Needing more than a spark test?

[graham-xrf]

Having something that can be used to hold down the SMD 0402 without it slipping about, and allow that whatever you grab it with is unobstructed can be awkward. Then I discovered that sometimes a better way is not to get a grab from the sides, but to press on it in the middle with a sharp point of a dentist's pick tool.
--> HERE
Second from the bottom in the picture. The sharp point has high friction, so stops any sideways movement, without pressing down hard at all. It allows to get at the pads with a soldering iron, way easier than working past tweezers. Most other ways seem clunky in comparison.

 

[homebrewed]

Having something that can be used to hold down the SMD 0402 without it slipping about, and allow that whatever you grab it with is unobstructed can be awkward. Then I discovered that sometimes a better way is not to get a grab from the sides, but to press on it in the middle with a sharp point of a dentist's pick tool.
--> HERE
Second from the bottom in the picture. The sharp point has high friction, so stops any sideways movement, without pressing down hard at all. It allows to get at the pads with a soldering iron, way easier than working past tweezers. Most other ways seem clunky in comparison.

I typically use 0603's, somewhat easier to handle -- especially with SMD-specific tweezers. However, sometimes the tweezers don't hold them flat against the PCB. That's where your technique could come in handy. But don't you have to change tools for each part you put down that way? A dental pick isn't going to pick the part out of the bag (or off the tape). I also can see a bit more futzing around to align the parts to their PCB footprint but I'm probably just not getting the whole picture.

One test fixture I made had to have relatively high resistance on one input, with minimum capacitance (it actually was an active probe to measure AC signals on the internal circuit nodes of IC's, up into the GHz range). I had thought that the more-widely-spaced contacts of the 1206 size would give me lower capacitance, but the 0201 quite handily beat it out. However, they really were a bear to handle and solder down.

 

[graham-xrf]

It's a two-handed affair. Right hand with tweezers puts the component down.
Then while it is in position, left hand pins it down with the dental pick point.
This gets over the problem of it shifting about from the attempted let-go.
Also, if it needs some position perfection, it's easier while using both tools.
Once in place, right hand uses soldering iron, while left is still keeping it held under the pick-point.
This method suits solder paste, unless there is enough tinning on the pad already, or you are good at carrying a smidge of solder on the iron.

More of a flap to use, and not so successful without a bit of practice, is a ghetto lash-up spring point for when you definitely need the left hand to feed in fine solder, instead of having the pre-applied paste. I used a sharpened piece of piano wire bent into a almost semi-circle, one end clamped into a PanaVise, which has a heavy base. I set the point to press on the component position without the component there. Then I use two tweezers, to lift the point, and get the component under, and into position.

At this stage, both hands are free. I used this method to hold down a 0.5mm pitch quad-pack while I secured the corners. I will admit, getting it right was fiddly, but I got it to work. I also did this to hold down a microwave FET which had tabs to be soldered to microstrip coplanar waveguide. That was a really awkward thing, because the gate was so small, even low charge transfers would have it expire, so all the connections had to be kept shorted together by a fine wire spring coil thing that gets pulled off after the part is down and soldered.

I guess I didn't have the skills of a line operator, and I needed the "third hand". Definitely this sort of stuff cannot be done if you have just had an expresso cappuchino laté !

 

[graham-xrf]

PocketGeiger Type 5

For just initially poking around among the circuits downloaded from the SparkFun site, clearly the board is over-produced spillover from the little white plastic goodie that plugs into a smartphone. We note the 2012 date. We have a little smile at the Japanese transposition of "R" and "L" that extends from the way they pronounce into the documents spelling. "Open corrector output" is exactly what you think it should be, if ever you encountered "lestaurant".

First the PocketGeiger-5
I note there is the SIG radiation detection pulse, and the NS noise detection pulse.
The explanation is there is output a positive HIGH pulse when "vibration noise" has been detected a NS signal is delivered.
Hmm.. It would seem that in addition to detecting radiation, the chip is somewhat microphonic. I suppose most any crystals are by the piezoelectric effect. That is concerning because of the very high outputs possible, and the fact we are to have some gain on the end of it.

The text at the end gives the clue that the NS pulse is used to trigger a dump to ignore the last 200mS of whatever is going on at the SIG connection. Quoting..
"Pulse width differs depending on vibration pattern and regardless of pull-up resistance R.
Dispose measurement data for last e [sec] {e<200[msec]} when vibration noise has been detected".

The circuit
We were agreed that the circuit is not best for what we want, and I get it that you may be using anything handy for now while checking it out.
The LMC662 is a OK CMOS OP-Amp. The original circuit is asking a lot to have a follower with 66M feedback resistor. The actual feedback current would be tiny. For most practical purposes, the 66M is almost not there! All the bias current the CMOS ever needs for the (+) input comes from the divider chain. I think the 1pF feedback is to keep it stable.

The (-) input works from only those few electrons that can fight their way through 66Mohms, and of course whatever might come out of the diode.
The bias condition of the diode is not obvious, done via 100K to a little switch-mode supply. At this stage, I admit I don't quite understand the thing.

It follows with a gain stage G=100, then offered at standard voltage window comparators in LM393. These are set to flip when the signal goes below 2.68V (assuming the 9V is exact). and when it goes above 3.09V. If it goes low, U3A output is open collector, looking to be pulled up hard by something it is hopefully connected to down the audio jack. current limited by 470 ohms. If it goes high, then U3B yanks the output low. If it is somewhere in between, regardless of diode noise and other information, things are quiet. I note the 4.7nF directly on the output of U3B open collector. There is no excuse for this! Let us bury this circuit now!

But we are not doing that!
We go linear instead, to try and extract whatever was yanking those comparators. I don't yet understand all of your Teensy circuit. How P2 selects the gain, and indeed how the signal makes it to the output. My search for HDR-IDC-2-54-2X3P did not find a datasheet - only 7 results of unrelated gabble.

PIN diode performance
The X100-7 SMD data sheet is encouraging. It is operated reverse biased. Looking at the dark current, I would choose to operate it at very little reverse bias, instead of the tens of volts as in the PocketGeiger. We see the "Absorption Probability" vs "Gamma Energy" graph. The main part of the curve is what I would call X-Rays. The graph is about the probability of absorption, which is not the same thing as "pulse height" nor "duration".

This diode could apparently detect Chromium (5.4KeV), Vanadium (4.95KeV). From the graph, even Calcium and Potassium have more than 80% probability of being absorbed. Even Aluminium has a 30% probability. These probabilities are just that, and would not be directly related to the pulse size, only to how often they occur - whenever they do. We don't really know how much a given energy from incoming photons, once absorbed, will contribute to diode current above the dark current (which will be noisy racket)! It may not be much. I suppose we could estimate it, but I would just go for showing it some metal.

First thoughts about the detection
The thing is, the energy from the photons are not showing their presence with the help of huge gain, such as we would have from a PMT, or a avalanche diode. Therefore, I think, one needs to operate with as little bias as possible, limited by the PIN capacitance getting large enough to soak up the signal. Perhaps about 6V, where the dark current is is only 2nA, and capacitance (ugh) as much as 100pF. If the detection circuit can
still get pushed around enough in the face of 150pF, then 4V bias gets you 1.5nA dark current.

From then on, it is about capturing that (current) pulse and applying the sort of gain one would get from a PMT or avalanche diode. The amplifier has to have a bonkers low noise figure, and a gain that I am thinking should be more than 100. Perhaps 100,000? I cannot know without knowing what actually happens with a MX-100. Avalanche diode gains are around 10E5, and PMT gains are like 1.7E6.

A crazy thought.
The capacitance curve is so steep, it makes me think that if one used the PIN diode as part of a RF oscillator, the frequency shift when a pulse arrives just might be useful. We get here to the parametric amplifier techniques that used varactor diodes, very popular for avoiding noise, and widely used for radio telescopes and satellite receivers. Now, with availability of pHEMT FETs , one does not go to the expense of them, but for extreme low noise, they are still ultimate!
Enough with the crazy thought!

Why a buffer?
Of course - to stiffen up a high impedance signal. Is it that? It's from a diode, with a whole 3nA of dark current. A stage with a gain of 1 adds the noise of it's first semiconductor molecules bashing around, and then it gets increased by as much again when it reaches the gain stage, so worsening the S/N ratio without yet any gain. You get an admittedly lower impedance signal, delivering a stiff, low impedance pile of extra noise!

I am thinking that a very low noise high gain stage is needed at the beginning. Bias, offset, signal clamping, etc, can all be dealt with, but the basic grab at the little current in excess of the dark current needs to be done right at the beginning, and brought to safety.

So, if I am wrong about all this, you can bet that I will just build yours!

 

[homebrewed]

Hi Graham,

A lot to unpack in your comments, so I apologize in advance for (most likely) missing something. Just to preface everything, the circuit I've designed (and will build up soon) really is just a shot across the bow. I have built a myriad of circuits for all kinds of things, and quite a few didn't pan out at first, often because I didn't understand all the factors involved. I have hopes that some signal processing can address some of the problems encountered along the way. Also keep in mind that the whole idea behind this scheme is "cheap as possible to get the job done".

First, regarding the noise detection function, I know from experience that charge amp and electrometer setups and the like are extremely sensitive to vibration, mostly due to triboelectric effects. Attempting to use your average coaxial cable in a high-gain circuit is a sure way to get frustrated due to that problem. It's possible that your average FR4 PCB is pretty bad w/regard to stress-related issues as well. I'd just assume that is the case, given the fact that just about EVERYTHING seems to play a part when it comes to introducing noise into high-Z circuits. I haven't seen noise issues in silicon devices due to vibration, but I can say for a fact that stress in the active devices in an integrated circuit can dramatically change their operational characteristics. It blew one high-gain RSSI design right out of the water because that wasn't properly taken into account (I could swing the output by volts by simply pressing a microprobe down on top of one of the input transistors). I didn't design it, I just got to figure out why it had serious problems. However, in the situation of this particular sensor residing inside a lead-lined box, vibration shouldn't be a big issue -- I hope....

Regarding the input circuit, I've been interpreting it as a charge amplifier. If so, the capacitor actually determines the gain: if Q = CV, then V = Q/C. A small C = large gain. The feedback resistance is there to provide the bias/offset current, but does of course introduce a time constant as well.

I agree with burying part of that circuit now . Certainly not useful for us, anyway!

About the reservations about the PIN diode's high capacitance. Since the charge amp input is a virtual ground, all the voltage variation across the diode is internal to it, so, ideally, there shouldn't be any delta-V across the diode -- so it shouldn't play a part in the system's bandwidth. That assumes an amplifier with infinite bandwidth so not precisely true. But (hopefully) good enough. Yet another open question. I do like the approach of using a high bias voltage, mostly to reduce the diode's internal capacitance as much as possible. Below breakdown, the leakage current is mostly due to thermal generation so it's not highly dependent on the bias voltage. I agree that the PIN diode "gain" is far below what can be had with an APD or PMT. THAT is one of the major unknowns which may skuttle this approach. I'm hoping that a least-squares curve fitting approach will address some of the noise issues you are concerned about; but, if not, a step up in the amplifier technology will be needed.

Regarding the diode's sensitivity vs X-ray photon detection probability. Not knowing what criteria the vendor is using to define "probability", it's hard to tell what it means in terms of energy resolution. Yet again, another unknown. I THINK it just indicates what the detector output is for a given X-ray photon. So a slope is good. Flatness = no energy resolution. Maybe.

Your thoughts about using the detector's capacitance variation are very interesting. A pure capacitance is noiseless so that's pretty attractive. One potential issue I can see is that the voltage change necessary to swing that capacitance suggests some sort of resistive load -- and its thermal noise voltage will, in effect, modulate the voltage across the PIN diode. Gotta think about this one some more. I just thought about a capacitive load with a high-value resistor to sink the DC current, but will need to run some numbers to see if that's reasonable or not. The capacitive load + drain resistor's time constant would have to be long enough to attenuate noise in the frequency range of interest.

Finally, about the gain-of-one buffer. Yep, not so good w/regard to minimizing noise. But I'm looking at the simulation results I got and thinking that ***maybe*** it's not a killer. Maybe...

Oh, one more thing. You were wondering about the 2x3 header. It's just an on-board jumper setup to select the 1X, 10X or 100X signals.

If you want one of my boards, just let me know. 5 boards was the minimum quantity so I've got some extras. I could probably mail one or two for just a few dollars. That probably depends a lot on what kind of downstream data acquisition scheme you've got, but PM me if you're interested.

 

[graham-xrf]

I'm learning PIN
The photocurrent from a ~5KeV or ~10KeV contribution might not shift the PIN bias voltage enough to be useful to a oscillator shift. PIN diodes do not work like that. Below a certain frequency, they work like normal diodes. Above the frequency that the carriers in the depetion region have time to re-form, it acts like a variable RF resistor. That said, there are such things as PIN diode RF modulators and phase shifters. I think a PIN diode, while it does change capacitance with bias, is not like a varactor diode in an RF oscillator.

The pre-amp is trans-impedance!
When we look harder at the "no good" op-amp follower we have been so critical of, it might not be altogether a follower. Instead, the reverse biased diode may be leaking it's current into a 66M "perfect" resistor. The amplifier presents a low impedance load to the photodiode, forcing that what works it is the current. In effect, a transimpedance -6.6E+7 gain amplifier, driven by direct injection of XRF KeV inspired electrons + dark current + thermal noise into the inverting input.

The reverse bias signals that a PIN diode photodetector is operating in photoconductive mode, as opposed to photovoltaic.

With such a large gain, any OP-Amp input offset voltage will result in output offset. There are ways to deal with this. A few pF across the feedback resistor collapses the gain at frequencies which would have it become a powerful oscillator all by itself. That capacitor, along with the input capacitance provided across the diode (50pF - 150 pF) define the roll-off. In this case, about 16Hz to 48Hz. Given that the recombination time of carriers in the PIN layer is about 5uS to 10uS, it is still the wrong design. A lower gain first stage, to get the roll-off around 100KHz to 300KHz or even 500kHz would be better. The gain, if need be, can be made up in the second stage.

About microphonics
The way to get past cable flex, substrate flex, and all other piezo voltages that can only live in a high impedance world is to not offer them a high impedance to live in - and chose the material. e.g. that thin PTFE coax is so active, it is used threaded through fences as intrusion detection microphones.

I don't think the stuff we are dealing with is necessarily intrinsically extreme impedance, like say, an electret microphone, but whatever it is, so long as it finds it's amplifier within as few mm as possible, gets some low noise transimpedance gain change , then microphonic effects won't have much chance. Better still, if it finds its way into becoming numbers within the next few mm, low frequency pulse noise of that kind has even less chance.

Camera X-Ray sensors?
Another passing thought is - many image sensors use pinned photodiodes (PPD) CMOS sensors that matched and exceeded CCD sensors.
So - is it possible a cheap camera sensor can see X-Rays? After all, X-Rays would go straight through the color filters!
[OK - I just ask out of ignorance]! What happens if one waves a smoke-detector source, or a gas mantle right up to a web cam sensor?
I don't know if this can ever be true. I think X-Ray sensors need to have a certain thickness. We only got GaAs sensors when it became practicable to make the stuff pure, and thick enough without laying it on by diffusion epitaxy.

SPICE?
Meantime, it should be possible to find an equivalent circuit model for a PIN diode with a blob in it representing electrons contributing to the current from external KeV inputs, just like you can have for other injected carriers, or thermal noise sources. Note that in some circuits, we see a 1 GigaOhm across it, purely to avoid a SPICE circuit dividing by zero. This is much the same as orphaned capacitors in filter circuits with no DC path to GND. I can't find application notes on MX-100. Maybe we can find a similar thing among it's competition?

One wonders what is seen if an ordinary oscilloscope is connected to a reverse biased PIN diode (say MX-100)?

 

[homebrewed]

The cutoff frequency of the pocket geiger's input amp circuit is about 2.4KHz so the 1pf capacitor has a substantial impact on the pulse response. Given the fast current pulse coming out of the detector, I think the circuit more resembles a charge amplifier. The 66M resistor is there mostly to take care of the amplifier's bias current. It is a which-is-it puzzle because TIAs also need a capacitor in the feedback loop so they don't oscillate. The capacitance of the photodiode introduces a pole that (in the absence of a feedback capacitor) destabilizes the circuit. So it depends on the bandwidth of the amplifier relative to the input signal BW if you call it a TIA or charge amplifier.

Since a charge amp acts like an integrator, the output pulse's peak amplitude is proportional to the total charge delivered to its input. No need to numerically integrate the waveform coming out of one of these, the integration has already been done for you.

Here's an explanation of charge amplifiers: charge amps.

I hadn't known about using PTFE coax as a kind of intrusion detector. That's a pretty cool use for an effect most others want to get rid of, like taking a lemon and making lemonade.

 

[graham-xrf]

Hi Mark.
Thanks for the Hamamatsu article. I think it is one I already have. I should have looked first.

The Hamamatsu circuit, which is essentially the identical same circuit we are talking about, has a very complete analysis of the simple version circuit, and it is in all respects, the transimpedance amplifier, which (when not mistaken for a follower by me), is extremely well known and developed. Nearly all the Hamamatsu analysis can carry across to the more elaborate circuits.

The 66M resistor is not to take care of the amplifier's bias current. It is fundamental to the device gain, and is not usually 66M. The best JFET low noise devices have input resistances of 10^14 ohms. What happens with a charge amplifier is the input current onto the (very small) gate will have the voltage between the OP-Amp inputs attempt to change, resisted mightily by the enormous gain to make the output change enough to correct the situation via the feedback resistor (Rf) , and it has to move volts to do it. The gain is essentially the value of the resistor. The capacitor, values between 0.1pF and about 10pF, provides stability for the gain stage, with local roll-off decided by the capacitor and the Rf.

Then, when the photodiode is added, it brings about a new roll-off, decided by the capacitor across Rf and the capacitance of the PIN diode.
The more usual value for the transimpedance stage gain is nearer 12M or 20M.

First stage goodness
It's two decades old technology, and costs a whole 5 bucks for one, but I like it, even with the added expense of a another Op-Amp to servo-stabilize the offset. LT1793 input bias is 3pA, which is significant given it has a voltage gain more than 4 million. Noise is 0.8fA/√Hz, and low voltage noise is 6nV/√Hz.

Pushing the limits!
It is possible, at relatively low cost, to use a single low-noise JFET, put inside an OP-Amp feedback loop, and have a charge amp that will uncompromisingly force something that approaches the limit of theoretical noise performance. In the circuit attached, it uses a IFN147 JFET, which is extreme, but there are many other low cost JFETs that would do instead. That circuit is from the Analog Devices (Linear Technology) collection. The gain-bandwidth product is high enough (2.4GHz) to deliver near 4.8MHz bandwidth at gain 500K. Having this available allows all sorts of roll-offs, compensations, pulse damping, and refinements. One could, of course, just use a 200Hz roll-off, and go for 12M gain in one hit, but I would not be inclined to. To get enough gain in one device, and a bandwidth more than about 200Hz, this is what you need. Having a Gain-Bandwidth product of (say) 500MHz, would likely require another stage.

Now that I have searched, I am astonished at the sheer number of published variants of this circuit! Many have servo-controlled anti-drift arrangements. LT1793 (without extra JFET) can be had on standard 2 x single Op-Amp evaluation kits (we don't need them). On some of the circuits, one has to add bias, since it is not a photovoltaic type.
Easy to be overwhelmed by simply entering "Photodiode Amplifier" into Google, and look for Images.

In the near ultimate JFET input circuit, there is no real need to expend on two kinds of op-amp. Finding a dual JFET OP-Amp will do, and you can take your pick over gain-bandwidth product. I may opt for something beginning with "LT" simply because we are likely to find the model in the LTSpiceVII simulation store.

One can also get fully compensated photodiode charge amps with additional stage gains, compensations, etc, in the same package as A/D converter(s). That's found in Analog.com. I have not yet checked if the prices would make eyes water, but it might be a one-device-does-all thing that only need digital numbers to be sampled by SDI.

Enjoy the --> Photodiode Circuit Design Wizard

There is a --> How-To article from Digi-Key

There are, of course, lots more articles, and YouTube videos, but I think that if we can hang together a nice diode model, we have all we need to completely handle this thing. We are pretty much overwhelmed by choice. There are some pitfalls - like having to be aware that ESD protected products can have the input bias current run away with temperature sensitivity, which is something the LT products have addressed. We can still stumble, but I feel a whole lot better about this part of the design, and I am perusing the possibilities of amp + A/D conversion one-chip stuff.

 

[graham-xrf]

This YT video is very clear, and very good.
--> Photodiode/Transimpedance Amplifier Design

It makes me think we need to cover the diode with kitchen foil.

 

[homebrewed]

The pocketgeiger simulation I have put together suggests that the first amplifier is basically acting like an integrator -- i.e., a charge amp. Here's how I came to that conclusion:

The input current pulse is a triangle, with a peak current of 100nA, lasting 100ns. This is a total charge of 5E-15 coulomb. The output of a perfect integrator obeys the relationship Q = CV, or V = Q/C. When C = 1e-12, we find that V = 5mV. The simulation predicts a delta-v of 4.17mV on the output of the LMC662. On the other hand, 66E6*100e-9 = 6.6 volts. Either the opamp's nonideal bandwidth is turning the circuit into a fairly decent resemblance to a charge amp, or the feedback capacitor is. Probably some of both.

You've found some good reference material -- thanks for the info! I haven't looked at the YT video you found, but, FYI, there is a copper foil shield around the detector. It may be necessary to remove it so it doesn't filter out the XRF-generated photons, but the whole board will be inside a lead shield anyway.

The schematic of the IFN147 application shows a cute way to null the DC component in the input. It took me a moment or two to figure out that the bottom amplifier is a non-inverting integrator, after that I knew what it was doing.

Hands-on work on this thing is going slowly right now because I'm working on a barn rehab project and trying to get it into reasonable shape before the winter rains come. I'll probably pour at least one concrete footing today. Today's high will be around 70F (21C), perfect for pouring concrete.