40m QIL Cryo_Lab CTN SUS_Lab TCS_Lab OMC_Lab CRIME_Lab FEA ENG_Labs OptContFac Mariner WBEEShop
 40m Log, Page 38 of 339 Not logged in
ID Date Author Type Category Subject
11391   Sun Jul 5 18:14:13 2015 IgnacioUpdatePEMWilcoxon Accelerometer Huddle Test

Updated: On Thursday/Friday (sorry for late elog) I was messing with Eric's Wilcoxon 731A accelerometer huddle test data that was taken without the box and cables being adjusted properly. Anyways, I performed the three cornered hat analysis as he had done but I also performed a six cornered hat method as well instead of permuting around in pairs of three accelerometers. The following plots of the ASD's show the results,

It is interesting to note the improvement at low frequencies when six accelerometers are used instead of six while at higher frequencies we can clearly see how the results are worst than the three hat results.

I decided to take a mean of all six accelerometers measured ground signal as well as that for their computed selfnoises, this is plotted below,

Notice the obvious improvement along the entire frequency band of the measurements when all accelerometers are used in the data analysis.

I also performed some Wiener filtering of this data. There was an obvious improvement in the results,

The mean of the signals is also plotted below, just as I did with the cornered hat methods,

I also compared the mean self noise of the accelerometers against the manufacturers calculated selfnoise that Rana put up on Github. Both methods are compared against what the manufacturer claims,

As expected the measured noise curves of the Wilcoxon is not as good as what the manufactures stated. This should improve once we redo the huddle test with a better experimental setup. We have some pretty interesting results with the six cornered hat method at around 5 Hz, it is surprisingly accurate given how rough the calculations seemed to be.

I have attached my code for reference: code_accel.zipselfnoise_allsix.png

SEE attachments for better plots of the six accelerometers...

Attachment 5: code_accel.zip
Attachment 6: selfnoise_allsix_miso.png
Attachment 8: selfnoise_allsix.png
11407   Tue Jul 14 10:23:27 2015 IgnacioUpdateGeneralOptimal detector array placement thoughts

Over the past few days, I've been thinking about how to workout the details conerning Rana's request about a 'map' of the vicinity of the 40m interferometer. This map will take the positions of N randomly placed seismic sensors as well as the signals measured by each one of them and the calculated cross correlations between the sensors and between the sensors and the test mass of interest to give out a displacement vector with new sensor positions that are close to optimum for better seismic (and Newtonian) noise cancellation.

Now, I believe that much of the mathematical details have been already work out by Jenne in her thesis. She explains that the quantity of interest that we wish to minimize in order to find an optimal array is the following,

$R = \sqrt{1-\frac{\vec{C}_{SN}^T C_{SS}^{-1}\vec{C}_{SN} }{C_{NN}}}$

where  $\vec{C}_{SN}$ is the cross-correlation vector between the seismic detectors and the seismic (or Newtonian) noise, $C_{SS}$ is the cross-correlation matrix between the sensors and $C_{NN}$ is the seismic (or Newtonian) noise variance.

I looked at the paper that Jenne cited from which she obtained the above quantity and noted that it is a bit different as it contains an extra term inside the square root, it is given by

$R' = \sqrt{1-\frac{\vec{C}_{SN}^T (C_{SS}^{-1}+C_{\Sigma\Sigma})\vec{C}_{SN} }{C_{NN}}}$

where the new term, $C_{\Sigma\Sigma}$ is the matrix describing the self noise of the sensors. I think Jenne set this term to zero since we can always perform a huddle test on our detectors and know the self noise, thus effectively subtracting it from the signals of interest that we use to calculate the other cross correlation quantities.

Anyways, the quantity $R$ above is a function of the positions of the sensors. In order to apply it to our situation, I'm planning on:

1) Performing the huddle tests on our sensors, redoing it for the accelerometers and then the seismometers (once the data aquisition system is working... )

2) Randomly (well not randomly, there are some assumptions we can make as to what might work best in terms of sensor placement) place the sensors around the interferometer. I'm planning on using all six Wilcoxon 731A accelerometers, the two Guralps and the STS seismometer (any more?).

3) Measure the ground signals and use wiener filtering in order to cancel out their self noises.

4) From the measured signals and their present positions we should be able to figure out where to move the sensors in order to optimize subtraction.

i have also been messing around with Jenne's code on seismic field simulations with the hopes of simulating a version of the seismic field around the 40m in order to understand the NN of the site a little better... maybe. While the data aquisition gets back to a working state, I'm planning on using my simulated NN curve as a way to play around with sensor optimization before its done experimentally.

i have as well been thinking and learning a little bit about source characterization through machine learning methods, specially using neural networks as Masha did back in her SURF project on 2012. I have also been looking at Support vector machines. The reasons why I have been looking at machine learning algorithms is because of the nature of the everchanging seismic field around the interferometer. Suppose we find a pretty good sensor array that we like. How do we make sure that this array is any good at some time t after it has been found? If the array mostly deals with the usual seismic background (quiet) of the site of interest, we could incorporate machine learning techniques in order to mitigate any of the more random disturbances that happen around the sites, like delivery trucks, earthquakes, etc.

11423   Fri Jul 17 02:46:07 2015 IgnacioUpdateGeneralNew huddle test data for Wilcoxon 731A results

On Thursday, new huddle test data for the Wilcoxon 731A was aquired by Eric.

The difference between this new data and the previous data, is:

1) We used three accelerometers instead of six this time around.

2) We used a foam box, and clamped cables on the experimental set up as shown in the previous elog, http://nodus.ligo.caltech.edu:8080/40m/11389

I have analyzed the new data. Here I present my results.

The following plot shows the ASD's for the three accelerometers raw outputs as well as their error signals computed using the three cornered hat method,

As before, I computed the mean for the output signals of the accelerometers above as well as their mean self noise to get the following plot

,

Now, below I compare the new results with the results that I got from the old data,

Did the enclosure and cable clamping do much? Not really, according to the computed three hat results. Also, notice how much better, even if its a small improvement, we get from using six accelerometers and calculating their self noise by the six cornered hat method.

Now, I moved on to analyzing the same data with Wiener Filtering.

Here are again, the raw outputs, and the self noises of each individual accelerometer calculated using Wiener filtering,

The accelerometer in the Y direction is show a kind of funky signal at low frequncies. Why? Anyways, I calculated the mean of the above signals as I did for the three cornered hat method above to get the following, I also show the means of the signals computed with the old data using wiener filtering,

Is the enclosure really doing much? The Wiener filter that I applied to the huddle test old data gave me a much better, by an order of magnitude better self noise curve. Keep in mind that this was using SIX accelerometers, not THREE as we did this time. I want to REDO the huddle test for the WIlcoxon accelerometers using SIX accelerometers with the improved experimental setup to see what I get.

Finally, I compare the computed self noises above with what the manufacturer gives,

,

As I expected, the self noise using six accelerometers and Wiener filtering is the best I could work out. The three cornered hat method works out pretty well from 1 to 10 Hz, but the noise is just too much anywhere higher than 10 Hz. The enclosed, clamped, 3 accelerometer wiener filter result is an order of magnitude worse than the six accelerometer wiener filtered result, and two orders of magnitude worse than the three cornered hat method in the 1 to 10 Hz frequency band.

As I stated, I think we must performed the huddle test with SIX accelerometers and see what kind of results we get.

Attachment 1: selfnoise_allthree_threehat_enclosed.png
Attachment 2: selfnoise_3hat_enclosed_averages.png
Attachment 3: selfnoise_3hat_6hat_enc.png
Attachment 4: miso_wiener_enclosedall.png
Attachment 5: selfnoise_wiener_enclosed.png
Attachment 6: compare_encl.png
11424   Fri Jul 17 04:56:37 2015 IgnacioUpdateGeneralMCL Wiener filtering + FIR to IIR conversion using vectfit

We took data for the mode cleaner a while ago, June 30th I believe. This data contained signals from the six accelerometers and the three seismometers. In here I have only focused on the seimometer signals as witnesses in order to construct Wiener filters for each of the three seismometer signals (x,y,z) and for the combined seismometers signal. The following plot showing the ASD's shows the results,

Wiener filtering works beautifully for the seismometers. Note that subtraction is best when we use all three seismometers as the witnesses in the Wiener filter calculation, as can be clearly seen in the first plot above.

Now, I used vectfit to conver the Wiener FIR filters for each seismometer to their IIR versions. The following are the bode plots for the IIR filters,

For the x-direction seismometer,

For the y-direction seismometer

,

And for the z-direction seismometer,

The IIR filters were computed using 5 zeros and 5 poles using vectfit. That was the maximum number of poles that I could use wihtout running into trouble with matrices being almost singular in Matlab. I still need to figure out how to deal with this issue in more detail as fitting the y-seismometer was a bit problematic. I think having a greater number of poles will make the fitting a bit easier.

Attachment 1: Wiener_MCL_seismometers.png
Attachment 2: seisx_mag.png
Attachment 3: seisx_mag.png
Attachment 4: seisx_mag.png
Attachment 5: seisx_phase.png
Attachment 6: seisy_mag.png
Attachment 7: seisy_phase.png
Attachment 8: seisz_mag.png
Attachment 9: seisz_phase.png
11425   Sat Jul 18 06:12:07 2015 IgnacioUpdateGeneralMCL Wiener filtering + FIR to IIR conversion using vectfit (Update)

(updateAfter Eric gave me feedback on my previous elog post, I went back and fixed some of the silly stuff I stated.

First of all, I have come to realized that it makes zero sense to plot the ASD's of the mode cleaner against the seismometer noise. These measurements are not only quite different, but elementary, they posess different units. I have focused my attention to the MCL being Wiener filtered with the three siesmometer signals.

One of the major improvements that I make in the following analysis is,

1) Prefiltering; a band pass filter from 1 to 5 Hz, in order to emphasize subtraction of the bump shown in the figure below.

2) I have used vectfit exclusively in the 1 to ~5 Hz range, in order to model the FIR filter properly, as in, the kind of subtraction that we care about. Limiting myself to the 1 - 5 Hz range has allowed me to play freeley with the number of poles, hence being able to fit the FIIR filter properly with an IIR rational transfer function properly,

The resulting ASD's are shown below, in blue we show the raw MCL output, in blac the Wiener filter (FIR) result, and finally in black, the resultant data being filtered with the calculated IIR Wiener filter.

Now, in the following plots I show the IIR Wiener filters for each of the three seismometers,

X Seismometer,

For the Y seismometer,

and for the Z seismometer,

The matlab code for this work is attached: code.zip

Attachment 1: Wiener_MCL_seismometers_iir.png
Attachment 2: seisx_mag.png
Attachment 3: seisx_mag.png
Attachment 4: seisx_mag.png
Attachment 5: seisx_ph.png
Attachment 6: seisy_mag.png
Attachment 7: seisy_mag.png
Attachment 8: seisy_mag.png
Attachment 9: seisy_ph.png
Attachment 10: seisz_ph.png
Attachment 11: seisz_ph.png
Attachment 12: code.zip
Attachment 13: seisz_mag.png
Attachment 14: seisz_mag.png
Attachment 15: seisz_ph.png
11432   Tue Jul 21 05:17:09 2015 IgnacioUpdateGeneralMore clear accelerometer huddle tests results

I generated the following plots from the two sets of huddle test data we have for the accelerometers.

Old data: 6 accelerometers, no cables clamped, no box, 55 mins worth of data.

New data: 3 accelerometers, cables clamped, foam box put on placed and completely sealed, 20 mins worth of data.

I made sure to use the same Impuse response time (6 sec) and sampling frequency (256 Hz), as well as every other parameter for the calculations.

Top left: The resultant self noise curve using the new data, there is definitely and improvement in the 0.5-2 Hz band.

Top right: Resultant self noise using the old data, for the first set of three accelerometers.

Bottom left: Old data result for the remaining three accelerometers.

Bottom right: Old data result, using all six accelerometers as witnesses instead.

Attachment 1: new_data.png
Attachment 2: new_data.png
Attachment 3: old_data_1.png
Attachment 4: old_data_1.png
Attachment 5: old_data_2.png
11442   Thu Jul 23 21:12:14 2015 IgnacioUpdateGeneralNewer accelerometer huddle test data + detrending

In the last post concerning the self noise of the accelerometers, I mentioned the differences between the two data sets I was playing with. In order to give a more concrete analysis of the accelerometers self noise, we came to the conclusion that data taking time should be the same.

The plots below show the analysis for the following two datasets:

Old Data:  6 accelerometers, no cables clamped, no box, 55 mins worth of data.

Newer data: 3 accelerometers, cables clamped, foam box put on placed and completely sealed, 57.66 mins worth of data, (we had 20 mins of data in the previous data set).

Filter parameters were kept the same in all calculations, the only change that was added to the analysis was the detrending of the signals using the detrend function on Matlab, this improved the results as the plots show. I also plotted the error bars for the Wiener filtered result for reference as well as the manufactures claimed self noise.

After detrending the data and taking a longer dataset we can see the improvement brought upon by the foam box in the low frequency band of 0.5 - 10 Hz, as shown in the bottom left plot. There is a lot of noise that needs to be cancelled out from 10 Hz and on, which brings to our attention the plot on the bottom right corner. This plot uses the old data but uses all six accelerometers as witnesses, it also improved overall after having detrended the data, but what is peculiar about this plot is the fact that it manages to mitigate the higher frequency 10 - ~100 Hz band noise.

Attachment 1: old_data_1.png
Attachment 2: old_data_2.png
Attachment 3: new_data.png
Attachment 4: six_old_data.png
11446   Fri Jul 24 23:08:53 2015 IgnacioUpdatePEMMC1 accelerators moved for future huddle test

I have moved the MC1 accelerators and cable to the huddle test set up, in order to see how a six witness huddle test with the improved set up will do.

Here is a picture of the accelerometer set up,

Our motivation for doing this is to see if more witness signals used in the Wiener filter really does indeed improve subtraction, as it was seen from previous huddle results, specially in the region above 10 Hz.

11449   Tue Jul 28 05:00:03 2015 IgnacioUpdateGeneralSeismometer cans

I've have been talking a little bit with Steve about the seismometer enclosures.

We want to improve on the current stainless steel cans that cover the two Guralps at the end of the arms. In order to do this, we want to cover the interior of the cans with copper foil to improve the thermal conductivity of the enclosure to better control the temperature inside it. Ideally, we would want to copper plate the cans, but cost and difficulty has been an issue.

I have done some rough calculations and it seems that we need a copper layer of thickness being about a third that of the stainless steel can. This happens to be around 0.5-0.6 mm since we have 16 gauge (~1.6 mm) stainless steel cans.

After wrapping the cans interior with copper, we will insulate them with foam in order to improve its thermal inertia. We want to probably use the same foam that Megan has been using for her seismometer enclosure. I have yet to think about a heater, but something similar to Megans resistor thing would work only smaller. I would be placed inside the can, right on the center of its bottom in order to ditribute heat evenly.

11450   Tue Jul 28 09:31:35 2015 IgnacioUpdateGeneralNewest accelerometer huddle test

I downloaded new accelerometer huddle test data from last night and analyzed it. This new data set is ~55 mins and uses the same Wiener filter parameters as previous huddle test analysis, the main difference being six accelerometers used in the Wiener filter and the improved experimental set up.

After computing the ASD for the self noise for each of the six accelerometers, (being witnessed by the remaining five), we get,

Now computing the mean of the above signals and the corresponding error bars gives the following result,

Comparing to prevoius huddle tests, I can note two trends on the Wiener subtraction:

1) When using six accelerometers, the subtraction above ~8 Hz drastically improves.

2) When using three accelerometers, there is better cancellation in the 1-5 Hz region, see http://nodus.ligo.caltech.edu:8080/40m/11442. This might have to do with how much more closer the accelerometers are to each other?

11457   Wed Jul 29 10:34:42 2015 IgnacioSummaryLSCCoherence of arms and seismometers

Jessica and I took 45 mins  (GPS times from 1122099200 to 1122101950) worth of data from the following channels:

C1:IOO-MC_L_DQ (mode cleaner)
C1:LSC-XARM_IN1_DQ (X arm length)
C1:LSC-YARM_IN1_DQ (Y arm length)

and for the STS, GUR1, and GUR2 seismometer signals.

The PSD for MCL and the arm length signals is shown below,

I looked at the coherence between the arm length and each of the three seismometers, plot overload incoming below,

For the coherence between STS and XARM and YARM,

For GUR1,

Finally for GUR2,

A few remarks:

1) From the coherence plots, we can see that the arm length signals are coherent with the seismometer signals the most from 0.5 - 50 Hz. This is most evident in the coherence with STS. I think subtraction will be most useful in this range. This agrees with what we see in the PSD of the arm length signals, the magnitude of the PSD starts increasing from 1 Hz and reaches a maximum at about 30 Hz. This is indicative of which frequencies most of the noise is present.

2) Eric did not remember which of  GUR1 and GUR2 corresponded to the ends of XARM and YARM. So, I went to the end of XARM, and jumped for a couple seconds to disturb whatever Gurald was in there. Using dataviewer I determined it was GUR1. Anyways, my point is, why is GUR1 less coherent with both arms and not just XARM?  Since it is at the end of XARM, I was expecting GUR1 to be more coherent with XARM. Is it because, though different arms, the PSD's of both arms are roughly the same?

3) Similarly, GUR2 shows about the same levels of coherence for both arms, but it is more coherent. Is GUR2 noisier because of its location?

Code: ARMS_COH.m.zip

Attachment 1: PSD_ARMS_MCL.png
Attachment 2: XARM_STS_COH.png
Attachment 3: YARM_STS_COH.png
Attachment 4: XARM_GUR1_COH.png
Attachment 5: YARM_GUR1_COH.png
Attachment 6: XARM_GUR2_COH.png
Attachment 7: YARM_GUR2_COH.png
Attachment 8: ARMS_COH.m.zip
11460   Wed Jul 29 17:51:56 2015 IgnacioUpdatePEMAccelerators moved back to MC1 and MC2

We are done taking accelerator huddle test data. So I moved back all six accelerometers and cables to MC1 and MC2. I also relabel each of the accelerometers properly since the labels on them were confusing.

QED

Attachment 1: FullSizeRender.jpg
Attachment 2: FullSizeRender_2.jpg
11462   Thu Jul 30 02:06:20 2015 IgnacioUpdateIOOMC2 <-> MCL Actuator TF fitted

Eric downloaded MC2 to MCL transfer function data (H) as well as its inverse, MCL to MC2 (Hinv). He also downloaded new MCL and MC2 data.

I used vectfit to fit the MC2 to MCL transfer function,

The ZPK parameters for this fit were,

Zeros              1278.36719876674 + 0.00000000000000i                    -100.753249679343 + 0.00000000000000i                    -18.6014192997845 + 13.0294910760217i                    -18.6014192997845 - 13.0294910760217i

Poles              -1.11035771175328 + 7.03549674098987i                    -1.11035771175328 - 7.03549674098987i                    -18.8655320274072 + 0.00000000000000i                    -690.294337433234 + 0.00000000000000i

Gain               0.00207206036014220

Using the above vectfit model, I filtered the raw MC2 signal to get 'MCL'. The PSD's of the raw MCL data and the filtered MC2 result is shown below,

The lack of accuracy of the transfer function at replicating MCL at frequencies lower than 0.7Hz is expected, the vectfit model I generated fails to follow accurately the raw transfer function data. My question: Does it matter? My guess: Probably not. In order to mitigate seismic noise from the mode cleaner we are mainly concerened with the 1-3 Hz region.

I also used vectfit to fit the transfer function for MCL to MC2,

This one was harder to fit accurately for some reason, I could do it with four pairs of zeros and poles but it took some preweighting.

The ZPK parameters for the above fit were,

Zeros              0.173068278283995 + 0.00000000000000i                    0.995140531040529 + 0.0268079821980457i                    0.995140531040529 - 0.0268079821980457i                    0.894476816129099 + 0.00000000000000i

Poles              -19.9566906920707 + 18.0649464375308i                    -19.9566906920707 - 18.0649464375308i                    -109.275971483008 + 0.00000000000000i                    -1791.88947801703 + 0.00000000000000i

Gain               1237.46417532120

Similarly, using this ZPK model, I filtered the MCL signal to get 'MC2'. I plotted the PSD for the MC2 signal and the filtered MCL to get,

Again, the lack of accuracy of the filtered MC2 at replicating MCL below 0.7 Hz and above 12 Hz is due to the inverse transfer function failing to converge in these ranges.

Attachment 1: TF_BODE.png
Attachment 2: MC2_2_MCL.png
Attachment 3: TF_INV_BODE.png
Attachment 4: MCL_2_MC2.png
11472   Thu Jul 30 19:12:52 2015 IgnacioUpdateIOOYAW and PIT WFS Wiener filtering

Rana pointed out that another way to mitigate seismic motion at in the mode cleaner would be to look at the YAW and PITCH output  channels of the WFS sensors that control the angular alignment of the mode cleaner.

C1:IOO-WFS1_YAW_OUT_DQ

C1:IOO-WFS1_PIT_OUT_DQ

And did some quick offline Wiener filtering with no preweighting, the results are shown in the PSD's below,

and

I'm quite surprised at the Wiener subtraction obtained for the YAW signal, it required no preweighting and there is about an order of magnitude improvement in our region of interest, 1-3 Hz. The PIT channel didn't do so bad either.

Attachment 1: YAW.png
Attachment 2: PIT.png
11485   Thu Aug 6 21:03:45 2015 IgnacioHowToWienerFilteringHow to do online static IIR Wiener filtering

In order to do online static IIR Wiener filtering one needs to do the following,

1) Get data for FIR Wiener filter witnesses and target.

2) Measure the transfer function (needs to be highly coherent, ~ 0.95 for all points) from the actuactor to the control signal of interest (ie. from MC2_OUT to MC_L).

3) Invert the actuator transfer function.

4) Use Vectfit or (LISO) to find a ZPK model for the actuator transfer and inverse transfer functions.

5) Prefilter your witness data with the actuator transfer function, to take into account the actuator to control transfer function when designing the offline Wiener FIR filter.

6) Calculate the frequency response for each witness from the FIR coefficients.

7) Vectfit the frequency reponses to a ZPK model, this is the FIR to IIR Wiener conversion step.

8) Now, either, divide the IIR transfer function by the actuator transfer function or more preferably, multiply by the inverse transfer function.

9) Use Quack to make SOS model of the IIR Wiener filter and inverse transfer function product that goes into foton for online implementation.

10) Load it into the system.

The block diagram below summarizes the steps above.

Attachment 1: iir.png
11488   Mon Aug 10 22:18:19 2015 IgnacioUpdateIOOReady to do some online mode cleaner subtraction

I'm attaching a SISO IIR Wiener filter here for reference purposes that will go online either tonight or tomorrow evening. This is a first test to convince myself that I can get this to work, MISO IIR filters are close to being ready and will soon be employed.

This Wiener filter uses the STS-X channel as a witness and MCL as target. The bode plot for the filter is shown below,

The performance of the FIR and IIR Wiener filters and the ammount of subtraction achive for MCL is shown below,

Output from quack to be loaded with foton: filter.zip

K bye.

Attachment 1: stsx.png
Attachment 2: performance.png
Attachment 3: filter.zip
11492   Tue Aug 11 11:30:19 2015 IgnacioUpdateIOOSISO (T240-X) FF of MCL

Last night we finally got some online subtraction going. The filter used is described in the post this eLOG is @eLOG 11488

The results were as follow:

The filter worked as expected when subtracting noise out of MCL,

There is about a factor of 6 subtraction at the ~3Hz resonant peak. The static IIR filter predicted a factor of 6-7 subtraction of this peak as well.

The 1.2 Hz resenonant feature improved by a factor of 3. This should improve quite drastically when I implement the y-channel of the T240 seismo.

There is some high frequency noise being injected, not very noticeable, but present.

We then took a look at the power in the MC when the filter was on,

The power being transmitted in the cavity was not as stable as with the feedforward on. We believe that the filter is not at fault for this as Eric mentioned to me that the MC2 actuator lacked some sort of compensation that I need to understand a bit better.

YARM was then locked when the filter was on and we took a look at how it was doing. There was stationary sound arising from the locking of the YARM, leading us to believe that the filter might have injected some noise in the signal. IT DID.

The filter injected nasty high frequency noise at YARM from 11 Hz and on. This is to be expected since the filter did not roll off to zero at high frequencies. Implementing a 1/f rolloff should mitigate some of the injected noise.

Also, as one can see above, subtraction by around a factor of 2 or so, was induced by the mode cleaner feedforward subtraction.

Attachment 1: MCL.png
Attachment 2: MCTRANS.png
Attachment 3: YARM.png
11496   Wed Aug 12 01:32:18 2015 IgnacioUpdateIOOImproved SISO (T240-X) FF of MCL

In my previous elog:11492, I stated that in order to improve the subtraction and reduce the injection of high frequency noise we want the filter's magnitude to have a 1/f rolloff.

I implemented this scheme on the filter SISO filter previously analyzed. The results are shown below.

The filters bode plot:

The nice 1/f rollof is the main change here. Everything else remained pretty much the same.

The predicted FIR and IIR subtractions:

Everything looks right but that hump at 8 Hz. I used 8 pairs of poles/zeros to get this subtraction.

The online MCL subtraction:

This looks better than I expected. One has to keep in mind that I ran this at 1 AM. I wonder how well this filter will do during the noisier hours of the day. The RMS at high frequencies doesn't look great, there will definitely be noise being injected into the YARM signal at high frequencies.

Measuring the YARM signal:

There is still noise being injected on YARM but it is definitely much better than the previous filter. I'm thinking about doing some IIR subtraction on the arms now to see if I can get rid of the noise that is being injected that way, but before I embark on that quest I will rething my prefiltering.

The plot below shows the ratio of the unfiltered versus filtered ASDs for the FIR and IIR subtraction predictions as well as for the measured online IIR subtraction. Positive dB means better subtraction.

Attachment 1: filter.png
Attachment 2: stsx.png
Attachment 3: mclonline.png
Attachment 4: yarmonline.png
Attachment 5: sub.png
11499   Wed Aug 12 16:39:46 2015 IgnacioUpdateIOOMISO WIener (T240-X and T240-Y) FF of MCL

Last night I performed some MISO FF on MCL using the T240-X and T240-Y as witnesses. Here are the results:

Filter:

T240-X

T240-Y

Training data + Predicted FIR and IIR subtraction:

Online subtraction results:

MCL
YARM

Subtraction performace:

Attachment 1: stsx.png
Attachment 2: stsy.png
Attachment 3: performance.png
Attachment 4: sub.png
Attachment 5: mcliir.png
Attachment 6: yarmiir.png
11500   Wed Aug 12 16:48:26 2015 IgnacioUpdateIOOBetter? Nope. MISO WIener (T240-X and T240-Y) FF of MCL

Last night, I also worked on a "better" (an improvement, I thought) of the MISO Wiener filter (T240-X and T240-Y witnesses) IIR filter. The FF results are shown below:

Filter:

T240-X

T240-Y

Training data + Predicted FIR and IIR subtraction:

Online subtraction results:

MCL
YARM

Subtraction performace:

Although the predicted FIR and IIR results are "better" than the previous MISO filter, the subtraction performance for this filter is marginally better if not worse (both peak at 15 dB, in slightly different regions).

Attachment 1: stsx.png
Attachment 2: stsy.png
Attachment 3: performance.png
Attachment 4: mcliir.png
Attachment 5: yarmiir.png
Attachment 6: sub.png
11501   Wed Aug 12 22:33:36 2015 IgnacioUpdateIOORe-measured MC2 -> MCL TF

Since I will need to do transfer function measurements in order to implement FF for the arms and the MC2's yaw and pitch channels, I decided to practice this by replicating the transfer function measurement Eric did for MC2 to MCL. I followed his procedure and the data that I aquired for the TF looked as shown below,

About five minutes of data were taken (0.05 Hz resolution, 25 averages) by injecting noise from 1 to 100 Hz. The TF coherence looked as below,

Attachment 1: bode_TF.png
Attachment 2: Coherence.png
11503   Thu Aug 13 20:32:07 2015 IgnacioUpdateLSCWorking towards YARM FF

The mode cleaner FF static filtering is by no means done. More work has to be done in order to succefuly implement it, by the means of fine tuning the IIR fit and finding better MISO Wiener filters.

I have begun to look at implementing FF to the YARM cavity for several reasons.

1) Even if the mode cleaner FF is set up as best as we can, there will still be seismic noise coupling into the arm cavities.

2) YARM is in the way of the beam path. When locking the IFO, one locks YARM first, then XARM. This means that it makes sense to look at YARM FF first rather than XARM.

3) XARM FF can't be done now since GUR2 is sketchy.

I'm planning on using this eLOG entry to document my Journey and Adventures (Chapter 2: YARM) to the OPTIMAL land of zero-seismic-noise (ZSN) at the 40m telescope.

11504   Thu Aug 13 23:57:33 2015 IgnacioUpdateLSCYARM coherence plots

I took data from 1123495750 to 1123498750 GPS time (Aug 13 at 3AM, 50 mins of data) for  C1:LSC-YARM_OUT_DQ, and all T240 and GUR1 channels.

Here is the PSD of the YARM_OUT, showing the data that I will use to train the FIR filter:

Coherence plots for YARM and all channels of T240 and GUR1 sesimometers are shown below. This will help determine what regions to preweight the best before computing FIR filter. They also show how GUR1 is back to work compared to those of elog:11457.

Attachment 1: YARM_psd.png
Attachment 2: YARM_GUR1_COH.png
Attachment 3: YARM_STS_COH.png
Attachment 4: YARM_GUR1_COH.png
11508   Fri Aug 14 21:40:26 2015 IgnacioUpdateLSCQuick static offline subtractions of YARM

Plotte below are the resultant subtractions for YARM using different witness configurations,

The best subtraction happens with all the channels of both the GUR1 and T240 seismometers, but one gets just as good subtraction without using the z channels as witnesses.

Also, why is the T240 seismometer better at subtracting noise for YARM compared to what GUR1 alone can acomplish? Using only the X and Y channels for the T240 gave the third best subtraction(purple trace).

My plan for now is as follows:

1) Measure the transfer function from the ETMY actuator to the YARM control signal

2) Collect data for YARM when FF for MCL is on in order to see what kind of subtractions can be done.

Attachment 1: arms_wiener.png
11510   Sat Aug 15 02:10:35 2015 IgnacioUpdateLSCMCL FF => YARM FF

In my last post I calculated the different subtractions (offline) that could be done to YARM alone just to get a sense of what seismometers were better witnesses for the Wiener filter calculation.

In this eLOG I show what subtractions can be done when the MCL has FF on (as well as Eric's PRC FF), with the SISO filter described on elog:11496.

The plot below shows what can be done offline,

What is great about this results is that the T240-X and T240-Y channels are plenty enough to mitigate any remaining YARM seismic noise but also to get rid of that nasty peak at 55 Hz induced by the MCL FF filter.

The caviat, I haven't measured the TF for the ETMY actuator to YARM control signal. I need to do this and recompute the FIR filters with the prefiltered witnesses in order to move on to the IIR converions and online FF!

Attachment 1: YARM_LIVES.png
11515   Wed Aug 19 00:55:35 2015 IgnacioUpdateLSCLSC-YARM-EXC to LSC-YARM-IN1 TF measurement + error analysis

Yesterday, Rana, Jessica and I measured the Transfer function from LSC-YARM-EXC to LSC-YARM-IN1.

The plot below shows the magnitude and the phase of the measured transfer function. It also shows the normalized standard error in the estimated transfer function magnitude; the same quantity can be applied to the phase, only in this case it is interpreted as its standard deviation (not normalized). It is given by

$\frac{[1-\gamma_{xy}^2(f)]^{1/2}}{|\gamma_{xy}(f)|\sqrt{2n_{d}}}$

where $\gamma_{xy}^2(f)$ is the ordinary coherence function and $n_{d}$ is the number of averages used at each point of the estimate, in the case here we used 9 averages. This quantity is of interest to us in order to understand how the accuracy of transfer function measurement affects the ammount of subtraction that can be achieved online.

Since this transfer function is flat from 1-10 Hz (out of phase by 180 deg), this means that we can apply our IIR wiener filters direclty into YARM without taking into account the TF by prefiltering our witnesses with it. Of course this is not the case if we care about subtractions at frequencies higher than 10 Hz, but since we are dealing with seismic noise this is not a concern.

The coherence for this transfer function measurement is shown below,

11516   Wed Aug 19 01:45:10 2015 IgnacioUpdateIOODoubly Improved SISO (T240-X) FF of MCL

Today I tried and doubly-improved SISO FF filter on MCL. This filter has a stronger rolloff than the previous SISO filters I have tried. The rolloff most definelty helped towards reducing the ammount of noise being injected into YARM. Below is the usual stuff:

Filter:

T240-X (SISO)

Training data + Predicted FIR and IIR subtraction:

Online subtraction results:

MCL
YARM
MCL TRANS

Subtraction performace:

11521   Thu Aug 20 18:08:28 2015 IgnacioFrogs40m upgradingFatality. Something broke.

So I made coffee at 1547 and was astonished to find the above. Its a sad, very sad day.

At first I thought that something (a gravity wave?) or someone, accidentally hit the thing and it fell and broke. But Koji told me that the janitor was cleaning around the thing and it did indeed fell accidentally.

11532   Thu Aug 27 01:41:41 2015 IgnacioUpdateIOOTriply Improved SISO (T240-X) FF of MCL

Earlier today I constructed yet another SISO filter for MCL. The one thing that stands out about this filter is its strong roll off . This prevents high frequency noise injection into YARM. The caviat, filter performance suffered  quite a bit, but there is subtraction going on.

I have realized that Vectfit lacks the ability of constraining the fits it produces, (AC coupling, rolloff, etc) even with very nitpicky weighting. So the way I used vectfit to produce this filter will be explained in a future eLOG, I think it might be promising.

Anyways, the usual plots are shown below.

Filter:

T240-X (SISO)

Training data + Predicted FIR and IIR subtraction:

Online subtraction results:(High freq. stuff shown for noise injection evaluation of the filter)

MCL

YARM

Subtraction performace:

11535   Fri Aug 28 00:59:55 2015 IgnacioUpdateIOOFinal SISO FF Wiener Filter for MCL

This is my final SISO Wiener filter for MCL that uses the T240-X seismo as its witness.

The main difference between this filter and the one on elog:11532 is the actual 1/f rolloff this filter pocesses. My last filter had a pair of complex zeroes at 2kHz, that gave the filter some unusual behavior at high frequencies, thanks Vectfit. This filter has 10 poles and 8 zeroes, something Vectfit doesn't allow for and needs to be done manually.

The nice thing about this filter is the fact that Eric and I turned this filter on during his 40 min PRFPMI lock last night, Spectra for this is coming soon.

This filter lives on the static Wiener path on the OAF machine, MCL to MC2, filter bank 7.

Anyways, the usual plots are shown below.

Filter:

T240-X (SISO)

Training data + Predicted FIR and IIR subtraction:

Online subtraction results:(High freq. stuff shown for noise injection evaluation of the filter)

MCL

YARM

Subtraction performace:

11536   Fri Aug 28 02:20:35 2015 IgnacioUpdateLSCPRFPMI and MCL FF

A day late but here it is.

Eric and I turned on my SISO MCL Wiener filter elog:11535 during his PRFPMI 40min lock. We looked at the CARM_IN and CARM_OUT signals during the lock and with the MCL FF on/off. Here is the spectra:

11541   Sat Aug 29 04:53:24 2015 IgnacioUpdateIOOMCL Wiener Feedforward Final Results

After fighting relentlessly with the mode cleaner, I believe I have achieved final results

I have mostly been focusing on Wiener filtering MCL with a SISO Wiener filter for a reason, simplicity. This simplicity allowed me to understand the dificulties of getting a filter to work on the online system properly and to develope a systematic way of making this online Wiener filters. The next logical step after achieving my final SISO Wiener filter using the T240-X seismometer as witness for MCL (see elog:11535) and learning how to produce good conditioned Wiener filters was to give MISO Wiener filtering of MCL a try.

I tried performing some MISO filtering on MCL using the T240-X and T240-Y as witnesses but the procedure that I used to develope the Wiener filters did not work as well here. I made the decision to ditch it and use some of the training data I saved when the SISO (T240-X) filter was runing overnight to develope another SISO Wiener filter for MCL but this time using T240-Y as witness. I will compare how much more we gain when doing MISO Wiener filtering compared to just a bunch of SISO filtering in series, maybe a lot, or little.

I left both filters running overnight in order to get trainining data for arm and WFS yaw and pitch subtractions.

The SISO filters for MCL are shown below:

The theoretical FIR and IIR subtractions using the above filters:

Running the filters on the online system gave the following subtractions for MCL and YARM:

Comparing the subtractions using only the T240-X filter versus the T240-X and T240-Y:

11543   Sun Aug 30 10:57:29 2015 IgnacioUpdateIOOMCL Wiener Feedforward Final Results

Big thumbnails? Could it have been this? elog:11498.

Anyways, I fixed the plots and plotted an RMS that can actaully be read in my original eLOG. I'll see what can be done with the MC1 and MC2 Wilcoxon (z-channel) for online subtractions.

11546   Sun Aug 30 13:55:09 2015 IgnacioUpdateIOOSummary pages MCF

The summary pages show the effect of the MCL FF on MCF (left Aug 26, right Aug 30):

I'm not too sure what you meant by plotting the X & Y arm control signals with only the MCL filter ON/OFF. Do you mean plotting the control signals with ONLY the T-240Y MCL FF filter on/off? The one that reduced noise at 1Hz?

11547   Sun Aug 30 23:47:02 2015 IgnacioUpdateIOOMISO Wiener Filtering of MCL

I decided to give MISO Wiener filtering a try again. This time around I managed to get working filters. The overall performance of these MISO filters is much better than the SISO I constructed on elog:11541 .

The procedure I used to develope the SISO filters did not work well for the construction of these MISO filters. I found a way, even more systematic than what I had before to work around Vectfit's annoyances and get the filters in working condition. I'll explain it in another eLOG post.

Anyways, here are the MISO filters for MCL using the T240-X and T240-Y as witnesses:

Now the theoretical offline prediction:

The online subtractions for MCL, YARM and XARM. I show the SISO subtraction for reference.

And the subtraction performance:

11549   Mon Aug 31 09:36:05 2015 IgnacioUpdateIOOMISO Wiener Filtering of MCL

MISO Wiener filters for MCL kept the mode cleaner locked for a good 8+ hours.

11550   Mon Aug 31 14:15:23 2015 IgnacioUpdateIOOMeasured the MC_F whitening poles/zeroes

I measured the 15 Hz zero and the 150 Hz pole for the whitening filter channels of the Generic Pentek board in the IOO rack. The table below gives these zero/pole pairs for each of the 8 channels of the board.

 channel zero [Hz] pole [Hz] Chan 1 15.02 151.05 C1:ASC-POP_QPD_YAW 2 15.09 150.29 C1:ASC-POP_QPD_PIT 3 14.98 150.69 C1:ASC-POP_QPD_SUM 4 14.91 147.65 C1:ALS-TRX 5 15.03 151.19 C1:ALS-TRY 6 15.01 150.51 --- 7 14.95 150.50 C1:IOO-MC_L 8 15.03 150.93 C1:IOO-MC_F

Here is a plot of one of the measured transfer functions,

and the measured data is attached here: Data.zip

EQ: I've added the current channels going through this board.

More importantly, I found that the jumpers on channel one (QPD X) were set to no whitening, in contrast to all other channels. Thus, the POP QPD YAW signals we've been using for who knows how long have been distorted by dewhitening. This has now been fixed.

Hence, the current state of this board is that the first whitening stage is disabled for all channels and the second stage is engaged, with the above parameters.

Attachment 1: Data.zip
11552   Tue Sep 1 06:58:11 2015 IgnacioUpdateWienerFilteringMCL FF => WFS1 and WFS2 FF => ARMS FF

I took some training data during Sunday night/Monday morning while the MCL MISO FF was turned on. We wanted to see how much residual noise was left in the WFS1/WFS2 YAW and PITCH signals.

The offline subtractions that can be achieved are:

For WFS1

For WFS2

I need to download data for these signals while the MCL FF is off in order to measure how much subtraction was achived indirectly with the MCL FF. In a previous elog:11472, I showed some offline subtractions for the WFS1 YAW and PITCH before any online FF was implemented either by me or Jessica. From the plots of that eLOG, one can clearly see that the YAW1 signal is clearly unchanged in the sense of how much seismic noise was mitigated indirectly torugh MCL.

Koji has implemented the FF paths (thank you based Koji) necessary for these subtractions to be implemented. The thing to figure out now is where we want to actually actuate and to measure the corresponding transfer functions. I will try to have either Koji or Eric help me measure some of these transfer functions.

Finally, I looked at the ARMS and see what residual seismic noise can be subtracted

I'm not too concerned about noise in the arms as if the WFS subtractions turn out to be promising then I expect for some of the arms seismic noise to go down a bit further. We also don't need to measure an actuator transfer function for arm subtractions, give that its essentially flat at low frequencies, (less than 50 Hz).

11553   Tue Sep 1 10:26:24 2015 IgnacioUpdateIOOMore MCL Subtractions (Post FF)

Using the training data that was collected during the MISO MCL FF. I decided to look at more MCL subtractions but this time using the accelerometers as Rana suggested.

I first plotted the coherence between MCL and all six accelerometers and the T240-Z seismometer.

For 1 - 5 Hz, based on coherence, I decided to do SISO Wiener filtering with ACC2X and MISO Wiener filtering with ACC2X and ACC1Y. The offline subtractions were as follows (RMS plotted from 0.1 to 10 Hz):

The subtractions above look very much like what you would get offline when using the T240(X,Y) seismometeres during MISO Wiener filtering. But this data was taken with the MISO filters on. This sort of shows the performance deterioration when one does the online subtractions. This is not surprising since the online subtraction performance for the MISO filters, was not too great at 3 Hz. I showed this in some other ELOG but I show it again here for reference:

Anyways, foor 10 - 20 Hz, again based on coherence, I decided to do SISO Wiener filtering with ACC2Z and MISO Wiener filtering with ACC2Z and ACC1Z (RMS plotted from 10 to 20 Hz):

I will try out these subtractions online by today. I'm still debating wether the MISO subtractions shown here are worth the Vectfit shananigans. The SISO subtractions look good enough.

Attachment 4: mclxycoh.png
11563   Thu Sep 3 00:45:25 2015 IgnacioUpdateIOORemeasured MC2 to MCL TF + Improved subtraction performance

Today, I remeasured the transfer function for MC2 to MCL in order to improve the subtraction performance for MCL and to quantify just how precisely it needs to be.

Here is the fit, and the measured coherence. Data is also attached here: TF.zip

OMG, I forgot to post the data and any residuals. LOL!

The transfer function was fitted using vectfit with a weighting based on coherence being greater than 0.95.

I then used the following filters to do FF on MCL online:

Here are the results:

Performance has definelty increased when compared to previous filters. The reason why I think we still have poor performance at 3 Hz, is 1) When I remeasured the transfer function, Eric and I were expecting to see a difference on its shape due to the whitening filters that were loaded a couple days ago. 2) Assuming the transfer function is correct, there is poor coherence at 3 Hz 3) The predicted IIR subtraction is worst at this frequency.

Attachment 1: TF.zip
11573   Fri Sep 4 08:00:49 2015 IgnacioUpdateCDSRC low pass circuit (1s stage) of Pentek board

Here is the transfer function and cutoff frequency (pole) of the first stage low pass circuit of the Pentek whitening board.

Circuit:

R1 = R2 = 49.9 Ohm, R3 = 50 kOhm, C = 0.01uF. Given a differential voltage of 30 volts, the voltage across the 50k resistor should be 29.93 volts.

Transfer Function:

Given by,

$H(s) = \frac{1.002\text{e}06}{s+1.002\text{e}06}$

So low pass RC filter with one pole at 1 MHz.

I have updated the schematic, up to the changes mentioned by Rana plus some notes, see the DCC link here: [PLACEHOLDER]

I should have done this by hand...

Attachment 1: circuit.pdf
11574   Fri Sep 4 09:23:32 2015 IgnacioUpdateCDSModified Pentek schematic

Attached is the modifed Pentek whitening board schematic. It includes the yet to be installed 1nF capacitors  and comments.

Attachment 1: schematic.pdf
11584   Wed Sep 9 11:00:49 2015 IgnacioUpdateIOOLast Wiener MCL subtractions

On Thursday night (sorry for the late elog) I decided to give the MCL FF one more try.

I first remeasured the actuator transfer function because previous measurements had poor coherence ~0.5 - 0.7  at 3 Hz. I did a sine swept to measure the TF.

Raw transfer function:

The data is attached here: TF.zip

Then I made Wiener filters by fitting the transfer function data with coherence > 0.95 (on the left). Fitting all the data (on the right). Here are the filters:

The offline subtractions (high coh fit on left, all data fit on right). Notice the better IIR performance when all the TF data was fitted.

The online results: (these were aquired by taking five DTT measurements with 15 averages each and then taking the mean of these measurements)

And the subtraction performance:

Attachment 3: TF.zip
11590   Thu Sep 10 09:37:34 2015 IgnacioSummaryIOOFilters left on MCL static module

The following MCL filters were left loaded in the T240-X and T240-Y FF filter modules (filters go in pairs, both on):

FM7: SISO filters for MCL elog:11541

FM8: MISO v1 elog:11547

FM9: MISO v1.1 Small improvement over MISO v1

FM10 MISO v2 elog:11563

FM5 MISO v3.1 elog:11584 (best one)

FM6 MISO 3.1.1 elog:11584 (second best one)

11713   Mon Oct 26 18:10:38 2015 IgnacioUpdateIOOLast Wiener MCL subtractions

As per Eric's request, here is the code and TF measurement that was used to calculate the MC2 FF filter that is loaded in FM5. This filter module has the filter with the best subtraction performance that was achieved for MCL.

code_TF.zip

Attachment 1: code_TF.zip
11493   Tue Aug 11 11:56:36 2015 Ignacio, JessicaUpdatePEMWasps obliterated maybe...

The wasp terminator came in today. He obliterated the known wasp nest.

We discovered a second wasp nest, right next to the previous one...

Jessica wasn't too happy the wasps weren't gone!

11530   Tue Aug 25 16:33:31 2015 Ignacio, SteveConfigurationPEMSeismometer enclosure copper foil progress

Steve ordered about two weeks ago a roll of 0.5 mm thick copper foil to be used for the inside of the seismometer cans. The foil was then waterjet cut by someone in Burbank to the right dimensions (in two pieces, a side and a bottom for each of the three cans).

Today, we glued the copper foil (sides only) inside the three seismometer cans. We used HYSOL EE4215/HD3561(Data Sheet) as our glue. It is a "high impact, low viscocity, room temperature cure casting" that offers "improved thermal conductivity and increased resistance to heat and thermal shock." According to Steve, this is used in electronic boards to glue components when you want it to be thermal conductive.

We are going to finish this off tomorrow by gluing the bottom foil to the cans. The step after this involves soldering the side to the bottow and where the side connects. We have realized that the thermal conductivity of the solder that we are using is only ~50. This is 8 times smaller than that of copper and wil probably limit how good a temperature gradient we will have.

Some action shots,

6003   Thu Nov 24 15:48:27 2011 IllustratorUpdateelogelogd gained an immunity to googlebot

5020   Fri Jul 22 17:01:41 2011 Iron ManFrogsGeneralProof that Alberto lived through his Iron Man!

 208 Alberto Stochino 67/129 585/1376 36:02 1:52 830 6:41 2:38:58 21.1 296 4:58 56:33 2:13:40 - 5:40:19

4854   Wed Jun 22 12:29:57 2011 IshwitaSummaryAdaptive FilteringWeekly summary

I started on the 16th with a very intense lab tour & was fed with a large amount of data (I can't guarantee that I remember everything....)

Then... did some (not much) reading on filters since I'm dealing with seismic noise cancellation this summer with Jenne at the 40m lab.

I'll be using the Streckeisen STS-2 seismometers & I need to use the anti aliasing filter board that has the 4 pin lemo connectors with the seismometers & its boxes that require BNC connectors. I spent most of the time trying to solder the wires properly into the connectors. I was very slow in this as this is the first time I'm soldering anything.... & till now I've soldered 59 wires in the BNC connectors....

ELOG V3.1.3-