We are testing the SUServo, with a specific frequency noise mix with AOM RF(Urukul output) to disturb the light and using a PD to capture the optical power as the input of Sampler. But we found that the SU Servo can only confusingly surpress the noise which has a frequency less than 10Hz. Should that be correct?
No; the bandwidth will be limited by the loop latency (SUServo + AOM delays) and any low-pass filtering in photodiode and AOM response, but unity-gain bandwidths of >20 kHz should be reachable.
SUServo is, of course, a closed-loop system, so I was referring to the bandwidth at which the disturbance rejection reaches unit magnitude (i.e. noise is no longer suppressed). Below that, the disturbance rejection will improve as expected from the loop design (20 dB/decade in the low-frequency limit).
I'm not sure what you mean by "self-excited", but typically, you'll want both P gain and I gain. The appropriate gain settings strongly depend on the system parameters, but to give you a vague starting point, for a typical well-configured system, we might use P gain 0.5 and I gain 500 kHz.
