Sunteți pe pagina 1din 9

Infrared Physics & Technology 44 (2003) 9199

www.elsevier.com/locate/infrared

Non-contact multiband method for emissivity measurement


a,* b
Adam Mazikowski , Krzysztof Chrzanowski
a
Gdansk University of Technology, Department of Optoelectronics, Narutowicza 11/12, 80-952 Gdansk, Poland
b
Military University of Technology, Institute of Optoelectronics, Kaliskiego 2, 00-908 Warsaw 49, Poland
Received 17 June 2002

Abstract

During the last decade an increasing interest in passive multiband systems for temperature measurement was noted
and quite a few such systems have been developed. However, recent studies showed that multiband systems are capable
of producing accurate results of non-contact temperature measurement only in limited number of applications and that
multiband systems will not become a real rival for single band systems in temperature measurement applications.
Available literature about passive multiband systems concentrated exclusively on the problem of temperature mea-
surements with these systems in situation when these systems can be used for non-contact emissivity measurements too.
A model of a passive multiband system for non-contact emissivity measurement has been developed in this paper.
Simulations carried out using this model showed that it is possible to achieve reasonable accuracy of emissivity mea-
surements with passive multiband systems and these systems can be considered as an attractive solution for emissivity
measurements in industrial conditions.
2002 Elsevier Science B.V. All rights reserved.

1. Introduction parameters: temperature, contamination, rough-


ness or viewing angle. To make matters worse, not
Emissivity describes the objects ability to emit only is the object emissivity usually unknown, but
thermal radiation, and measurement of emissivity in many applications it may change during the
is of signicant importance in science, industry and measurement.
other areas. When heat transfer is modelled, Several methods for emissivity measurements
emissivity must be known. Most of commercially have been developed [26]. Direct measurement of
available thermal instruments, such as pyrometers, emissivity using the denition: ratio of sample
thermal scanners or thermal cameras need the emissive power to blackbody emissive power is the
emissivity of the object examined as an input pa- most popular method. This technique involves
rameter [1]. It is usually obtained using literature measurement of the examined object and spe-
data. Unfortunately, emissivity depends on many cially prepared reference plate of known emissiv-
ity. Indirect emissivity measurement using the
relationship between object reectivity and ob-
ject emissivity has found many applications, too.
*
Corresponding author. The latter method requires integrating bi-direc-
E-mail address: adamazik@eti.pg.gda.pl (A. Mazikowski). tional reectivity measurements over an entire

1350-4495/02/$ - see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S 1 3 5 0 - 4 4 9 5 ( 0 2 ) 0 0 1 8 2 - 2
92 A. Mazikowski, K. Chrzanowski / Infrared Physics & Technology 44 (2003) 9199

hemisphere to work out the hemispherical-direc- systems is developed. Calculations of emissivity


tional reectivity. Another group of techniques measurement errors in possible measurement
that have been developed recently are techniques conditions were carried out. Conclusions about
based on periodic radiometry [5]. It is shown that capabilities of multiband systems in industrial
thermal modulation of a sample allows direct and applications in comparison to other methods of
indirect emissivity measurements even at low emissivity measurement are presented.
temperature, with ordinary equipment. The emis-
sivity measurement methods mentioned above
allow us to perform reliable and accurate labora- 2. Basic concept of the multiband radiometer
tory emissivity measurements. However, they
have some disadvantages, when in eld measure- In passive multiband systems measurements of
ments are needed. In most cases they involve the emitted radiance of the object are taken at
knowledge of the sample temperature, tempera- several wavelengths (usually more than four). In
ture modulation, sophisticated equipment or result we obtain a set of n equations with m un-
specially prepared samples as reference source. knowns, as presented below:
To conclude, they cannot allow us to perform
emissivity measurement in working and un- S1 f Tob ; ea1 ; a2 . . . am1 ; k1 . . .
known conditions often met in many industrial S2 f Tob ; ea1 ; a2 . . . am1 ; k2 . . .
1
applications. ...
During the last decade an increasing interest in
Sn f Tob ; ea1 ; a2 . . . am1 ; kn . . .
passive multiband systems for temperature mea-
surement was noted [714]. The method is based where n is the number of detection bands, Sn is the
on the assumption that object emissivity can be signal measured at n band, Tob is the true object
well approximated by function of wavelength of m temperature, ea1 ; a2 . . . am1 ; k is the object emis-
parameters where m  1 must be lower or equal sivity at wavelength k.
to number of system spectral bands. A few multi- Both the temperature Tob and the emissivity
band pyrometers and thermal cameras have been ea1 ; a2 . . . am1 ; k of the object tested can be cal-
developed [1518]. However, recent studies culated by tting the measurement data to the
showed that multiband systems are capable of theoretical model [7,8,10,20]. For this purpose a
producing accurate results of non-contact tem- few dierent techniques are used. The most pop-
perature measurement only in a limited number ular is the least-squares technique [21,22]. Re-
of applications and that multiband systems will searchers interested in temperature measurements
not become a real rival for single band sys- analysed the accuracy of determination of the
tems in temperature measurement applications temperature Tob by solving set of equations (1)
[10,11,19]. omitting the problem of accuracy determination of
Available literature about passive multiband the emissivity ea1 ; a2 . . . am1 ; k1 [714]. In this
systems concentrates exclusively on problem of paper we will have an inverse situation.
temperature measurements with these systems in Graphical presentation of emissivity (or tem-
situations when these systems can be used for non- perature) measurement process with passive in-
contact emissivity measurements, too. A study of frared radiometer is shown in Fig. 1. Radiation
the accuracy of non-contact emissivity measure- emitted by the investigated object is collected by
ment with multiband systems is in this paper. The the optical system and then transmitted through
aim of the study is determination of accuracy of band-pass lters to the detector (detectors). The
emissivity measurement with multiband systems in electrical signal is then amplied and converted to
real industrial conditions. A mathematical model digital form. Next, using the chosen technique the
that enables determination of errors generated emissivity (or temperature) of the measured object
during measurement process with real multiband is calculated.
A. Mazikowski, K. Chrzanowski / Infrared Physics & Technology 44 (2003) 9199 93

Fig. 1. Emissivity measurement system using multiwavelength infrared radiometry. BRbackground radiation, ORobject radia-
tion.

3. Mathematical model of emissivity measurement 


gn  dgn Rn  dRn Adn
system Sn
4F 2 1
Z
The signals S1 ; . . . ; Sn in the set of equations (1)  eob kMTob ; kso ksFn ksn k dk
Dk
are signals measured at the output of the multi- Z n
band radiometer. The components f Tob ; ea1 ; 1  eob keback kMTback ; kso ksFn ksn k dk
Dk
a2 . . . am1 ; k1 ; . . . ; f Tob ; ea1 ; a2 . . . am1 ; kn are Z n 
the assumed signals calculated using a theoretical eopt kMTopt ; ksFn ksn k dk
Dkn
model of the measurement process and let us call  Van  Vdn 3
these signals Sa1 ; . . . ; San .
The assumed output signal Sa in channel n can where eob k, eback k, eopt k are, respectively, real
be modelled mathematically as [19,24] object, background and optics emissivity, dgn is the
standard deviation of the gain gn , dR is the stan-
Z 
gn R Adn dard deviation of the detector peak spectral re-
Sn 2n eas kMTas ; kso ksFn ksn k dk sponsivity R , Van and Vdn are the standard
4F 1 Dkn
deviations caused by analog and digital noise re-
2
spectively.
As we can see there are signicant dierences
where F is the optics F -number that equals the between the assumed signal (2) and the real signal
ratio of focal length f and the aperture diameter shown in (3). Eq. (3) takes into account additional
D [19,24], gn is the electronic system amplica- parameters: background radiation, radiation
tion in n spectral channel, R is the detector emitted by optical elements, electronic noise and
peak spectral responsivity at n spectral channel, non-ideal emissivity approximation of the real
Ad is the detector area, eas k is the assumed ob- body. Depending on measurement conditions
ject emissivity, MT ; k is the blackbody spectral some of these factors can be negligible, while
radiant exitance in temperature T and wave- others can cause signicant errors of object emis-
length k, sn k is the detector relative spectral re- sivity measurements. Now let us dene the quan-
sponsivity, so k is the transmittance of optics and tities presented in Eqs. (2) and (3).
sFn k is the lter transmittance in n spectral To choose an emissivity approximation function
channel. it is necessary to make a few assumptions. From
The real output signal Sn in channel n obtained analysis of analytical functions for approximation
in working measurement conditions can be pre- of object emissivity available in literature we can
sented as follows [19,24]: conclude that a function of one to two parameters
94 A. Mazikowski, K. Chrzanowski / Infrared Physics & Technology 44 (2003) 9199

can well approximate emissivity functions of most LSBn Smax n


objects [19,23]. As emissivity function used in Vdn p p 6
12 122k
multiband systems must very well interpolate ob-
ject emissivity at system spectral bands, let us as- where Smax n is maximal value of the analog signal
sume that a function of at least three parameters is to be digitised at electronic channel n, k is bit
needed. From a few possible functions a mth de- number of A/D converter.
gree exponential polynomial was chosen. Additional electronic errors may be caused by
variation of system channel gain and variation of
ek expa0 a1 k a2 k2 4 detector responsivity due to the system tempera-
ture changes. These errors cannot be modelled
Such a form for emissivity makes possible to use a mathematically, but can be well estimated. Let us
linear least squares method for unknown param- consider a case when the same electronic channel is
eters calculations. This method is fast and simple, used for all optical bands, so the errors occur si-
gives good results but is limited to use exponential multaneously in each of system bands. Such errors
polynomial as object emissivity approximation may be assessed as the dierence between param-
[10,14]. eters during calibration and those existing in real
The output signals in each channel suer from measurement conditions.
electronic errors. The output analog noise consists All multiband infrared systems require a cali-
of two components: the noise generated by the bration procedure. During calibration it is possible
detector and amplied by electronics blocks and to precisely determine the signal caused by the
additional noise generated by these blocks. For a optics and geometrical factors. As a reference
well designed electronics the output noise is mostly source a blackbody is usually used. The limited
caused by the detector. In case of pin-type detec- accuracy of the blackbody temperature DTcal is the
tor, working in current-to-voltage converter con- main source of calibration errors as other errors
guration, the output analog error can be assessed can be successfully minimised in laboratory con-
as ditions. The calibration signal is usually deter-
s
2 mined experimentally.
rn Rn
Van Adn Dfn iN rn 2 Dfn 4kTsyst rn Dfn
Dn
5 4. Calculations

where rn is the current-to-voltage converter trans- Having dened multiband radiometric system
resistance in n spectral channel, iN is the spectral components as mathematical formulas, it is pos-
density of current noise of operational amplier, k sible to perform computer simulations. Because
is the Boltzmann constant, Tsyst is the electronic multiband systems can be designed in many ways,
device temperature, D is the detector peak nor- let us make a few general assumptions about de-
malised spectral detectivity (cm Hz1=2 W1 ), Df is sign of the system.
the noise equivalent bandwidth to the output of First, a system with six spectral bands, located
the analog channel n. between 1.3 and 1.85 lm, will be analysed. The
The error generated by limited resolution of the number of system spectral bands is greater than a
digitisation system used in channel n can be cal- minimum number. Additionally, in this spectrum
culated as standard deviation of rectangular range atmosphere attenuation is negligible. Sec-
probability distribution. In the range determined ond, we assume that the optical signal emitted by
by least bit (LSB) of the converter, the digital optical elements is also negligible or can be pre-
output is equal to 0 for the analog input less cisely determined during the calibration process.
than LSB/2 and equal LSB for the analog input Third, the limited blackbody temperature accuracy
greater than LSB/2. Using simple integration we will be considered as the only source of calibration
obtain [19] errors. Fourth, the extended InGaAs pin-type de-
A. Mazikowski, K. Chrzanowski / Infrared Physics & Technology 44 (2003) 9199 95

tector, working with current-to-voltage converter


as preamplier will be used for calculations.

4.1. The emissivity accuracy criteria

For emissivity accuracy assessment the emissiv-


ity accuracy criteria must be determined. The ob-
ject emissivity is a function of wavelength, so
simple comparison of parameters may not work
properly. In such a situation a new way of as- Fig. 2. The emissivity measurement error caused by method
sessing emissivity accuracy is proposed. We in- and numerical operations.
troduce a parameter hDei that equals the average
of error modulus for all wavelengths in the used
and was eliminated in further computer simula-
spectral range
Z tions.
1
hDei jeas k  ecalc kj dk 7
Dk Dk 4.3. Background radiation and object temperature
where Dk is the used spectral region, eas k is the uctuations
assumed real object emissivity, ecalc k is the cal-
culated object emissivity. Based on this parameter The inuence of background radiation on
the emissivity measurement accuracy will be de- measurement results depends on the background
termined by further computer analysis. In most temperature Tback , its emissivity eback and the object
cases, when any pseudorandom value is added to reectivity. Let us consider a typical situation
the system parameter examined, the calculations when the background emissivity is high, close to 1.
must be repeated several hundred times so we Additionally we will choose iron as the measured
obtain good statistical results. The statistical error object, whose emissivity can be approximated
can be calculated as shown by the following function: eob exp0:361
0:667k2 [22,23]. The simulation results for three
1X dierent object temperatures equal to 300, 400 and
hDeistat hDeik 8
k k 500 C respectively are presented in Fig. 3.
As we can see, the error caused by background
where k is the number of repetitions, hDeik is the
radiation reected by the object strongly depends
average error (7) obtained in kth trial.
on background temperature and object tempera-
ture. The calculated curves shift right when the
4.2. The method accuracy
object temperature rises. The correlation of signals
At the beginning of emissivity accuracy analysis
the accuracy of calculations should be assessed.
Some errors may be eected by numerical opera-
tions and do not completely fulll the least squares
method assumption [9,21]. For this purpose let us
assume ideal measurement conditions, without
inuence of any error causing factors. The results
of the simulation are shown in Fig. 2.
The results obtained show that the error gener-
ally decreases as the object temperature rises.
However, all the errors can be thought of as ac-
ceptable or, in higher temperatures, negligible. Fig. 3. The emissivity measurement error caused by inuence
Additionally, the error has systematic character of background radiation.
96 A. Mazikowski, K. Chrzanowski / Infrared Physics & Technology 44 (2003) 9199

Fig. 4. The emissivity measurement error caused by object Fig. 5. (1) The real and (2) the calculated object emissivity as
temperature uctuations. function of wavelength.

received in dierent spectral bands cause a few function. In the whole analysed spectral range the
local minima. However, it is clear that for object dierence between assumed and calculated emis-
temperatures over 300 C and background tem- sivity is about 0.04. Additionally, other tests have
perature below 100 C the emissivity errors are been carried out to analyse the inuence of object
relatively small, almost negligible. For higher temperature on these errors. The errors seem to be
background temperatures these errors can be more practically independent of temperature of the
signicant. measured object.
The system channel signals are measured si-
multaneously or quasi-simultaneously. Therefore, 4.5. Analog and digital electronic noise
let us assume uctuations of object temperature
during measurement of about 0.1 C. Multiband The noise in analog electronic blocks repre-
systems are sensitive to such errors which result in sented by the standard deviation of the output
quite a signicant emissivity measurement errors electronic signal Van introduced in point 3 causes
(Fig. 4). variations of the indications of the radiometer.
Because analog noise depends mainly on the nor-
4.4. Object emissivity approximation malised detector detectivity D , calculations of
standard deviation of output emissivity variations
The multiband method of emissivity measure- were carried out for three dierent normalised
ment is based on a fundamental assumption that detectivity values for objects made from iron.
the real object emissivity can be very well ap- Other parameters included into calculations are
proximated by a mathematical function with a few spectral density of current noises for operational
unknown parameters. For applying the linear least amplier iN and system gain g. The rst was set to
squares method the exponential polynomial is 0.01 pA Hz2 , a value typical for standard ampli-
needed. In fact this assumption is not exactly ful- er, while the latter was set to 106 .
lled so emissivity measurement errors can occur. As shown in Fig. 6, the emissivity measurement
To assess these errors the following test was per- errors caused by noise in the analog electronic
formed. Let us assume the object made from N- channel decrease when the object temperature ri-
155ASTM steel of oxidised surface, so the real ses. For detectors of normalised detectivity equal
object emissivity is given by a function presented to 5  1011 or greater, which is typical for most of
in Fig. 5(curve 1) [19,23]. In Fig. 5 the spectral pin-type detectors, these errors are relatively small,
characteristics of calculated object emissivity is while for lower detectivity they can be more sig-
also presented (curve 2). nicant.
The test performed shows serious emissivity er- As well as analog noise digital errors can inu-
rors if the real object emissivity cannot be exactly ence the emissivity measurement accuracy due to
approximated by the assumed mathematical the limited resolution of digitisation system. Be-
A. Mazikowski, K. Chrzanowski / Infrared Physics & Technology 44 (2003) 9199 97

Fig. 6. Emissivity measurement error caused by analog elec- Fig. 8. Emissivity measurement error due to variation of gain
tronic channel noise. and detector responsivity.

cause of wide range of object temperatures and DTsyst 2 K as a common value for all channels.
emissivities it is almost impossible to perform the Additionally, system temperature can change
analog-to-digital conversion using one xed range. during measurements, so potentially the measure-
Let us make some assumption then. First, signals ment in each channel can be taken at dierent
from all optical channels are measured using the temperatures. Let us assume that this dierence
same range. Second, we assume the situation, to DTn 0:2 K. Greater temperature changes
when the sub-ranged system allows us to obtain can be well controlled and compensated. In Fig. 8
near 100% of full analog-to-digital converter range the errors presented are relatively small, about
for the highest output analog signal. 0.01 and are rather independent of object tem-
Calculations were performed for three dierent perature.
resolution A/D converters. As we can see in Fig. 7
the errors caused by the limited resolution of the 4.6. Calibration errors
A/D converter are signicant. Additionally, they
strongly depend on bit number of A/D converter. In point 3 we assume the blackbody temperature
Even applying 16-bit converters do not reduce and emissivity accuracy as sources of calibration
these errors below 0.01. It can also be noticed that errors. The typical blackbody temperature uncer-
limited digital resolution errors decrease as the tainty DT is about 0.25% of absolute blackbody
object temperature rises. temperature, while the emissivity uncertainty is
Other errors are errors that exist due to the gain smaller than 0.005.
and responsivity variations. They can be assessed The calculations showed that errors caused by
as 0.25% per one degree of system temperature calibration may be considered as negligible in
change [19]. Let us assume that the temperature comparison to errors previously discussed.
dierence between calibration and real conditions
4.7. Final results

Let us now calculate the emissivity measurement


accuracy in real condition, when all previously
discussed factors are present. For calculation we
assumed the following parameters: the real object
made from N-155ASTM steel of oxidised sur-
face, Tback 100 C, eback 1, 12 bit A/D con-
verter, uncooled G5851-01 detector of normalised
detectivity D 1  1011 , the object temperature
uctuations 0.1 C, the common gain and detector
Fig. 7. Emissivity measurement error caused by limited reso- responsivity uctuations 0:5  0:05% in each sys-
lution of digitisation system. tem channel. Results are presented in Fig. 9.
98 A. Mazikowski, K. Chrzanowski / Infrared Physics & Technology 44 (2003) 9199

Errors caused by background radiation reected


by the object strongly depend on background and
object temperatures. For background temperature
below 100 C and object temperature higher than
300 C, the inuence of this source of errors can be
considered as non-signicant, too.
Generally, simulations carried out using the
developed model of a passive multiband system for
non-contact emissivity measurement showed that
Fig. 9. Emissivity measurement error in assumed real condi-
it is possible to achieve reasonable accuracy of
tions.
emissivity measurements with passive multiband
systems and these systems can be considered as an
Although some of the previously analysed fac- attractive solution for emissivity measurements in
tors cause quite signicant errors, in general case industrial conditions.
these errors are partially compensated. Measure-
ment accuracy is relatively small only for lower
temperatures, while for higher values the results
obtained can be considered as attractive in indus- References
trial conditions. The measurement accuracy of
other methods in laboratory conditions is 0.005 [1] R.P. Madding, Emissivity measurement and temperature
correction accuracy consideration, Proc. SPIE 3700 (Ther-
0.01. However, for methods useful in industry mosense XXI) (1999) 393401.
conditions the accuracy is reported to be 0.020.03 [2] L. Chen, B.T. Yang, X.-R. Hu, Design principle for
[15] which is a comparable value to the one simultaneous emissivity and temperature measurements,
achieved in the method developed here. Opt. Eng. 29 (1990) 14451448.
[3] R.B. Johnson, C. Feng, J.D. Fehribach, On the validity
and techniques of temperature and emissivity measure-
ment, Proc. SPIE 394 (Thermosense X) (1988) pp. 202206.
5. Conclusions [4] D. Especel, S. Mattei, Total emissivity measurement
without use of an absolute reference, Infrared Phys.
A model of a passive multiband system for Technol. 37 (1996) 777784.
non-contact emissivity measurement has been de- [5] M. Siroux, E. Tang-Kwor, S. Mattei, A periodic technique
for emissivity measurements of insulating materials at
veloped. It shows that multiband method of moderate temperature, Meas. Sci. Technol. 9 (1988) 1956
emissivity measurements is based on validity of 1962.
some assumptions about measurement conditions, [6] V. Webbeking et al., Emissivity measurement apparatus
system design and calibration process. In such a and method, US Patent 5,505,543, 1996.
situation radiometric and electrical disturbances [7] W.F. Kosonocky, M.B. Kaplinsky, et al., Multi-wave-
length imaging pyrometer, Proc. SPIE 2225 (1994) pp. 26
must be analysed as potential sources of mea- 43.
surement errors. [8] Z. Bielecki, K. Chrzanowski, T. Piaz tkowski, M. Szulim,
Simulations carried out using a model of emis- Multiband infrared pyrometer, in: KKM98Polish Na-
sivity measuring multiband radiometer showed tional Metrology Congress, 1998.
that non-ideal approximation of object emissivity [9] A. Mazikowski, R. Hypszer, M. Jez drzejewska-Szczerska,
Modelling of non-contact temperature measurement sys-
by assumed mathematical function and uctuation tem using multiwavelength pyrometry, Proc. SPIE 4516
of object temperature during measurement process (2001) 120124.
are the most important sources of errors. For [10] K. Chrzanowski, Z. Bielecki, M. Szulim, Comparison of
a well-designed system the inuences of other temperature resolution of single-band, dual-band, and
sources as caused by analog electronic noise, multiband infrared systems, Appl. Opt. 38 (13) (1999)
28202823.
blackbody non-accuracy, limited resolution of [11] K. Chrzanowski, M. Szulim, Errors of temperature mea-
analog-to-digital converter can be considered as surement with multiband IR systems, Appl. Opt. 38 (10)
non-signicant. (1999) 19982006.
A. Mazikowski, K. Chrzanowski / Infrared Physics & Technology 44 (2003) 9199 99

[12] V. Tank, H. Dietl, Multispectral infrared pyrometer for [18] M. Khan et al., Emissivity independent multiwavelength
temperature measurement with automatic correction of the pyrometer, US Patent 5,132,992, 1992.
inuence of emissivity, Infrared Phys. 30 (1990) 331342. [19] K. Chrzanowski, Non-contact thermometry. Measure-
[13] R.R. Corwin, A. Rodenburgh, Temperature error in ments errors, Res. Dev. Treat. (Polish Chapter of SPIE)
radiation thermometry caused by emissivity and reec- 7 (2000).
tance measurement error, Appl. Opt. 33 (10) (1994) 1950  Andreic, Distribution temperature calculations by tting
[20] Z
1957. the Planck radiation curve to a measured spectrum, Appl.
[14] C. Schietinger et al., Non-contact optical techniques for Opt. 31 (1) (1992) 126130.
measuring surface conditions, US Patent 5,490,728, 1996. [21] M.A. Khan, C. Allemand, T.W. Eagar, Noncontact
[15] V. Tank, Method of and device for contactless temperature temperature measurements. Least squares based tech-
measurement of an object independently of radiation niques, Rev. Sci. Instrum. 62 (1991) 403409.
emissivity, US Patent 4,924,478, 1990. [22] M.A. Khan, C. Allemand, T.W. Eagar, Noncontact
[16] U. Kienitz, Pyrometric measurement procedure and multi- temperature measurements. Interpolation based tech-
channel pyrometer, US Patent 4,880,314, 1989. niques, Rev. Sci. Instrum. 62 (1991) 392402.
[17] Z. Bielecki, K. Chrzanowski, R. Matyszkiel, M. Szulim, [23] A. Sala, Radiant Properties of Materials, PWN-Polish
Infrared pyrometer for temperature measurement of ob- Scientic Publishers, Warsaw, Elsevier, Amsterdam
ject, emissivity of which depends of wavelength and time, Oxford, 1986.
Optica Applicata 29 (1999) 287294. [24] C.L. Wyatt, Radiometric System Design, London, 1987.

S-ar putea să vă placă și