WDS preamplifier design

[sem-geologist]

Ok, this is some of my take on the schematics (made in KiCad) based on visual inspection and probing the zero resistance between pins of components:


C3 on it is with question mark as its value is incredibly hard to read as those metal plates surrounding high voltage part of circuit obstacles any clear view. Bypass capacitors of A847 probably are some standard values and is not so important to signal handling.
A203 - the "black box" I had redrawn as a component based on the website presented version for A203 documentation:

yes that schematics of AMPTEK is very unrealistic and highly abstracted.
Considering more detailed information on CSP and shapping amplifier workings at Cremat Inc site, I guess A203 uses 2 OPAMPS for shapping amplification. On A203 webpage in picture with metal-cover-off I think there is 3 tiny IC's visible. 1 would be OPAMP for C.S.P and other two for Shapping amplifier.
As I wrote before there is clear trace connecting from pin 8 and going to that mica capacitor. So A847 does differentiation and drives the output voltage, where two BJT transistors together with two diodes and resistors forms so call Class AB amplifier and provides the signal with enough of current so that there would be no voltage drop. Finnaly output signal is decoupled with 47uF electrolytic capacitor.

Regarding the A203, the datasheet is frustrating, as it provides very little information on the circuit inside.  The feedback capacitor for the charge amplifier is 2 pF, but the fall time is stated as being only 30 μs.  If this represents the Cf * Rf time constant, then this would imply Rf = 15 MΩ, which is strangely small.  At 250 ns, the shaping time constant is also quite small.  I guess this would help prevent pulse pileup at the expense of signal-to-noise ratio.  Maybe someone could scavenge an A203 from a decommissioned SX-100 and pry the top off?  Wishful thinking.  I haven’t gotten to the JEOL shaping amplifier yet; the main amplifier board and power supplies are located in the “intelligent unit.”

EDIT:  I believe that differentiation occurs between the A203 and the AD847 buffer amplifier and involves the odd-shaped capacitor and probably the resistor next to it.  Like I noted above, I can see a trace from the capacitor that goes underneath the A203 (i.e., toward its output).  Also, one lead of the resistor oriented parallel to the capacitor is located very close to pin 3 (noninverting input) of the AD847.

What?, 30µs for decay is good - not bad. That allows this preamplifier to cope with very high count rates and the feedback capacitor... well, actually rather C.S.P. OPAMP feedback to not hit the Vss or get saturated. (A203 CSP OPAMP is not "rail-to-rail" type).

250ns is also very suitable (or golden spot) shapping time, if You had payed previously attention in my previous post, someone  demonstrated that proportional counter produced pulse takes about 200ns. Check it again if You had missed it:
https://physicsopenlab.org/2017/07/23/x-ray-proportional-counter-2/
and/or the key figure:


So 250ns is just very right shapping time constant for GPC as it is neither too long (witch would increase troubles for high counting rate) or too short (then there is not all charge collected, which is not good too, and less precise amplitude). 250ns is enough to collect all charge or in other words - to integrate the pulse of event fully, as that is the primary function of C.S.P. for GPC. (for G-M tube it would be dominated with differentiation, for Tl screen scintilation as well discrimination, but for GPC we need integration).
I think working of A203 is quite clearly described, and more information can be obtained rather with trying it with artificial testing board, than dismantling. I think Cremat Inc has much better description, and I find that it rather works very similary as those parts from Creamat Inc, A203 is just nicely packed in single metal box. For these kind of low signals this looks very suitable and shapping time looks to be optimized for GPC, so being more resilient against noise, CPS+SA integrated, and easy replacable is bigger selling point than transparency of components.

As for discussion about those capacitors, thanks to all this thread I accidentally found out that 220nF capacitor is connected wrongly on one of our spectrometer, which is current out of order (it is extended spectrometer, crystal turret stuck  ). Again my question is how far high voltage power supply is from the preamplifier box (approximate length of cable) on Jeol. I have no idea what the "intelligent box" is or where it is situated.

[Brian Joy]

Ok, this is some of my take on the schematics (made in KiCad) based on visual inspection and probing the zero resistance between pins of components:

Cool!  Thanks for posting this.

I didn’t necessarily mean to be critical of the time constants for the A203 preamplifier and shaper – I was just making an observation that these times are relatively short compared to others.  For instance, compare the 30 μs A203 preamplifier time constant with that of the A250 (300-1200 μs), the Cremat CR-110 (140 μs) or the JEOL preamplifier (100 μs).

Below is a table from Cremat that shows the tradeoff between pulse width and voltage noise for the CR-200 shaping amplifier (noting that FWHM = shaping time constant * 2.4)  Clearly, both count rate and signal amplitude must be considered when choosing a shaping time constant.  I’ve been searching for shaping circuits, and I find that time constants of 1-2 μs appear to dominate (though amplitude of the signal from a solid state detector is only ~1/10 that from a proportional counter):



I intend to post considerably more on this subject, as I’ve decided that I’m going to build my own gamma ray detector.  I’ll use the following webpage at Cremat as a starting point, but I’m going to build my own preamplifier and shaping amplifier so that I can understand these circuits better.  I’ve ordered three of the elusive Hamamatsu S1223-01 (13 mm² active area compared to 6.6 mm² for the S1223) from RS Components in the UK (20 USD apiece, excluding shipping, etc.); they should arrive in about a month, though I can get started building and testing the preamplifier before that (when I get time, that is):

https://www.cremat.com/femtojoule-detection/

Yes, I remember your post/figure showing experimental determination of the pulse width.  However, the pulse will acquire a tail simply due to the capacitance of the counter, cable to the preamplifier, and especially the coupling capacitor (which has capacitance many times greater than the counter and cable).  The pulse shape will be altered significantly by the time it reaches the preamplifier.  Not only will it possess a long tail, but the peak will be rounded.

The cables to/from the preamplifiers to the intelligent unit, which is situated immediately behind the microprobe, appear to be ~3 m long.  Considering the 51 ohm resistor at the output of the JEOL preamplifier, the signal is passed through RG58 shielded cable with capacitance of about 100 pF/m.

[Brian Joy]

As I wrote before there is clear trace connecting from pin 8 and going to that mica capacitor. So A847 does differentiation and drives the output voltage, where two BJT transistors together with two diodes and resistors forms so call Class AB amplifier and provides the signal with enough of current so that there would be no voltage drop. Finnaly output signal is decoupled with 47uF electrolytic capacitor.

I agree that you’ve drawn the circuit correctly, however, your interpretation of the function of the AD847 is not correct.  In the circuit as you’ve drawn it, C5 and R6 constitute a passive differentiator in series with a noninverting amplifier with gain = R10/R9 + 1 ≈ 20.  As I noted before, an active differentiator also inverts, but this circuit does not; the polarity of the pulse remains unchanged between the A203 and output.

I’ve attached an LTSpice “equivalent circuit” for the portion of the circuit between the A203 and output to cable.

[sem-geologist]

Hmm, You can be right about that. I also had spot error in my previously sent schematics after trying LTSpice simulation (BTW, thanks for LTSpice file). C3 should be 220pF, not 220nF, that makes a huge difference. I am attaching modified simulation file, with updated input pulse description to be more closer to real-like output from A203 (Attached below).
at simulation I get something like this which resembles what is observed on oscilloscope. Please note that pulse input with straight line raise, straight line "on-top" and straight line fall is much approximated from more round pulse shape from A203. Nevertheless I think this demonstrates this satisfactory well:


So what is the function of AD847? - that gets here obvious. it is used  in non-inverting configuration so that input would be infinite impedance (non influencing CR passive differentiator depending from load...) and it is amplifying that differentiated pulse. Also differentiation as expected reduces the amplitude of pulse, and thus it needs amplification before being sent on cable. The question then is why engineers implemented own passive differentiator if such passive differentiator is already built in A203 (pin 9). Why not just directly forwarding the output of pin 9 straight into positive input of AD847?

[sem-geologist]

Ok, while this is slightly out from "preamplifier design" topic, I think it can be useful for better understanding:
https://doi.org/10.1119/1.4923022
Yes "can" here is a pun, as that proportional gas flow detector which uses P10 gas it is made from aluminum can  .
Looks it would not be practical for WDS purpose, as aluminum can walls would absorb most of X-rays... but for gamma rays - that rather would be efficient enough. Damn, I don't know how, but that just made me thirsty... Should be the best cold beverage advertisement ever

Brian, but Yes, I think your aim to make Your own Gamma detector based on PIN chips is better idea as 1) won't need gas, 2) don't need HV generator. Go for it!

However, signal incoming to preamplifier will be a bit different than from GFPC It will take few µs not few hundred ns. Also because signal is much smaller, it is common to have no feedback resistor on feedback of preamplifier (all EDS preamplifiers have like that) and have reset circuit. Feedback resistor is kind of source of additional thermal noise (As all resistors do) and because solid state detectors produce much smaller signals the preamplifier design is "resetting" kind, where on GFPC due to much higher amplitude there is this resistor which decays the signal.

[Probeman]

Yes "can" here is a pun, as that proportional gas flow detector which uses P10 gas it is made from aluminum can  .
Looks it would not be practical for WDS purpose, as aluminum can walls would absorb most of X-rays... but for gamma rays - that rather would be efficient enough. Damn, I don't know how, but that just made me thirsty... Should be the best cold beverage advertisement ever

I'm reading the book "Superabundance" which is quite fascinating:

https://www.amazon.com/Superabundance-Population-Innovation-Flourishing-Infinitely/dp/1952223393

and there is a brief mention (regarding resource utilization) on how the first aluminum cans weighed about 6 times more than today's cans. 

So the can walls are getting thinner...   

[Brian Joy]

However, signal incoming to preamplifier will be a bit different than from GFPC It will take few µs not few hundred ns. Also because signal is much smaller, it is common to have no feedback resistor on feedback of preamplifier (all EDS preamplifiers have like that) and have reset circuit. Feedback resistor is kind of source of additional thermal noise (As all resistors do) and because solid state detectors produce much smaller signals the preamplifier design is "resetting" kind, where on GFPC due to much higher amplitude there is this resistor which decays the signal.

I’ll probably start with a reproduction of the JEOL preamplifier, though I’m aware of the “Johnson”/thermal noise problems that high value resistors can create.  I have a schematic for a circuit that discharges the feedback capacitor without use of a resistor, and I’ll go this direction if I find that the magnitude of the thermal noise is too great.  Note that the Cremat CR-110 preamplifier used in the femtojoule detection demonstration contains a 100 MΩ feedback resistor (and 1.4 pF feedback capacitor).

I’ve now gotten a shaping amplifier to work along with the JEOL preamplifier in LTSpice, though a few problems are present.  I’ll post further about this.

[sem-geologist]

I just wan't to point out that LTSpice simulation results can get colossally different depending from how good the components are described. In particular at these moderate length (200ns) type of signals (decades ago it would be classified as ultra short, but at current trends of THz signals, they are only moderate) the effects of internal induction, internal resistance and capacitance of components plays very important role. It looks that LTSpice use some default values then those secondary parameters are not specified. I seee more expected results if I replace U1 as preamplifier OPAMP replacement with theoretical OPAMP4 in LTSpice component list (I can then simulate more correctly the circuit without feedback resistor (EDS-like preamplifier), Where i.e. using some fast amplifier like AD 744 would make some saturated output from time 0.

[Brian Joy]

Keeping in mind that I’m learning as I go…

Output from the WDS preamplifier typically is next passed to a shaping amplifier.  The purpose of the shaping amplifier is to amplify the signal from the leading edge of a pulse while attenuating pulse tails and electronic noise (see figure below).  A shaping amplifier constructed with multiple stages of integration after initial differentiation produces “quasi-Gaussian” pulses – in terms of V(t) – with large signal-to-noise ratio and narrow widths (with FWHM generally ranging between 100 ns and 20 μs).  The circuit, which constitutes a band-pass filter, could be constructed simply with passive components (resistors and capacitors), but, on a plot of gain versus frequency, steeper gain “roll-off” above the passband is obtained from cascaded active filters constructed with op amps.  In the two commercial designs that I’ve seen so far, initial passive C-R differentiation, constituting a high-pass filter, is followed by two cascaded active second-order low-pass filters, producing four stages of integration (“poles” in the frequency response).  The Bessel filter response appears to be the most appropriate for pulse processing, as it shows better stability during transient events than the Butterworth or Chebyshev responses.  A drawback of the Bessel response is that, when viewing a plot of gain versus frequency, it attenuates the signal above the “corner frequency” less steeply than the Butterworth and especially the Chebyshev response.  Another drawback of the Bessel response is that gain within the passband is not as uniform as that of the second-order Butterworth response.  Cremat uses the Sallen-Key topology for their CR-200 shaping amplifier, but they do not specify the filter response (see attached CR-200 application guide).  In a design that I’ll address in more detail below, the multiple feedback (MFB) filter topology is used in conjunction with the Bessel response.

Below is a figure that shows output from the Cremat CR-110 preamplifier (bottom) as well as simultaneous filtered output from the CR-200 shaping amplifier.  The shaping amplifier (with 1 μs shaping time or FWHM = 2.4 μs) dramatically reduces pulse width and equally impressively increases signal/noise:



Ideally (see below), the width of the peak produced by the shaping amplifier is controlled by the initial passive differentiation stage, with “shaping time” given as C_in * R_in; FWHM is obtained by multiplying by 2.4, ideally.  Shaping times typically range between tens ns and 10 μs, with longer times giving better signal-to-noise ratio.  A pulse from a solid-state detector would typically require a longer shaping time than one from a proportional counter because the amplitude of its signal is roughly an order of magnitude smaller than the ~3 mV signal produced by a proportional counter.  Since the rise time of the pulse from the preamplifier is small, only some portion of the leading edge of the pulse (as modified by the preamplifier coupling capacitor and noise) will be characterized by dV/dt of appropriate magnitude to place it within the passband of the filter.  Note that, in the figure above, only regions of steep dV/dt are amplified, with amplitude of the resulting pulse proportional to ΔV above the surrounding low dV/dt + noise region (commonly consisting of superimposed pulse tails).  For shorter shaping times, the passband is moved to higher frequency (greater dV/dt for this case), The figure below from Cremat shows gain/attenuation as a function of frequency for the CR-200 (application guide attached) for various shaping times.



I’ve attached a paper by Perevertaylo et al. in which schematics are presented for a commercial shaping amplifier and charge-sensitive preamplifier.  I’ve simulated the shaping amplifier in the LTSpice simulator.  It comprises a passive C-R differentiator followed by two cascaded second order low-pass Bessel filters constructed using the “multiple feedback” (MFB) topology.  The circuit includes adjustable “pole-zero cancellation” used to compensate for undershoot on the trailing edge of the shaped pulse.  The small-signal voltage gain is also adjustable by means of a variable resistor.  The output pulse FWHM is stated as 4 μs.  Here is the schematic (attached as a .asc file for LTSpice):



The figure below is a so-called Bode plot that shows both gain (heavier curve) and phase shift of the Perevertaylo et al. filter as a function of frequency.  The passive differentiator (high-pass filter) is responsible for the 20 dB per decade slope of the gain curve at frequency below the passband.  (Voltage gain [dB] = 20 × log₁₀ [output voltage / input voltage], i.e., 20 dB corresponds to a voltage gain of ten.)  The passband is centered at about 100 kHz, which is roughly consistent with placement of the passband in the CR-200 (for interpolated shaping time = 1.5 μs).  At frequencies above the passband, the active low-pass filter attenuates the signal at -40 dB per decade.  According to Perevertaylo et al., gain should range between 10 and 140 (20-43 dB), while I get values between about 14 and 400 (23-52 dB); I’m not sure of the reason for the discrepancy. 



Combining the JEOL preamplifier and Perevertaylo et al. shaping amplifier produces the following circuit (attached as .asc file):



As before, when simulating the circuit in LTSpice, I’ve simply used the leading edge of a square wave as input to the preamplifier; I’ve set its amplitude at -300 μV.  The preamplifier coupling capacitor, C3, can produce substantial effects on performance of the shaping amplifier, and I’ve used the simulator to assess variation in pulse width and distortion (mostly just visually) as a function of the value of C3.  (The value of C3 can be viewed as a sum of C3 itself and detector and cable capacitances, which should be much smaller.)  For instance, in their preamplifier, Perevertaylo et al. use a 200 pF coupling capacitor in conjunction with a (Cd,Zn)Te (CZT) detector (with 5 pF capacitance).  In the femtojoule detection demonstration, Cremat uses a 10 nF coupling capacitor with a PIN photodiode (though its capacitance is only a few tens pF).  The JEOL WDS preamplifier contains a 2.2 nF coupling capacitor and CAMECA, 10 nF(?).  If I use a 1 nF coupling capacitor, then I get the results below.  The broad, low pulse is the input signal from the preamplifier, while the tall, narrow pulse is the output from the shaping amplifier (at minimum shaping amplifier gain).



For the shaping amplifier output shown above, FWHM is about 4.3 μs, with some asymmetry in the shaped pulse apparent.  The output of the shaping amplifier appears to be centered on the uppermost part of the trailing edge of the preamplifier tail pulse, but this is misleading, as formation of the shaped pulse can require a few microseconds.  This becomes much clearer when the value of C3 is lowered to 100 pF (shown below).  Along the tail of the input pulse, the value of dV/dt is too low for it to fall within the filter passband, and so the centroid of the shaped pulse must represent some portion of the leading edge of the input pulse and is delayed ~2.7 μs (at least at the centroid).



Here is a view with the time axis stretched out a little:



In the plots above, the value of FWHM has decreased to 3.8 μs, which is much closer to the ideal value of 3.6 μs and also consistent with the value advertised by Perevertaylo et al. (4 μs).  If C3 is increased to 10 nF, the pulse from the preamplifier becomes much more rounded; voltage gain within the preamplifier increases dramatically but not linearly.  Increase in C3 causes asymmetry of the shaped pulse to increase and also creates a greater range of time over which dV/dt along the leading edge of the pulse falls within the passband; these two effects produce an increase in FWHM.  For this case (C3 = 10 nF), FWHM = 9.3 μs.  I would guess that it’s possible to make the peak more symmetrical with a final stage of differentiation, but I haven’t explored this in any detail.



But enough with the simulator and onto construction!!  The audiophile transistors that JEOL uses in its WDS preamplifiers are no longer manufactured, and so instead I’ll use the 2N4416 JFET for Q3 and the 2N5087 for Q4 (as I did in the simulation) in my replica.  For Q5 and Q6, I’ll probably use the BC550C.  For the detector coupling capacitor, I’d like to use no more than 1 nF – I need to contact Cremat and see why they used 10 nF.  In order to reduce noise, I’ll probably lower the value of the feedback resistor to 50 MΩ and increase the value of the feedback capacitor to 2 pF, noting that tolerances at that capacitance are generally +/- 0.5 pF (i.e., 50% at 1 pF).  I’ll replace transistors Q1 and Q2 with diodes and will keep resistor R3 in the circuit.  I happen to have a couple LT1360 op amps on hand, and I’ll probably use one for the output buffer.  I’ll probably end up building the shaping amplifier of Perevertaylo et al. with the LM6171 op amp (x 2).  Those are just a few initial thoughts…