Here's a Finesse modeling of what we're expecting to observe with this test. It uses Gautam's base model of the 40m IFO with appropriate modifications for the needed configuration.

The idea is to lock the IFO in the SRMI configuration, with the phase-locked AUX beam injected from the AS port. The AUX beam is imprinted with AM sidebands and slightly misaligned relative to the SRC so as to transfer power into HOM1. The RF network analyzer provides the drive signal for the AOM, and its frequency is swept to coherently measure the transfer function [reflected AUX beam / drive]. The reflected AUX beam is sensed by the AS110 PDA10CF.

It is also possible to drive PM sidebands as Koji suggests, but the squeezer group has encouraged using AM for practical advantages. The SNR with AM is a bit higher (less power lost into harmonics at large modulation index), there is a broadband AOM already available aligned to the SQZ beam at LLO, and there is also concern that driving strong PM could interfere with the SQZ control loops.

Expected SRMI Response

Attachment #1 shows the expected response to swept-AM in SRMI. Resolving just the FSR and the first-order mode splitting is sufficient to extract the SRC Gouy phase.

Expected DRMI Response

Since the 40m has not been opearted in SRMI since ~2016 (last done by Eric Q.), Gautam believes it may take some time to relock this configuration. However, the modeling indicates that we can likely obtain sufficient sensitivity in DRMI, which would allow us to proceed faster. Attachment #2 shows the expected response to swept-AM in DRMI. The PRC leakage signal turns out to be significantly smaller than the SRC reflection (a factor of ~30 in amplitude), so that the signal still retains its characteristic shape to a very good approximation. The tradeoff is a 10x reduction in SNR due to increased PSL shot noise reaching AS110.

Based on this, we should proceed with DRMI scans instead of PRMI next week.

I have connected a 4TB disk to chiara via a USB-SATA adapter. This disk has been recognized as /dev/sde. A GUID Partition Table (GPT), not MBR was made with gdisk to make a partition with the size beyond 2TB.
I tried to use "dd" to copy /home/cds (/dev/sdb1) to /dev/sde1, but failed. The copy was done (taking ~12h) and the partition was not recognized as a complete filesystem.

So I decided to use rsync instead.

sudo mkfs -t ext4 /dev/sde1
sudo mkdir /media/usb4g
sudo mount -t ext4 -o rw /dev/sde1 /media/usb4g
sudo rsync -a --progress /home/cds/ /media/usb4g

Progress
14:33 Copied     33G/1831G
14:38 Copied     36G/1831G
17:02 Copied   365G/1831G (~2.2GB/min)
01:18 Copied 1449G/1831G (~2.2GB/min)
04:36 Completed
> sent 1907955222607 bytes  received 126124609 bytes  37010956.31 bytes/sec
> total size is 1907271994803  speedup is 1.00

I'm running a comsol job on optimus in a tmux session named cryocavs. Should be done in less than 24 hours, judging by past durations.

The initial local backup with rsync was done. Now the new 4TB disk is (supposed to be) automatically mounted at boot as /media/40mBackup so that we can run the daily backup on this disk. (<- This was confirmed by "sudomount -a")

controls@chiara|~> sudo blkid
...
/dev/sde1: UUID="92dc7073-bf4d-4c58-8052-63129ff5755b" TYPE="ext4"

controls@chiara|~> cat /etc/fstab
...
UUID=92dc7073-bf4d-4c58-8052-63129ff5755b   /media/40mBackup    ext4    defaults    0   0

Here I've used UUID rather than the device name "/dev/sde1" because the device name can get altered depending on the order of the usb connection.

This new disk is just a bare HDD drive sitting on the top of the chassis. We eventually want to accommodate this disk in the chassis so that we can recover the function only with the modification of /etc/fstab.  We need to wait for a next chance to have chiara down. In fact, when we can isolate chiara, we want to use this disk as the main disk and install another 4TB disk as a backup.

Yesterday, I moved the following optics:

1. Lens in front of AS110 PD.
2. BS splitting light between AS110 and AS55.

After moving these components around a bit, I locked them down once I was happy that the beam was pretty well centered on both of them, and also on AS110 and AS55 (measured using O'scope with single bounce from one ITM, other optics misaligned).

The beam was close to clipping on the lens mentioned in #1, probably because this wasn't checked when the 90-10 BS was installed for the AUX laser. Furthermore, I believe we are losing more than 10% of the light due to this BS. The ASDC (which is derived from AS55 PD) level is down at ~110cts as the Michelson is fringing, while it used to be ~200 cts. I will update with a power measurement shortly. But I think we should move ahead with the plan to combine the beam into the IFO's AS mode as discussed at the meeting last week.

Unrelated to this work, but c1psl and c1iscaux were keyed. 

ASDC has something weird going on with it - my main goal yesterday was to calibrate the actuators of ITMX, ITMY and BS using the Michelson. But with the Michelson locked on a dark fringe, the ASDC level changed by up to 50 counts seemingly randomly (bright fringe was ~1000 cts, I had upped the whitening gain to +21dB), even though the CCD remained clearly dark throughout. Not sure if the problem is in the readout electronics or in the PD itself.

Is the 10% specified for P-Pol or for UNP? I contacted CVI about beamsplitters, since their website doesn't list a BS1-1064-90-... option on the website. They say a R=90% beamsplitter would be a custom job. The closest stock item they got is BS1-1064-95-2025-45UNP specified at R=95% for UNPolarized beams. They were kind enough to sent me the measured transmission curves for a recent lot of these, which is attached was uploaded to the wiki [Elog Police K: NO PROPRIETARY DOCUMENTS ON THE ELOG, which is public. Put it on our wiki and put the link here]. The figure is not labeled, but according to the contact Red is S-Pol and Blue is P-Pol, which means that this one actually has R=~90% for P, pretty much what we want. We'll need to buy two of these to make the swap in the setup.

Back to your original point: There's only a BS1-1064-10-2025-45UNP on the website, so unless we got these as custom items, the R for P-Pol is probably NOT actually 10%, just somewhere between 0% and 20%

Of course, many (but no all) of the optics were custom-ordered back in ~2000.

Summary:

The actuator (pendulum) gains for the Beam Splitter and the two ITMs were measured to be:

BS: 9.54 +/- 0.05 nm/ct [100 ohm series resistor in coil driver board]

ITMX: 2.44 +/- 0.01 nm/ct [400 ohm series resistor in coil driver board]

ITMY: 2.44 +/- 0.02 nm/ct [400 ohm series resistor in coil driver board]

Counts here refers to DAC counts at the output of the coil filter banks (as opposed to counts at the LSC servo output). The dominant (systematic) uncertainty (which isn't quoted here) in this measurement is the determination of the peak-to-peak swing of the dark port sensor, AS55_Q. I estimate this error to be ~1ct/33cts = 3%. These values are surprisingly consistent with one another once we take into account the series resistance.

Details:

The last time this was done, we used ASDC to do the measurement. But I don't know what signal conditioning ASDC undergoes (in PD or in readout electronics). In any case, in my early trials yesterday, ASDC was behaving unpredictably. So I decided to do redo the measurement.

[Attachment #1]- Flowchart describing the calibration procedure.

[Attachment #2] - AS55_Q output while the Michelson was freeswinging. I had first aligned the ITMs using ASS. The peak-to-peak value of this corresponds to . So we know AS55_Q in terms of cts/m of MICH displacement.

[Attachment #3] - Magnitudes of transfer function from moving one of the MICH optics, to the now calibrated AS55_Q. Fits are to a shape , with  being the fitted parameter. Coherence during the measurement is also plotted.

• Note that the excitation is applied to the channels C1:SUS-<optic>_LSC_EXC, for <optic> in [BS, ITMX, ITMY]. But since my de-whitening board re-work to remove the analog x3 gain, there is a digital x3 gain in the coil driver filter banks. So while the calibration numbers given above are accurate, be aware that when using them for sensing matrix measurements etc, you have to multiply these by x3.
• Furthermore, moving the BS by results in a Michelson length change of , and this has been factored into the above number.

Next Steps:

1. Now that I have a calibration I trust more, re-analyze my DRMI sensing matrix data. Actually the sensing response numbers aren't significantly different from what I have been assuming. It's just that in terms of counts applied at the LSC input of a suspension, there is a digital x3 gain that has to be explicitly factored in.
2. Calibrate POX and POY by locking the arms and driving the now calibrated ITMs by a known number of counts.
3. Calibrate the ETMs, and MC1/MC2/MC3 by looking at calibrated POX/POY.
4. Lock DRMI, and calibrate SRM and PRM.

Reference:

[1] - http://www.phys.ufl.edu/~bernard/papers/CQG20_S903.pdf

Motivation:

1. I want to use the QPD at POP, calibrate it into physical units, and quantify the amount of angular jitter in the PRC (which I claim is what limits DRMI stability atm).
2. I want to revive the PRC angular feedforward to try and mitigate this a bit. But is feedforward even the best approach? Can we use feedback using the POP QPD?

POP QPD checkout:

• The POP QPD sits on the ITMX optical table. 
• It is interfaced to the CDS system via an OT301 and then a Pentek whitening stage (z:p = 15:150). 
• The OT301 claims to have a switchable offset nulling capability - but despite my best efforts tonight, I couldn't use the knobs on the front to null the offset (even with the PRC locked on carrier and a strong POP beam on the QPD).
  • We don't have readbacks of the individual quadrants available.
  •  
• So I moved the QPD with the PRC locked, to center the CDS readback of the spot position at (0,0).
• Next step is to calibrate the POP QPD readback into physical units.
  • I'm thinking of using the EricG diode laser for this purpose.
  • I can calibrate counts to mm of displacement on the QPD active area.
  • After which I can use the estimated position to PR2 (from which POP is extracted) to convert this to angular motion.
• I guess I should check for coherence between the POP QPD signal and all angular sensors of PRM/BS/MC1/MC2/MC3 to try and confirm the hypothesis that the folding mirrors are dominating the angular noise of the cavity. Unfortunately we don't have readbacks of the angular positions of TT1 and TT2.
• I moved the POP camera a bit in YAW so that the POP spot is now better centered on the CCD monitor.
• I also wanted to check the centering on the other POP QPD (POP22/POP110/POPDC?) but I think the POPDC signal, used for triggering the PRCL LSC servo, is derived from that PD, so everytime I blocked it, the lock was lost. Need to think of another strategy.
• MC3 has been rather glitchy tonight.
  • So I will wait for a quieter time when I can collect some data to train the WF for angular FF.

Aim: To measure the optical power from led using a powermeter.

Yesterday Gautam drilled a larger hole of diameter 5mm in the box as an aperture for led (aperture angle is approximately 2*tan^-1(2.5/7) = 39 deg). I repeated the measurements that I had done before (https://nodus.ligo.caltech.edu:8081/40m/13951). The measurents of optical power measured using a powermeter and the corresponding input voltages are listed below.

So we are able to receive optical power close to the value (1.6mW) given in Thorlabs datasheet for LED1050E (https://www.thorlabs.com/drawings/e6da1d5608eefd5c-035CFFE5-C317-209E-7686CA23F717638B/LED1050E-SpecSheet.pdf). I hope we can proceed to BRDF measurements for CCD calibration.

Steve: did you center the LED ?

Not much progress today with the AUX cavity scans. I've determined there still are some alignment issues.

At the start of today a large AUX/PSL beat note was visible on the AS110 sensor, at a similar power as where we left off last night (-60 dBm). Proceeding from there, I attempted to reproduce Johannes' measurement of the cavity transmission resonances. I misaligned the X-arm, locked the Y-arm cavity, and scanned the AUX RF offset approximately 8 MHz in 2 kHz steps. This should have swept through two FSRs, but nothing was visible.

Further inspection revealed that none of the PSL light was making it back to through the AUX fiber to the PSL table. I take this to mean that the beam seen earlier on AS110 was the ITMY reflection, and that the AUX injection axis was no longer reaching ETMY. I also found that the AUX beam size just after the 90/10 beasmsplitter looks anomolously large. Maybe a lens was recently changed? In any case, the mode-matching looks like it is going to need to be readjusted.

Per discussion today eve, barring objections, I will do the following tomorrow morning:

1. Remove ETMX coil driver board from 1X9
   • Change series resistances on the fast path to 2x4k in parallel. One will be snipped off once we are happy we can still lock.
   • Remove AD797s, potentiometers.
   • Thick film-->thin film for important components.
2. Remove ETMX de-whitening board from 1X9
   • Remove x3 analog gain.
   • Thick film-->thin film for important components.

We did swap a lens as discussed in elog 13968, but they both had f=100mm specified, the difference being one was AR-coated for 1064 and bi-convex, while the other one was plano-convex and had a different coating. The reason for the large beam spot was something else: The fiber wasn't sitting in the coupler properly. When reconnecting the fiber after taking it out make sure to align the key on the fiber end with the notch in the coupler before tightening. After discovering this the following was done:

• Fixed fiber mounting situation
• Tested AUX alignment into fiber on PSL table, was still good
• The AUX polarization was aligned to the wrong fiber axis. I fixed this. The coupler on the PSL table has it's noth oriented vertically since we're using s-polarized light. The AS-table coupler is rotated by 90 degrees, such that the notch points to the side. This way we technically don't need any halfwaveplates for rotation. However, there are still current HWPs installed.
• Locked both arms and ran dither alignment until satisfactory
• Misaligned ITMX and ETMX, and further set the ITMX pitch offset to 0.0
• Started overlapping the expectedly misaligned beams by eye. For this I turned the power of the deflected beam down to 50mV bias voltage, which gives the PSL and AUX lasers similar card-brightness on the shared path
• Misaligned SRM more because there was still the strong prompt reflection coming out the AS port.
• Restored phaselock between AUX and PSL, with beat at 30MHz between 1st-order diffracted in fiber and PSL
• Immediately saw STRONG 30MHz RF signal on AG4395. Disappeared when blocking AUX, and optimized alignment brought the signal up to -10dBm, as shown in attachment #1
• Checked YEND PDA10CF and saw a -80dBm RF signal at 30 MHz (#2), compatible with earlier observations.

Before leaving I restored the XARM alignment. SRM remains misaligned, LSC off. Alignment shouldn't change drastically over night, so I suggest when picking this work up tomorrow to directly look for the beats after phaselocking AUX and PSL

You should wipe off the table cover before you take it off next time.

It is important to turn up the PSL encloure HEPA Variac voltage if you are working in there. It takes less than 10 minutes to reach lab condition.

Lab air count normal. It is not logged. I have a notebook of particle count on the SP table next to the Met One counter.

Yes.

I finished the re-soldering work today, and have measured the coil driver noise pre-Mods and post-Mods. Analysis tomorrow. I am holding off on re-installing the board tonight as it is likely we will have to tune all the loops to make them work with the reduced range. So ETMX will remain de-commissioned until tomorrow.

I decided to take a quick look at the data. Changes made to the ETMX coil driver board:

1. Fast path series resistances: 400 ohm ---> 2.25 kOhm (= 2x 4.5 kohm in parallel). Measured (with DMM) resistance in all 5 paths varied by less than 3 ohms (~0.2%).
2. All thick film resistors in signal (fast and bias) paths changed to thin film.
3. AD797 ---> Op27 for monitor output.
4. Above-mentioned mon output (30Hz HPF-ed) routed to FP LEMO mon via 100ohm for diagnostic purposes.
5. 4x Trim-pots in analog path removed. 

I also took the chance to check the integrity of the LM6321 ICs. In the past, a large DC offset on the output pin of these has been indicative of a faulty IC. But I checked all the ICs with a DMM, and saw no anomalies.

Measurement condition was that (i) the Fast input was terminated to ground via 50ohm, (ii) the Bias input was shorted to ground. SR785 was used with G=100 Busby preamp (in which Steve installed new batteries today, for someone had left it on for who knows how long) for making the measurement. The voltage measurement was made at the D-Sub connector on the front panel which would be connected to the Sat. Box, with the coil driver not connected to anything downstream.

Summary of results:

[Attachment #1] - Noise measurement out to 800 Hz. The noise only seems to agree with the LISO model above 300 Hz. Not sure if the low-frequency excess is real or a measurement artefact. Tomorrow, I plan to make an LPF pomona box to filter out the HF pickup and see if the low-frequency characteristics change at all. Need to think about what this corner freq. needs to be. In any case, such a device is probably required to do measurements inside the VEA.

[Attachment #2] - Noise measurement for full SR785 span. The 19.5 kHz harmonics are visible. I have a theory about the origin of these, need to do a couple of more tests to confirm and will make a separate log.

[Attachment #3] - zip of LISO file used for modeling coil driver. I don't have the ASCII art in this, so need to double check to make sure I haven't connected some wrong nodes, but I think it's correct.

Measurements seem to be consistent with LISO model predictions.

*Note: Curves labelled "LISO model ..." are really quad sum of liso pred + busby box noise.

My main finding tonight is: With the increased series resistance (400 ohm ---> 2.25 kohm), LISO modeling tells me that even though the series resistance (Johnson noise) used to dominate the voltage noise at the output to the OSEM, the voltage noise of the LT1125 in the bias path now dominates. Since we are planning to re-design the entire bias path anyways, I am not too worried about this for the moment.

I will upload more details + photos + data + schematic + LISO model breakdown tomorrow to a DCC page. 

gautam noon 21 June 2018: I was looking at the wrong LISO breakdown curves. So the input stage Op27 voltage noise used to dominate. Now the Bias path LT1125 voltage noise dominates. None of the conclusions are affected... I've uploaded the corrected plots and LISO file here now. 

[Jon, Keerthana, Sandrina]

Yesterday we carried out preliminary proof-of-concept measurements using the new AS-port-injected AUX laser to resolve cavity mode resonances.

At the time we started, I found the beat note levels consistent with what Johannes had reported the night before post-realignment. Hence we did not change the AUX alignment.

Test 1: YARM Mode Scan

• IFO locked in YARM configuration on carrier.
• Confirmed the presence of a -80 dBm beat note on the temporary YEND broadband PD (i.e., at the cavity transmission).
• Slowly canned the RF offset of the AUX laser from 50 MHz (nominal) to 60 MHz in 10 kHz steps.
• Attachment 1 shows the measured scan in "max hold" mode. The bottom panel is the transmission spectrum and the top panel is the reflection, with the AUX/PSL carrier-carrier beat note visible to the far left. In addition to the FSR, it looks to me like the scan resolves at least two HOMs.

Test 2: PRC Mode Scan

• IFO locked in PRMI configuration on carrier.
• Moved the temporary 150 MHz PDA10CF from the YEND to an unused pickoff of the REFL33 beam (i.e., the PRC transmission of the AUX beam). There was an existing 50-50 beamsplitter just before REFL33 whose reflected beam was directed onto a beam dump. The PD is now placed in that location. The modification to the AS table is shown in Attachment 2.
• We made a similar slow scan of the AUX RF offset over ~35 MHz in 10 kHz steps.
• We resolve the 22 MHz FSR, but it is apparent that an incoherent "max-hold" analyzer measurement is inadequate. The problem is that in max-hold mode, because the 11 MHz-spaced PSL sidebands also beat with the AUX subcarrier, we measure a messy superposition of the PSLcarrier-AUXcarrier beat AND all of the PSLsideband-AUXcarrier beats. The next step is to use the AOM to make a coherent measurement at only the frequency of PSL/AUX carrier-carrier beat.

The SURFs have the data from last night's scans and will be separately posting plots of these measurements. We'll continue today with mode scans using AM sidebands rather than the AUX RF offset.

(Jon, Keerthana, Sandrine)

We tried to align the AUX and PSL laser yesterday. We collected the data from the spectrum analyser for the Y-ARM reflection and also for the Y-ARM transmission from the ETM mirror. I am attaching the plots here.

[steve, gautam]

I took this opportunity of EX downtime to change the supply voltage for the AA unit (4-pin LEMO front panel) in 1X9 from +/-5V to +/-15V. Inside the AA board are INA134 and DRV135 ICs, which are rated to work at +/-18V. In the previous state, the inputs would saturate if driven with a 2.5Vpp sine wave from a DS345 func. gen. After the change, I was able to drive the full range of the DS345 (10Vpp), and there was no saturation seen. This AA chassis is only used for the OSEM signals and also some ALS signals. Shadow sensor levels and spectra are consistent before and after the change. The main motivation was to not saturate the Green PDH Reflection signal in the digital readout. The steps we took were:

1. Confirm (by disconnecting the power cable at the back of the AA box) that the power supplied was indeed +/- 5 V.
2. Remove DIN fuse blocks from DIN rail for the relevant blocks.
3. Identify a +15 V, -15 V and GND spot to plug the wires in. 
4. Effect the swap.
5. Re-insert fuses, checked supply voltage at connector end of the cable was now +/- 15 V as expected.
6. Re-connect power cable to AA box.

Initial tests look promising. Local damping works and I even locked the X arm using POX, although I did it in a fake way by simply inserting a x5.625 (=2.25 kohm / 400 ohm) gain in the coil driver filter banks. I will now tune the individual loop gains to account for the reduced actuation range.

Now I have changed the loop gains for local damping loops, Oplev loops, and POX locking loop to account for the reduced actuation range. The dither alignment servo (X arm ASS) has not been re-commissioned yet...

pianosa has been upgraded to SL7. I've made a controls user account, added it to sudoers, did the network config, and mounted /cvs/cds using /etc/fstab. Other capabilities are being slowly added, but it may be a while before this workstation has all the kinks ironed out. For now, I'm going to follow the instructions on this wiki to try and get the usual LSC stuff working.

We (Rana and I) are re-assembling the temperature controls on the seismometer to attempt PID control and then improve it using reinforcement learning.

We tried to re-assemble the connections for the heater and in-loop temperature sensor on the can that covers the seismometer.

We fixed (soldered) two of the connections from the heater circuit to the heater, but did not manage to get the PID working as one of the wires attached to the MOSFET had come off. Re-soldering the wire would be attempted tomorrow.

Equipment for undertaking all this is still left at the X-end of the interferometer and will be cleared soon.

Two out of the four over-head fluorescent lights in the X end of the interferometer were flickering today.

MEDM, EPICS and dataviewer seem to work, but diaggui still doesn't work (it doesn't work on Rossa either, same problem as reported here, does a fix exist?). So looks like only donatella can run diaggui for now. I had to disable the systemd firewall per the instructions page in order to get EPICS to work. Also, there is no MATLAB installed on this machine yet. sshd has been enabled.

We may lost the UL magnet or LED

Aim: To find a model that trains the simulated data of Gaussian beam spot moving in a vertical direction by the application of a sinusoidal signal.

All the attachments are in the zip folder.

The simulated video of beam spot motion without noise (amplitude of sinusoidal signal given = 20 pixels) is given in this link https://drive.google.com/file/d/1oCqd0Ki7wUm64QeFxmF3jRQ7gDUnuAfx/view?usp=sharing

I tried several cases:

Case 1:

I added random uniform noise (ranging from 0 to 25.5 i.e. 10% of the maximum pixel value 255) using opencv to 64*64 simulated images made in the last case( https://nodus.ligo.caltech.edu:8081/40m/13972), clipped the pixel values from 0 to 255 & trained using the same network as in the previous elog and it worked well. The variation in mean squared error with epochs is given in Attachment 1 & applied signal and output of the neural network (NN) (magnitude of the signal vs time) as well as the residual error is given in Attachment 2.

Case 2:

I simulated images 128*128 at 10 frames/sec by applying a sine wave of frequency 0.2Hz that moves the beam spot & resized it using opencv to 64*64. Then I trained 300cycles & tested with 1000 cycles with the following sequential model:

(i) Layers and number of nodes in each:

                                             4096 (dropout = 0.1) -> 1024 (dropout = 0.1) -> 512 (dropout = 0.1) -> 256  ->  64 ->  8   ->   1

           Activation :                        selu                   ->                 selu             ->         selu                -> selu ->  selu -> selu -> linear

(ii) loss function = mean squared error ( I used mean squared error to easily comprehend the result. Initially I had tried log(cosh) also but unfortunately I had stopped the run in between when test loss value had no improvement), optimizer = Nadam with default learning rate = 0.002

(iii) batch size = 32, no. of epochs = 400

I have attached the variation in loss function with epochs (Attachment 3). It was found that test loss value increases after ~50 epochs. To avoid overfitting, I added dropout to the layer of 256 nodes in the next model and removed the layer of 4096 nodes.

Case 3: 

Same simulated data as case 2 trained with the following model,

(i) Layers and number of nodes in each:

                                             1024 (dropout = 0.1) -> 512 (dropout = 0.1) -> 256 (dropout = 0.1) ->  64 ->  8   ->   1

           Activation :                              selu             ->         selu                ->              selu ->             selu -> selu -> linear

(ii) changed the learning rate from default value of 0.002 to 0.001. Rest of the hyperparameters same.

The variation in mean squared error in attachment 4  & NN output, applied signal & residual error (zoomed) in attachment 5. Here also test loss value increases after ~65 epochs but this fits better than the previous model as loss value is less.

Case 4:

Since in most of the examples in keras, training dataset was more than test dataset, I tried training 1000 cycles & testing with 300 cycles. The respective plots are attached as attachment 6 & 7. Here also, there is no significant improvement except that the test loss is increasing at a slower rate with epochs as compared to the last case.

Case 5:

Since most of the above cases were like overfitting (https://machinelearningmastery.com/diagnose-overfitting-underfitting-lstm-models/, https://github.com/keras-team/keras/issues/3755) except that test loss is less than train loss value in the beginning , I tried implementing case 4 with the initial model of 2 layers of 256 nodes each but with Nadam optimizer. Respective graphs in attachment 8, 9 & 10(zoomed). The loss value is slightly higher than the previous models as seen from the graph but test & train loss values converge after some epochs.

I have forgot to give ylabel in some of the graphs. It's the magnitude of the applied sine signal to move the beam spot. In most of the cases, the network almost correctly fits the data and test loss value is lower in the initial epochs. I think it's because of the dropout we added in the model & also we are training on the clean dataset.

4  std cataloge item fused silica  BS1-1064-95-2025-45UNP 

ordered today. They will arrive no later than July 13, 2018

Seems like DTT also works now. The trick seems to be to run sudo /usr/bin/diaggui instead of just diaggui. So this is indicative of some conflict between the yum installed gds and the relic gds from our shared drive. I also have to manually change the NDS settings each time, probably there's a way to set all of this up in a more smooth way but I don't know what it is. awggui still doesn't get the correct channels, not sure where I can change the settings to fix that.

DON"T RUN DIAGGUI AS ROOT

I think if the magnet fell off, we would see high DC signal, and not 0 as we do now. I suspect satellite box or PD readout board/cabling. I am looking into this, tester box is connected to ITMY sat. box for now. I will restore the suspension later in the evening.

Suspension has now been restored. With combination of multimeter, octopus cable and tester box, the problem is consistent with being in the readout board in 1X5/1X6 or the cable routing the signals there from the sat. box.

• Tester box hooked up to sat box ---> UL coil still shows 0 in CDS.
• Tester box hooked up to sat box ---> Mon D-sub on sat box shows expected voltages on DMM. So tester box LEDs are being powered and seem to work.
• Sat box re-connected to test mass ---> Mon D-sub on sat box shows expected voltages on DMM. So OSEM LEDs are being powered and seem to work.
• Sat box remains connected to TM ---> Front panel LEMO monitor points on readout board shows 0 for UL channel, other channels are okay.

[Jon, Keerthana, Sandrine]

Thu.-Fri. we continued with PRC scans using the AUX laser, but now the "scanned" parameter is the frequency of AM sidebands, rather than the frequency of the AUX carrier itself. The switch to AM (or PM) allows us to coherently measure the cavity transfer as a function of modulation frequency.

In order to make a sentinel measurement, I installed a broadband PDA255 at an unused pickoff behind the first AUX steering mirror on the AS table. The sentinel PD measures the AM actually imprinted on the light going into the IFO, making our measurement independent of the AOM response. This technique removes not only the (non-flat) AOM transfer function, but also any non-linearities from, e.g., overdriving the AOM. The below photo shows the new PD (center) on the AS table.

With the sentinel PD installed, we proceeded as follows.

• Locked IFO in PRMI on carrier.
• Locked AUX PLL to PSL.
• Tuned the frequency of the AUX laser (via the RF offset) to bring the carrier onto resonance with the PRC.
• Swept the AOM modulation frequency 0-60 MHz while measuring the AUX reflection and injection signals.

The below photo shows the measured transfer function [AUX Reflection / AUX Injection]. The measurement coherence is high to ~55 MHz (the AOM bandwidth is 60 MHz). We clearly resolve two FSRs, visible as Lorentzian dips at which more AUX power couples into the cavity. The SURFs have these data and will be separately posting figures for the measurements.

With the basic system working, we attempted to produce HOMs, first by partially occluding the injected AUX beam with a razor blade, then by placing a thin two-prong fork in the beam path. We also experimented with using a razor blade on the output to partially occlude the reflection beam just before the sensor. We were able to observe an apparent secondary dip indicative of an HOM a few times, as shown below, but could not repeat this deterministically. Besides not having fine control over the occlusion of the beams, there is also large few-Hz angular noise shaking the AS beam position. I suspect from moment to moment the HOM content is varying considerably due to the movement of the AS beam relative to the occluding object. I'm now thinking about more systematic ways to approach this.

How much was the osc freq of the marconi? And then how much was the resulting freq offset between PSL and AUX?

Are we supposed to see two dips with the separation of an FSR? Or four dips (you have two sidebands)?

And the distance between the dips (28MHz-ish?) seems too large to be the FSR (22MHz-ish).
cf https://wiki-40m.ligo.caltech.edu/IFO_Modeling/RC_lengths

Summary:

For a series resistance of 4.5 kohm, we are suffering from the noise-gain amplified voltage noise of the Op27 (2*3.2nV/rtHz), and the Johnson noise of the two 1 kohm input and feedback resistances. As a result, the current noise is ~2.7 pA/rtHz, instead of the 1.9 pA/rtHz we expect from just the Johnson noise of the series resistance. For the present EX coil driver configuration of 2.25 kohm, the Op27 voltage noise is actually the dominant noise source. Since we are modeling small amounts (<1dB) of measurable squeezing, such factors are important I think. 

Details:

[Attachment #1] --- Sketch of the fast signal path in the coil driver board, with resistors labelled as in the following LISO model plots. Note that as long as the resistance of the coil itself << the series resistance of the coil driver fast and slow paths, we can just add their individual current noise contributions, hence why I have chosen to model only this section of the overall network.

[Attachment #2] --- Noise breakdown per LISO model with top 5 noises for choice of Rseries = 2.25 kohm. The Johnson noise contributions of Rin and Rf exactly overlap, making the color of the resulting line a bit confusing, due to the unfortunate order of the matplotlib default color cycler. I don't want to make a custom plot, so I am leaving it like this.

[Attachment #3] --- Noise breakdown per LISO model with top 5 noises for choice of Rseries = 4.5 kohm. Same comments about color of trace representing Johnson noise of Rin and Rf. 

Possible mitigation strategies:

1. Use an OpAmp with lower voltage noise. I will look up some candidates. LT1028/LT1128? LISO library warns of a 400 kHz noise peak though...
2. Use lower Rin and Rf. The values of these are set by the current driving capability of the immediately preceeding stage, which is the output OpAmp of the De-Whitening board, which I believe is a TLE2027. These can source/sink 50 mA according to the datasheet . So for +/-10V, we could go to 400 ohm Ri and Rf, source/sink a maximum of 25mA, and reduce the Johnson noise contribution by 40%.

Other comments/remarks:

I've chosen to ignore the noise contribution of the high current buffer IC that is inside the feedback loop. Actually, it may be interesting to compare the noise measurements (on the electronics bench) of the circuit as drawn in Attachment #1, without and with the high current buffer, to see if there is any difference.

This study also informs about what level of electronics noise is tolerable from the De-Whitening stage (aim for ~factor of 5 below the Rseries Johnson noise).

Finally, in doing this model, I understand that the observation the voltage noise of the coil driver apparently decreased after increasing the series resistance, as reported in my previous elog. This is due to the network formed by the fast and slow paths (during the measurement, the series resistance in the slow path makes a voltage divider to ground), and is consistent with LISO modeling. If we really want to measure the noise of the fast path alone, we will have to isolate it by removing the series resistance of the slow bias path.

Comment about LISO breakdown plots: for the OpAmp noises, the index "0" corresponds to the Voltage noise, "1" and "2" correspond to the current noise from the "+" and "-" inputs of the OpAmp respectively. In future plots, I'll re-parse these...

At some point we want to change the AUX injection on the AS table to interfere less with the normal interferometer path, and avoid 10/90 beamsplitters which produce a fair amount of ghosting. The plan is to replace the 99/1 BS whose reflection goes to AS110 and AS55, while the transmission goes to the AS camera, with a 90/10 BS as shown in the attachment. This results in ~10% less light on the PDs compared to the pre-AUX era. Between this BS and the AS camera there will be a second 90/10 BS that sends the AUX light into the IFO, so we end up with marginally less AUX power into the IFO and the same PSL power on the AS cam. We're short optics, so this has to wait until two new beamsplitters arrive from CVI.

After removing all the clamping screws from the heater circuit board, I soldered the wire connecting IRF630 to the output of OP27, which had come off earlier. This can only be a temporary fix as the wire was not long enough to be able to make a proper solder joint. I also tried fixing two other connections which were also almost breaking.

After re-assembling everything I found out that one of the LEDs was not working. The most likely cause seems to be an issue with LM791, LM 781 or the LED itself. Due to the positioning of the wires, I was unable to test them today but will try again possibly tomorrow.

Equipment used for this is still lying at the X end.

(Jon, Keerthana, Sandrine)

I am attaching the plots of the Reflected and transmitted AUX beam. In the transmission graph, we are getting peak corresponding to the resonance frequencies, as at that frequency maximum power goes to the cavity. But in the Reflection graph, we are obtaining dips corresponding to the resonance frequency because maximum power goes to the cavity and the reflected beam intensity becomes very less at those points.

Aim: To track the motion of beam spot in simulated video.

I simulated a video that moves the beam spot at the centre of the image initially by applying a sinusoidal signal of frequency 0.2Hz and amplitude 1 i.e. it moves maximum by 1 pixel. It can be found in this shared google drive link (https://drive.google.com/file/d/1GYxPbsi3o9W0VXybPfPSigZtWnVn7656/view?usp=sharing). I found a program that uses Kernelized Correlation Filter (KCF) to track object motion from the video. In the program we can initially define the bounding box (rectangle) that encloses the object we want to track in the video or select the bounding box by dragging in GUI platform. Then I saved the bounding box parameters in the program (x & y coordinates of the left corner point, width & height) and plotted the variation in the y coordinates. I have yet to figure out how this tracker works since the program reads 64*64 image frames in video as 480*640 frames with 3 colour channels and frame rate also randomly changes. The plot of the output of this tracking program & the applied signal has been attached below. The output is not exactly sinusoidal because it may not be able to track very slight movement especially at the peaks where the slope = 0.

I wanted to investigate my coil driver noise measurement technique under more controlled circumstances, so I spent yesterday setting up various configurations on a breadboard in the control room. The overall topology was as sketched in Attachment #1 of the previous elog, except for #4 below. Summary of configurations tried (series resistance was 4.5k ohm in all cases):

1. Op 27 with 1kohm input and feedback resistors.
2. LT1128 with 1kohm input and feedback resistors.
3. LT1128 with 400 ohm input and feedback resistors.
4. LT1128 with 400 ohm input and feedback resistors, and also the current buffer IC LM6321 implemented.

Attachments:

Attachment #1: Picture of the breadboard setup.

Attachment #2: Noise measurements (input shorted to ground) with 1 Hz linewidth from DC to 4 kHz.

Attachment #3: Noise measurements for full SR785 span. 

Attachment #4: Apparent coupling due to PSRR.

Attachment #5: Comparison of low frequency noise with and without the LM6321 part of the fast DAC path implemented.

All SR785 measurements were made with input range fixed at -42dBVpk, input AC coupled and "Floating", with a Hanning window. 

Conclusions:

• I get much better agreement between LISO and measurement at a few hundred Hz and below with this proto setup. So it would seem like the excess noise I measure at ~200 Hz in the Eurocrate card version of the coil driver could be real and not simply a measurement artefact.
• I am puzzled about the 10 Hz comb in all these measurements:
  • I have seen this a few times before - e.g. elog13655.
  • It is not due to the infamous GPIB issue - the lines persist even though I disconnect both power adaptor and GPIB prologix box from the SR785.
  • It does not seem to be correlated with the position of the analyzer w.r.t. the DC power supply (Tektronix PS280) used to power the circuit (I moved the SR785 around 1m away from the supply).
  • It persists with either of the two LN preamp boxes available. 
  • It persists with either "Float" or "Ground" input setting on the SR785.
  • All this pointed to some other form of coupling - perhaps conductive EMI.
  • The only clue I have is the apparent difference between the level of the coupling for Op27 and LT1128 - it is significantly lower for the latter compared to the former. 
  • I ruled out position on the breadboard: simply interchanging the Op27 and LT1128 positions on the breadboard, I saw higher 10 Hz harmonics for the Op27 compared to the LT1128. In fact, the coupling was higher for the DIP Op27 compared to an SOIC one I attached to the breadboard via an SOIC to DIP adapter (both were Op27-Gs, with spec'ed PSRR of 120 dB typ).
  • To test the hypothesis, I compared the noise for the Op27 config, on the one hand with regulated (via D1000217) DC supply, and on the other, directly powered by the Tektronix supply. The latter configuration shows much higher coupling.
  • I did have 0.1uF decoupling capacitors (I guess I should've used ceramic and not tantala) near the OpAmp power pins, and in fact, removing them had no effect on the level of this coupling 
  • As a quick check, I measured the spectrum of the DC power used to run the breadboard - it is supplied via D1000217. I used an RC network to block out the DC, but the measurement doesn't suggest a level of noise in the supply that could explain these peaks.
  • The regulators are LM2941 and LM2991. They specify something like 0.03% of the output voltage as AC RMS, though I am not sure over what range of frequencies this is integrated over.
  • But perhaps the effect is more subtle, some kind of downconversion of higher frequency noise, but isn't the decoupling cap supposed to protect against this?
• The 19.5 kHz harmonics seem to originate from the CRT display of the SR785 (SVGA).
  • The manual doesn't specify the refresh rate, but from a bit of googling, it seems like this is a plausible number.
  • The coupling seems to be radiative. The box housing the Busby preamp provides ~60dB attentuation of this signal, and the amplitude of the peaks is directly correlated to where I position the Busby box relative to the CRT screen.
  • This problem can be avoided by placing the DUT and preamp sufficiently far from the SR785.

Punchlines:

1. The actual coildriver used, D010001, doesn't have a regulated power supply, it just draws the +/- 15V directly from Sorensens. I don't think this is good for low noise. 
2. The LM6321 part of the circuit doesn't add any excess noise to the circuit, consistent with it being inside the unity gain feedback loop. In any case, with 4.5 kohm series resistance with the coil driver, we would be driving <2.5 mA of current, so perhaps we don't even need this?

Aim:  To find a model that trains the simulated data of Gaussian beam spot moving in a vertical direction by the application of a sinusoidal signal. The data also includes random uniform noise ranging from 0 to 10.

All the attachments are in the zip folder.

I simulated images 128*128 at 10 frames/sec by applying a sine wave of frequency 0.2Hz that moves the beam spot, added random uniform noise ranging from 0 to 10 & resized the image frame using opencv to 64*64. 1000 cycles of this data is taken as train & 300 cycles as test data for the following cases. Optimizer = Nadam (learning rate = 0.001), loss function used = mean squared error, batch size = 32,

Case 1:

Model topology:

                         256 (dropout = 0.1)  ->           256 (dropout = 0.1)   ->       1

Activation :             selu                                         selu

Number of epochs = 240.

Variation in loss value of train & test datasets is given in Attachment 1 of the attached zip folder & the applied signal as well as the output of neural network given in Attachments 2 & 3 (zoomed version of 2).

The model fits well but there is no training since test loss is lower than train loss value. I found in several sites that dropout of some of the nodes during training but retaining them during test could be the probable reason for this (https://stackoverflow.com/questions/48393438/validation-loss-when-using-dropout , http://forums.fast.ai/t/validation-loss-lower-than-training-loss/4581 ). So I removed dropout while training next time.

Case 2:

Model topology:

                         256 (dropout = 0.1)  ->           256 (dropout = 0.1)   ->       1

Activation :             selu                                         selu                          linear

Number of epochs = 200.

Variation in loss value of train & test datasets is given in Attachment 4 of the attached zip folder & the applied signal as well as the output of neural network given in Attachments 5 & 6 (zoomed version of 2).

But still no improvement.

Case 3:

I changed the optimizer to Adam and tried with the same model topology & hyperparameters as case 2 with no success (Attachments 7,8 & 9).

Finally I think this is because I'm training & testing on the same data. So I'm now training with the simulated video but moving it by a maximum of 2 pixels only and testing with a video of ETMY that we had captured earlier.

I checked out the elog from the vent in October 2016 when the OMC was removed from the path. In the vent in a couple weeks, we'd like to get the beam going through the OMC again. I wasn't really there for this last vent and don't have a great sense for how things go at the 40m, but this is how I think the procedure for this work should approximately go. The main points are that we'll need to slightly translate and rotate OM5, rotate OM6, replace one mirror that was removed last time, and add some beam dumps. Please let me know what I've got wrong or am missing.

[side note, I want to make some markup on the optics layouts that I see as pdfs elsewhere in the log and wiki, but haven't done it and didn't much want to dig around random drawing software, if there's a canonical way this is done please let me know.]

Steps to return the OMC to the IFO output:

1. Complete non-Steve portions of the pre-vent checklist (https://wiki-40m.ligo.caltech.edu/vent/checklist)
2. Steve needs to complete his portions of the checklist (as in https://nodus.ligo.caltech.edu:8081/40m/12557)
3. Need to lock some things before making changes I think—but I’m not really sure about these, just going from what I can glean from the elogs around the last vent
   1. ​Lock the IMC at low power
   2. Align the arms to green
   3. Lock the arms
   4. Center op lev spots on QPDs
   5. Is there a separate checklist for these things? Seems this locking process happens every time there is a realignment or we start any work, which makes sense, so I expect it is standardized.
4. Turn/add optics in the reverse order that Gautam did
   1. ​Check table leveling first?
   2. Rotate OM5 to send the beam to the partially transmissive mirror that goes to the OMC; currently OM5 is sent directly to OM6. OM5 also likely needs to be translated forward slightly; Gautam tried to maintain 45 deg AOI on OM5/6.
   3. A razor beam dump was also removed, which should be replaced (see attachment 1 on https://nodus.ligo.caltech.edu:8081/40m/12568)
   4. May need to rotate OM6 to extract AS beam again, since it was rotated last time
   5. Replace the mirror just prior to the window on the AP table, mentioned here in attachment 3: https://nodus.ligo.caltech.edu:8081/40m/12566
      1. ​There is currently a rectangular weight on the table where the mirror was, for leveling
5. Since Gautam had initially made this change to avoid some backscattered beams and get a little extra power, we may need to add some beam dumps to kill ghosts
   1. This is also mentioned in 12566 linked above, the dumps are for back-reflection off the windows of the OMC
6. Center beam in new path
7. Check OMC table leveling
8. AS beam should be round on the camera, with no evidence of clipping on any optics in the path (especially check downstream of any changes)

I used the following commands to get diaggui to run on rossa/SL7:

controls@rossa|lib64> ls -lrt libsasl*
-rwxr-xr-x. 1 root root 121296 Feb 16  2016 libsasl2.so.3.0.0
lrwxrwxrwx. 1 root root     17 Dec 18  2017 libsasl2.so -> libsasl2.so.3.0.0
lrwxrwxrwx. 1 root root     17 Dec 18  2017 libsasl2.so.3 -> libsasl2.so.3.0.0
controls@rossa|lib64> sudo ln -s libsasl2.so.3.0.0 libsasl2.so.2
controls@rossa|lib64> ls -lrt libsasl*
-rwxr-xr-x. 1 root root 121296 Feb 16  2016 libsasl2.so.3.0.0
lrwxrwxrwx. 1 root root     17 Dec 18  2017 libsasl2.so -> libsasl2.so.3.0.0
lrwxrwxrwx. 1 root root     17 Dec 18  2017 libsasl2.so.3 -> libsasl2.so.3.0.0
lrwxrwxrwx. 1 root root     17 Jun 26 22:02 libsasl2.so.2 -> libsasl2.so.3.0.0
 

Basically, I have set up a symbolic link to point sasl2.so.2 to sasl2.so.3.0.0. I've asked LLO again for some guidance on whether or not to find some backport in a non-standard SL7 repo. IF they reply, we may later replace this link with a regular file.

For the nonce, diaggui runs and is able to show us the spectra. We also got swept sine to work. But the FOTON launched from inside of AWGGUI doesn't inherit the sample frequency of the excitation channel so we can't filter noise injections from awggui yet.

Summary:

I've been thinking about what we need to do to the de-whitening boards for the ITMs and ETMs, in order to have low noise actuators. Noting down what I have so far, so that people can comment / point out things I've overlooked. 

Attachment #1: Block diagram schematic of the de-whitened signal path on D000183 as it currently exists. I've omitted the unity gain buffer stage at the output, though this is important for noise considerations. 

Some considerations, in rough order of priority:

1. Why do we need de-whitening?
   • Because we want the Johnson noise of the series resistor (4.5 kohm) in the coil driver path to dominate the current noise to the coils at ~200 Hz where we want to measure the squeezing.
2. What should the shape of this de-whitening filter be?
   • The DAC noise was measured to be ~1 uV/rtHz at 200 Hz. 
   • The Johnson noise spectral density of 4.5 kohm at 300 K is ~9 nV/rtHz. 
   • So we need ~60dB of attenuation at 200 Hz relative to DC. Currently, they have ~80dB of attenuation at 200 Hz.
   • However, we also need to consider the control signal multiplied by the inverse of this shape in the digital domain (required for overall flat shape). This should not saturate the DAC range.
   • Furthermore, we'd like for the shape to be such that we don't have a large transient when transitioning between high range and low noise modes. We should use the DARM control signal estimate to inform this choice.
3. What about the electronics noise of the de-whitening filter itself?
   • This shows up at the input of the coil driver stage, and gets transmitted to the coil with unity gain. 
   • So we should aim for < 3nV/rtHz at 200 Hz, such that we are dominated by the Johnson noise of the 4.5 kohm series resistance [the excess will be 5%]. 
   • This can be realized by using the passive network which is the final stage in the de-whitening (there is a unity gain output buffer stage implemented with LT1128, which we also have to account for).   

I will experiment with a few different shapes and investigate noise and de-whitened digital signal levels based on these considerations. At the very least, I guess we should remove the x3 gain on the ETM boards, they have already been bypassed for the ITMs. 

UNELOGGED: someone has changed Pianosa from Ubuntu/Dumbian into SL7. Might be hackers.

Donatella is now the only Ubuntu machine in the control room. I propose we keep it this way for another month and then go fully SL7 if we find no bugs with Pianosa/Rossa.

I moved the N2 check script and the disk usage checking script from the (sudo) crontab of nodus to the controls user crontab on megatron .

The fardest I can go back on channel C1: Vac_N2pres is  320 days

C1:Vac-CC1_Hornet Presuure gauge started logging Feb. 23, 2018

Did you update the " low N2 message"  email addresses?

we disabled logging the N2 Pressure to a text file, since it was filling up disk space. Now it just sends an email to our 40m mailing list, so we'll all get a warning.

The crontab uses the 'bash' version of output redirection '2>&1', which redirects stdout and stderr, but probably we just want stderr, since stdout contains messages without issues and will just fill up again.

Earlier today I cleared up most of the equipment at the X end near the seismometer to make the area walkable. 

In the process, I removed the connections to the temperature sensor and placed the wires on top of the can.

Shruti and Sandrine received 40m specific basic safety training this morning.

For the upcoming vent, we'd like to rotate the SOS towers to correct for the large YAW bias voltages used for DC alignment of the ITMs and ETMs. We could then use a larger series resistance in the DC bias path, and hence, reduce the actuation noise on the TMs. 

Today, I used the calibrated Oplev error signals to estimate what angular correction is needed. I disabled the Oplev loops, and drove a ~0.1 Hz sine wave to the EPICS channel for the DC yaw bias. Then I looked at the peak-to-peak Oplev error signal, which should be in urad, and calibrated the slider counts to urad of yaw alignment, since I know the pk-to-pk counts of the sine wave I was driving. With this calibration, I know how much DC Yaw actuation (in mrad) is being supplied by the DC bias. I also know the directions the ETMs need to be rotated, I want to double check the ITMs because of the multiple steering mirrors in vacuum for the Oplev path. I will post a marked up diagram later. 

Steve is going to come up with a strategy to realize this rotation - we would like to rotate the tower through an axis passing through the CoM of the suspended optic in the vertical direction. I want to test out whatever approach we come up with on the spare cage before touching the actual towers.

Here are the numbers. I've not posted any error analysis, but the way I'm thinking about it, we'd do some in air locking so that we have the cavity axis as a reference and we'd use some fine alignment adjust (with the DC bias voltages at 0) until we are happy with the DC alignment. Then hopefully things change by so little during the pumpdown that we only need small corrections with the bias voltages.

Some remarks:

1. Why the apparent difference between ITMs and ETMs? I think it's because the bias path resistors are 400 ohms on the ETMs, but 100 ohms on the ITMs. 
2. If we want the series resistance for the bias path to be 10 kohm, we'd only have ~800 urad actuation (for +10V DC), so this would be an ambitious level of accuracy.