UDK 543.428.2:544.171.7	ISSN 1580-2949
Original scientific article/Izvirni znanstveni članek	MTAEC9, 49(3)435(2015)
USING SIMULATED SPECTRA TO TEST THE EFFICIENCY OF SPECTRAL PROCESSING SOFTWARE IN REDUCING THE NOISE IN AUGER ELECTRON SPECTRA
UPORABA SIMULIRANEGA SPEKTRA ZA PREIZKUS UČINKOVITOSTI PROGRAMSKE OPREME PREDELAVE SPEKTRA PRI ZMANJŠANJU ŠUMA SPEKTRA AUGERJEVIH ELEKTRONOV
Besnik Poniku1,2, Igor Belic1, Monika Jenko1
1Institute of Metals and Technology, Lepi pot 11, 1000 Ljubljana, Slovenia 2Jožef Stefan International Postgraduate School, Jamova 39, 1000 Ljubljana, Slovenia besnik.poniku@imt.si
Prejem rokopisa - received: 2015-01-04; sprejem za objavo - accepted for publication: 2015-01-21
doi:10.17222/mit.2015.013
When attempting to automate Auger spectra analyses it becomes necessary to have a deeper knowledge of the constituent elements of the spectra. In order to obtain a reliable analysis, the unavoidable spectral noise must be reduced, thus giving a clearer view to the spectral peaks and the spectral background. Therefore, the necessary step is to analyze the spectral noise and to find a way to evaluate the noise-reduction algorithms. A method in which simulated Auger electron spectra are used for testing the efficiency of noise-reduction routines has been proposed. The performance of noise-reduction procedures on measured spectra cannot be evaluated since the intrinsic noiseless spectra is never available for reference; therefore, the spectra were simulated and the noise-reduction routines were used on the simulated spectra. After the processing, the simulated noiseless spectrum is subtracted from the complete spectrum, leaving the remaining noise for further analysis and a comparison with the exactly known simulated noise. For each spectrum data point the noise ratios are calculated by dividing the remaining noise levels by the initial noise. When plotting the noise ratios for each respective processing route, it was found that most of the noise ratios lie in the interval -1 to +1, indicating an improvement in regard to the initial noise. Such a plot of the noise ratios offers a convenient way for assessing the efficiency of the noise-reduction routine at a glance. Keywords: Auger electron spectroscopy, spectra simulator, spectral noise, noise reduction
Avtomatizacija postopka analize Augerjevih spektrov zahteva dobro poznanje posameznih sestavnih elementov spektra. Zanesljivost avtomatske analize je v prvi vrsti odvisna od tega, v kolikšni meri nam uspe zmanjšati spektru primešan šum, ki sicer zamegli tako spektralne vrhove kot tudi spektralno ozadje. Zato moramo najprej analizirati lastnosti šuma, primešanega spektrom, in poiskati načine za ovrednotenje delovanja orodij, ki šum zmanjšujejo. V članku predlagamo uporabo simulatorja Augerjevih spektrov, ker sicer pri izmerjenih spektrih nikoli ne poznamo oblike primešanega šuma in torej nimamo osnove za dobro ovrednotenje delovanja uporabljenih orodij. Po uporabi orodja za zmanjševanje šuma, ki deluje na simuliranem spektru, odstranimo natančno poznano spektralno ozadje in spektralne vrhove. Tako dobimo preostali šum, ki ga primerjamo z znanim začetnim šumom. V vsaki točki spektra so izračunana razmerja med začetnim in končnim šumom. Razmerja so nato prikazana v grafu in v veliki večini spektralnih točk ležijo v intervalu -1, 1. Tako dobimo vizualno predstavitev delovanja orodij za zmanjševanje šuma, ki omogoča hitro oceno učinkovitosti preizkušanega orodja.
Ključne besede: Augerjeva elektronska spektroskopija, simulator spektra, spektralni šum, zmanjševanje šuma
1 INTRODUCTION	tra.12 Very little can be said about the efficiency of such
procedures in reducing the noise when applying them in Auger electron spectroscopy is a technique often	^	,	••,,,•
measured spectra, because noise in both the input and the
used for the elemental characterization of the surface of
conductive samples.1-8 Apart from a high surface sensiti- output spectra is at unknown levels. The aim of this work vity,9 due to the fact that the primary electron beam can is to show a simple way in which the performance of the be focused down to approximately 10 nm in diameter,10 noise-reduction techniques can be assessed using analyses with very good spatial resolution can also be simulated spectra. Using simulated spectra to assess the performed. This fact makes it possible to analyze efficiency of noise-reduction routines is very appro-features on a nanometer scale on the surface through this priate. This comes about due to the fact that the values technique.	for the different components of the simulated AE spectra
To interpret the measured spectra the measured data
(including the noise) are known before processing, and
have to be manipulated by software for signal processing. This manipulation inevitably leaves its mark on thus any change due to the processing route may be the results obtained.11	found and then compared to the initial preprocessed
Smoothing is one of the methods that are used for the values. purpose of reducing the noise in Auger electron spec-
2 EXPERIMENT
The construction of the simulator for gathering the simulated spectra used for this assessment is described in detail in11. Eor the construction of this simulator a number of measured AE spectra obtained from spring-steel samples were closely inspected. The neural network was used to model the primary background by selecting a number of representative points for the background and including them in the training data set for the neural network. After carefully observing the behavior of the background in the measured spectra an equation was derived, which then would be used for generating various primary backgrounds that would resemble those observed in the measured spectra. After removing the primary background defined in this way the peak base and the peaks remained. The peak base was also modeled in the same way using the neural network, and the removal of the defined peak base left only the characteristic peaks. The peak base and the peaks of various elements were saved in the database. Combining the generated primary backgrounds, on the one hand, and the peak base and characteristic peaks from the database, on the other, produced the simulated spectrum. The generated noise that was then added to such a spectrum was also made to resemble the noise observed in measured spectra. It is important to note that while the components of the simulated spectra such as the background and noise are made to resemble those of the measured spectra, their exact values are simulated and therefore known and stored in the computer (Figure 1).
Through the modeling of the background, which was performed using the neural network, we have found that the AE spectra consist of three main components: the primary background, the peak base, and the peaks (Figure 2).
Erom the set of standard AE spectra that were obtained using COMPRO10, a freely available online spectral database, the peak base and the peaks of elements such as Al, C, Co, Cu, Ee, Au, Ni, O, Si, Ag, Ti, and V
Figure 2: The AE spectra constituent elements: the primary background, the peaks base, and the peaks
Slika 2: Sestavni elementi AE-spektra: primarno ozadje, podlaga spektralnih vrhov in spektralni vrhovi
were extracted (as shown in Figure 3 for the case of iron) and were stored separately.
The AE spectra simulator combines the extracted peak base and the peaks from various standard elements, and it combines them with the randomly defined primary background (Figure 4) to form the complete simulated spectrum without the noise.
At the end of the simulation process the random noise is added and also stored separately for further use. We have ensured that the properties of the simulated noise resemble the properties of the noise in the measured AE spectra.
Other AE spectra simulators can be used for this purpose as well. One such simulator is SESSA (Simulation of Electron Spectra for Surface Analysis). SESSA is
Figure 1: Simulated Auger electron spectrum
Slika 1: Simuliran spekter Augerjeve elektronske spektroskopije
Figure 3: a) The AE spectra peak base and b) spectral peaks
Slika 3: a) Podlaga spektralnih vrhov in b) spektralni vrhovi AE-spek-
tra
Figure 4: The randomly defined primary backgrounds Slika 4: Naključno določena primarna ozadja
intended for facilitating the quantitative interpretation of electron spectra (Auger and XPS spectra), and therefore a lot of attention is paid to the detailed physical phenomena related to the excitation and emission of the Auger electron or photoelectron. The database of SESSA contains the data of many physical parameters needed in quantitative electron spectroscopy (AES and XPS).13 The simulations needed for the purpose discussed in this paper do not require such detailed simulations. The key factor here is that the spectra resemble the real measured ones, and that the values of the different components of the spectra are known before the processing starts. This fact is of utmost importance for the comparison of values of any of the spectral components before and after the processing.
As mentioned in the introduction, smoothing is often used for reducing the noise in Auger electron spectra. Eor spectra measured with a energy step size 1 eV a 5-point averaging window is recommended.14 Since most of the measured spectra that were used when building the simulator were of energy step size 1 eV, the same averaging window was used for processing the spectrum
Figure 5: Processed spectrum from Figure 1 using a 5-point window smoothing in CasaXPS
Slika 5: AE-spekter s slike 1 po uporabljenem glajenju z oknom širine 5 točk (program CasaXPS)
Figure 6: Processed spectrum from Figure 1 using a notch filter in Audacity
Slika 6: AE-spekter s slike 1 po uporabi ozkopasovnega filtra (program Audacity)
shown in Figure 1. The processing was performed using CasaXPS. The resulting spectrum is given in Figure 5.
The same simulated spectrum (Figure 1) was processed by applying a notch filter. The threshold frequency was selected arbitrarily for this case, just for a comparison. This procedure was completed using Audacity, where this procedure is used to reduce the noise in sound files. The resulting spectrum is given in Figure 6.
3 RESULTS AND DISCUSSION
After applying the smoothing procedures to the simulated spectrum, the noiseless simulated signal (^noiseless) was subtracted from the processed spectrum (^processed), thus obtaining the remaining noise after processing (-^remaining):
^remaining ^processed Sn
s	(1)
Such a procedure was used for obtaining the values of the remaining noise for each data point. These values were then compared to the initial simulated noise, ^initial, which is the random noise added to the simulated spectrum, thus obtaining the noise ratios that serve as a measure of the efficiency of the processing software in reducing the noise and bringing the signal closer to the noiseless one:
N ■ = N
-i » rQtir» — -i » rf
,/Ni,
(2)
Figure 7 illustrates the concept behind the use of the noise ratios for this kind of evaluation of the efficiency in noise reduction.
As may be inferred from Figure 7, when manipulating the signal for reducing the noise the obtained signal will have a new value with a different deviation from the target point, the noiseless signal. This new difference from the noiseless signal, the remaining noise, will be smaller than, equal to, or greater than that of the initial noise. Thus, for one specific data point a noise ratio of
B
H
Noiseles Kinetic Fe Simulatec) AE Energy Peaks Background spectrum
50	0 85.521594 85.52159
51	0 83.173313 83.17831
52	0 80.927765 80.92777
53	0 78.764702 78.7647
54	0 76.664264 76.68426
55	0 74.681944 74.68194
56	0 72.753556 72.75356
57	n 7nFI95?1 7nRq591
_ _5S _ _ Ii_69.1032a2_6a.lD32S.
9
JQ.____
r595~	"643"
1596	1644
1597	1645
1598	1646
1599	1647 1800	1848 1601	1649 ifin?	iR5n
Initial Noise -1.29378 0.421041 -0.41938 -0.88762 1.545949 0.384103 -0.69507 -0.83461 _-L1E4a4.
Connplete AE
spectrum 84.22782 83.59935 30.50839 77.87708 76.23021 75.06605 72.05849 70.0606 _67.33834
■ "4"3Ö1336 74.366858 74.432383 74.49791 74.563441 74.628974 74.69451 74 7Rnni9
74"3Di34'
74.36686 74.43238 74.49791 74.56344 74.62897 74.69451 74 7finn5
■-".60"46 72~699B8~ 1.49713 75.86399 -1.13928 73.29311 0.11125 74.60916 -1.35235 73.21109 -1.08518 73.5438 0.224889 74.9194 n4innfi4 75 i7nii
5 Point Smoothed 84.3649 83.225 80.5747 78.6823 77.369 75.3379 72.2577 69.9403
"4"4497 74.1172 74.7135 73.6919 73.6805 73.7001 74.6885 75
Remaining Noise (G-D) -1.1566939 0.04683741 -0.3530651 -0.0824024 0.70473564 0.65595638 -0.4958559 -0.9549098
Ö. "48364 ~3'
-0.249653 0.28111716 -0.3060105 -0.8829409 -0.928374 -0.0060099 n 4^355147
Noise
Ratios (H/E) 0.89404339 0.11088572 0.84188131 0.09283547 0.45565976 1.70776009 0.71339476 1.14414195
■-Ö"09264303 -0.16675773 -0.24675032 -7.24505201 0.65289337 0.85596404 -0.02672388 1 'SPffiFPA
Table 2: Data sheet with the noise ratios from the AE spectrum processed using a notch filter
Tabela 2: Razmerja amplitud {umov pri filtriranju spektra z ozkopa-sovnim filtrom
Figure 7: Concept of noise ratios as a measure of the efficiency in noise reduction
Slika 7: Koncept razmerja {umov kot merilo u~inkovitosti zmanj{e-vanja {uma
less than one means that the processing was successful in reducing the noise, a noise ratio of one means that the noise level is kept the same, and a noise ratio of more than one means that the noise level is actually increased due to the processing of the spectrum. A graphical plot of the noise ratios would serve as a quick assessment at a glance with respect to the success of the noise-reduction routine, as will be shown later on.
By using Equation (1), first the values for the remaining noise were found for each data point, and then by dividing at the respective data points according to Equation (2), the noise ratios were found and recorded in the data sheet, as shown in Table 1 for the spectrum processed in CasaXPS, and in Table 2 for the spectrum processed in Audacity.
By plotting the obtained noise ratios for each data point according to Equation (2), the graph obtained will give, at a glance, an indication of the improvement with regards to the noise. Figure 8 shows such graphs for the two processing routes discussed in this paper.
Table 1: Data sheet with the noise ratios from the 5-point smoothed Auger spectrum
Tabela 1: Razmerja amplitud {umov pri glajenju spektra s 5-to~-kovnim povpre~enjem
D
H
Kinetic Energy
50
51
52
53
54
55
56 '57
Fe
Peak	Simulated
s__Background
O	85,5215939"'
O	83.1783126
O	80,9277651
O	78.7647024
O	76.6842642
O	74.6819436
O	72.7535559
' 'o	70.3952098
O	89,1032824
O	67,3743966
O	65,7054003
Noiseles AE
spectrum ■ 85.52159 83.17831 80.92777 78.7647 76.68426 74.68194 72.75356 70.89521 69.10328 67.3744 65.7054
Initial Noise
-1,294 0,421 -0.419 -0,888 1.5459 0.3841 -0,695 -0,835 -1,165 -1,086 -0,177
Complete AE
spectrum " 84T227B2 83,59935 80.50839 77.87708 78.23021 75.06605 72.05849 70.0606 67,93834 66,28791 65,52819
Audacity SjTiooJ^d "19,85671" 45,33394 64,56315 68 37446 75,32746 73,34311 75,92464 7r43B66 72,53164 68,05102 69.1224
Remaining NoiseJG^P) ■ -65r664886 -37,844375 -16.364618 -10.390246 -1.3568013 -1.3338349 3.1710817 0.5434543 3.4783531 0.6766247 3.4169989
Noise
Ratios (H/E) 50,7543595" -89.882924 39,0213159 11,7057686 -0,8776499 -3,4725938 -4,5622793 -"O", 6511494 -2,9358672 -0.622762 -19.281932
1599
1600 1601 1802
" 1646" ~ 0~74,4'9791Ö5~ 74,49791 ~0~ fi2~ 74,6li"9r6
1647	0 74,5634409 74,56344 -1.352 73,21109
1648	0 74,623974 74,62897 -1.085 73.5438
1649	0 74.6945099 74.69451 0.2249 74.9194
1650	0 74.7600485 74.76005 0.4101 75.17011
'74.49693 -0.00Ö9344 73.67407 -0.8893705 0.6576478 74.11453 -0.5143922 0.47401608
73.63111	-1.0633957 -4.7285443 74.74124 -0.0188118 -0.0453752
As can be seen in Figures 8a and 8b, in both cases most of the points representing the noise ratios occupy the region between -1 and 1, while some of the peaks lie outside these boundaries. If the ratio of the remaining noise to the initial noise is less than 1, this indicates that the new signal after processing is actually closer to the real signal than the one before processing, thus indicating an improvement with respect to the noise. The noise ratios whose values lie outside the [-1,1] interval
Figure 8: Noise ratios from: a) the 5-point smoothed spectrum and b) the spectrum smoothed using a notch filter
Slika 8: Razmerja amplitud {umov pri: a) filtriranju spektra s 5-to~-kovnim povpre~enjem in b) pri filtriranju spektra z ozkopasovnim filtrom
indicate that for those specific data points the processing has actually worsened the situation with respect to the noise. Plotting these noise ratios for all the data points in the spectrum offers a convenient way to see at a glance whether there is an improvement in terms of the noise or not, as well as for comparing different processing techniques for this purpose. Again, the usefulness of the simulated spectra must be stressed in this regard, because no such comparison can be made if the values for the noise are not known at the beginning, as it is in the case of the measured spectra.
4	CONCLUSIONS
Simulated spectra have been used to assess the performance of two noise-reduction techniques. A simple idea of using the noise ratios as a measure of the efficiency of the noise-reduction routines was presented. By applying this idea on a simulated spectrum, which was processed using two different procedures, the values for the noise ratios at each data point were found for the respective procedures. Plotting the noise ratios provided a convenient way to assess, at a glance, the efficiency of the processing route in reducing the noise. The noise ratios for the majority of the data points lie in the [-1,1] interval, indicating an improvement with respect to the noise.
The only way in which the values of the noise ratios can be obtained is if the values of the noise and the noiseless signal are known before and after the processing. Such a condition can be fulfilled if simulated spectra are used in the assessment stage, as shown in this paper.
5	REFERENCES
1	D. R. Baer et al., Challenges in Applying Surface Analysis Methods to Nanoparticles and Nanostructured Materials, Journal of Surface Analysis, 12 (2005) 2, 101-108
2	D. R. Baer et al., Characterization challenges for nanomaterials, Surface and Interface Analysis, 40 (2008), 529-537, doi:10.1002/sia. 2726
3	D. R. Baer, D. J. Gaspar, P. Nachimuthu, S. D. Techane, D. G. Cast-ner, Application of Surface Chemical Analysis Tools for Characterization of Nanoparticles, Analytical and Bioanalytical Chemistry, 396 (2010) 3, 983-1002, doi:10.1007/s00216-009-3360-1
4	A. S. Karakoti et al., Preparation and characterization challenges to understanding environmental and biological impacts of ceria nanoparticles, Surface and Interface Analysis, 44 (2012), 882-889, doi:10.1002/sia.5006
5	D. R. Baer, Application of surface analysis methods to nanomate-rials: summary of ISO/TC 201 technical report: ISO 14187:2011 -surface chemical analysis - characterization of nanomaterials, Surface and Interface Analysis 44 (2012), 1305-1308, doi:10.1002/sia. 4938
6m. p. Seah, Summary of ISO/TC 201 Standard: XXIII, ISO 24236:2005 - Surface chemical analysis - Auger electron spectro-scopy - Repeatability and constancy of intensity scale, Surface and Interface Analysis 39 (2007), 86-88, doi:10.1002/sia.2493
7	M. P. Seah, Summary of ISO/TC 201 Standard XI. ISO 17974: 2002
-	Surface chemical analysis - High-resolution Auger electron spectrometers - Calibration of energy scales for elemental and chemical-state analysis, Surface and Interface Analysis, 35 (2003), 327-328, doi:10.1002/sia.1529
8	M. P. Seah, Summary of ISO/TC 201 Standard XII. ISO 17973:2002
-	Surface chemical analysis - Medium - resolution Auger electron spectrometers - Calibration of energy scales for elemental analysis, Surface and Interface Analysis, 35 (2003), 329-330, doi:10.1002/ sia.1530
9N/A. Auger Electron Spectroscopy (AES): What is AES? http://www.phi.com/techniques/aes.html (accessed November 2014). Physical Electronics, Inc. Chanhassen, MN, 2006 - 2014.
1"VG Scientific, Microlab 310 - F: Operators Manual, VG Scientific, East Sussex, 1997
11 B. Poniku, I. Belie, M. Jenko, The Auger spectra recognition and modeling: Modeling Auger spectra for effective background removal and noise reduction, Lambert Academic Publishing, Saarbrücken 2011, doi:10.13140/2.1.3620.7045
121. S. Gilmore, M. P. Seah, Savitzky and Golay differentiation in AES, Applied Surface Science, 93 (1996) 3, 273-280, doi:10.1016/0169-4332(95)00345-2
13W. S. M. Werner, W. Smekal, C. J. Powell, NIST Database for the Simulation of Electron Spectra for Surface Analysis (SESSA) -User's Guide, Version 1.3, National Institute of Standards and Technology, Maryland, 2011
14 N. Fairley, CasaXPS Manual 2.3.15: Introduction to XPS and AES, Casa Software Ltd., Devon, 2009