The data was taken from a 5-hour window from 8 am to 1 pm yielding 288000 data points per channel. First I take the channels and subtract them in three different ways, 0-1 1-3 and 3-0 to yield all possible unique differences. Then I take the PSD of the data and take the square root. The output is given in the first graph showing very similar trends for all three differences.
Then I assumed all channels have the same transfer function but different offsets, so each PSD of the difference channels has the sum of the noise of both devices in quadrature. By adding difference PSDs with the channel we want and subtracting the other we get the noise of a single device squared. Taking the square root we can pull out the individual noise spectral density for each channel. It is interesting to note here some values from the PSD difference operation resulted in a negative value which was lost when square rooted.
The third graph shows the ASD of the voltage noise coming from the 15V power supply which powers the IC's and AD590s on the temperature board. There is a spike at 4 Hz in the plus and minus channel, which is a mystery as of now. I also shorted the input of the ADC to ground on channel two in order to measure the input-referred noise of the ADC itself. I used the ADC noise and the noise from the plus and minus rail to model this power noise contribution coming in through the AD590, OPA827, and OP27 of the circuit and added it to the noise sum at the output modeled in zero. To determine how much the noise propagates through the circuit, I multiplied each noise source to an IC first by its power supply rejection ratio and then by the transfer function to the output of the circuit.
The fourth graph shows this power noise added to the noise sum, which includes real data taken to estimate the power supply noise for each source as well as the simulated noise values from zero. The data plotted as a scatter plot is the same as in graph two and provides a real-world measurement of the overall noise measured at the ADC from the circuit as a whole. It seems the noise level measured is well below the expected noise, which is dominated by the OPA827 transimpedance amplifier stage.
To get input-referred noise I divided everything by the transfer function of the circuit yielding units of A/root(Hz), and then I divided by the 1u K/A TF of the AD590 to yield K/root(Hz) in the final graph.