The G1 Assessments - CompAct
This page reports an assessment of the results generated by Gordon Gilmore's spectrum analysis program CompAct in anlysing the IAEA G1 test spectra.The assessment is particularly favourable in terms of peak area measurement performance and it's least squares fitting routine. It fails, of cours, in automatic doublet deconvolution, for which it is not designed. If nothing else, this page provides an example of a rigorous assessment.
The IAEA G1 Spectra, and the justification for the indexes, are described in full in Practical Gamma-Ray Spectrometry (Section 15.5.2 in the Second Edition).
The spectra comprise:
- G1100 - Reference Spectrum containing 20 well defined peaks
- G1200 - Peak Search Test Spectrum containing 22 very small peaks on high and low backgrounds.
- G1300 to G1305 - Consistency Test Spectra, each containing 22 easily detectable peaks of
varying size. Apart from differing statistical scatter the peaks are identical in all six spectra.
- G1400 - Doublet Fit Test containing 9 doublets.
The peaks in the various spectra were all produced by shifting the peaks in G1100 by integral
numbers of channels and attenuating by known factors. The position and ratio of all the
peaks is known, therefore, relative to the G1100 spectrum.
a) All spectra were searched using sensitivity 0.91 and the peak areas estimated using
automatic background width selection.
b) G1200 peaks were not all found in the peak search. All peaks were measured using the
manual interactive peak fitting process. The G1400 doublets were
not resolved by the search process, these were also measured using the fitting procedure.
Full results can be obtained from Nuclear Training Services Ltd. Alternatively use
CompAct itself to analyse the G1 Spectra which are provided with the
program.
Results and Commentary
|
G1 Intercomparison Spectra - Summary Sheet |
Program Tested:
Sreadsheet Filename:
Version:
Analysis Conditions:
Backgd. Width:
|
CompAct
G1COMPAC.WB1
2.1
Batch search, Sensitivity: 0.91
Auto
|
| Spectrum |
Number of Peaks |
Search
Index |
Peak Position |
Normalised Area |
R
Factor |
| Expected |
Found |
Spurious |
Error |
± |
Std.Dev |
Error |
± |
Std.Dev |
|
| Targets: |
1.00 |
0.000 |
± |
0.000 |
0.00 |
± |
1.00 |
1.00 |
|
|
| G1200 |
22 |
17 |
0 |
0.77 |
0.117 |
± |
0.274 |
0.37 |
± |
0.72 |
1.17 |
| G1200Fitted |
22 |
22 |
0 |
1.00 |
0.000 |
± |
0.000 |
-0.05 |
± |
0.93 |
0.72 |
NOTES:
The peak search routine, in common with most other programs, misses the smallest peaks.
17 peaks detected is good performance, particularly as no spurious peaks were detected. Two
more peaks can be detected by increasing sensitivity but then 4 spurious peaks are detected. The
R Factor indicates that all the 'errors' are atributable to statistical factors alone.
The large edge around channel 1024 is detected as an 'odd' peak. This edge is unreal and the peak has been
ignored for the purpose of these tests.
Fitting the peaks gave a significant area for all peaks in spite of the fact that the smaller
peaks are not clearly identifiable by eye. The accuracy of the peak area estimations is an improvement
on the simple ROI estimations and is very
satisfactory. Since the peak positions are supplied to the fitting process the position error
is zero.
|
|
| Consistency Tests 1-6 |
| G1300 |
22 |
21 |
0 |
0.95 |
-0.048 |
± |
0.139 |
0.11 |
± |
1.16 |
0.58 |
| G1301 |
22 |
21 |
0 |
0.95 |
0.005 |
± |
0.098 |
0.17 |
± |
0.84 |
1.55 |
| G1302 |
22 |
21 |
0 |
0.95 |
0.036 |
± |
0.145 |
0.47 |
± |
0.96 |
2.92 |
| G1303 |
22 |
21 |
0 |
0.95 |
0.025 |
± |
0.114 |
0.10 |
± |
1.12 |
1.42 |
| G1304 |
22 |
21 |
1 |
0.91 |
0.021 |
± |
0.087 |
0.13 |
± |
1.30 |
1.02 |
| G1305 |
22 |
21 |
0 |
0.95 |
0.040 |
± |
0.172 |
0.33 |
± |
1.00 |
1.30 |
| Overall |
132 |
126 |
1 |
0.95 |
0.012 |
± |
0.127 |
0.22 |
± |
1.06 |
1.52 |
NOTES:
Peak position accuracy is within 0.05 channels and the precision of the position estimation better
than 0.2 channels. The acuracy of the area measurement is good.
All peaks were, in fact, detected with only 1 spurious peak in the whole set of spectra.
The problem peak was that at 1010 which lies at the at the top of the curve of the edge feature
around channel 1024. The automatically selected ROIs are clearly in error and areas, even with
manually set ROIs, are consistently high. Taking into account the shape of the continuum
underlying the peak this is not unexpected. This peak has been omitted from the analysis.
Taking that into account the overall peak search index would be 0.98.
In all but one spectrum (G1302) the R Factor indicates that area errors can be attributed to
statistical scatter alone.
|
|
| Doublet Resolution Test |
| G1400Fitted |
18 |
18 |
0 |
1.00 |
-0.101 |
± |
0.036 |
0.21 |
± |
0.80 |
0.53 |
NOTES:
The peak search routine does not, and is not designed to, resolve peak doublets as close
the peaks encountered here. Three of the doublets have peak only 1 channel apart, another
three 3 with peaks 3 channels apart (a little less than 1 FWHM) and the remaining three with
components 6 channels apart. In all cases the area accuracy was good; in most cases the
area was within one standard deviation of the correct area. This is reflected in the
low area error indexes and the low R Factor.
|
|
| Chi-squared table |
| Range |
Found |
Expected |
| <-2 |
3 |
3 |
| -2<<-1 |
14 |
17 |
| -1<<0 |
34 |
43 |
| 0<<1 |
45 |
43 |
| 1<<2 |
25 |
17 |
| >2 |
5 |
3 |
| Total |
126 |
126 |
Chi-Squared: 7.60
Deg. of Freedom: 5
R Factor: 1.52
|
|
| The Chi-squared table shows the number of area results found to be within different
standard deviation bands. (For example, the first line looks at the number of results
more than two standard deviations below the true value.) The actual numbers found is
compared to the number statistically expected.
The comparison shows that the distribution of errors is symmetrical about zero, i.e. there
is no bias on the area estimates, and the numbers within the various bands compares well
with expectation. The chi-squared factor of 7.6 for 5 degrees of freedom is well within
statistical expectation.
|
|