How crystal mounting was done.

We did "in-between" method (alog 88900) and inserted two pieces of 3/8IDx5/8ODx0.025" thick shim washers between the front plate and the input side plate.  shim_washers_spacers.jpg shows the washers and the front plate before the rest of the EOM structure is placed on top. Note that the front plate still has the alumina piece on top, not the RTP. Important points in this picture:

1. We're using two pieces of alumina pieces as a convenient 4mm thick shim to raise the front plate. It doesn't have to be alumina pieces, but the front plate must be raised enough so the shim washers "clear the ground".
2. Three set screws are pre-adjusted so each sticks out just shy of 3.5mm. This is the height of the crystal (4mm) minus the depth of the shallow groove (nominally 0.02" or 0.508mm). This made it somewhat easier to to make sure that there's a good contact between the crystal and the metals.

(Added later: If I were to do this again, I'll set the set screws with alumina piece in place such that all touch the board when there's a good contact between board/plate/alumina (of course no indium). Insertion of indium foil with the real RTP later will automatically ensure that the screws are just shy.)

Next we cut a 40mmx4mm piece of indium sheet with a clean pair of scissors and installed it in a groove in the front plate. I tried to set the edge of the sheet to be 7.6mm from the outside edge of the front panel (this is 7mm plus the thickness of 0.025" washer). Judging from indium.jpg, I was mildly successful, maybe it's more like 7.4mm but it's not 7mm nor 8mm.

rtp.jpg shows the side view of the crystal. Maybe it's hard to see but it's wedged, in this picture the shortest face is down, the longest side is up.

I placed the crystal on top of the indium sheet, making sure that the edge of the crystal is well aligned with the edge of the indium as good as I can. We also made sure that the shortest face mates with the front panel. You cannot see any of that in the crystal_installed.jpg but you can at least see that the crystal is there.

Then we went through the same procedure we've already done more than several times, i.e. tighten the screws on the input panel until the panel touches the washer and the washer touches the front panel, tighten the scrwes on the output panel so the screws touch that panel, then go balanced tightening, moving just a tiny amount at a time, always applying small downward pressure for the EOM side/board/bottom assy otherwise the assy will shift when the screws are tightened. When all of the screws are tightened stronger than finger tight, screws on the input side are tightened just a bit more. After this, neither Elena nor I weren't able to undo screws by finger.

Sorry no picture of the assembled unit.
 

Tuning is done, larger capacitance than obtained before with alumina, looks good. (But why do the dips have to be so narrow?)

We noticed that the frequencies were lower than what we have previously obtained with alumina, i.e. the capacitance is bigger. This is probably a good sign even though we don't know if this is due to the indium or something else. Especially, 118MHz dip was a MHz or two lower than nominal.

We were able to tune all four frequencies using trim cap, except that the trim cap for 45.5MHz hit the minimum and we could not increase the frequency any more, so we bent the winding of the coil a bit to spread loops apart. Below is the table of center frequencies (actual vs nominal). In the attached photos, cursor is placed close to the nominal frequency. They all look good in that the frequency is close enough to nominal that we're only loosing less than a dB, but the resonances are all very narrow (for my preference, anyway). The Q values are from 640 for 9MHz to 1300 for 118MHz, going higher as the frequency. A small change in frequency will result in a big degradation in the modulation depth.

(Added later: Read Valera's entry below, thanks Valera! The numbers here are not the real Q, they should be smaller. To make it more embarrassing, I somehow mixed up 3dB and 6dB. Given that all dips are smaller than -20dB, a quick thing to do is probably to define the width of the peak as full width of the -3dB points in S11, not +3dB points from the bottom but really -3dB measured from 0dB full reflection. If you do that the numbers are more like 100 instead of 1000.)

If you look at the pictures, you'll also notice that the reflection dips at the center are -23dB (9MHz), -24.5dB (24MHz), -23dB (45MHz) and is -25dB (118MHz) so a bit smaller than 10% in amplitude is coming back. It's not really matched to 50 Ohm transmission line, and that on its own is OK, but because of that, I wonder if we can add a bit of resistance to bring down Q values without any negative impact (like worse matching with increased reflections) in the future design. 

After this first round of tuning, three set screws on the front plate were all extracted, and there was no change in the tuning of 118MHz dip.

I changed the orientation of the EOM and the frequency jumped a bit

Up to this point, tuning was done with the front plate facing down and put on the ceramic insulator placed on the EOM base (because it's convenient to access trim caps that way).

When I changed the orientation of the EOM so the front plate becomes upfront (i.e. like intended), the frequency of 118MHz dip shifted a bit, from 118.3055MHz to 118.31745MHz, it's just 12kHz shift so not the end of the world but it's still meaningful. Maybe it's the interaction of the magnetic field from the coil and the metals nearby?

Then I tapped the front plate and side plates and it shifted again, fortunately by smaller amount (from 118.31745 to 118.3113MHz, a negative 6kHz jump).

What we'll do tomorrow is to fully assemble the unit, tune it again as good as we can, then tap and retune if necessary. Hopefully, tapping enough and things will settle to the bottom of the potential.

Once the EOM goes into chamber we'll measure the resonances again, and we might have to retune in chamber.

I was also wondering why the S11 (return loss) is narrower then the transmission curve back when we did the prototype testing at LLO. So I did the math myself - the attached plot shows the calculated curves for voltage across the crystal for 1 W incident RF power and S11 for 9.1 MHz (similar for other f's). Initially I also made the same mistake estimating the Q - the Q is actually about 100 not ~1000 as one can see from the transmission curve (voltage on the crystal).