Not logged in
Thu Aug 6 04:36:41 2015, ericq, Update, ASC, Reviving PRC angular feedforward
Tue Aug 11 02:26:46 2015, ericq, Update, ASC, PRC Angular FF Lives!
Tue Aug 11 02:26:46 2015
In reply to:
PRC Angular FF Lives!
PRC Angular FF is back in action!
Short and sweet of it:
Took witness (T240 channels) and target (POP QPD) with DC coupled oplevs on. About 25 minutes of nice stationary data.
Downsampled everything to 32Hz, since coherence suggests subtraction only really possible from 1-5Hz.
Prefiltering done by detrending and
4 second FIR impulse time was enough
Filtered target by inverse actuator TF before sending to wiener code. The only difference between this and filtering the witnesses with the actuator TF directly is an effective RMS cost function, i.e. prefiltering.
Spending time tweaking IIR fitting pays off. Divided out zpk(0, [p3, p3*],1), where p3 is some well fit stack/suspension resonance, so that vectfit fits remaining portion with equal numbers of poles and zeros, guaranteeing AC coupling and 1/f rolloff to prevent noise injection
Quack->foton->OAF all worked fine
All in all, seems to work well. POPDC RMS goes down by a factor of 2
Code used lives in /users/ericq/2015-08-PRCFF and the NoiseCancellation github repo
Subtraction prediction vs. reality (positive dB is good)