"Activity" Question (too much Cs-137)

[isoenzyme]

This is a question and post about determining "activity" from a dose-response curve. I'm hoping that the "Spectrum" board will be an OK place to post this. Since I'm still a new user of the GS-USB-PRO and PRA software the main question I have might already be known to experienced users and gamma ray spectroscopy experts, but I did the following experiments to try to determine how to find the lowest level of activity of an isotope in a sample.

First of all - an admission... I bought WAY too much Cs-137. The photograph below shows my collection (0.01, 0.05, 0.1, 0.25, 0.5, 1, 2 and 5 uCi sealed samples).

Sealed Cs-137 Sources.png (348.77 KiB) Viewed 9036 times

Why I think I have too much Cs can be seen from the following spectra collected for 5 minutes (300 seconds). I wanted to ensure that I had a wide spread of Cs-137 activities (amounts) to provide a good dose-response curve but am learning that one can't measure such a wide spread of activities using identical conditions and times and have the spectra be that useful for analytical purposes.

300 second spectra for 0, 10, 50, 100, 250, 500, 1000, 2000 and 5000 nCi Cs-137 1" sealed sources.

In the previous figure showing all the spectra, the areas totaled are shown with gray shading (since I haven't learned how to do that in PRA it was done in Excel). Plotting the total counts in the shaded area vs. the activity of the calibration sources, and fitting those curves with either a linear (straight-line) and nonlinear (single-phase associative) fit yields the following data:

The results are shown with both standard (linear) and logarithmic (base 10) axes to highlight how the linear fit really stinks at the lower activity levels. This would lead me to think that the nonlinear fit is more appropriate -but this could be wrong!

Averaging five independent spectra for each source for all of my sources (measured using identical GS-USB-PRO volume settings, analysis times and the like) gives the following data:

Dose Response Data.png (4.85 KiB) Viewed 9036 times

Using the modified GS-STANDUP and holder allowed each sample to be presented to the shielded detector in a precise way each time resulting in such great statistics! Plotting the average total counts (in the areas of interest) against the activities resulted in the following figures (plotted with linear and logarithmic axes for my 'old-school' eyes):

Cs-137 Dose Response Curve. (Error bars for each data point are too small to be seen.) These data are fit with a single-phase associative nonlinear model.

In the log-plot the background total counts are shown as the horizontal dash-dot-dash line.

A careful observer will notice that the graphs do not show the same number of points as I've disregarded the data from anything above 500 nCi when calculating the fit. [More accurately, when I let the fancy-schmancy graphing and statistics software calculate the fit.] I don't know if this is appropriate for gamma spectroscopy but I did it for the following reasons:

(1) My years of experience in optical spectroscopy and chromatography tell me that when "peaks" stop looking Gaussian (or Lorentzian) it usually means that detectors are being saturated with signal or columns are overloaded with too much sample; the shapes of the 662 keV peaks in the spectra for the 2 and 5 uCi sources show that this might be the case for the conditions under which the data was collected.

(2) Unless the activity of the calibration sources are evenly spaced the higher concentration samples have a greater influence (weight) on the curve fit than the lower activity standards. Since one of the reasons I ran these tests was to determine how low of an activity I can reliably detect (with statistical significance) I wanted to use as many of the "good looking" spectra without the higher and less-perfect spectra.

Using the statistical fit from the software indicates that the lowest detection limit for these conditions would be around 1-2 nCi (>68% confidence) or around 4 nCi (99.5% confidence). The logarithmic plot using the data from 10 - 500 nCi indicates that these results are not unreasonable. (Sorry, but I'm so old I remember curve fitting with calculators and graph paper so I am still prone to examining plots rather than just believe what some computer program told me the answer is...) I'm assuming that I can go lower (down in the pCi range) if I change the GS-USB-PRO settings and up the count times but I can't talk the guy at Spectrum Techniques to make me a source below 0.01 uCi (yet) so I can test it.

NOW FOR THE NEWBIE QUESTION: Is it already known that activity vs total counts fits a linear model or am I getting a nonlinear model due to my novice conditions? Any wisdom from experienced users or theoretical gurus would be welcome!

[Sesselmann]

Christopher,

Nice detailed report....

For this kind of measurement the count rate is obviously very important, and doing it with relatively slow sound card sampling places some extra restrictions your experiment..

DEAD-TIME
First you need to understand the concept of 'dead-time', it is the time it takes for PRA to sample a pulse, and during this time you can obviously sample only one pulse. Depending your settings the dead-time could be 300 µs at worst or 100 µs at best. With the latest version of PRA we can push it a bit lower, but at the expense of resolution.

By opening the Interval Histogram you will see that PRA actually calculates the dead-time for you, so for this type of experiment you should do a dead-time correction.

\[ ActualCountsPerSecond = cps * \frac{1s}{(1s - deadtime)} \]

REGION OF INTEREST
There are two ways to show the counts within a region of interest.

a) Select the ROI by clicking on a bin (channel) to the left of your peak and press 'b' for beginning, then select a bin to the right of your peak and press 'e' for end, click inside the ROI to see the calculated stats.

b) After activating the advanced filter you can specify the precise energy R.O.I. which effectively removes everything but the ROI in the histograms (note it will not affect your recording)

COUNT RATE
Spectrometry with a sound card can produce excellent resolution, but this comes at the expense of count rate. PRA achieves this with something Marek has called "Pulse Shape Method" , it basically discards badly formed pulses. These badly formed pulses are generally caused by PPU (Pulse Pile Up). All MCA spectrometers suffer from PPU, but it is far more pronounced in long pulse sampling. By setting your tolerance high you can minimise discarded pulses, PRA will show you how many counts have been dropped.

The only way you can minimise PPU is to keep the count rate low and the dead-time short.

It would be useful if you noted the dead-time and the count rate per second along side your charts, as this may give an indication as to why it is non linear.

PS: It is worth learning all the PRA keyboard shortcuts, they are in the manual which downloads with PRA.

Others might have more to add :)

Steven

[isoenzyme]

Steven:
Thanks so much! That was very helpful and it makes sense that reducing dead time would be at the expense of resolution after reading both what you wrote and the PRA pdf manual. (With that fantastic CsI detector you made for me it could be that I can afford to lose some resolution to reduce dead time, especially when the rest of the background is relatively low due to great shielding and the 'purity' of the sample in this case.) I've been using PRA version 20 (1/10/19 update) and, especially after reading what you wrote about PRA22 (https://www.gammaspectacular.com/phpBB3 ... ?f=7&t=495), it makes me want to try it with the much higher sampling rate. Off the top of your head do you know if the GS-USB-PRO will handle the 384 kHz sampling rate? (My owners' manual instructs me to use the 96 kHz rate.)

[Sesselmann]

Christopher,

What goes on between the audio codec (ADC) and the computer is, as far as I can tell, "black magic". The ADC in the USB-PRO is technically 48 kHz, which normally leaves you with a pretty jagged looking pulse, but when we oversample it by using 96, 192 or even 384 kHz, we don't see the extra samples dotted along straight lines, they miraculously fall on the curve.

So for whatever reason we get better results with oversampling, but that said, it does not mean that we can shorten the pulse right down, because the input is still 48 kHz so when our input pulse is too short we fall below the Nyquist sample rate and our pulses do not get sampled.

At 48 kHz the analogue signal gets sampled every 20 µs so to get a reasonable shape you need at least 5 samples or 100 µs pulse, which determines the dead-time.

Faster sampling is the ultimate goal, but so far I haven't found a common USB audio codec with faster than 48 kHz rate, scientists determined that the human ear can't hear above this frequency, so with few exceptions 48kHz is the norm.

I am always on the lookout for a faster CODEC so if anyone knows of one, please let me know.

Steven

[Sebastien_billard]

Professional souncard are often 24 bits and 96 kHz or 192 kHz so codec should exist ?

[Sesselmann]

Sebastian,

Yes, they certainly exist, but most of them like the PCM3060 (24/96) are not usb compatible, they have a serial output, which makes the board a little bit more complicated.

I need to build up a bit of confidence before I start designing a digital circuit on a high voltage mixed board ;)

Steven

[Peter-1]

Hello,
I made a quick test with Pechblende and 2 different counting rates. Green is normal < 500 cps and red is over 1000 cps. I use a soundcard with 192 khz. A clear difference. A good example for pileup.

Peter

[Sesselmann]

Peter,

That looks quite dramatic...

I doubt PRA would behave like this, because it filters out the badly formed pulses, so I suspect it would start dropping a lot of pulses as the count rate goes up.

You should be able to handle 2000 cps at 92 kHz.

Steven