A while ago Oli found that some earthquakes cause IMC splitmon saturations, possibly causing locklosses. I asked Daniel this with respect to the tidal system to see if we could improve the offloading some. After some digging we found that some of the gains in the IMC get lowered to -32db at high power, greatly reducing the effective range of the IMC SPLITMON. He and Sheila decided that the best place to recover this gain was during the LASER_NOISE_SUPPRESSION state (575) so Sheila added some code to that state to redistribute some the gain (lines 5778 -5788):
self.gain_increase_counter = 0
if self.counter ==5 and self.timer['wait']:
if self.gain_increase_counter <7: #icrease the imc fast gain by this many dB
#redistribute gain in IMC servo so that we don't saturate splitmon in earth quakes, JW DS SED
ezca['IMC-REFL_SERVO_IN1GAIN'] -= 1
ezca['IMC-REFL_SERVO_IN2GAIN'] -= 1
ezca['IMC-REFL_SERVO_FASTGAIN'] += 1
time.sleep(0.1)
self.gain_increase_counter +=1
else:
self.counter +=1
We tried running this, but an error in the code broke the lock. That's fixed now, the lines are commented out in ISC_LOCK and we'll try again in some other day.
This caused 2 locklosses, so it took a little digging to figure out what is happening. The idea is to increase H1:IMC-REFL_SERVO_FASTGAIN, to compensate for reducing H1:IMC-REFL_SERVO_IN1GAIN and H1:IMC-REFL_SERVO_IN2GAIN, all analog gains used in IMC/tidal controls. It turns out there is a decorator used in almost every state of IMC_LOCK that sets H1:IMC-REFL_SERVO_IN1GAIN to some value, so when ISC_LOCK changes all 3 of these gains, IMC_LOCK was coming in after and resetting FASTGAIN. This is shown in the attached trend,on the middle plot the IN1 and IN2 gains step down like they are supposed to, but the FASTGAIN does a sawtooth caused by two guardians controlling this gain. The decorator is called IMC_power_adjust_func() in ISC_library.py and is called as @ISC_library.IMC_power_adjust in IMC_LOCK. The decorator just looks at the value of the FASTGAIN, Daniel suggests that it would be best for this decorator to look at all of the gains and do this a little smarter. I think RyanS will look into this, but looks like redistributing gain in the IMC is not straightforward.
tagging this with SPI. This would be good to compare against. The SPI should reduce HAM2-3 motion and reduce IMC length changes coming directly from ground motion. If IMC drive is to match the arm length changes then it won't help. (unless we do some feedforward of the IMC control to the ISIs?)