Needing more than a spark test?

[graham-xrf]

As for this application, I'm just trying to avoid feature creep before I even know if the basic approach is going to be good enough or not. My signal conditioning board really isn't much of a step up from a jumper-wire-infested bread board so I know it's not "production ready" by any stretch of the imagination. At this point I'm willing to live with some warts, even if they're big ones .

Agreed - and yes, the military design background has shaped my life. Even before that. it was aviation, which also had to be the best.
I had thought of adjustables as extras. If I could find an A/D with suitable op-amps on board, it would be lovely. No extra features here. Bonkers bare-bones leaves out the USB interface. I think the interface is perhaps a "feature" we need.

I was thinking only three main ICs besides the main sensor. Come to that, I am hoping your trawl that discovered the MX100 - 7 was reasonably extensive. Be sure that if we discover a $35 or $20 PIN diode example, then it gets considered. The three devices would be a quad op-amp, and a 16-bit A/D converter, and a tiny something, PIC, whatever, responsible for driving a USB connector, and obeying whatever the driving program asks. That would be one version.

I had thought a more expensive development version could leave out the USB interface (so fractionally cheaper), and be driven directly from a (in my case) Raspberry Pi GPIO. It could be Arduino, or any other small thing that can at least wiggle the SDI stuff. The Pi delivers the OS, the computing, the screen display - everything! Hardware-wise, it's the cost of a Pi, + cooling fan heatsink, + screen, and all the bits that add up to a Pi-4 desktop, additional to the XRF front-end,

After consideration, I think one design only, with the high speed USB, is better. The program that works it from a Raspberry Pi can have a variant that runs in Android. I don't know how apps are done for iPhones. If you and I are the only ones doing this, then we don't much care. If it gets a wider liking, then it only stays attractive if it is one design, with the same basic program functions, though eventually implemented to suit some different phone OS. This is how it becomes useful to many more folk.

If we get it going, just for ourselves, we can appeal to others to pitch in software interfaces for the various smartphones, and computers out there.
XRF identification kit is normally into the several $K. I have no idea how much HM members care about knowing what steel is in their lathe before they start. Maybe they already know, because they always bought it, and kept it marked. Who tries to use any bit of material in the junk box that happens to have enough volume? Me actually!

For the present, you and I can experiment. The diode was in the little Geiger counter pulser. Has anybody used it for XRF?

 

[homebrewed]

For the present, you and I can experiment. The diode was in the little Geiger counter pulser. Has anybody used it for XRF?

I found two blog posts on the physicsopenlab web site that feature experimental results with silicon PIN diodes and XRF. This one features a Hamamatsu S1223 photodiode and circuitry based on a couple of demo boards from TI. They used Americium-241 x-ray emissions to test the system's response. I believe they used a sound-card based MCA to analyze the pulses. And here are their notes on using the X100-7, but in combination with a couple of off-the-shelf electronic modules for the analog signal processing. They did not include schematics of the modules, nor did they describe how the X100-7 was biased.

In the latter post they state that the relatively high capacitance of the X100-7 makes it unsuitable for x-ray spectroscopy, but continue on to show results that say otherwise. The S1223's junction C is 10pF vs. 40pF for the X100-7 (with bias applied). I'm a bit skeptical about the assertion regarding the disadvantages of higher junction capacitance, because the input to the first-stage amplifier, whether you call it a CSA or TIA, is basically a virtual ground. Zero voltage change across the diode capacitance = no current lost to charging the capacitance (for a theoretically-perfect amplifier anyway). This being said, the "stiffness" of the bias supply then becomes important. They show a 100M series R in the S1223 test circuit, but we don't know what the bias circuit looked like for their X100-7 circuit. The pocketgeiger bias circuit places a 1uF capacitor on the input side of the detector so it should look pretty stiff relative to the short current pulses. This is a good thing. The bias network used for the S1223 is, in my opinion, not optimal.

Interestingly (and confirming my own hunch), they indicate that you want to use a high detector bias voltage in order to increase the depletion region width as much as possible -- to maximize the collection efficiency. This also minimizes the junction capacitance, also good -- but at the cost of increased dark current. I've been mulling over the idea of cooling the detector using a thermoelectric cooler stack to reduce thermally generated current, but the scheme has its drawbacks once the temperature drops below the dew point. It would require a sealed chamber for the detector and dessicant (or a source of dry gas like CDA or N2), all for some as-yet-unknown degree of improvement in dark current. And it would need a significant step-up in the complexity of the mechanical components in the detector system. Temperature regulation and dark current offset compensation would be easier to implement. I probably would still use a TE cooler, but the set point would only be a few degrees below ambient.

BTW, I just did a search to see who stocks the Hamamatsu part. Newark has 'em for a bit over $10USD in single quantities. I also found some on ebay, but (as usual) some of them are priced significantly higher than Newark. Thorlabs has a model FDS1010 that is a 10x10mm detector, single quantity price is $55.73USD (not including shipping). Its dark current is specified to be 600nA @ 5V, and its junction capacitance is very high compared to the X100-7. Since its capacitance is so much higher but its area is the same, I don't think it is a PIN diode (and likely not a good candidate for XRF). Their FDS100 photodiode looks comparable to the Hamamatsu, but costs a bit more (but larger in area compared to the S1223). It is 13mm^2 in area, the same as the Hamamatsu S1223-01.

The disadvantage of using photodiodes is the necessity of keeping light out. The optical window over the S1223 (and FDS100) also might be a bit problematic for lower-energy X-rays. But for $10-15 a pop it might be worth experimenting with ways to remove the window. A mill with a CNC setup makes it easy. We had a CNC'd Sherline mini mill at work that worked like a champ for this kind of thing. It only took about a dozen lines of g code to move a small end mill in a circular path -- but now that I'm retired I don't have access to the Sherline any longer.

 

[graham-xrf]

We need not worry about keeping light out of any good (meaning low cost) candidate. cover in 35um Al kitchen foil should work! The aluminum pot down the middle of the PMT scintillator is much thicker than that!
[I still like your suggestion of putting a lead disc down there, and force it to only collect sideways scatter]!

I have been trawling. e.g. -->HERE
It's a small thing, for less than a buck .. BUT .. it links all over the place to circuits.
--> CCT HERE

It's actually a Fairchild thing (so from the past). A good read about capacitance, and sensitivity and noise trade-offs, and a waveform.
I am not done with this, but am called away. I will get back sometime today.

 

[graham-xrf]

Re: A/D converter cost, and other passing stuff

Not in detail (yet), but I am interested in what you would think appropriate for the A/D cost.
I have been selecting from Linear Tech / Analog Devices, and I was punting for something between $7 and $20, with much attempt to keep it below $14.

There are some temptations to Linear Technology devices because they seem lower cost, are good stuff, and we get the simulation models, but I am considering all I can find. We can use devices generally up to 10MS/sec in that price range.

Trading off bandwidth, and other factors while selecting, I consider the pulse duration, and the reasonable bandwidth we need to sample it. I know that happens with PMTs. We can expect pulse widths of 3uS to maybe 10 uS. From the fastest risetimes and spectral content, I multiply by 2.8 to get a crude minimum estimate (loosely applying Shannon), and then add some. Taking (say) 20 samples out of a 3uS pulse is sampling at 6.6Ms/sec It sets the upper reasonable limit at about 10Msps. There are excellent and affordable A/D at 2Ms/sec, but these could only get a coarse shape representation, say about 6 samples out of the very fastest pulse decays.

PIN diodes go slower .. I think.
Maybe not OK for a PMT, but is a PIN diode is a good deal slower, then we are probably OK for A/D costs.

We can use slower A/D (500kHz) if we filter the hell out of the pulse blob first, but we lose level, and energy information. Everything I have tried about "pulse stretching" shows that the stretched pulse amplitude and duration is still somewhat related to the energy that caused it, but is at a much lower level, and is a distorted version, this done only to still have something to offer to lower speed (audio) samplers.

I don't know what the PIN diode response time and carrier recombination times would be. Still trawling that!
I sneak out a hint from the first minute of --> HERE from about 0:28 to about 0:42, with the info at 0:36 revealing that the RD2014 device does 40uS to 150uS, depending on energy level.



Something that slow could be sampled accurately at 500kHz, and easily allows S/N ratios beyond 88dB or maybe 92dB if an affordable 1MS/Sec A/D was used. I would prefer that the back end be good, even for PMTs, and that is still actually affordable, but perhaps you want to do something different. I take into account that this is a thresholded Schmidt-triggered style TTL around the real pulse.

Still hazy - what exactly reads the A/D, and exactly how does it ship the measurements to the main computing and display platform?
Low power PIC? Power for the interface, and the digital power pin for the A/D can be taken from the USB. Only the instrumentation amplifier, and the A/D reference end need use a battery. i.e. isolated.

It occurs to me that a Raspberry Pi Zero is £4.80 and has the ready made USB to get at something with a screen. The W version with WiFi and BlueTooth on board is about $10, and has enough hit to do all the computing too. The amplifiers and A/D board can live direct connected to the GPIO. It also has built-in USB for the option of hooking to a laptop or phone or other PC by cable. That little 5 bucks computer has 1080p HDMI video on board as well. One micro-USB is a power connector, and the other USB is for keyboard/mouse or whatever you want.

I imagine a $5 Pi-zero version, taking it's power from the USB lead. That may be too much to ask of a phone. Another concept has it using it's own 5V power supply adapter - a USB charger would do. All sorts of ways this could go. For the present, I am not sold on anything. I might just use the Pi-4 I already have, for now.

In any event, do we have a shot at keeping the components to under $100?
Would that be likely OK for for folk who might want to attempt this?
I am thinking here that folk who do play can dream up all sorts of their own physical/mechanical screened measure box stuff, with one of these in the back of it.

 

[homebrewed]

I've got the feeling that the Fairchild PIN diode wouldn't work for our application -- further down in the Maxim app note they mention that the 60Kev gamma is near the bottom of the diode's detection range. Milling off some of the molding compound might improve the situation. The discussion on minimizing the noise contribution is good info.

I hadn't found the RD2014 in my online searching. I note that the intro portion of the video mentioned a different model, I believe it was an RD2007, that is sensitive to lower-energy photons -- and that is the realm we are interested in.

The frames showing the detector specifications indicates that the pulse width (not height) depends on the photon energy, so the MCA S/W would be different compared to the Theremino's pulse height discrimination. If the pulse width is proportional to the photon energy, why mess with an A/D? Gate a fast clock and send it to a counter chain.

Regarding your comment on using aluminum foil to block light, black paint might be a better choice if the pigment is carbon black (or an organic dye). Carbon's absorptivity @10Kev is about 1/5 that of aluminum. However, plain old black paint won't provide electrostatic shielding. I don't know if aquadag would go on heavy enough to block all the light, but if it does it could serve as an electrostatic shield, too. Or perhaps a double layer coating, aquadag on top of black paint? Or an electroless silver layer? A light-tight box?

I haven't found anyone on ebay (or anywhere else) that is offering RD2014 or RD2007 detectors. Mikroe sells a radiation detector based on the BG51 from Teviso (called the "Mikroe radiation click", Digikey has 'em for $134), but that particular detector's energy range doesn't go low enough. The Teviso AL53 would be better suited but so far I haven't found anyone selling them. Teviso doesn't have an online store so you'd need to contact their sales department to see if they'd sell us any. I bet they cost at least as much as the X100-7, since their part is a module with electronics in it.

 

[graham-xrf]

We are OK to just keep an eye out for detectors. X100 - 7 will do for now. It's big enough that even if we find some other goodie, it can still likely be arranged to "fit". I just keep seeing various photodiodes ID numbers in papers and literature that, given the date of the publications, makes them very old. I check, and yet I find they are still available. eg Microsemi's glass axial leaded UM9441 can be had from Mouser for $108.81 (not that we ever want this thing). It delivers photo-currents in typically 6mA in 10nS from 2.5MeV Flash X-Ray, and the spec sheet test condition was 10^6 rads(Si)/sec.

To some extent, we just have to build it and test to see what it does. While my SPICE simulation is coming together OK, I don't theorize it to death before building. I do just enough to get some idea of response speed, noise, etc. so as to choose the nice parts, and explore various candidate circuits, and get through the major decisions to give the proto a fair chance.

Gain?
I am hoping a fixed gain would be OK, but I do like your selectable gain feature.
Anyway, it prompts that I at least try for a first-order guess at the signal range needed. When huge dynamic range is needed, I usually go dB logarithmic. but I don't think anything like that is needed for this. When signals will range over several orders of magnitude, the electronics does not care. It's only us humans that need dBs.

My point is, if there is need for switching gains, then log compression amplifiers are the automatic way to do it. Given the range 2KeV to 60KeV, I am thinking we don't need programmable gain. we only need the "right" gain. I get it that not all of the incoming energy turns into photo-current. Some gets lost heating up the package and leads, carbon black paint, whatever. One hopes the absorption probability curve for X100 - 7 is right. It scrapes 100% at some energies.

The frames showing the detector specifications indicates that the pulse width (not height) depends on the photon energy, so the MCA S/W would be different compared to the Theremino's pulse height discrimination. If the pulse width is proportional to the photon energy, why mess with an A/D? Gate a fast clock and send it to a counter chain.

Hmm.. OK , even before I get my SPICE model finished, let us address this.
I think that when a photon arrives at a PIN diode, it will put energy into a TIA gain stage.
That comes as a current. It will cause a amplifier voltage to "go up".

I think what you are saying is the X100 - 7 has a specification behaviour whereby the energy absorbed delivers a current pulse related to the charge in a way that current rises to a value, and then stays at that value for a time, then drops away. Yes it does, but every description I have seen does not feature a plateau in the current waveform. Maybe X100 - 7 really does that, but that does not mean we ignore the amplitude.
Perhaps point me to the frames showing detector specifications that show this.

There are many circuits that, if sufficiently filtered, bandwidth limited, integrating, non-linear, whatever, can turn the transient into something with a flat top and of longer duration. The "pulse-stretcher" idea even depends on it.

What we measure
For us, we know that the voltage we measure is an analogue of the current pulse. It represents the rise and decay of what the photon delivered into the depletion region. The energy in the pulse is the integral of the current x time. The energy we want to measure is, as I am understanding it, accurately proportional to the area under the curve of the pulse.

Getting at the energy (accurately)
If we want a number to put in a bucket, we don't just try for the pulse maximum, nor do we just measure a duration.
Instead, we add up all the values of all the valid samples of the pulse duration. That represents the energy.

The "duration" is when the measure was above a threshold caused by the dark current.
If a value is "unrealistic", or a duration too long, it can be weeded out, or simply put in a bucket anyway, and let itself be seen as statistically unlikely. Either way, it gets recognized!

Even if we offset the majority of the dark current out of the measure before A/D, we need the dark number subtracted anyway.

PIN Diode Model
So far as I can tell, the equivalent circuit model for the PIN diode is the same as for a regular Si photodiode (for visible light). In PIN diodes, the undoped "I" region is still devoid of carriers when reverse biased. It's just that it is thick enough that incoming photons don't just go straight through without stopping. I am still getting myself sure of the model.

I know we have a high impedance charge amplifier. It has to be that way because the charge is tiny, but I don't treat it as a high impedance situation at the TIA input. The 40M resistance is not in series with the charge source. It is a shunt resistance. The diode is presented with what looks like a near ideal zero impedance load. The op-amp does not allow it's inputs to develop volts between them. All the charge current goes into the freedback resistor, and the op-amp obligingly makes its output voltage exactly what it needs to be to accept this current.

From the Hamamatsu chapter2



You can see from equation (1) that the wanted charge current IL has to be seen as a change in the face of an existing diode dark current ID and I'
A version of this is what drives my SPICE simulation.

If the M100 - 7 delivery of energy back from the diode is a flat top duration-type pulse as the intrinsic region gives up its carriers and becomes depleted again, then so be it, but that makes the "pulse" unlike most others I have seen in what I have read. I am keeping in mind that this is PIN diode XRF, not PMT electron tube pulses. I don't think it matters, so long as we count the energy. I am not too keen on assuming we only need count a duration.

 

[homebrewed]

If the M100 - 7 delivery of energy back from the diode is a flat top duration-type pulse as the intrinsic region gives up its carriers and becomes depleted again, then so be it, but that makes the "pulse" unlike most others I have seen in what I have read. I am keeping in mind that this is PIN diode XRF, not PMT electron tube pulses. I don't think it matters, so long as we count the energy. I am not too keen on assuming we only need count a duration.

Sorry if I wasn't all that clear. I was looking at the youtube video frames that showed the RD2014 specifications, not the X100-7. The spec indicates that the pulse height coming out of the module is a TTL level signal whose width varies according to the photon energy. I expect the analog signal going into the pocketgeiger's comparators to exhibit pulse height variations w/respect to photon energy.

Out of curiosity I took a look at the UM9441. The spec sheet doesn't indicate the diode area, and the package type doesn't really lend itself to detecting lower-energy X-rays. So I think we have been spared the need to spend $108 more on this project.

Regarding your simulation work, you will likely need to make some guesses on some of the parameters. The X100-7 spec sheet gives you the diode area and capacitance. The saturation current Is can be derived from the dark current vs. Vr curve, and they actually give you the shunt resistance (40Mohms). The forward resistance is the biggest unknown. If you have a SPICE model for a different PIN diode and know (or can derive) the area of the diode you might be able to scale Rs accordingly. The most straightforward approach would be to make the (rash?) assumption that the two diode's dark currents only differ because their areas are different. This because the saturation current in the standard diode model is proportional to the area of the diode. So by using yet another rash assumption, that Rs only depends on the area, wallah.

 

[graham-xrf]

Ah --ha. I gotta say good morning. It has to be near 08:00 where you are. It's getting on 16:00 here. I have just come back in from a tree-pulling scene. The rest of the trees and rooted shrubs came out with a chain and a tug from a short wheelbase 4WD Land Rover Defender. One, which is not even a proper tree, refused. It's going to take some digging and/or the tractor and a bigger chain! The outbuilding area is now clear. Farmer G said I should put the building where I want to, then we place the boundary to suit the regulations, and agree it. Nice neighbor!

OK on the flat level pulse. So it was a TTL thresholded thing, with duration somewhat related to the energy. I feel better now!

Re: the X100 - 7 dark current. We are OK on that, because the measured values are in the data sheet plot. It depends on the bias, and my model allows to vary the bias, which is a quite important parameter.

The shunt resistance is 40M, which for the kit we are using, may be considered (low). That one is a major noise contributor.

On the forward resistance, we don't care. It is going to have reverse bias anyway.

The bias
One option is to have it operate at a tiny reverse bias, and big 350pF capacitance. Reverse bias between there, and about -30V is the range, and somewhere in there is the sweet spot. Choosing (say) 2V lets the gadget operate from one 3.3V 2032 Li cell. OK , make it two cells, and go for 5V or 6V. Perhaps we need to stand the cost of a micropower regulator, but I am hoping we can "borrow" the precision clean band-gap reference from the A/D, isolate it and use it for the bias. Anything else needs more components.

I decided we need not tangle with the 20fA electrometer grade ADA4530-1, with the built-in internal driven guard rings, low gain-bandwidth product (2MHz) and £19-20 price tag. It's noise is greater than some cheaper chips anyway.

I am still perusing candidates, like LTC6240 (LTC6241 and 6242 also). The industrial (to 85°C) quad version is £4.77, or £2.72 if you buy 25.
There are others, like AD8615 at £1.71/1 or £1.53/25
I think I am OK with a 5 bucks chip for the gain and buffering, and maybe in a quad version, can do differential drive of the ADC.
If we get lucky, this kit is a 2-chip thing + the sensor. It needs another "something", and a bunch of software to have the "USB to any phone" feature.

Also mulling over various circuits, but not much longer.

[Early morning, unless of course, you are on MT. I thought yours was Pacific time]

 

[graham-xrf]

X100 - 7 data sheet does not mention quantum efficiency. Without that, it's kind of hard to get at how much photo-current happens when an incoming photon hits. I am assuming we would have the occasional hit, making a pulse, however many per second, instead of an "illumination intensity".

The low probability hits may not happen as often, but when they do, they deliver the energy. The probability curve is not something that affects amplitudes.

 

[graham-xrf]

I have been messing with it a bit, and trawling circuits and parts costs.
A first feel for the beginnings - the first TIA stage.

Gain-Bandwidth Product.
It turns out that this is very important. The photon-generated carriers make a pulse that first has amplitude that depends in the PIN diode capacitance. The shape is the same, but smaller if the capacitance is high. To get lower capacitance, you have to wind up the bias, which brings on the noise. There is a compromise to be found.

To get the very high gain, while still preserving the pulse shape information, one needs to use high bandwidth amplifiers, or split the task over several stages. Using Rf =10M in the first stage is not so good, because even a very fast op-amp would deliver a very slugged pulse shape. This does not prevent may designs out there to have the diode bias isolated by using capacitive DC blocking, and very high Rf in a first op-amp with GBW way too low, with too much phase lag, and several pF across the feedback resistor. The resulting pulse smudge is then only vaguely related to the original energy, and has lost it's bottom reference, so now subject to the pulse swinging back below it's start point. Charging the DC blocking capacitor much modifies the shape. Yuk!

It turns out that there are many low cost op-amps with pA bias, and tiny offset that can go at 10MHz GBW, or 500MHz GBW, usually costing £2 to £5 each. Having op-amps with (say) 10M GBW to deliver a nice pulse shape in 4uS can only stand a gain of 20. We try to put as much gain as possible in the first stage, consistent with having enough bandwidth to keep the approximate pulse shape. This gets the best S/N ratio.

The value of Cf (here called C2) is important!
The capacitor across the feedback resistor. It decides the first frequency roll-of, and controls the peaking (overshoot) and stability.
Even with a high bandwidth precision amplifier, the effect is striking.

Here is a (passing) example. Note: it is unlikely to be in the final circuit!
The condition is a 20nA peak diode event current. Reverse bias was set to about -6V, to get a relatively unwelcome 100pF. I have not put in the noise generator yet. V3 is just a zero volts source, to let me measure the input current. The LTC6269 is way too expensive at about $12, but that does not matter for what I am exploring. We use something else in the end.



The green trace is an approximation to the 20nA pulse. The several output traces are for C2 from 0.25pF to 2pF in steps. I fumble a bit. The steps are 0.5pF, starting at 0.25pF. You can also see the constant lift from the dark current. You have to imagine that will be a noisy line. The capacitor that best suits seems to be the 0.5pF, which I guess would be two 1pF in series, to help with the stray board capacitance. Time shifted, and scaled, the slopes reasonably track the pulse shape.

It is in fact handy to have the "low" 470K resistor there. To get this 500kHz bandwidth takes amplifiers with 500MHz GBW, so gain down to 1 at 350MHz. We do this example in 2 stages, but I don't yet know how much we need.
We can do one like this, but with different op-amps, some tricks to the bias, etc. The diode model is still coming together.

Maxim's Radiation Detector
The PDF is attached. I say at the outset that I consider it to be a *#@z! poor design! It does, however give some insights. As much as 10M for the first stage gain, with a whole 4.7pF, and the op-amp has GBW 10MHz. The first gain of 10milliion via ac-coupled 10nF would imply the bandwidth is 1Hz (do I get that wrong?), but beyond that there follows three stages of gain = 10.

So the total gain in that, for some low frequency, would be 10million x 1000. That is a ten billion voltage gain of 200dB ! The smudge of the waveform in Figure 2 has everything I hate, including the classic negative lurch. In the text, they suggest replacing C1 with a digitally trimmable MAX1474 to have gain adjustment (Huh?). There are more mistakes. eg. "A 60keV gamma from Am241 is close to the circuit's noise floor". Hmm maybe it is with their diode. They suggest R2 could be much larger than the 10M (wow!). I think all they want is a comparator pulse to signal something happened!

Now you can see why I think you and I could design these guys into the dust! Ten billion gain to end up with just over 100mV of nudge means something is wrong! Even my half a circuit has 1.5V output. I could have used a 2nA pulse, or 500pA and still done better.

I am going to try out Analog's TIA designer, if only because it has pretty plots with noise in them.

ADS and AWR Microwave office had noise generators, but they were on a "work" computer. I found an excellent article on how to trick LTSpice into making noise generators.