Needing more than a spark test?

[homebrewed]

I finally got around to turning on my pocket geiger board to take a look at its analog signal. Looking at the board vs. schematic, I was becoming suspicious that the connections to JP2 did not agree, so I needed to fire it up and make some measurements. It turns out they do NOT agree. The board swaps P2 and P4! If you look at the bottom of the board, the pad marked with an "A" is the analog signal; and the pad opposite it is ground.

Since I had the board running, I decided to get out one of my Americium smoke-detector capsules to see what kind of pulses I'd get out of the thing. I clamped the board in a Panavise with the detector facing down, then slid the Am241 source underneath. The copper foil shield was left in place so I'd (hopefully) see at least a few fluorescence photons. BTW this isn't an ideal configuration -- the only XRF photons I'd detect are the ones emitted on the exit side of the copper foil, from the somewhat-attenuated 60Kev primaries, PLUS some primaries (and reduced pulse height because the detector's response rolls off above 10Kev). I'm assuming the copper will strongly absorb its own 8Kev XRF photons & that's why only ones emitted near the exit side will be detected. Anyway, here are some photos showing a few pulses I captured using a DS212 mini-scope:



BTW, the count rate was quite low, maybe a few CPS....hopefully due to the nonideal test setup I used. It's encouraging that the time scale agrees with my simulations. The noise level is concerning to me though. I also occasionally observed some very large negative transients, possibly popcorn noise. I did not see any large positive transients.

I did buy 6 AM241 capsules so if the count rate really IS this low I can bump it up by using more, similar to what the Theremino folks did.

 

[homebrewed]

The largest-amplitude pulse I showed probably is not due to copper XRF or the Am241. It's either background/cosmic rays or popcorn noise, because those pulses show up with no Am241 nearby. When I remove the americium, the lower-amplitude pulses pretty much vanish.

Now it's time to work on a setup with some proper geometry. I'm not going to even think about the MCA portion until I can get a decent physical setup. I have some lead sheet for shielding and some punches to create an aperture for the XRF photons to enter (and to shield ME from the 60Kev primaries). I will probably cut away the copper foil that's over the detector and replace it with aluminum foil, to reduce the attenuation of lower energy X-rays.

I wish there was a way to employ pulse averaging to improve the SNR, but since they can't be classified apriori that's not possible. The best approach probably is to use least-squares curve fitting on each pulse. I HAVE seen a few instances of what looks like pulse pileup, so -- hopefully -- the RMS error between the polynomial fit and pulse will serve as a way to reject excessive pileup. I'll have to examine the Arduino code I found for polynomial curve fitting to see if the error is computed at the same time or not.

 

[graham-xrf]

Yay! Something actually bumps it! I do like your pictures!

Try not to worry about the low amplitude, and come to that, not even the noise. We can get it to where what (our) kit can see to the noise floor. Significant that when the Am241 is not there, the smaller pulses go away. The detector is built to count, not to get to linear accurate pulse analysis info. I am glad that the pulse does at least have the same shape as my simulation assumption.

When you say "copper shield left in place", is that to hit it with Am241 from one side, and have the XRF from copper come out the other side?

I know this sounds nuts, but try uncovering the sensor and showing it a banana.

My sensor has not arrived yet. The whole import thing from the USA is odd. I can get everything from Sharpie pens to router cutters from China in the mail in a few days. The baggies have the standard Chinese customs form stuck on them. The whole business from the USA is needlessly expensive, and delayed, with no arrangement to get it to the door without some sort of exchange and payment to bring it from wherever it landed, by some carrier, even if it was the Royal Mail.

Regarding pulses, pulse pile-ups, noise, signal processing etc. Lesson 1 is we are not going to get very far with Analog's photodiode design "wizard". I get it that it is supposed to be used to recommend their products, but the circuit options are stupidly limited. I do not want a fixed choice about gains, filters, number of stages, etc.

Higher Education
Staring from a sensor, and a photon energy, and getting to a current pulse is actually not so easy. Incoming keV photons will be in the range 5keV to 10keV with highest probability, and probably 0.5keV to 60keV in all. Any others might be more direct stuff, scatter from cosmic ray excitation, and generally radioactive stuff around, like wine, or the strontium in the bones in your fingers. I never knew before that a simple radiation check of a wine can date it quite accurately, and that wine made before the 1950's nuke tests is simply not radioactive from fallout, and pretty much everything that grew after those test - just is.

Incoming photons whack into the Silicon. Is it Silicon? I think it has to be. The only other candidates are Germanium and Gallium Arsenide. They work over higher energies, and Silicon is the only one that has the correct dark current range corresponding to out X100 - 7. The energy pings loose an electron. The work function of all the metals is generally less than 6eV, so a 5keV hit will have most left over as a deflected photon, and go on to liberate more electrons. That's a bit of an over-simplification. Sometimes less is left over, and the liberated electron flies off with some high kinetic energy, and can dislodge electrons of its own. In the end, all are absorbed, and the electrons collected by the field from the bias, with a rise time, and a tail-off. A 5keV photon should liberate at least 800 electrons.

That is depressingly small, and may be less that we can see in the face of other noise. Loosely, if half of them the pulse rise-time of 500nS, the current involved is about 128pA. The input current for the op-amp is 3pA. It sits on top of a 1500pA dark current, which can be cancelled by offset adjustment. There is a noise hash which at 6.1E1-14 A/√Hz. At 500KHz bandwidth, that's 30nA, but as a peak-to peak.

If we accept the slowdown, and narrow the bandwidth, we can lose a lot of noise, but now I am feeling I made a calculation mistake.
If the makers of X100 - 7 were able to make a plot down to 600pA dark current, and claim 5keV was detectable, then we should be able to see it too!

The AREA under my pulse curve represents the electron charge knocked loose. In all this, I am trying to get at a realistic peak value of Ip, the photodiode current.

On the way, I go through selected lectures from the absolute expert Professor at an Indian University. --> HERE
Lecture #40 gets close to what we do.

None of the energies get as far as Compton scattering or electron-positron pair formation and annihilation - thankfully.

Basically we get 5000 to 10,000 electrons appearing in a 100pF capacitance being discharged by a 40M ohm resistance. The trouble is, the mere presence of electrons is not actually a current. I know current is supposed to be a "flow" of electron charge in a time, but that's the point. Electron flow is about 2 to 3 cm per hour in copper and an electron can take a week to get around a circuit. Also, they never go through capacitors. The charge transfer is different. At near light speed.

Do it backwards
Let's first start with your 200mV to 600mV pulses. Then work back through the circuit to get at the Ip that caused it.

Capacitor AC coupled.
It can bring all sorts of problems, but is convenient to get bias onto the sensor without upsetting the charge amp. My simulations show exactly the same pulse, un-shifted, AC coupled or not.

Anyway, one justification I have seen for AC coupling into differentiator circuits was to "separate" other pulses that may happen during the trailing tail of a pulse. Edge detection like that may be good for circuits dedicated to getting simple count rates captured with a threshold comparator, but in our case, I think it loses the amplitude information, which for us is the only grip we have on what it's origin was. There may be some programming trick by which multiple pulses during a tail decline can still be analyzed, subtracting the right amounts before putting them in the correct bucket, but I think it is simpler to just reject them.

The "right" gain.
I think enough so that the A/D maxes out it's range a bit beyond the highest energy we expect from XRF, say 60keV. We can have a "zoom in" gain, mode, to analyze lower height pulses, and a lower gain setting in case we want to check out the direct radiation coming from isotopes (like the banana). Possibly some XRF identification can happen from seeing materials directly, instead of set glowing by Am241. Cobalt in an alloy might have a direct signature. Anyway, programmable gain ranges sounds useful. Is this "feature creep"?

Potassium 40 (bananas) does MeV beta decay, but no gamma, so useless for XRF, but your sensor might see it!

 

[homebrewed]

Yay! Something actually bumps it! I do like your pictures!

Try not to worry about the low amplitude, and come to that, not even the noise. We can get it to where what (our) kit can see to the noise floor. Significant that when the Am241 is not there, the smaller pulses go away. The detector is built to count, not to get to linear accurate pulse analysis info. I am glad that the pulse does at least have the same shape as my simulation assumption.

When you say "copper shield left in place", is that to hit it with Am241 from one side, and have the XRF from copper come out the other side?

I know this sounds nuts, but try uncovering the sensor and showing it a banana.

My sensor has not arrived yet. The whole import thing from the USA is odd. I can get everything from Sharpie pens to router cutters from China in the mail in a few days. The baggies have the standard Chinese customs form stuck on them. The whole business from the USA is needlessly expensive, and delayed, with no arrangement to get it to the door without some sort of exchange and payment to bring it from wherever it landed, by some carrier, even if it was the Royal Mail.

Regarding pulses, pulse pile-ups, noise, signal processing etc. Lesson 1 is we are not going to get very far with Analog's photodiode design "wizard". I get it that it is supposed to be used to recommend their products, but the circuit options are stupidly limited. I do not want a fixed choice about gains, filters, number of stages, etc.

Higher Education
Staring from a sensor, and a photon energy, and getting to a current pulse is actually not so easy. Incoming keV photons will be in the range 5keV to 10keV with highest probability, and probably 0.5keV to 60keV in all. Any others might be more direct stuff, scatter from cosmic ray excitation, and generally radioactive stuff around, like wine, or the strontium in the bones in your fingers. I never knew before that a simple radiation check of a wine can date it quite accurately, and that wine made before the 1950's nuke tests is simply not radioactive from fallout, and pretty much everything that grew after those test - just is.

Incoming photons whack into the Silicon. Is it Silicon? I think it has to be. The only other candidates are Germanium and Gallium Arsenide. They work over higher energies, and Silicon is the only one that has the correct dark current range corresponding to out X100 - 7. The energy pings loose an electron. The work function of all the metals is generally less than 6eV, so a 5keV hit will have most left over as a deflected photon, and go on to liberate more electrons. That's a bit of an over-simplification. Sometimes less is left over, and the liberated electron flies off with some high kinetic energy, and can dislodge electrons of its own. In the end, all are absorbed, and the electrons collected by the field from the bias, with a rise time, and a tail-off. A 5keV photon should liberate at least 800 electrons.

That is depressingly small, and may be less that we can see in the face of other noise. Loosely, if half of them the pulse rise-time of 500nS, the current involved is about 128pA. The input current for the op-amp is 3pA. It sits on top of a 1500pA dark current, which can be cancelled by offset adjustment. There is a noise hash which at 6.1E1-14 A/√Hz. At 500KHz bandwidth, that's 30nA, but as a peak-to peak.

If we accept the slowdown, and narrow the bandwidth, we can lose a lot of noise, but now I am feeling I made a calculation mistake.
If the makers of X100 - 7 were able to make a plot down to 600pA dark current, and claim 5keV was detectable, then we should be able to see it too!

The AREA under my pulse curve represents the electron charge knocked loose. In all this, I am trying to get at a realistic peak value of Ip, the photodiode current.

On the way, I go through selected lectures from the absolute expert Professor at an Indian University. --> HERE
Lecture #40 gets close to what we do.

None of the energies get as far as Compton scattering or electron-positron pair formation and annihilation - thankfully.

Basically we get 5000 to 10,000 electrons appearing in a 100pF capacitance being discharged by a 40M ohm resistance. The trouble is, the mere presence of electrons is not actually a current. I know current is supposed to be a "flow" of electron charge in a time, but that's the point. Electron flow is about 2 to 3 cm per hour in copper and an electron can take a week to get around a circuit. Also, they never go through capacitors. The charge transfer is different. At near light speed.

Do it backwards
Let's first start with your 200mV to 600mV pulses. Then work back through the circuit to get at the Ip that caused it.

Capacitor AC coupled.
It can bring all sorts of problems, but is convenient to get bias onto the sensor without upsetting the charge amp. My simulations show exactly the same pulse, un-shifted, AC coupled or not.

Anyway, one justification I have seen for AC coupling into differentiator circuits was to "separate" other pulses that may happen during the trailing tail of a pulse. Edge detection like that may be good for circuits dedicated to getting simple count rates captured with a threshold comparator, but in our case, I think it loses the amplitude information, which for us is the only grip we have on what it's origin was. There may be some programming trick by which multiple pulses during a tail decline can still be analyzed, subtracting the right amounts before putting them in the correct bucket, but I think it is simpler to just reject them.

The "right" gain.
I think enough so that the A/D maxes out it's range a bit beyond the highest energy we expect from XRF, say 60keV. We can have a "zoom in" gain, mode, to analyze lower height pulses, and a lower gain setting in case we want to check out the direct radiation coming from isotopes (like the banana). Possibly some XRF identification can happen from seeing materials directly, instead of set glowing by Am241. Cobalt in an alloy might have a direct signature. Anyway, programmable gain ranges sounds useful. Is this "feature creep"?

Potassium 40 (bananas) does MeV beta decay, but no gamma, so useless for XRF, but your sensor might see it!

View attachment 340489

A lot of good comments. So in (approximate) order of questions/comments --

Yes, the first test had the 60Kev gammas enter on one side and the copper XRF photons exit on the other. When we x-rayed IC packages, 60Kev was about the lowest we could go and get decent images through the molding compound and silicon chip itself so they will be strongly absorbed by the copper foil.

I happen to have some potassium chloride solution for storing pH meter sensors and had thought of exposing the detector to the bottle, so I immediately got the banana reference .

Yes, the detector just about has to be silicon-based. Regarding the yield of carriers per photon, I think the threshold difference between the noise comparator and signal comparator could be used as a rough guide to the magnitude of "real" current pulse coming out of the detector. Or just compare the actual pulse height coming out of the pocket geiger vs. the simulated result and scale the current pulse in the simulation accordingly. So I'm on the same page as you w/regard to working backward to see what the detector really is putting out.

The comparison to the velocity of carriers in a metal is not relevant to a semiconductor. For lightly-doped silicon, the mobility of electrons and holes is around 1000 cm^2/(volt-second). Sounds pretty slow, but the junction width is pretty small!

Regarding your simulations and AC coupling (or not): I see the same thing. According to the Theremino docs, they're really concerned about preserving the baseline -- and that's why they use that pole compensation network along with the differentiator. Their approach is all about getting a really good measure of the peak height, so it's important to establish a good baseline. I'd thought of some sort of reset circuit to force a known baseline (after a good pulse comes through), but, again, it's something that needs some preliminary results to see what we really need to do in order to get decent data. Software can come in handy here as well, to provide an averaged and quiet baseline voltage.

I think just simply rejecting overlapping pulses is best, at least for now. Trying to de-convolve data in order to resolve the individual pulses is not something I want to try -- at least, not at first. If we end with a couple of reasonably stable test platforms then we can test different ideas in order to improve the resolution.

 

[graham-xrf]

So far as I can see in the simulations, when the diode is the "right" way around, we get the nice pulse, with the stable baseline that corresponds to the dark current. In theory, turned the other way up, it should be able to put electrons into GND, and make the signal at the op-amp go the other way, but the sim does not work like that. I don't yet know why.

That way up, when AC coupled, we get the pulse with the daft baseline. It goes up to a (sort of) peak, drops down below the point it started, and then drifts back up to a point almost half up the total size of the peak.

Turning it back to the "nice pulse" direction, I think the amplified output from the dark current is actually useful. We can apply some offset at one of the later stages to shift most of it away, but we see the photon current poke up out of it. Add the thermal noise, and it expands into a "noise band".
That's OK. The level between pulses is what we want. I can limit the bandwidth to clean up the noise, to make the trace look cleaner. That would reduce the variation in the sampled numbers the A/D sampler will give.

I m also doing the trawl for a good low cost A/D that has a built-in band-gap reference, 16 true bits.
Adding the A/D is not, to my mind, "a feature creep".
If the pulse baseline wanders about for any reason, it's going to get clamped, like a old-style composite video.

- - - -- - - - - - - - - - - - -
I am now spending most daytime setting out for the hardstand land dig, and working on "essential" internal (boring) DIY.

Using USA units approximations, the debate for width is 10ft..? Hmm 11ft..? er OK.. 12ft..? Gulp! Hmm!
One of the HM members says something like, "You can't have too many tools, nor a shed that's too big"!
5m long. That's 16.4ft.
A 16' x 12', less the wall thickness, would probably be considered "small" by many, but for me, that much is a big deal!

 

[homebrewed]

I m also doing the trawl for a good low cost A/D that has a built-in band-gap reference, 16 true bits.

When I bought the SMT components for my signal-conditioning board I also got an LTC2378 to play with. It's a SAR style 16 bit ADC and goes up to 1MSPS. Its SNR is pretty good, claimed to be 97 dB -- close to theoretical for 16 bits. For some reason many ADC's don't include a built-in Vref circuit. Of course, Linear Tech has plenty of reference generator parts. The data sheet for the ADC shows a "typical" application using an LTC6655 Vref generator. The ADC cost me $31.37. The LTC6655 isn't all that cheap, surprisingly enough -- $9.13. These are Digikey single-item prices.

To minimize package size and pin count, ADCs like this one use high speed SPI....so for a 1MSPS sample rate you'd need a clock that at least runs at 16MHz. Actually faster, to handle the extra bits needed for the chip address and R/W bits.

I also have looked at higher-end ADCs that are used for digitizing audio but they have all sorts of features that are overkill for this kind of thing, like 24 bits, different data-framing options and so on. I thought I'd try to keep the ADC (and therefore programming) as simple as possible.

One nice thing about the Teensy's SOC is that it supports DMA so the processor doesn't have to deal with the overhead. You can set it up for a circular buffer so you can do things like capturing pre-trigger data. I don't know how easy it is to set up an RPi's GPIO subsystem for DMA, although it's likely that it IS possible. In a perfect open-source world there's someone out there who has already done it, eh?

 

[graham-xrf]

When given a heatsink, and a USB3 SSD, the Raspberry Pi4 is a reasonable desktop computer in it's own right. When given over to single task instrumentation code, it can go as fast as most things want to. I am pretty sure one can have a continuously running buffer loop snatching whole registers. One can also have software running fast enough to control the A/D for a data transfer, serial or otherwise, while the A/D is building it's next sample. Pipelined is just fine!
For speed, this one is order of magnitude faster than Thermino. How fast does Teensy go? It may be way more than enough!

I have a set of 1MHz, 2MHz, to 10MHz candidates in my notes.

It says things like AD4000 2M 93.0dB £21.67 ($28.17) or £11.09 ($14.42 ) in the MSOP pack sans pins
_ LTC2380 2M 96.2dB £35.99 ($46.79) (don't want to go there!)
_ AD7890 1M 91.5dB £15.53 ($20.19)
_ LTC2310-16 2M 82.0dB £14.67 ($19.07)

There are some affordable 10M parts. I am staying resolutely at 16 bits.
None of the above is decided.

Aside from making the sampling reach to the noise floor, my reason for wanting built-in analog side powered bandgap reference was to use it (or something derived from it), for the diode bias, and for any offsets inputs. The digitals side can use the computer power, and only meet the A/D part at one place inside the A/D chip. I would be looking to have it that the energy for getting the digitals communicated did not share a return current path in any way that could bounce the analogs, especially the parts trying to gather pico-amps. If not opto-isolated, then at least with enough resistance in the digital leads to limit that energy, consistent with keeping the pulses slightly sharper than sine waves.

I have not thought through the digital comms yet. High speed serial is not what SPI and the like is good at.
LVDS goes at the kind of speeds and low power that I like. Indeed, transferring a whole 16 bits register in a parallel chunk can be done, but I have used serial for a long time now. There has to be some sort of buffer, so the A/D can keep clocking regardless, possibly policed by the computer to have it snatch and deliver only proper, compete pulses.

Using a dedicated high speed sample/hold can accurately capture the peak. If in calibration, the peak value can represent the energy in the rest of the pulse well enough to be able to identify it as separate, the data gathering can be simpler. I don't want to build a fractionally better Thermino. (Strictly speaking, Thermino is only the general purpose display device, free, but not open source).

We will have more complete circuits to discuss soon. If the cost of stuff, including sensor, can be kept to under about $120, I would feel almost cheerful about it!
The thing is, we are trying for a modicum of quality. Use fewer bits, and a tiny diode, and a Geiger-type amp, compromise it all in an audio lead to a mic channel, and we get something that I think would be difficult to make work right. Ahh well.. I still have a scintillator that twinkles - I think.

 

[homebrewed]

For speed, this one is order of magnitude faster than Thermino. How fast does Teensy go? It may be way more than enough!

The Teensy4.0 runs at 600MHz and has 512K of on-chip RAM that is used for buffering peripheral I/O; and apparently the DMA channel(s) can access the RAM at 1/4th the system clock rate, i.e., 150MHz. There are two RAM blocks and the other is used for programming so I/O doesn't compete with code execution. The data sheet for its MCU shows that a number of device pins can be configured as LVDS I/O, as well. However, I didn't find much on the pjrc.com forum regarding a successful use of LVDS I/O. DMA can be used in combination with the serial I/O subsystem -- I believe that is how the PJRC audio codec board is used.

Apparently, the default register assignments for the Teensy4 are not set up for doing parallel 16 bit reads (or writes). But I did find some info in this thread that suggests it can be done, at least for 12 bits. No idea if LVDS functionality could be added in addition to this: but the Teensy4 only has 23 digital I/O's connected to the headers so there aren't enough pins for LVDS anyway.

So it looks like the teensy won't work for 16 bit parallel LVDS I/O. But if the lines are short and have proper impedance matching I'd think that single-ended I/O could go up to at least 10MSPS (the SPI clock for the PJRC audio codec runs at 30MHz).

For anything much faster I'd start looking at an FPGA solution. But I really don't want to go there -- I don't know Verilog that well, so that would turn into a real bootstrapping kind of project. For now I'm just going to try using the Teensy's 12 bit 1MSPS on-chip ADC and see how that works out. There already is a library for using the ADCs so that makes things a bit simpler. I hope.....

 

[graham-xrf]

Re: SparkFun --> Pretty much snafu.
For SEN-14209 $69.95 + $5.19 shipping and handling = $75.14.
No arrangement to pre-pay customs charge, so apparently +£18.97 => +$24.79 extra.
So total is £99.93, but that is the least of it.

I had repeatedly (3 times) tried to get Sparkfun to fix the misspelled first letter of my name, and they assured me that they would do it right away.
I checked, and it seems I succeeded. My correct name and address is there - 3 times repeated. Their system does not edit replace. It deletes and creates a new extra.

All this did not prevent them sending it with the wrong name. That causes an identity problem when it arrives, and it is time to pay customs and delivery charges online to get it delivered. The name must match the name on the card or the payment systems will never accept. To (maybe) fix this I have to come out of isolation, travel to Alton, and try to extract the package from the mail depot, in person.

I do not know how the fees are comprised. I get stuff from China, and other places, Ukraine even, rather easier. They come with a customs sticker indicating prepaid. Regardless the charges, these traders seem to have it all set up. eBay may be a bit special, with standardized schemes to help traders. I am not sure about SparkFun. Are they Canadian?

We can try and make everything about the kit we are trying to make be relatively low cost - except, it seems, the photodiode!

 

[homebrewed]

Re: SparkFun --> Pretty much snafu.
For SEN-14209 $69.95 + $5.19 shipping and handling = $75.14.
No arrangement to pre-pay customs charge, so apparently +£18.97 => +$24.79 extra.
So total is £99.93, but that is the least of it.

I had repeatedly (3 times) tried to get Sparkfun to fix the misspelled first letter of my name, and they assured me that they would do it right away.
I checked, and it seems I succeeded. My correct name and address is there - 3 times repeated. Their system does not edit replace. It deletes and creates a new extra.

All this did not prevent them sending it with the wrong name. That causes an identity problem when it arrives, and it is time to pay customs and delivery charges online to get it delivered. The name must match the name on the card or the payment systems will never accept. To (maybe) fix this I have to come out of isolation, travel to Alton, and try to extract the package from the mail depot, in person.

I do not know how the fees are comprised. I get stuff from China, and other places, Ukraine even, rather easier. They come with a customs sticker indicating prepaid. Regardless the charges, these traders seem to have it all set up. eBay may be a bit special, with standardized schemes to help traders. I am not sure about SparkFun. Are they Canadian?

We can try and make everything about the kit we are trying to make be relatively low cost - except, it seems, the photodiode!

Ouch! I'm sorry to learn about your problem -- rather surprising, overall. I wouldn't have expected Sparkfun to be so inept when it comes to shipping stuff outside of the US. BTW, Sparkfun is located in Colorado.

I found a European electronics distributor called Swatee Electronics selling them, but for about 2X what Mouser wants for them.