[Kiwamu, Koji, Suresh]
After correcting several errors in the WFS loops, we turned them on today and saw them working!
A while back (last week actually) I noticed that the WFS1 and WFS2 QPD segments were numbered in a different order but that their input matrices did not reflect this change. As result the WFS pitch and yaw definitions were pretty much mixed up. However even after clearing this up the signals still showed significant amount of cross couplings.
This problem was finally traced to the relative phase between I and Q channels of the WFS segments. Koji suggested that I check the relative phase between all the segments to be sure. I then repeated the procedure that Valera and I followed in our earlier elog # 5321 , and found that the phases indeed required to be adjusted. The excitation of MCL was at 6Hz, 100mVpp, as before. The WFS response after this was much improved i.e. the pitch yaw cross couplings were not visible when we misalign the MC with sliders in MC_ALIGN. And the magnitude of the response also increased since the signal was transferred from the Q to I channels. The the phases were tweaked by hand till Q< 1% of I. However when I repeated this measurement an hour later (I wanted to save the plots) I found that the phases had changed by a few percent!
Koji noticed that the MC_REFL camera image showed significant intensity fluctuations and advised that we try a higher frequency and lower amplitude to avoid nonlinear effects in the WFS and in the MCL to PSL lock. So we repeated the process at 20Hz and 20mVpp, introduced at the IN2 of the MC_Servo. The fig below shows the level to which we reduced the signal in Q.
We then checked the relative phase between various quadrants by looking at the time series in dataviewer. WFS2 Seg4 phase had to be flipped to bring it into phase with all the rest.
After this I tried to see the WFS response to moving the MC1 and MC3 with the sliders and determined the following relations:
Disregarding the MC2 for now and assuming arbitrary gains of 1 for all elements we inverted these matrices inserted them into the WFS_servo_outmatrix. We then found that the with a sign flip on all elements the loops were stable. In the servo filters we had turned on only the filter modules 3 and 4. There was no low frequency boost. We gradually increased gain till we saw a significant suppression of the error signal at low frequencies as shown below. There was also an associated suppression of Intensity noise at REFL_DC after a single bounce from PRM.
To see if the locks can actually realign the MC if it were manually misaligned, we turned the loops off and misaligned MC by moving MC3 pitch by 0.05 (slider position), and then turned on the loops. The locks were reengaged successfully and the MC regained alignment as seen on the StripTool below:
We can now proceed with the fine tuning the servo filters and understand the system better:
Q1: Does the WFS (I to Q) phase drift rapidly? How can we prevent it?
Q2: How is that we do not see any bounce or roll resonances on the WFS error signals?
Q3: How do we include the MC2 QPD into the WFS Servo?
I will proceed with determination of the actual transfer coefs between the MC DoF and the WFS sensors.