Needing more than a spark test?

[graham-xrf]

Hi Mark
Lets go with the undershoot first.
I can only guess that with the op-amp you are using, the long phase change, and way-behind slew rate, and maybe what it takes to get the capacitor across the feedback resistor to change it's mind about which direction it is charging, is what is responsible for the undershoot.

The voltage only undershoots because the op-amp is driving it there, because it is too slow in responding to the new information at the input that the pulse is over. If the circuit was fast enough, it would refuse to undershoot. Also, if 66MEG is still in there, consider just little currents are coming back through the feedback to persuade the inverting input not to allow an overshoot. Lose the 66MEG, no matter what op-amp you use!

You don't have to mess with the coupling capacitor, nor consider DC coupling, to not have overshoot. Aside from the extremely necessary property that it allows us to bias the diode at will, it also can limit the slowest surges of shot noise.

See what happens in TIA-Amp3-dev. This circuit is incomplete. It does not yet have the last gain/filtering stage, nor the differential driver with white noise filter, but it is enough to show 800mV of output that started from 10nA pulse (I guessed that might be a maximum).

Note that the op-amp in the top left corner is just a temporary simulation convenience to set the power supply rails. I was trying out how close the "Rail-to-Rail" could get.



Notice that the current pulse 10nA stimulus did not persuade the actual current to get any higher than 8.6nA, because some got leaked into the 40Meg, and some tried to charge C1. That's OK. It all still works!

One can increase the C2, and see the delay increase. Decreased to 0.5pF, it finds the best peaking, and closest tracking of the pulse. I don't really want to put two 1pF in series, and that was a bit close to oscillation anyway. Reducing the Rf to 510K also makes it track closer, but I did not care that the signal would be 1 or 2 microseconds "late". I did care that it would not have late feedback overshoot, which it does not. I believe everything about struggling with overshoot, and elaborate tricks to establish a "zero" are unnecessary. When the pulse dies to zero, so should our amplifier!

- - - - - - - - - - - - -
Yep - I do like the views. When I first moved here, my place was a disheveled bungalow with asbestos roof and derelict outbuilding at the sh**ty end of the village. So I built another place more or less around it, and demolished the middle. I knew the Telegraph house on the hill was a listed semaphore historic building, as was the cottage opposite. Over time, I found I was living in a sought-after des res spot.

- - - - - - - - - - - - -
Your last calculation worries me a bit.
A gamma energetic enough to get X-rays glows out of a chunk of metal, might do damage to a DNA protein, and persuade it to start replicating itself. Come to that, when we have medical X-Rays, these come as billions of photons, with energies, I think, low end is about 25keV. A 60keV "gamma" is in the range we would call "harder X-Rays". Only from ignorance, I will try and avoid too much "thumb gamma", but it happens when extract the Am241 from the electrode tag disc.

While we are on the subject of these gammas, may I assume that 2mm (about 5/64") of lead is easily enough to be a "shield" for the photodiode, and outer skirt? In the design, the lead shield includes behind the sensor. It has two holes to let through the wires to reach the circuit board.

One of the things I think about is whether one can turn a 2mm thick flange in lead.

 

[graham-xrf]

60 KeV is around the KV used for medical extremity radiography so I would expect significant absorption. 55% seems reasonable, maybe higher. The photon flux from your source is pretty low however.
In imaging the patient dose generally goes down as the KV goes up since more photons reach the image receptor. One other difference is the 60KeV represents a monochromatic beam. In imaging that would be the peak KV so the average energy is about 2/3rd.
R

Hi Robert.
Thanks for that.
When I did get to operate a hospital X-ray machine, the energy of the photons was decided by the voltage, and the flux intensity (how many photons/sec) was controlled by the filament current. I had not thought that if the kV was increased, the image receptor would see more because they went through instead of getting stopped in the patient.

Of course, our flux is tiny, to the extent that I cannot use responsivity formulae for PIN diode carriers that depend on it. We do not get photons/sec/area. Instead, we get a photon hit every now and then at a rate of Hz only. Each hit releases only one X-Ray.

Maybe Mark can say roughly how many pulses/sec happen (when the Am241 carrier is turned the right way around).

 

[homebrewed]

I can only guess that with the op-amp you are using, the long phase change, and way-behind slew rate, and maybe what it takes to get the capacitor across the feedback resistor to change it's mind about which direction it is charging, is what is responsible for the undershoot.

I'm afraid you are over-thinking it. The slow pulses used by Theremino-style MCAs aren't going to produce slew rate related problems like that. It's all about charge conservation and doesn't require an active circuit in order to demonstrate the problem. I've inserted a simulation of the RC coupling circuit to show what I mean:



However, in the overall scheme of things, I'm not sure this really is a problem. As long as the peak voltage passes through the RC network accurately, why should it matter if it then has some undershoot?? Time to re-read the Theremino stuff I guess.

 

[homebrewed]

Maybe Mark can say roughly how many pulses/sec happen (when the Am241 carrier is turned the right way around).

On average, I see somewhere between 1-4 reasonably well-formed pulses per second. This is the capsule facing the correct direction, with the copper shield in place. Now I can see why we want multiple Am241 capsules. A set of 8 should produce something like 8-32 CPS, all else being equal. The count rate seems to be lower than I would have expected, based on the info in the Theremino-MCA literature. Maybe the copper shield is absorbing more than I think.

On the subject of machining lead, I'm going to try using my knockout hole punch set -- this -- to make the aperture. The first hole in grabby/sticky lead will probably be alright, but subsequent ones could be overly exciting. The lead shielding doesn't need to be hermetic, the pieces just need to overlap enough to block unwanted x-rays....so the shielding box may be assembled using duct tape and/or silicon glue (pretty much like my home-made spectrometer). After all, the first setup is to demonstrate proof of concept, not design a commercial piece of kit. That's reserved for Version 2 .

I'm hoping that the shielding can also be used to shield the pocketgeiger's detector and amplifier from line noise, so I can get rid of the copper-foil shield altogether.

 

[graham-xrf]

I'm afraid you are over-thinking it. The slow pulses used by Theremino-style MCAs aren't going to produce slew rate related problems like that. It's all about charge conservation and doesn't require an active circuit in order to demonstrate the problem. I've inserted a simulation of the RC coupling circuit to show what I mean:

However, in the overall scheme of things, I'm not sure this really is a problem. As long as the peak voltage passes through the RC network accurately, why should it matter if it then has some undershoot?? Time to re-read the Theremino stuff I guess.

Thanks for counting the rate. I have 6 sources, but I am wondering whether 8 or 12 would be better. I think even with that many, we can hardly call this thing "ragingly radioactive"!

- - - - - - - - - - - - - - - -
Undershoot need not happen, unless you inadvertently measure within a feedback loop. I believe it does matter. OK - call me hardball, but I set out to measure the pulse, and find it's energy, pretty much exactly! I was quite turned off by the Thermino smudge that had a very tenuous relationship with the event that caused it.

Putting a pulse through a capacitor, and letting the other side settle at something related to the average value of the pulse shape is not something I thought would be OK if the attempt was to measure the energy (area sum).

PocketGeiger 5 was worse - and also in it's way better. It was never for XRF measuring. It was a pulse counter.

Re "overshoot".
OK I do understand that the overshoot pulse, and what caused it, is a slow event. Your initial pulse is also very long. We know from the scintillator that the entire event is over in 4uS to 10uS

I think that what is going on in the example is the initial (very long) exponential pulse starts charging the capacitor, racing the 10k resistor intent on discharging it, and the resistor is the short-term loser. As the exponential starts reverses direction, it is again the winner, but this time against a partly charged capacitor. This is much the same as feeding a square wave into a capacitor, and noting that the waveform on the other end is still a square wave, but going both positive and negative, about a DC value which is the average of the waveform shape.

If your simulation time is set to something longer, eventually all returns to zero. Here is your circuit with the capacitor current monitored.



You can see that the current reverses (for a while) at about 60uS

If there is a place in Geiger circuit where that is happening, we don't have to put up with it, and we don't have to mess with a series capacitor to have our way. At the charge summing point, we do not have a 10K resistor. We have instead is a op-amp that should be very insistent on forcing the voltage to be within a few uVof the positive input at GND - except this one isn't. We get to monitor the op-amp output efforts to bring this about.

I can show this (apparent undershoot current) on the LTC6269 circuit at speed. It requires two plots, to allow the scale of the small value to be seen. This one plots the photodiode current from the carriers released (green trace), and the current through the capacitor C3 (yellow trace).

Also the voltage at the non-inverting input (red trace). That is a less meaningful thing to be doing, because voltages inside a feedback loop can swing harshly all over the place as the amp tries to correct the stimulus, but you do see exactly the same "undershoot" going below 4.6uV at about 1.2uS


When the op-amp output at op1 is plotted, you see the real signal, with no undershoot.



When we get to op2, you have more gain (blue trace), there it is, with no shenanigans. The red trace scale is now overwhelmed, and has to plot near zero.

Overthinking it?
I don't know. When I think something through, I burn it until I satisfactorily understand it. "Overthinking" is not really a concept I get, nor recognize as "a thing". If it exists, then "overthinking" is still not the same as "wrong thinking". So maybe I do that.
I see less problem in "overthinking" anything than, shall we say "underthinking it"

 

[homebrewed]

While reviewing the LT6268/9 DS I realized it won't be a drop-in replacement for the pocketgeiger's LMC662 because the LT626X abs max for Vcc-Vee = 5.1V. The LDO on the board outputs 9V, but there is a pin-compatible 5 volt version available. I don't know if the comparators can go that low, but it doesn't matter in this application.

 

[graham-xrf]

I saw those issues - but I was into major destruction anyway. I never thought to use those supplies anyway. Reluctantly, the convenience of a USB (or Raspberry's) 5V supply, cleaned up with little common mode chokes and filters, feeding a low noise LDO for around 4.7V, and a reference at 4.096V, seemed to be a plan.

Another variant tried for 3.3V from a battery cell. I never considered the 9V for anything, and whatever voltage comes out of the switcher gets binned, along with the switcher components. It promises to work just fine using the positive supply for bias, with lower dark current noise. Even with 2V bias, and a big capacitance, it still works.

Swiftly glancing at it - do I see comparator pin-outs are actually compatible with standard op-amps?
That is a open invitation to replace the comparators with a op-amp pair, and cut and lash up circuits around it. For me, that allows a differential driver which includes noise filtering, and can even supply some gain.

I now see there is some development value in parts of that little circuit board, but for me, the question is marginal. Except I got distracted into something at the time, the board very nearly got the heatgun treatment to get the photodiode off, and the hot remains dropped into the trash!

One thing I have confirmed is Pin 7 of U3B does indeed go to a junction of R18 and C16, and the other side of C16 is on GND.
I don't think it is a fumble. It was intentional, even if plain wrong! If C16 was supposed to be on the other side of R18, then 1/RC = 452693.5
In this circuit, that would be ambition indeed!

I get it that 470R in a wire at a audio jack was to squash the ringing from a comparator transition.
I am open to suggestions about what the capacitor was ever supposed to do.
Maybe the heatgun is again under consideration.

 

[homebrewed]

I saw those issues - but I was into major destruction anyway. I never thought to use those supplies anyway. Reluctantly, the convenience of a USB (or Raspberry's) 5V supply, cleaned up with little common mode chokes and filters, feeding a low noise LDO for around 4.7V, and a reference at 4.096V, seemed to be a plan.

Another variant tried for 3.3V from a battery cell. I never considered the 9V for anything, and whatever voltage comes out of the switcher gets binned, along with the switcher components. It promises to work just fine using the positive supply for bias, with lower dark current noise. Even with 2V bias, and a big capacitance, it still works.

Swiftly glancing at it - do I see comparator pin-outs are actually compatible with standard op-amps?
That is a open invitation to replace the comparators with a op-amp pair, and cut and lash up circuits around it. For me, that allows a differential driver which includes noise filtering, and can even supply some gain.

I now see there is some development value in parts of that little circuit board, but for me, the question is marginal. Except I got distracted into something at the time, the board very nearly got the heatgun treatment to get the photodiode off, and the hot remains dropped into the trash!

One thing I have confirmed is Pin 7 of U3B does indeed go to a junction of R18 and C16, and the other side of C16 is on GND.
I don't think it is a fumble. It was intentional, even if plain wrong! If C16 was supposed to be on the other side of R18, then 1/RC = 452693.5
In this circuit, that would be ambition indeed!

I get it that 470R in a wire at a audio jack was to squash the ringing from a comparator transition.
I am open to suggestions about what the capacitor was ever supposed to do.
Maybe the heatgun is again under consideration.

I'm holding off on the transplant-surgery approach until I see just how broken the pocketgeiger is w/regard to an XRF application. I've got my suspicions though. The similar pinout for comparators vs amplifiers probably is no accident -- I'm willing to bet that the first "comparators" WERE op-amps. I've certainly used 'em for that in the past, at least for low-speed things; but they aren't very nice if you want to interface them to digital circuitry.

Re-using the pocketgeiger board does have some value if we consider that other interested folks may not have the resources or experience we do w/regard to circuit surgery. Just something to keep in mind.

R18/C16 may be there to slow the comparator's output down a little, or to tame an oscillation problem; or to deal with something on the receiver end of that line. The LM393's output is open-collector so the capacitor will produce a relatively slow transition from "0" to "1" and a relatively fast transition from "1" to "0" when the output transistor turns on. If we consider it as a voltage reset circuit, maybe it's some sort of pulse-pileup detection scheme? Missing pulse detector?? A simple one-shot pulse stretcher (my vote is for this one)?? If so, the design is a bit more elegant than first thought. Which of these really depends on what sort of pullup circuit is on the receiver end. Arduino digital inputs can be configured to place a 20K pullup resistor on them, so the effective time constant is much longer, 94uS. That's long enough for a simple polling loop to detect a "0" pulse, assuming you don't want to write an ISR to do it. I also had wondered why the main signal comparator was wired to produce a negative-going transition when radiation was detected, and that's probably why. This approach would establish an upper limit to the count rate, beyond which the apparent count rate saturates. A cell phone would likely have a different pullup and most certainly a different polling rate so I have no idea how all this translates to a cell phone app/hardware.

I had initially wondered how the comparator outputs could be compatible with an Arduino's digital inputs, since the pocketgeiger's internal supply voltage is 9V...until I realized that the comparator outputs are open-collector. I'm embarrassed I didn't figure that one out sooner.

 

[homebrewed]

I just bought a set of 10mmx10mm metal blocks to use as references for calibrating the MCA S/W. They are aimed at folks who want to collect elemental samples, but (if pure) will work fine for this application as well. The blocks are: iron, chromium, zinc, cadmium and titanium. With the exception of cadmium, that pretty much spans the range of energies expected for ferrous alloys. Amazon, $32.95.

 

[homebrewed]

I had another thought regarding the mechanical portion of the XRF. Folks may recall a discussion about using energy filters to improve the resolving power of the system, with the most useful being a thin sheet of manganese or iron. For this to work, the sheet cannot be exposed to the primary 60Kev gammas from the Am241, or the spectrum will be "contaminated" by a manganese or iron peak. For convenience, it must be relatively easy to swap out filters -- suggesting a kind of filter holder on the outside of the lead shield -- but, obviously, it still must be shielded from the gammas.

Also regarding the energy filters, I did some math to see if there is an optimum thickness for the filter in order to achieve the best effective resolution -- in the case of a manganese filter, to distinguish iron and cobalt. The filter preferentially absorbs the cobalt peak. It turns out that the ratio of iron to cobalt counts improves without limit as the thickness of the filter increases -- at the expense of lower and lower count rates. So there IS an optimum, and it is a balance between count rate and energy resolution. This can be messed with some, by using more Am241 sources to increase the primary gamma flux. So I don't have an actual number for that ideal thickness.