Because of the way RMS works, I think the dither's 1st harmonic (i.e not the fundamental) is not contributing much to the RMS. Its only 1/5 or 1/10 of the total, so if you removed it the RMS would be:

RMS_now^2 - dither_RMS^2 = RMS_better^2

I think Lee's suggestion of using a broader band is good:

1. Use a SBP-70 or SBP-100 to bandpass the 1811 output

2. square it using a double balanced mixer (e.g. ZAD-1 or ZAD-3)

3. Low pass the IF output of the mixer using a low pass like SLP-5 or SLP-10 or SLP-30

4. Demodulate as you already are doing

But check the spectrum, as it is, to make sure there are no weird lines in there which would give you an anomolous error signal.

{Shruti, Yehonathan}

For some reason, we were not able to raise the loop gain using the Moku screens despite having at least 10db of gain margin so we just stay with the current loop for now.

Attachments 1 and 6 shows the updated OLTF with a UGF at ~ 55Hz (depending on the green laser pump power). We set out to measure the calibrated suppressed and unsuppressed LO phase noise.

First, we calibrate the LO phase readout by opening the loop and looking at the extrema at the error point (attachment 3). We estimate the plant to be 1.209 V/rad.

Then, we close the loop and calibrate the actuation by injecting a swept sine to the error point and looking at the control point. The resulting transfer function is shown in attachment 4 in red. In the control band, the actuator is estimated to be

3.6 rad/V. We now use these values to plot the calibrated error and control points shown in attachment 5. 

We also show the RMS of the ctrl and the error point. Notice how bad the RMS of the error point is because of the unsuppressed 2nd harmonic of the dither, but it can be considered part of the sensing noise.

{Shruti, Yehonathan}

Summary

- switch to summing after demodulation (Tried FIR, IIR, etc)

- Measured OLTF

Details

Following our unsuccessful OLTF measurement we realized that we put our summing junction before the demodulator which made the loop weird.

We switch to the more conventional approach, and by doing so we found a more compact setup using only 2 MokuGos + MokuLab (Attachments 5 and 6).

As before the MokuLab is low-noise and is used as a preamplifier for the BHD + bandpass filter. Its output is connected to the Moku shown in attachment 5. Lock-in on the left (attachment 2) is the envelope detector. The lock-in on the right (attachment 3) is used to modulate the LO phase (using the piezo on the green path which seems to have a reaction mass) at 5kHz and demodulate the signal. A low pass at 1kHz with 24db/octave is used for the LO phase detection.

Then, the LO phase signal is amplified by 50dB before being sent to the 2nd MokuGo (attachment 6). There, a PID controller (attachment 4) is used as a control filter and a summing junction for the OLTF measurement. We tried other filter modules like IIR and FIR filter boxes but non of them had a first-order LPF :/ so we ended up using the PID screen.

The LO phase signal (In 1) is summed with an excitation signal from the frequency response analyzer. The summed signal goes through the filter (attachment 7) and is fed back to the piezo on the red path. The summed signal also goes through an all-pass filter for analysis in the frequency response analyzer.

The OLTF measurement is done in the frequency response analyzer (attachment 1). There, we calculate using the math channel the ratio between the unsummed signal and the summed signal to get the measurement (yellow line). It can be seen that our UGF is around 40Hz with a 10db gain margin, which we can use to easily boost our UGF to 200Hz. Unfortunately, the instrument doesn't have a coherence measurement. Maybe we should remeasure using the SR785.

* a diagram of the Moku setup

* steps at the input of PID control

* TF measurement

Offset locking

When we had tried locking to LO phase angles other than 0 and 90 deg, we were not able to do it yesterday by adding output offsets to the controller. In the setup shown in Attachment 4, between the envelope detector and the controller we added another instrument to perform small signal injection and also offsets (the best choice we had was the PID controller app, with only a 0 dB proportional term).

From the point with only squeezing, by applying steps of 200 mV in the input offset location of the PID (also shown in Attachment 3), we saw that we were able to lock to different offsets. In Attachment 2, we see the steps as measured by the envelope detector (before the summation point, i.e., after going around the entire loop). At 1.4 V (around -25 s in the timeseries) we see that the loop becomes noisy and does not increase further. This is where there is only antisqueezing.

OLTF measurement

We used a third Moku:Go as a frequency response analyzer. We injected the excitation at the error point (using the summation), and measured before and after excitation in the loop. In attachment 3, we see the signal at these two points. In our attempts we were not able to get a sensible transfer function so far. (side note: 1 Moku:Pro = 3 Moku:Go's for our needs, or the Pro could be better with the digital signal bus connections instead... See Attachments 1,4,5)

[Yehonathan, Shruti]

Today we tried the quantum noise locking idea described in this paper, and after a bunch of troubleshooting we finally got it to work!

Implementation:

We used the Moku:Lab as a huge band pass filter between 4-6 MHz, then fed this to the envelope detector that we were already using which was set up on the Moku:Go. The signal from this was sent, in Multi-Instrument mode, to another Lock-In amplifer.

The Lock-In is shown in Attachment 1 and 3.

- A 5 kHz signal was sent to the PZT in the green path because we could see no evidence of a coupling-related intensity modulation. When we modulated the PZT in the red path, we saw that the difference channel still had a peak even when the waveguide was NOT being pumped. This was because of some coupling modulation into the fiber that we had seen earlier.

- We set up the Lock-In where the two ports of the mixer were the LO dither signal and the output of the envelope detector.

- Initially we were using a low pass signal above a 100 Hz, but were not able to acquire lock. Attachment 3 shows how we appeared to 'lock' to shot noise when we had very high gain [bogus].

- We then realized that we needed an integrator in the servo and so we changed the low pass to 1 Hz with a 6dB/octave dropoff. We were then able to see a good lock! This is a bit ad-hoc and would need something better for our next version

- The output of the lock-in goes to the piezo controller that can output up to 100 V, which is then connected to the PZT in the LO path

Results: Attachment 1

- The panel shows the "servo" settings for the lock-in

- {Timeseries: -19s to-14s} From an initially locked state, blocking the green pump beam shows the shot noise level at ~170 mV

- {Timeseries: -14s to -9s} After unblocking we see that it locks to squeezed quadrature, here shown at noise level ~150 mV. This corresponds to ~0.55 dB of squeezing and is this low since we are averaging over a huge band

- {Timeseries: -9s to -7s} By inverting the gain, we see that it locks to the antisqueezed quadrature at ~ 195 mV. The lock here is less stable (also explained in the paper). When we decreased the gain we were able to get a better lock (not shown here).

- {Timeseries: -7s to -1s} When we block the beam we still recover a similar shot noise level

- We were able to see a lock that lasts >> 30 s on the squeezed quadrature (not recorded here though...)

Attachment 2 is the spectrum at the output of the envelope detector, where we see the 5 kHz peak 10 dB above the noise floor.

We also began to set up a second Moku:Go to measure squeezing in a band different from the one we use for locking.

In our new set-up, every time we return to the lab, we have to spend some time re-aligning to recover green coupling...

[Yehonathan, Shruti]

1. Simple rotation to match 532 nm polarization to SHG

After yesterday's alignment, since we had just arbitrarily placed the bare fiber on the mount, we had to make sure we had the right polarization between the LO and squeezed field.

For this, we first looked at the polarization of the light generated with SHG and aligned the PBS right before the green coupling into the fiber such that we had maximum transmission (we actually checked for minimum transmission and then rotated it by 90 deg). Then with the green from the Diabolo, we rotated the HWP before the PBS to get maximum transmission after the PBS. Then, we had to realign again and more or less got the same results as yesterday.

When we pumped with 100-150 mW of green, we once again saw squeezing at the level of 1 dB.

2. Using the 3-paddle polarization controller and polarization detection setup on the LO path

To further improve the polarization control, we thought of adding the FP030 3-paddle polarization controller. From the ThorLabs website for this model, 2 loops corresponded to  and 3 loops to . We used the yellow fiber that we found with the part  and therefore assume that this was the non-polarization maintaining fiber for 1064nm required for controller (no label). We made the appropriate number of loops (3->2->3) to get it into the roughly QWP->HWP->QWP configuration.

We added another PDA100A to the reflection port of the PBS in the QWP->cube PBS->PDA100A setup (see Attachment 2) that was on the table to make some sort of a polarization homodyne detector. Without the polarization controller, with the LO fiber, we rotated the QWP and saw that we got very close to 0 on one of the channels suggesting that we had pretty much linearly polarized light. With the polarization controller in its supposedly nominal position (all paddles vertical), we did not get the same polarization. Also, moving this non-PM fiber around or even touching it caused glitches and weirdness in the polarization. We mounted everything as fixedly as possible to cause minimal disturbance to this fiber.

We moved around the paddles till we saw roughly the same polarization as directly from the 1064 nm LO fiber, and then connected this in between the LO fiber and the LO port of the BHD BS (see Attachment 1).

We once again looked for squeezing and tried to further optimize the paddles to get maximum squeezing/anti-squeezing. We were somewhat successful with the HWP, but with the phase drifts and shot noise level changing, we could not do much better today.

Attachment 3 shows some uncalibrated polarization drift even without the 3-paddle controller. The max signal seen on each PD was almost 4 V, so a 100 mV drift is approximately 10 mrad. We still have to measure noise and drift more carefully with this in the setup, but it is unlikely to give us a large enough improvement as improving the coupling/pumping or reducing the out-coupling loss would.

{Shruti, Yehonathan}

After fiddleing with the fiber coupling unit similar to this one we decided wo don't like it and resorted to just taping the fiber on a piece of metal. We took the 20X objective out of the unit and put it on a post mounted translation stage.

We generate SHG light and place the fiber close to objective lens. We collimated the beam as best as we can steered the beam such that it goes through both irises in the green path to get good starting point.

Setup is shown in the attachment.

After alignment process we were able to get 1.3mW at the waveguide-output-fiber when 10mW is measured after the Diabolo. It is slightly worse then our best coupling efficiency which was 1.6mW. We didn't check the SHG mode size which could be made better if we wanted by choosing different telescope lenses.

[Yehonathan, JC, Shruti]

On Friday, with a stripper and cleaver borrowed from the Marandi group, we initially practiced stripping and cleaving on a bare fiber we found at the 40m. Then, we proceeded to do it on the input (532 nm) fiber of our waveguide.

We stripped about half a cm of the jacketing, then inserted it into a fiber tip holder. After which we cleaned the stripped region with 100% IPA and a lens tissue. Placing it securely in the CT50 cleaver where the blade was touching the stripped region, we pressed it down and cleaved it.

We then pumped the other side with 1064 nm to see SHG. Initially, when we tried it the mode seemed to have Bessel-type fringes. But, we realized we were straining the fiber, and the mode actually looked very beautifully cleaved!

[Yehonathan, Shruti]

TL;DR: After meeting with Alireza Marandi's group, we found out that removing the fiber connectors from the Advr waveguide is not too hard since they are just glued to the waveguide. The tricky part is in the setup of the waveguide chip once we remove the packaging and fibers, and the changes to be made in the BHD and in-coupling setup.

Process to remove the fiber connectors

• Removing the packing will cause the glued fiber connectors to come off
• Wash in acetone to remove any additional glue and clean the surface. We expect the waveguide edges to be polished

Mounting the bare chip

The present butterfly connectors and diode holders do not offer enough space to place the objectives near enough to the waveguide, so we have to think of a better way to mount the bare chip while also being able to use the TEC, somewhat based on what we saw in Alireza's labs (attachment 2). We intend on meeting his students tomorrow to learn more about how to get the parts that they use for their mount. Until we get these parts we suggest we continue with the plan to cleave the fiber.

Changes to the in- and out- coupling optics

Attachment 1 shows the schematic and plan of how we plan to do this.

• BHD now requires a free-space BS
• Another 1064 nm beam splitter to be able to align the BHD and measure visibility
• Two additional 1064 nm flip-mount mirrors to be able to switch between being the LO for SPDC and pump for SHG
• Dichroic mirror/ filter for 532nm

The steps for undertaking the experiment:

1. We couple 1064 nm into the "green" side of the waveguide in order to see the mode shape that the waveguide outputs from the "red" side.
2. Measure the mode profile of 1064 nm on the "red" side. This is the mode shape that the LO beam must match to
3. Build a mode-matching telescope for the 1064 nm LO
4. Align the BHD setup. This includes polarization, spatial mode, etc
5. Flip the mirrors to pump the "red" side of the waveguide with 1064 nm
6. Generate 532 nm by SHG with strong 1064 nm pumping. The 1064 nm mode entering the WG must be efficient for SHG with the previous mode-matching solution.
7. Optimize the waveguide temperature for SHG
8. Profile the mode of the 532 nm beam coming from the "green" side of the waveguide.
9. Build a mode-matching telescope for 532 nm to match the above profile

{Shruti, Yehonathan}

While setting up a new green coupling set up a spacer fell on the bare waveguide input fiber and broke it.

The broken piece is not very large and we still have a lot of fiber (attachment 1) to work with.

Attachment 2 shows the mode of the broken fiber illuminated with SHG from the waveguide.

Next Steps

We need to find a fiber stripper and a cleaver. This is a common tool in fiber optics labs and I'm sure we can find a place on campus to borrow from. If we don't find it we can buy a fiber stripper (like this) and a cleaver (like this) from Thorlabs.

Once we have a cleaved fiber end we are back in business.

In the longterm, we will have to make a new connector for the fiber using one of these kits. Something I have some experience with.

From Thorlabs

S1FCA2 - Ø1" FC/APC Fiber Adapter Plate Without Threads, Narrow Key (2.0 mm) X 3

AD12NT - Ø1" Unthreaded Adapter for Ø12 mm Cylindrical Components X 1

TC06APC-532 - 532 nm, f=5.95 mm, NA=0.28, FC/APC Triplet Collimator X 1

SPW801 - Adjustable Spanner Wrench X 1

SPW301 - Spanner Wrench for an M9 x 0.5 Optics Housing, Length = 1" X 1

SM1ZP - Z-Axis Translation Mount, Post Mountable, 1/4"-20 and 8-32 Taps X 1

SM1A3 - Adapter with External SM1 Threads and Internal RMS Threads X 3

E08RMS - Extended RMS to M8 x 0.5 Adapter X 1

E09RMS - Extended RMS to M9 x 0.5 Adapter X 1

C110TMD-A - f = 6.24 mm, NA = 0.40, WD = 1.59 mm, Mounted Aspheric Lens, ARC: 350 - 700 nm X 1

A110TM-A - f = 6.24 mm, NA = 0.40, WD = 2.39 mm, Mounted Aspheric Lens, ARC: 350 - 700 nm X 1

LM1XY - Translating Lens Mount for Ø1" Optics, 1 Retaining Ring Included X 1

HW-KIT6 - Imperial Screw Thread Adapter Kit X 1

From Coastalcon

https://coastalcon.com/wp-content/uploads/2020/10/Polarization-Maintaining.pdf 480 1 m patch cable FC/APC

[Yehonathan, JC, Shruti]

In order to improve the green pumping and coupling into the fiber, we purchased a separate fiber collimator and began changing the setup to accomodate it.

Attachment 2 shows the new commilator and a translation stage contraption we would use for the fiber head. Required some machining and crafting to get it to the right height while also being about 6 mm away from the lens. We removed the previous collimating lens from the coupling head.

Attachment 1 shows a plastic tube we placed around the bare fiber to protect it. This is fiber from which light is directly coupled into from free space and is attached to the waveguide.

[Yehonathan, Shruti, JC]

Today we first rotated both the PBS and HWP before the green fiber coupler so that the polarizations reaching the homodyne BS would be matched. After doing this, we improved the alignment a bit to get 1.6 mW out of the 1064 nm port of the waveguide with 10 mW green being sent from the Diabolo.

We repeated the measurements to see if we could still see squeezing. We did see squeezing with a 5 mW LO. While playing with the filter BW we also noticed that a smaller BW upto ~50kHz seemed to consistently improve the level of squeezing we were able to measure! We saw up to about 1.7 dB of antisqueezing. See Attachment 1.

We did measurements of squeezing both at high frequency (2.45-2.5 MHz) and low frequency (100-150 kHz) as a function of power. The error bars were derived from the standard deviation for a single shot timeseries. There were drifts in the levels of squeezing and LO power over the course of the measurement that probably require larger error bars and a more careful measurement.

The bandpass filters used for high and low frequencies are shown in Attachments 2 and 3 respectively.

Also, the squeezing arcs for these measurements with 125 mW with a high frequency bandpass and low frequency bandpass are in  Attachments 4 and 5 respectively. We used 200 Hz and 2Vpp to the LO PZT. The lower frequency one was noisier. Also, overall these arcs measured in a smaller BW do not look as smooth as ones with larger BW but do show more squeezing.

We measured the level of squeezing and antisqueezing plotted in Attachment 1 using the average of "min hold" and "max hold" for a combination of free running and PZT dithered timeseries of the RMS^2. The shot noise was measured once for the LF and HF measurements by blocking the beam and averaging the RMS.

The notebook used for analysis, and data are uploaded to gitlab.

[JC, Radhika]

On Wednesday we started a cooldown of a 40um alumina strip by itself - not epoxied to a Si wafer. This was motivated by a desire to decouple the emissivity of the alumina from the emissivity of the Si+epoxy+alumina sample, and see if there is a non-negligible difference in cooldown. Granted this 40um strip is quite thin and transparent in visible - it will be interesting to see how much thermal coupling is present with the enclosure.

First we replaced the RTD at the cold head with a PT100 (previously a PT1000 giving unphysical temperature values for the coldhead - not fixed by recalibration). We spring-clamped the new RTD to the cold head and closed up the bottom conflat of the T [Attachments 1, 2].

A PT100 was varnished to a new 40um alumina strip. We opened up the main chamber lid and swapped out the previous sample (epoxy layer on Si wafer) for the alumina strip [Attachment 3, 4]. We then closed up the chamber.

Vacuum pump turned on at 4:15pm, crycooler turned on at 4:50pm.

[Yehonathan, Shruti, JC]

Yesterday: Coupling efficiency

With 10 mW of green light from the laser, we saw 1.6 mW at the other end of the waveguide when the light was directly being coupled into the PM480 fiber. We expected 3 mW. For sanity, we repeated measurements of coupling with the patch cable.

With 10 mW of green light from the laser, with a patch cable we were able to see 65% coupling at its other end (6.5 mW). At the end of a second PM460 patch cable, we saw ~3 mW. Then at the 1064 nm otuput of the waveguide we saw 1.1 mW.

With the same 10 mW green light, we again then coupled it to the waveguide. Since we would not have the 50% mating sleeve loss, we expect 3 mW at the 1064 nm output. But, again, we could not seem to improve the coupling to obtain more than 1.6 mW.

Mode-matching cross-checks

- With the WinCam beam profiler placed right after the piezo in the green path, when we shone green light from the Diabolo we saw a beam with a diameter ~900 um as shown in attachment 2. When we pumped the 1064 nm and obtained the waveguide-generated SHG light at roughly the same location, the beam diameter was ~1100 um as shown in attachment 1. For perfect mode-matching to the waveguide, we would expect the diameters to be the same...

We also checked where the green beam and the SHG focus in the imaging telescope. The two beams seem to focus at slightly different planes along the path which indicate that the radii of curvature of the beams are mismatched as well which could be a serious issue for mode-matching. Unfortunately, the fiber is coupled from free space using a fixed focus collimation package Thorlabs F240APC-A which prevents us from moving the fiber along the optical axis. To proceed we either need an adjustable focus collimator or separate the focusing lens from the fiber tip and use a translational stage for adjustments.

Assuming a Gaussian mode we can use the formula for the mode field diameter (MFD)

, where NA is the numerical aperture of the collimating lens, f is its focal length  = 7.86mm and, EFF_2W = 1.15mm is the measured beam collimated diameter which can be read off from attachment 1. MFD is estimated to be 4.6um. The datasheet spec 4um.

For an adjustable aspheric collimator with a focal length of 7.5mm the needed beam diameter at the fiber coupler should be 1.1mm.

For an aspheric lens with a focal length of 6.2mm the needed beam diameter at the fiber coupler should be 0.9mm which is what we currently have.

Shopping list

https://www.thorlabs.com/thorproduct.cfm?partnumber=CFC8A-A

https://www.thorlabs.com/thorproduct.cfm?partnumber=C171TMD-A

https://www.thorlabs.com/thorproduct.cfm?partnumber=S1TM08

https://www.thorlabs.com/thorproduct.cfm?partnumber=SPW308

[JC, Shruti]

At the laser output: 10.1 mW,

The end of the first fiber after the fiber launch: 5.7 mW,

Mating sleeve ADAFCPMB4 (PC to PC or APC to APC): Nominal 2.2 mW but by loosening fibers, moving them around we could measure upto 2.9 mW (~60% loss)

Mating sleeve ADAFCB4 (APC to APC): Nominal 2.6 mW but also varied depending on connection (~54% loss)

This was with the new key position, and cleaning the additional patch cable again.

We decided to directly connect the PM480 fiber that is attached to the 532 nm port of the WG to the free space to fiber coupler to get rid of the mating sleeve loss. We cleaned the fiber tip, attached it, and looked at green power at the 1064 nm port. Since the alignment changes, we spent some time maximizing this power. Initially when looking at the beam shape on a piece of paper at the 1064 nm fiber we tried to see if we could still get close to a Gaussian beam, but after playing around several local minima, the 532 nm scatter to higher order modes in that fiber  is probably too high to preserve the beam shape. At the end, for 5 mW at the Diabolo output, moving from a position where we got >50% coupling into a patch cable, we saw 0.77 mW at the WG output. Without the extra mating sleeve loss, if there is a coupling efficiency of 50% into the fiber and 50% at the waveguide output, we'd expect 0.8 mW. 

{Shruti, Yehonathan}

Dhruva pointed out that even with 85% loss we should observe at least a few dB of anti squeezing. The fact that we see ~ 0.7 dB of anti squeezing means that we are not yet limited by loss, we are just not pumping hard enough.
 

We check the green fiber mating to the waveguide input fiber coupler. We found that they both had some residual index-matching gel that we applied a few months ago. We removed the gel using IPA and some tissues. Inspection shows that the fibers are clean (see attachments 1,2).

This didn't seem to have an effect on the squeezing signal. Suspecting that the coupling between those two fibers is bad. We measure the power at waveguide output. Even though the fiber output coupler is designed for 1064nm, it should still have a good transmission for 532nm (but multimode). We measure the power at the green fiber output to be 30mW. At the waveguide output it was 5.5mW. We take an identical fiber PM460 and check the coupling between it and the green fiber and measure a power of ~ 9mW which gives 70% loss!. This means there is a great loss in the coupling of the green fibers. We should either replace the mating sleeve or couple the green light directly to the waveguide input fiber coupler.

We connected the fibers using a different mating sleeve but the result stayed the same.

We decided to couple the green light into the waveguide directly from free space. Fortunately, the WG fiber is long enough. After removing the green fiber we noticed that its key is pointing sideway instead of up unlike the mating sleeves where all the keys are pointing upwards (see attachment 3). We decided to rotate the key upwards and put back the green fiber, thinking it might solve the issue with the lossy mating of the fiber. Once we regained good coupling efficiency from free space we tested the mating again with the two identical fibers which again showed 70% loss. We will continue with the plan to couple the light to the WG input fiber directly from free space.

JC reminded me about this. I checked the legs on the lower table in QIL. Due to the rigid connections between the legs, it's not physically possible for the table to tip over (unless it is thrown into the air by several feet, enough to rotate/roll the cage onto its side). At the moment, it's sitting on top of the legs and the main danger, in an EQ, would be for the table to travel across, relative to the legs underneath, and slip off in the in the North or South directions. The table would need to move 2 feet in order to fall onto the floor - which could happen in a big quake. There's no holes on the underside of the table, so we'll need to drill and tap into the plate to create points that we can bolt clamps into.

[JC, Radhika]

Today we opened up the chamber for the first time after its transport. The contents remained in tact; nothing looked shifted or damaged [Attachment 1].

We worked towards mounting the previously-prepared "control" sample: Si wafer with a ~100um layer of epoxy. The goal is determine if the epoxy itself has a non-negligibly high emissivity. I soldered a new PT100 RTD to existing leads and varnished it to the bottom side of the sample [Attachment 2]. This was because I wasn't sure how the varnish would interact with the cured epoxy, and the temperatures of the top and bottom sides of the wafer would track with negligible time constant.

Next we mounted the sample in the chamber and made the necessary RTD connections [Attachments 3,4]. We closed up - vacuum pump switched on at 5pm; cooler switched on at 5:30pm.

This is the first cooldown since Megastat's transport, so we will monitor the run closely to make sure everything is operating as expected.

{Shruti, Yehonathan}

We move on with the measurement and open the green shutter. We measure 3mW/5mW=60% green coupling efficiency where 5mW is measured right after the polarizer in front of the Diabolo and 3mW is measured at the green fiber output just before it is mated with the waveguide fiber coupler.

We couple green light with different powers measured at the Diabolo output (25, 50, 75, 100, 125 mWs). A mirror mounted on a piezo in the LO path is dithered to modulate the LO phase using a function generator.

The dithering frequency is 200Hz and the piezo Vpp is 3V. LO power used is 5.2mW.

The noise RMS^2 is measured using the readout scheme described in previous elogs. At each green power, we note the shot-noise level by blocking the green beam. Filter box settings are shown in attachment 1. We decided to increase the filter BW to minimize the fluctuations in the measured RMS^2.

We observe what seem to be squeezing arcs. These are not due to the LO coupling efficiency modulation because they do not show when the green is off and the A-B channel is insensitive to this modulation.

https://git.ligo.org/cit-qil/data/-/blob/main/WOPO_SQZ/20230213/PXL_20230213_201708103.TS.mp4 shows a video of the MokuGo lock-in screen. In the beginning, the green light shutter is closed and at around ~ 5-sec mark the green is exposed.

Attachment 2 shows a screenshot of one of the measurements when the green power was 125mW. The central dashed line shows the shotnoise level.

Attachment 3 shows the squeezing-antisqueezing levels for each green power. This was estimated by just taking the maximum and minimum values at each time series. The noise level relative to the shot noise is then just 10*log10(noise/shot). The 10 here is for measuring noise RMS^2 instead of RMS.

It seems like the squeezing saturates at the -0.5db level. Which implies a loss of 90%.

The next thing is optimizing the squeezing:

1. Check the polarization of the signal and LO.

2. Play with the crystal temperature again.

3. Stabilizing LO phase maybe.

We should also add errorbars to the squeezing levels plot.

The New Focus (aka Newport) 1811 RFPD has a tiny active element, ~0.3 mm in diameter. The photodiode is readout by the negative input of a transimpedance amp, so, within the bandwidth of the feedback, the photodiode voltage drop is mostly just the bias voltage: don't need to worry about the PD capacitiance modulation nonlinearity mechanics probably (cf. paper by me and Hartmut on RFPD saturations).

With 1 mA of DC current, the shot noise is sqrt(2 e I) ~ 18 pA/rHz. The RMS from 0.1-100 MHz = 0.2 uA. The transimpedance of the PD is 40 kOhms. So that's only a RMS voltage of 7 mV. Not enough to saturate any analog elecronics.

Yet, the manual claims the CW saturation limit is at 120 uW. Is this just for the DC coupled version? I think so, but also don't know when the RF stage actually saturates. Would be interesting to see if the spectrum looks funny as you change the input power.

[Yehonathan, Shruti]

Before moving forward we check again the scaling of the noise RMS^2 as a function of LO power.

Unfortunately, today we find that it doesn't scale lineary as expected. We realize that this means we are saturating the electronics/Moku somewhere.

We first notice that at low powers, when the LO is less than 2mW (right after the polarizer in front of the laser), the scaling works.

By playing with the digital gains in the filterbox on the MokuLab we see that we can get the right scaling, which suggest there is some issue with the MokuLab filterbox implementation.

In more detail, changing the input gain by +10 dB caused saturation for the same LO powers. We saw that the same thing happened even when we decreased the number of filter coefficients. This is consistent with the saturation taking place in the filter box.

We have decided to proceed with the 0 dB gain where the scaling works as expected. The data we recorded is in Attachment 1.

Megastat was returned to the north table using the engine hoist [Attachment 1].

The pump hose was reattached and the compressor returned to its original location [Attachment 2].

We reconnected the temperature/gauge controllers and ensured that the RTD EPICS channels were reading correctly [Attachment 3]. 

Next week we plan to open up the chamber and do a post-transport health check of the contents. The aim is to cool down a silicon sample with epoxy (no alumina) to determine the emissivity of a thin epoxy layer. 

{Shruti, JC, Yehonathan}

We set up the RMS measurement of the noise around 4MHz. The setup is illustrated in attachment 1.

The PDs are connected to the input channels of MokuLab since it's low noise. In the filter box instrument, they get subtracted from each other by the input matrix, amplified by 50db (10db before filtering and 40db after), and filtered by a band pass filter in the range 4.185 - 4.853 MHz and sent to the output port of the MokuLab. That output port is then connected with a BNC cable to the input of MokuGo. There, the input is split digitally and fed the lock-in instrument. In the lock-in screen the signal gets multiplied by itself and filtered by a 1kHz 18dB/Octave low-pass filter.

We measure the noise RMS squared when the LO beam is off (attachment 2), on (attachment 3), and when its power is halved by two (attachment 4). As expected, the mean value of the shot noise RMD squared is 100 times higher than that of the PD dark noise and 2 times lower when the LO power is halved.

Outlook on measuring squeezing:

A crude measurement with no averaging in our present configuration will show us squeezing if it is above    , 20 mV is the standard deviation of the RMS^2 estimated on the Moku and 545 mV was the RMS^2 at the higher power configuration. Roughly the same result, but slightly worse, is also obtained with the halved LO power.

The north table legs have been swapped.

[JC, Stephen, Radhika]

Yesterday we cleared off the north optical table in QIL to prepare it for leg replacement. Detailed photo documentation of the "before" state has been uploaded to the WB Google photo dump here.

1. Disassembly of optical setup leading to Megastat viewport

- Attachment 1 is a photo of the untouched setup. The optics were transferred to the cabinet above the solvent cabinet [Attachment 2].

2. Complete transfer of PD testing experiment to CAML

- The IR Labs dewar, vacuum pump, and most electronics for PD testing had been previously moved to CAML; but the laser diode, controllers, and ref PD had remained in QIL [Attachent 3]. These components were moved to the north table of CAML - all PD testing equipment is now located there.

4. Preparation for Megastat transport

- The vacuum pump flexible hose was detatched from Megastat [Attachment 4], and we covered the viewport + hose with foil [Attachment 5]. 

- We disconnected the pressure gauges and TIC controller, along with the RTD connections to the CTC100 [Attachments 6, 7]. These electronics were moved to the rack underneath the table.

- The plumbing from the compressor to the cryocooler was left connected.

- A cooler box under the table [Attachment 8] was moved to CAML.

- The electronics rack was moved out from under the table.

5. Megastat transport

- The engine hoist was used to lift the chamber off of its aluminum supports [Attachment 9]. 

- While moving the engine hoist, we made sure to move the compressor + plumping lines with it.

- Megastat was lowered onto a table pushed against the north wall of the lab [Attachment 10]. 

- Attachment 11 shows the final state of the north optical table.

Added power units, for two of the plots from elog 2832

[Shruti, Yehonathan, JC]
Last time, we tried to observe anti-squeezing but were not able to. Figuring that we might not be at the right conditions for parametric down conversion we set out to observe SHG in crystal.

We connect the output of the LO fiber to the 1064nm port of the waveguide and connet the 532nm to a PDA100A with 70db gain setting (~5Mohm transimpedance gain).

1064 power was increamentaly increased until we saw a signal on the PD. At first, we saw a very small signal. For 100mW of 1064 going into the waveguide fiber coupler we got only few nW of green.

After a coffee break we went back disconnected the LO fiber from the waveguide fiber coupler and connected the LO to a different 1064 fiber to measure the mating sleeve loss. We found it to be around 1 percent.

When we connected the LO fiber to the waveguide, we suddenly saturated the PD. We lower the LO power to 25mW and measured 6uW with a power meter. The crystal temperature was 62.3 deg as the specsheet. We turned off the TEC and watched the green power jump up as the temperature went through the optimal SHG temperature. With careful scanning of the crystal temperature we find the optimal temperature to be 58.7 deg with 61.3 deg set point. 
 

At that temperature the green power was measured to be 67uW, comparable with previous measurement (elog 2135).

After removing the LO fiber to measure its power again and reconnecting it, we got 140uw. This amounts to 20%/W efficiency (without coupling losses), it is consistent with the spec-sheet efficiency (they send 78mW and get 1.2mW of green power).

On a sidenote, one of our lenses is producing a significant loss, ~ F2 in the LO path, but it doesn't really affect our experiment

It couldn't go beyond a 100 kHz

why FIR instead of IIR? Normally IIR is better, but there are some few areas where FIR might be better.

[Shruti, Yehonathan]

Here's how we decided to readout our squeezed/anti-squeezed light, for now:

1. Within the FIR Filter Box on the Moku:Lab, subtract the two channels with any relative gains (the matrix in Attachment 1 can be changed)

2. Send the subtracted signal through a FIR filter roughly around 6 MHz with minimum BW (shown in Attachment 2)

3. Apply a digital gain after filtering, if needed

4. Look at the V rms of the time series obtained after filtering and possibly amplifying

In the lab today, we were slowly increasing the 532 nm light being sent to the waveguide and looking at the RMS

Later, we will try to modulate the LO phase and add lock-in detection.

(Attachment 3) By modulating the piezo on the LO path at 3 kHz with 20 Vpp, we saw intensity modulation of the light on each channel at the same frequency. The DC level of the PD output also looked modulated at the same frequency. We believe this is because the coupling into the fiber is being modulated by the PZT creating an intensity modulation.

In our diff channel A-B, looking directly at the oscilloscope, without filtering, we saw a suppressed modulation.

[Stephen, JC, Radhika]

Yesterday Stephen and I opened up the chamber and removed the 120um alumina sample. We noticed the edge of the alumina was in contact with the screw head holding down the dog clamp [Attachment 1]. Looking at photos from close-up, the sample must have shifted during cooldown or during the opening of the chamber. We decided to proceed with cooling down the 40um sample, and after estimating the magnitude of conduction in analysis we can choose to re-test the 120um sample. 

I noticed objects resting on the prepared 40um alumina sample, so in the future I will be sure to clearly mark anything fragile. The alumina had chipped around the circumference of the silicon [Attachment 2]. The parameter we care about is the surface area of the alumina, and we can set bounds on these by the silicon wafer area and the square alumina area. 

We used an aquadag-coated piece of aluminum foil to shield the sample from the shiny heater body [Attachment 3].

We mounted the sample and decided to further chip off a corner of the alumina to prevent it from coming in contact with the screw head [Attachment 4]. This also made the alumina protrusions more symmetric: the final surface area of the sample can be approximated by half the Si wafer area + half the square alumina area. 

The last cooldown showed some odd cold head RTD behavior, where it displayed as being significantly warmer than the inner shield when fully cooled. This discrepancy cannot be explained by calibration, so we opened up to take a look at the RTD element. Upon inspection the spring clamp mechanism looked in tact, and the RTD element was making contact with the cold head. I confirmed the room-temperature resistance as ~1.09 kOhm, and it succesfully displayed room temperature on the CTC100. I re-clamped the RTD, tightening the nut and increasing the spring force this time [Attachment 5]. We will continue to keep an eye on it, and if it appears wonky on the next cooldown I will change out the RTD entirely.

Today JC and I closed up the chamber and the bottom conflat of the T. Attachment 6 is a close-out photo. The vacuum pump was engaged on 2/3 at 3pm, and the cryo cooler was started at 4:15pm.

[Shruti, JC]

Shruti and I came in this morning and talked over our plan for reviewing the number from the powerpoint slide shown during Tuesday's meeting. On the beam splitter datasheet, it was shown that our fiber beam splitter has 3% loss.

To double-check this, Shruti and I began by cleaning the tips of all the fibers. Attachment #1 shows the dirtiest fiber we came across and Attachment #2 shows the fiber after we cleaned it. (The is the fiber from the Mephisto Laser)

After cleaning all the fiber tips, we checked the power from the laser. To do this, we added the fiber-tip adapter to the power meter as shown in Attachment #3. The power of the Mephisto Laser is 1.857 mW when we initially began.
 

We have roughly a 10% loss. Not good .

Also ~~~~ Laser Sign Light has been fixed.~~~~

Shruti:

Beam splitter imbalance and loss estimate in Attachment 4,5: 13%

- Can this be improved with index matching gel or being more careful with connector torque?

- Or is it intrinsic to the beamsplitter?

Attachment 1: Moku noise floor, dark noise, laser noise

- The two photodiodes seem to have different dark noise spectra, but both are 10x smaller than the expected shot noise level above 1 MHz. ChA on the Moku: RFPD3, ChB: RFPD1.

- Relaxation oscillation noise peaks above 1 MHz with the noise eater ON

Attachment 2: Diff channel PSD statistics

These are various PSDs for the channel A-B, roughly 15 measurements were taken at 2.4 mW LO and around 10 at 1.2 mW LO power
- The gray bunch of curves were measured when the power incident on each PD was around 1.2 mW (LO power=2.4 mW)

- The blue bunch of curves were for roughly half the above LO power

- The dashed lines were eyeballed and differ by sqrt(2) in the flat region (higher frequencies) as expected for shot noise

Attachment 3: Expected shot noise vs measured (initial attempt)

This inspired the measurements in Attachment 2.

- At 1.1 mW, we accidentally did not save the 'HighRes' data and only had 1024 instead of 16k so had a lower high frequency limit.

- Without really changing anything, we remeasured the channels A and B. This difference was higher than earlier. Then we found that the power was 1.2 mW -- a little higher but not enough to explain the discrepancy

- The shaded region corresponds to one std devation (as outputed by the welch PSD code). The expected shot noise for 1.2 mW incident on each PD is within this bound but lower than what was measured by sqrt(2)

(02-Feb-23) Attachment 4: discrepancy between expected and measured shot noise

The formula for shot noise used: where QE~ 97%, TIG ~ 40 V/ mA, R~ 1 mA/mW. We convinced ourselves that QE, R, TIG are out of the square root because we're measuring the photonic shot noise and not the electronic one which is almost absent in the RF coupled port of the PD because the current here is much smaller.

The PSDs plotted here were all one-sided PSD. Taking a two-sided PSD would reduce the levels by sqrt(2) and resolve the discrepancy here, but the formula above is for the one-sided case right?

[JC, Shruti, Yehonathan]

- Found 3 NF 1811s (Attachment 1)  and beginning to characterize them

- Measured noise floors of  Moku:Lab and Moku:Go. Data and Script (plotted in Attachment 2)

The power connector on the unlabeled RFPD was found to be busted.

We measure the DC responses of RFPD1/3:

Power on RFPD1 was 1.11mW and the measured voltage at the DC port was 1.07V

Power on RFPD3 was 1.03mW and the measured voltage at the DC port was 1.02V

Assuming TI gain of 1V/1mA (according to the specs) we get QEs of 96% for RFPD1 and 99% on RFPD3.

On Friday 1/20 we prepared a 40um alumina sample bonded to a Si wafer, using silver epoxy. We mixed 2.7g of substance A with 2.6g of substance B (using tools from B - adding about a gram). Out of this mixture, 1.7g was deposited onto the wafer to form the bond. (This amount should be safely within the recommended bond thickness range of 1-4 mils.) We used a much heavier weight on the sample this time, to help the epoxy distribute evenly.

Attachments 1 and 2 show the bond area after application and a few minutes of weighting, and after 3 days of curing with the weight. The epoxy spread significantly more than on the last sample, although it still didn't quite make it to the wafer edge. Note that a corner of the thin alumina strip snapped at some point in the process, as seen in Attachment 3. The approximate area lost is 52.79 mm^2, or ~1% of the alumina strip area.

We also prepared an epoxy-only sample by dispensing 1.3g onto a Si wafer. I used the dull edge of a razor to spread out the epoxy into an even layer [Attachment 4].

Today I varnished the previously prepared RTD (L) to the prepared sample [Attachment 5]. I prepared a new RTD (I) for the epoxy-only sample, to be used in a future cooldown.

I started venting the chamber today, and tomorrow we can proceed with swapping out the samples and closing up again.

I realized our original calculation for epoxy mass was overestimated by a factor of 4 (we used the wafer diameter instead of radius in the volume calculation). This means that for the max epoxy layer thickness, we would need only 1.55 g. We applied a 2.3g blob to this sample, which should have been more than enough. My guess is the mixed epoxy was sitting out for a while before application, and it started to cure before it could fully spread out. We will take care to mix and apply the epoxy as quickly as possible to avoid this. Also we can use a larger weight that covers the entire wafer area. I'll stick to mixing 5 g of epoxy moving forward, accounting for the amount irretrievable from the cup.

Note that this epoxy does seem to contain filler that would set a minimum layer thickness. The manual does not indicate this minimum, but we can plan to measure the bond thickness of our samples.

On Tuesday we checked on the previously epoxied sample (Si wafer + 120um alumina strip - Attachment 1). The bond area did not spread much more since we last checked on it on Friday. For the next bonding attempt, we will increase the amount of epoxy to 8g and aim for better centering of the bond area.

We prepared and installed a new resistive heater to aid in warming up the chamber [Attachment 2]. The heater was clamped down to the coldplate with an indium gasket in between the surfaces [Attachment 3]. Then we re-covered the heater with the pulled-back aquadag foil [Attachment 4] and left for the day.

On Wednesday we soldered a new RTD (PT1000 - K) for the sample. We added crimp pins to the RTD leads to plug-and-play into the existing connections, and then soldered the RTD to the top of the alumina surface [Attachment 5]. We mounted the sample and closed up [Attachment 6]. 

The vacuum pump was turned on at 3pm on 1/18; cryocooler switched on at 3:30pm.

[Stephen, Radhika] 

[pics to come]

Today we bonded a 120-um alumina strip to an undoped silicon wafer, using EP21TDCS-LO silver epoxy [Attachment 1]. The user guide specified an optimal layer thickness between 1-4 mils (0.0254-0.1016 mm). The max thickness called for ~5 grams of epoxy, factoring in the wafer area and epoxy density. We used a scale to measure out 2.5 grams each of parts A and B in a cup (we had ordered 10 g each).

We mixed the paste and applied it as a blob to the center of the Si wafer. (In reality, we managed to scoop out 2.3 g from the cup.) We placed the alumina strip on top and put weights on the sample. Attachments 2-4 show the dispersion of the epoxy immediately after applying, and after 2 increments of 5-10 minutes. We are leaving the sample to cure over the weekend, as per the room-temperature curing schedule found in Attachment 1.

On Monday we plan to mount the sample in the chamber and start a cooldown.

How to move the large engine hoist through the narrow door

See http://nodus.ligo.caltech.edu:8080/Mariner/122

Cryocooler turned off at 1:30pm.

With the heavy rain and flooding basement level, there was a small amount of water instrusion into QIL around the base of the desk - seems like only the floor was affected. I mopped that up.

Also, there are a lot of ants in the lab. The trails loop through the entire lab. They're also in the ... wait for it ... antechamber. I put down some ant bait in a couple of places. 

[Stephen, Radhika]

We opened up Megastat, and luckily there were no visible signs of water or ice in the chamber [Attachment 1]. See the Google photos dump for more detailed photo accounting.

However the contact between the Si wafer and alumina strip did not hold up during pumpdown/cooldown [Attachment 2, 3]. The varnish may not have been completely set prior to closing up, and the air pockets present in the initial contact were forced out when we pumped down. There were also cracks in the alumina. The thermally conductive epoxy has arrived, so at this point it may be worth going forward with an epoxy bond. Next week I'll apply it to a new Si wafer and give it the recommended room-temp time to cure.

Since preparing a new sample won't take place until next week, we decided to proceed with cooling down the existing sample. It will be interesting to see how much conduction we get to the alumina strip given the uneven contact area. Data will be collected over the weekend, and the chamber will be available for the new epoxied sample once cured.

Vacuum pump turned on at 4pm, cryocooler turned on at 4:40pm.

[Stephen, Radhika]

We wiped/scraped the floor to clean up from the last flooding incident.

seems iffy to have a cryo-cooler running without vacuum. I guess we should make some kind of interlock to prevent this, otherwise we'll end up with a giant ice cube in Megastat. I am guessing at 3 Torr we probably already have ice build up on the cold surfaces.

Today I checked and there is no flooding in QIL from recent rain.

The unexpected power outage turned off the vacuum pump for Megastat, and the pressure was back up to 3.9 torr. Meanwhile the cryocooler was still on; the steady-state temperature inside the chamber was ~220 K. This disrupted the current cooldown, so I'll have to restart it. I've switched off the cryocooler and once it's reached room temperature I'll restart the pumpdown/cooldown sequence.

Thanks Chris! Did you run startc4iop in /opt/rtcds/caltech/c4/scripts/?

CW: Yes, that's right. Perhaps after this cooldown we can find some time to upgrade the system to the new RCG that starts things automatically on reboot (like at the 40m).

I think that's a good idea - I'll get started on it.

The framebuilder is now running again. It relies on the c4iop front end model for timing, and that was not running either. (Maybe there was a power outage and the system wasn't brought back up afterward.)

The cryo lab has a gitlab CI script that complains to us on mattermost when its DAQ stops working. The QIL can probably use something like that, too.

[UPDATE]

I realized on Sunday that the frame builder was not storing channel data. I turned off the cryocooler so that the cooldown could be restarted once the past temperature data could be extracted. 

In the meantime, I wrote a python script to query the EPICS channels associated with the temperatures of the workpiece, inner shield, outer shield, cold head, and the heater power. The script writes out the data to a .txt file. Once Megastat reaches room temperature, I'll have someone switch on the cryocooler to start the cooldown. Until then I'm working to debug the cds issue, but we can at least use the new script to record the data if it comes down to it.

The pipe is currently being replaced. Should be finished up by the end of the day.