As a response to 9290, I doubled check the TFs of the TM and found that the signals were very coupled (~10%). Therefore I went through the OL diagonalization again. This time I did something different from the others. I first maximized the output in OPLEV_LEN_DIAG by moving IP in L and see how much it changed in OPLEV_LEN_PIT and OPLEV_LEN_YAW. From here I decide the rotational transformation for the length OL. I found that when I move IP in L, only 70% of the TM L was shown in OPLEV. Without doing any correction, I plugged in the numbers to enable reading OPLEV_PIT_DIAG and OPLEV_YAW_DIAG. If the TM L is accurate, when I move IP in L, OPLEV_PIT_DIAG should not change, but it did. That matches the fact that TM L is only measuring 70% of actual L, so it didn't completely cancel the L2P coupling. This could due to mismeasurement in calibration factor, misplacement of LEN QPD and error in various length measurements. I put in a correction factor so the TM L matches the IP L. After that, I did the same test again, OPLEV_PIT_DIAG barely changed. Then, I proceed to the normal diagonalization process, basically measuring L, P, Y TFs and calculate coupling ratios and found the matrix that decouples the degrees of freedom to below 1% level.
After that, based on the TFs measured, I roughly redid the TM damping filter but failure to suppress the "residual motion" seen by the OL. It fails regardless of the gain is large or small. I think that this is due to the fact that this OL still haven't had a windshield so the control is, to some extend, injecting noise. This was the same for the other type-Bs. So, I think if we really need to suppress the motion, we need to implement the windshield for SR3.