Compton damping

[RobBryant]

Hi.. I thought this might be interesting on Compton scattering suppression
UK research.. perhaps the neural learning might be expensive..
Machine Learning Based Compton Suppression for Nuclear Fusion Plasma Diagnostics https://www.researchgate.net/publicatio ... aFVHamM%3D
regards
Robert

[Sesselmann]

Robert,

Interesting approach, but I am not sure if machine learning is the right tool to remove Compton scatter. At best it can imitate reality which doesn't actually give us any more information about reality. Similar to how photoshop can remove something is a photo and fill in the background, it tells us nothing about what actually is behind the item that was removed.

Steven

[jneilson]

Aha! This is actually the research of one of my colleagues. Our lab has quite an interest in Compton suppression techniques - one of our main HPGe systems utilises an anticoincidence veto Compton Suppression System, but we're also interested in Compton suppression by Pulse Shape Discrimination, which this work uses machine learning to try to achieve. PSD and other discrimination tasks are something machine learning is in fact quite well suited to.

Steven, indeed, there's no new information being inferred by the ML, but just being able to throw away some of the Compton noise still allows the real spectrum data to stand out more and not be lost in noise. So far the ML PSD approach seems to be equally effective as the anticoncidence veto system, although I suspect a combination of both approaches may do even better

[Sesselmann]

Robert,

The interesting thing about Compton scatter is that every scattered count actually belongs in the main photopeak. What I found is that the product of energy and counts in the Compton plateau appear to be a constant.

Based on this idea I attempted writing a script to put the counts back where they belong. Follow the link below to my old post....

viewtopic.php?f=5&t=1072

[jneilson]

The interesting thing about Compton scatter is that every scattered count actually belongs in the main photopeak.
...
Based on this idea I attempted writing a script to put the counts back where they belong.

Indeed - there's actually been quite a lot of work around this idea. Some of the keywords to look for are Compton Addback and Compton Unfolding. It's a bit more involved than your script though - I think you're assuming there's only a single photopeak - that assumption might work for a single Cs-137 source, but it's not going to hold for a more complex spectrum, especially one where some of the photopeaks that contribute to the Compton might not be visible in the spectrum. The idea of unfolding is exactly what you're trying to do though - take those counts and return them the photopeak they belong to. With a good understanding of what the contribution to the continuum shape is from any given photopeak energy, algorithms like Maximum Likelihood Analysis can do a great job of figuring out which parts of the continuum belong to which photopeaks.

Actually, in some types of detectors, especially organic scintillators, you don't really see photopeaks so you need to rely on ideas like unfolding or looking at Compton edges to determine what peaks you really have.

What I found is that the product of energy and counts in the Compton plateau appear to be a constant.
...[snip]...
Follow the link below to my old post....

viewtopic.php?f=5&t=1072

Yes, I've been meaning to have a proper look at that topic when I get a bit more time. I'm not particularly convinced by your hypothesis; certainly not the way you've arrived at it - I think what you're seeing in terms of a constant energy-count product is more a coincidence based on the type of detector you're using and probably has more to do with detection efficiency and factors like that rather than being anything fundamental about Compton effects. I work with several detectors that exhibit quite different Compton shapes - I can think of one detector optimised for lower energy measurements that the Compton is pretty flat in channel-space, so I doubt it is still flat when you do the energy product. I'll have to see if I can grab some spectra and do the multiplication to demonstrate at some point, but I think you're barking up the wrong tree.

We actually have a good explanation of the intensity of Compton scatter as a function of angle in the Klein-Nishina formula - so understanding what Compton actually does ought to just be a case of combining Klein-Nishina with the Compton energy loss formula to get out what histogram you ought to see with an ideal detector - I doubt it's flat in energy-product space but when I get some time I'll do the maths and check.

[Sesselmann]

Joe,

You are right, it becomes a lot more complicated when there are more peaks, but for demonstration purposes Cs137 was nice and easy.

My general thought about Compton scatter being energy constant was based on Compton shine being isotropic. If we think of the exited crystal as being hot body, it seems reasonable that energy would flow outwards isotropically and further that the higher energy shorter wavelength photons carry the heat away faster and the lower energy photons carry the heat away slower and therefore require more photons to exhibit the same cooling.

So again in the case of Cs137 662 keV is dumped into the crystal, this "heat" needs to dissipate and go somewhere, it either gets fully absorbed, in. which case it all becomes light and results in a 662 keV count or it scatters and gets partially absorbed.

When looking at a single event it does not look like heat, but when we look at millions of counts it becomes statistically equivalent to an isotropic heat source.

[jneilson]

Steven,
Just a quick counterexample to your flat energy-product spectrum: