The I.A.E.A G1 test spectra

This is a set of nine spectra distributed originally in 1976 as an intercomparison exercise. (Reported IAEA 1979) The spectra are public domain and can be copied freely. The spectra were prepared by measuring pure radionuclides with high precision - just below 1 million counts in the highest channel of each spectrum - using a detector of, what was at the time, 'average' performance:

60 cc Ge(Li) detector of resolution
2.8 keV FWHM at 1332 keV and a
Peak-to-Compton ratio of 40:1.

It was noted that peaks had a slight asymmetry. Spectra were recorded in 2000 channels with gain of about 0.5 keV/channel giving a range of about 1000 keV. One FWHM is equivalent to 5.6 channels at 1332 keV.

The task is to find all the true peaks in each spectrum, with no spurious detections, and then measure the peak area, after deconvolution if necessary. The positions and peak areas can then be compared with the expected values. A spreadsheet to allow that comparison and calculate performance indexes is available.

To download and save the entire .zip files, Right-Click on the link and select 'Save Targe File...'.

The spectra were smoothed slightly to suppress residual statistical uncertainty and then combined in various ways after shifting by whole numbers of channels and dividing each channel count by a known factor. The reason for shifting peaks was originally to provide peaks at positions unknown to the participants in the original intercomparison exercise. In order to allow programs to be tested in the presence of high and low peak backgrounds about 10000 counts per channel was added to the bottom 1000 channels and about 200 counts per channel to the top 1000 channels. The 'join' between these two regions was adjusted to simulate a Compton edge. (It must be said that this is the one unsatisfactory aspect of these spectra in that this pseudo-Compton edge is much larger that encountered in real spectra and many analysis programs produce spurious peaks at this position.) The spectra as constructed at this point had peaks in known positions and of known ratios to their reference peaks but were essentially 'noiseless'. The spectra were then subjected to a random number generator process to simulate the effects of Poisson distributed counting statistics. While this process will alter individual channels it will not alter the position of the peaks and should not alter the number of counts within the peaks.

Go Back to Top

G1100 - Reference Spectrum

Contains 20 peaks to be regarded as independent nuclides. The spectrum is the sum of the complete spectra of all 20 pseudo-nuclides over the whole energy range. Each peak contains near to 65,000 counts and can be measured with an uncertainty of about 0.4%.
G1100 Spectrum

G1200 - Peak Search Test Spectrum

Contains 22 peaks all shifted and attenuated relative to G1100. Many of these are difficult to detect and measure. This spectrum also serves as a very good test of the ability of programs to measure the area of poorly defined peaks. Experience has shown that many programs can detect 17 peaks without reporting spurious peaks. The highest number of real peaks detectable is, perhaps, 19 at the expense of reporting spurious peaks. The pseudo-Compton edge in this spectrum seems to cause problems with many programs and, as this is an unreal feature, any peaks reported in the range 1020 to 1030 keV should be ignored for the purposes of assessment. (Nevertheless, it is instructive to examine the way in which the program handles this feature as an indicator of what could happen, perhaps in a less extreme fashion, to real Compton edges.)
G1100 Spectrum

Go Back to Top

G1300 to G1305 - Consistency Test Spectra

These six spectra each contain 22 peaks and are derived from the same prototype spectrum but subjected to the 'noise generation' process separately. The idea of these spectra is to check the consistency of analysis of spectra which are identical except for the statistical scatter. Twenty peaks are in the same position as the G1100 reference peaks but attenuated. Another two peaks are attenuated and shifted. One notable feature of this set of spectra is that the peak at channel 1010.6 is near to the pseudo-Compton edge and, because of this, the area measurement may be affected by the downward slope at the high energy side of the spectrum.
G1100 Spectrum

Go Back to Top

G1400 - Deconvolution Test Spectrum

This spectrum contains nine well defined doublet peaks formed by shifting and attenuating peaks from G1100. The peak separations are 1, 3 and 6 channels and the peak ratios are 10:1, 3:1 and 1:1. These separations represent about 0.2, 0.5 and 1 FWHM.
G1100 Spectrum

Go Back to Top

Objections to the I.A.E.A. G1 spectra

It should be pointed out that there are objections to the continued use of these spectra in some quarters. These objections revolve around the fact that the spectra are of only 2000 channels, are 'computer generated' and are measured on an old detector. In rebuttal of these objections the following observations can be made:

The fact that only 2000 channels are available is irrelevant, the spectra can be regarded as a 2000 channel slice of a 4096 channel spectrum - or 8192 or 16384 channel spectra for that matter. What matters is the number of channels within a peak. These spectra were recorded at about 0.5 keV/channel and this is, in many respects, optimal for best peak area measurement precision.

The argument that because the spectra have been 'computer generated' they are not relevant is not valid. The shapes of the peaks have not been altered by the mathematical manipulation. In fact, the correspondence of peak shape between standard and sample spectra may be better than in an actual measurement set where differences in count rate could, in principle at least, cause peak shapes to depend to a small extent on count rate.

The only real problem may lie in the fact that the peaks in the IAEA are slightly asymmetric. In practice any detector system has the potential to produce spectra where the peaks are not ideal. For example, one would continue to use a detector after sustaining slight neutron damage and one would hope that the analysis software would cope. In fact, the peaks in the G1 spectra are not grossly asymmetric and are close to a pure Gaussian with an underlying step function.

References

The I.A.E.A. intercomparison of methods of processing Ge(Li) gamma-ray spectra - preliminary report, in 'Computers in Activation Analysis and Gamma Ray Spectrometry. I.A.E.A. CONF-780421, Ed. Carpenter et al.,
Parr, R.M., Houtermans, H., Schaerf, J., (1979)
(This document now seems to be unavailable.)

Additional results for the 'G-1' IAEA intercomparison of methods for processing Ge(Li) gamma-ray spectra.
Zagyvai, P., Parr, R.M., Nagy, L.G., (1985), J.Radioanalyt.Nucl.Chem 89,589-607.

Quantitative evaluation of gamma-spectrum analysis methods using IAEA test spectra.
Nielsen, S.P., Nucl. Instrum. Methods 192 (1982) 433-438.

This website was created by Dr Gordon Gilmore for the benefit of readers of Practical Gamma-Ray Spectrometry (3rd Edition).

Problems with this page? Report it on the Feedback page.