Summary: The QOSEM LED feedback loop is working, as it can be seen reducing noise of the raw QPD difference channels. Despite this, we still can't get down to the expected shot noise limit for the QOSEM readout, for reasons that are still not clear. Further modeling of this feedback loop and readout scheme are underway.

This is one in a series of posts investigating the excess noise seen on the BBSS M1 QOSEMs. In short, at both LHO and LLO, the noisefloor of the QOSEMs appears to be around 8pm/rtHz at 100Hz, quickly increasing to 50pm/rtHz by 1Hz; where we expect the noisefloor to be flat at ~5pm/rtHz across this bandwidth. See more details in LHO alog 91466. 

During a period of time with the BBSS M1 damping loops off (see LHO:92011), the QOSEM LED feedback circuit was switched on and off at the sat amp, to further investigate the LED behavior, which is the leading suspicion for the excess noise. In short, the LED feedback circuit takes the sum channel from each QOSEM, and uses this as an error signal to control the LED current. The idea being this suppresses the intensity noise on the LED. 

In Figure 1, we plot the 6D motion of the BBSS M1, as measured by the QOSEMs. Here we see that the state of the LED feedback being on or off has no noticeable effect on the noisefloor. Likewise the damping being on and off has no effect on the noisefloor observed,

In Figure 2, we plot the OSEM basis motion of the BBSS M1, as measured by the QOSEMs. Here we see the state of the LED feedback has some effect on the F1 and RT QOSEM, just in there y axis. This was not observed in the figure 1 6D Euler basis plot, as this is formed by combining many QOSEM measurements in the OSEM2EUL matrix conversion, which obscured things if only a channel or two are off.

In Figure 3, we plot raw voltage spectrums from the QOSEMs from the same datasets as figure 1&2, showing the raw QPD Sum, XDiff and YDiff signals. Here we see a few interesting things:

• Looking at the sum channels the feedback loop is certainly doing something, as we see the sum channel go from about 1uV/rtHz with no feedback, then suppressed down to the ADC noise level, as one would expect from an in loop signal.
• The XDiff channel does not change significantly with feedback on or off
• The YDiff channel does drop significantly when the feedback is switched on
• The sum channel for F1 and RT is not included as these channels were saturated at some points on the ADC

In Figure 4, we plot the DC position of the QOSEM flag, against how much the noise on the QPD difference channels reduced when switching the LED feedback off to on. We see an interesting correlation, that as the offset increases, so too does the amount of suppression provided by LED feedback. This is actually expected; when the QOSEM flag is far from centered, the common mode rejection of the QPD difference channel reduces (See slide 52 of G2500799), making it more susceptible to LED intensity noise. So when the QOSEM is far from centered, the LED feedback is seemingly doing it job, and suppressing the LED noise such that the difference channel is less noisy, and when the QOSEM is relatively centered, the LED feedback has no effect, as the good common mode rejection in this regime is already suppressing the LED noise.

While I'm still thinking over what this all means, the take aways are, the LED feedback loop is definitely doing something, as its helping the QOSEM difference channels which are far from center be less influenced be LED noise. However, this isn't actually having a huge impact on the QOSEM readouts, as the normalization we do in the readout scheme (dividing the difference channels by the sum), seems to be getting rid of this LED noise anyway, even if the feedback loop isn't active. The question remains though, why can't we push below this limit we are seeing, why are the sum normalization steps and the LED feedback not allowing us to be anymore sensitive at 10Hz. I am going to try run some models to predict the LED noise coupling through the circuity and readout chain, and I think measuring the coherence between the QPD Sum + Diff channels to the LED drive current at the sites will be useful.