuine_spectrum
The genuine spectrum is free from instrumental distortions of
whatever source /optical, electronic, mechanical/. The character of these
effects can be described by the convolution
-f-OO
+0°
T(Vi) = ^ К (v - v.^) s(vi)dv/
— oo
Where T (v^
and
respectively; and
К (v)dv
— oo
s(v^) are the experimental and the genuine spectrum,
K(v) is the instrument function /it simplifies the
mathematical procedure if this is normalized to 1/.
Although the convolution broadens the bands of the spectrum and
decreases its amplitude, it does not change the integral intensity, lor
practical spectroscopy the genuine spectrum is preferable because:
- it is completely independent of the instrument;
- the fine structure is more distinct than in the experimental
spectrum;
- the determination of the energy value is more accurate
- the assignment is more reliable.
The elimination of instrumental distortions is a typical deconvolu
tion problem
— OQ
In the solution of this equation the determination of the inverse function
of К (v) is problematic. The direct method is the Fourier transform, but
the usefulness of this method is limited to cases where the noise is low. [,'jJ
The indirect iterative computer method proves to be the best,
since its application does not require complicated mathematical considera
tions and conditions. Details of the method are presented in a valuable
summary by Seshadri and Jones. [4 ]
In our work the instrumental distortions are expressed by an
experimental instrument function wich can be obtained from singlet spectral
line. The finite slit effect is dominant in this function, wich is in the
ideal case a triangle. As a result of optical aberrations this triangle
becomes asymmetric and Gaussian in character. Fig. 7 shows an instrument
function measured with a quartz spectrometer of Cary-Beckman type.
The deconvolution method we used was based on the
Y -or pseudo
deconvolution technique developed by Jones et al. [б]
In order to improve the convergence multiplication by 2 was applied
for correction in the first step.
In Fig. 9 the experimental /е/ and genuine /g / spectrum of the
fluorescence from uranyl-tris-carbonate are presented, where Fig. 8 has
been used as an instrument function. The accuracy of the result is better
than 0.1 per cent for all points /as can be tested by convolution/.
b / Determinatign_of_the_number_of_spectrum_cgmponents
It was pointed out in the introduction to this section that there
are various problems in the determination of the number and the apparent
/suitable for calculation/ shape of component bands in a spectrum.
16
s
Fig. 8
Experimental instrument function
In a favourable case the number of component bands can be determined
by simple inspection, but if this is not possible, it is necessary to use
another method for the determination.
The derivative method is employed to determine the x-coordinates
of the peaks and inflections. For this type of calculation the SavitzkyGolay method [з] was used. In our experience the derivative method does
not give acceptable results if the fine structure is blurred or masked by
noise.
The most elegant method is the Fourier transform, although this
can only be used when the spectrum is free from noise. Assuming that the
shapes of the component bands can be approximated by a Gaussian distribu
tion function, it is valid for
fk(v 'X) = A k(ak /(°k - A))
If
X
exp["(v‘mk)2/(2ok"X)]
is suitable chosen, the different components are separated and thus
the number of peaks /п/ can be counted and then the positions of the bands
/m^ / read off. The variance a2 can be evaluated from X and the bandwidth
Fig.
Experimental /2/ and genuine /1/ spectrum of uranyl-tris-carbonate. The experimental
instrument function is used for the deconvolution.
18
associated with the peak
/dk / from the formula
a
2
2
« dk + \
while the amplitude
can be calculated from the intensity /1к /
k-th peak and the bandwidth d as
\
■ \
of the
dk(dk * O ' 1
According to definition, the spectrum is
S(v) = [ fk (v,X)
к K
An example for that can be seen in Fig. 10 where the 0.0-bands of the
fluorescence spectrum of uranyl-tris-carbonate are analysed. In this case
the convolute function is almost as broad as the narrowest band in the
spectrum.
Fig. 10
Analysis of the component numbers of the 0.O-transition
of uranyl-tris-carbonate fluorescence spectrum by Fourier
transform.
An effective method proved to be the cyclically applied iterative
deconvolution using a Gaussian band of narrow shape for the genuine spectrum.
Fig. 11. shows the results of a calculation of this type. This method does
NUMBER
Fig. 11
OF
SMOOTHING
CYCLES
Determination of the number of spectrum components by cyclical
deconvolution.
20
not give better results than the Fourier transform, but it imposes fewer
conditions of application .
An interesting method for determining the number of components is
the so-called successive elimination method. In this case it is supposed
that the low-frequency side of the spectrum is shaped by only one component.
The parameters of a Gaussian band are calculated from the points of the
low frequency side by the least squares method and the original spectrum
is reduced by that component. The•procedure is repeated until the residual
spectrum is small enough. This method requires repeated visual control.
с/ Fitting_grocedure
This procedure serves to fit the experimental spectrum S(v) with
a calculated spectrum f(v) defined by a parameter set to minimize the
sum of square deviations
minimum
where
is the weighting factor due to the i-th point. Both spectra
consist of
N
corresponding points.
The algorithm of the spectrum calculation is selected by the user.
The only conditions are:
- the spectrum must be the sum of the positive component bands;
- the component bands can have only a single peak.
In our computation work the following types of component bands are used:
Gaussian components
where
A
is the amplitude,
vq
is the position of the peak and
the variance.
Cauchy components
where 2b is the half-bandwidth.
Voigt components /Gauss-Cauchy product function/
a
2
is
21
f(vi)
A
Sum of Gaussian and Cauchy functions
- v.
where AQ and
respectively.
Ac
+ A,
are the amplitudes of the Gaussian and Cauchy components,
The Voigt and sum function gives information about the contributions
of the Gauss and Cauchy functions to the experimental spectrum.
There are 3 parameters in the Gauss and Cauchy functions, 4 in the
Voigt function and 5 in the Sum function /only the parameter
vq
is common
to both functions/.
The calculated spectrum
f (v ^,
a)
is produced by summarizing the
adequate points of the component bands:
F( V
a)= ^
f SL vi' a
where f^v^, a ) represents the values of the component bands at
frequency
and with a given parameter set. The forthcoming task is to find the minimum
with respect to
a:
N
Q(ä) =
I
г
wi [ s ( v i ) - F ( v ± , 5 )
The condition for the minimum is
-I
where
■J
1"i[s(',l) - F( V
5)] ЭГ^»а'
* 4j(;) - °
j = l,2,...,m.
Then the transcendent equation system
5(5) = о
is solved using the Newton-Raphson iteration
method. If the k-th approximation
22
of the solution vector a is denoted by
be obtained by the iteration formula:
, the next approximation can
- (k+l) _ -(k)
(k)a
= a
+ t
p
where
i(S)
1da
and
j
is the inverse of the derivative of the
function with respect to
q
vector-vector
a.
In the case of more than 20 parameters it is convenient to analyse
the spectrum in parts, and sometimes preferable to use Jones's moving
subspace method. [7]
The system of normal equations becomes "ill conditioned" if the
individual components are too near to one another and will thus cause
convergence difficulties in the fitting procedure. Our program operates in
the ill-conditioned case according to the steepest descent method [5]. This
convergences in all cases although its rate of convergence may sometimes
be exceedingly slow.
Fig. 12 demonstrate a least squares fit applied to the 0.0 - transi
tions of the fluorescence spectrum of uranyl-tris-carbonate. The spectrum
was drawn by line printer.
Fig. 12
Least square fit of the O.O-transition band of the uranyltris-carbonate ammonium salt fluorescence spectrum.
23
7/ Storage of spectra
During an experimental investigation a number of spectra are obtained.
From the storage standpoint it is important to reject all unnecessary spectra
containing redundant information. This category includes those spectra wich
can be calculated from other spectra by a well defined mathematical process:
spectra wich cannot be determined mathematically are called "core spectra".
The core spectra are selected from the measurements and stored.
The core spectra are subjected to data selection for the reversible
elimination of redundant data, as described in Chapter 4.
The organization of spectrum storage is not without its own problem,
because there is no internationally accepted code system for chemical compounds.
In our work we use simplified alphanumeric code system.
The standard spectra /see section 5/ are stored separately on magnetic
tape. The blockhead of each standard spectrum contains the name of the compound
and its most characteristic spectral parameters.
The core spectra are also stored on magnetic tape, but in this case
the names of compounds form the basis of organization, and the blockhead
gives information about the type of spectra stored.
8/ Usage of the spectrum library
The analytical standard spectra can be addressed sequentially in
order of storing. The computer first compares the stored and analysed spectra
with respect to their most important features /accumulated in the blockhead/.
Once the most probable spectrum has been selected from the library, it is
compared point to point with the normalized, investigated spectra. In many
cases the data of the experimental spectra are not in agreement with the
stored spectra and thus analysis fails. It has been found that logical methods
of spectrum analysis by which
the structural elements of the investigated
compounds are estimated from the experimental spectrum using the computer
version of the classical spectrum evaluation procedure /see section 5/ are
more reliable.
The main adventage of having core-library is that it is capable of
storing a very large number of spectra in a form easily accessible by computer.
Detailed abstracts of the spectra stored for each compound, specifying the
structural parameters,the number of spectra stored, the experimental condi
tions and the range of the calculable spectra can be made available by line
printer for the investigator in printed form. The algorithms needed for the
calculations are stored on magnetic tape and can be actuated by software.
24
At present only Lagrange's interpolation is used as an algorithm for calcula
tion of the required spectra, multiplication of spectrum point, etc. The
degree of the polynomial can be chosen freely. The spectra are mobilized
in a suitable form by standard programming procedures.
SUMMARY
In the present report we summarized the experiences and results of
the digital optical spectroscopic work performed in the Central Research
Institute for Physics in Budapest. This work involves the development of
the on-line spectrometers, the computerization of the spectrum evaluation
work and the digital storing of the measured spectra.
In conclusion it can be stated that the digital methods of
spectroscopy are suitable for the fast and precise evaluation of a multitude
of data. The versatility of the instruments permits to obtain collect spectra
under different conditions. It must be emphasized that the mathematical
methods do not free the investigator from the detailed measuring work. They
are valuable tools to obtain distinct and quantitative information from
favourable experiments, and the experiments can be regarded as the primary
source of information.
25
REFERENCES
[l3
I.В. Berlman; Handbook of Fluorescence spectra of Aromatic Molecules
Academic Press New-York, 1965.
[2]
P ,L ■ Richards: J. Opt. Soc. Am. 1964 , 54_, 1474.
[3]
A. Savitzky and M.J.E. Golay; Anal. Chem. 1964, J36, 16 27
[V]
K.S. Seshadri and R.N. Jones; Spectrochim. A. 1963, 19, 1013
[5]
J Szőke, L. Varga and I. Naqypál: Proc. of Coll. Spectr. Interntl.
XIV. Debrecen 1967, page 1905.
[jf)
R.N. Jones and R. Venkataraqhavan; Spectrochim A. 1967. 2ЗА, 925; 941.
[У]
R.N.
Jones and R.P. Young: Natl. Res. Counc. of Canada Bull.
13 page
1969.
C l .
Kiadja a Központi Fizikai Kutató Intézet
Felelős kiadó: Szabó Elek, a KFKI Kémiai
Tudományos Tanácsának elnöke
Szakmai lektor: Kötél Gyula
Nyelvi lektor: T. Wilkinson
Példányszám: 310
Törzsszám: 72-6290
Készült a KFKI sokszorosító üzemében,
Budapest, 1972. március hó