I noticed that when the power to both cavities increases (6 mW to 12 mW), the beat noise get smaller at high frequency, broadband (200Hz and above.)

The problem is when I increase the power, I change several things:

1) + 2) error signals from both cavities

3) shot noise on RFPD

4) SNR on the beat

,so I haven't show the improved beat signal before I can pinpoint what actually limits the signal.

I checked the noise from both RFPDs, from 1MHz to 50MHz, ACAV's RPFD looks very different from RCAV's RFPD.

ACAV's RFPD noise also changes between morning and evening measurements, I'm not sure what is happening here.

The first set of data (in solid lines) is taken with the amplifier to increase the noise level, then divided by 30 which is

the increasing factor of the amplifier, see below. The second set in dotted lines are taken without the amplifier.

The signal without the amplifier is quite low ~ 5 - 10 nV which is close to the noise floor of the analyzer already. 

This might cause the factor of 3 in the signal of RCAV's RFPD noise.

About the changing noise shape of ACVA's RFPD, I stil don't have a clue. I'll check again tomorrow.

so I tried to switch the RFPDs to see if the beat signal can be improved.

Since the error signal on RCAV's loop is better, I thought that in can compensate with the effect from

bad RFPD. However, I could not get it locked, I increased the signal with the low noise amplifier before the mixer.

Still could not lock it. I returned the setup back to original and still could not lock it. No error signal coming out during the scan.

I worked around and then it's back on lock. So I guess there might be loosen connectors in ACAV's RFPD.

Note about the low noise amplifier I used:

The amplifier increases the signal by a factor of 30 (29.5 dB) flat from 1MHz to 100MHz, I used this to amplify the RFPD noise so that it's higher than the noise floor of the analyzer.