Thanks for the update. On the schematic it is hard to discern, simply because a different GND symbol was used. So I wasn't sure. Better to ask than not know.
-
Welcome back Guest! Did you know you can mentor other members here at H-M? If not, please check out our Relaunch of Hobby Machinist Mentoring Program!
Needing more than a spark test?
- Thread starter graham-xrf
- Start date
I was thinking the same thing
. Luck of the draw I guess....I found a demo board for the ADP1031 that can be used as an isolated power supply, and the whole thing costs about as much as that transformer Graham mentioned. $13.85 from Digikey. The transformer costs about $4.78 from Mouser, coilcraft/yaego YA9293-1548212. Inductance is 300uH. Going that route we've got a switcher back in the picture but it's in a lower-gain portion of the system.
On the other hand, a 5VAC output wall-wart will cost you $6.95 from Jameco. No switching noise, decent power line isolation. Simple diode bridge/capacitor/regulator. I always hang on to wall warts like that, even if the equipment it ran is long-gone.
Edit: I corrected the P/N for the Analog Devices part. And the mouser P/N for the transformer is YA9293-ALD.
Last edited:
That part of the schematic is sloppy on my part, sorry about that. I'll try to do it better on the next iteration.
Shorted out C11 to ground after isolating the AD8655. And I get 16K samples of zeros. Average count is 0.00000, RMS variation = 0.00000. It is the AD8655. Nothing wrong with the ADC. Running at 500KHz and 1MHz - mean count is 0.00000.
Schematics for private consumption can definitely be a bit looser. One's that go out the door get scrutinized, because that is all the end user has to go by. All this stuff is a learning process.
Been learning some of the same lessons myself as someone else is looking over my ELS code, and trying to adapt it to a different lathe.
Schematics for private consumption can definitely be a bit looser. One's that go out the door get scrutinized, because that is all the end user has to go by. All this stuff is a learning process.
Been learning some of the same lessons myself as someone else is looking over my ELS code, and trying to adapt it to a different lathe.
Yes indeed. This one happens to be a 1:1.3 ratio, so that you can start with (say) 3.3V and end up with 3.3V. You can use 1:1, so long as one is centre-tapped. The Mouser part was HALO brand TGMR-320V6LF.
The very lowest noise converters are the Ćuk topology in it's fully isolated form, which I thought way too expensive, and elaborate, needing two inductors, plus transformer, and external FET switches.
The MAX253 is a cheap, old-school (1994) part which, if operated at high enough frequency, and followed with the LDO, plus (possibly) some low pass filtering, one can get the PSU supply noise down to microvolts. In our case, it is not the PSRR ratio for opamps that matters, so much as what a decoupling capacitor can do to our precious 0V.
We cannot treat 0V GND as if it is a solid "connected to all the whole planet Earth" black hole sink thing into which we can cheerfully dump currents. Noise on a power rail, if connected to GND by effective decoupling capacitors, becomes noise on the GND, and so also, is then noise on various opamp inputs. In our case, inputs with truly spectacular gain. Our AGND is never going to be "stiff". We rely on the common mode cancellation for both inputs of an opamp. If the response is good enough to cancel "noise", at least up to the bandwidth we need, we win.
I have very carefully gone over all the options about ways to give an opamp linear range in this application.
"Rail-to rail" opamps cannot be truly "rail to rail". The best of them get to within 20mV of the rail, like my LT1807, but more commonly the limit is are nearer the 200mV off the rail, typical of the LT6268-10.
There are several strategies that can be used to fix this. One is to commit to having both positive and negative rails. In out case, the whole thing can be done within the positive rail, except for the very first amplifier. The TIA is fed into its inverting input. The output is tiny, but that is not much help if it is not at least enough away from the rail.
The very low noise, low power, dual rail generator.
To have isolation, it still needs a switched high frequency supply (like MAX253) or a wall wart upstream of it, but the ADM27762 is a bit of a gem. It supplies both, and it has the whole charge pump negative supply, and LDO regulators built-in, and it runs at 2MHz.
For convenience, I have attached the datasheet here.
Another approach
This is to use a deliberate, controlled offset, and choose non-inverting amplifiers, such that the signal is always between 0V and the positive rail.
In my case, I have the differential input, AD7622, so I can use a resistor divider from Vref, and offer it at the buffer to (IN-), and also at (IN+). The ADC still only sees the difference, but everything upstream is shifted upwards to the full accuracy, sufficiently away from the 0V (rail) if using single power supply.
There comes the problem of the negative going pulse at the TIA. This too, we can turn around. I put the capacitor in the input to remove the offset from the dark current through the 40MΩ reverse biased PIN diode. 4nA in 40MΩ would be a whole 160mV, but that is not what happens in a TIA. The actual output would be Idark x Rf = about 1.6mV. That is still many times greater than the microvolts of pulse, but can, in theory, be offset.
The "shift the Rails with an opamp" trick
This is one I have mentioned before. If the rails are a bit inconvenient, then shift them.
It does, of course, require a supply that is floating in the first place, or batteries.
We can "turn around" the PIN diode.
I know I am using an AC coupled TIA input, and I think I only get away with it because the charge current is so tiny. The AC coupled pulse seems unable to end up around the 0V either side of it's average value. This idea only works if one deliberately allows the dark current, to direct DC couple, and at the same time, use an integrator offset remover. Alternatively, use AC coupling, and add a deliberate Vref derived offset. As I have it at this time, I supply a little negative charge pumped supply to the first opamp only.
See the TIA first stage problem
Here is a fully simulated (simple type) TIA, using a 400pA pulse. The output at V(n003) TIA first stage output is -90uV. Even if we turned the PIN diode around, it's all too close to the 0V if that was also a single supply rail. Note that this circuit is delivering 200mV output at Vout3, with -35pV offset. There is a spare opamp in one of the packages if that is not enough, but we did start with 400pA.
Note that to play with this one, set Tools -> Control panel -> Absolute Current tolerance =1e-012, and Absolute Voltage tolerance 1e-007. The latter may be OTT. Likely 1e-006 will do.
My circuit.
Given some of these final things are still not proven in real assembly for me, I am still somewhat keeping the options open, but I think choosing a deliberately shifted negative rail too little to hurt the ADC if something goes wrong, could be the way. There are Schottky protection diodes in the ADC, which, along with the resistor in the driver output, can protect it anyway,
I know it's a distraction, but developing a nice ultra low noise, hopefully cheap, low power isolated power supply set, possibly with more than one output, is what I am after right now.
Last edited:
So you were looking at the AD8655 natural input offset voltage?
That's good! It means you have the ADC looking at a short circuit, and seeing what you expect.
Is the AD8655 a gain=1 driver?
Regardless, if we adapted the circuit in the previous post, it might work for you.
Thinking about this, the first stage is the one that really needs the tiny voltage offset, maybe 200mV. The signal is so small at the first stage it is nowhere near ground. The second stage can be inverting to flip the pulse up. Finally the third stage can be non-inverting. Whether one wants to use the integrator to hold adjust the bias or just just AC couple at the end,is an implementation choice.
More or less. Actually I was looking at the lowest possible output of the AD8655, on a zero volt rail. It depends on the sampling rate I use, but it is 0.52 mV for 500 KHz sampling and under (roughly) and about 0.62mV for 1 MHz sampling. 400 KHz is 0.52 mV, 300 KHz is 0.52 mV 200 KHz is 0.52 mV. 100 KHz the offset goes up. I am using a 64K buffer to run means and rms, to reduce my variance. A 1K buffer was not long enough for the measurements.So you were looking at the AD8655 natural input offset voltage?
That's good! It means you have the ADC looking at a short circuit, and seeing what you expect.
That is exactly the recipe in the circuit in post #1626.
Give me a few minutes, and I will try for one with the AD8655 driver in there.