I am studying the schematics of an analog isolator with variable gain (SIM 984 from SRS0, the isolator is bipolar and can handle positive and negative voltages.

In the isolation part of the schematics (see picture below), the author uses two voltage references circled in red.

Having read the manual of the analog optocoupler used here (HCNR200, see p.19 of the HCNR manual (https://docs.broadcom.com/doc/HCNR200-HCNR201-High-Linearity-Analog-Optocouplers-DS)) I think that these voltage references are used as current sources, the first one on the top left offsets the current so that it appears unipolar to the photodiode, and the second one offsets it again the other way so that the signal recovers its original polarity.

The thing is, on the input side for example, the voltage Vin can reach up to +/-1000V according to my understanding of the device and given the input resistor the current can reach +/-10mA. So if say the current reaches -10 mA the voltage reference must source 10 mA for the current to stay positive, this is because I am assuming the op amp to be ideal and that no current flows into the inputs. But at best the current sourced by the voltage reference is 5/38300<<10 mA. So how does this work ?

The full schematics can be found on (https://xdevs.com/doc/Stanford_Research_Systems/SIM900/sim984_sch.pdf)



The input part noted Vin above is shown below, J201 is the input, the GA net can be connected either to G1A or G10A, the GB net can be connected either to G10B or G1B, when I say that Vin can reach 1000V, I assume that the input is 10V with respect to GND-ISO and that G1A and G1B are connected to G10A ang G10B respectively:

 

What are you trying to achieve? It looks awfully complicated.

 

What are you trying to achieve? It looks awfully complicated.

I am just trying to understand the SIM984 design and how the designer from Stanford Research Systems made it, this is the only thing I do not understand yet: what makes this isolator bipolar, I know it is related to these voltage references, I just do not understand the components that have been chosen. In the end I will want to make an analog isoalter board of my own.

 

So, let's see...
At first, we see the two main things in isolated part, these particular things:

1. There is no any voltage divider at input.
2. Opamps are powered by +/- 12V.

In total these means input signal (related to the isolated ground) cannot be bigger than +/- 12V (it can of course but it's senseless because opamp will be saturated and then damaged).
So, no 1000V at input of course.
1000V is about grounds levels difference but not about the input signal range.

OK, it's for the signals in range from -12+(rail-to-rail reality) to +12-(rail-to-rail reality) and input impedance is 1M.

The next part is very simple also but a bit tricky:



U201 is voltage follower and no more. Jumper GA-G1A or GA-G10A (it should be jumper) and 90.9K resistor are interesting in purpose but not important at the "bird view", because it's voltage follower and no more, gain is 1 anyway.

U202A is a typical differential amplifier, the R206 || R207 in series with R208 are a kind of single resistor with non-standard very close to 110K value (I'm too lazy to calculate but 100K || 909K should be < 100K but very close to it), and there is R209 which gives out that this is still a differential amplifier. So, the gain should be very close to -10 (because of R205).
OK, but let's see here:


U202 is powered by the same +/- 12V.
And now we can know something about the input voltage at net labelled +IN_ISO.
This voltage should be in range -1V...+1V or slightly more, no matter.
Otherwise U202A will saturate or will work too close to rails.
OK.

And now the last part in the isolated domain:



Of course, linear optocouplers are "unipolar" devices in as in sense they can't transmit a current direction through a "galvanic barrier" as in sense there is only one LED in these devices and current can flow through it only in one direction.
So, because LEDs are current controlled and unipolar devices, there should be a very simple trick - bipolar voltage should be shifted to be unipolar.
That's why U204 exists.
And that's why because it's 5V Vref (and it's formidable Vref BTW) we know more about input voltage range, it should be in range -0.4-(a bit) V ... +0.4+(a bit) V, otherwise it will not be shifted right to unipolar range.

So, now it's much simpler and it what it is.
U203A with Q201 forms voltage to current converter.
There is feedback of course, look at junction between R213 and inverting input U203A.
The trick with paralleling of linear optocouplers is used to minimize optocoupler's non-linearity I'm pretty sure (despite I never used this trick before).

I hope it will help.
Linear optocouplers are really beautiful devices but...
For a long time it's much easier to place ADC in isolated part of design and have a deal with digital signals isolation.
And the results will be better because non-linearity of linear optocouplers will be avoided forever.

 

Interesting! I designed & built a similar circuit in the early 1970s using a single red LED illuminating a pair of tiny CdSe photo-resistors to make a wide range audio compressor/limiter. One cell was in the feedback loop of an opamp that detected the signal amplitude & drove the red LED, and the other was in the feedback loop of another opamp that acted as a variable-gain audio stage with low distortion & wide dynamic range [> 60 dB]. The trickiest part was aligning the two photo-resistor cells mechanically to "see" the same brightness, then packaging the assembly in an opaque container without disturbing the alignment. Took several attempts to succeed. The good ol' days of Fairchild 741 opamps with popcorn noise. LOL