Needing more than a spark test?

[graham-xrf]

I truly wish I was at a concrete pouring stage. The outside work got stalled shortly after I won the "Battle of Ye Olde Tree Stump". Everything else just does not go as well as I would like, and takes longer than I thought. I have quite a lot of house internal stuff still to be done. We have had about a month's worth of rain in a few days, and it's steadily getting colder, between 10C (50F) and 14C (57F). We are into October. Next month, the heating will kick in, if not before. Halloween usually sees frost out back.

Do check out the YT video. It makes the whole thing we are doing abundantly clear.

Another thing I have figured out is that amid some noise, an incoming energy will make a difference of some kind, usually a lump taking 4uS to 12uS to be over with. If this charge is grabbed by a slow enough amplifier, it becomes an integrator. Whether we use a local peak detector to capture it to measure at leisure, , or sample it to bits and roll in the software, does not much matter, so long as it looks bigger for a more energetic photon.

Provided the rise and decay of the energy in the detector is not too much modified by local filtering storage in the electronics, there is some value in thinking that the height of the pulse, and to some extent the area under it correlates with the energy of the photon. The re-combination time of the carriers in the diode does have information. We need some program smarts to dump pulses that crashed each other.

I have found a Op-Amp available as both 2-channel and 4-channel (3 to 5 bucks) that can be the transimpedance charge amp, gain stage, offset regulator, and maybe peak detector all in one. There are certainly other nice ones. So many out there!

The integrated photodiode amplifiers, some including the photo-diodes, with all the built-in digitals, are all many channels monsters, used for those medical imaging tomagraphy X-Ray detection array things. I think we may end up with at least these..

1) A 4-parts Op-Amp, perhaps with extra FET
2) An A/D conversion.
3) Some kind of back-end arrangement to get the numbers away into a small computer, unless the serial data can be taken direct from the A/D conversion IC.

It would need a small, low power uP down there if something like a USB cable is used to get the info into a smartphone. The latter is quite appealing to many who would like a gadget that does it's stuff using a phone.

Other fall-out is possibly needing some good substrate PCB, or using lacquer on regular FR4, and a guard ring around the input to take up surface currents to GND instead if to the charge input. With input resistance between 100Giga and 450Giga-Ohms, the whole circuit has to be seen as a kind of variable conductive soup. This is how you would measure the resistance of insulators!
FR4, though not exactly short-circuit, is maybe not insulating enough in this league. but maybe can be made so.

There has to be every effort to keep the output screened away from the input, preferably sampled into digits as soon as possible. This gadget has a gain so high, it will make us pay if we ignore the size of the low-frequency pickup loop we make with circuit conductors. Like an open oscilloscope probe being waved about, it will let you see the 50/60 Hz mains wiring in your house if you don't take care of it.

The digital signal return current has to be resistor limited. We can't have the hard-won input be jerked around by powerful serial pulses common mode noise. Easy stuff - but we have to anticipate it.

Having a "clean" power supply may best be done by having the critical instrumentation stuff powered from a little cell battery - like in the YT video.
This is all extremely low current stuff. A cheap 2032 Lithium cell, or two, may be all that is needed. If power is to be taken from the mother kit, then it needs isolation, local capacitor storage, low noise regulators fast enough to control out the racket, and other design care. I admit it is nice to be "USB-powered", but in this case, the instrumentation may be better served by a little battery.

The serial data route can be powered from the receiving gadget, be it phone, Raspberry Pi, Arduino, whatever. It should touch with only a single star point at the A/D converter, and with as much isolation as can be contrived. An opto-coupled A/D would be good.

I think that whatever the value seen in exploiting the audio channels A/D as a "cheap A/D already there" could be hopelessly lost it trying to get an analogue signal up a cable to the 192K music input. Exact instrumentation, even if low-cost, is, I think, too precious to be risked in that way.
I allow that others may just love using the audio channels, and I take into consideration that there may be lots of ready-made software out there.

My MX-100 has not arrived yet. When I have it, I suppose I might give it a few minutes hooked to a scope, and be shown some Am241, but from then on, it's only the diode that I want. I have not abandoned my PMT. Its a juicy goodie, and everything it has to do is known from past decades. It can hardly "not work" except in the physical arrangements around it. For the present, however, I wanted to stay more in sync with your developing kit. Honestly, it may be approaching Christmas before I truly get attempting any signals measured in a part-proto.

 

[homebrewed]

Other fall-out is possibly needing some good substrate PCB, or using lacquer on regular FR4, and a guard ring around the input to take up surface currents to GND instead if to the charge input. With input resistance between 100Giga and 450Giga-Ohms, the whole circuit has to be seen as a kind of variable conductive soup. This is how you would measure the resistance of insulators!
FR4, though not exactly short-circuit, is maybe not insulating enough in this league. but maybe can be made so.

I've done some PCB designs using guard rings. It gets a little hairy because, although guard rings are recommended by very knowledgeable app engineers at Analog Devices, LTC and the like, the package pin-outs don't lend themselves to it. I really hate the idea of routing traces _between_ package pins, just seems like a recipe for ending up with shorts where you don't want them. Instead, I've done things like lifting sensitive pins and soldering components to them. Dry air is quite resistive . Not a production-worthy approach, but this project ain't destined for that anyway.

Oh, regarding the issue of dealing with pulse pile-up. I've been thinking that the second-order peak-fitting approach might handle that. Well, actually, the squared error between the data and polynomial. A second pulse hiding underneath the main pulse will screw up the fit; so if the squared error exceeds some threshold, it has been "contaminated" by a second pulse. Or excess noise. Either one is grounds to reject that pulse.

 

[graham-xrf]

I've done some PCB designs using guard rings. It gets a little hairy because, although guard rings are recommended by very knowledgeable app engineers at Analog Devices, LTC and the like, the package pin-outs don't lend themselves to it. I really hate the idea of routing traces _between_ package pins, just seems like a recipe for ending up with shorts where you don't want them. Instead, I've done things like lifting sensitive pins and soldering components to them. Dry air is quite resistive . Not a production-worthy approach, but this project ain't destined for that anyway.

Ahh .. yes. Taken to it's logical extreme, turn the DIP or SO8 upside down, super-glued to the board. I think it is called "Dead Bug" construction. Staying right way up, and lifting one pin to a little wire that finds a ring is OK. That said, I think there is just about enough room to do this on SO8 style sizes.


I am a bit purist in having the ring around the pin only, with the only electrons arriving into it from the components directly connected, and even then, they may have to sneak in under a ground plane. Above, the guard ring is connected to the positive input. I have also seen guard rings that are "driven" from an active output.




The AD8552 is only there for a visual example, but on the way one spots that it is a "Zero Drift" thing supposedly competitive to auto-zero or chopper-stabilized amps. The little circuit with the extra JFET and anti-drift integrator, I think, would qualify for being "auto-zero". This one costs £3.80 ($4.90) from Mouser. 0.005uV/°C is quite good. 20pA bias, 700uA supply 2.7V operation. It would run off a 2032 cell, but we don't need to be mean with the pulse size. The more volts we have to sample after gain, the better. 1uV offset? It's interesting, so I attached the data sheet, but my point is, I think there is enough space between the pins, provided the SMD pads don't use it all up.

Oh, regarding the issue of dealing with pulse pile-up. I've been thinking that the second-order peak-fitting approach might handle that. Well, actually, the squared error between the data and polynomial. A second pulse hiding underneath the main pulse will screw up the fit; so if the squared error exceeds some threshold, it has been "contaminated" by a second pulse. Or excess noise. Either one is grounds to reject that pulse.

Yes - agreed. If the pulse is out of range for calibrated energies, it gets rejected. If there is a very long duration pulse, with two peaks, it may be possible to use software to detect it, and use the peaks, but why bother? Even if two (smaller) pulses arrive together, are they going to add to spoof a valid (larger) pulse? Even if they do, the bucket collection statistices will diminish the significance of the false positive. A pulse must look like a single energy event, so it needs also to fit in a time-frame, to weed out the arrive-nearly-togethers. The PyMCA and similar software may already do much of this.

I think, in principle, we may have just about got a handle this stuff!

 

[graham-xrf]

Seebeck Voltages?
You already mentioned unwanted voltages from flexing/piezo/vibration etc.

Umm.. On page 16/24 of the AD8552 data sheet, we get to address Seebeck (thermocouple) voltages from the junction of dissimilar metals, these being solder-to copper on board, and solder-to component lead. I thought that so long as one starts with copper, and ends with copper, any other metals in between would cancel (if at the same temperature). These folk use "dummy" components, to cancel the Seebeck voltages at the input!

This could be taking things a bit too far! I would hope we won't be needing stuff like this!

 

[homebrewed]

Seebeck Voltages?
You already mentioned unwanted voltages from flexing/piezo/vibration etc.

Umm.. On page 16/24 of the AD8552 data sheet, we get to address Seebeck (thermocouple) voltages from the junction of dissimilar metals, these being solder-to copper on board, and solder-to component lead. I thought that so long as one starts with copper, and ends with copper, any other metals in between would cancel (if at the same temperature). These folk use "dummy" components, to cancel the Seebeck voltages at the input!

This could be taking things a bit too far! I would hope we won't be needing stuff like this!

View attachment 339429

I hope it won't be necessary, either! This sort of thing would affect the baseline, but on a slow timescale compared to the pulses. A calibration step using a few pure elements done just before you analyze your unknown would address a lot of generic thermal drift effects. For instance, does the detector itself has an efficiency temperature coefficient? I know that the mean free path of carriers has a temperature dependency (this directly affects avalanche breakdown). Current carriers do have a finite lifetime, so the longer it takes for them to get to the anode or cathode, the fewer there are to contribute to the signal. So I'd expect the efficiency to have at least a small dependency on temperature.

BTW carrier lifetime is one reason for a high detector bias voltage, so the carriers are collected as quickly as possible. Reducing the junction capacitance is another.

 

[graham-xrf]

Zero bias - i.e. operating entirely in photovoltaic mode would be nice, but as you say, the capacitance is so high the pulse gets lost in it, and not what one needs on a charge amplifier input. For us, it is a fine line compromise.

Looking at XRFSpec_ENG.pdf, I regret the inability to have Magnesium (1.25kEV), and Aluminium (1.48KeV), but we can hardly expect to be getting returns from the stuff we use because we know it it allows X-Rays to go right through. We know that we can stop the radiation with lead, so we can use it for shielding, but is that the same thing as not expecting a nice X-ray photon pinging out of it when slammed by a gamma photon? Whatever it does, the Pb excitation energy is 74.9KeV, with about a 2% chance of it getting a X100-7 excited!

I consider all the elements 22 through 30 as the very desirable minimum set. Titanium, Vanadium, Chromium, Manganese, Iron (obviously), Cobalt, Nickel, Copper, Zinc. At the poor end, we want to see 4.5KeV. At the other end, it's 8.6KeV. The graph is about probabilities of the incoming gamma proton to hit, and clearly the interior of Aluminium must be mostly empty space, because the chance seems to be about 2% to 3%.

Unfortunately, off the left side of the curve, the low probability ones (below about 4KeV) are also the lighter ones that will deliver the least energy X-Rays when they do get hit. I haven't figured it out yet, but suppose Aluminium delivers 3% of the time, but when it does, it provides a measurable signal. I just don't know if this is realistic. X100-7 is not claimed to do much below 5KeV, or above 100KeV.



The Bias
Moving on to bias. I am not sure why carrier lifetime is a problem. Is it so slow that the junction lets loads of hits go by? I get it about the capacitance.
Fortunately, the graph scale for capacitance is no longer logarithmic, showing about 50pF at 30V bias.
The diminishing returns point happens at about 10V bias, with capacitance about 80pF.
For charge amp design, things are manageable up to about 200pF, (but one hopes for less)!

The Dark Current
Oh, the noise of it! Logarithmic scale and all. About 5nA for 30V bias, and 50pF.
At 6V bias, it is down to 1nA. The price of this is the capacitance is 100pF.
At 2V bias, the dark current is about 800fA, and we have reached the little plateau where the pips squeak, and the capacitance is about 140pF.



The stuff I don't yet know, but will be getting to soon, is how much bump we can expect from a 4.5KeV X-Ray, times the quantum efficiency of this gadget. If it is generously above 5nA, then bias away, no worries. If not, then we look to bias less.

Then I throw in the notion that, if one can make the amplifier work OK with 80pF or 100pF across it's input, there is no need to find a 30V bias!
I like it if it can work off the same battery, without a 30V generator.

1/f Low frequency Shot Noise
Unfortunately, these can look like a wanted XRF target hit, and it gets worse, the lower the frequency we go. Learning how the internal offset auto-correction circuits, already built into the good op-amps, is a relief. They do not have exponentially increasing flicker noise at lower frequencies, because low frequency noise is treated as a slowly varying offset error.

I think my first lash-up is going to have the means to initially tweak the bias.

 

[graham-xrf]

What can we expect to see?

I suppose this is a question where @RJSakowski would know, but it is about what we can expect to detect.
So we have a detector where we can push our luck a bit so that it can see from about 2KeV to 100KeV, with unequal probabilities of it doing it's thing, as per the graph in post #376

We have Am241 as the first source. It decays 5.486MeV, 5.443MeV, and 5.388MeV alphas, which can't even get through a piece of paper, but are handy for detecting smoke particles. It does have a relatively feeble gamma output at 59.54KeV.

So now we come to the materials. Looking at the Kα1 energies, we can see Tungsten (W) at 74, which glows at 59.318KeV
So, it being that teeny bit less than the gamma, does that mean we get to detect Tungsten?
The next Tungsten state Kβ1 is 67.24KeV, so cannot be excited by Am241.
But.. the Lα1 for Tungsten is only 8.397KeV, and the Lβ1 is only 9.67KeV

So my question is, does the incoming gamma energy have to first excite the "K-shell" states fully, before it gets to put excess into the "L-shell" states? Can a incoming gamma get glows from the (lower) energy L-shell states first?

I ask this because I can see something like Lead (Pb) No. 82, with K-electron states impossible to get excited by Am241, but with handy, detectable 10.55KeV and 12.61KeV responses from the L-electron states.

Besides Lead (Pb), we have others, like Gold (Au) No.79. Pretty much all the elements can get their L-electron states hiked up by energies less than 18KeV, (which could come from Am241). I say that - but is that how it works? Do the lower energy transitions in the L-electron states happen first?

[The table is in XRFSpec_ENG.pdf, attached to the previous posting #376, just above this one]

 

[homebrewed]

What can we expect to see?

I suppose this is a question where @RJSakowski would know, but it is about what we can expect to detect.
So we have a detector where we can push our luck a bit so that it can see from about 2KeV to 100KeV, with unequal probabilities of it doing it's thing, as per the graph in post #376

We have Am241 as the first source. It decays 5.486MeV, 5.443MeV, and 5.388MeV alphas, which can't even get through a piece of paper, but are handy for detecting smoke particles. It does have a relatively feeble gamma output at 59.54KeV.

So now we come to the materials. Looking at the Kα1 energies, we can see Tungsten (W) at 74, which glows at 59.318KeV
So, it being that teeny bit less than the gamma, does that mean we get to detect Tungsten?
The next Tungsten state Kβ1 is 67.24KeV, so cannot be excited by Am241.
But.. the Lα1 for Tungsten is only 8.397KeV, and the Lβ1 is only 9.67KeV

So my question is, does the incoming gamma energy have to first excite the "K-shell" states fully, before it gets to put excess into the "L-shell" states? Can a incoming gamma get glows from the (lower) energy L-shell states first?

I ask this because I can see something like Lead (Pb) No. 82, with K-electron states impossible to get excited by Am241, but with handy, detectable 10.55KeV and 12.61KeV responses from the L-electron states.

Besides Lead (Pb), we have others, like Gold (Au) No.79. Pretty much all the elements can get their L-electron states hiked up by energies less than 18KeV, (which could come from Am241). I say that - but is that how it works? Do the lower energy transitions in the L-electron states happen first?

[The table is in XRFSpec_ENG.pdf, attached to the previous posting #376, just above this one]

XRF excitation:
I believe that all possible fluorescent wavelengths can be generated, as long as the incoming photon has sufficiently high energy to excite it. IIRC, the Theremino blogs concerning their experiments indicate they are "seeing" L-alphas from some of the elements they played with.

Detector efficiency:
Detector efficiency vs. X-ray wavelength is also something that scintillators exhibit, so the real question is if lower efficiency is accompanied by a larger spread in the pulse energy distribution for a lower photon energy. If it is weakly correlated, a set of known elemental references could be used to create a compensation curve. If the sigma is strongly dependent on efficiency, well, that's a problem. I suspect that the sigma at least has some dependency -- at least, for photons that are incompletely absorbed by the detector. So let's look at the curve you show for efficiency vs photon energy in Kev's. It starts out low, climbs to a maximum around 100% and then decreases. In general, material absorptivity is inversely proportional to photon energy -- the more energetic, the lower the absorptivity. That explains the decline above 10Kev. Below 10Kev, I think the curve is dominated by absorption in the encapsulation material over the detector. To put it another way, up to 10Kev, I think that photons which make it through the molding compound are completely absorbed by the detector. That's a good thing when it comes to achieving the best statistics. The varying attenuation of x-rays getting into the detector indicates that a good set of elemental references is highly desirable, but, based on my back-of-the-envelope hand-waving, yadayada, I don't expect a significant shift in pulse height sigma vs photon energy in the range of (our) interest.

1/F:
I was thinking that shot noise might be differentiated from a "real" pulse by its shape, but that won't be the case for the heavily filtered output of the pocket geiger's analog block. That's where a wider bandwidth acquisition system could really shine, because photon-generated pulses would be pretty fast compared to 1/f noise. I'm really hoping that 1/f noise isn't a big problem: but, on the other hand, the pocket geiger comparator block has one that appears to be designed to detect noise. It's all related to the amplitude of the pulse coming out of the analog circuit, so it ALSO might be possible to differentiate the two by the amplitude of the pulse. I'd say that might be the case for some, perhaps most of the pulses -- but statistics being what they are, we're guaranteed that at least some of the noise pulses will look a lot like an x-ray photon.

Offset compensation:
My signal conditioning circuit uses a very simple offset compensation scheme -- a 10 turn trimmer. Yet again going with the KISS principle when I can.....knowing that variations in the supplied VCC/VEE will directly affect the offset. It's easy enough to address THAT particular design flaw. I can see a V1.1 version of my circuit in the future .

 

[graham-xrf]

XRF excitation:
I believe that all possible fluorescent wavelengths can be generated, as long as the incoming photon has sufficiently high energy to excite it. IIRC, the Theremino blogs concerning their experiments indicate they are "seeing" L-alphas from some of the elements they played with.

That is what I wanted to hear. The buckets with low probability samples won't be as full, but so long as they are consistent, the stuff is there, so in calibration the profile gets identified.

1/F:
I was thinking that shot noise might be differentiated from a "real" pulse by its shape, but that won't be the case for the heavily filtered output of the pocket geiger's analog block. That's where a wider bandwidth acquisition system could really shine, because photon-generated pulses would be pretty fast compared to 1/f noise.

There is value in having the instrumentation bandwidth fast enough to capture the shape. It allows (us) to recognize the good ones.

Offset compensation:
My signal conditioning circuit uses a very simple offset compensation scheme -- a 10 turn trimmer. Yet again going with the KISS principle when I can.....knowing that variations in the supplied VCC/VEE will directly affect the offset. It's easy enough to address THAT particular design flaw. I can see a V1.1 version of my circuit in the future .

I did that too - a long time ago. Pretty much all external offset compensation input circuits were a load of extra bits that needed a tweak far too often. Some types (like the original 741) had internal offset null circuits that could be fed from high precision stable resistors. I praised the day when op-amps got so good externals were not needed. My kit was military. It would get heated, and frozen, and shaken, and shocked, and vibrated to bits! I could never beat on-chip, laser-adjusted, temperature compensated ready-made offset adjust. There was never a time I needed to tweak the last nano-volt away. My goal was to have exact predictable circuit boards without any adjustments.

You get a bit messed when the design has gain circuits with programmable resistors. Watching software zip through a setup procedure to tweak the vector display for setting up a PAL colour TV camera is good fun. I get it that sometimes, one needs to adjust things. That's not the same as adjusting away an offset that should not be there. For me, that was still not hardball uncompromising enough!

If we need to set a zero, or clamp to a reference, I would still look to some automatic way. Even my pH meter, that constantly needs to be dunked into this or that buffer solution, can irk me. Sure, dunk it, and push a CAL button. But when the instructions say, "now dunk in a different solution, and tweak a trimmer down a little hole in the plastic case", then I just want to throw it in a bin!

 

[homebrewed]

Military, aerospace, telecom, medical, ATE and automotive applications are all really demanding when it comes to circuit design requirements and/or reliability. If you designed for any one of those, I'm not surprised that you abhor manual "tweaks". Hard to do that inside a chest cavity anyway, the cost of the medical-grade screwdrivers is just a killer . OK, don't beat me too hard!

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 .