Ooh - Lots of goodies! Well done Mark
OK - I let my thoughts roam a bit here, and I may sometimes get things wrong, but you deserve a considered response.
1. Looking at the spectra, just for a at-a-glance impression, the
1018 sample shows definite "blocks" of similar counts that have a "width". Interspersed are "thin" jumps to higher counts that may not be noise. The buckets under the distribution, even if some had smaller counts, cannot show, because they appear to agree to be at least as full as their neighbors, up to the point one found it's maximum, whereupon one or the other might go on to make a "higher thin peak".
2. We do know that low counts of high energy events are possible, because there are places away from the distribution peaks which show some single buckets having consistent counts that came from bigger pulses, but less often. It tells us quite a lot!
3. I do assume that any one of the very thinnest lines there would be the count from a single bucket. This (thin) would be the resolution from one bucket to the next. We get the "shape" because the energy count decision as to which bucket it qualifies for, compared to its neighbor is that bit uncertain, from noise, and wave shape consistency.
We expect some of this, even from the very best resolution spectrum histogram kit. What is perhaps a bit unexpected here is that there are many random, non-typical high counts right in the middle of them.
This characteristic having largely disappeared from next two spectra (aluminium and zinc) is what jumps out at us!
4. I do get it that the
1018 steel spectrum was raw, not filtered by any digital processes. I do not think you should lose faith in your pulse qualification, and matching. Clearly the aluminium and zinc spectra have a long floor histogram to the right, indicating low probability of higher amplitude pulses. Also, the random untypical counts of buckets compared to immediate neighbors has greatly reduced, or disappeared. The "clean-up" of the latter two spectra gives them a look of entirely better cred.
5. Forgive that the words are a bit non-scientific sloppy. I have to keep reminding myself that a high-looking peak on a spectrum display does
not mean the pulses had higher amplitude. It does mean there were a whole lot more of pulses that had a range of amplitudes, and that range was quite wide in two places.
6.
The count rate?
The nicest thing you tell us is that when you take the sample away, the count rate drops to less than 1CPS.
I am a little confused at comparing that to the count rate displayed on the output.
1.72/sec for aluminium, counting 2925, apparently taking 28.3 minutes.
0.39/sec for zinc counting 2002, apparently taking 85.5 minutes.
Wow! Do I have this right? 1CPS is a bit close to (say) 1.72CPS.
That might make too large a proportion of the count belong to those that would have come anyway.
Regardless, I was hoping for wait times to be a few minutes, rather than half-hour events. Do we really need 16 or 24 sources?
Without the sample there, and expecting to let it run for half an hour, and hoping that the "unwanteds" represent (say) 10%, that would make the unwanteds contribute about 180. I guess you might have let it run for a time with no sample there?
7.
Re: Probability scaling
One does notice the shape to the left side (low energy end) of the plots. There is clearly a sudden cut-off. It is there also on the 1018 spectrum, but the 1018 sample was not contributing much at that low energy, so perhaps expected. What we do see is that both spectra have two clusters. Also, we see that the zinc spectrum has several buckets showing peaks of their own at energies between the two large distributions.
Starting from "what we expect", getting returns from aluminium at all was a bit of a long shot, something to hope for.
There would be
1.48KeV and
1.55KeV. These might end up in the same bucket at times, or more likely, populate several buckets to either side.
The probabilities of the photodiode responding to these energies is about 1.5% or so. The counts we get to, at those energies, deserve to be "scaled up" by (count/0.015), or about x66.
Alternatively. the counts of higher probability responses deserve to be "scaled down", essentially according to the inverse of the X-100-7 probability vs energy curve. This may well already be built in to the MCA code. In the end, we have to be pleasantly surprised we see counts from aluminium. It gives hope that we might in future discriminate to see magnesium, if the area count-up can be consistently, even if only slightly, lower. (
1.25KeV and
1.3KeV)
8.
The Zinc spectrum
Of course, we have to notice that both the aluminium and the zinc results both come up with similar looking plots, with the "extra" counts in between the peaks on the zinc sample. Zinc should deliver
8.6KeV and
9.5KeV, which at our resolution, may well make counts over a range of some boxes either side. These counts would be inflated by the near 100% probability at that energy, which again, I guess is taken into account. Zinc also delivers a couple of very low energy returns
1.01KeV and
1.03KeV at less than 0.1% probability. That energy is less than even magnesium. They would surely not have "helped" the count at all.
9.
Looking at the amplifier choices
More and more, I think we are getting there, and I envy your progress. Deciding to change the TIA may be a good idea, but if the choice is driven by voltage noise 8nV/√Hz only, I think there may be others, not too costly, that you might like. In a TIA, the bias current, and the current noise density become very important. Also, in a TIA, one desires the feedback resistor noise be way lower than the inherent voltage and current noise from the opamp front end, and in most TIAs, this is normally the case. It might not be with 66MΩ.
ADA4177 Noise (compared to say LTC6268)
Voltage noise density 8nV/√Hz vs 4.2nV/√Hz Not a huge difference. About half in LTC6268
Current noise density 0.2pA/√Hz vs 5.5fA That is a difference x36 more in ADA4177
Input Offset Voltage 2 to 60 uV vs 200uV The LTC6268 has more input offset voltage.
Input Offset Current 1nA vs +/- 6fA
Input Bias Current -400pA to 1nA vs +/- 3fA The difference is huge! x 66,000
Gain Bandwidth Product 3.5MHz vs 500MHz If you want high gain, you need this.
In a TIA, what counts for me are the last two. We are trying to capture the tiny current from a photon.
The LTC6268 can see to within +/- 3 femto-amps! The ADC4177 has more than the signal even before it starts.
Allowing the LTC6268 has a uncertain range 6fA, and choosing the typical 400pA for the ADC4177, I compare 400pA from to the calculated 45pA to 100pA expected. I would try not to have input bias current more than the expected signal, even if it can be a "steady" value.
About the Rf - less noise than you think!
While RF=66MΩ has "more" Nyquist generated rms noise voltage √(4*k*T*B*R), where k is Boltzmann’s constant, T is temperature in Kelvin, R is the resistance, and B is the bandwidth, this does not necessarily decrease performance just because it's "higher", and one does not necessarily have to lower impedances to lower noise. It all depends on other things.
The current is measured by passing it through a (opamp assisted) resistor and measuring the resulting voltage. The voltage developed is
proportional to the resistor value. BUT.. the Johnson noise of the resistor is proportional to the
square root of the resistor value. There is a 3dB improvement in the signal-to noise ratio each time the resistor value is doubled. 66MΩ instead of 1MΩ gets 18dB better S/N ratio.
One would think there is no downside to simply going as high as one pleases, but there are practical limitations. One is simply being able to keep electrons that belong to the current we want to measure from escaping across the PCB as we try to. Another is the integrator filter time constant of the circuit.
For me, the main blockage is the Gain Bandwidth Product. If we want to preserve the waveshape of a pulse that happens in 13uS, and has components to 300kHz, and have gain at the same time, then 3.5MHz is too low. We already have the full analysis of what happens with LMC662. I admit I am surprised that it manages as well as it does, and I have a lot of catching up to do in the practical constructions.
In getting there with some certainty about what is going on, I would short-circuit the input, and be curious about the counts. I would short-circuit the input of the ADC, and be curious about what inevitable (low) count we always see. There needs to be enough bits in the count to see this, and I imagine Mark can hardly wait to tro out his new ADC(s). I would try to create artificial voltage pulses, and test the route, maybe not including the TIA at first..
I suppose it is possible make an artificial photon test input using an attenuated signal generator pulse, put into a series resistor. 1mV into 10MΩ amounts to a 100pA input current. That is the sort of order we might be interested in. Say a range of about 50Pa to about 1nA. I never tried stuff like this before. Even measuring on this thing is going to be tricky. Scopes can show one quite a lot that isn't really there!
When one knows with some certainty what the TIA does, and what counts we get, then maybe go after some diode photons. One hopes to be able to get single, quite narrow spikes in the plot, from a constant amplitude pulse. In testing and use, changes by movement, etc. should not alter the amplitude of an arriving photon. It should only affect t the count of how often it happens, not the bucket it ends up in.
Regardless everything I think, and what I may be wrong about, I think what Mark is doing, and the stuff he is looking at, is just fantastic!
