Application Note Behavior of the MCA 166 at different Temperatures and Gain settings and limits of centroid accuracy 10.1.2001 Jörg Brutscher GBS Elektronik GmbH 1. Problem 2. Basic limits 3. Centroid error 4. Temperature drift of the MCA 166 5. Temperature drift of detectors 6. Stabilisation 7. Linearity and accuracy of gain setting 8. Conclusions 1. Problem For very accurate measurements, all possible influences on measurement performance have to be known and understood. This helps to estimate errors and to improve measurement conditions. In this document, the temperature drift and the linearity of the gain and its dependence on settings have been examined. 2. Basic limits Gamma spectroscopy can be a very accurate measurement method in terms of measuring photon energy. The limits are mostly given by detector resolution. For example, the FWHM of the Co60 1333 keV peak with measured with a HPGe (High purity Germanium) detector is typical 1.85 keV. A useful assumption is that the centroid of a peak can be calculated with an error of 10% of its FWHM. This is a measurement error in this case of 1.4E-4=140 ppm. With other detectors the error is larger. Table 1: typical photon energy measurement errors for different detectors. detector condition relative error HPGe 1.85 keV FWHM at 1333 keV 1.4E-4 CZT 12 keV FWHM at 662 keV 1.8E-3 NaI 7.5% FWHM at 662 keV 7.5E-3 Gamma spectroscopy with semiconductor or scintillation detectors is a relative measurement method which relies on calibration by standard photon sources. So main requirements are linearity and stability. The most interesting point is the peak drift caused by temperature change. Temperature drift may be caused by the detector crystal, the preamplifier and the main amplifier within the MCA. For accurate measurements it is desirable to keep conditions such that the drift does not exceed the relative errors mentioned in table 1. Absolute values of gain are not critical because the MCA has to be calibrated anyway. The absolute accuracy of the MCA166 gain setting is a few percent. 2. Centroid error h The peak centroid is the sum of the background corrected channel i * Spectrumi contents multiplied by the channel number and divided by the sum of i =l (1) Centroid = h the background corrected channel contents. Only the channels above the half maximum are used (MCA measurement software algorithm). Spectrumi From this definition it can be derived that there is a systematical and a i =l statistical component of the centroid error. Statistical error The measurement of a peak in a spectrum can be seen as N repeated measurements of the corresponding photon energy, where N is the peak area. The standard deviation of a single measurement is 42.5% of the FWHM of the gaussian distribution. The standard deviation of the average is normally the standard deviation of a single measurement divided by the square root of the number of single measurements. In the definition only the ∑ ∑ channels above half maximum are used, which represent 76% of the total peak area. This yields in a statistical ∆E ≅ 0.5 × centroid error depending on peak area N of FWHM N (2). systematic error: From the definition above, it can be derived that the centroid is not a linear function of the photon energy, but has some discontinuities. This discontinuities occur, when a channel at the edge has about half of the maximum counts and it is decided whether to include it into the centroid calculation or not. The difference when including another channel is ∆n = ∑ is ∑s i − m ⋅ sm + ∑ isi i . sm + ∑ si The edge channel m is about 0.5*FWHM from the centroid: ∑ is ∑s i − m = 0.5 ⋅ FWHM , i (i − n )2 (i − n )2 *8 ln 2 0 0 area area ⋅ 8 ln 2 2 FWHM 2 2 the gaussian distribution s of the peak is si = , e 2σ = e σ 2π FWHM 2π the sum of the channels above FWHM is ∑ si = 0.76 ⋅ area , the content of the edge channel is just half of the maximum channel sm = 2 ⋅ ln 2 0.618 sm 1 area ⋅ 8 ⋅ ln 2 and the ratio is . = = 2 FWHM 2π ∑ si 0.76 ⋅ FWHM 2π FWHM Assuming now that the FWHM (in channels) is large (>4), the difference can be estimated as 0.5 ⋅ FWHM ⋅ ∆n = 1+ sm ∑ si sm ∑ si ≅ 0.5 ⋅ FWHM ⋅ 0.618 = 0.309 (3) FWHM So almost independent from FWHM, this centroid calculation algorithm causes discontinuities of about 0.3 channels, which can be seen as an systematic error of +/- 0.15channels. As this discontinuities are caused both by channels on the left and the right side of the peak, it is better to multiply this value with error of +/- 0.21 channels to be assumed. 2 which results in an 11 1 8 .3 1 C e n tro id e rro r (in F W H M u n its) FW H M 3 ch a n n els FW H M 1 1 .5 c h an n e ls FW H M 4 6 ch a n n e ls 3 51 .99 0.1 6 0 9 .1 4 6 0 9.2 3 3 5 2.0 7 35 1 .9 6 0.01 10 100 1000 P e a k a re a Fig. 1. Centroid error dependend on peak area using peaks which are large compared to background. If the FWHM is only a few channels, the formula derived for statistical centroid error can be applied. If FWHM is too large, then the centroid error is increased due to strong fluctuations of the channels used for evaluation. In an experimental study of the centroid error it was found that there is a third contribution to the centroid error if the peak area is distributed to many channels. In this case, the statistics of a single channel content is very bad. This causes the FWHM determination algorithm to work incorrect and the fluctuation of channels used for centroid calculation is much larger than one channel. This also leads to an increased centroid error. To avoid this, it is recommended to adjust the MCA resolution in a way that the FWHM is about 4 - 8 channels. To reduce the range of possible centroid errors, it can be stated that peaks with an area <30 counts are hardly recognizable and centroid errors < 5%FWHM seem hard to believe. So for practical purposes the following may be assumed (if FWHM=3...12channels and the peak is large compared to background): Table 2 Area peak area<30 30<peak area <400 Centroid error not a peak peak area>400 FWHM*0.05 ∆E ≅ FWHM peak area 4. Temperature drift of the MCA 166 At first, it is evaluated how the drift changes with time. Knowing the thermal time constant allows to judge how long it necessary to wait until the MCA runs stable. It also tells that for measurement times short compared to the time constant, resolution losses due to gain drift can be reduced. 1 C e ntroid d rift 0.998 0.996 M CA 166 w arm ing up from about -7°C 0.994 0.992 0.99 0.988 0 50 10 0 1 50 200 25 0 30 0 350 40 0 T im e (m in ) Fig. 2. Reaction of the MCA166 on a sudden temperature change from -7°C to +20°C. The thermal time constant (1/e) can be evaluated as 33 minutes. planar high resolution HPGe detector type GL0310 which was kept at constant temperature measuring a Co60 source was connected to the MCA.Spectra were taken in 5 minute intervals and the drift of the 1333 keV peak was evaluated. Shaping time was 2µs, Gain =5*0.6. The drift reaction caused by switching on the device is similar to that on a thermal change of 4-7°C. So, for perfect stability it is a good idea to leave the MCA at least 3 hours running to come to a thermal equilibrium. For short measurements (few minutes) drift will not affect resolution, as drift is very slow. However, energy calibration may have to be readjusted a few times. In the next experiment it was measured how gain changes with temperature and if there are other dependencies. For this experiment a GL1015R with a Ra226 source was used. The MCA was serial number 140 (one of the first series). The count rate was about 6500 cps and 13% dead time (2µs shaping time). The MCA was in a climatic chamber and all the time connected to the charger. After temperature adjustments the next measurement was started earliest after 4h to allow the MCA to find its thermal equilibrium. This temperature test was also meant to check the reliability of the MCA electronics. Below -10°C, the battery voltage was too low to allow correct MCA operation without charger. The tests were stopped at -40°C due to lack of time and because this is far out of specifications (normal minimum temperature -5°C). In first order and within the specified temperature range, the gain change with temperature can be considered linear. Only at very low temperatures there seems to be a nonlinear effect. Much more interesting is that the drift depends strongly on the gain setting. The drift is sometimes positive, sometimes negative and sometimes nearly negligible. Fig. 3. Gain change depending on temperature. The drift depends very much on the gain setting. This behavior was evaluated a bit closer. So the drift between 20°C and 0°C was measured for gain settings between 2.7 and 560 using logarithmic steps of the E12 series. A second measurement was made for the drift between 10°C and -10°C using values between 10 and 150 and smaller steps (E24 series). A few measurements were additionally performed using a different MCA (serial number 366, recent series) and a different detector (GL0310) 6E -4 T em pe rature D rift 1/°C 4E -4 2E -4 0 1 10 100 G a in -2E -4 D rift 2 0 °C - 0 °C D rift 1 0 °C - -1 0 °C -4E -4 D rift M C A #3 6 6, D et. G L0 3 10 -6E -4 Fig. 4. Temperature drift dependence on gain setting. At the first glance, the drift dependence on gain seems to be absolutely erratic. It becomes only understandable by knowing about the internal hardware of the MCA166. The most temperature sensitive component within the MCA is the adjustable gain amplifier [1]. This contains internally a 6dB step (factor 2) attenuation ladder and a circuit for interpolating between the interpolation steps. This corresponds to the drift coefficient oscillating with a factor 2 period. At the gain of 15 and the gain of 30 this oscillation is disturbed. This can be explained by additional attenuators which are switched at these gains. The results also seem to be reproducible within 100 ppm/°C between different MCAs. But this was not further examined as these measurements are quite time consuming. 5. Temperature drift of detectors For practical purposes, it is useless to look only at the MCA drift; the drift of detectors have always to be taken into account. For this experiment, a CZT 500 detector was warmed up from -5°C to 22°C and a HPGe was cooled down from 22°C to 12°C. re lative peak shift 1.01 1 0.99 C ZT500 detector w arm ing up from -5°C to +22°C 0.98 -20 -10 0 10 20 30 40 50 T im e (m inutes) Fig. 5. The thermal time constant of a small CZT 500 is in the order of 5 minutes. The peak drift is about 1E-3 /°C 1.0006 1.0 004 P e a k sh ift o f th e C o 6 0 1 3 3 3 ke V P e a k w he n th e d e te cto r (G L 0 3 1 0 ) is co o le d d o w n fro m 2 2 °C to 1 2 °C . T h e M C A is ke p t a t co n sta n t te m p e ra tu re . M C A g a in : 5 *0 .6 , 2 µ s sh a pin g tim e . relative pe ak sh ift 1.0002 1 0.9998 0.9 996 0.999 4 0.9992 Tem perature coefficent detector: ~1E-4/°C 0 .99 9 0.9988 0 20 40 60 80 10 0 T im e (m inu te s) Fig. 6. The time constant of a planar HPGe detector is around 25 minutes. The drift is about 1E-4 /°C It can be seen that the time constant of a CdZnTe detector is quite short, which corresponds to its small size. The thermal time constant of a HPGe detector is comparable to that of the MCA, and the drift is small, but not negligible. It has to be mentioned that the drift of a HPGe is always the drift of its preamp, as the detection crystal is always at liquid nitrogen temperature. 6. Stabilisation If for a certain measurement task, a well known peak is always present, then the method of peak stabilization can be applied to compensate peak drift regardless if caused by the detector or by the MCA. For stabilization, the reference peak has to be marked with a special stabilization ROI. The centroid of this ROI is evaluated periodically and the gain is adjusted correspondingly. 1 D rift (re lative ) w ith sta b iliza tio n 0.99 8 0 .9 96 w ith o ut stab iliza tio n 0 .99 4 0 30 60 90 1 20 T im e (m in ) Fig. 7. Peak drift of the MCA166 with detector GL0310 and a gain of 2.8 when exposed to a sudden temperature change of 10°C, with and without stabilization. The period of stabilization is determined for the MCA166 by the counts in the ROI. Default is at the moment a stabilization cycle every 25000 counts, only in newer versions of WinSpec this area can be adjusted. The stabilization cycle time is something which can be optimized. If the stabilization cycle time is too long, then the drift may be faster and may not be fully compensated. If the stabilization cycle time is too short, then the error of centroid evaluation acts as spectrum broadening noise and adds to the FWHM. The centroid evaluation error FWHM (4) N E ⋅ϑ ⋅ N The centroid drift within a stabilization cycle is: ∆E D = E ⋅ ϑ ⋅ t = (5) n depending on FWHM and area N is assumed as: ∆E c = with the gain drift rate ϑ, the stabilisation peak photon energy E and the stabilization peak count rate optimum can be found where the sum of both contributions is minimized. n . An d (∆Ec + ∆E D ) FWHM E ⋅ ϑ (6) = 0 = − 21 + 3 dN n N2 2 FWHM ⋅ n 3 This leads to an area N opt = (7) 2 ⋅ E ⋅ϑ Examples: -For the example of Fig. 7 a maximum drift rate of 4 ppm/s can be assumed. For stabilization a peak at 352 keV with a FWHM of 1 keV and a peak area count rate of 52 cps is used. So the optimum stabilization area is 2 N opt 3 1keV ⋅ 52cps = = 700 −6 2 ⋅ 352 keV ⋅ 4 ⋅ 10 1s This means that a stabilisation cycle here is 13s and a 4% peak broadening due to centroid calculation error has to be accepted. -A HPGe is to be stabilized on the weak K40 1460 keV peak for environmental measurements. FWHM at 1460 keV is 2 keV and a peak count rate of 0.4 cps can be assumed. As drift rate the maximum drift rate immediately after switching on (1.6 ppm/s, same as a temperature change of 5°C) is assumed. Here, an optimum stabilization 2 area of N opt 3 2 keV ⋅ 0.4cps = = 30 . ⋅ 10 −6 1s 2 ⋅ 1460keV ⋅ 16 is calculated. This is near the lower limit for detection of the peak. Areas lower than 30 should not be used at all. Here, the stabilization time is 125s and a peak broadening of 18% has to be accepted. -For U235 enrichment measurement with a NaI, the 186 keV peak used for stabilization. FWHM there is 22 keV, the peak rate is assumed with 1400cps and the NaI is considered ten times less stable than the MCA (16 ppm/s). In this case, the optimum area calculates as: 2 N opt 22 keV ∗1400cps 3 = = 30000 . ⋅ 10 −5 1s 2∗186keV ∗16 The stabilization cycle time is here 21s and the expected peak broadening is neglible (0.6%). The default value for stabilization cycle area of 25000 seems to be rather at the upper end of useful values. Especially with low count rate applications, it is very much recommended to use a newer version of WinSpec and decrease the stabilization area value. C h an n e ls/ke V 7. Linearity and accuracy of gain setting In the scope of this study, also the absolute accuracy of the gain setting was investigated (Do not mix this with the linearity of the ADC!). This measurement was also done with the MCA166 #141 and a GL1015R detector. The absolute value of the scale / gain ratio depends severely on detector type and preamplifier. Even the variations between several detectors of the same type may be large. E.g. the typical variation of signal amplitude for CZT500 detectors from Ritec is about 20% [2]. Compared with that, the gain variation of different MCA166 units (typical 2-3%) can be neglected. 10 1 10 100 G ain settin g Fig. 8. gain setting vs. energy scale using a GL1015 planar HPGe detector S ca le /G a in ra tio (ch a n n e ls/ke V /g a in ) 0 .1 2 8 0.1 2 4 0 .1 2 0 .1 1 6 10 100 G a in Fig. 9. variation of the scale / gain ratio in dependence of the gain setting. The absolute gain setting is correct within +/- 3%. The nonlinearity of the gain shows the same oscillating behavior as the temperature drift coefficient, which shows that the most critical component here is the adjustable gain amplifier. 8. Conclusions -The temperature drift of the MCA166 is a complex oscillating function of the gain setting. Positive as well as negative values for the drift are found. -A positive aspect of this complicated behavior may be that it is possible to find a point with almost no gain drift by varying the gain in a range of only +/- 30%. -The MCA166 temperature drift gets important when doing optimum resolution germanium detector spectrometry. With CdZnTe- and NaI-detectors, the temperature drift of the detectors will dominate. -The thermal drift time constant of the MCA166 is about half an hour. If the measurement time is short compared to this thermal time constant, then resolution will not degrade by drift, but it may be necessary to redo the energy calibration. For long time measurements with best resolution it is recommended to switch on MCA and detector a few hours before starting the measurement. -Peak stabilization can do in some applications a great job of drift compensation, especially when stabilisation parameter settings are optimized. -The MCA166 works principally also at temperatures much deeper (-40°C) than specified; the primary limit is the battery performance at low temperatures. [1] A. Wolf, FZR, private communication [2] Alexander Braun, diploma thesis, dated 16.9.2000; Hochschule für Technik, Wirtschaft und Sozialwesen Zittau/Görlitz (FH) and JRC Ispra. GBS -Elektronik GmbH, Bautzner Landstr. 22, 01454 Großerkmannsdorf / b. Dresden Tel.: (0351) 217007-0 Fax: (0351) 217007-21 homepage: http://www.gbs-elektronik.de/ email: contact@gbs-elektronik.de