Content uploaded by Annick Barlier-Salsi
Author content
All content in this area was uploaded by Annick Barlier-Salsi on Jan 12, 2016
Content may be subject to copyright.
1
Stray light correction on array spectroradiometers for optical
radiation risk assessment in the workplace
A Barlier-Salsi
Institut national de recherche et de sécurité (INRS)
E-mail: annick.barlier-salsi@inrs.fr
Abstract. The European directive 2006/25/EC requires the employer to assess and if
necessary measure the levels of exposure to optical radiation in the workplace. Array
spectroradiometers can measure optical radiation from various types of sources;
however poor stray light rejection affects their accuracy. A stray light correction
matrix, using a tunable laser, was developed at the National Institute of Standards
and Technology (NIST). As tunable lasers are very expensive, the purpose of this
study was to implement this method using only nine low power lasers; other
elements of the correction matrix being completed by interpolation and
extrapolation. The correction efficiency was evaluated by comparing CCD
spectroradiometers with and without correction and a scanning double
monochromator device as reference. Similarly to findings recorded by NIST, these
experiments show that it is possible to reduce the spectral stray light by one or two
orders of magnitude. In terms of workplace risk assessment, this spectral stray light
correction method helps determine exposure levels, with an acceptable degree of
uncertainty, for the majority of workplace situations. The level of uncertainty
depends upon the model of spectroradiometers used; the best results are obtained
with CCD detectors having an enhanced spectral sensitivity in the UV range. Thus
corrected spectroradiometers require a validation against a scanning double
monochromator spectroradiometer before using them for risk assessment in the
workplace.
1. Introduction
Workers may be exposed to optical radiation from specialized [1–3] or general use lamps [4,5]
or industrial processes [6,7]. In all these working situations, optical radiation has the potential
to be harmful. Some guidelines on limits of exposure to incoherent Ultraviolet [8], Visible and
Infrared [9,10] radiation have been published by the International Commission on Non-Ionizing
Radiation Protection (ICNIRP). These Exposure Limit Values (ELVs) [10-11] have been used
in the European directive 2006/25/EC [11], which requires the employer to assess and, if
necessary measure and/or calculate the levels of exposure to optical radiation. ELVs are divided
into seven potential health effects on the spectral region from 180 to 3000 nm.
2
Due to the variety of sources and the wide range of the spectrum, there are no accurate, fast,
simple, and affordable methods for measuring risk in the workplace. Radiometers can rapidly
measure an entire wavelength region [12]; nevertheless to avoid errors, the radiometer should be
calibrated with a standard source similar to the tested source [13,14]. This is unconceivable, as
radiometers are intended to measure various types of sources in the workplace. Moreover risk
assessment covering the seven spectral regions involves using several filtered detectors and the
attempts to produce a filter precisely matching the various hazard weighting functions has met
with a limited success. Since these filters are related to the ELVs, modifying in the ELVs means
changing the device and, furthermore, old measurements would become unusable. This problem
currently occurs with the new ELVs [10] proposed by ICNIRP in the Visible and near-IR wave
range.
Spectroradiometric measurements, on the other hand, exclusively characterizing the source are
independent of ELVs. Consequently, on the wavelength range from 200 to 1000 nm, CCD array
spectroradiometers appear to be the more suitable instrument for this application, being
portable, affordable and fast compared to scanning double monochromator devices. Although
having many advantages, their accuracy is limited by the presence of spectral stray light
originating from inter-reflections on the optical elements of the instrument, higher order
diffraction, fluorescence, etc. In the best array spectroradiometer, the level of stray light
measured at any pixel, is around 10-5 for the measurement of a monochromatic source or 10-3 for
the measurement of a broadband source [15]. Due to the cumulative effects, the stray light
observed with a broadband source such as incandescent lamps may cause significant errors,
especially in the UV range where the signal is weak compared to the visible range.
Several stray light rejection methods have been described in scientific literature. Optical
methods can be applied to suppress the stray light, blocking the unwanted light by using an
optical filter wheel [16,17]. This technology is commonly used in scanning spectroradiometers.
Various bandpass filters are available in the UVA, visible or near infrared region. However, a
suitable shortpass filter which transmits UV radiation at wavelength shorter than 250 nm and
rejecting signal at longer wavelength, to our knowledge, does not exist. Numerical methods to
reduce stray light errors have also been proposed [15,18–20]. Yianttila et al [18] proposed a
scatter light correction algorithm where the slit scattering function is determined by measuring a
laser line at 633nm. Applying this method allows to obtain adequate accuracy not only for
sunbed UV radiation measurements [18] but also for measurements of phototherapy and
photobiology sources [21]. Nevertheless this method has only been experimented for UV
measurements. An another numerical approach developed by Zong et al. [15] at NIST
demonstrates that the stray light errors can be reduced by one or two orders of magnitude. The
spectrometer’s response, using a tunable laser, is characterized for a set of laser lines that cover
the instrument’s spectral range to derive a correction matrix. The corrected signal is obtained by
a simple matrix multiplication. This method based on a tunable laser is also used by
Physikalisch Technische Bundesanstal (PTB) [22–24] to characterize the spectrometer stray
light behaviour within the wavelength range of 200 to 1100 nm. This simple solution, easily
implemented in acquisition software, has several advantages, with the exception of the high cost
3
of the tunable laser. For this reason, the method has been simplified [25] for solar applications
or adapted [26] using a double monochromator to produce monochromatic lines.
Following the same approach, a method where only a few laser lines are used, is proposed in
this study. The correction matrix is based on monochromatic lines produced by nine low power
lasers in the 266-980 nm wavelength range; other elements of the correction matrix are
completed by interpolation and extrapolation. The experiment consists of measuring several
lamps with two commercial CCD array spectroradiometers and a scanning double
monochromator device as reference. The correction efficiency was assessed by comparing CCD
spectroradiometers with and without correction and the reference instrument on the basis of
relative spectral measurements and quantities Eeff, EUVA, EB [11] required to assess exposure
levels.
2. Methods
2.1. Stray light correction matrix
Principle. The stray light correction method by Zong et al. [15] is clearly explained in 2.1.1.
paper Ref. [15] and illustrated in figure 1. A spectrometer with n pixels array detector is fully
characterized with the help of two indices i and j referring to output and input signal,
respectively; i is the pixel index of the array detector and j relates to the wavelength of the
excitation line, expressed as pixel space. The method is based on measuring the spectrometer
signal on the full spectral range for different monochromatic lines. Response of each pixel i to
the monochromatic excitation at wavelength
j is called spectral line-spread function (LSF),
denoted fLSF i,j. LSF has two parts: the narrow peak region around pixeli=j, called in-band
response IB, representing the instrument’s bandpass, and the low signal measured at other pixels
arising from spectral stray light.
Figure 1. Illustration of the LSF determination with an n pixel array
detector.
4
The stray light distribution function (SDF) is obtained by normalizing fLSF i,J to the sum of the
relative signal in the in-band region IB and setting values within the IB to zero. The normalized
signal di,J, at each pixel i due to the monochromatic excitation line J is given by:
(1)
SDF vectors determined at each monochromatic excitation for wavelength
j enable the SDF
matrix D (n x n) to be built.
When a broadband source illuminates the n pixels of the array detector, the total stray light
signal at a given pixel i, denoted ysl,i is the sum of stray light contributions from each IB signal
yIB,j(j=1..n) at wavelength
j(j=1..n) of the broadband source spectra. Thus, the measured signal ymeas,i
at pixel i includes the true IB signal yIB,i plus the stray light signal ysl,i:
(2)
Expressed in terms of matrix, equation 2 gives:
(3)
where column vectors YIB and Ymeas represent the in-band signal and the measured signal
respectively, and I the identity matrix. Then the true spectra YIB of the broadband source is
obtained as follow: (4)
where C is the stray light correction matrix.
Choosing the IB width. Unlike the scanning monochromator device, CCD array 2.1.2.
spectroradiometers do not have a constant bandwidth across the spectral range. The bandwidth
generally increases at higher wavelengths, which means it is not easy to define the width of IB.
For this reason, the effects of the chosen width of IB on the correction matrix and consequently
on the corrected signal, were studied for several configurations. On the one hand, IB was
constant on the full spectral range and defined as IB=2.4xFWHM or IB=3xFWHM; on the other
hand, IB varied according to the wavelength so that IB matches all the values around the peak
region for which the signal was higher than 1% of the peak value [27] and defined as
IB=values>1%peak.
Monochromatic sources. Nine low power lasers in the wavelength range 266-980 nm 2.1.3.
were used as monochromatic sources to create the correction matrix. The monochromatic line at
IBidIBi
f
f
dJi
IBiJiLSF
JiLSF
Ji
for 0or for ,
,
,
,
n
jjIBjiiIBisliIBimeas ydyyyy 1,,,,,, )(
IBIBIBslIBmeas YDIDYYYYY
measmeasIB CYYDIY 1
5
wavelength
j =266 ±1nm required a Q-switched pulsed laser with 3mW of power; other
sources were Continuous Wave (CW) lasers with 1mW of power for wavelength
j = 405, 447,
473, 532, 589, 633, 808, and 980±5 nm.
Process to generate the correction matrix. The initial matrix consists of n rows depending 2.1.4.
of the number of usable pixels on the detector, and 9 columns. Each column J contains the
matching measured spectral line-spread function fLSFmeas i,J (i=1…n); The correction matrix C is
obtained from the initial matrix as follow:
Normalizing fLSFmeas i,J by its peak value to obtain fLSF i,J ,
Interpolating the columns between the known columns and extrapolating columns
outside the known columns to generate an n x n LSF matrix. Two methods of
interpolation/ extrapolation were experimented: cubic spline and linear. Interpolation
and extrapolation were performed along the matrix diagonals. Because extrapolation
becomes impossible where there is only one point left, which is the case in the both
high right and low left quarters of the matrix, the last calculated diagonal was copied to
complete the remaining diagonal.
Normalizing the n x n LSF matrix according to equation 1 and adding identity matrix.
Thus, [I+D] matrix has the following form :
(5)
Inverting [I+D] matrix to obtain the correction matrix C.
Double correction. In order to improve the correction of the signal, a double correction 2.1.5.
was tested. A first correction matrix C1 based on the 9 lasers is created as described in the sub-
section 2.1.4. This correction matrix C1 is applied to correct the signal from the 9 lasers used to
create it. The corrected signal from the 9 lasers is then used to generate a second correction
matrix C2. The final signal from a source can be obtained by multiplying the original signal by
the matrix product C1C2.
2.2. Signal measurement procedure
100.... 010.... 001.... .
.
.
.
.
.
..
.
.
..
.
.
.
.
.
...1.. .
.
.
.
.
.
..
.
.
..
.
.
.
.
.
.....100 ....010 ....001
,3,2,1,
,13,12,11,1
,23,22,21,2
,1,2,3,2,1,
,31,32,3,3
,21,22,2,2
,11,12,1,1
jnnnn
jnnnn
jnnnn
nininiiii
nnnj
nnnj
nnnj
dddd dddd dddd
dddddd
dddd dddd dddd
DI
6
A small signal, depending on the integration time, is generated by the CCD array detector even
when no radiation falls on the CCD element. This dark current increases with the integration
time and depends of the characteristics of each pixel [18][28]. Therefore, whenever possible,
integration time was chosen to be shorter than 1s [28]. Then, after the reading of raw data Sigi
(counts), the dark current dci measured with the same integration time is subtracted. The
measured signal ymeas,i expressed in counts.s-1 is determined by normalizing it by the integration
time. To enhance the dynamic range, measurements are carried out in two stages. The first
measurements are done with a suitable integration time t1 to acquire the spectrum in the region
of its peak value. Second measurements are done to acquire the spectrum outside this region and
a higher integration time t2 is used to improve signal-to-noise. During the second measurement
the instrument then saturates in the region of its peak value, but this region is disregarded and
replaced with the data from the first measurement to obtain the final spectrum; this technique is
called bracketing. Applying this technique is particularly important when the correction matrix
is generated, because the signal-to-noise ratio is relatively weak outside the peak region. Figure
2 illustrates the efficiency of the bracketing technique whereby the signal is more smooth and
lower compared to the original signal. This technique has also been used during measurements
of actual sources and during calibration.
2.3. Filters
As mentioned by Zong et al. [15], equation 4 does not take into account the stray light δsl arising
from signals outside the spectral range of the spectroradiometer. The level of δsl depends on the
measurable range of the instrument compared to the spectral band response of the CCD. Zong et
al. [15] proposed to block this out-of-range stray light by filtering these radiation that is using a
shortpass filter. Up to now, shortpass filters transmitting radiation at wavelengths within 200 -
1100 nm and rejecting radiation at wavelengths longer than 1100 nm are not available. Except
for interferential filters, spectral characteristics of filters are rarely available for wavelengths
Figure 2. Signal from a laser measured with and without bracketing
technique.
7
cfor /)( , cfor /)(
tdcSigytcSigFSigy iimeasiimeas i
i
shorter than 250nm. Thus effects of δsl on the signal measured in the UV spectral range was
studied using a longpass filter with a cut-on wavelength
c=1100nm as a shutter. Consequently
the measured signal SigF1100 with the filter F1100 includes the dark current plus the stray light
δsl and the signal measured at each pixel or wavelength is given by:
(6)
Two other longpass filters F425 and F800 help identify which spectral ranges induce constant
stray light in the UV range.
Note that filters have an optical density of 4, hence they can be used as a shutter only if the
highest signal over the range of measurement of SigF
ci, with the same integration time and in
the absence of the longpass filter, would be lower than 10,000 which is the case with UV
radiation from a tungsten halogen lamp.
2.4. Experiment
Spectroradiometers. The two AvaSpec (Avantes, Netherlands) CCD array 2.4.1.
spectroradiometers used for the experiment cover the wavelength range 200-1100 nm. Both
have a 300 lines/mm grating, order sorting filters and a 16-bit AD convertor. The input optics
consists of a cosine corrected diffuser coupled to the spectroradiometers using optical fibre.
Other individual technical features are described in table 1. A scanning OL 750
spectroradiometer (Optronics Labs, USA) with an integrating sphere for input optics, a double
monochromator, order filters and two detectors: photomultipliers and silicon, was used as
reference. Table 1. Technical features of tested spectroradiometers.
Product
Detector
Integration time
FWHM
2048x14
Back-thinned CCD array, 2048x14 pixels
2.2 ms – 10 min
2.4 nm
2048TEC
CCD linear array, 2048 pixels
1.1 ms – 10 min
4.3 nm
Spectral irradiance calibration. CCD spectroradiometers and OL 750 reference was 2.4.2.
calibrated using two standard sources of spectral irradiance OL 752-12 (Optronics Labs, USA)
and OL 752-10 (Optronics Labs, USA). The 30W deuterium lamp (OL752-12) provided the
spectral irradiance within the wavelength range of 200 nm to 400 nm and the 200 W tungsten
halogen lamp (OL 752-10) the spectral irradiance within the wavelength range of 250 nm to
1100 nm. Spectral calibration factors (W.m-2.nm-1 / Counts.s-1) are calculated as the ratio of the
spectral irradiance of standard lamps to the signal measured at each wavelength.
8
Evaluation of correction efficiency. Risk assessment in the UV-Visible range according to 2.4.3.
ELV from the European directive 2006/25/EC [11] requires the calculation of the irradiance
EUVA and effective irradiances Eeff and EB defined as follow:
(7)
where S(
) and B(
) are the spectral weighting function taking into account UV radiation effects
on eye and skin and blue light radiation effects on eyes, respectively.
Several lamps with various features were measured using CCD spectroradiometers and the OL
750 reference; the purpose was, on the one hand, to compare spectral distributions, and on the
other hand, values of Eeff, EUVA and EB, resulting from CCD spectroradiometers to the OL 750
reference.
3. Results
3.1. Correction matrix
The effect of separately changing the width of the IB region as described in sub-section 2.1.2.
and the interpolation/extrapolation modes used to generate the correction matrix is illustrated in
figure 3. Different correction matrixes of the 2048 x14 spectroradiometer were applied to a
hypothetical source for which the signal was constant and equal to one at each wavelength.
As shown in figure 3, there were no significant differences between the cubic interpolation
curve (thick solid yellow line) and linear interpolation curve (thin solid red line) when IB=
values>1% peak. Similar results between cubic and linear interpolation methods were obtained
when IB=2,4xFWHM. Conversely, the levels of corrected signals with the three configurations
of IB widths are significantly different within the near infrared region. As the FWHM is wider
in the near infrared region than in other spectral ranges the level of the corrected signal
decreases when the value fixed for IB decreases. This simulation with a constant spectrum in
.)(
400
180
SEE nm
nm
eff
.
400
315
nm
nm
UVA EE
.)(
700
300
BEE nm
nm
B
Figure 3. Effect of separately changing the width of the IB region and the
interpolation/extrapolation modes used to generate the correction matrix.
9
particular highlights the correction matrix effects near to the peak region. Due to the lack of
constant bandwidth across the spectral range, the adaptive method with IB= values>1% peak
theoretically gives better results than the fixed IB values.
3.2. Calibration factors
Experiments with longpass filters. Before applying the stray light correction, both 3.2.1.
2048x14 and 2048 TEC calibrations were normally carried out for both deuterium and tungsten
standard lamps as described at §2.4.2 and also by using longpass filters F1100, F800 and F425
for the tungsten standard lamp as in (6). As the deuterium lamp emits no radiation in the visible
and near IR region, the experiment with filters was only applied to the tungsten standard lamp.
Matching calibration curves were plotted on figure 4 between 250 and 400 nm where the spectra
of the deuterium and tungsten lamps overlap. The calibration curves without stray light
correction from the tungsten lamp fall clearly below that of the deuterium lamp (thick solid
lines). Both 2048x14 and 2048 TEC calibration curves using F1100 filter (dotted-dashed line)
and those without filter (thick solid lines) are alike, confirming that out-of-range stray light δsl
for wavelengths greater than 1100 nm is negligible. 2048TEC’s behaviour is nearly the same
with F800. On the other hand, the 2048x14 calibration curve with F800 is slightly higher than
the curve without filter proving that a part of the stray light in the UV range arises from near IR.
Dotted curves matching both instrument’s calibration with F425 for the tungsten lamp are very
close to the calibration curves for the deuterium lamp; thus the stray light on the measured
signal in the UV spectral range mainly originates from visible spectral range.
Figure 4. 2048x14 (a) and 2048TEC (b) calibration factors obtained for deuterium and
tungsten standard lamps without stray light correction but using longpass filters with the
tungsten standard lamp as in (6).
10
Stray light corrected calibration factors. Both 2048x14 and 2048 TEC calibrations were 3.2.2.
carried out for each one of the standard lamps as described in sub-section 2.4.2. The signal from
the instruments had been stray light corrected beforehand by using the correction matrix as in
equation (4). Like figure 4, figure 5 illustrates these stray light corrected calibration curves from
250 to 400 nm.
The close match of the two curves in figure 5a highlights the fact that the correction matrix
enhances the 2048x14 signal from the tungsten lamp between 260 to 400 nm. Nevertheless, the
poor signal (weak signal-to-noise ratio) from the tungsten lamp below 260 nm affects the
corrected calibration factors.
As the level of stray light from 2048TEC is higher than that of 2048x14 (see tungsten
uncorrected ‘sig-dc’ in figure 4), the correction matrix can reduce only a part of the 2048TEC’s
stray light. Consequently the best results (dotted line) for this instrument were obtained by a
combination of the correction matrix C that removes the stray light component near to the peaks
and the F425 filter by subtracting the constant component of the stray light arising from visible
range. Comparing figure 4b and figure 5b the improvement offered by the combination
correction matrix-F425 against the single F425 is about 11% within the wavelength range 250-
350 nm. Note that in this configuration, the matrix is only calculated and applied between 200
and 400 nm.
The concatenation of the deuterium curve and the tungsten curve provides the final calibration
curve of the spectroradiometers from 200 to 1100 nm used in the study follow-up. To reduce the
stray light error the deuterium curve was used from 200 to 380 nm and tungsten lamp beyond
380 nm.
3.3. Spectral distribution
Figure 5. Stray light corrected calibration curves from both 2048x14 (a) and 2048TEC (b)
for deuterium and tungsten standard lamps.
11
Figure 6 shows examples of comparison of the OL 750 (reference spectroradiometer) and both
2048x14 and 2048TEC spectra for a halogen spotlight and a LED. Each spectral irradiance
curve was normalized by its peak value for convenience.
Figure 6. Corrected and uncorrected spectra from a 1500 W halogen spotlight without the
protective glass (a) and a LED (b).
Figure 7. Corrected and uncorrected spectra from a 1500 W halogen spotlight without (a)
and with (b) the protective glass.
* One time corrected (Sig-dc) based on a correction matrix generated without bracketing.
12
As well as results observed during the calibration, the stray light level from 2048TEC is higher
than that of 2048x14. There is also stray light for the 2048TEC in the near IR region where a
LED is measured (figure 6b). Applying the correction matrix reduces the stray light from the
2048x14 below the minimum detectable level where both spotlights without the protective glass
(figure 6a and 7a) and LED (figure 6b) were measured and can reduce that of 2048TEC by 1 or
2 orders of magnitude in the UV and IR regions.
In figure 7b one time correction of the signal from 2048x14 is insufficient to suppress the stray
light within the 270-320 nm region where the spotlight is measured with the protective glass.
Nevertheless, applying a two time correction matrix provides a spectrum that follows OL 750
closely. Note that one count level of a 16 bit spectroradiometer corresponds to a signal level of
1.5x10-5 which accounts for the minimal sensitivity of these spectroradiometers. However, the
dynamic range is improved by using bracketing; the impact of bracketing on the results is
developed in the sub-section 3.5.
3.4. Risk assessment
Quantities Eeff, EUVA and EB were calculated in compliance with equations 7 on the base of both
2048x14 and 2048TEC corrected and uncorrected spectra. Test lamps were chosen according to
their spectra features. The LED, the spotlight and the infrared lamp have continuous spectra
while the xenon and the mercury lamps have line spectra. The LED emits only in the visible
region, the halogen spotlight emits in all regions and the peak of the emission lies in the near
infrared region. Adding a protective glass in front of the spotlight largely reduces the UV
radiation below 300 nm. The mercury lamp yields lines in the UV region and the xenon lamp
both in the UV and the visible regions. Comparison of results from 2048x14 or 2048TEC with
OL 750, are given in tables 2 and 3, respectively.
13
Table 2. Comparison of Eeff, EUVA and EB calculated on the basis of 2048x14 corrected and
uncorrected spectra against OL 750.
OL 750
2048x14
Differences between 2048x14 and OL 750 (%)
Source
Quantities
Values (W m-2)
Uncorrected
Corrected 1x
Corrected 2x
LED
Eeff
< 10-4 a
4800
0
0
EUVA
< 3.5x10-2 a
0
0
0
EB
1.92
-4
-3
-3
Spotlight without
protective glass
Eeff
2.1 x 10-2
424
14 (-99)b
-10
EUVA
0.89
11
0 (-17)b
-1
EB
3.38
-4
-5 (-6)b
-5
Spotlight with
protective glass
Eeff
< 10-4 a
75900
1700 (2300)c
0
EUVA
0.56
16
0
-2
EB
2.98
-4
-4
-4
Infrared lamp
Eeff
< 10-4 a
2570
0
0
EUVA
< 3.5x10-2 a
0
0
0
EB
1.9 x 10-3
62
-24
-24
Xenon lamp
Eeff
0.32
16
9
9
EUVA
1.40
4
3
3
EB
1.60
-5
-4
-4
Mercury spectral
lamp
Eeff
5.5 x 10-2
0
-4
-4
EUVA
0.12
-6
-9
-8
EB
8.1 x 10-2
-8
-6
-6
a EN14255-1 [29] and EN 14255-2 [30] standards relating to the measurement and assessment of personal
exposure to incoherent radiation in the workplace, require minimal instrument sensitivity set at 1/10 of
the applied ELV. As the ELV is expressed in terms of exposure, this minimal value is the ratio of the
ELV to the maximal exposure duration. Thus Eeff min=10-4 W.m-2, EUVA min=3.5x10-2 W.m-2and EB min=10-
3 W.m-2.
b Corrected spectrum on the basis of a correction matrix generated without bracketing
c Source measurement without bracketing
As shown in table 2, the percentage difference between the uncorrected 2048x14 and the OL
750 Eeff varies from 0 to 75900%; results from the uncorrected 2048TEC Eeff being in the same
order (table 3). Significant differences occur when the low level spectral component of
broadband sources as LED, spotlight or infrared lamps are measured, confirming what has been
recorded in another works [20]. The infrared lamp that emits only in the near IR waveband
generates the same effects on the uncorrected 2048x14 Eb (62%); since CCD sensitivity is better
in the visible range than in UV, the level of stray light is lower. On the other hand, the
percentage difference between the uncorrected 2048x14 and OL 750 Eeff, EUVA and Eb from
xenon and mercury lamps falls below 16%.
14
Table 3. Comparison of Eeff, EUVA and EB calculated on the basis of 2048TEC corrected and
uncorrected spectra with OL 750.
OL 750
2048TEC
Differences between 2048TEC and OL 750 (%)
Source
Quantities
Values
(W m-2)
Uncorrected
Corrected 1x
Corrected 2x
Corrected 1x
(Sig-F425)
LED
Eeff
< 10-4
45900
1400
690
EUVA
< 3.5 10-2
14
0
0
EB
1.92
-5
-3
-3
Spotlight
without
protective glass
Eeff
2.2x10-2
1627
200
150
50
EUVA
0.93
29
5
3
2
EB
3.50
-3
-4
-4
Applying the correction matrix once on the 2048x14 is sufficient to reduce the percentage
difference compared to OL 750 to a level within 0 to 24% for all tested lamps, except the
spotlight with the protective glass. However, for this spotlight the difference can be set to 0 by
using a two-time correction. Table 3 shows that even a two-time correction of the signal from
2048TEC is insufficient to obtain results close to that of the OL750. Nevertheless, a
combination of applying the one-time correction matrix and using a filter F425 for the spotlight
measurement reduces the difference from 1627 % to 50%.
3.5. The effect of bracketing
As describe in sub-section 2.2, the bracketing technique was used during the measurements of
laser’s signal to generate the correction matrix. To assess the bracketing impact on the results,
the spectrum of the spotlight without the protective glass calculated with a correction matrix
generated without bracketing is illustrated at figure 7a. The observed effect is an excessive
correction of the stray light which leads to suppress any signal below 300 nm. Consequently the
risk in the UV region is largely underestimated: the percentage difference in Eeff between the
2048x14 and the OL 750 is ~ -100% (table 2). Note that the same measurements carried out
with the spotlight with protective glass gives Eeff < 10-4W.m-2 as the OL 750 and in this case
the excessive correction cannot be identified. This experiment also shows that a comparison
between the corrected spectrum and the reference spectrum is essential to clearly identify an
excessive correction and correctly evaluate the efficiency of a numerical correction of the stray
light. The effect of the bracketing technique used during the measurement of the sources has
also been studied. As shown in table 2, a measurement of the spotlight with the protective glass
performed without bracketing produces a noisy signal in the UV region which results in an
increase of the percentage difference in Eeff.
4. Discussion
As mentioned by Zong et al.[15], the stray light correction matrix is largely independent of the
bandwidth of the monochromatic radiation because the LSF is normalized by the sum of the
15
signals within IB band. For this reason, simple lasers instead of tunable lasers could be used in
this study. Similar to results recorded by NIST [15] and PTB [22–24] based on about 80 laser
lines, experiments carried out in this study show that it is possible to reduce the spectral stray
light by one or two orders magnitude by using only nine simple lasers. Because the corrected
spectrum has not been compared with a reference spectrum in studies [15, 22–24] and also
because the results are dependent of the model of spectroradiometers, as shown in this study, it
is not possible to assess the impact of reducing the number of lasers from 80 to 9 on the
measurement uncertainty. The method limitations can be tested by further reducing the number
of lasers but the selection of the suitable wavelengths of the remaining lasers requires many
configurations to be tested, which is time-consuming. For this reason, a single experiment was
carried out using only three lasers (266, 589 and 980 nm) to generate the correction matrix.
When applying this correction matrix to measurements of the spotlight without the protective
glass, a shift of the beginning of the spectrum from 250 nm to 300 nm occurs, showing an
excessive correction in the UV range. The percentage difference in Eeff between the 2048x14
and the OL 750 is -88%.
Based on a single laser, the correction method developed by Ylianttila [18] appears more simple
and cheaper. It provides an adequate accuracy for sunbed UV radiation [18] measurements and
for clinical ultraviolet dosimetry [21]. Nevertheless these specific applications need the
spectrum for wavelengths above 280 nm to be obtained instead of 200 nm, as required by the
European directive 2006/25/EC [11]. Therefore this method might not be efficient for
wavelengths below 280 nm. Furthermore extending the algorithm of Ylianttila [18] beyond 400
nm assumes that CCD array spectroradiometers have a constant bandwidth across the spectral
range. However,this is not the case. In figure 3, the simulation with a constant spectrum shows
that the levels of the corrected signals with IB=2.4xFWHM and IB=value>1% are significantly
different within the near infrared region; the percentage difference between the two corrected
signals at 980 nm is about 10%. Even if the UV region suffers from stray light more than other
regions, results of the infrared lamp without correction (table2) show that Eb is also affected by
the stray light with a measurement uncertainty of 62% compared to the OL 750.
Although this experiment shows that same results are obtained by using only nine simple lasers
instead of tunable lasers, a key question is whether this stray light decrease is sufficient when
spectroradiometers are intended to assess risk in the workplace. EN14255-1 [29] and EN 14255-
2 [30] standards specify that uncertainty shall not exceed 30% for measurements where the
results are to be compared with exposure limit values. Corrected quantities EUVA and EB satisfy
this condition; percentage difference between both corrected spectroradiometers (2048x14 and
2048TEC) and the OL 750 do not exceed 24%. Mixed results are obtained regarding Eeff,
depending on the instrument; 2048x14 provides better results than 2048TEC. Low level of
residual stray light in the wavelength band from 220 to 300 nm or in the wavelength calibration
[12] results in large errors in the quantification of Eeff due to the spectral weighting function
S(
), being greatest in this region. Moreover, the required minimal instrument sensitivity to
assess Eeff is lower by one or two orders of magnitude compared to EB and EUVA, respectively,
16
while CCD detectors have lower spectral sensitivity in the UV band than in the visible range.
Consequently, the Eeff measurement involves using an enhanced detector in the UV band such as
the back-thinned CCD array of the 2048x14 instrument for example. If the matrix correction is
not sufficient, a combination of correction matrix and a filter can be applied to enhance the
results.
The largest stray light errors were observed during measurements of the spotlight with a
protective glass for reasons previously described. Thus it is highly inadvisable to assess
efficiency of protective goggles in the workplace by measuring radiation from a source through
the goggles. It is better, whenever possible, to separately measure radiation emitted from the
source in the workplace and transmittance of goggles in the laboratory. Similarly, measurements
in the workplace should be intended to assess the risk of industrial processes or specific lamps
integrated in industrial processes which cannot be moved. In the case of lamps, measuring the
spectrum in laboratory with a double monochromator spectroradiometer would be preferable
and give more accurate results.
Since the operational wavelength range (200-1100 nm) of the spectroradiometers used matches
the spectral range of the silicon array detector, the stray light δsl from signals at wavelengths
greater than 1100 nm is negligible. Conversely, when the instrument’s measurable range is
limited to the UV and Visible bands, δsl is significant, as demonstrated in figure 4a. In this case,
a longpass filter which cuts out the signal at 800 nm such as the F800 filter can replace the
shutter. To ensure the measured signal with the filter only comes from the stray light and the
dark current, the transmittance of the filter must be below 10-5 for a 16 bit AD convertor.
In this study, the assessment of blue-light risk [11] was based on irradiance EB for convenience,
while most situations require radiance LB to be calculated instead of irradiance EB. As luminance
L of a source is easily measured, LB can be obtained as follow:
where K=683 lm.W-1 and V(
) is the spectral luminous efficiency.
Workplace risk assessment has to cover the whole range 200-3000 nm. In order to complete the
spectral range of the CCD spectroradiometers (200-1100 nm), array spectroradiometers
covering the region from 900 to 2500 nm can be used. A few measurements with an IR array
spectroradiometer have shown that the stray light error is not significant and does not affect
spectral measurements to be compared with the ELV.
5. Conclusion
It is important that people in charge of risk assessment of optical radiation in the workplace are
aware that using array spectroradiometers can dramatically overestimate the risk when the stray
L
VEK
E
Lnm
nm
B
B
)(
780
380
17
light error is not corrected. A spectral stray light correction method developed by Zong et al.
[15] at NIST reduces the stray light error by one or two orders of magnitude and has advantages
to be applied in real time. The current study, based on the NIST method, has shown that the
same results were obtained by using only nine simple lasers to build the correction matrix,
making this method cheaper and more easily implemented by most laboratories.
Regarding the risk assessment in the workplace, this spectral stray light correction method helps
determine the exposure level with an acceptable uncertainty for the majority of workplace
situations. In some cases, applying a double correction enhances the results. In spite of the
efficiency of the matrix correction, Eeff quantity may remain affected by a residual stray light
error particularly where the UV signal is close to zero or weak compared to the signal in the
visible range. The level of uncertainty depends upon the model of spectroradiometers used; the
best results are obtained with CCD detectors having an enhanced spectral sensitivity in the UV
range. Thus corrected spectroradiometers require a validation against a scanning double
monochromator spectroradiometer before using them for risk assessment in the workplace.
To have access to a suitable instrument for risk assessment, further work will consist of
implementing a correction matrix and corrected calibration factors in the spectroradiometer’s
software. Measurements with this instrument will be validated in the workplace.
References
[1] Murray W E 1990 Ultraviolet radiation exposures in a microbacteriology laboratory
Health Phys. 58 507–10
[2] Barlier-Salsi A and Salsi S 1998 Lampes à rayonnement ultraviolet - quantification des
risques associés à leur utilisation Cah. Notes Doc. 170 49–56
[3] Salsi S and Barlier-Salsi A 2013 Exposition aux dispositifs d’éclairage scénique : risque
pour la santé des professionnels du spectacle vivant ou enregistré Radioprotection 48 391–
410
[4] Martinsons C 2013 Les diodes électroluminescentes et le risque rétinien dû à la lumière
bleue Photoniques 63 44–9
[5] ANSES 2010 Effets sanitaires des systèmes d’éclairage utilisant des diodes
électroluminescentes (LED) (ANSES)
[6] Whillock M J, Bandle A M, Todd C D and Driscoll C M H 1990 Measurements and
hazard assessement of the optical emissions from various industrial infrared sources J.
Radiol. Prot. 10 43–6
[7] Barlier A 1994 Rayonnements optiques dans une forge - mesures et moyens de protection
Cah. Notes Doc. 155 181–93
[8] ICNIRP 2004 Guidelines on limits of exposure to ultraviolet radiation of wavelengths 180
nm to 400 nm (incoherent optical radiation) Health Phys. 87 171–86
[9] ICNIRP 1997 Guidelines on limits of exposure to broad-band incoherent optical
radiation (0.38 to 3 µm) Health Phys. 73 539–54
[10] ICNIRP 2013 ICNIRP guidelines on limits exposure to incoherent visible and infrared
radiation Health Phys. 105 74–91
[11] European Union 2006 Directive 2006/25/EC of the european parliament and of the
council of 5 April 2006 on minimum health and safety requirements regarding the
exposure of workers to risks arising from physical agents (artificial optical radiation) Off.
J. Eur. Union L114 38 – 59
18
[12] Trevelyan F 2011 The use of clinical broadband UV radiometers for optical radiation
hazard measurement J. Radiol. Prot. 31 453–65
[13] Reed N G, Wengrattis S and Sliney D H 2009 Intercomparison of instruments used for
safety and performance measurements of ultraviolet germicidal irradiation lamps J.
Occup. Enviromental Hyg. 6 289–97
[14] Barlier-Salsi A and Salsi S 2010 Mesures des rayonnements optiques aux postes de
travail : comparaison de différentes méthodes et matériels de mesure Radioprotection 45
307–20
[15] Zong Y, Brown S W, Johnson B C, Lyckke K R and Ohno Y 2006 Simple spectral stray
light correction method for array spectroradiometer Appl. Opt. 45 1111–9
[16] Shaw M and Goodman T 2008 Array-based goniospectroradiometer for measurement of
spectral radiant intensity and spectral total of light sources Appl. Opt. 47 2637–47
[17] Shen H S, Pan J, Feng H and Liu M 2009 Stray light errors in spectral colour
measurement and two rejection methods Metrologia 46 129–35
[18] Ylianttila L, Visuri R, Huurto L and Jokela K 2005 Evaluation of a single-
monochromator diode array spectroradiometer for sunbed UV-radiation measurements
Photochem. Photobiol. 81 333–41
[19] Brown S W, Johnson B C, Feinholz M E, Yarbrough M A, Flora S J, Lyckke K R and
Clark D K 2003 Stray light correction algorithm for spectrographs Metrologia 40 81–4
[20] Zong Y, Brown S W, Lykke K R and Ohno Y 2007 Correction of straylight in
spectroradiometers and imaging instruments Proceedings of 26 CIE th session (BEIJING)
[21] Coleman A, Sarkany R and Walker S 2008 Clinical ultraviolet dosimetry with a CCD
monochromator array spectroradiometer Phys. Med. Biol. 53 5239–55
[22] Sperling A, Larionov O, Grusemann U and Winter S 2005 Stray-light correction of array
spectrometers using tuneable pulsed and cw lasers Proceedings of the 9th international
conference on new developments and applications in optical radiomety (NEWRAD) Davos
p 93
[23] Sperling A 2011 Tuneable lasers at PTB for photometry and radiometry NewRAD 2011
Abstract collection NEWRAD conference (Maui (Hawaii)) pp 10 –11
[24] Nevas S, Sperling A and Oderkerk B 2012 Transferability of stray light corrections
among array spectroradiometers AIP conference proceedings 1531,821(2013) AIP
conference (Berlin)
[25] Kreuter A and Blumthaler M 2009 Stray light correction for solar measurements using
array spectrometers Rewiew Sci. Instrum. 80 1–3
[26] Salim S G R, Fox N G P, Hartree W S, Woolliams E R, Sun T and Grattan K T V 2011
Stray light correction for diode-array-based spectrometers using a monochromator Appl.
Opt. 50 5130–8
[27] Feinholz M E, Flora S J, Brown S W, Zong Y, Lykke K R, Yarbrough M A, Johnson B C
and Clark D K 2012 Stray light correction algorithm for multichannel hyperspectral
spectrographs Appl. Opt. 51 3631–41
[28] Zonios G 2010 Noise and stray light characterization of a compact CCD
spectrophotometer used in biomedical applications Appl. Opt. 49 163–9
[29] EN 14255-1 2005 Measurement and assessment of personal exposures to incoherent
optical radiation - part 1: Ultraviolet radiation emitted by artificial sources in the
workplace AFNOR 29
[30] EN 14255-2 2006 Measurement and assessment of personal exposures to incoherent
optical radiation - part 2: Visible and infrared radiation emitted by artificial sources in the
workplace AFNOR 43
