Even with loop-shaping in Moku:Pro, passing the control signal through a unity-gain Moku:Go buffer resulted in oscillation at 10s of kHz.
I at first suspected this is due to extra input voltage noise on Moku:Go's input ports. The Moku:Pro does some smart blending of signals between its low-frequency and high-frequency ADC, which suppresses input voltage noise below ~10 kHz. Attached is a voltage spectral density for
The input voltage noise of Moku:Pro is about 4x less than that of Moku:Go between 100 Hz and ~10 kHz, mostly due to the ADC blending (similar input voltage noise curve available on liquid instruments' webpage about ADC blending).
If the loop instability were due to input voltage noise on Moku:Go, I'd expect to be able to mitigate this by boosting the signal in Moku:Pro and reducing gain in Moku:Go... but this did not significantly change the residual noise or oscillation.
Now I suspect the ADC/DAC handoff on a second device is just introducing too much loop gain. I think our phase margin was only at best 45-90 degrees with maybe 65 kHz UGF (can dig through the elog). A delay of ~4 us would completely eliminate the phase margin at the UGF. I did find adding some derivative stabilized the loop a bit, but couldn't completely eliminate oscillation. Next I'll check the delay when passing a signal through a Moku:Go buffer.
We want to inject signal both at the input of our amplifier and at the PDH (and maybe PLL) error points to null the phase modulation in cavity reflection and avoid unnecessarily driving the pump frequency (S laser current).
Previously, we used 2x lock-in instruments and 2x PID controllers in our Moku:Pro to achieve lock. However, this configuration does not let us modulate the PDH error point.
Yesterday, I tried using 2x lock-in instruments in Moku:Pro to generate PDH and PLL error signals, then routing the signals to a Moku:Go for loop shaping and feedforward. Unfortunately, I wasn't able to lock either the PDH or PLL when passing the error signal between Moku:Pro and Moku:Go. I got both loops locking independently with the error signal shaped internal to the Moku:Pro, but the same combination of filters split between Moku:Pro and Moku:Go could not acquire lock. It could be that some combination of DC offset drift between the Moku boxes and input ADC noise is to blame, but I wasn't able to resolve the issue by shifting around where gain is applied between Pro and Go.
Today I'll try handling loop-shaping in Moku:Pro, and just do feedforward with the Moku:Go. This way, the low frequency signal is boosted above 1/f noise in the ADC.
We retrieved four wooden blocks from the 40m to place under the vacuum chamber during bake out.
We then vented the chamber with dry nitrogen. Because we don't have a second up-to-air valve in the system, we loosened the KF flange on valve 1 to avoid overpressuring the chamber. While the chamber was open, we flowed nitrogen at 2.5 psi at the regulator. We also cleared out the area around the chamber, and disconnected the vacuum gauge from power.
We used the hoist to lift the chamber body and move it away from the bottom plate. We placed the bottom plate on wooden blocks wrapped in UHV foil, and removed the styrofoam and remaining wrapping that had been holding the plates.
We wiped the bottom o-ring and plate with 100% IPA. We photographed a large indentation in the bottom o-ring, which we suspect could be responsible for our leak. We also tried to remove the o-ring from the groove using teflon forceps applied at the notch in the o-ring groove, but were unable to do so. The o-ring still has vacuum grease inside the groove.
We lowered the chamber body back onto the bottom plate. We reattached the KF flange on valve 1 and gave it a new o-ring. After waiting ~20 minutes, we recalibrated the vacuum gauge's 'atmosphere' point (as it was reading only 100 torr or so at atmospher). After being disconnected from power for about an hour, then being allowed to equilibrate and calibrate at atmosphere, the gauge is once again giving 'lower noise' readings. Good to know.
The pumpdown is proceeding very slowly, and I'm restarting the roughing pump several times as it times out (as usual). The turbo kicked on after about an hour of pumping, so I changed the roughing cycle timer to 90 minutes to avoid having to repeatedly restart roughing. I suspect a leak, once the turbo spun up fully the vacuum pressure exponentially decayed to >10 utorr.
I did a rate-of-rise test on the vacuum now that we've swapped CF flanges. The vacuum pressure did stabilize at a level below 1e-4 torr this time, and squinting at the leak rate plot you can make out a downward trend... but I didn't even add a trend line because the gauge has become much noisier since I last ran this test. I don't know why the gauge is so noisy, but it has up to 300+% deviations upwards in pressure that last for less than a minute before going back to a baseline under 1e-4 torr (even with the pump valved off, so there shouldn't be any gas sinks in the system).
Since the rate-of-rise test is promising, I think we should bake out the chamber over the weekend instead of venting to clean the o-rings. We need to vent and open the chamber anyway to put wooden blocks under the chamber for baking, so we'll just wipe the bottom o-ring today.
Shruti and I finished replacing gaskets and bolts on the remaining 3 CF flanges. The subsequent pumpdown is still proceeding slowly enough that I expect a leak (despite not yet performing a rate-of-rise test). The next step is to clean and reseat the large viton o-rings on the top and bottom of the chamber.
I received compressed dry nitrogen this afternoon. I connnected the regulator we had been using, and valved off the line going to the cryo cavs experiment (there was already a valve at the cryo cavs table). Despite the low-pressure-side gauge on the regulator having a 100 psi range, the regulator can only supply ~10 psi. This is awkward because I spec'ed the calibrated leak to fill at ~30 psi; it will still leak, but the rate will be lower. We could decide to swap regulators if it's taking too long to vent through the leak (I've seen the regulator on our He line supply 100 psi).
After Maty gave us a tutorial on CF flanges, JC and I
One of the CF flanges we removed had at least one bolt that galled well before it was fully seated in the plate nut; others had minor galling on some bolts. We put all the steel bolts, washers, and plate nuts in a bag labelled "BAD," and can either throw out the entire batch or just the ones with noticeable galling.
In the afternoon, we installed 3 out of 4 CF35 flanges and 2 out of 4 CF125 flanges. We'll finish tightening the remaining 3 flanges tomorrow.
Aaron, Madeline, and I met up in CRYO to discuss what would be the best move to check the PSOMA chamber. Attachment 1 has the plans for the upcoming week and what we plan to do.
Here is a notebook that takes into account the change in the interference condition and phase-lock phase offset to estimate how much AM and PM we expect.
(More details coming soon...)
We were supposed to receive dry, compressed N2 in cryo lab today... but it never arrived. I've asked JC to follow up with airgas (or whoever was supposed to deliver).
In the meantime, I found fittings to adapt the 1/4" VCR male flange on our calibrated leak to the 3/8" tubing that carries nitrogen to the PSOMA experiment (1/4" male VCR -> 1/4" female VCR coupling body -> 1/4" VCR to swagelok tubing adapter -> 1/4" gas line filter -> 1/4"-to-3/8" right angle connector). See photo for assembly. I borrowed the VCR connectors from Hutzler lab, and identified swagelok SS-4-WVCR-6-400-SC11 as the adapter we want to purchase.
I don't think it's necessary to have to do the whole matrix calculation again for breaking up blocks with no cross-coupling because each block is just a product of sub-blocks...
To make this proposal more explicit... since it's easy to measure , , and the error signals in loop without additional calibration, we could estimate by exciting at and measuring
I'm assuming we inject the excitation in such a way that can be identified with their respective nodes . This seems like a reasonable measurement between ~1 kHz and ~100 kHz, but might be impossible at lower frequency near the cantilever resonance where it's hard to measure the loop transfer functions. I included the equivalent estimate based on , but in practice we would measure the ratio of error signals and would need to add a node inside of to derive the appropriate transfer matrix.
We could also excite at or at the error points. Breaking into the plant and controller and labelling the new nodes corresponding to error signals , we have a new open loop transfer matrix
The matrix is identical to the matrix above in the 6x6 upper-left quadrant, except with all references to . The rest of the matrix is
It's not easy to FM the laser at low frequency using an EOM (requires phase modulation proportional to ), so let's assume we can't directly excite the laser frequency either before or after the beamsplitter. Instead, we can only sum into the control and error signals at . Therefore, we only have access to columns 2, 5, 7, and 8 of . Likewise, we can directly measure only the control and error signals either before or after the injection point. This limits us to accessing rows 2, 5, 7, and 8 of .
(Note that we could consider measuring at node 3 using the driver's current mon, but since this is related to the control signal by a trivial transfer function I'm leaving it out).
This gives us a list of 4x4+1=17 transfer functions in the full system... the extra 1 is due to always being able to measure with the transfer function measured just after the injection point. The transfer functions in the table below are normalized by , and one could measure any ratio of these transfer functions just as easily (assuming both are coherent at the relevant frequencies).
I'm having some trouble discerning which of these transfer functions are coherent at what frequencies, and I think we should just start trying them out in-lab.
**Note: I reproduced in the 6x6 case, and I think there are a few minor typos above. Since there are several, someone should check again for final resolution, but there isn't a pattern to the differences so I suspect simple typos in the unwieldy matrix
The PSOMA chamber reached ~9e-5 torr overnight. I'm trying to identify and correct the leaks in the PSOMA chamber. Using the rate-of-rise test, I measured an O(1e-5 torr*L/s) source of gas in the chamber, based on ~1000 L chamber volume. The test seems to indicate an actual leak, because the rate of rise is steady or increasing after ~1 hour with the pump valved off. Attachment 1 shows the rate-of-rise test and subsequent pumpdown. Attachment 2 shows the leak rate during the test.
After replacing the CF flange I suspected of being leaky based on methanol spritz, the vacuum quickly pumped down to ~20e-5 torr. I'm leaving the system heated to ~80 C overnight to bake out water.
I replaced the gasket of the CF flange with broken bolt.
After a very slow roughing cycle (~1 hour with roughing pump only), the vacuum pressure reached a sufficiently low level that the turbo kicked on. Ultimately the vacuum pressure reached 25e-6 torr. (attachment 1 is the pumpdown curve)
The rate-of-rise test (valve off the pump only and watch the rate of pressure increase) suggests the presence of a real leak, not simply outgassing, because the rate of pressure increase remains approximately constant over time. The test is visible in the attachment where the pressure rises after bottoming out, then falls again to ~25 utorr. I stopped the test somewhat early to avoid the vacuum gauge switching over to Pirani readout, and to avoid needing to spin down the turbo before pumping down again.
I sprayed methanol on all flanges and wasn't able to identify a single limiting leak. However, there are some candidate CF flanges:
I'm leaving the system pumping overnight to see if it reaches a lower ultimate pressure.
Here are some comments on the proposed measurements. Intended to let me understand fully, no fault assumed.
Requires same calibration transfer functions as used in (A) to calibrate (6) and (1). I don't understand why the modulation at the N EOM should be converted to voltage at current driver input. Doesn't 6 correspond to frequency of the N laser, so we want to refer to N laser frequency?
What are and ?
Since the current driver is a sub-block of , wouldn't the equations instead apply to transfer functions when monitoring the control signal for either PLL or PDH? I may be misunderstanding whether and appear before or after the EOM.
Just as an aside, the only alternative I know to measuring the various loop transfer functions above is to pick off the beam just before the cavity and directly measure amplitude and phase fluctuations on the pickoff using BHD (or whatever). I'm not sure where we'd get the clean (unaffected by unwanted loop couplings) LO for such a pickoff measurement.
Since it seems pretty straightforward to measure all of the blocks independently, maybe the following computationally complicated measurement would actually be experimentally simpler:
We could do some meta-analysis of how much information each monitor point contributes (as a function of frequency), and eventually simplify the measurement. But since we wouldn't need to worry about calibrating and comparing multiple excitation points, this might minimize calibration uncertainty.
This could also be expanded to directly measure the optomechanical transfer function, as in the other log entries in this thread.
The phase offset in the PLL sets both the signal quadrature (relative to the pump) and also the interference condition between the two lasers in the current setup. We always apply phase modulation at the north EOM...
Assume, , is when the modulation is AM wrt the pump.
- By measuring the DC port of the 1811 REFL PD directly, the AM can be measured (I think?)
- PM can be measured in the PDH error signal
1) The ratio of PM measured in the PDH error signal when AM is injected to PM measured when PM is injected
[ Considering the change in power with the change in interference condition ]
1) Residual amplitude modulation
In principle, when AM is injected, and gain=0, this signal should not be detectable in the PDH error signal unless there is some residual amplitude modulation in the north EOM.
The two loops with feedforward that currently make PSOMA controls using two lasers are shown in Attachment 1.
The south laser is the pump and is locked to the cavity using PDH locking. The north laser is the probe/signal simulator and is locked to the south laser with a phase-locked loop. In practice, the beat between the two lasers is locked to a local oscillator at (150 or 300 MHz was used) and an additional modulation at is applied, one of whose sidebands is resonant with the cavity. The signal is an audiofrequency sideband of the cavity resonance. There is an additional feedforward from the PDH to the PLL to allow the phaselock to follow the PDH loop when the cavity is very noisy.
The two lasers are mixed on a 50/50 BS and the PLL uses the beat in the AS port and symmetric port with (mostly) constructively interfering light is sent to the cavity. Since the frequency noise measured around the cavity sees the north laser/path frequency noise only when the additional modulation is applied, the two loops' error signals depend on the north and south paths as:
where is the frequency noise of the south laser and is that of the north laser.
is shown in the bottom-center of Attachments 1 and 2.
is the error signal seen by the PDH loop (in frequency units). For the PLL, the error signal is proportional to in the Laplace domain.
Also, are the control signals of the PDH and PLL loops respectively.
For the nodes as numbered in Attachment 2, the open-loop transfer matrix is
[convention: row index-> column index for both S and T]
In closed loop, the full set of relations are
With the feedforward turned off, , while both loops are on, this can be measured as .
Experimentally, this would mean measuring two transfer functions, the first by modulating the north EOM (the secondary AF sideband) and measuring the PDH error signal, and the second, by modulating the south EOM and measuring the PLL error signal in frequency units.
is just the constant factor (=0.5) specified in the Moku cross-coupling matrix, but can be easily measured between the PDH control signal and the North laser HF MON (by dividing out the voltage->current conversion of the north current driver).
Since both loops can be individually locked at the same servo settings, the A's and B's can be measured separately in each loop.
The full/overall loop-suppression = , where
This factor can be measured directly (with or without feedforward) , once is known.
where 6->1 is the measured by modulating the north EOM (but converting it to voltage units at the current driver input) and measuring the PDH error signal.
When a signal, , is injected as the secondary audio sideband in the north EOM and both laser driver current monitors are measured, the following transfer functions are obtained:
For the south monitor,
since the south EOM does not change affect the frequency between the south laser and the cavity. If is calibrated to units of frequency of the north laser, .
For the north monitor,
is the transfer function of the north laser in units of Hz/A.
Before pointing out all faults and reasons why these measurements are not immediately possible in the lab, please consider
1. These are what directly follow from the calculations as what occurred to me as the easiest way to measure those quantities theoretically when I couldn’t think of a more practical method. For e.g., X=0 cannot be possible while acquiring lock, but maybe it could be slowly decreased when the setup is stabilized. Please let me know if you do think of a better way to measure k.
2. Of course, from where we actually end up injecting the signal there may be an additional calibration of the drivers, etc that may be necessary which I can easily add. But I have taken into account the units whenever I thought was necessary.
Spritzing methanol around the CF flanges securing valves 2 and 3 reduced the vacuum pressure by ~3 ntorr (10%). I'll keep an eye on these flanges as we test the rest of the chamber.
I proceeded to vent the chamber, roughly following the steps to reach chamber isolated, vent ready, and vented in the elog above... with the exception that valve 2 remained closed at all times and I abridged steps related to venting with dry nitrogen (I'm just using room air at 1 atm).
I opened the chamber lid and wiped the o-ring and mating surface with a dry lint-free cloth. The o-ring already has some vacuum grease applied. Then, I pumped down the chamber.
I turned off the pump and re-tightened the CF flanges. I had been tightening the flanges on the chamber to a torque specification, but this time I tightened until the metal surfaces of the flange completely touched.
On one of the CF35 flanges, the bolt broke off in the nut desite a visible gap present between the CF surfaces. Maybe I need to use a lubricant with these bolts? Since I now need to garb up and completely open at least one flange, I'm going to push this leak hunt to next time.
Attachment 1 shows the pressure during the above. The linear plot focuses on the leak rate with valves 1, 2, and 3 closed.
Shruti and I checked out the ergonomics of the PSOMA cryostat and table. We determined that we should acquire new legs for the optics table that are 8.25-8.5" shorter than the current legs. The distance from the floor to the bottom of the rail around the enclosure is 28.25", and the table is 12" thick. The current legs are 22.25", and have a protrusion for the floating attachment to the table which ends ~24" from the ground.
We want the new legs to attach to the bottom of the optics table (28.25"-12") = 16.25" from the ground. Up to ~1" longer legs would be fine.
This would put the top surface of the optics table ~28.25" from the ground. This height is comfortable for both Shruti and I to work on from a standing position with no stool or bending over, and it would be comfortable to reach over the 19.5" tall cryostat especially with assistance from a stool or short platform.
The PSOMA vacuum chamber is closed. The attachments describe the current system.
It would be helpful if someone could review and discuss these proposed procedures for operating this vacuum system.
We could consider adding a second gauge on the opposite side of the chamber, in which case valve 2 could be closed here instead of 1. This would allow continuous monitoring of both the chamber pressure and pressure in the line between the pump and chamber.
I should make a state diagram of these processes.
I'm having trouble with nds...
I tried restarting the epics services on cominaux, restarting rtcds using the rtreset script on cymac1, and rebooting all of cominaux, cymac1, and spirou workstation. systemctl on cominaux reports the usually services are running on that machine, and I don't know what to do based on the status screens on sitemap (attached).
How to move the large engine hoist through the narrow door
I continued closing the PSOMA chamber, and will post a sketch of the vacuum system comorrow when all the bolts are tightened.
We saw some promising signals, but need to say definitively what is happening in our amplifier cavity. We want to observe the dynamics of our plant (cavity), and preferably while minimally resorting to detailed modeling of the coupled PDH and PLL loops. So what measurements should we make? I'm writing this up following discussion with Chris to help clarify for myself.
Ignoring intracavity noises, the fields reflected from the cavity are
Where the quadratures are amplitude (1) and phase (2), is a frequency dependent phase due to the cavity pole, and is the ponderomotive coupling. To full characterize the optomechanical transfer function, we want to measure the transfer function from excitation to each of . I think measuring all of these transfer functions using the same excitation point lets us characterize the cavity transfer function no matter how complicated our controller is (even if the controller itself involves many coupled loops), because the above equation must hold in steady state. We could even measure them simultaneously with a single excitation.
The DC MON on REFL is sensitive to , while the cavity pole lets us measure after demodulating at the PDH sideband frequency (the usual PDH technique). Likewise, the BEAT PD is sensitive around DC to , and is sensitive to by demodulating at the PDH sideband frequency. The last claim is justified because the acoustic signal sidebands around the probe are not (directly) present on the pump laser or its RF sidebands.
Formally, the Beat PD sees a slightly different field than the cavity, but the transfer function measurements might be mappable between and . As long as and either the pump or probe dominates in the amplitude quadrature, measuring phase fluctuations at BEAT PD is equivalent to measuring the fields just before the cavity.
However, in practice I'm not sure we can monitor only BEAT PD. The PDH loop will inject any signal in back to by driving the pump frequency. Therefore, we probably cannot approximate , especially in the limit of amplitude-quadrature signal injection (meaning direct coupling of signal to is small) with high ponderomotive gain (meaning can be much larger than ); furthermore, the PLL drives , so we would expect . In other words, we might be able to use the BEAT PD to estimate fields entering the cavity, but only if we measure some transfer functions of the coupled PDH-PLL loop.
We may need to pick off some light just before the cavity instead (easy enough), or fully characterize the coupled control loops (probably worth it to avoid loss in an eventual sub-SQL demonstration). We also would need to measure the phase of the pickoff using BHD, since we want to know the contributions to of both the pump and the probe and suspect that the pump contribution may dominate. I'm not sure where to pick off a BHD LO -- both pump and probe current drivers are controllers in either the PDH or PLL, so neither laser has a clean carrier free from any injected signal.
Alternatively, can we excite below the PLL UGF then assume and simply measure at the pickoff before the cavity and demodulating at PDH sideband frequency?
To summarize a few possible approaches
In either (1) or (2) above, we might lift the assumptions about pump and probe fields by making appropriate (and unkown to me) measurements of the PDH-PLL transfer function.
I closed the 6.25" CF flanges on the PSOMA chamber with blanks, though I didn't completely seal the copper gaskets (I left two bolts finger tight on each flange).
On three of the flanges, I used Kapton tape to secure the outer edge of the copper gasket to the flange so I could easily install the blank without dropping the gasket. I wasn't satisfied with the result since I suspect at least some of flanges have a small piece of Kapton tape between the gasket and the groove. The tape is well outside the knife edge, but I'm concerned it will prevent the copper from fully forming into the groove. I'm going to remove and reseat these flanges.
I also started re-shelving some optics from the cryo-cantilevers experiment to make room for the chamber on the PSOMA table.
I'm trying to understand the changes in DC power on BEAT MON and TRANS MON channels when the resonant probe field is tuned to constructively or destructively interfere with the pump. In particular, the change in DC power are larger than I would expect based on the sideband amplitude in the beat note spectrum.
Here are some observations. In all cases, I assume that minimizing power on BEAT mon (maximizing power on TRANS mon) implies the pump and probe are in-phase.
The beat note at the PLL LO frequency is proportional to , while the probe's sideband is proportional to . Therefore, the ratio of field amplitudes for the probe's carrier and first sideband is
When one of the sidebands is made homodyne with the pump by driving the probe EOM at the PLL LO frequency, the probe's sideband field interferes with the pump with some detuning set by the relative phase () of the PLL LO and EOM drive tone. Beat mon voltage is proportional to the incident power at DC, which is
Since the pump and probe lasers have approximately the same power, I'd expect the sideband to cause up to a ~1/250 fractional change in power at the pump carrier frequency. Even doing the calculation more carefully finds that the ratio of Beat mon levels when the pump and probe are constructively / destructively interfering should be
However, I instead measured
I still can't explain this discrepancy. When I made an analogous measurement with the cavity locked and comparing TRANS mon levels instead of Beat mon, I saw a similar discrepancy. I was driving the probe EOM with -10 dBm, and I saw ~10% difference between \phi=0 and \phi=\pi. The measurement above confirmed for me that the scaling is approximately linear and not a result of unexpected coupling between the PLL and PDH loops. It's great that I can see a lot of interference between my pump and probe, but disconcerting that I can't explain the effect. Have I missed something in this calculation or measurement?
While making this measurement I was able to drive the probe EOM with up to 3 dBm RF power and maintain PLL lock and even (noisy) PDH lock. I also drove the probe EOM at the PLL LO frequency detuned by <kHz and saw low frequency beating on Beat mon. If the PDH can remain locked with a frequency-detuned probe, this might let us make a lock-in measurement of the gain rather than controlling the pump-probe phase.
Chris found my error -- electronic voltage is proportional to optical power, so beat note power on the spectrum analyzer is proportional to not . Everything above is consistent with this modification.
I saw a signature of amplification, but I can't explain everything I observed so don't want to attribute it definitively to ponderomotive effects.
Attachments 1 and 2 show the PDH control signal in the top display (measured at pump's HF mon) and the MON of the beat note PD (amplitude fluctuations into the cavity) on the bottom display. In attachment 1, I've tuned the pump-probe phase to minimize the signal in the amplitude quadrature entering the cavity. Attachment 2 adds 90 degrees to the pump-probe phase, so the signal should be maximally in the amplitude quadrature (up to drifts during the measurent time).
When the signal is in the phase quadrature, it couples directly to the "output" of the amplifier since the PDH servo compensates phase fluctuations on the pump laser. When the signal is in the amplitude quadrature, it should only couple to the "output" of the amplifier if the cavity reflection transfer function rotates amplitude into phase (due to ponderomotive effects, or possibly due to the cavity sitting off-resonance).
The attachments show that the signal peak at the amplifier output is 2 dB higher when injected in the amplitude quadrature compared to the phase quadrature.
It's a bit premature to claim this is a ponderomotive effect because there are a number of unexplained features...
I was remiss in my eloging, but here's the update from last week...
tldr: I saw my AM signal before the amplifier above the noise floor, but wasn't able to see the optomechanically-induced PM signal after the amplifier. Then, I messed around with the PLL loop shape longer than I should.
I removed the amplitude modulator from the pump path to get us back to ~1 W circulating pump power. I also changed the PLL LO frequency (offset between the pump and signal lasers) to 150 MHz, the bandwidth of our fiber EOM. The loop seemed to work fine at 300 MHz offset, but if the EOM response is falling off there could be some unnecessary phase delay, and 150 MHz seems far enough from the cavity resonance and PDH sidebands.
I tried two signal injection schemes. In both cases, the Moku and Marconi are locked to the same 10 MHz timing signal.
In both cases, the signal can appear as either PM or AM on the pump according to the relative phase of the LOs used for the PLL and for driving the EOM.
I didn't actually notice that one of these methods generated more AM around the pump, but I didn't make a careful enough comparison. I expected that modulating the probe after the PLL would lead to the PLL suppressing the signal. Will have to re-measure and re-think this.
To find out, I'm looking at REFL and BEAT MONs on the SR785 spectrum analyzer. I was able to produce AM sidebands above the noise floor between 40 Hz and 10 kHz, and could change the height of these AM sidebands by roughly tuning the relative phase of pump and homodyne probe.
With both the PLL and PDH servos locked, I can change the transmitted light level by changing the relative phase of the pump and homodyne probe. This makes sense: when the pump is in-phase with the pump, it just looks like more carrier power to the cavity. However, the magnitude of the changes in transmission power is larger than I expect (O(10%) changes in transmitted power, but I think the homodyne probe field amplitude is only 1% that of the pump).
Assuming the signal is AM, it would only appear in the PDH loop after being rotated to PM by pushing on the cavity mirror. Below the UGF of the PDH servo, I would expect to see a signal peak in the PDH control spectrum as the loop compensates. I couldn't find this peak in the pump laser's HF mon when the cavity was locked. I tried integrating the spectrum for O(5-10 minutes), as well as using a lock-in amplifier with similar integration times.
Messed around with the PLL, and tried to understand the open questions.
I tuned up the PLL servo for several hours by measuring the PLL open loop transfer function then adjusting a two-stage PID filter (with up to 4 integrators and 2 derivatives total) a) increase gain below the UGF in the region with significant residual error, b) increase the UGF frequency, c) increase the phase margin at the UGF, and d) increase the gain margin where the phase margin vanishes. I made and unmade many marginal improvements, and found a few heuristics about our PLL.
We can push the UGF above 150 kHz and maintain lock, but the intrinsic phase delays in the loop limit the stability of this loop. For example, I could get a marginally stable loop with UGF above 200 kHz by adding a second derivative, but since this reduces the gain rolloff above the UGF I still get gain peaking wherever the phase margin vanishes.
There is some
I checked the general function of the intensity modulator. The internal PD port responds to input laser power (reads ~10 mV with S laser off, ~865 mV with S laser on). I'm using the beat note PD to monitor transmitted laser power, and checked that the changing the DC bias level with a function generator can change the DC output of the beat note PD from 0 V to ~325 mV.
I created a couple of new cds filters to monitor and control the intensity modulator
I make'd and installed the new x1oma model, and ran rtreset to restart the front ends. Rather than wait for Chris' cronjob to run this week, I ran "./mdl2adl.sh" from /opt/rtcds/tst/x1/scripts/mdl2adl directory on cominaux to generate updated medm screens. Note that running the script only on the x1oma model (as in "./mdl2adl.sh x1oma") threw an error.
I used a function generator to sweep the DC bias level with a +- 10 V triangle wave, and monitored the DC power level on the beat note PD and internal PD of the modulator (ndscope traces in attachment 1, all y-axis units in counts). The internal PD is supposed to monitor a 90:10 or 99:1 tap fiber on the transmission side, so I'm surprised to see a nonlinear output.
I've left the system under vacuum for about a week. On early Sunday morning ~4am (nobody was in the lab), the pressure apparently dropped to 1e-7 torr for several minutes then increased to ~600 utorr. Each step change was in about a minute. The pump is still operating as normal, and I wasn't able to find a leak by spritzing methanol around the flanges. The vacuum pressure over the last 7 days is attached.
I added the MXAN-LN-10 amplitude modulator to the pump (S laser) fiber path, just after its EOM. I do see the expected ~3dB of insertion loss due to the AOM (though I should measure this more carefully). I was able to lock the cavity with no trouble with the new modulator in the path (no modulation applied). I'll measure the modulator's transfer functions and Mach-Zehnder DC tuning next time.
I connected the RF ports of the modulator to the front panel with LM-200 SMA cables. While I had the box open, I swapped out some lossier SMA cables for LM-200 cables: between the beat note PD RF port and the front panel, and between the S EOM RF port and the front panel. The cables were not of the same length, so I expect the PDH phase will have changed.
Due to space constraints, I mounted the amplitude modulator on top of the S EOM, separated by the breakout breadboard and a heat sink as shown in attached photos. I used rubber to soften the clamp, and placed another small heat sink between the rubber and the amplitude modulator. I didn't use any conductive interface between the heat sinks and the modulators.
1. Fiber setup before modification
2. fiber setup after modification
4. heat sink
5. breakout PCB is supported not dangling
I'd like to add the MXAN-LN-10 fiber-based amplitude modulator to our pump path to allow us to amplitude modulate our pump independent of our PLL-and-probe-laser scheme for signal injection. Since we aren't yet using a Mach-Zehnder to reject pump RIN, we won't really be taking a hit from the perspective of SNR, and it would be nice to test the optomechanical response of the cavity independent of the PLL and transfer functions involving the probe laser.
Zach had been using wire clips to tune the DC electrodes of our MXAN. Since this is a semi-permanent installation, I made a PCB board to connect the electrode pins to SMA connectors. The result is attached.
I'm not sure whether our model actually has an internal photodiode, but I included a connector for pins 3 and 4 anyway. I checked that the electrical connections work and there isn't unexpected shorting. I also found some heat sinks that I can place under the MXAN, since the body of the modulator won't press directly on the optical breadboard.
For about 9 hours after turning off the heater and removing insulating foil, the vacuum pressure was below 1e-7 torr. Then, the pressure increased to several 1e-7 torr, where it has remained. Not sure why this order of magnitude increase in vacuum pressure.
The vacuum pressure minute trend for the last day is in attachment 1, and the raw data for the hour or so where the pressure was rising is in attachment 2.
This vacuum handbook from Agilent that I return to often notes that at high vacuum, most of the remaining gas in the system is adsorbed on the walls rather than residing in the volume. Slide 7 notes that at 1e-9 torr, there are 500,000 molecules on the surface for every one in the volume, and it takes 2200 seconds for a monolayer to form (with both of these numbers inversely proportional to vacuum pressure in the molecular flow regime). Maybe for the first ~9 hours after the heater was turned off, the vacuum pressure was determined by (pumping speed + adsorption to the walls) - (desorption from the walls + permeation through KF flanges + backflow + outgassing + etc), with the adsorption rate limited by the surface area of the vacuum system. If there's some relation between number of monolayers and adsorption/desorption rate, at some point the number of monolayers could limit the net adsorption rate and lead to a higher equilibrium pressure.
At a pressure of ~4e-8 torr, the walls would form 1 monolayer of gas in about a minute. So in 9 hours, ~500 monolayers would form. I'd sort of expect the adsorption/desorption rate to equilibrate with just 1 monolayer of gas, so maybe I need another explanation.
I tuned up the PDH servo by widening the region with positive phase above 100 kHz and setting the PI corner near the cavity pole. I measured the plant transfer function from 10 kHz to 100 kHz and found that the phase is -45 degrees relative to ~DC near 37 kHz; the recent PDH sweep implied cavity pole ~39 kHz, so this is reasonable.
I measured the transfer functions and noise spectra (at error and control points) for the S laser locked to the (higher finesse) cavity. I kept the N laser off during these measurements, so it's the same 1-loop system we worked with in September. The cavity pole is now lower, ~37 kHz, and the UGF is ~115 kHz.
I plotted the results, but wasn't completely satisfied that I caught all factors of few in the units from the HP spectrum analyzer, so I'll update with attachments when I do so.
Update: The results are attached.
I pushed this updated noise spectrum to the gitlab, and debugged the pipeline so the results are again available on this webpage.
I turned off the heater and let the system return to room temperature. The pressure settled between 2-6e-8 torr (attachment 1 shows pressure decreasing for a couple hours after turning off the heater). This far exceeds our requirement for this vacuum system.
The pressure did reach 6e-6 torr, but then jumped to ~mtorr for after less than an hour.
I could not identify leaks by spritzing methanol around all flanges, but after going up to atmosphere noticed one KF flange was somewhat loose.
I noticed that the gauge reports >1000 torr at atmospheric pressure. After spinning down the turbo and venting to atmosphere, I calibrated the atmosphere reading by shorting pin 7 of the gauge to its negative supply. Even after the atmosphere calibration, the gauge read >900 torr, so I introduced an offset voltage on my calc channel to send the gauge's pressure reading at atmosphere to 10 V (the nominal gauge output at atmosphere). The -0.082 V 'ad hoc' offset decreases the absolute pressure reading by 25% at any given pressure.There is a Vacuum gauge manual available online
I then pumped down to several utorr and again shorted pin 7 to the negative supply to provide a 'vacuum' reading for the pirani. I only somewhat trust the absolute pressure readings (maybe to 25%), but presumably the pumpdown curves can now be consistently compared going forward.
After tightening the KF flanges and calibrating the gauge, I pumped down again and maintained at < 1 utorr for over 1 hour. I tried tightening the CF flange, but in doing so caused a leak that limited vacuum pressure to ~600 utorr (one of the KF flanges was torqued while tightening the CF bolts, and I'm not sure whether the leak is at the CF or one of the KF flanges since methanol reveals no leaks at any flange). I vented back up to atmosphere, cleaned the effected KF flanges, and replaced the copper gasket and resealed the CF flange. Then, I pumped down again.
The latest pumpdown curve is in attachment 1. I wrapped the cross in heat strap and foil, and when the pressure reached ~utorr turned the heater on to equilibrate at 85 C.
With the Edwards WRG-S gauge borrowed from QIL, I'm continuing with planned tests of the PSOMA chamber.
I don't have a controller for this gauge, so I cut off an RJ45 cable and added BNC connectors to the appropriate wires for +18V power supply and pressure signal output.
I'm testing the gauge by pumping on a minimal system: right angle valve -> KF40 and KF16 cross -> [KF16 blank, up to air valve, KF40-to-DN40CF adapter] -> WRG-S with DN40CF connection. Pumpdown curve is attached, I'll leave it pumping overnight to see how low the pressure gets.
On Friday, I noticed that the PSOMA S laser diode had suddenly reduced power (we've been seeing 1.7 mW at the beam launch, and suddenly I was getting only 0.5 mW). I was close to wrapping up, and was able to recover full power by hopping to a different hysterisis curve with TEC set to ~10 kOhm. Lee also pointed out that he wasn't seeing as much laser power as expected out of the cryo cavs diodes.
However, today I'm seeing odd behavior in the Acromag channels (such at X1:AUX-ACROXT_AI_1, which is the analog input channel for PSOMA vacuum pressure). The channels alternate between reading 0 and reading the correct value, for 10s of seconds at a time in either state.
On investigating the acromag chassis' power supply, I found that the +18 V rail is only supplying ~11 V. This could explain the sudden loss in laser power -- the current driver is probably supplying less current than it would with full voltage supplied. Probably turning on the cryo cavs lasers (plus whatever else we've plugged in since they were last used) drew too much current.
I increased the current limit on that Sorensen, and the DC current increased from 3.5 mA to 3.8 mA. The acromag-related channels are now fully functioning.
I'm measuring the beam properties at a few more locations
These measurements and the alamode fit for the beam profile are in attachment 1. The x data best fit waist radius is 405.6 um at -103 mm from the front plane of the fiber launch. The y data best fit are 421.3 um at -103 mm.
Attachment 2 shows the >0.99% mode matching solution least sensitive to errors in modematching lens position. It uses a 500 mm FL lens at 0.3023 m from the launch plane, and a 850 mm FL lens 0.4608 m from the launch plane. The target beam waist is larger than yesterday's plot because I looked up the quote for our coastline concave lens, and found it has a 1m ROC (not a 1m FL).
After these changes, I am seeing a bit more transmitted power. Power measurements suggest 72% mode matching and 450 ppm loss. I added a second stage ID filter to give some extra gain between 10 Hz and 100 Hz to suppress the cantilever motion. The PDH servo filters are in attachment 3 and 4. I measured the open loop transfer function to determine that the UGF is between 100 kHz and 125 kHz, which helped in choosing the derivative corner and saturation. When adjusting mode matching lenses, I increased the servo gain by 6dB and tuned the output HWP to adjust the amount of light sent to REFL PD; this allowed me to lock on the 00 mode most of the time just by roughly placing the lenses such that the beam spot maintains the same position on REFL camera.
I'm not convinced my mode matching solution is any better than the previous one, or perhaps the lenses are misplaced. The reflected beam still looks like a 1st order Laguerre-Gauss mode. I'm still placing the first mode matching lens at a 30-45 degree angle to correct astygmatism. THe reflected beam is the top monitor in attachment 5. Since there seems to be >1W of circulating power, I'll move on to signal injection tomorrow, but it would be good to know how to diagnose and correct this mode mismatch.
Thanks Koji! Shruti also suggested that WinCamD might be Si-based... I was looking at their TEL Phosphor response curve! I'm using the InGaAs camera on the Beam'R2-DD now.
I recorded the following beam parameters, with no lenses in the beam path. I left one steering mirror and the input HWP in the beam path. I need to check out the manual to understand some of these measurements... I think it's something like the camera scans a slit along two orthogonal directions (parameterized by 'theta'), and uses this to measure the beam power along the slices defined by the slit.
I measured the distance from the front surface of the beam launch to the HR surface of MC1 to be 523 mm. MC1 to MC2 is 230 mm. MC2 to MC3 is 230 mm. And MC3 to MC1 is 53 mm.
I found some mode matching solutions with a la mode (eg attached), and will implement one tomorrow.
WincamD is a silicon-based CCD camera (I believe). And may not have sufficient sensitivity at 1550nm.
Instead, use Beam'R2, which you probably brought back together with the WincamD combo. It has Silicon PD and exInGaAs PD and the exInGaAs can measure the beam profile even at the wavelength of 2um. So, your success is promised. You just need to make sure to use the exInGaAs side.
We adjusted MC3 by 1 full knob turn clockwise in yaw, and recorded the following power measurements after realigning the incoming beam to the cavity. These imply ~65% mode matching and 460 ppm loss.
I also retrieved the beam profiling cameras from OMC lab, and am using WinCamD to profile our beam and improve mode matching... but wasn't able to find the beam in about an hour of searching with the camera in a couple locations on the input path, using different ND filters, gain/exposure settings, etc. The camera responds to ambient light, but maybe I'm missing something.
I tuned MC1's yaw knob 1/2 turn clockwise, which should move the cavity axis closer to center on MC1 and MC3. I didn't get a stable enough lock to characterize the cavity with the box off, but it seemed to have moved the beam closer to center as expected (based on where the razor blade cuts off the cavity) and may have decreased the cavity pole (based on where I needed to put the PI corner to get more stable locks).
that's great news! I guess it makes sense to move things into the vacuum, since the table HEPA situation is not great yet.
We would really like to be able to close the sliding doors without a gap and install a switch to adjust the speed of the HEPAs. We should strategize something with JC.
Today I'm removing the dust from the PSOMA cavity optics by drag wiping their HR surfaces in-situ.
I prepared the environment by wiping down surfaces inside the enclosure with kimwipes and 70% isopropanol. With both HEPA FFU on "high," I first wiped the surfaces that go inside the box covering the cavity (optics table, optomechanics, interior and exterior of the plastic box) and covered the cavity with the box. Then I wiped the other surfaces inside the enclosure (optics table, optomechanics, vacuum chambers, oscilloscope, cables, TV monitors, cables, interior and exterior surfaces of enclosure walls, etc).
Per Koji's drag wiping tutorial and advice, here's my procedure for cleaning the optics:
Our pure IPA and acetone supplies were opened more than 2 years ago, so Koji lent me two newer bottles for today.
I completed the steps above, drag wiping at least 3x in each of steps 4-6. There is still a visible streak on the left side of MC1 (close to the beam spot) that wasn't affected by wiping with either solvent. Both MC1 and MC3 still have some visible dust, and MC1 still looks a bit hazy.
As soon as I finished cleaning the optics, I saw several times stronger flashes in TRANS PD. I needed to reduce the PDH servo gain to lock to the 00 mode, and reshape the high frequency response a bit to avoid oscillation. The TRANS PD is saturating, so I reduced its gain setting (from 70 dB to 50 dB).
Based on the above power measurements, the mode matching is 54% and intracavity loss is 575 ppm. Rana points out that these power measurements aren't very reliable, because they depend on our mediocre estimates of mirror transmissivities. Still, we can see a factor of >20 increase in transmitted light implying ~1 W circulating in the cavity. The implied cavity finesse based on power measurements is 2600, and the linewidth would be 230 kHz. The cavity pole based on power measurements is ~58 kHz. The cavity should now be overcoupled.
The new PDH slope based on attachment  is -4 MHz/V. The implied cavity linewidth is 150 kHz, and finesse is F~3900. I'm a bit wary of this measurement, because based on REFL levels I'd expect a PDH slope closer to -6 MHz/V. Sweeping the laser current more slowly resulted in the error signal being dominated by cantilever motion at 40 Hz; sweeping laser current more rapidly resulted in ringing near the peaks of the error signal. The implied cavity pole based on PDH sweep is ~39 kHz.
I set up a script to autogenerate medm screens from simulink models in /opt/rtcds/tst/x1/scripts/mdl2adl, and added it to the crontab on cominaux to run on a weekly schedule. Screens are available from a link button on the sitemap (see attachments).
We think we're seeing ~5000 ppm intracavity losses, which is limiting the power buildup in our cavity to about 5% of the lossless case.
I removed the box covering our cavity to investigate the source of these losses.
There is dust visible by eye across all three optics. MC2 (the cantilever) is clear across most of its front surface; MC1 (curved mirror) is the most uniformly coated with dust. I used a white LED flashlight for the attached photos.
With the cavity locked, I moved a razor blade (by hand) in front of each of the cavity mirrors. I identified that the beam is far left of center (closer to the edge than the center of the optic) on MC1 and MC3, but only slightly right of center of MC2. The beam is close to centered in pitch on all three mirrors .
I wiped down all o-rings and KF flanges with isopropyl alcohol and lint-free tekwipes, then closed up and am once again pumping down on just a single 4-way cross.
turbo -> right angle valve -> 4-way cross -> [up-to-air valve, vacuum pressure gauge, KF40 blank]
I noted some black particulate on the steel mesh filter just before the turbo pump. It looks like it could be silicon flakes? Attachments show the same flakes on a KF16 blank.
Update: After the group meeting, JC and I took the 4-way cross with up-to-air valve, gauge, and KF40 blank over to "camel" (formerly TCS) lab where we tried pumping down with their pumping station. We were still unable to reach below 6 mtorr, so concluded that the turbo pump is not to blame.
Next, we removed the MKS 748B gauge and used instead the combination gauge that had been in use for the photodiode testbed cryostat. The pressure was immediately able to reach below 1e-5 torr (the limit of this gauge). It seems the MKS 748B gauge we were using was either faulty (not providing an accurate pressure reading below ~6 mtorr) or actively leaking. I'll find a different combination gauge not currently use and continue testing the PSOMA vacuum system until a new gauge arrives.
I returned the "camel" lab pumping station and photodiode testbed to its previous state, and returned the 4-way cross and accessories to cryo lab.
Attachment #3 is the photo of pressure gauge reaching 'The pressure was immediately able to reach below 1e-5 torr (the limit of this gauge)'.