Needing more than a spark test?

[graham-xrf]

If you took that as anything more than tounge in cheek I apologize! I'm well aware of the low levels you're dealing with, and how little clue the general public actually has. I hold an NRC license to operate a reactor.
My first thought in this thread was that you'd get most your components just taking an old detector that was being surplused out of a power plant, but it appears theres not much luck to be had with the internal components after sitting on the shelf for 40 years. General Atomics will gladly make you new, for dumptrucks of cash it appears

Hey! That's neat! I've never seen that graphic in Seivert before, only REM.

Things that caught my interest
One can find various descriptions of the the 1.6kg "Demon Core" of plutonium that killed Harry Daghlian, and later, Louis Slotin (1945/1946). Harry took an estimated 5.1 Sieverts from the instantaneous blue flash, and wave of intense heat. He died after 25 days.
The Slotin accident exposed Louis to about 10 Sieverts. He knew he was a walking dead man, and he died 9 days later. I would guess that might be the highest dose anyone has ever received!

Recently, I have been somewhat distracted by the goings on in Ukraine involving the nuclear reactors, and the attentions of the Russians.
When the Russian troops took over the Chernobyl site, they must have been operating in complete ignorance. It beggars belief that there was not enough basic knowledge right through the command chain to be aware that the entire evacuated area around Pripyat got that way for a reason!

Digging in to defensive trenches in the Red Forest, regardless the warnings, was what caused their acute radiation sickness, and hasty withdrawal. They were taken by buses to the Belarusian Radiation Medicine Center in Gomel, and I doubt the full story about their fate will be known for some time.

One of the employees at the exclusion zone management taunted them on Facebook with..
“Have you dug trenches in the Red Forest, b*tches? Now live the rest of your short life with this. There are rules for handling this area. They are mandatory because radiation is physics — it works regardless of status or shoulder ranks".

My XRF project feel?
It's great! To figure a way to use the feeble leftovers from a few smoke detector innards, re-purposed into a materials identification tool, delivers a nice first buzz. Learning some nuclear physics, low noise electronics design, and small computer programming on the way? Now that is just delightful!

 

[K30]

The Slotin accident exposed Louis to about 10 Sieverts. He knew he was a walking dead man, and he died 9 days later. I would guess that might be the highest dose anyone has ever received!

Ummm, let me dig! I think it might not be! Let me see if I can dig up the estimated dose for the guy that didn't immediately die in the SL-1 "accident". I think there might have been a higher one outside the US too, but I can't remember the circumstances.

On a funny note. I actually tried to follow the rules once, and send 5 or 6 30+ year old detectors back to the manufacturer for disposal. They had ceased operation at least a decade before...

 

[graham-xrf]

I have not posted here for a while, because of other critically important things in my life that have to come first. When I can, I keep pressing on, mostly using LTSpice, and using components from Analog Devices that I do have model data for. Also, I keep adding to my KiCAD circuit.

Regulators - suggestions?
As I get the circuit line-up gathered together, I run into some quite tricky power supply component problems. The TIA likely is able to see below the noise floor of what is coming out of the photodiode. The photodiode bias is clean, coming from batteries. Everything else needs a variety of voltages, some of which need to be pure as the driven snow. We can rely on high PSRR to some extent to help.

There are good, extremely low noise high bandwidth low dropout regulators, like (say) LT3042, but the darn thing is in a 10-lead plastic MSOP package 3mm x 3mm body, and 5mm from tip-to-tips of the leads, 0.5mm lead pitch. So - I am still looking for good low noise LDO that mortal fingers with tweezers have a sporting chance of getting soldered in.

My basic line-up allows for two transimpedance amplifier circuit configurations on the same board, depending on which components are left out, and where some resistors are replaced by zero-ohms links. It has the TIA stage, then a 3-range gain switched stage, allowing "zoom" onto low level pulses. The next stage can add more gain if needed, but it's main job is the 50Hz/60Hz deep notch filter. Last is the ADC driver, using a dual op-amp pair to make the ADC differential signal from the single ended input. Control can be from Raspberry Pi, Teensy, Beaglebone, and possibly a few others.

The ADC is the AD7622, using a 2.5V regulator devoted to its (clean) analog conversion supply. It also needs a 2.5V digital supply, and the temptation is to steal some of the analog supply via a 10 ohms resistor and a couple of capacitors. I have a dedicated ADM7160AUJZ-2.5 regulator for 2.5V analog need, which takes from the 3.3V supply. (Yes - I get it that that is an over-long part number for a little regulator!).

What total gain does one need?
Hopefully not long now, I will post the circuit, it's simulations, and a little explanation of my thinking on it. For present entertainment, and to imply that I am still alive, I attach a couple of plots. One shows a simulated 5nA pulse TIA amplified to about 2V ADC level (red trace). The other shows the FET input TIA doing it's thing on a 45pA worth of photodiode electrons (green trace) with claimed 900pV/√Hz input referred voltage noise. I think the noise from the diode will be above that! Do not take this as final. In practice, I might use 200K (2 x 100K in series) for the TIA feedback resistor. Maybe even lower than that.

We can have all the gain we can ever want from the subsequent stages. Even if the first TIA gain was 100K from 2 x 50K, that is a power gain of 50dB, which is more than enough to get to the point the signal/noise ratio cannot be degraded by subsequent stages. The idea here is to make the noise contribution from the feedback resistor lower than from the photodiode. We use two in series to control the stray capacitance, so we can have our own low value 1pF capacitance there to keep it from oscillating

I don't (yet) know how to make LTSpice do its "integration" function. I don't really know what current pulse height will come from the photons it sees. We might be able to see less than 2KeV, if we allow some time for the count to gather at 20% probability. I never expect to see aluminum (pity). There is hope. Al is theoretically (just about) within reach at 1.5KeV at about 2% capture probability, meaning wait even longer.

Perhaps wishfully, silicon, phosphorus, and sulphur might also be in range. Most other heavier stuff should display OK.

 

[homebrewed]

Yeah, there's been a bit of a silence from all involved parties. Summertime is not a good time for stuff like this because we know the nice weather only lasts for so long. Myself, between the garden, house-related items and the occasional hike, right now there isn't much time left over for the other stuff. But summer is waning so I should be able to return to some other fun projects.

One way around your assembly-related problem might be to roll your own regulator(s) using low-noise op-amps. Based on my browsing through DigiKey and the like, it appears to me that high-performance regulators cost about as much (or more) as good op-amps --- so why not? You could have a "master" regulator and resistive dividers off that to provide lower voltages. Good bypass capacitors will drop the thermal noise from the dividers. If you worry about such things

On a slightly different note, I also have noticed that many so-called "low noise" regulators are pretty lame compared to decent op-amps. 'Course, one could say the same about vendor claims about many of their op-amps...

 

[graham-xrf]

Yeah, there's been a bit of a silence from all involved parties. Summertime is not a good time for stuff like this because we know the nice weather only lasts for so long. Myself, between the garden, house-related items and the occasional hike, right now there isn't much time left over for the other stuff. But summer is waning so I should be able to return to some other fun projects.

One way around your assembly-related problem might be to roll your own regulator(s) using low-noise op-amps. Based on my browsing through DigiKey and the like, it appears to me that high-performance regulators cost about as much (or more) as good op-amps --- so why not? You could have a "master" regulator and resistive dividers off that to provide lower voltages. Good bypass capacitors will drop the thermal noise from the dividers. If you worry about such things

On a slightly different note, I also have noticed that many so-called "low noise" regulators are pretty lame compared to decent op-amps. 'Course, one could say the same about vendor claims about many of their op-amps...

Yes - I have done this before, and I agree that this could be a convenient way. Hanging a follower op-amp between the supply terminals, and connecting the input to a voltage divider using two resistors. There is enough current availability from any one of the low noise op-amps. The resistors set the output to drive the 0V reference, so making available a positive, and a negative to use for the other op-amps supplies. They need not be symmetrical. One can set the "negative" supply to be (say) -0.8V, and the "positive" end to be (say) 2.5V. That way, one can have inverting stages without saturating at the rail, all taken from a 3.3V master regulator.

The tiny regulator I mentioned happens to be exceptional. It has 800uV noise, and 1MHz bandwidth, 200mA current rating, and is capable of cleaning up switched regulator supplies. It's just too darn small to be attractive.

I don't know how much gain to use
The 16 bits has 65536 counts. Allowing the unknown threshold for the least significant bit, then 32768 states can be had with certainty. That would be 90.3dB dynamic range. The data sheet claims 92.5dB (but I don't know how that happens). The SINAD is claimed 91dB minimum.

So, thinking about the range needed. The biggest pulse we would ever be curious about is a direct hit from a smoke detector. 60KeV.
The smallest energy that we can reasonably expect the photodiode to respond to is about 1.5KeV.
Thus the range is 20*log(60/1.5) = 32dB
Therefore, we don't need any gain range switching. The ADC should easily manage the range the photodiode has, and more. We only need set the gain to deliver about 2V for the ADC when it sees 60KeV.

 

[homebrewed]

We use two in series to control the stray capacitance, so we can have our own low value 1pF capacitance there to keep it from oscillating

I used a similar approach in an active probe I designed for probing internal nodes of RF IC's. For that one I used three series-connected 0201-sized resistors. It worked reasonably well but the technician who assembled them said those resistors were a PITA to solder down.

 

[WobblyHand]

@graham-xrf you may find that the better newer components are only available in tiny packages. Since you are making a PCB (from the KiCad reference above) you might consider the boards being all SMD and just having it assembled. Personally, I wouldn't even attempt 0201 stuff, about the smallest these days I would consider for hand work is 0805, or maybe 0603. Fine motor control and eyesight are not where they used to be... For the critical parts, maybe consider hand assembly using very slow hot air or an oven using a lower temp solder.

 

[WobblyHand]

Just recently taught myself KiCad and did my first design. What a lot of stuff to learn to get things done! It was a simple through hole design, for a home made electronic lead screw controller, but it taught me a lot. In ancient times, for work, I did a few PCB designs, but I had access to rules based auto-routers and other high end tools. DIY hand routing is an art. Fortunately, remembered enough from 40 years ago to get through the design. I had the PCBs fabricated recently. Just waiting for them to arrive. Next design will be primarily SMD, since the parts availability is so much better. Also the smaller parts make it easier to make more compact designs. Those DIP packages take up a heck of a lot of room!

 

[graham-xrf]

I guess I do agree with you in general. One gets to use the nice goodies if they are put down on a PCB. It's just that, when one is messing with the first prototypes, it is necessary to be able to work on them. Given the cost of things like ADCs, I can't assemble a PCB, have a problem with it, and start from scratch with a new one in the hope of accidentally coming across why the first did not work! Diagnosing requires dissasembly.

Since I will have a KiCAD circuit, it does make sense to follow through to the PCB, and then sure , one can site the components, and when the solder paste melts, they "pull" exactly into place. The thing is, unlike Mark, who at least has powered up a modified pocket geiger, all my "experimenting" has been on separated circuit fragments. I need to at least get the first together, so I was thinking to try for manageable component sizes where possible.

Being able to mess with it - and variations.
I had in mind to have the thing so that others could use the PCB, with options to change things on it.
We can have a straight op-amp TIA, or we can add the JFET version, That trick also comes in several flavours.

I used a 1.5nF coupling capacitor to be able to bias the photodiode without causing offsets at the input of an extreme gain amplifier. The simulations showed it worked without need for all sorts of return overshoot and baseline compensations. Even so, I saw how it was done for Hamamatsu photodiodes using LTC6244HV. There it is in Fig 6a, a way of connecting with DC coupling so that you can have arbitrary bias VBB. I could not get it to work in simulation, and the circuit I chose has lower noise anyway. Regardless, the layout can still have the circuit, just by putting in 0-ohms links in place of some of the resistors.



Further in considering a basic PCB which is capable of accepting "experimental variations", I had in mind to have headers so that a user can choose to interface with Teensy, Beaglebone, or Raspberry Pi. Maybe having a header separated from the works by a means to have wire links to get to any interface pinout would be the most flexible. It does not have to be this way, but a PCB might be as many as will step and repeat on 12" x 12", and I don't mind sending some to anyone who wants to play. I will only need a couple. Hence, trying for bigger bits where possible.

I deliberately purchased a some LTC6268, which is a single op-amp package in SO8, just because it then becomes possible to lay out a guard ring at the input, and you even get given a guard ring recommended layout. Of course, the other thing I was thinking was, an SO8 can also be handled by fingers like mine.



The gain calculation.
Someone who is used to teaching sciency stuff had better check my calculation.
When one is dealing with a photon with energy in KeV, you don't get to know the currents involved straight away. It all depends on the capacitance those electrons are spread into. 60KeV would be 60,000 electrons on a capacitor with one volt across it, or 30,000 electrons with 2V, and so on. 60KeV energy is 9.613059804E-15 Joules. We simply don't know how many electrons were involved, without knowing a voltage.

Those Joules ended up in about 230pF capacitor, before getting leaked away into driving a TIA.
Energy in the capacitor has equivalent expressions (C/2) * V^2 = (Q*V)/2 = Q^2/(2*C).
It is the last of those that lets us in. Charge Q also equals current x time. We actually have an approximately triangular pulse, with base time about about 13uS. We can get to the area under the curve using 1/2 base x height. Suppose the pulse peak is Ip, and energy was E joules.
Q~= 6.5uS x Ip where the 6.5uS came from being half of 13uS

E = ((6.5E-6 * Ip)^2)/(2*C) where C was 230pF

then 6.5e-6*Ip = √(E*2*C), and Ip = √(E*2*C)/6.5e-6
Doing the arithmetic, Ip = 3.2351e-7A, which is better said as 323.5 nA.

Wow!, My 5nA simulation even is way too small. I think that is huge! A third of a micro-amp!

We can now get at least a reasonable estimate of the components to deliver the required gain, and make a final adjustment when we have a chance to assault the photodiode with a direct view of a smoke detector.
Something feels wrong! We don't need as much gain as I have been thinking about,
If I fumbled any of that, please do tell.

 

[WobblyHand]

E = 9.613e-15 J
t = 13e-6
C = 230e-12
Q^2 = E * 2 * C
Q = sqrt(E*2*C) = 2.1e-12 coulombs
Q = i * t
i = Q/(t/2) = Q/6.5e-6 = 2.1e-12/6.5e-6 = 3.235e-7A

3.235e-7A * 1e9 nA/A = 323.5 nA = 0.3235 uA

You slipped 3 orders of magnitude. This is assuming the numbers are as you stated. If this is true, you have lots of current to play with, a veritable walk in the park!