Needing more than a spark test?

[homebrewed]

I had figured it probably was the particular choice of FET but just had to verify it. Really no big deal since it's highly unlikely any of us will use a FET like that in our pulser, if we make one. I've got a bunch of small-signal BJT's and have no problem with a small loss in pulse amplitude due to Vsat so would just use them. I bought them by the pound when a local electronics company was clearing out their stockroom. Mostly 2N3904 and 2N3906 devices )

Back to the lathe, it also was frustrating because I had decided to make a punch and die set for the little lead ring around the base of my "focus ring". That had to wait until I put the lathe right again. It also didn't help that my bandsaw was acting up as well, and I needed it to make some of those bearing-installation fixtures. Sort of a cascade of problems but now resolved.

I'm currently machining the die portion -- an 1/8" thick steel rectangle that will have two holes in it to match the ID and OD of the ring. Lead is very soft so I'm just using low-carbon steel for the punch and die, no hardening required. The approach is a lot more wasteful than my original scheme but I think the end result will be much better.

 

[homebrewed]

I finally have been able to do more work on my XRF hardware. Before my lathe went out I turned the aluminum "focusing ring" but my punch and die scheme for making that little shield ring was thwarted -- until just recently. After getting the lathe up & running again I made the punch and die, formed the ring and epoxied the focusing ring and shield ring to my lead shield. After all that set up I mixed up another batch of epoxy and dipped the edge of each source into it, then placed them in the focusing ring. The shield ring acted as a "stop" so they didn't slip down too far.. That's setting up as I type this. So I should be able to re-assemble everything pretty soon and see what kind of count rate I get as a result of getting everything _much_ closer to the detector.

In addition to the shield ring I made a shield piece to wrap around the outside perimeter of my focus ring. I thought that would be a good way to further confine the sensitive volume, to reduce and/or eliminate false counts from fastener hardware, like the latch on the exterior of the box. That shield also is freshly-glued to the focus ring.

 

[homebrewed]

I've made some progress on my hardware and software for this project. First, a couple of photos:


The first shot was taken looking down the axis of the focus ring. Some of the Am241 sources can be seen, and the bottom edge of the detector can be seen on the other side (because the photo was taken a bit above the center-hole). The second shot is a top-down view showing my signal conditioning board, detector board and lead shield/focus ring in left-to-right order. I also cut a piece of lead to wrap around the OD of the focus ring to shield me and other extraneous hardware from 60Kev x-rays. I'm not as rigid as aluminum so in that context I guess I could be called "exterior software"

Some preliminary tests have shown that the count rate increases when an iron sample is placed up against the focus ring, but I didn't succeed in completely eliminating stray counts sneaking in from the sources. It's possible I'm seeing some counts from the lead shielding. If so it's going to be difficult to totally eliminate that with my setup. Backing off from the "get everything as close together as possible" approach may be required.

Advancement on the "real" S/W front: I adapted some serial I/O code I wrote for a different project, to set acquisition parameters (used by my Teensy front-end) and also to capture pulse data sent out by the Teensy. For now it will output pulse data in a CSV style format, making it easy to plot using a spreadsheet. It will be run in a terminal window, no fancy GUI for now.

I also found an ADC library written for the Teensy series of processor boards that looks pretty good. It should be relatively straightforward to implement a continuously-running setup using a circular buffer and (hopefully) relatively simple pulse-identification code. The open question is if the library + Teensy can deliver a high enough sample rate for this application.

 

[graham-xrf]

Well done on getting together the hardware.
I am also going for 8 sources. When you say "count rate", approximately how many would that be? I was always after how many events happen, and we have to hope that 60keV hits coming from 8 sources is enough.

You should be OK for ADC sample rate, because the pulses are "stretched". If you have pulses that happen too close to each other, that can get awkward.

You would have been hoping for almost dead quiet zero pulses, except when the iron is offered at it. It sounds as if you have a some part responding from everything and everywhere, and the rest coming from the iron. There cannot be responses from lead K-shell (energy too high), but there might be some at 10.55keV and 12.61keV coming from lead L-shell. Who knows, maybe these might be quite numerous.

I am thinking that "background subtraction", if that's what the software does, might be something like gathering the counts from when the iron is not there, and sorting their energies into buckets. A sort of "pre-testing calibration". Then, with iron present, the new set gets all those buckets presumed to have come from the lead "subtracted". I think that the fact you find the counts increase when you bring the iron up to it is just great. The darn thing is working OK !

When you get to it, your kit will want to see buckets of 6.4keV, 7.05keV, from Iron K-shell. The feeble energies (0.7keV) from iron L-shell won't wiggle the diode anyway. There may be other energies coming from the iron. Basically, if you see pulses in two energy groups, one being about double the energy (though not the count) of the other, one peak is lead, and the other is iron. (I think).

 

[homebrewed]

Well done on getting together the hardware.
I am also going for 8 sources. When you say "count rate", approximately how many would that be? I was always after how many events happen, and we have to hope that 60keV hits coming from 8 sources is enough.

You should be OK for ADC sample rate, because the pulses are "stretched". If you have pulses that happen too close to each other, that can get awkward.

You would have been hoping for almost dead quiet zero pulses, except when the iron is offered at it. It sounds as if you have a some part responding from everything and everywhere, and the rest coming from the iron. There cannot be responses from lead K-shell (energy too high), but there might be some at 10.55keV and 12.61keV coming from lead L-shell. Who knows, maybe these might be quite numerous.

I am thinking that "background subtraction", if that's what the software does, might be something like gathering the counts from when the iron is not there, and sorting their energies into buckets. A sort of "pre-testing calibration". Then, with iron present, the new set gets all those buckets presumed to have come from the lead "subtracted". I think that the fact you find the counts increase when you bring the iron up to it is just great. The darn thing is working OK !

When you get to it, your kit will want to see buckets of 6.4keV, 7.05keV, from Iron K-shell. The feeble energies (0.7keV) from iron L-shell won't wiggle the diode anyway. There may be other energies coming from the iron. Basically, if you see pulses in two energy groups, one being about double the energy (though not the count) of the other, one peak is lead, and the other is iron. (I think).

My guesstimate (which could be totally wrong due to my DSO's holdoff time) is that the incremental count rate is somewhere around 10CPS. A decent acquisition program with a circular buffer will really tell the tale. Working on that now.

I think the background counts I'm seeing are mostly due to the 10Kev photons from lead (but there are always other environmental contributors). One thing I didn't take into account is that the top inside corner of my shield ring isn't nearly as thick as the bulk so they could be coming from that region. Perhaps one reason to back off a bit from the get-everything-close approach, eh? Geometry seems to be more complicated than I originally thought. Something to consider is that lead should be pretty good at blocking its own XRF photons so perhaps "fringe effects" are talking. Apparently there's still room for improvement....

But as long as the counts from lead are relatively low, it should be OK -- subtracting a baseline spectrum if we have a low pileup rate should be fine. Hoping so at this point.

 

[graham-xrf]

Thanks for the count rate estimate. My admittedly crude calculations suggested a pessimistic 1 every 2 per second, while another was looking to be about 60 or 100 per second.

I have been thinking geometry.. Quoting Mark..
"One thing I didn't take into account is that the top inside corner of my shield ring isn't nearly as thick as the bulk so they could be coming from that region. Perhaps one reason to back off a bit from the get-everything-close approach".

We know that once a photon gets going, it will keep going to the end of the universe unless it hits an atom of something. What it encounters is pretty much empty space, even when it finds some material. I don't really know what happens as a photon encounters material, but I think it goes on in until it finds electrons to mess with. The resulting XRF photons then have to find their way out. The solid angle of candidates that got so lucky they end up aimed at the diode is small enough. The closer you get, the better the chances. Thus, lacking a diode about 150mm diameter under it, and maybe another over it, I think I still believe that getting up close is best.

Our sources can't see the diode directly, but they can beam their stuff "past the corner". How close you get may be governed by that.

I invested in a new soldering iron..


It had to happen! I take my chances that replacement tips may not be altogether straightforward to find, because although the supplier was in UK, I think it is probably Chinese. This one has the pump that sucks the smoke and fumes up a little pipe. The rest of it is the "slow hot air" stuff. I am taking a bit of a gamble here. The kit cost £85 ($112). I was not thinking I would be madly into circuit construction, so I am thinking it had better be the last soldering iron I ever buy.

I had another rake through all the available low noise amplifiers, with particular emphasis on stuff with SO8 packages. There are some extremely low noise types out there, even some with RF-type bandwidths, but my original choice LTC6269 is still clearly the best.

 

[homebrewed]

Thanks for the count rate estimate. My admittedly crude calculations suggested a pessimistic 1 every 2 per second, while another was looking to be about 60 or 100 per second.

I have been thinking geometry.. Quoting Mark..
"One thing I didn't take into account is that the top inside corner of my shield ring isn't nearly as thick as the bulk so they could be coming from that region. Perhaps one reason to back off a bit from the get-everything-close approach".

We know that once a photon gets going, it will keep going to the end of the universe unless it hits an atom of something. What it encounters is pretty much empty space, even when it finds some material. I don't really know what happens as a photon encounters material, but I think it goes on in until it finds electrons to mess with. The resulting XRF photons then have to find their way out. The solid angle of candidates that got so lucky they end up aimed at the diode is small enough. The closer you get, the better the chances. Thus, lacking a diode about 150mm diameter under it, and maybe another over it, I think I still believe that getting up close is best.

Our sources can't see the diode directly, but they can beam their stuff "past the corner". How close you get may be governed by that.

I invested in a new soldering iron..

View attachment 402267
It had to happen! I take my chances that replacement tips may not be altogether straightforward to find, because although the supplier was in UK, I think it is probably Chinese. This one has the pump that sucks the smoke and fumes up a little pipe. The rest of it is the "slow hot air" stuff. I am taking a bit of a gamble here. The kit cost £85 ($112). I was not thinking I would be madly into circuit construction, so I am thinking it had better be the last soldering iron I ever buy.

I had another rake through all the available low noise amplifiers, with particular emphasis on stuff with SO8 packages. There are some extremely low noise types out there, even some with RF-type bandwidths, but my original choice LTC6269 is still clearly the best.

The LT1028 has pretty amazing noise performance and is available in a DIP-8. The question is if its 75MHz GBW is sufficient for your purposes. I've been looking at low noise amplifiers as well, but with emphasis on very good low frequency performance like very low 1/F and (or?) flicker noise. The LT1028 is specified as having just 35nv peak to peak noise amplitude in the .1-10 Hz band and .85nV/sqrt-hz (typ) @1KHz. Current noise is not as good as the LTC6269 @ 1pA/sqrt-hz. I'm not looking at it specifically for a TIA but mulling over attempting an ultralow field NMR spectrometer.

I've found that a lot of so-called "low noise" amplifiers don't even specify what their noise current is. The omission usually means "not great".

 

[graham-xrf]

Absolutely agreed! If the feature was something to be proud of, it would be in the headline!
While I was rummaging about in the (rather small) cellar, finding junk from the last century, I came across this free sample promo that I never got around to opening. It hails from about 1996 when Comlinear was still a thing, though the datasheet is from 2001. There is even a small double-sided with ground-plane evaluation PCB stuck on the back of the card. From the days of through-hole mounting, and big enough to not disappear if you dropped it!



It's in a DIP package! GBW is 1.7GHz! Vnoise is 1.05nV/√Hz.

It's not bad, but..
It needs +/- 5V power supplies.
It has 100uV input offset voltage.
.. and various other things that make it fun only for some other one-off project.

For me, the 5.5fA bias current of the LTC6269 is the truly amazing spec. It allows having quite high value resistors in the circuit. Thinking about it, that amounts to (the charge effect of) only about 34000 electrons in a whole second!

This is not a good thought. Unless the manufacturer test getup was pretty special (glass?), the leakage across the PCB between the pins under a tiny MSOP or TSOT-23 might account for a fair fraction of that!

[EDIT: I have the datasheet of LT1028 in my XRF collection, so I have must have visited it before. Like the LTC6269, it's expensive! £10.49 ($13.84). 75MHz sounds like a lot, but if the first TIA gain is (say) 500,000, then the bandwidth is 150Hz, or put another way, needing nearly 1.66mS to rise or fall.

The published TIA Photodiode Amplifier on the LT1028 datasheet is a variant of the (several) TIA amps from Glen Brisebois, all featuring the discrete FET input. ]

 

[homebrewed]

Absolutely agreed! If the feature was something to be proud of, it would be in the headline!
While I was rummaging about in the (rather small) cellar, finding junk from the last century, I came across this free sample promo that I never got around to opening. It hails from about 1996 when Comlinear was still a thing, though the datasheet is from 2001. There is even a small double-sided with ground-plane evaluation PCB stuck on the back of the card. From the days of through-hole mounting, and big enough to not disappear if you dropped it!

View attachment 402360

It's in a DIP package! GBW is 1.7GHz! Vnoise is 1.05nV/√Hz.

It's not bad, but..
It needs +/- 5V power supplies.
It has 100uV input offset voltage.
.. and various other things that make it fun only for some other one-off project.

For me, the 5.5fA bias current of the LTC6269 is the truly amazing spec. It allows having quite high value resistors in the circuit. Thinking about it, that amounts to (the charge effect of) only about 34000 electrons in a whole second!

This is not a good thought. Unless the manufacturer test getup was pretty special (glass?), the leakage across the PCB between the pins under a tiny MSOP or TSOT-23 might account for a fair fraction of that!

[EDIT: I have the datasheet of LT1028 in my XRF collection, so I have must have visited it before. Like the LTC6269, it's expensive! £10.49 ($13.84). 75MHz sounds like a lot, but if the first TIA gain is (say) 500,000, then the bandwidth is 150Hz, or put another way, needing nearly 1.66mS to rise or fall.

The published TIA Photodiode Amplifier on the LT1028 datasheet is a variant of the (several) TIA amps from Glen Brisebois, all featuring the discrete FET input. ]

Sometimes I wonder if some of those amazingly good Ibias specs are just specsmanship -- the vendor comes up with the part and then puts it in a package that isn't at all that friendly for doing things like guard rings. Analog Devices has some (older) designs that are packaged in a way that actually makes sense if the user wants to make a truly low input-current circuit but that seems to be an exception rather than the rule.

 

[graham-xrf]

Yes, you are right. I wanted guard rings, but the spacing between the pins is just too close. One can surround the input pin most of the way, and hope that the capacitive fields around the pin go to it more than the capacitance pin-to-pin. If one could get a trace between the pins, then it would split pin-to-pin capacitance to be in series, but I think this is not an option for us (unless we go for big DIPs, or even SO8s).

Related to this, on my PCB, I split the transimpedance Rf gain resistor into two in series, and used two 3.3pF capacitors in series to go past the strays over Rf