Needing more than a spark test?

[homebrewed]

I had an opportunity to fire up my pocketgeiger board and did some tests to see how vibration sensitive it is. Yep, it's quite sensitive. I used the eraser end of a wood pencil and tapped the detector and also tapped the board near the signal/power jack. The board was gripped in my Pana vise on the detector end so the tapped end of the board was free to flex.

It appears that board flex generates a much larger signal than tapping the detector itself. See the photos below. Note the vertical scale. Also note that the _direction_ of the leading edge is different. The most vibration-sensitive path generates a positive-going signal; but it also generates a large negative-going excursion after that. I didn't flip the board over and tap it to see if the positive-going leading edge would become a negative-going pulse. BOTH signals show a damped waveform that has both a positive and negative component, as opposed to an x-ray generated pulse.

The time frame for vibration-generated signals also is about 5X longer than x-ray generated pulses. So the damped waveform or longer timescale could be used by a very simple S/W based routine to reject these things. It doesn't make any sense to use a curve-fitting approach for this situation.

 

[graham-xrf]

Excellent work Mark!
The signal is AC coupled after the first amplifier with a whole 0.1uF, so it will work even to quite low frequencies.
Some of that is board vibration, some, I think, is circuit ringing overshoot,
I hope shortly to have opportunity to get at mine.
I would be trying first disconnecting the diode, and go again.
I would slowly bend on the board this way and that to feel this out.

What I think is happening is a capacitor soldered across two pads is being "stretched" or jiggled by board stress or vibrations.
The ceramic C4 1uF on the diode bias point is the favourite suspect right now.


(Top Half Excerpt of Geiger Type 5 Circuit)

We have been through this circuit before, but since it is all we have to play with for the present - consider..
The potential divider to 9V, with values that set the voltage at pin3 of LM662 to 2.88V
I nearly did not spot that connection, because in my world, for my whole life involved with circuits, it is verboten to place a spot on a wire crossover, to make it a "connection". They should always be staggered, to make it unambiguous!

Going after this noise should be just a matter of finding the offender. I don't think it's FR4, but I do think it is a ceramic being wiggled by FR4.
The obvious stuff I would do. Take the diode out, or cut the track to pin 2.
If it still "makes a noise" when bent or tapped, then C6 is next!

The response of this circuit is limited by the 1.1v/uS slew rate, related to the 1.4MHz gain-bandwidth product.
To get a signal even as slow as 1kHz through it unmolested leaves only 1400Hz before the gain is down to 1.0
The gain of this circuit by component values is 66 Million x 100. I do not understand it. It has a credibility problem!
I think the delivered gain of this circuit is much lower than the component values would imply, and is not decided by 66Meg Rf.

C16 is unforgivable. That is something that should never be done to an op-amp. Maybe it is a circuit drawing mistake.
Perhaps it was supposed to go after R18, to give a CR roll-off at 452.693kHz

There is not a whole lot there that can make microphonics. C4 is there to filter the noise from a switching bias supply (another bad idea).
Removing the diode is the fastest way to cut off that route - just to eliminate C4. If the microphonics persist, then the culprit is downstream.
C6 is the next suspect. Only a gain of 100, but 100nF ceramic could get loud!

C3? Another "smoothing capacitor" on the 2.88V. It is connected directly onto non-inverting inputs. No series resistor. OK, I get it the designer did not want to ground them. The circuit is operating with GND on the pin 4 V-. It was to lift everything up by 2.88V.

I think we can find the cause of this "sensitivity" and lose it.

 

[homebrewed]

The circuit has some definite hits against it. The gain relative to the short current pulses generated by the x-ray photons is mostly set by the 1pF capacitor -- V = Q/C. But the low frequency gain is very high so disturbances due to low-frequency vibration get the hell amplified out of them. I'm not sure there is an easy around this if a charge amp topology is used. Maybe a bandpass filter, but we're already throwing away an awful lot of signal away now as it is.

The second is, indeed, C6 -- what a great way to destabilize an op amp. Seeing as how I have found at least one error on the schematic already, it is possible (I hope true) that C6 really is on the other side of R18.

 

[graham-xrf]

The charge amp circuit can do this. Admittedly with much smaller PIN diodes, there are communications circuits that go at fibre-optic speeds.
Just using fewer Megs in the feedback, and more stages of reasonable gain, and op-amps with GBW 300MHz and above, gets you there.
I am now less worried about microphonics.

The choices for A/D converter are a bit limited. Those with differential input need a dual op-amp extra chip to drive them. I am trying for one with internal reference, because that solves a noise problem. I am trying for one that does not cost £39. There is another at £27. There is even a candidate at £11, but it needs the differential driver, and external reference. I will get there. I may even go for it without doing anything with the SparkFun kit other than liberate the diode.

 

[graham-xrf]

TIA Gain Bandwidth Analysis
Other casual HM readers will get well turned off by this, but I wanted to do a walk-through. It delves the electronic technical using currents of less than 100,000 electrons in about 10uS, and is way far away from making hot metal lathe chips and surface metrology!

A little passing check on the issues of what it takes to extract the signal in a credible way.. motivated by the Pocket Geiger 5.
The circuit looks so simple, but I can tell you almost none will "just work" easy.

I am now rather further along into A/D conversion selection, costs, and weighing up things like multiple supply voltages, including those that would trash the amplifier front end op-amp. Some good affordable ADCs need external references and reference buffers vs ADCs with the reference built in. I try for using the full Vref range, even with differential inputs, and very low noise high bandwidth negative voltage generators. The circuit fragments shown here are not the final, just test circuits to investigate what it takes to believe that the pulse was preserved and measurable, instead of some smudged artifact.

The LTSpice test example shown compares a lower end photonic signal of 500pA (which may be unsuitable as too big, but it compares to Id=2nA dark current), with a much bigger 80nA signal (which may be unsuitable for being too small, but we already hit the rail)! That's OK. One could make the minimum signal smaller, and increase the gain, to no point, but stay with these for now.



It took some work getting the LTSpice app even to make these. To get the tool to work, I set the Absolute Current Tolerance to 5E-13A, and the Absolute Voltage Tolerance to 1e-7V. It may need more tweaking. I set Gmin to 1E-14, and that may yet not be small enough to compete with FET input currents in femtoamps.

This time, I use an op-amp circuit where the first stage "gain" is not defined by a silly feedback capacitance in integrator mode, in parallel with 66Meg resistor that may as well not be there! The gain is now defined by the feedback resistor. To be able to sustain a first stage gain as high as 510k, and still hope to follow to follow a pulse that will be over in 10uS, you need an amplifier having GBW of 300MHz and more. The LTC6269 is a dual op-amp, with GBW of 350MHz in S8 size package, or 500MHz in MSOP. I don't know how the LTSpice model can know how to choose.

I fully explored the value of C2, across the feedback resistor. At 0.2pF, and 2MEG , it oscillates madly. 0.5pF stops the madness, but still overshoots and distorts. 1pF is safe, and is already introducing bigger phase delays. The delay seen is near 1uS.

I fully explored the value of C3, which I use to take out the offset of the dark current. from 1pF to 1000pF. The signal captured gets bigger, and is running out of advantage at 600pF. I put a 1nF value there, and I would consider making it smaller.

The higher range for A/D converters chooses between 3.3V supplies, and 5V supplies. Choosing the larger, the Vref will be 4.096V from a low noise, buffered, bandgap reference.

In the circuits shown, the 500pA current delivers 26mV. Without the full noise analysis, I don't know yet if that is 26mV competing with highly amplified noise, but I think it, as design example, is OK, it being already -101.2dB down for a devices which would only manage no more than 96dB dynamic range anyway, and more likely to be nearer 90dB. This also lets us know that using a 2.048V reference only halves the small signal, and we are already 10dB below. THere would be no advantage.
[ These dBs are 20*Log10(voltage ratio) ]

We could make the first stage gain a bit less, and the second stage gain a bit more, but to no point. We should make the gain suit the minimum signal we can have tickle the LSB of the A/D converter, starting with 4.096V. We will have a maxed out gain distribution that can see to the noise floor.

Driving ADCs
One needs a low impedance forcing signal to settle the speed of the pulse at the ADC input, which itself is going to take current suck right then. A differential driver, if it has gain, would allow we operate the TIA at lower gains, potentially allowing cheaper, lower GBW op-amps.

I get it that this stuff will look like gobbledegook to some, and is, to be truthful, relatively amateur level electronics design, but I wanted to take the mystery out of just about every XRF-type Geiger counting circuit I have seen so far, starting with Theremino.

 

[homebrewed]

The LMC662 spice model I downloaded from TI had a commented line indicating that GMIN=1e-16 should be used to properly model its low input current, so that's what I used for simulating the pocketgeiger circuit. Right now my simulations aren't converging all that quickly, but that just means a few more seconds to wait for the results. I probably have the minimum time step set too small.

In the case where the input amplifier is acting more like a TIA, the feedback capacitor is needed to ensure circuit stability -- it basically compensates for the parasitic capacitance(s) hanging on the inverting input node of the amplifier. I'm sure you know this, just clarifying to the "casual reader" (if there are any at this point) on what's going on with the oscillation problem you described.

I note that your latest simulations could apply to a modified version of the pocketgeiger -- just different R's and amplifiers. So am I correct in guessing that your development path involves swapping out some parts on the preexisting board?

 

[graham-xrf]

Even if the pin-outs were the same in places, I had not considered the preexisting board in any way.
I might not even get around to powering it up. The only thing on it we want is the photodiode.

My circuit goes right on into low cost differential drivers, and a low noise external reference, to allow affordable a high performance A/D convertor. Pseudo-differential, buffered ADCs with onboard reference cost £36 and up. A reference is about £5. AD4000 is about £11. A major thing in the metal debate is whether or not to mount the diode flat on the PCB at all. Depending on the lead stuff around it, I considered an end-on mounting, with low capacitance fine wires, and a lifted pin.

There are only six parts in the Analog Devices stable that fit the requirement 16 bits, and 2Msps. Three of them are Linear technology.
Maxim Integrated has only one product at 1.6Msps, and needs a reference.

AD4000 (£11.09) Mouser site is down)
AD4001
LTC2310-16
LTC2370-16 (Pseudo-Differential, Single-Ended £36)
LTC2380-16
AD7622

It's OK - we will get there in a sweet way.
- - - - - - - - - - -
It does lead to some questions about the specks of Am241, but I have to go now. I will return to it later.
- - - - - - - - - - -
Later is here now!
Re: Am241. The radioactive half a microgram, whatever, is in there somewhere

So after taking it apart, we get at the little button cup with the AM241 in it.
My question is, are the gamma rays squirting out of it in all directions, regardless that only one end of the button cup is open?
Going deeper. the little thing in the middle does not look like any oxide. the surface in there is yet another tiny disc of metal - peened down.

Our photons cannot be reflected, aimed, nor collimated, except by putting into a little lead tube with a thick base, and what happens to come out the open end are those that are not wasted by expending themselves warming up lead. Not energetic enough to provoke any X-ray, they just knock electrons off the outer region of probable existence, and leave them to ping around in the lead with enhanced KE until they become exhausted.

It's nice to know that the "gamma exposure constant" is 1.3R-cm^2/mCi-hr ????
It's much better to know that the amount of lead necessary to reduce the exposure by a factor of ten is 0.03mm.
That means a 0.15mm thick sheet reduces the radiation by a factor 100,000.

Do we have to have copper over the sensor? Can we use aluminium?
The thing is, I know copper to be used as a filter to cut down the strength of X-Rays. I did that myself once using 0.5mm and 1mm square plates with a hospital X-Ray on minimum, to an image intensifier. Maybe the adhesive-backed copper tape over the X100-7 is too thin to matter. It seems to be 0.05mm thick.

Since the assembly would shield the copper (or aluminium) from seeing any gamma, we would be OK. For total light shield, + electric field screening, I guess it has to be metal. That would stop any radiated fields in both electric and magnetic components. The problem comes from static and very low frequency magnetic fields. We don't want to be making the best 60Hz house stray field pickup on the planet!

I am going to have to check - I never tried to turn lead stuff in a lathe before!

 

[graham-xrf]

About the Pocket Geiger Teardown
Checking out the possibilities for the new mounting for X100-7, we start on the Pocket Geiger circuit.
The copper tape has got very tenacious adhesive.Under is a insulation layer of white plastic - also sticky with blue stuff.

_ _

It came with two brass plates 25mm x 35mm x 0.5mm. Maybe they go in the plastic box that comes with it.

_ _

The green wire under is soldered to the copper tape only, but it burns a niche for itself.


When we get to it, it is clear the only way it comes off is by heating it up for solder reflow.
The other possibility is to leave it on the PCB, but cut it away between the chip and the sensor through C3.

The sensor window is potted - filled with black epoxy. Light cannot get in.

The only pins used for operation are the two centre pins on each side. All the others serve to hold it down, but are not electrically connected.

 

[homebrewed]

Gamma rays are emitted from the americium in all directions, but their intensity depends on what they go through. So the metal back of the capsule will attenuate them some.

Using the NIST mass attenuation coefficient table for copper, at 6Kev (close to iron's K-alpha line) we see u/p = 115.6. They are using a density of 8.96g/cm^3. If we plug those numbers into the attenuation equation, we get an attenuation factor of 177 for .05mm thick copper, not a good situation for our low-energy XRF setup. For 60Kev photons copper reduces the intensity by only .93. So my simple setup of placing an americium capsule directly over the copper foil shield should be capable of exciting detectable XRF copper photons @ about 8Kev, although only ones near the exit side of the foil would make it out. I believe it also would work for a thin sheet of iron (as in, shim stock). This scheme will only work for very thin samples, not bulk, so an arrangement like the Theremino would be needed for analyzing chunky stuff.

If we replace the copper foil with standard-gauge kitchen aluminum foil, Wikipedia indicates it is .016mm thick. At 6Kev it reduces the 6Kev intensity by a factor of .6 instead of 177. Since the detector is covered by molding compound we don't need a perfectly pinhole-free piece of aluminum foil (to keep light out) -- it's just for electrostatic shielding. The super-thin fake silver leaf might work OK for this.

 

[homebrewed]

You posted your teardown info while I was writing a post on this subject. I think the brass sheets probably are there to filter out beta particles but I could be mistaken about that.

It would be interesting to see how bad the AC noise is with the shield removed.