There may be some variation in PSRR between op amps, but they all drop off as the frequency goes up (from a few hundred Hz up). I would think it's factors like the internal current sources not keeping up and internal capacitive coupling becoming a problem, and things like that, but I don't know for sure. At least the one you've chosen is balanced for +PSRR and -PSRR. A lot of older bipolar parts were quite unbalanced, with one rail much better than the other.

This is what my simulation shows if I step the LED resistor 100R, 200R, 300R. No effect up to 100kHz, but the 300 is slightly better from there up.

Hi Jon,

I believe it's related to supply noise for that LED and it's op-amp PSRR, but was also wondering if the op amp was able to control well at that combination of drive current and LED output and photodiode response, but varying that resistor makes no noticeable difference to the output. I could try a OPA2192 chip instead of OPA2197 for a slight improvement, but I don't know if that will help enough (the charts look identical in the datasheet , but in the electrical characteristics it shows there could be a 20dB improvement), since the charts look identical I will probably leave that for now.

I am going to try a couple of 9V batteries (I have a low-noise +-5V regulator board to use with that), to see what happens.

The opto LED doesn't really have much PSRR, does it? If anything, it might be a slight gain, as the curve is non-linear and a small change in the supply voltage will give a higher proportional change in the current (which defines the light output). The one thing that might affect things is the intrinsic capacitance of the LED junction, but I don't think that will have a filtering effect until up into the MHz area.

This is the PSRR simulated. This is purely indicative - there's no hope it accurately models what you've got because there's nothing there to model the effect of the supply distribution and decoupling.

The green curve is LED and op amp running off noisy supply.
The red curve is op amp noisy and LED perfectly clean.

I've only injected a signal into the +5V, so it's also only looking at part of what's going on. Looks like a factor of a little over 20, or so, between the two curves. The cursors are on 72kHz.

So, it suggests there is at least some benefit to cleaning up the LED supply. [As a quick hack, as long as you keep the LED current to less than 8mA, you could try running the LED from the 4V reference output, just to see how much difference a cleaner supply makes, though 3x AA batteries would be even quieter.]

The PSRR of the op amps is only mid 40s at that frequency, so local supply filtering close in to the op amp would probably be of benefit too. Even just a few ohms in front of the decoupling cap would probably help.

Personally, I'd concentrate on a cleaner supply, though. Could the front end be made to work on a single PP3 if the fake earth was biased to give a bit more headroom on the positive side [perhaps +5V and -4V]?

For anyone that's interested, this is the circuit in the simulator. The free TI Tina is sometimes a bit clunky with the parameter stepping and it doesn't seem to control switch settings, so instead I used a VCVS to mirror the 'noise' signal to the LED where I could use a gain of x0 or x1 to select either there or not there. Set to 'parameter stepping' mode, that then allowed me to have both curves on the same Bode plot.

Hi Jon,

Great points! Compensation is definitely something that I'll need to consider.  I'm hoping the parasitic capacitance won't vary too much from board-to-board, to try to make it somewhat repeatable, but it may still need trimming due to capacitor tolerance. I H-field probed the DC-DC converter, and the field extends over a lot of the board, but I still wasn't sure if it was radiated or conducted issues, so I unsoldered the DC-DC converter and put it on extension wires, and that made absolutely no difference. Next I tried a fairly low-noise DIY supply (I couldn't use batteries for now because the 4.096V reference requires 5.5V max) and the output noise changed characteristics, so it is definitely influenced by the supplies. I connected the supply with as much ferrite as I could fit on the cable.

The screenshot here shows the output noise on the SMA connector J1, with the DC-DC converter replaced by an external supply (this is with R6 being 10k, I need to replace it back to 100k).

I checked the noise of the external supply out-of-circuit (I measured this for now without a load, so it will be different in-circuit). This is the positive rail, about 277uV RMS noise according to the automated measurement at the bottom of the screen:

The negative rail is very similar:

I'm now speculating maybe the current through the LED needs to be altered, maybe there is a region where it could be less susceptible to power rail noise : ( But this will affect the dynamic range. Or, it needs a further order of magnitude lower-noise supply rails, perhaps by replacing the DC-DC converter to a +-9V output one, followed by more than one regulator! but this is getting to be excessive, it seems unusual that it should be so sensitive to the supply rail. The OPA2197 has reasonable PSRR too.

Ignore the part about the open-loop response. I had the circuit slightly wrong AND misremembered the result into the bargain.