Figure 1. Looking at the CARM sensor RFC 6MHz I, it has a noise with a shape of 1/sqrt(f) between 10Hz and 200Hz. On other signals (for example B2), we do know that this is demodulation amplitude noise, and that its amplitude is proportional to the signal amplitude and can be subtracted using the amplitude noise of a different RF frequency demodulated on the same photodiode. There is no coherence with RFC 12MHz mag, but this is likely because 6MHz I is in loop and changes sign several times per second which would destroy any coherence.

Figure 2 shows that there are times during the lock acquisition when the RFC 6MHz I signal has a large offset from zero

Figure 3 during those times RFC 6MHz I and RFC 12MHz mag are coherent, so one could be used to subtract noise from the other.

Figure 4 looking at the RFC RF spectrum, the 12MHz line is the highest, so it is the best candidate for amplitude noise subtraction, and in time domain it is has a roughly constant non-zero value.