Needing more than a spark test?

[WobblyHand]

Got a circuit to simulate using two LT6230-10's. A 1nA exponential pulse generates a 131mV pulse at the output of the second stage. Not a very fancy circuit, but it seems to work over a 3 decades of current. At 33pA, it outputs about 4.4mV, at 33nA it is just below saturation with a peak voltage of 4.2V. Yes it will have to be reduced in amplitude at the output. At the very low levels there's all kinds of weird stuff happening, bias current and input offset that is getting amplified up. I'm sure there are things that can be done. But having enough GBW is making this a lot easier.

It's not clear to me, what currents should be used. I used a dark current of 1nA, diode resistance of 40Meg, and diode capacitance of 150pF. I capacitively coupled the diode to the TIA using 100nF. Powered with 9V at the moment. At low impulse currents the output gets lumpy in the tail of the exponential. High currents look good. Pulse doesn't change characteristics, it just gets taller as it should. Whether this circuit actually works in real life remains to be seen.

10pA impulse. It settles out after a while. It might in fact be a simulation artifact. I have a minimum 100ps time step.

Impulse at 32nA.

On a completely different note, I was surprised to see today that @homebrewed Mark's circuit board arrived! Two days from the West Coast, that's pretty quick. Thank you so much!

Now I have to go shopping for parts... I'd like to settle on this front end though. Seems it is a critical element.

 

[homebrewed]

Hi Mark
Yes - the X-rays in the aluminium come in all sorts of directions, and may well be headed towards the diode. They keep on going through the relatively transparent and the arrive at the diode.

At least, that is what I think you might mean.

Sure, if you go with plastic, I guess you still need a lead ring fitted into it, surrounding the diode, to stop the initial Am241 radiation from simply heading sideways into the diode, releasing whatever they find in the ceramic carrier, the connection metals, etc.

Maybe I should be looking at 3D-printers?

Or is it that we just get some plastic rod, and turn that?

I have Acetal rounds that are big enough. Although I have PVC rod I wouldn't use it because the chlorine in it would be just as problematic as aluminum.

The way it's going to work for me is that the source ring will be glued to a lead plate with a hole in it for the XRF photons to pass through & strike the detector. I'll enlarge the ring so the carrier thingies (that's a technical term ) won't be visible from the detector's point of view.

 

[homebrewed]

On a completely different note, I was surprised to see today that @homebrewed Mark's circuit board arrived! Two days from the West Coast, that's pretty quick.

Excellent! I thought the Monday delivery date was a worst-case estimate, good to hear it got there so much sooner.

I realized I don't have enough header pins to populate my Teensy4.1 that I'm going to use as the test bed for my ADC driver so I ordered some from PJRC. The header pins came to less than $2 so I "forced" myself to get another teensy to play with

On the subject of the driver, I discovered that Teensyduino has a delayNanoseconds function. I don't know how accurate it is for relatively short delays. It uses the cycle clock that's running at 600MHz but it also has some setup code and a loop. I didn't see any tweaks for the time loss due to the overhead so I'd expect delays in the low nS range to exhibit a relatively high error. I may use it for the 45nS delay needed for data to settle after toggling the BYTESWAP input.

 

[graham-xrf]

Got a circuit to simulate using two LT6230-10's. A 1nA exponential pulse generates a 131mV pulse at the output of the second stage. Not a very fancy circuit, but it seems to work over a 3 decades of current. At 33pA, it outputs about 4.4mV, at 33nA it is just below saturation with a peak voltage of 4.2V. Yes it will have to be reduced in amplitude at the output. At the very low levels there's all kinds of weird stuff happening, bias current and input offset that is getting amplified up. I'm sure there are things that can be done. But having enough GBW is making this a lot easier.
View attachment 435983
It's not clear to me, what currents should be used. I used a dark current of 1nA, diode resistance of 40Meg, and diode capacitance of 150pF. I capacitively coupled the diode to the TIA using 100nF. Powered with 9V at the moment. At low impulse currents the output gets lumpy in the tail of the exponential. High currents look good. Pulse doesn't change characteristics, it just gets taller as it should. Whether this circuit actually works in real life remains to be seen.

10pA impulse. It settles out after a while. It might in fact be a simulation artifact. I have a minimum 100ps time step.
View attachment 435984
Impulse at 32nA.
View attachment 435986
On a completely different note, I was surprised to see today that @homebrewed Mark's circuit board arrived! Two days from the West Coast, that's pretty quick. Thank you so much!

Now I have to go shopping for parts... I'd like to settle on this front end though. Seems it is a critical element.

Oh yes! That is great! A working simulation!
I can see some of the things that I started out with.
When you get a waveform that "needs explaining", like the output at V(vo2), the first thing you check is the Absolute Current tolerance is set in the Tools --> Control Panel to 1e-012. It can help if the Voltage tolerance is set to 1e-008, or 5e-009

If it won't stop stepping after about 10 seconds, hit Escape, wait about 5 seconds, and hit it again.

I will have a look at LT6230. If OK with you, put the .asc circuit into a zip and post it.

Regarding the dark current, get it off the diode graph. 10V bias has it at 2.8nA. 30V bias puts it at 4.5nA.

From what I can see, it is an arrangement with the same resistor-chain offset from the Pocket Geiger, so it can work between 0V and +9V
My solution to that was the opamp rail driver I posted before, where you connect a opamp output to 0V, and use a voltage divider with capacitor shunts to make a very low noise dual rail supply.

The thing about LT6230 specification is the input bias current of 5uA with maximum 10uA. That is huge! It compares with currents for TIA amps like LTC6268 of 2fA to 4fA. I have to look at those numbers again. The difference is a factor of 2.5 trillion!. Basically, an amplifier that takes a bias current massively bigger than the little current pulse from perhaps only 20,000 to 80,000 electrons is maybe not such a good idea!

I think what is also happening is you have a x1000 voltage gain in the second stage. another 120dBV, all in one go?
300K for the first stage is OK, it should be able to take more, but I would not go above 1MΩ, and I would have a strong preference for the lower values around 200K to 500K. Use another stage as needed. Make the gain of the second stage x100, and the follow with x10.
I do get it that if this is happening on a Pocket Geiger board, you don't have that option, but now you have @homebrewed Mark's board, you have much more freedom to play.

The other specifications seem OK. I guess you would use the S6 package TSOT-23, if it goes onto the same tracks used by the LMC662.
I deliberately went for the single opamp in a 8-pin S8, because that leaves enough room between the pins to put a guard ring.

Still , congratulations! Now I am not the only one posting simulation plots.

 

[homebrewed]

Would a plain metal bud box be sufficient? How good does the grounding have to be all the way around? I have some circular connectors I could use. DB connectors are no fun to file out, and the punches are ridiculous.

I have left the sample end of my enclosure open to do some testing and didn't observe a noticeable increase in power line noise so depending on where they are a few relatively large holes probably wouldn't have much of an impact on noise. But the detector still _was_ behind a 1/16" thick lead shield, and the aperture hole is just .56" in diameter.

I priced out punches for DB connectors as well. Ouch! Round DIN connectors clearly were the way to go. Since my enclosure is made out of .25" thick aluminum I started with drills & finished the job with my boring head.

 

[WobblyHand]

Oh yes! That is great! A working simulation!
I can see some of the things that I started out with.
When you get a waveform that "needs explaining", like the output at V(vo2), the first thing you check is the Absolute Current tolerance is set in the Tools --> Control Panel to 1e-012. It can help if the Voltage tolerance is set to 1e-008, or 5e-009

If it won't stop stepping after about 10 seconds, hit Escape, wait about 5 seconds, and hit it again.

I will have a look at LT6230. If OK with you, put the .asc circuit into a zip and post it.

Regarding the dark current, get it off the diode graph. 10V bias has it at 2.8nA. 30V bias puts it at 4.5nA.

From what I can see, it is an arrangement with the same resistor-chain offset from the Pocket Geiger, so it can work between 0V and +9V
My solution to that was the opamp rail driver I posted before, where you connect a opamp output to 0V, and use a voltage divider with capacitor shunts to make a very low noise dual rail supply.

The thing about LT6230 specification is the input bias current of 5uA with maximum 10uA. That is huge! It compares with currents for TIA amps like LTC6268 of 2fA to 4fA. I have to look at those numbers again. The difference is a factor of 2.5 trillion!. Basically, an amplifier that takes a bias current massively bigger than the little current pulse from perhaps only 20,000 to 80,000 electrons is maybe not such a good idea!

I think what is also happening is you have a x1000 voltage gain in the second stage. another 120dBV, all in one go?
300K for the first stage is OK, it should be able to take more, but I would not go above 1MΩ, and I would have a strong preference for the lower values around 200K to 500K. Use another stage as needed. Make the gain of the second stage x100, and the follow with x10.
I do get it that if this is happening on a Pocket Geiger board, you don't have that option, but now you have @homebrewed Mark's board, you have much more freedom to play.

The other specifications seem OK. I guess you would use the S6 package TSOT-23, if it goes onto the same tracks used by the LMC662.
I deliberately went for the single opamp in a 8-pin S8, because that leaves enough room between the pins to put a guard ring.

Still , congratulations! Now I am not the only one posting simulation plots.

I was happy to get anything like a pulse. Especially after the crashing and burning I had earlier. Still, there's some stuff that doesn't look right, like the output going below the rail. Definitely is too much gain in each stage. Was just plain getting greedy for something to work. But I do know that dynamic range will suffer - which is sort of what I am seeing.

I will look for some TIAs. I did find the selection confusing, which is why I chose something with a large GBW, just to get started. At low current levels, the weirdness was in the first stage, the second simply amplified it.

Here's a copy of the file.

 

[WobblyHand]

Tried out simulating the better TIA, the LTC6268-10 and the LTC6268. Put the faster one in the front end. Chain of 3. Now a 1pA pulse looks ok. Adjusted the gain to get 2V for 1nA. For some reason the slower parts cost more. Go figure. About ~ $5/amplifier for the fast ones, $10/amp for the slower wider temp range ones. Seems to be a bit better behaved now. Simplified the schematic some more. Made the middle stage non-inverting, for some reason the LTC6268-10 did not like driving a lower impedance. Then added an inverting stage.

1pA puts out 2mV at the peak and looks like its bigger brother. Quite a bit better. Now to see if I switch over the parts to the faster ones in the back end, the circuit still works. That's for tomorrow.

 

[graham-xrf]

@WobblyHand Hi Bruce
Thanks for posting the simulation. I have had a little fun with it, watching the traces slowly drawing themselves . Then I removed the 100pS maximum timestep, and it then fairly zips along.

Next was to have a change to the capacitors. Way back, when I was exploring exactly how the photodiode charge gets into a TIA, I found as the value is increased from 1pF all the way to 2.2nF, there was a definite "best place". The point of diminishing returns is between 1nF and 2nF. In practice, trying to charge a 1uF or higher from less than 100,000 electrons is not something I am sure of,

10uF and 100uF in later stages is also way too much for me. That is what one does for exxtended bass response in audio hi-fi. Here, we are happy to have a really poor low frequency response, to lose the low frequency noise. It will not stop the random "shot noise" pulses that can happen, but our pulses are over in 20uS.

I do get it that in this circuit, shifted halfway up the V+ rail, you need to lose the offset before trying it onto a ADC, unless you use differential input, or shift everything up for the ADC as well.

(Nearly) DC coupled, and three stages
OK - I had a separate go at trying to get a good result using only three opamps, in two packages, hiking the gain more.
Here is one that you can see does not have AC coupling except right at the TIA input, and it might be possible to lose even that capacitor. The optimum value is somewhere between 1nF and 2nF before we get into diminishing returns. Also, it's very simple, all in a row, opamps.

The assumed input current from the photodiode is only about 280pA in this example, and I am taking that to be a "maximum". If real pulses get to 500pA, or 1nA, then we can reduce gain. Note the exponential that generates it starts at 500pA. It "loses" some when we don't give it time to to get there, because that is an infinite wait.

I removed the 100pS maximum timestep. I set the Absolute Current tolerance to 1e-014, and the Voltage tolerance to 5e-009.
I shortened the plot time to 20uS. It is delivering slightly over 2.0V with about -1.1mV offset at the output, without any fancy loops to remove that offset.

Given that we do not have any idea what the true size of a "maximum" pulse should be, and that we may have to amplify 20pA pulses, and that we may want to "zoom in" at times, there is a case for being able to switch gain ranges.

Regardless - I now think we have the measure of this thing. If the photodiode does anything, I think we can capture it.

 

[WobblyHand]

@WobblyHand Hi Bruce
Thanks for posting the simulation. I have had a little fun with it, watching the traces slowly drawing themselves . Then I removed the 100pS maximum timestep, and it then fairly zips along.

Next was to have a change to the capacitors. Way back, when I was exploring exactly how the photodiode charge gets into a TIA, I found as the value is increased from 1pF all the way to 2.2nF, there was a definite "best place". The point of diminishing returns is between 1nF and 2nF. In practice, trying to charge a 1uF or higher from less than 100,000 electrons is not something I am sure of,

10uF and 100uF in later stages is also way too much for me. That is what one does for exxtended bass response in audio hi-fi. Here, we are happy to have a really poor low frequency response, to lose the low frequency noise. It will not stop the random "shot noise" pulses that can happen, but out pulses are over in 20uS.

I do get it that in this circuit, shifted halfway up the V+ rail, you need to lose the offset before trying it onto a ADC, unless you use differential input, or shift everything up for the ADC as well.

(Nearly) DC coupled, and three stages
OK - I had a separate go at trying to get a good result using only three opamps, in two packages, hiking the gain more.
Here is one that you can see does not have AC coupling except right at the TIA input, and it might be possible to lose even that capacitor. The optimum value is somewhere between 1nF and 2nF before we get into diminishing returns. Also, it's very simple, all in a row, opamps.

The assumed input current from the photodiode is only about 280pA in this example, and I am taking that to be a "maximum". If real pulses get to 500pA, or 1nA, then we can reduce gain. Note the exponential that generates it starts at 500pA. It "loses" some when we don't give it time to to get there, because that is an infinite wait.

I removed the 100pS maximum timestep. I set the Absolute Current tolerance to 1e-014, and the Voltage tolerance to 5e-009.
I shortened the plot time to 20uS. It is delivering slightly over 2.0V with about -1.1mV offset at the output, without any fancy loops to remove that offset.

Given that we do not have any idea what the true size of a "maximum" pulse should be, and that we may have to amplify 20pA pulses, and that we may want to "zoom in" at times, there is a case for being able to switch gain ranges.

Regardless - I now think we have the measure of this thing. If the photodiode does anything, I think we can capture it.

View attachment 436008

Thanks for the write up. I originally put in the minimum time step because spice was non cleverly using a much larger step and the pulse itself was oddly distorted. Spice often figures out the correct time stepping, but for some reason it wasn't working correctly. So I forced it.

High gain DC systems simulate fine, but in practice they saturate easily. Spice doesn't simulate all the process spreads of bias and offsets. I have had to change various compensating resistors to hand tweak the outputs back to mid range. A well configured DC chain is very nice, but we would need adjustments to get it to work. I believe several of us could set the chains up properly, but for many, it could be difficult.

AC coupled chains are easier to setup, but one has to deal with baseline offsets due to pulse rate variability. At higher count rates the baseline will lower, to bring the waveform to it's average. So the SW has to compensate, and that process may introduce error. As I write this it's unclear to me which I prefer. Both techniques have their pluses and minuses.

That being said, I am drawn to the DC approach.

 

[graham-xrf]

Thanks for the write up. I originally put in the minimum time step because spice was non cleverly using a much larger step and the pulse itself was oddly distorted. Spice often figures out the correct time stepping, but for some reason it wasn't working correctly. So I forced it.

High gain DC systems simulate fine, but in practice they saturate easily. Spice doesn't simulate all the process spreads of bias and offsets. I have had to change various compensating resistors to hand tweak the outputs back to mid range. A well configured DC chain is very nice, but we would need adjustments to get it to work. I believe several of us could set the chains up properly, but for many, it could be difficult.

AC coupled chains are easier to setup, but one has to deal with baseline offsets due to pulse rate variability. At higher count rates the baseline will lower, to bring the waveform to it's average. So the SW has to compensate, and that process may introduce error. As I write this it's unclear to me which I prefer. Both techniques have their pluses and minuses.

That being said, I am drawn to the DC approach.

On the circuit I sent, if you short circuit the capacitor, the output changes polarity, and messes up.

Given that you are cautioning what may happen in a practical build, should I use preemptive caution, and include 0603, or 0805 pads, with 0Ω links, just in case on power-up, one needs to introduce a capacitor or two?