AGFA PMT module AGFA-IUP3, reverse engineering

[Mads Barnkob]

The original article on my website, has more information on similar modules and more pictures: http://kaizerpowerelectronics.dk/high-voltage/photomultiplier-tube/

Schematics are not yet done.

I got this module from a AGFA ADC Compact Plus computed radiography developer machine.

The photomultiplier assembly sits in a aluminium housing with the tube going out through a plastic cover, the inside of the plastic cover has a grounded layer of mu-metal for shielding the tube from electrostatic interference.

The 15-pin SUBD connector is the only connection to the module and is both for supply voltages, input and output signals.





Opening the lid of the aluminium enclosure, the entire assembly can be taken out with 4 screws undone. The photomultiplier tube is soldered directly to the circuit board with a supporting aluminium base ring to sit tight in, along with another grounded layer of mu-metal around the rest of the base.

It is not possible to see any kind of markings or stickers on the tube to identify its exact type. From the number of pins only, I guess this tube only has 8 dynodes and in the computed radiography application this gives plenty of sensitivity. The thick blue glass front window makes is impossible to make out how the tube is constructed on the inside.

The tube is soldered to the circuit board on which the 15-pin connector connects to, upwards sits another circuit board on which the high voltage power supply is. Underneath the tube there is a LED coupled with a light receiving diode. This LED is used for testing the tube.

The different markings on the boards are as follows: AGFA-IUP3 on the main circuit board along with serial number SN 02583 and photomultiplier tube number PMT 111296. High voltage power supply circuit board is marked with AGFA-HV1.







On the following picture of the bottom side of the main circuit board I have marked pin header for the 15-pin connector and the pins connecting the high voltage board to the main circuit board. These markings are essential in relation to the attached schematic.

The arrangement of pin 1 and 9 quickly shows that it is a meant to be supplied from a positive and negative power supply, from the polarization of diodes and electrolytic capacitors. The +/- 15 VDC power supply is the only power source connected and the +5 VDC house keeping power supply is generated with the LP2951 IC.

Pin 2, 6, 10 and 12 is connected directly to ground plane and pin 13 is connected to the ground plane through a 1K resistor.

Pin 3, 4 and 11 goes straight to the high voltage power supply circuit board. Pin 3 through a 1K resistor. Pin 3 is the high voltage enable input.

Pin 5 is the processed data output from the AD823 op-amp, a signal between ground to +15 VDC is outputted on this pin in regard to light detected by the tube. There is also a pin header near the tube connections where the raw signal from the tube can be taken out.

Pin 7 and 14 are not connected to anything.

Pin 8 and 15 connects to a DG202 IC which is a quad SPST CMOS analogue switch. Two switches each have their inputs connected in parallel, but only one of the outputs are utilized. Pin 15 activates the test LED with its auxiliary circuitry of the coupled light receiving diode on the second op-amp of the AD823 IC. Pin 8 imposes a +400 mV reverse bias on the negative input of the first op-amp of the AD823 IC and is used to reset the tube output data faster than the fall time of the tube itself so it is ready to read the next pixel faster.

The upside of the main circuit board only has pin header connector for the 15-pin SUBD connector, 4 electrolytic capacitors and 2 inductors.

From the pin HV9 comes the negative high voltage supply for the photomultiplier tube. The cathode of the tube is connected to a large orange painted capacitor that is tied to ground, to ensure a stable voltage at the start of the high voltage chain.

As more electrons are accelerated from the first dynode to the next and so on, less current is needed to drive them on, this can normally be done with a pure resistive divider network between the dynodes, but it has its drawbacks with loss of linearity and output deviation in the region of 10-20% at high output current. A improved divider network uses capacitors to insure stable supply voltage at peaks and even individual power supplies for each dynode can be used to do the same.

The solution on this circuit board is however much more elegant and complicated. A network of resistors are used between the first 4 dynodes and there after capacitors, but load dependent driven transistors are used to linearise the output depending on load and that brings the output deviation to around 1-2%.



The high voltage power supply circuit board has no components on the backside, only a few traces that is marked on the following picture.

From the photomultiplier tube handbooks from Hamamatsu, the requirements to a PMT high voltage power supply reveals that quality control is needed to achieve a line/load regulation at +/- 0.2% and ripple noise/temperature drift at 0.05%.

Pin 3 and 11 from the main circuit board connects to pins HV1 and HV2 and are somehow related as they both goes to the positive and negative input of the TL032A op-amp and the output from this goes to a LM393 op-amp positive input and with negative tied to ground through a 50K resistor. The output of the LM393 drives a transistor that ties the ZVS Royer oscillator to ground and thus enables it to oscillate. The ZVS oscillator uses two IRFL110 MOSFETs, 100 VDC/12 A rating, driving the small transformer through a isolation toroid and the output of the transformer is fed through a Cockcroft -Walton multiplier to generate around -600 VDC.

When the circuit is applied with power it is however only pin 3 that has to be pulled high with +5 VDC to enable the high voltage generator and get the photomultiplier tube circuit running.

From the high voltage output end of the Cockcroft-Walton multiplier there is two 100M resistors, one leads back to the -15 VDC supply rail and the other to the second part of the TL032A op-amp, possible for high voltage measurement feedback.



5th March 2016

After having drawn a fairly accurate schematic from reverse engineering the circuit boards, I was confident to test the module with power on. I was using a power supply with +15 VDC, -15 VDC and GND. On the oscilloscope I have a differential probe on the high voltage supply output on channel 1 (yellow), raw signal output from the photomultiplier on channel 2 (blue) and processed signal output from pin 5 on the main circuit board on channel 3 (purple).

I shielded the window of the PMT with aluminium foil to block out any light



Judging from the house keeping power supply IC that generated +5 VDC and the datasheet for the DG202 analogue CMOS switch IC, I tried my luck with putting +5 VDC on the input pins I had located. This is how I found the high voltage enable on pin 3.

When +5 VDC is applied to pin 3 the high voltage supply is generating -600 VDC.

Now that the module was up and running it was no problem doing a few tests, as the test LED was controlled by the CMOS switch, it was merely to pulse + 5 VDC on pin 15 to activate the test LED and capture a shot of the output pulse. The result was a raw signal of -20 mV and processed signal of +4.2 VDC on pin 5.



To test the sensitivity and assumption of maximum processed output of being positive supply voltage for the op-amp, +15 VDC, I used a LED flash light to blink at the tube through a pin hole in the aluminium foil covering the window. The result was a raw signal of -770 mV and a processed signal of +12.6 VDC on pin 5.



As a photomultiplier tube can be a very sensitive instrument, there must be a explanation on why a blink of a flash light is not completely drowning the tube and giving a maximum output. From the PMT handbooks we can see on graphs for common PMT behavior that the amplification is highly dependent on the high voltage supply.

The -600 VDC this module uses gives a mere amplification of 40.000 times, marked with a red line on the graph. This level of amplification is sufficient for the computed radiography, but for a more interesting task like radiation detection with a scintillation crystal, it is suggested that 1-2 kV is needed and as it can be seen in the marked blue area that gives a amplification of 20.000.000 times and up towards 1.000.000.000 times.



Future experiments involves:
- Scintillation radiation detection with current power supply
- Adjust high voltage power supply for higher voltage to increase amplification

[futurist]

How large is the PMT window?

It would be nice to try NaI:Tl scintillation crystal, the blue window shouldn't be a problem because NaI:Tl emission maximum is at 415 nm
Downside is that they don't come cheap

[Mads Barnkob]

How large is the PMT window?

It would be nice to try NaI:Tl scintillation crystal, the blue window shouldn't be a problem because NaI:Tl emission maximum is at 415 nm
Downside is that they don't come cheap

It is a 3" / 7.5 cm window.

I did search ebay and other places for a 3" NaI:Tl crystal, but those in the buy-able region was all yellow'ish, damaged or otherwise in a state where it was nearing its end of life, those clear and looking new were out of my budget for a experiment

So I settled for a 2" round piece of BC408 scintillation plastic, which has an almost identical wave length output, 425 nm, compared to NaI:Ti's 415 nm. I later borrowed a large piece of unknown scintillation plastic from a friend that works in radiology.

From my experiment notes...

9th June 2016

To conduct this experiment I knew that I needed a crystal or plastic scintillator of a fair size to make up for the low tube voltage with more detector volume. Plastic are the cheapest and for a reason, in the detection of gamma rays, organic plastic scintillators have a low absorption coefficient and exhibit less probability for the photoelectric effect, making them unsuitable for energy analysis applications. BC408 is however very good for fast neutron detection.

I bought a BC408 cylinder piece that measures 50 mm in diameter and is 30 mm thick, this corresponds to a volume of 59 cm3.

I borrowed a unknown square piece from a friend with the measures 32 x 32 mm and a length of 100 mm, this corresponds to a volume of 102 cm3.

The test setup was fairly simple, power supply, oscilloscope and photomultiplier tube with a scintillator on it and covered in 2-3 layers of aluminium foil for blocking out light.

I captured the waveforms in single shot mode and just kept raising the trigger limit to see how high output levels I would get. It was however necessary to use a scan range of about 200 uS to catch the signals and then zoom in for the following screenshots.



In the following screenshots from the oscilloscope is the test done with the square plastic scintillator and  I was able to catch up to 4.16 VDC pulse out of the processed output pin.



In the following screenshots from the oscilloscope is the test done with the round plastic scintillator and  I was able to catch up to 5.6 VDC pulse out of the processed output pin.



The noise floor is well below 200 mV with this preliminary test I do know for now that I have a sensitivity between 200 mV to 5600 mV. To be rough and assume this is the maximum values I had found already, have a step of 100 mV that actually gives me a resolution of 54 steps. This is a much better result than what I expected from the standard setup of the tube, future experiments with a proper data collection setup will reveal much more information about its sensitivity.

[bktemp]

I recently bought the same PMT module.
While reverse engineering the circuit I found your page with all the information.
That confirmed the pinout I had figured out and helped me understanding what the analogue switch does. Maybe the blue LED is used for calibrating the gain of the PMT because it injects a known amount of light?

After reading your page, I stumbled across the description of the HV power supply using a royer ZVS converter.
That made me draw its schematic to find out how the circuit exactly works, because I was previously unsuccessful using a royer conveter in a regulated power supply. My circuits didn't liked to be controlled over a wide output voltage range.
After drawing the schematic I figured out, it isn't a royer converter, but a simple flyback converter with some interesting additions:
One mosfets acts as a linear regulator for stabilizing the 30V (+/-15V) input voltage and maybe isolation the switching noise from the remaining circuit. The second mosfet is the actual switch, switching the transformer and the series connected toroidal core transformer.
The function of the toroidal core seems to be for reducing the turn on current spike: Because the transformer drives a capacitive voltage multiplier, during turn on of the mosfet a large current flows when charging the capacitors. The toroidal core transformer acts a series inductor, limiting this current. It's secondary winding pumps the stored energy back into the supply rail. So basically it is a lossless leakage inductance.
The switching mosfet is driven by the LM393 comparator. One comparator acts as an oscillator, the second one as the PWM comparator.
The opamp is used for regulation, the second one drives pin 4 of the 15pin connector for voltage readback (4.7mV/V PMT voltage).
Pins 3 and 11 go the the feedback loop opamp and control the PMT voltage. They are actually not enable pins but analogue control inputs.
Pin 3 is the positive input, generating approximately 120V PMT voltage per 1V input voltage. The same applies to pin 11 going negative.

When adjusting the voltage up to 1kV, the PMT gets really sensitive, easily detecting a dim white LED across the room.
I can see about one pulse >0.1V at the output every 2s with the input window fully covered.
With a small piece of Am241 in front of the tube, the pulse rate increased to about 50 pulses/s. So this PMT seems to be sensitive to radiation, even without a scintillator!

I wonder if it is possible to remove the blue filter in front of the tube. It seems to block everything except a rather small range of blue light. Using a green laser, the PMT didn't show
any signal at all. I could aim it directly at the PMT without the slightest signal at the output.
The filter seems to be a seperate glass disc glued in front of the actual PMT using some optical adhesive.

[Mads Barnkob]

Hi bktemp and welcome to forum

I am glad that it helped you on the way and it is good to hear that you went further with the reverse engineering than I did, I just took a look of the power supply schematic I had saved, there is only 6 components and 15 lines in it, I never finished it. Would you mind sharing the schematic here?

I agree on the blue LED, its properly both for calibration and self-check, I am pretty sure on the self-check since it sits with a photodiode to register if the LED actually works.

I would love to update my website and forum post here with your corrected information, again, if I could see the completed schematic first

I also dug out some email correspondence I had with another guy that had such a module and back then he also described how the HV enable inputs was in fact analogue inputs for controlled the HV voltage. I completely forgot about those when I wrote up the article again. He also mentions some of the same things as you do now... Which actually makes me think that they are from you? Are you Malte?

[bktemp]

I've attached the preliminary version of the schematics I have drawn. It is based only on photos I have taken from the pcbs and the ones from your page, because the marked traces from the backside really helped seeing all connections without having to probe around on each via.
I haven't measured any component, so some values could be wrong or have a wrong connection, but the overall circuit looks plausible to me. I also had to guess for diodes being a zener diode or not.
When I have time, I plan to complete, verify and clean up the schematic (but this probably won't happen anytime soon, since the schematic is already accurate enough for what I was interested in).
The schematics are missing some less important parts like the 2951 voltage regulators used as reference voltages for 5V and 10V, the supply rail filtering or decoupling capacitors.
It is interessting that there seems to be no decoupling capacitor near the highspeed opamp. Actually, there seems to be absolutely no decoupling capacitor at all on the pcb, except the ones at the input filter.

My PMT module is a slightly newer version (made in 2009), but it seems to be almost identical to yours except some minor changes in component values or packaging (like using SMT HV resistors).


The PMT itself is a XP3314/FL1. I couldn't find a datasheet for this specific variant, but the one for the XP3314 is available here:
http://www.qsl.net/k0ff/016%20Manuals/PMT/Photonis/XP3314.PDF
I also found an image of the tube without the blue filter:
https://duncanadamsfusion.wordpress.com/2011/09/24/14/


The current to voltage amplifier looks well engineered:
The FET input opamp AD823A is used because of the high input impedance and gain for providing DC precision while the LM6171 with its 100MHz GBW provides the high AC gain for fast transient response.
So this circuit seems to be accurate at both DC and AC up to maybe 5MHz (I simulated the circuit with guessed and tuned capacitor values for no overshoot and got around 3MHz -3dB bandwidth).
The gain is 1V/6.5uA, so the fullscale range is somewhere around 80uA (~110pA cathode current at 1200V PMT voltage).

My guess is, they first use the 400mV offset for DC offset measurement, then the LED gets turned on, providing a known amount of light because it is inside a closed loop, using the 5V from a 2951 as the reference voltage.
With the known light source, the PMT voltage gets adjusted until the output voltage reaches a certain value for calibrating the gain.
For scanning x-ray images they probably want a very consistant behaviour of the PMT image sensor over temperature or its lifespan. They could easily do the calibration every time before scanning an image.
It looks like the resistor values for the LED light sensor feedback are different on every pcb, they are probably manually soldered during production. Around the MELF resistor on your pcb there is some flux residue.
I haven't figured out why there are transistors at both ends of the PMT dynode voltage divider chain. The zener diode at the source/emitter lead is unusual. Maybe the voltage divider at the gate/base amplifies the zener voltage for generating a better high voltage zener diode than using an actual zener diode?


The HV power is a typical flyback converter with the addition of the toroidal inductor and the unusual differential control input.
Q6 at the top left acts as a current limiter for the output current by sensing the dc signal at the bottom end of the voltage multipler. Since I did't measured any component values I can only guess, but the current limit is probably somewhere around a couple of 100uA.
It looks like they really tried to make the output voltage as clean as possible by filtering both the output of the converter and also the input. Seeing a linear voltage regulator in front of a SMPS is quite unusual, but maybe necessary for rejecting low frequency noise, because they don't seem to use current sensing at the switching transistor, which is odd, because there is a 2.2ohms resistor.

Having all the circuit inside the module with a robust metal case makes it a nice device for playing around. I'm going to add a small controller box providing the power supply and an adjustment knob for the PMT voltage and also a protection circuit for automatically reducing the PMT voltage when the output signal is saturated for a prolonged time.
Although it reduces the detectable light sources substantially, having the narrow optical filter is actually nice, because it allows to operate the PMT while having a visible (red) light turned on.

And no, I am not Malte. These PMT modules got sold frequently over the last couple of years on a surplus store in Germany, so there are probably many people playing around with this or the other variants of the AGFA PMT modules you have also listed at the bottom of your page. This round PMT seems to be more seldom than the the square Hamamatsu tubes. I this variant looks much nicer to me, because it can be disassembled without unsoldering the tube (the Hamamatsu PMTs are glued to the metal case).

[Mads Barnkob]

That is a great reverse engineering job you did there, you went much further than I did, when I first understood the basics and could get it to run, I kind of stopped working on the documentation. But with your work I can easily finish my schematics for my version of the board and get it online

If you allow me, I would like to add your pictures/schematics to my website, so this version is also there along with the others.

The transistors in the PMT high voltage divider setup is there to ensure linearity in a high count rate application like a image scanner is, they are using the transistors to supply additional current at the later stages that could normally suffer from lack of current at the last stages. See attached image and handbook about it.



Could it be that the 2R2 resistor at the transistor is the actual current limiting and therefor a over current circuit is not needed?

I also planned to build a project box out of it, if I could fit it inside a small as possible aluminium box with a door in the side to exchange scintillation material, and then make a larger steel box that it fits inside of, with a larger door in the side and so on, to have some different particle filters all around the sensor.

I am not sure if I wrote it on my website, but I planned to use a STM32F7 discovery development board to make a multi channel analyzer for measuring mu decay events and present them on the attached touchscreen of the board. That idea have however been on the shelve for a good time, never got the time to start up on that.

[bktemp]

The transistors in the PMT divider I was talking about are the one between cathode and dynode 1 and the one between anode and dynode 8. Unlike the transistors buffering the voltage divider, those two have a zener diode at their emitter/source pin.
I measured all zener diodes to get a better idea of the voltages and currents in the circuit. Both zener diodes near the transistors measure 6.2V.
I also measured the voltages at both stages and they match the calculated voltages based on the voltage divider around the transistors and the zener voltage. It looks like both transistors circuits act as simple zener diodes using the low voltage zener diode as the reference.
Both 160V and 22V are common zener diode voltages, so I have no idea why they used a transistor here instead of the appropriate zener diodes. Maybe they wanted to be able to adjust the voltage? Or maybe they selected 6.2V because at 6.2V zener diodes have the lowest temperature drift? We can only guess why they did it that way, but it seems the voltages at the first and large stage needed to stay fairly constant regardless of voltages on all other stages.
The current flowing through the voltage divider is around 440uA (at 1200V). This is much higher than the measurement range of the anode current amplifier, so it should give a good linearity even for high DC levels.
They probably needed this because when scanning images the circuit must be able to measure large bright sections without having time dependend linearity problems.


Having a flyback converter with no current sensing is often considered a bad design, because it risks blowing the transistor when there is an overload condition at the output.
But maybe it is not possible to use the current ramp due to the additional current draw of the voltage multiplier circuit (in a typical flyback converter, the current during on phase only flows into the inductance of the primary winding, therefore the ramp is linear. But here it has additional peaks making the current waveform useless for peak current control)
The active current limiting circuit starts to limit at around 700uA. That makes sense for a 440uA current draw of the PMT voltage divider at 1200V.
I can not see any connection to the 2.2ohms resistor except between GND (actually -15V) and source of the mosfet. The only way it can limit the current is when the voltage drop gets so high enough the gate voltage seen by the mosfet gets reduced and it goes into linear mode.
That happens at around 5V voltage drop, so the peak current limit is set to about 2.5A. Considering an avarage current draw of the circuit of <50mA, this is a rather high value. The resistor is probably only used for avoiding a self destruction of the mosfet in case the transformer goes into saturation. At that current level, the resistor dissipates >10W.


Using a 0-10V control input voltage, the PMT voltage can be adjusted to almost 1200V. The useful operating range is maybe 500-1200V. Although the datasheet of the PMT says minimum 800V, I have set the lowest voltage to 500V (4.2V at the control input).
At 500V it works fine, but going below 400V the PMT voltage divider current gets too low for the PMT to reach the full range anode current.

How did you plan to get the pulses into the microcontroller?
When researching how to build a multi channel analyzer as cheap as possible wihthout affecting its performance too much, I found two possible solutions:
The first one was using a high speed ADC connected to an FPGA. Based on the pulse width, the ADC needs to have at least 20MS/s to make sure it samples close enough to the peak value of each pulse. The biggest advantage of this solution is the zero dead time, because the ADC is always sampling and the peak detection is done in the digital domain. So after seeing a pulse, it can rearm immediately when the falling slope of a pulse is over.
Today fast ADCs are cheap, but a couple of years back, high speed ADCs mit >10bits were more expensive so I decided for an analogue sample&hold stage, followed by a medium speed ADC.
The S&H stage is basically a peak detector with a reset switch. There is also a highspeed comparator for triggering the ADC when a pulse is detected going above a threshold level.
Then the ADC samples the voltage level and the peak detector gets reset. A small microcontroller collects all the samples. The circuit has a dead time of a couple of microseconds, but otherwise works quite well after a lot of tweaking of the S&H stage.

Feel free to use my schematics/picture on your page. It is always nice to see somebody collecting all the information for such interesting devices.
I knew the PMT came from some medical device, but I had no idea that it was an x-ray image scanner.

[Mads Barnkob]

Could it be that the transistors are there to handle extreme load /overvoltage conditions? Depending on the base voltage distribution through the resistive divider it can let off some steam through the zeners?

Maybe the life span of the tube is shorter than of the circuit, so there was only implemented explosion/fire preventing mechanisms instead of a full fledged current control.

I wanted to use the 12-bit ADC on the STM32F7 discovery board, I did not do any math yet if its even fast enough, which it properly isn't, when looking at this paper: http://www.st.com/content/ccc/resource/training/technical/product_training/group0/0b/e4/af/01/4a/92/44/dc/STM32F7_Analog_ADC/files/STM32F7_Analog_ADC.pdf/_jcr_content/translations/en.STM32F7_Analog_ADC.pdf

Using all 3 ADCs on the board, interleaved and sequenced, it can get up to 7.2MS/s

[RickyTerzis]

Hi....i am a new user here. The FET input opamp AD823A is used because of the high input impedance and gain for providing DC precision while the LM6171 with its 100MHz GBW provides the high AC gain for fast transient response.
So this circuit seems to be accurate at both DC and AC up to maybe 5MHz.The gain is 1V/6.5uA, so the fullscale range is somewhere around 80uA .

[Mads Barnkob]

I never got around to update and complete the schematics, but I figured that the work I did complete, still poses some value to other people.