Needing more than a spark test?

With just a little work it should be possible to get a handle on the dark current. Solder a wire to the TIA's output terminal and bring it out. We don't exactly know what the derived ground reference is since it's produced by 3 series-connected resistors connected between Vcc and Gnd but we can get close enough for some ballpark figuring.
 
The dark current is a DC current, not shot current. Its major component has to be thermally-generated carriers, the Is (I-sub-s) component of ye olde diode equation: Id = Is *exp(Vd/(kT)). When reverse-biased Vd is negative so the exponent rapidly declines to Is. You can see this is mostly true because the reverse current vs Vd curve is approximately linear on the lin-log plot shown on the data sheet.

If taken close to its breakdown voltage you will get avalanche multiplication and the curve will start to bend up. For a PIN diode this probably is not a good thing -- so-called "hot carriers" can generate traps which enhance the recombination of photogenerated carriers. If taken too far, the detector's efficiency could be severely degraded. Similar to what happens to a bipolar transistor's gain if its emitter-base junction is taken into reverse breakdown.

I-sub-s is proportional to the diode area so it's no surprise that this gigantic diode has a fairly high dark current. It also sez that we can't do much about it, unless we cool it with a Peltier. Don't laugh, I'd already considered that possibility.....and I happen to have a couple on hand....

On a slightly different subject, I thought the baseline noise had a suspicious look to it -- it's quite fixed in its max/min range, not what I expect from noise. So I just shorted the 'scope probe to ground, and guess what -- the baseline noise level stayed about the same. This could be an artifact of the display's limited vertical resolution of 480 pixels, rather than digitizer noise. I can download a csv of the captured waveform and examine it to see if that's the case.

It also would be a good idea to see if the waveform on either side of the peak really is accurately modeled by a least-squares second-order polynomial fit, since that's the crux of my MCA. Being able to average over multiple samples of the waveform should significantly improve the effective SNR, which, in turn, should improve the energy resolution of the XRF system.

I'm still learning things about my 'scope. The input configuration for each channel allows you to specify if its input comes from a 1X, 10X or 100X probe. The scope isn't smart enough to figure that out by itself, so I was misteakenly (sic) using my 1X probe in the 10X setting. So the vertical scale shown above is off by a factor of 10 -- the pulse height really is 90 millivolts, not 900. My signal conditioning board may come in handy anyway!
 
Re: Scope probe measuring. You would be looking at the pulse at the output of the transimpedance amplifier. It should be OK to use a scope probe at the X1 setting. The output impedance of the amplifier should be low enough to drive the 1Meg Ohm of the scope probe, and it's 20pF (or so) capacitance. No need to use X10, with it's signal attenuation.

The DC component of the diode current is removed OK by the series AC input capacitor to the current junction. The photon pulse will be able to wiggle the current input, along with any other racket able to go through the capacitor. Any other amplifier input DC offset can be dialed out with bias on the positive input. I would have thought your signal conditioning board is pretty much essential.

My experience with Peltier coolers was all about their low efficiency. I was trying to cool CCD imagers to lose the dark noise colour patterning when they got hot, mostly because of the incandescent lighting on the same assembly. The Peltier metals are highly conductive shunts. To get one end cool, you need an elaborate construction on the other side to suck that heat away into a heatsink, and by far the most heat having to be removed is the self-heating done to run the Peltier current in the first place, then plus a bit more to make the cooling happen. Generally they were awkward, complicated, inefficient things. I was running substantial powers, like 10W and 20W, to cool a CCD.

While we certainly can get the dark current shot noise, and other wideband noise lower by cooling, it seems possible to get the design to work without needing cryogenics. If the smallest energy pulse we want to see is sufficiently above the noise to see it, that is good enough.
 
I have attached a spreadsheet with the raw csv data and a chart showing the pulse, which occurs near the middle of the data. Some of the "fuzz" I'm seeing may well be digitizer noise so there is a possibility of some improvement with a better ADC.

I'm concerned about the count rate I'm seeing. It's low, far lower than I would expect with that many Am241 sources (1-2 counts per second). I may need to cobble together a simple electrometer to verify the current generated by the alpha particles, if any. Another approach would be to get my hands on some pitchblende and see what the pocketgeiger thinks of that.
 

Attachments

The noise wobbles are, I think, constant and random, showing up better on the slower, more horizontal parts of the pulse. Definitely the lower order bits are being jangled. The causes could be many. Disconnecting the diode, or just shorting it out, so long as what you do does not offer a way for noise to couple in, is a reasonable test.

Another is to disconnect all from the ADC, except a low resistance across the input. The "zero" point at the ADC common is the usual very awkward place where the sampling gets bumped. If an ADC is to measure a voltage, then using differential inputs with equal resistances is best, and the power to the digital parts of the ADC must not share the return common with the analog input return. In effect, it has a digital supply, and a separate analog one. The bottom connects at one point only at the ADC chip.

In practical builds, I have found it near impossible to make the least significant bits of an ADC go completely silent. On 16-bit ADCs, when trying hard, the usual zero has still counts of 40 to 120 remaining. Supposedly, with the ADCs I have, the spec says more than 90dB measure range. I would have to check out a Analog Devices reference circuit to find out how they do that. I think doing stuff like making sure even the energy transfer in reading the digital outputs might be limited by resistors.

Basically, doing various gross things to try and make the base line go silent is the first step.
 
Last edited:
I did an experiment today, varying the detector bias voltage between 9V (the minimum-possible before some internal diodes in the PocketGeiger take over and clamp the voltage) and 40V, the maximum my bench supply can output. The detector's specified minimum breakdown voltage is 50V so I felt safe doing this. At the low end, I started to see some distorted pulses , and perhaps a slight reduction in the already-low count rate. Above 15 volts or so, the pulse shape and count rate stayed pretty constant.

As a part of troubleshooting the low count rate I've decided that I need a way to get a handle on how active my Am241 sources really are. After all, they're from China. I found an open-source ionization type radiation detector that uses a pair of Darlington-connected transistors here. The final design, which can be found on github, uses an obsolete IC to drive a thermometer type display to give a rough indication of the input radiation flux -- so I have laid out my own version that will just output an analog voltage I can measure by various means: a DVM, or some processor board that has an ADC.

The Darlington transistors are still available on Digikey, although it appears to me that they are nothing special. At the low collector current this design uses, the current gain still is around 1E4 (I found some that look to have hfe's in the 50K region at the same Ic). The collector load resistors are 10M so the overall gain is pretty high, approximately 10^11. I like the design because it uses a second Darlington to provide a certain amount of temperature compensation.

One additional "nice" aspect of the design is that I found a decent instrumentation amplifier that doesn't cost an arm and a leg :)
 
rwm said:
Even Chinese radiation is of poor quality? LOL.
R
Click to expand...
:) I got that too!

For this stuff, we are not measuring a flux intensity, we are measuring the amplitude of a detected photon.
Chinese poor quality radiation might be from a Am241 chunk so tiny, the counts don't happen so often, but even a Chinese can't change the half-life, nor the energy height of the X-Ray pulse.

That said, I kinda knew what Mark meant :)
 
I've been doing more analysis regarding the low count rate on my XRF prototype. I removed the lead shield/aperture I made that goes between the Am241 sources and detector and confirmed that the count rate jumped up by a substantial amount. So the sources definitely are not bogus (that was the concern I indicated, indirectly, in a previous post).

Since the sources and detector are on opposite sides of my aluminum aperture plate, I know that they are generating gamma rays -- plugging the aperture hole made no difference in the count rate, so the counts are not due to alpha particles being reflected back through the aperture hole (if such a thing is even possible). The counts have to be from gamma rays. Besides, alphas wouldn't make it through the epoxy on top of the detector.

Another (remote) possibility was that the sources contain a different isotope that's emitting gammas that are significantly lower in energy than the 59Kev ones from Am241. To get a rough idea of the gamma energy, I placed a .25mm thick piece of copper over the detector. My calculations indicate that about 30% of 59Kev gamma rays would be absorbed by that thickness. I don't have a great pulse counting setup, basically slowing the 'scope's timebase down and then counting the pulses I see in a particular sweep; but it appears I'm seeing a count-rate reduction that is in the right ballpark.

So. I've shown there's no functional problem with the PocketGeiger and signal conditioning boards, and my smoke detector sources probably ARE active and contain Am241. You might think that there's a problem with the geometry of my setup but I've checked the alignment of the aperture hole (and hole in my lead plate) and they look OK. It's hard to believe that 8 sources can't produce more than .1 to .5 counts per second, but I guess I need to do some calculations to determine what percentage of gammas will strike my sample, and from there how many of the fluorescence x-rays will make it to the detector. It will be a low percentage, but my results are far from what the Theremino folks have reported, with count rates in the 100's of counts per second. Time to revisit their physical setup, I suppose.

Another theory is that my detector's sensitivity in the ~6Kev range where vanadium, chromium, manganese, iron, cobalt and nickel emit XRF photons, is much lower than advertised. The data sheet indicates that the detector's sensitivity at 60Kev is about 3%, compared to about 100% @10Kev, so it's suspicious that I'm seeing a pretty high count rate coming directly from my Am241 sources.

Looking over my periodic table, I see that tin's k-alpha line is at 25Kev, about 3X the energy of the iron-sequence of elements. 60-40 solder might be an interesting alloy to try. Another is Tungsten, whose k-alpha line is 58Kev. That might be too close to 59Kev to get many fluorescence photons. I also have a mineral sample of barite (barium sulfate). Barium's k-alpha line is 32Kev. I'll try the barite and a roll of 60/40 solder to see if I get an increased pulse count. I've got some carbide inserts I can try as well, but they're pretty small.
 
You must log in or register to reply here.
Back
Top