Needing more than a spark test?

[graham-xrf]

I'm throwing this out there to suggest that there may be different ways to look at the detector circuit, and (perhaps) allow the use of parts that have a lower GBW compared to some high GBW parts that seem to be in short supply.

Admittedly, this particular implementation is aimed at the Theremino approach where the pulse is heavily filtered: but in this case the peak voltage is pretty much determined by the charge coming out of the detector divided by the load capacitor. No ambiguity regarding the relationship between the current coming out of the detector and the voltage coming out of the amplifier chain.

I can't say this is a perfect solution because there still is the possibility of pulse overlap causing problems but that's going to be an issue regardless of what the front end looks like.

Mark - I guess you must be psychic or something..
Thanks for the new analysis.
You join the party. Three of us now posting LTSpice plots!
If it's OK, could you post the .asc here?

The great pulse you showed us on Page 62, #617 has about a volt amplitude, so I am guessing it had the benefit of extra gain from your signal conditioning board. Was that amplified from the U1B output at R4?

 

[graham-xrf]

How others do it (xrf), and some fundamental stuff + hard sums
Like Mark, I am in review mode. I have be trawling everything, all the way back to Theremino Mca. At the end, perhaps a basic consequence to our computing.

Theremino is, of course, a general purpose plot display program written for Windows, with interface kit described as "modular" It is supposed to give visuals to all sorts of projects from keyboard music, atmospheric monitoring, data loggers, stepper motors, etc. The "modules" are such a close clone of Arduino, it can be completely misleading. Maybe, back then, they started out playing with Arduinos.

"Theremino" name alludes to that (ineffective) music maker tone generator called Theremin where one waves one's arms about in the space near it, and all tone progressions are forcibly glissando.

What we know from Theremino
Theremino Mca was a XRF project, driven by a separate hardware called Theremino_PmtAdapter. From the name, you can deduce that it is about making a high voltage for, and using, photomultiplier tubes. We, of course, are taking a different approach, using a large area PIN diode that has specifications promising useful operation in X-ray band up to 100KeV, where probabilities of detection are still above 1%, although the data shows it still functional (0.6%) at 1MeV. We steal recycle re-purpose get it from a Pocket Geiger, which basically triggered counts, as opposed to trying to measure the energy.

You can tell that I was trying to wring out why we think the pulse shape we get should ever look like the one that comes from scintillators, although the way photon does it's thing when it hits has to be much the same, even in a diode. When Mark @homebrewed provided us with a lovely measured pulse, I stopped looking.

Lead shielded test boxes
Amid the old documents, check out a 8-source array on the outside of a lead plate box. It was used to check out the coating on a camera lens.
It loses some in the translation, but I quote the text..
-------------------------------
"7 uCi of Am241 (not 8 as in the above image) has been used, arranged around a hole in a lead plate, behind
this hole a RAP47 25.4 x 1 mm probe was placed.
Please note that good results may be achieved using just one uCi of Am241, but
carefully for the geometric relationships between probe, source and target material".
-------------------------------



Am241 smoke detectors make "better" spectrum plots
So OK, we find in another site, using the same gadget. This time being used to discover what some antique stuff is made of.
Apparently, it is a Tibetan ink pot, and the green plastic disc contains 0.1uCi of Strontium-90 !
Thus, that source is about one ninth the strength of a Am241 smoke detector disc.

The Strontium disc goes right over the detector input, so would be making a peak of it's own



The whole series of experiments was about comparing Am241 smoke detector sources with the Sr-90, and how much better is using Am241.
They came from here --> https://www.theremino.com/en/blog/gamma-spectrometry/xrf
For convenience, I include the spectrum displays in the PDF attached.

What do the pulses look like at the start - really?
I have read through all of Livio Cicada's stuff on processing the pulses he gets, and all about stretching, and base line shifting, etc.
The key thing to note is that a huge chunk of the gain comes from the photomultiplier tube itself. What follows in the Pmt_Adapter is not to my taste, but anyway..


He was having a struggle with the (substantial) portion of the signal traversing below the baseline. This happened because the signal was already hugely amplified by the PMT before it got taken off the high voltage end of the tube via a 4.7nF 1.5KV capacitor.



_

---------------
OK, I understand what he does. he gets some nice spectra as well, although I would want better.
For all this, I am still going to be going my own way on this. We know that..

1. One of the spectra published were produced by a single smoke detector source. One was produced by four.
Another was produced by seven, regardless the image showed eight. Hence the clue is "do-able"!

2. The X100 - 7 will produce outputs of X-ray energy fluorescence.

3. The gain we give it is comparable to a photomultiplier, and more.

4. We don't really know what the current pulse looks like. What the PMT gives is a second hand highly amplified photocathode response to scintillator stuff that glows. The diode pulse may be very much faster, limited by the diode capacitance. Small area diodes can work at GHz.

5. We have a theoretical justification that the simple height of a pulse is proportional to the energy, only provided the waveshape form factor of all pulses is constant, and that the duration, even of the small ones, is also constant. Generally, we can't rely on this, and I think "low pass pulse stretching" maintains the relationship in a approximate way, becoming more marginal as the duration gets longer. The longer the time, the less the Y-axis fraction changes of the area represents energy.

Maybe if we just measure the peak, and take it that the duration no longer matters, because we preserved the peak, and we pretend the waveshape is constant, and was shorter, and we only stretched it to have more time to measure it on a slow ADC. This might even work!
OR..
We stay with integration as best way of counting up the energy, and insist on best resolution.
For now, I stay with integration. I think it is because I lack trust!

What we measure
We move to a theoretical summation of the charge as the area under the curve, so long as the Y-values were currents. We integrate it!
The relationship between charge and energy involving a capacitor is well enough known. It depends on the capacitor, and the voltage it gets to.



Of these, I choose the last. That is what a TIA does.
Here we have to keep very clear in the mind that we were measuring currents in the TIA, and representing them by a voltage analogue. From the output of the TIA onward, the voltage is the proportional analogue of the current.

Then when these were added up, the number we get represents the charge Q.
The only worrisome thing is that the energy we think is being measured is proportional to the square of what we summed up.

Hmm..
Either we use a nonlinear squaring opamp circuit, or we ask the little computer to find the square of the added up sum before we put it in a bucket. Otherwise, the relative energies will end up plotted too far crowded to the left all being square-rooted in a kind of anti-log(x) base 2 crush.

Have I got this right?
Also - my apologies for what has to be some tough stuff for members of a machine enthusiast's site

 

[rwm]

You are machining electrons...

 

[WobblyHand]

How others do it (xrf), and some fundamental stuff + hard sums
Like Mark, I am in review mode. I have be trawling everything, all the way back to Theremino Mca. At the end, perhaps a basic consequence to our computing.

Theremino is, of course, a general purpose plot display program written for Windows, with interface kit described as "modular" It is supposed to give visuals to all sorts of projects from keyboard music, atmospheric monitoring, data loggers, stepper motors, etc. The "modules" are such a close clone of Arduino, it can be completely misleading. Maybe, back then, they started out playing with Arduinos.

"Theremino" name alludes to that (ineffective) music maker tone generator called Theremin where one waves one's arms about in the space near it, and all tone progressions are forcibly glissando.

What we know from Theremino
Theremino Mca was a XRF project, driven by a separate hardware called Theremino_PmtAdapter. From the name, you can deduce that it is about making a high voltage for, and using, photomultiplier tubes. We, of course, are taking a different approach, using a large area PIN diode that has specifications promising useful operation in X-ray band up to 100KeV, where probabilities of detection are still above 1%, although the data shows it still functional (0.6%) at 1MeV. We steal recycle re-purpose get it from a Pocket Geiger, which basically triggered counts, as opposed to trying to measure the energy.

You can tell that I was trying to wring out why we think the pulse shape we get should ever look like the one that comes from scintillators, although the way photon does it's thing when it hits has to be much the same, even in a diode. When Mark @homebrewed provided us with a lovely measured pulse, I stopped looking.

Lead shielded test boxes
Amid the old documents, check out a 8-source array on the outside of a lead plate box. It was used to check out the coating on a camera lens.
It loses some in the translation, but I quote the text..
-------------------------------
"7 uCi of Am241 (not 8 as in the above image) has been used, arranged around a hole in a lead plate, behind
this hole a RAP47 25.4 x 1 mm probe was placed.
Please note that good results may be achieved using just one uCi of Am241, but
carefully for the geometric relationships between probe, source and target material".
-------------------------------

View attachment 436116

Am241 smoke detectors make "better" spectrum plots
So OK, we find in another site, using the same gadget. This time being used to discover what some antique stuff is made of.
Apparently, it is a Tibetan ink pot, and the green plastic disc contains 0.1uCi of Strontium-90 !
Thus, that source is about one ninth the strength of a Am241 smoke detector disc.

The Strontium disc goes right over the detector input, so would be making a peak of it's own

View attachment 436117

The whole series of experiments was about comparing Am241 smoke detector sources with the Sr-90, and how much better is using Am241.
They came from here --> https://www.theremino.com/en/blog/gamma-spectrometry/xrf
For convenience, I include the spectrum displays in the PDF attached.

What do the pulses look like at the start - really?
I have read through all of Livio Cicada's stuff on processing the pulses he gets, and all about stretching, and base line shifting, etc.
The key thing to note is that a huge chunk of the gain comes from the photomultiplier tube itself. What follows in the Pmt_Adapter is not to my taste, but anyway..
View attachment 436125

He was having a struggle with the (substantial) portion of the signal traversing below the baseline. This happened because the signal was already hugely amplified by the PMT before it got taken off the high voltage end of the tube via a 4.7nF 1.5KV capacitor.

View attachment 436136

View attachment 436137 _ View attachment 436139

---------------
OK, I understand what he does. he gets some nice spectra as well, although I would want better.
For all this, I am still going to be going my own way on this. We know that..

1. One of the spectra published were produced by a single smoke detector source. One was produced by four.
Another was produced by seven, regardless the image showed eight. Hence the clue is "do-able"!

2. The X100 - 7 will produce outputs of X-ray energy fluorescence.

3. The gain we give it is comparable to a photomultiplier, and more.

4. We don't really know what the current pulse looks like. What the PMT gives is a second hand highly amplified photocathode response to scintillator stuff that glows. The diode pulse may be very much faster, limited by the diode capacitance. Small area diodes can work at GHz.

5. We have a theoretical justification that the simple height of a pulse is proportional to the energy, only provided the waveshape form factor of all pulses is constant, and that the duration, even of the small ones, is also constant. Generally, we can't rely on this, and I think "low pass pulse stretching" maintains the relationship in a approximate way, becoming more marginal as the duration gets longer. The longer the time, the less the Y-axis fraction changes of the area represents energy.

Maybe if we just measure the peak, and take it that the duration no longer matters, because we preserved the peak, and we pretend the waveshape is constant, and was shorter, and we only stretched it to have more time to measure it on a slow ADC. This might even work!
OR..
We stay with integration as best way of counting up the energy, and insist on best resolution.
For now, I stay with integration. I think it is because I lack trust!

What we measure
We move to a theoretical summation of the charge as the area under the curve, so long as the Y-values were currents. We integrate it!
The relationship between charge and energy involving a capacitor is well enough known. It depends on the capacitor, and the voltage it gets to.

View attachment 436140

Of these, I choose the last. That is what a TIA does.
Here we have to keep very clear in the mind that we were measuring currents in the TIA, and representing them by a voltage analogue. From the output of the TIA onward, the voltage is the proportional analogue of the current.

Then when these were added up, the number we get represents the charge Q.
The only worrisome thing is that the energy we think is being measured is proportional to the square of what we summed up.

Hmm..
Either we use a nonlinear squaring opamp circuit, or we ask the little computer to find the square of the added up sum before we put it in a bucket. Otherwise, the relative energies will end up plotted too far crowded to the left all being square-rooted in a kind of anti-log(x) base 2 crush.

Have I got this right?
Also - my apologies for what has to be some tough stuff for members of a machine enthusiast's site

Math is easy for all the micros these days. Don't worry about that, if that's all that's needed. Floating point instructions are hardware based on a Teensy and are fast, especially for single precision.

At the moment, I'm dealing with frozen pipes. It got to -17F last night, with 20mph winds. Our bathrooms don't have water at the moment. The kitchen has hot and cold water so it's not hopeless. So far hairdryer on the pipes isn't working.

So I'm a bit preoccupied, I'm afraid. Added another 1500W of heat to the basement, hoping it helps. Certainly makes it more comfortable.

 

[homebrewed]

Mark - I guess you must be psychic or something..
Thanks for the new analysis.
You join the party. Three of us now posting LTSpice plots!
If it's OK, could you post the .asc here?

The great pulse you showed us on Page 62, #617 has about a volt amplitude, so I am guessing it had the benefit of extra gain from your signal conditioning board. Was that amplified from the U1B output at R4?

Yes, the photo shows a pulse coming out of the signal conditioning board, X10 setting. Based on the context it was taken just after I replaced the on-board diode bias generator with an external power supply.

I've attached the .asc file. My diode "model" is a bit different than what you're using, it will be interesting to see what difference yours makes. At the least, I'd expect it will be necessary to tweak the "offset adjustment" current source. In no way am I suggesting this circuit is anything but an exercise to present a different approach to the problem.

BTW I had to change the file extension from .asc to .txt before I could attach it. Is this another file extension we need to let @vtcnc know about?

 

[graham-xrf]

Here's a different approach to the input circuit, based on the idea that a charge amplifier, not a TIA, can accomplish much the same thing. The charge coming out of the detector is dumped into a capacitor that integrates the charge. Q = CV so V = Q/C (and the "gain" element is a noise-free capacitor!). The resultant voltage is input to a high-impedance (but not particularly high bandwidth) amplifier. I did have to extract a bit of current to reduce the offset but I think a secondary lowpass-filtered feedback loop could take care of that.

I really do not want to spoil it, and I do get it that everyone deserves a more compelling analysis/justification, but so sorry - this does not work for me.

Perhaps random, but I am minded of that school experiment where the charge on a variable capacitor is being measured with an electrometer, then you grab the adjuster rod, and pull the plates somewhat apart, and the voltage shoots up. The energy remains the same. That is not "gain".

I think what you are saying is that we have a passive integrator, building up (voltage) in the 1pF capacitor at the non-inverting input. Indeed we do, but we do not have enough energy there to allow anything to get up to sufficient voltage to get an opamp to work.

Actually, what happens is the charge carriers find themselves, very suddenly, in a 50pF (or 85pF) diode reverse biased capacitance. That would leave it a bit charged, but it does not get to do that. Firstly, it is leaking away through the 40MΩ equivalent leakage resistance, but that is incidental. It is also going into the opamp, and all the strays, and external leaks.

At that stage, the initial charge, if connected to anything at all, but particularly another capacitance, including strays, will have electrons flow to share that charge. The voltage will drop. Another 1pF would not amount to much change, except for all what else is there.

Looking at the circuit..
The little charge we have is shared onto a 1pF to GND, which itself has a much more serious 1MΩ discharge leak to GND.

If we wanted a pure integrator, that would be a voltage input device, through a resistor, into the inverting input. We then accept the help of a sensitive opamp, and put the integration capacitor in the feedback loop.

The most effective available way of getting that tiny charge to do something is to offer it into an opamp input, where the opamp will not allow more than a minuscule deviation before it's output goes bananas. The feedback connection allows it to correct the attempt at the input, with a monster gain. That would be a TIA.

The tiny charge we have, after being beefed up, does ultimately have to be shared in the switched capacitor array inside a ADC. The capacitors in the chip may be tiny, but the ADC will want sufficient volts to have them work.

This approach is based on my recollection of a high speed integrator using an all-passive RC network.

For expediencies sake I _did_ reduce the bias voltage to zero, so the offset current really will need to be greater to compensate for the PIN diode's dark current. Feedback will take care of that. Or maybe a digital pot that periodically acts to null the offset.

I'm throwing this out there to suggest that there may be different ways to look at the detector circuit, and (perhaps) allow the use of parts that have a lower GBW compared to some high GBW parts that seem to be in short supply.

The circuit is two stages of non-inverting voltage amplifier looking at a passive 1pF capacitor, and I am having a little trouble untangling the current generator on the non-inverting input. I can see why the SUBCKT is needed. AD8646 is not in the supplied product set.

Admittedly, this particular implementation is aimed at the Theremino approach where the pulse is heavily filtered: but in this case the peak voltage is pretty much determined by the charge coming out of the detector divided by the load capacitor. No ambiguity regarding the relationship between the current coming out of the detector and the voltage coming out of the amplifier chain.

I can't say this is a perfect solution because there still is the possibility of pulse overlap causing problems but that's going to be an issue regardless of what the front end looks like.

I believe overlap might be the least of it's problems!

Essentially, we need credible ways to measure maximum pulses of about 9.61e-15 Joules (from a 60KeV photon)
At the other end, we contemplate measuring minimum pulses of about 2.0e-16 Joules (from Magnesium, if we get lucky, and wait a long time).

By the definitions I see on Wikipedia, the coincidence of an electron volt being 1 eV = 1.602176634E-19 Joules,
and also the charge on an electron is the very same number 1.602176634E-19 Coulombs,
That must mean the potential was through 1 volt.

Therefore, reasonably, the first case involved 60,000 electrons, and the second involved only 1250.
That is so little, I don't think we can do it, or we will have a hard time getting there.
I definitely don't think we can do it across a 1pF capacitor, leaking 1MΩ, and an amplifier input gobbling up a whole 1pA input bias that is nearly 200 times the little current pulse we hope to capture.

This is why I offer the pulse up a TIA opamp input with only 2e-15A bias, and ask it to do it's best at a speed that will capture the shape of what little we had.

 

[homebrewed]

What we measure
We move to a theoretical summation of the charge as the area under the curve, so long as the Y-values were currents. We integrate it!
The relationship between charge and energy involving a capacitor is well enough known. It depends on the capacitor, and the voltage it gets to.

Of these, I choose the last. That is what a TIA does.

I don't think your premise is correct. The input waveform to the TIA is a time varying current -- it's not charge. Charge is the integral of current, so in the case of a TIA, where the output voltage is proportional to the current, the integral of the voltage waveform isn't proportional to the energy of the pulse, it's proportional to the charge contained in the pulse. No square-root calculation is required.

 

[graham-xrf]

I don't think your premise is correct. The input waveform to the TIA is a time varying current -- it's not charge. Charge is the integral of current, so in the case of a TIA, where the output voltage is proportional to the current, the integral of the voltage waveform isn't proportional to the energy of the pulse, it's proportional to the charge contained in the pulse. No square-root calculation is required.

Hmm - we do have to fix this!
The input waveform is indeed a time varying current. It happens over a time. I x T = charge in Coulombs.
When we integrate that, we do indeed have a value proportional to charge in coulombs.

Yes, I agree that the TIA output is a voltage analogue proportional to input current at any instant on the waveform.
Then when, over time, we add up a whole bunch of it's values, we have a number, in a computer, which represents the integral, over the pulse period, of the sampled currents. That would be the area under the current waveform curve. That then is the integrated currents. i.e. = the charge.

So finally, we have to figure out the energy of that charge.
We used a TIA, so we (deliberately) avoided it making a voltage, across any capacitor, at any point.
We look back to the equation, showing three ways of expressing energy in terms of charge, voltage, and capacitance.
We deliberately choose the last one Energy = (Q^2)/2C

The capacitor the original pulse was stored on was the 50pF (or 85pF). All the amplified analogues of current, summed, let us come to a value proportional to total charge. We can put a scaled value on it during calibration, when we show it some known pulses.

What we have is a good value proportional to charge Q
I think we have to square it to get a value proportional to energy.
The value of C is a constant. It gets taken up in the calibration.

In the entire thing, I kept in mind that we never made a fundamental voltage measurement, even once!
The ADC measures were voltage analogues of current.
I think I stand by my calculation, and that to get proportional to energy, we have to square the number representing the charge.

 

[homebrewed]

Hmm - we do have to fix this!
The input waveform is indeed a time varying current. It happens over a time. I x T = charge in Coulombs.
When we integrate that, we do indeed have a value proportional to charge in coulombs.

Yes, I agree that the TIA output is a voltage analogue proportional to input current at any instant on the waveform.
Then when, over time, we add up a whole bunch of it's values, we have a number, in a computer, which represents the integral, over the pulse period, of the sampled currents. That would be the area under the current waveform curve. That then is the integrated currents. i.e. = the charge.

So finally, we have to figure out the energy of that charge.
We used a TIA, so we (deliberately) avoided it making a voltage, across any capacitor, at any point.
We look back to the equation, showing three ways of expressing energy in terms of charge, voltage, and capacitance.
We deliberately choose the last one Energy = (Q^2)/2C

The capacitor the original pulse was stored on was the 50pF (or 85pF). All the amplified analogues of current, summed, let us come to a value proportional to total charge. We can put a scaled value on it during calibration, when we show it some known pulses.

What we have is a good value proportional to charge Q
I think we have to square it to get a value proportional to energy.
The value of C is a constant. It gets taken up in the calibration.

In the entire thing, I kept in mind that we never made a fundamental voltage measurement, even once!
The ADC measures were voltage analogues of current.
I think I stand by my calculation, and that to get proportional to energy, we have to square the number representing the charge.

I think you're losing track of what generates the pulse.

The current coming out of the detector is proportional to the xray photon energy, just spread out over time due to carrier transit time in the depletion region (somewhere around 20nS for a 26V bias) and the junction capacitance + other parasitic capacitances (package, TIA input C, etc.). So the integration basically "reconstructs" the energy.

 

[graham-xrf]

I think you're losing track of what generates the pulse.

The current coming out of the detector is proportional to the xray photon energy, just spread out over time due to carrier transit time in the depletion region (somewhere around 20nS for a 26V bias) and the junction capacitance + other parasitic capacitances (package, TIA input C, etc.). So the integration basically "reconstructs" the energy.

I am not sure that it is so. I am OK to get convinced.

I understood it differently.
The incoming the x-ray photon has a fixed energy. It causes the release of carriers. They can only be electrons released from the photoelectric effect, and quite exactly, the energy used in that action corresponds to the energy one might derive from dissipating those electrons.

That number of electrons becomes a charge, on a capacitance, all very suddenly, which will run away and dissipate as currents in whatever it discovers itself connected to. Part goes into the reverse bias resistance. The rest goes into whatever is outside, including a TIA input.

I never saw the current one gets out of a photodiode as a straightforward analogue of the photon energy in Joules of what arrived.
It's not a big problem, because if the content of the current pulse is indeed already a measure of incoming energy, then counting up the waveform of current conveniently does not need squaring.
That said - I will find out, one way or another.