I have made an unaided test: with the 2K7, the settling time is under 10s, thus perfectly acceptable.
I also tested the effect of a leakage resistor across the 470µ: with 47K, there is practically no discernible difference, except the settling time which is marginally improved.
I measured the offset of the TLO72 operator in circuit, and it came out as 0.9mV, which is on the low side, and may partly explain the good behaviour. However, even the max of 5mV wouldn't be problematic, and having the discharge resistor in the final build could be helpful, because unlike the lab supply, the internally generated rails will rise slowly and asymetrically
I also tested the effect of a leakage resistor across the 470µ: with 47K, there is practically no discernible difference, except the settling time which is marginally improved.
I measured the offset of the TLO72 operator in circuit, and it came out as 0.9mV, which is on the low side, and may partly explain the good behaviour. However, even the max of 5mV wouldn't be problematic, and having the discharge resistor in the final build could be helpful, because unlike the lab supply, the internally generated rails will rise slowly and asymetrically
I'm more worried by the input current multiplied by 22 Mohm times the voltage divider resistor ratio plus one. There is a bias current of 65 pA typical, 200 pA maximum at 25 degrees, 7 nA maximum at 70 degrees ambient temperature. If there is 1 Tohm capacitor leakage or PCB leakage shunting C1, that's another 5 nA.
If and only if you can keep the op-amp cool and high voltage capacitor leakage and PCB leakage negligible, a few gigaohm of transresistance should be no problem. Hence, I wouldn't make the discharge resistor any lower than needed.
If and only if you can keep the op-amp cool and high voltage capacitor leakage and PCB leakage negligible, a few gigaohm of transresistance should be no problem. Hence, I wouldn't make the discharge resistor any lower than needed.
Last edited:
Do you or does anyone here happen to know what happens with the leakage current of aging bipolar electrolytics at voltages well below 1.5 V?
I know that when you take a normal electrolytic capacitor that hasn't been used for many years and apply the nominal voltage, you initially get a large leakage current that gradually drops until, after an hour or so, it is typically decades below the specified leakage. After that, the leakage stays low if the capacitor is regularly biased at a substantial voltage. When you store it for a long time with zero voltage across it, the dielectric degrades and the leakage increases again.
The electrolytic capacitor in this circuit will have no big DC bias voltage across it to keep its dielectric in shape, but then again, it only has to handle voltages that are so low that even the oxide on an unformed electrode could handle them.
I know that when you take a normal electrolytic capacitor that hasn't been used for many years and apply the nominal voltage, you initially get a large leakage current that gradually drops until, after an hour or so, it is typically decades below the specified leakage. After that, the leakage stays low if the capacitor is regularly biased at a substantial voltage. When you store it for a long time with zero voltage across it, the dielectric degrades and the leakage increases again.
The electrolytic capacitor in this circuit will have no big DC bias voltage across it to keep its dielectric in shape, but then again, it only has to handle voltages that are so low that even the oxide on an unformed electrode could handle them.
An unbiased, unformed Al Ecap creates its own insulating barrier, in the absence of any external voltage.The breakdown voltage of this Al2O3 layer is around 1V. This means that up to ~1V, the resistance will be very high, in both directions. How high depends on the electrolyte, etching, temperature and other factors. BP caps have a lower degree of etching, a thicker initial insulating layer and are symetrical, all of which contribute to a better behaviour at zero bias, but I think that a regular Ecap will be sufficient in this case: if the circuit tolerates 47K of parallel resistance, it should cope easily with the leakage of a standard cap
That sounds like very good news! As long as the DC current leaking into the negative op-amp input (including its bias current) is less than 45 nA, the voltage across the electrolytic capacitor will be less than 1 V, and if I understand you correctly, its insulation at such low voltages should remain good as the capacitor ages.
How much parallel resistance the circuit can tolerate depends on how much current leaks into the negative op-amp input. Do you use ordinary PCB mounting with guarding for C1 or something with flying wires?
All in all, does this mean you have a usable single-op-amp solution now?
How much parallel resistance the circuit can tolerate depends on how much current leaks into the negative op-amp input. Do you use ordinary PCB mounting with guarding for C1 or something with flying wires?
All in all, does this mean you have a usable single-op-amp solution now?
If it works nicely on a lousy breadboard, it will certainly be OK with a half-decent perfboard, so yes, the single opamp solution looks highly workable, and I am going to test it very soon: the core HV supply is completed and tested (and it only consumes 3.5mA on the raw 7.5V supply).
The opamp will be a TLO62 for a low current consumption, and it will be fed from the HV converter. This part has been tested too, it can deliver up to 1mA if required (this won't be necessary in principle).
One unknown remains: will the +/-18V supply of the opamp be sufficient to counteract the mix of ripple + external perturbations?
Only a real world test will tell. If more dynamic is required, I will add a high-voltage post-amp to the opamp. It will have some influence on the closed-loop behaviour, but this doesn't scare me much: the bandwidth involved is very limited, and my limited skills in control theory should suffice. In case they don't, I know that I can count on your super-powers in that field...
The opamp will be a TLO62 for a low current consumption, and it will be fed from the HV converter. This part has been tested too, it can deliver up to 1mA if required (this won't be necessary in principle).
One unknown remains: will the +/-18V supply of the opamp be sufficient to counteract the mix of ripple + external perturbations?
Only a real world test will tell. If more dynamic is required, I will add a high-voltage post-amp to the opamp. It will have some influence on the closed-loop behaviour, but this doesn't scare me much: the bandwidth involved is very limited, and my limited skills in control theory should suffice. In case they don't, I know that I can count on your super-powers in that field...
I have noticed something funny: I was making additional tests, to check the effect of protection resistors in series with the 1.5nF caps among other things, and I also tested a regular Ecap instead of the BP one.
With the circuit live, I removed the BP cap in order to replace it with the polarized one, and I was astonished to see no change in the 50Hz rejection: the 47K was still present, and kept the circuit working.
In the sim too, having just the 47K in circuit changes nothing to the 50Hz behaviour: the most important change is a larger VLF peaking.
In reality, the 50Hz rejection is ~35dB against 39dB in sim, but for the rest it looks rather reliable
With the circuit live, I removed the BP cap in order to replace it with the polarized one, and I was astonished to see no change in the 50Hz rejection: the 47K was still present, and kept the circuit working.
In the sim too, having just the 47K in circuit changes nothing to the 50Hz behaviour: the most important change is a larger VLF peaking.
In reality, the 50Hz rejection is ~35dB against 39dB in sim, but for the rest it looks rather reliable
At 50 Hz, the capacitive feedback dominates over the resistive feedback, even with the extra 47 kohm. Just calculate the transconductance of a 10 Mohm-49.7 kohm-22 Mohm T-network and compare it to the admittance of 6.8 pF or 7.5 pF at 50 Hz.
I was under the impression that frequencies far below 50 Hz also matter for your application. Is that correct?
I was under the impression that frequencies far below 50 Hz also matter for your application. Is that correct?
Here is the HV supply, including the electronic filter (bottom right):
It seems to work as expected, this is the correction waveform at the output of the opamp:
It contains ripple from the converter, some 50Hz, and large amplitude, slow variations (which explains why two successive traces aren't superposed.
It never reaches the +/-18V limit though
It seems to work as expected, this is the correction waveform at the output of the opamp:
It contains ripple from the converter, some 50Hz, and large amplitude, slow variations (which explains why two successive traces aren't superposed.
It never reaches the +/-18V limit though
I have finalized my radiation detector:
In the end, I had to add a passive shield: the supply regulation, passive filter and the active filter (courtesy of Marcel) worked well, and kept the supply variations <1mV when subjected to hand effects, but it was still too much.
They caused significant deviations when the device was moved, or when a big object came close (the whole outer envelope sits at ~-5kV).
The centre electrode is a 0.05mm silvered steel wire, and it has capacitance of ~0.6pF to the chamber (the meshed pot, courtesy of IKEA).
The TIA has a physical feedback resistor of 27Gohm, but a programmable divider can be inserted, bringing the virtual value to 27Tohm in √10 steps. A feedback cap of 8.2pF is also included, but electrostatic disturbances came through, due to the insane gain: after the TIA, a post-amplifier can add a further x100 gain, also in 10dB steps.
Between the the TIA and the post-amp, I have included a LP filter, to eliminate the chopping residues, and a twin-T passive filter to kill the 50Hz.
All of this is made possible thanks to the TIA opamp: it a chopper-stabilized ICL7650, having a µV level offset, and its -input is left hanging in the air: the PCB and IC socket have been milled, and the input goes directly to the guarded PTFE passthrough connecting to the center electrode.
The passive shield is the coil of iron wire surrounding the chamber.
It works well, but the full sensitivity is unusable: the needle remains stuck full-scale all the time after the 5th range, due to the background radiation.
I know that it is radiation, not noise, because noise has an average value =0, but each ionization event causes a positive excursion, which is what I see here, and all the circuitry is bipolar, with +/5 and +/-18V supplies (on even more sensitive ranges, noise certainly comes into play, but they are completely unusable anyway).
I packed such an insane gain because it practically free: a few resistors on the selector.
The current time-constant is 0.47s, but I will probably include a switch to allow a much longer averaging, like 20s.
Ionization events are relatively rare, but at high gain they are violent and send the needle crashing full-scale.
The ideal would be an averaging method that limits excursions without increasing the response-time too much
This the chamber, with its guarding ring:
The electric shield has another advantage: it protects the user against an accidental contact with the 5kV can (I had a brush or two when testing!)
The units are arbitrary, as I have no way to calibrate it, but the sensitivity is astounding: it detects the americium source of a smoke detector at 20cm, thanks to the mesh chamber allowing alpha particles, and a small vase made of very lightly doped uranium glass gives a strong reading on the most sensitive usable range.
The construction is an horrible kludge made of disparate materials, angular and bulky, but that is the result of an evolution/growing process rather than a deterministic design.
The total current consumption is 21mA for a battery voltage of 7.5V, which is quite reasonable
In the end, I had to add a passive shield: the supply regulation, passive filter and the active filter (courtesy of Marcel) worked well, and kept the supply variations <1mV when subjected to hand effects, but it was still too much.
They caused significant deviations when the device was moved, or when a big object came close (the whole outer envelope sits at ~-5kV).
The centre electrode is a 0.05mm silvered steel wire, and it has capacitance of ~0.6pF to the chamber (the meshed pot, courtesy of IKEA).
The TIA has a physical feedback resistor of 27Gohm, but a programmable divider can be inserted, bringing the virtual value to 27Tohm in √10 steps. A feedback cap of 8.2pF is also included, but electrostatic disturbances came through, due to the insane gain: after the TIA, a post-amplifier can add a further x100 gain, also in 10dB steps.
Between the the TIA and the post-amp, I have included a LP filter, to eliminate the chopping residues, and a twin-T passive filter to kill the 50Hz.
All of this is made possible thanks to the TIA opamp: it a chopper-stabilized ICL7650, having a µV level offset, and its -input is left hanging in the air: the PCB and IC socket have been milled, and the input goes directly to the guarded PTFE passthrough connecting to the center electrode.
The passive shield is the coil of iron wire surrounding the chamber.
It works well, but the full sensitivity is unusable: the needle remains stuck full-scale all the time after the 5th range, due to the background radiation.
I know that it is radiation, not noise, because noise has an average value =0, but each ionization event causes a positive excursion, which is what I see here, and all the circuitry is bipolar, with +/5 and +/-18V supplies (on even more sensitive ranges, noise certainly comes into play, but they are completely unusable anyway).
I packed such an insane gain because it practically free: a few resistors on the selector.
The current time-constant is 0.47s, but I will probably include a switch to allow a much longer averaging, like 20s.
Ionization events are relatively rare, but at high gain they are violent and send the needle crashing full-scale.
The ideal would be an averaging method that limits excursions without increasing the response-time too much
This the chamber, with its guarding ring:
The electric shield has another advantage: it protects the user against an accidental contact with the 5kV can (I had a brush or two when testing!)
The units are arbitrary, as I have no way to calibrate it, but the sensitivity is astounding: it detects the americium source of a smoke detector at 20cm, thanks to the mesh chamber allowing alpha particles, and a small vase made of very lightly doped uranium glass gives a strong reading on the most sensitive usable range.
The construction is an horrible kludge made of disparate materials, angular and bulky, but that is the result of an evolution/growing process rather than a deterministic design.
The total current consumption is 21mA for a battery voltage of 7.5V, which is quite reasonable
Yes, that's one of the options I considered when I started the project, but in the end I opted for a linear output.
After the first real-life impressions, I feel that it would be very helpful, combined with a clever filtering/averaging scheme
Nothing really specific: I can check various objects and items in my house for activity, like the small uranium-glass vase, I can check for how long I remain a hazard after undergoing a scintigraphy or a PETscan, and in case Kim plays with one of his favourite toys, I'll be informed too
I have no intention of publishing the whole project, as it presents little interest to audio DIYers, but the HV section might be interesting: it is extra-simple, very clean, well regulated, uses only commodity parts and can be powered from an ordinary 9V battery whilst consuming 10mA @7.5V. It works from 6 to 10V and the consumption is inversely proportional to the supply voltage. Most of the power is used in the 1G feedback resistor.
It could be adapted for ESL's or special condenser mics for example
It could be adapted for ESL's or special condenser mics for example
I have described the supply in this thread: https://www.diyaudio.com/community/threads/a-simple-and-effective-hv-supply.397598/post-7309529