Spectrum Recorder PCI-E Card?

This area is for discussions involving any fusion related radiation metrology issues. Neutrons are the key signature of fusion, but other radiations are of interest to the amateur fusioneer as well.
User avatar
Rich Feldman
Posts: 1109
Joined: Mon Dec 21, 2009 11:59 pm
Real name: Rich Feldman
Location: Santa Clara County, CA, USA

Re: Spectrum Recorder PCI-E Card?

Post by Rich Feldman » Tue Jan 29, 2019 9:33 pm

>> I see,measurement technology is unbelievable expensive in comprehension to other electronic equipment.

Did you overlook digital multimeters that cost less than $5, and are often given away free for buying something at a Harbor Freight store?
Light bulbs and magnetrons are relatively inexpensive, because the technology is mature and millions are made every day.
Same goes for smart phones and large TV sets. Technically not too challenging, and the makers get revenue from tracking you long after the original purchase.
Super digitizers, like you originally asked for, are expensive because they are technically cutting-edge, and the quantities are too low to justify high volume production methods and cost reduction. I bet the $3500 IC's that Chris pointed to don't use a plain old xx-nanometer digital CMOS process, on plain silicon, through mainstream 12-inch wafer fab lines.

First exercise for you, Harald:
Find out and explain the meaning of 6 GHz and 20 Msps in the same software-defined radio spec. Front-end downconverter? Undersampling, with sampler aperture size and jitter very small compared to repetition rate?

As for how to separate nearly-coincident pairs of 20 ns analog-voltage pulses,
here are some guesses (without research) about what might be good enough for your interest.
How about one good sample every 2 nanoseconds? (500 Msps) You know the pulse shape, and should be able to fit the data samples no matter how the sample clock phase is aligned to any given pulse location.
Analog 3db bandwidth 200 MHz? (sub 2 ns risetime)
Digitized to render 10 or 12 bit numbers for your largest expected coincident-pulse voltage?
I think this is getting into mainstream oscilloscope territory, except maybe for the bit count.
Another critical consideration is voltage noise.

What fraction of the pulses will overlap, if they have ordinary random statistics and the average rate is 60,000 cpm?
Mike echo oscar whisky! I repeat! Mike echo oscar whisky, how do you copy? Over.

User avatar
Steven Sesselmann
Posts: 2108
Joined: Thu Aug 11, 2005 1:50 am
Real name: Steven Sesselmann
Location: Sydney - Australia
Contact:

Re: Spectrum Recorder PCI-E Card?

Post by Steven Sesselmann » Wed Jan 30, 2019 3:36 am

Harald_Consul wrote:
Tue Jan 29, 2019 2:33 pm
Meanwhile I have figured out, that the signal amplitude of a (negative voltage) scintillation tube typically is in the range of
~20 mV with a pulse length of ~20 ns.
What exactly are you trying to achieve with GHz sampling?

Can you amplify and stretch the pulse so you can sample at a lower rate?

This is how I achieve pulse height analysis from PMT's with only 48 Khz sampling and a few software tricks.

Steven
http://www.gammaspectacular.com - Gamma Spectrometry Systems
https://www.researchgate.net/profile/Steven_Sesselmann - Various papers and patents on RG

Harald_Consul
Posts: 96
Joined: Sat Oct 20, 2018 10:01 am
Real name: Harald Consul

Re: Spectrum Recorder PCI-E Card?

Post by Harald_Consul » Wed Jan 30, 2019 1:48 pm

Well, yes I thought it would require 100 or maybe even 1000 samples to disassemble a multiple signal voltage amplitude (of 20 ns in this case) into its originals properly. This is to some degree also a statistical/ mathematical question I have not researched fully, this time.

However I can say,
  • the more single event amplitudes are contained in a cumulated measured amplitude,
  • the more easily the amplitudes of two or more low energy events do exactly pile up to the amplitude of a higher energy event and
  • the more individual the curvature of each single event amplitude is
the (much) more samples will be required to properly decompose a multiple signal into its origins.

List above last edited on Thu 2019-01-31 12:06 pm

Richard, I will "do my homework" and come back.

Steven, is there a turn key solution for pulse stretching?
Last edited by Harald_Consul on Thu Jan 31, 2019 10:12 am, edited 4 times in total.

John Myers
Posts: 66
Joined: Wed Mar 08, 2017 12:13 am
Real name:
Location: SoCal

Re: Spectrum Recorder PCI-E Card?

Post by John Myers » Wed Jan 30, 2019 10:29 pm

A rule of thumb is to sample at a rate of 2.5 times the frequency of the signal. In order to avoid aliasing you need to sample above the Nyquist rate (2x the freq).
So to accurately reproduce a 20ns pulse you would sample at 125Msps.

It may be possible to use a slower rate (undersampling) if there were no other longer pulses that would alias with the faster ones.

Andrew Seltzman
Posts: 717
Joined: Mon Feb 02, 2004 1:02 am
Real name: Andrew Seltzman
Contact:

Re: Spectrum Recorder PCI-E Card?

Post by Andrew Seltzman » Thu Jan 31, 2019 12:20 am

Some things to think about:

You are trying to digitize a stochastic, non-repetitive signal. Frequency domain methods (eg a spectrum analyzer) will not be of much use.
The Nyquist rate is the BARE MINIMUM frequency that is theoretically required to reconstruct a REPETITIVE signal.
20 points per period of a sine wave will give you nice looking plot.
10 points per pulse will generally be ok to fit a gaussian pulse nicely if you have a shaping amplifier.
If you don't have a shaping amplifier, the best you can localize the time location of the edge is on the order of sampling period of the digitizer.
In this case, if you are looking to capture pulse height, capturing the peak value with an analog peak detector circuit and then using a slower ADC will be much more useful. As Steven said, using a shaping amplifier to slow down the pulse is another valid alternative.

Also look into something called a "baseline restoration circuit" which can help at high count rates.
Andrew Seltzman
www.rtftechnologies.org

User avatar
Steven Sesselmann
Posts: 2108
Joined: Thu Aug 11, 2005 1:50 am
Real name: Steven Sesselmann
Location: Sydney - Australia
Contact:

Re: Spectrum Recorder PCI-E Card?

Post by Steven Sesselmann » Thu Jan 31, 2019 12:48 am

Harald_Consul wrote:
Wed Jan 30, 2019 1:48 pm
Steven, is there a turn key solution for pulse stretching?
What I use is just a fast rail to rail opamp as a current amplifier with a high impedance drain, so when a pulse comes in via the coupling, it takes time to filter through.

Obviously this only works when the pulse count rate is slow, otherwise you get pulse pile up (PPU).

Thousands of people around the world are now doing gamma spectrometry with sound card sampling which everyone here said would be impossible when it was first suggested 10 years ago.

Steven
http://www.gammaspectacular.com - Gamma Spectrometry Systems
https://www.researchgate.net/profile/Steven_Sesselmann - Various papers and patents on RG

User avatar
Rich Feldman
Posts: 1109
Joined: Mon Dec 21, 2009 11:59 pm
Real name: Rich Feldman
Location: Santa Clara County, CA, USA

Re: Spectrum Recorder PCI-E Card?

Post by Rich Feldman » Thu Jan 31, 2019 6:57 am

Bit of clarification about continuous-time and discrete-time functions, and the Nyquist rate.

Consider a continuous-time function x(t) that has the important property of being bandlimited. It has no frequency content at or above the limiting frequency Fb. Periodicity doesn't matter. Frequencies that we don't care about do matter.

Suppose we generate a discrete time signal X(n) by sampling x(t) at regular intervals -- times separated by exactly 1/2Fb.
Then x(t) can be perfectly reconstructed from X(n)
, using the interpolation method called sin(x)/x or "sinc" or Whittaker-Shannon.
https://en.wikipedia.org/wiki/Whittaker ... on_formula

Digital storage oscilloscopes often support horizontal scale factors that have more pixels than samples. Then they render the waveforms using sin(x)/x interpolation, which is not fakery. It actually eliminates real errors that would appear if they drew straight lines to connect the sample dots.

Of course the key to lossless sampling is having a bandlimited signal to start with.
If you start with an arbitrary function y(t), you can generate x(t) with a low-pass filter ("anti aliasing filter") that passes no frequencies at or above Fb. Information lost at that point is gone. Whatever gets through into x(t), there's no fidelity benefit from using a sample rate higher than 2Fb.
In practice, designers need some room for the anti-alising filter to roll off between 100% passing with constant delay (in frequency range of interest) to 100% stopping (at and above half the sampling rate).
Mike echo oscar whisky! I repeat! Mike echo oscar whisky, how do you copy? Over.

Harald_Consul
Posts: 96
Joined: Sat Oct 20, 2018 10:01 am
Real name: Harald Consul

Re: Spectrum Recorder PCI-E Card?

Post by Harald_Consul » Thu Jan 31, 2019 10:15 am

For my understanding. Do you or don't you agree with:
Harald_Consul wrote:
Wed Jan 30, 2019 1:48 pm
However I can say,
  • the more single event amplitudes are contained in a cumulated measured amplitude,
  • the more easily the amplitudes of two or more low energy events do exactly pile up to the amplitude of a higher energy event and
  • the more individual the curvature of each single event amplitude is
the (much) more samples will be required to properly decompose a multiple signal into its origins.

List above last edited on Thu 2019-01-31 12:06 pm
?

Especially did I forget a characteristic of the mathematical problem, which determines the necessary number of samples (data points) to decompose properly?

Further questions to the mathematical problem:
  • Is the amplitude/duration relation among the amplitudes of singular high energy particles and singular low energy particles approximately constant?

Intermediate result of mathematical part of the problem:

From the first research into this topic it looks like mixed distribution models aka mixture models from the free statistical software R might be suitable to decompose complex amplitudes.

User avatar
Rich Feldman
Posts: 1109
Joined: Mon Dec 21, 2009 11:59 pm
Real name: Rich Feldman
Location: Santa Clara County, CA, USA

Re: Spectrum Recorder PCI-E Card?

Post by Rich Feldman » Thu Jan 31, 2019 5:16 pm

Harald,
Some of your particle detector questions need to be answered before we can properly tackle the pile-up problem.
If you can't find them by reading, or asking on forums, how about finding them by experiment?

First please help your helpers stay engaged.
* Clarify whether the goal is real-world spectrum measurements by you, or thought experiments, or something else.
* What is the maximum average count rate you need to deal with?
* What's the approximate numerical probability that any event will have another event close enough to make a combined pulse?
* How low must the probability be, to not interfere with your application?

* Why don't you reduce the overlap probability by using a detector that makes shorter pulses, as suggested by others in this thread? Or by moving the detector to reduce the maximum count rate?

Before speculating on samplers, you need to declare your expectations about the PMT output when mixed radiation is present.
* What is the pulse waveform for one representative particle? Please draw or point to a picture, with x-axis time units given.
* What is the range of pulse shapes for particles of the same energy, interacting at different places within the scintillator?
* How about the range of pulse shapes for all particles of a different energy? Repeat for other energies in the range of interest.
* Do you trust that the detector system is linear? That means that when two events happen at about the same time, their respective output voltages for each instant are simply added.

Each pulse shape (the real continuous-time waveform) has a frequency content, and that drives the required sampling rate.
If the typical pulses are smooth and well-behaved, they may be naturally sort of bandlimited. If faster-edge risetime were on the order of 2 ns, that suggests not much going on above 200 MHz, and high fidelity sampling might be done at 500 Msps.

The key to reconstruction (to get pulse locations much more precisely than the sample interval, or to resolve overlapping pulses)
is to do the interpolation properly. Connecting sample dots with straight lines is simplistic, and significantly wrong, unless you oversample by an unnecessarily large ratio. Fine if oversampling is cheap. Foolish if your money is better spent on good samples at a lower rate, with proper front end filter, as found in practical oscilloscopes.
Mike echo oscar whisky! I repeat! Mike echo oscar whisky, how do you copy? Over.

User avatar
Richard Hull
Moderator
Posts: 11348
Joined: Fri Jun 15, 2001 1:44 pm
Real name: Richard Hull

Re: Spectrum Recorder PCI-E Card?

Post by Richard Hull » Thu Jan 31, 2019 7:14 pm

Rich makes a fabulous point. GM counters have a dead time that must be accounted for with 60 usec being one of the best and 200 usec being the norm in many older tubes. True random events like radioactivity have issues with detection just not being counted due to a GM tube's dead time.

Moving the detector farther away with a GM set up might not allow for the inverse square law to compute back to the actual radiation emission rate!
This is especially true with mica windowed tubes as a tremendous amount of alphas just will not be counted due to their MFP in air! This can amount to, often, 50% of the total radiation. Add to this, the fact that many low energy betas can scatter before being counted in a more distant mica windowed detector.

For neutron tubes, especially like the 3He tubes and PMT gamma detectors, the pulses are very short compared to the gas amplification scenario within a GM detector. Moving a neutron detector can result in fairly accurate back figuring using the inverse square law provided you are not so far away from the source that scattering and reflections from nearby moderators do not significantly interfere.

It all comes back to......

1. Knowing what you are measuring and all of its oddball characteristics and gotchas.
2. Knowing what you are measuring with (detector and all of its electronics), and its limitations and gotchas.
3. Trying like the devil to stay out of statistics that grasp at straws where the statistics becomes like a drunk at a lamp post at night....Using it more for support that illumination.

Rich touched on all of this in a wise and steady manner above.

Richard Hull
Progress may have been a good thing once, but it just went on too long. - Yogi Berra
Fusion is the energy of the future....and it always will be
Retired now...Doing only what I want and not what I should...every day is a saturday.

Post Reply