There was some confusion as to the order in which we should go about trying to recover the Y arm. But here are the steps we decided on in the end.

1. Use the tip tilts to make sure the input beam is hitting roughly the center of ETMY, with ITMY left out.
2. Use the reflected beam from the ETM as viewed in the ITM chamber to set the pitch bias on ETM.
3. Center OSEM coils on ETM, rotate them to minimize bounce mode coupling into the sensor signals.
4. Install the ITM, look for cavity flashes, and use alignment biases to try and lock the Y arm in air.

Yesterday, Eric, Johannes and I tried to do step 1, but after some hours of beam walking, we were unsuccessful. Today morning, Koji suggested that the ITM wedge could be playing a part - essentially, over 40m, the wedge would shift the beam horizontally by ~30cm, which is kind of what we were seeing yesterday. That is, with 0 biases to the tip tilts, we could find the beam in the ETM chamber, towards the end of the table, ~30cm away from where it should be (since the input pointing is adjusted taking this effect into account, but we were doing all of our alignment attempts without the ITM in).

So, we shifted strategy today. The idea was to trust that the green beam was well aligned to the cavity axis (we had maximized the green transmission before the vent), and set the pitch bias voltage to ETMY by making the reflected beam overlap with itself. This was done successfully, and we needed to apply a pitch bias of ~-2.70 (value on the MEDM screen slider), which agrees well with what I was seeing in the cleanroom. We then adjusted the OSEMs to bring the sensor outputs to half their nominal maximum value. Next, we went into the ITMX chamber, and were able to find the green beam, at the right height, and approximately where we expect the center of the ITM to be (this supports the hypothesis that the green input pointing was pretty good). I am however concerned if this is truly the right value of the bias for making a cavity with the ITM, because the pre-vent value of the pitch bias slider for ETMY was at -3.7, which is a 30% difference from the current value (and I can't think of a reason why this should have changed, the standoffs weren't touched for ETMY). If we go ahead and fine tune the OSEMs rotationally assuming this is the right bias to have, we may end up with sub-optimal bounce mode coupling into the sensor signals if we have to apply a significantly larger/smaller offset to realise a cavity? The alternative is to put in the ITM, and set the pitch balance using the IR beam, and then go about rotating OSEMs. The obvious downside is that we have to peel the F.C. off, risking dirtying the ITMs.

For much of the rest of the day, we were trying to play with the rotation of the OSEM coils in order to minimize the bounce mode coupling into the sensor signals. We weren't able to come up with a good scheme to do this measurement, and I couldn't find any elog which details how this was done in the past. The problem is we have no target as to how good is good enough, and it is extremely difficult to gauge whether our rotation has improved the situation or not. For instance, with no rotation of the OSEMs, by observing the bounce mode peak height over a period of 20-30 minutes, we saw the peak height change by a factor of at least 3. This is not really surprising I guess, because the impulses that are exciting the bounce mode are stochastic (or at least they should be), and so it is very hard to make an apples to apples comparison as to whether a rotation has improved the situation on.

After some thought, the best I can come up with is the following. If anyone has better ideas or if my idea is flawed, or if this is a huge waste of time, please correct me!

1. Adopt this spectrum (except the side signal) as a reference for what constitutes "good" rotational orientation of the OSEMs (even though it is for ETMX not ETMY).
2. Start with one coil. The suspension assembly document tells us to expect the orientation with minimal bounce coupling to be located within 20 degrees of "the vertical", the vertical being defined as that orientation in which the line connecting the LED and PD as seen by eye is vertical. So start with the coil oriented vertically, as best as possible by eye.
3. Damp the optic for ~1min, with the curtain covering the chamber entrance. Ideally, we want the door back on, as this lowers the noise floor significantly, but it is too cumbersome to replace even the light door so I suppose we will have to compromise.
4. Take a reference spectrum. In the interest of time, I think a bandwidth of 0.1Hz on the Fourier Transform should be sufficient. (Tangentially related - the BW you specify in the measurement setup in DTT doesn't seem to be the BW with which the spectrum is computed, I wonder why that is?)
5. It is basically impossible to rotate the coil continuously. So divide the range to be explored into steps (so each step will involve rotating the coil by ~2 degrees (I don't know if this number is physically feasible, but some discrete step will be involved). Rotate the coil, center it such that the sensor output is close to half the maximum.
6. Pull the curtain down, damp the optic, and take another spectrum. If the bounce mode peak is higher, abandon this direction of rotation, and rotate the other way. We accept as the optimal position the one from which the bounce mode peak height gets worse by rotating to either side.

Of course, this method assumes that the excitation into the bounce mode is a constant over time. I'm also attaching the spectrum of the OSEM sensor signals right now - the optic is in the chamber, free swinging (no damping) with the door on (so it is fairly quiet). The LR signal seems to be the best (indeed seems to match the levels in this plot), but it is not clear whether the others can be improved or not.

There was also some concern as to whether we will be able to see the beam in the ETMX chamber once the ITM has been re-installed. Assuming we get 100mW out of the IMC, PRM transmission of 5.5%, and ITM transmission of 1.4%, we get ~35uW incident on the ETM, which while isn't a lot, should be sufficient to see using an IR card.