So far, we've been locking the south cavity by feeding back only on the laser frequency, not on the broadband EOM. This is because the demodulation path produces an error signal with the wrong sign for the EOM feedback, and the only way to flip the sign is to manually adjust cable lengths.
Last night I shortened the cable length between the LO 4-way splitter and the LO input to the frequency TTFSS in order to flip the sign of the error signal. The phasing is not quite optimal but it's pretty close (the cable length probably requires ~10 cm of adjustment at most). The power into the LO input was 5.9 dBm with the old cable, and is now 6.1 dBm with the new cable.
I then toggled the polarity of the inversion on the TTFSS from (+) to (−) and turned on the servo with laser frequency feedback only. I got the cavity to lock, but the error signal had huge 8 kHz sawtooth oscillations that I hadn't seen before (1st picture, below; trace 1 is cavity refl power, trace 2 is the error signal from OUT2 on the TTFSS). Then I added EOM feedback to the servo, and got huge 5 kHz sawtooth oscillations along with some spiky crap (2nd picture, below). Then I turned down the common gain knob to about 100 and the fast gain knob up to about 600, and got the loop to "stabilize" in the sense that the error signal now oscillates in the same way as before; i.e., with 200 kHz sinusoidal oscillations (3rd picture, below — note the similarity to the error signal oscillation in post 1123).
I took a quick spectrum of the error signal corresponding to the 3rd picture (attached below) with the SR785. Since the sinusoidal oscillations are happening at 200 kHz or so, this spectrum isn't capturing these oscillations.