I'm a little behind in the discussion since H-M was down for a bit. Anyway, to comment on some of the recent discussion points....
1. A/D resolution and speed. My comment here is based on reading through the Theremino gamma spectrometer information. They are using the 16 bit A/D on a USB sound card, so it's limited to a maximum sampling rate of 192KSPS. That's not nearly fast enough to capture the pulses coming out of a PMT, so they are stretching the pulses using a simple 2-pole low pass filter, then amplifying by 100 to get the pulse amplitude back high enough to get decent numbers out of the A/D. This is a "nice" approach because it permits the use of _very_ inexpensive A/D's; and the drivers are generally included with the computer OS (the Teensy approach would use the provided audio library to accomplish the same thing). The LP filters also improve the SNR which helps improve the resolution of the multi channel analyzer.
The cheap A/D kit from PJRC can be found
here
The Theremino group also has some information regarding the energy resolution (in FWHM) vs A/D bits. You really do want at least 14 bits. I believe more bits is better due to the method they use to find the pulse peak in the presence of noise and the inevitable misalignment of sample time relative to the peak -- it is highly unlikely you will actually sample the peak, so some form of curve fit or interpolation is needed. The energy resolution is completely dependent on how accurately the peak voltage can be measured. You'd then conclude that one of the fancier audio A/D's that can get to 24 bits would be better yet: but in fact those A/D's really don't give you true 24 bit resolution. From what I've read on the web about this, the lower 4 bits are basically noise. So 16 bits is the sweet spot in terms of "real" system performance.
I had mentioned in post #125 using a low-pass filter to "stretch" the pulse event, and I called it "smearing". I was maybe too hasty. It could be that all it loses is mostly noise, and leaves an easier pulse to capture.
Thanks for the link to the PJRC A/D converters kit.
Forgive that I still miss on this, but the $2.65 PT8211 Audio Kit the link leads to is a
DAC kit for turning number streams into sound waveforms. I am thinking this is not what we mean.
16 Bits.
I have been looking through A/D converters, considering 14-bit and 16-bit candidates at rates up to 40MSPS.
These would be Texas Instrument, Analog Devices, and Linear Technology.
I have decided it has to be 16 bits. Also, I go for the highest sample rate possible for about £20 for the device alone.
For the 16 bits, at sample rates like 2MSPS or 5MSPS or even 10MSPS or so, come in at near the cost I was aiming for. Many can be had on a evaluation board, or evaluation kit. That costs a bit more, but I see this as a part as vital as the scintillator, or PMT. Going on the information from
@RJSakowski in post #140, I am thinking the response scintillations may come at perhaps some kHz.
Other thoughts
Some A/D chips are 2-channel simultaneous sampling. I thought having two samples is a way to a fast first average.
If we use the known fastest rise time of the scintillator crystal, and the frequency response of the detector PMT or SiPM, then there is no harm in setting an input low-pass filter to something slightly above that, because we know what it cuts is noise. A low noise gain stage as part of the filter is a good idea, to set the A/D count range suitably near maximum.
Dolby-style noise removal (nonlinear gain to expand the low levels, filter, then compress back is not necessary if one has the sample numbers. If the genuine peaks are bouncing up and down from noise, or living in the middle of independent scatter noise counts, then replacing some least significant bits with a single set at some threshold level strips out those not making a valid contribution to the peaks. A vicious, crude, but effective digital noise filter.
I am still deciding the line between maximizing the effectiveness if the gadget, and the cost.
I still try for something that uses a USB link to a phone app, with some tiny thing like a Teesy 4.0, Arduino, PIC, or a Raspberry Pi Compute Module (like a Teensy, but 64-bit Coretex-A53, 1.2GHz, 1GB SDRAM, and 32GB Flash).
That last one is £30.90 in UK