A spiral shape is a very good choice for array configurations to measure spatial spectra. It produces small aliasing. How important is array configuration for NN subtraction? Again: plane waves, wave speeds {100,200,600}m/s, 2D, SNR~10. The array response looks like Stonehenge:

A spiral array is doing a fairly good job to measure spatial spectra:

The injected waves are now represented by dots with radii proportional to the wave amplitudes (there is always a total of 12 waves, so some dots are not large enough to be seen). The spatial spectra are calculated from covariance matrices, so theory goes that spatial spectra get better using matched-filtering methods (another thing to look at next week...).

Now the comparison between NN subtraction using 20 seismometers, 19 of which randomly placed, one at the origin, and NN subtraction using 20 seismometers in a spiral:

A little surprising to me is that the NN subtraction performance is not substantially better using a spiral configuration of seismometers. The subtraction results show less variation, but this could simply be because the random configuration is changing between simulation runs. So the result is that we don't need to worry much about array configuration. At least when all waves have the same frequency. We need to look at this again when we start injecting wavelets with more complicated spectra. Then it is more challenging to ensure that we obtain information at all wavelengths. The next question is how much NN subtracion depends on the number of seismometers.

We are trying to chopping the signal today.

   The low noise amplifier can be used as bandpass filters for 10-100 HZ.

   We are trying to figure out the signal squaring. The mixers in the lab only work for high frequency (> 500 KHZ).

   Frank recommends us to use AD734 4-Quadrant multiplier. 

   We checked the electronics lab in Downs and 40 m and couldn't find it. We plan to order some AD734.

 I ordered 5 of AD734 and thinking about how to make a circuit for squaring the signal.

    The "chopping" signal readout technique requires that we square the signals.  Basically we need to (as rana suggested):

(1) square the signal from PD, (after 10-100Hz bandpass) to convert it to power, and band pass it again.

(2) square the driving signal (might be varied from 0.1- 1Hz.) This is illustrated in the diagram as doubling the frequency ("2 x freq" box.)  The driving signal for PZT is offset. So the signal is  V drive = A + B xsin (2pi fdrive t) with A > B. This ensures that the voltage on one end of the PZT is always higher than another end. We might need to high pass this signal first, to get a signal with only 2 fdrive frequency after we square it.      

(3) multiply signal from (1) and (2) to demodulate the signal.

Basically, 3 multipliers are needed.

The first one is for (1), so the input frequency is ~ 10 -100Hz, and the output is 20-200 Hz.

The second multiplier is for (2), the signal is ~ 0.1 - 1 Hz, but this one might have large DC term after we square it.

The third one is for (3), this one has to multiply 2 low f signals together which is quite similar to (2), so the design can be the same.

I'll consult Frank and/or Koji again before finalize the multiplier circuit.

In the mean time, we might try this mixer to multiply the signal. I'll order one.

koji, mingyuan, tara: We designed the circuit for multiplying/ squaring signals with AD734. 

    The details for each signal are discussed here. 

    The "general multiplying circuit" box in the diagram shows how each AD734 will be powered/ fed input signal.

     For the signal from the PD, we need to bandpass(10-100Hz) it first. We plan to use a SR560. To split the signal to x and y input, we will use a T connector. Then square the signal and band pass it again at 0.1 - 100Hz bandwidth.

     For the signal from the function generator which drives the PZT. We will high pass it, by either SR560 or a high pass circuit. We might need a buffer here if the output impedance of the function generator is not high. Split the signal with a T again, and square it.

    After both signals are squared, we multiply them together. Send one to X1 input, another signal goes to Y1 input. Then we FFT the output signal from W.

 I tested the mixer, the demodulated signal from input at 10 - 100 Hz might be too small and too distorted to get reliable data.

  As we want to square/demodulate  signal in 10 - 100 Hz BW. a low frequency mixer might be a good tool. I asked Alastair to buy this mixer for me, and it arrived today.

  The lowest acceptable frequency in the design is 500 Hz, but I don't know how well it works at 10 - 100 Hz so I tested it.

   ==Setup and result==

   I used  SR785 to generate sine wave, then split it with a T and connected the output to LO and RF of the mixer. 

   I tested that the mixer works fine at the designed frequency.  The plot below shows the result from  1kHz signal input.

Next, I changed the frequency to 10 Hz, 50Hz, and 100Hz.

  The demodulated signal is then observed in frequency domain (left column of the plot) and in time domain ( right column of the plot)

I think the peaks at driving frequencies (10Hz, 50Hz, 100Hz and their harmonics) appear because of the offset of the sine input signal.

The results for low frequency  seem to be too distorted. We will test the AD734 chips tomorrow. I got the package this afternoon.

 We tested AD734 on the diagnostic bread board, the result is good.

     We want to square/multiply signals between 10 to 100 Hz, so we use AD734 chip to do the work. The circuit is connected as described here

We try to square the signal. the test signals are sine waves at 10 Hz, 50Hz. The output are nice sine waves, but the gain is high (72dB). The chip rails as the input exceeds 0.5 Vpkpk. We will have to check the signal from the PD in the setup to see if it is higher than 0.5 Vpkpk or not. If so we can change the gain of the chip. Otherwise we can go ahead and use it.

The spectrum of the output, for 10Hz input, there's a peak at 20Hz output. For 50Hz input, there's a peak at 100Hz. The response is flat between this bandwidth.

We brought the setup back. The interferometer is working and more stable. We will try extracting signal next.

    From this entry, we noticed the 180 degree phase shift in the signal when one arm was driven. The signal from PD followed the driving signal before drifted up and phase shifted by 180 with respect to the driving signal. We believed that this was the effect from the drift of the arm length. Suppose that we operate the IFO at the fringe's maximum slope. The drift in arm length will move the operating point on the fringe, and we might end up on the other side of the fringe which will show up in the 180 degree phase shift of the signal.

    The mirror was pushed by a piece of soft rubber which was glued to a pzt. Another end of the pzt was glued to a piece of plastic. This plastic piece was clamped on a translational stage. We thought that the soft rubber, the plastic and the translational stage caused the drift of the arm length.

So we tried to improved this by

• replacing the rubber and plastic with two pieces of magnets. One was glued on the back of the mirror, another one was glued to the pzt. This did not work, the combination of the force, and the shim stiffness, had to be matched so the mirror position can be adjusted without letting the magnets touch each other. So we tried
• replacing the rubber and plastic with stainless steel nuts, one nut is for clamping, another one is for pushing the mirror.
• 

we haven't got rid of the stage because we still need it for position adjustment purpose. We will use dc voltage offset on pzt to adjust the position later once we can add dc signal to the driving voltage.  Currently, we use a single function generator to drive both pzt simultaneously.

     With new pushing scheme, the drift becomes much less than before. The signal is in phase for more than a minute or two which should be enough for chopping technique later. The picture below shows the signal from driving voltage @ 2Hz(blue), and readout from PD at maximum slope (yellow).

     Once we made sure that the signal was quite stable, (that is, the operating point stays at the maximum slope most of the time), we measured the background noise. This is a readout from PD and maximum slope on the fringe without driving voltage applied on the pzt. Then we measured the signal when one arm was driven at 2 Hz. Finally, we drove two arms at 2Hz and adjust the voltage on the pzt so that the 2Hz common mode cancelled out.

 The plot shows the noise of the setup: 1) the background 2) when one arm was driven at 2 Hz. 3) Both arms are driven, with common mode at 2Hz minimized.

 We will try squaring the signal next. The read out from PD is ~ 200 mV. This value will determine if we need a divider for the signal or not.

We used a big box to cover the optical loop. The interferometer is more stable now.

We build other two AD734 chip circuits for signal square and multiplier.

We already tested that we could square the driving signal and PD signal.

The square of the PD signal has a big offset from the AD 734 circuit. We need figure out how to take the offset out.

We tried read out the signal from chopping technique.We could not see anything yet.

The signal when both IFO arms were driven were similar to the signal when there was no driving.

   After we made the necessary electronics for chopping technique, we tested if we could see the signal or not.

    ==setup==

    We used a 4 mW HeNe laser as a source with a simple Michelson interferometer setup. We tried to operate at the maximum slope of the fringe. Each mirror was attached to a metal shim which could be pushed by a PZT behind it, see here . We drove the mirror with the same distance so that the common mode was canceled and only incoherent noise from crackle in each blade could be detected.

    The diagram omits the IFO part and the blades. The output beam from the IFO was incident on the PD. We operated at the maximum slope of the fringe. The driving voltage Vdrive was send to the PZTs pushing blades (with mirrors attached on them) at the end of both arms.

    The 1/2 and 1/10 dividers are used to reduce the signal down below 0.5 V. This number comes from the square testing. When the input signal to be squared is larger than 0.6, the output starts to rail. So we use 0.5V to be the upper limit for now.

   ==result==

       The PSD of the signal output when two arms are driven are similar to the background signal (arms are not driven). It might be that the gain setting are not optimized, the setup is too noisy, or problems from offset from the AD734 chip. We will figure that out next. We will also make a sturdy box for multiplying chip. Currently we just use temporary test board to operate the chips for the read out.

 1) We removed the squaring circuit from the test board and built it on a board. The box for the circuit was prepared.

 2) We replaced the crappy beam splitter with a Thorlabs 20mm cube 400-700 nm beamsplitter. The beam power is evenly divided and has no multiple reflections. We measured the noise psd at the AS port.

      1) The circuit for squaring, multiplying signals was temporarily built on a plug-and-play test board which was neither sturdy nor compact. So We used a breadboard available in the EE lab to build the circuit.

The cartoon schematic is shown below.

      A) The signal from PD at AS port is band passed before squared (not shown here), then band passed again before.

      B) The driving voltage for PZT will be high pass to get rid of DC component (not shown here), then divided. We want a divider here because we might need to drive the pzts with higher voltage. The second divider might be unnecessary, but we have it just in case.

     C) Then we multiply  A and B and get the signal out for FFT.

     Currently, the chips have offset added to the output, ~ from -1 to -2 V. We tried adding the offset in Z2 let as suggested in the datasheet, but it killed the signal ??!!!. So we are planning to high pass signals that we care only their AC parts. Currently, we are not sure if we care about DC part of the V drive or not. We have to think about it.

     2) The beam splitter used in the original setup is not really for a beam splitter for Michelson IFO. It is not 50/50, and there are multiple reflections from the surfaces.

Thus, we ordered a cube beam splitter suitable the job and replaced it. It is mounted on a beam splitter mounted directly mounted on a 2" post, so we expect it to be more stable.

We measured the noise from AS port when the armed was not driven vs driven at 1 Hz. The result is shown below.

The calibration from V to differential arm length (Lx - Ly) is approximated from

dx ~ dV x  lambda/ 4 / (Vmax - Vmin)

At the maximum slope of the fringe, as we tap the table, the voltage will fluctuate between Vmax (from constructive interference)and Vmin (destructive interference.)  On the fringe, the differential arm length between maximum to minimum V output is lambda/4 (so the accumulated distance from round trip is lambda/2, a condition for changing from maximum Vout to minimum Vout). We can approximate the slope to be (Vmax - Vmin)/ (lambda/4).

Vmax - Vmin ~ 500 mV, lambda = 660 nm. so

dx = dV x 3x 10^ -7

The result is 5 - 6 orders of magnitude above the shot noise level (~ 1e-17 m/rtHz for this setup.) Noise characterization will be considered next, but from

a quick test of tapping, seismic is the dominating source.

[Giordon, Zach]

NB: The vacuum system was back at 10^-7 Torr when we started work on Friday. 

We began using the lock in to monitor demodulated time series while driving modes and then allowing them to ring down. We used two different driving methods:

1. Narrowband white noise. Using either the Agilent analyzer or FG as the broadband source, and then filtering to a 1-2kHz bandwidth with an SR650 filter.
2. Swept sine using the Agilent FG. Set the center frequency to the mode frequency as seen on the spectrum analyzer, and then used a ~10-Hz span with a sweep time of ~0.1-1 second to drive the mode.

Both used a DC bias of 3 kV. For the readout, we set the internal SR830 lock in reference signal to ~10-100 Hz offset from the measured mode peak, and the time constant such that there was a clear signal at these beat frequencies without excess high-frequency fuzz. The lock in input was of course the differential PD signal, with the difference taken by an SR560. It was basically like an SRS/Agilent product demo down there.

The noise method was obviously just not strong enough somehow. Looking at the AC drive signal, we increased the power until there was a rough peak-to-peak level of ~2 kV, above which I did not dare go. Looking at the readout, there was basically no difference at all with the noise drive on or off.

The coherent drive seemed to do something, but as expected it both excited and damped the mode as an unpredictable function of the mode drift and the sine sweep. That said, when it excited the mode, there was a notable increase in the signal. When the drive was turned off while the output was at relatively high amplitude, the signal immediately went down to its unexcited level. The timescale was well under a second.

There are at least two possibilities:

1. The excited signal seen while using the sinusoidal drive is actually a spurious EM coupling. This seems unlikely because of the low-frequency amplitude fluctuations seen while driving, which are suggestive of alternating excitation and damping.
2. The mode Q is just that low. The ~2 Hz linewidth (Q ~ 10⁴) I measured using the spectrum analyzer would correspond to decay times under 1 second. In all fairness, we don't know what it should be for this random sample a priori anyway.

It seems odd that we have a hard time ringing the mode up substantially, since the DC bias visibly excites the pendulum mode upon turn-on (so the electrostatic force is high enough to do something). It could be that the vibrational modes' admittance is just much lower than that of the pendulum, or that there is a great deal of cancellation from the particular symmetry. Another thing is that the sample might be getting subtly tapped by the ESD assembly every once in a while, and this destroys the coherence of the measurement.

We might consider slightly adjusting the arrangement (e.g. by translating it) to see if there is any improvement.

I got a chance to talk with Norna about the strain range we are in for our maraging steel blades:

We typically load the blades to a stress level of 800 MPa to 1000MPa. The upper value there is approx. 55% of the yield stress which is ~1.9 GPa. The Young’s modulus E = 186 GPa.

—> strain rate = d/dt [ (F sin(wt) + F0) / E ] = (wF/E) cos(wt) —> maximum rate = wF/E = (2 pi 0.125 Hz) (800 counts ~ 1um deflection) / 186 GPa

400 MPa ~ 1mm deflection

4 kPa ~ 1um deflection

—> max rate = 1.689e-08 /second

Also, the triangular shape gives equal stress along the length of the blade when loaded.

In addition, as [1996 Dahmen] "Hysteresis, avalanches, and disorder-induced critical scaling: A renormalization-group approach" shows a relationship between hysteresis and crackle, on page 14 878 they show how a model scaling hysteresis loop area with r, where r is associated with the average avalanche size: A_sing ~ r^(2-alpha) (MFT :alpha =0), I am thinking maybe materials with larger hysteresis loop area would generate crackling noise more “easily”. If this is the case, a good candidate would be the AISI 1085, which is also called music wire; it’s a high carbon steel. 

Reference: [1998 Beccaria] "The creep problem in the VIRGO suspensions: a possible solution using maraging steel"

A folder on the 40m svn server has been created to store crackle related files

The location is: /trunk/crackle

This currently includes the latest poster, a MATLAB subdirectory which holds all of the code from my last elog, a Lit subdirectory with a couple of papers in it, and an ExpChar directory which houses the bulk of the measurements I've made regarding the experiment.

Specifically listing what is in each sub-dir

/trunk/crackle/ExpChar/:

• Blades: transfer functions and ringdowns of the blades via shadow sensor
• Laser RIN: measurements of RIN in different configurations (fiber, no fiber, dark noise, etc.)
• Loop: Measurements of the Loop TF
• Mich: spectra of the Michelson error signal
• Servo: LISO files of the servo circuit, and transfer functions (liso and measured)
• ShadowSensor: Circuit of the shadow sensors and TF of the damping
• circuit.graffle: OmniGraffle drawing of the crackle NIM box circuitry

I have once again borrowed the floor-standing HiCube turbopump to use on the gyro chamber.

Since no one is actively using the CQ chamber right now, I will probably just keep it in the ATF for the moment. The bell jar seems to hold ~uTorr (I meant mTorr -- ZK) vacuum after a day of being sealed, so it shouldn't be a problem to leave it as is.

On the analog part, we modified the old circuit such that we can use it directly in the digital control.
The schematics of the original circuit goes as follows:

Basically, we only need the LED and coil drive part. We removed the resistors R17, R30, R31, R41 such that the
coil drive part is now disconnected from the derivative and integral control part [we will replace it by digital control].
We also replace the potential meter with fixed resistors. To make it more clear, let us look at only one signal path
of the coil drive [the first one]. The modifications are shown by the following two figures [left one is the original coil
drive and the right one is the modified one].
 

>>>>>>>>>>>

Now we have four analog channels for both LED and coils. By combining them with the DAC and ADC in 
the two National Instruments cards, we can move to phase of implementing the digital control.

On the digital part, we find an example on the Internet about using Labview for proportional-derivative-integral (PID)
control. The virtual instrument file is attached: pid_control_labview_example.vi

In this example, they used a simulated plant to demonstrate how to realize PID control in Labview, which is very useful to us.
The modifications that we need to make are the following:
1. Replace the simulated plant with our DAC and ADC vi module (Koji showed us yesterday)
2. Extend it to four input and four output channels (from SISO to MIMO);
3. Include the control matrix.

In the next few days, we will try the PID control first with a single input and single output to make ourselves get familiar with the
Labview interface and also demonstrate the control principle. We later can proceed to multiple channels.
 

Jan and I wound up cutting the connector from the photodiode DC power supply as opposed to ordering a cable from Thorlabs and connecting it to the electrical feed-throughs at the polycarbonate window (orange cable). And connecting the wires to the BNC box below:

We aligned the interferometer and connected the photodiode to the oscilloscope, after which Jan was able to produce the following image:

We attached the two shadow sensors, producing the following image. All that remains is aligning the shadow sensors.

Otherwise, I was also able to realign the laser and the optical fiber.

Update:

1. Got the rack settled and tidied up the lab a little bit.
2. Worked on the chassis power regulator for binary input and output.
3. Sketched a timeline for the project before the Christmas.

Near-term timeline:

During the discussion with Koji this afternoon, he suggested to sketch a timeline for coordinating the progress of different modules in order to have a constant flow of progress. Below is a schematic flow chart for the final setup:

Here I first divide the parts in the flow chart into five modules and show who are responsible for each of them:

I then estimate approximately the time needed for completing individual module before Christmas:

This cannot be precise, as there are various uncertainties, but it does give us some feelings about what need to be done in the near term, and how we should coordinate the progress of different modules.

Today we tried to use the national instrument card together with Labview to acquire data from the OSEM for measuring cross coupling.
The levitated plate is supported by four springs around the equilibrium point [below is a view from the side]:



One issue we encountered was that the flag can easily touch the edges of the OSEMs due to imperfection
in the design and the slight horizontal drift of the equilibrium point.

To fix this, we find a possible solution: we can use 1''inch diameter Polycarbonate Round Tube to replace the case of OSEM.

 For this purpose, I have ordered the following components today:
1. 1''inch diameter Polycarbonate Round Tube with length of 1 feet. Mcmaster Carr Part number: 8585K14
2.  IR Emitters with wavelength 890nm and also 935nm. Newark Part number: 08F2922 and 08F2957
3.  the corresponding IR photodiodes. Newark Part number: 91F1840 and 32C9152

Concerning the circuit board for LED and coil drive, I found that we might not need to order extra new board from PCBexpress.
We have an extra one left from the previous ordering. I will look into it more carefully to see whether we can use it or not.

Concerning the feedback control part, Yi has worked out the control matrix by assuming certain values for the sensing matrix elements [we are trying
to measure them]. This afternoon, Koji gave us very precious suggestions and materials for implementing the feedback control in Labview.

Yesterday, we test the ADC and DAC for the maglev. Specifically, we measured the time delay in the
digital path by using the method suggested by Koji. This is useful for us to model the stability of the system.
Basically, we connect the analog input (AI) to a function generator, and make a direct connection in the
Labview from the ADC and DAC.We then compare the time delay between the signals from the function
generator and from the analog output  (AO). The setup is shown by the figure below [we have rescaled
the two outputs for a clear display]:

Labview virtual-instrument file for the ADC and ADC:

We used function generator to produce square waves at different frequencies: 50 Hz, 100 Hz, and 200 Hz
to avoid the possibility of delaying by an integral number of periods if we only use one frequency. These three
frequencies measurement all give the same measurement result---the delay is around 2.8 ms.

We are now trying to put together all four AI and AO channels. We need to make corresponding change to the
SISO PID controller to MIMO. We are still working on it. A flash show of the unfinished PID controller.vi
is shown by the figure below [We are making very simple modifications to the  example given in
http://techteach.no/labview/lv85/pid_control/index.htm]:

An update on the current status of maglev:

(1) We installed the springs in our setup. Right now, the levitated plate is held stably by those springs.
     We are going to measure the cross coupling, once the Labview is working. In addition, we need to have
     four current buffers to drive the coils.

(2) The second BNC terminal block for the National Instrument card has arrived. Now, we have enough
     input channels (36 in total) and output channels (4 in total) to implement the feedback control. I am current
     learning labview and Jan Harms is helping me use it to take data.

(3) The new magnets have arrived, and we will make similar measurements of the strength distribution, as
      we did earlier. We try to find better matched magnets, possibly down to maybe 1%.

(4)  We measured the current force on the levitated magnets. The measurement setup is the same as
      the magnetic force measurement, except for that we now connect the coil to a DC power supplier to
      deliver a constant current flow to the coil. This measurement allows us to determine the DC biased current
      that needs to counteract the imbalance in the DC magnetic force. The data is under analyzing, and the result
      will be posted soon.

Finally a little bit of progress with the leak in the crackle chamber.

1) Replacing all copper rings we found that one copper ring had two intersecting cuts that may well have contributed to our leak.

2) The lid had an older L-shaped viton gasket to seal the vacuum. It is replaced with a new one, but since many things have changed it is not perfectly clear how much scratches in the old viton ring contributed to the leak. It seems that there was some contribution, but it was probably minor.

3) Finally, the most significant change came from replacing the polyarbonate window and viton ring with a metal flange and copper ring. The main problem with the polycarbonate window was that it took scratches quickly so that the viton ring could not seal properly anymore. Also at the chamber side of the flange, the viton ring had no proper groove and became a bit twisted under pressure.

The result is that the pressure got down from a previous 450mTorr record to now 56mTorr and still decreasing (very slowly though). If pressure does not decrease substantially over night, then we still have a bad leak, but at least we can already hope for a positive effect on our sensitivity.

I bet that the backing pump that you have for the turbo will bottom out at ~10-20 mTorr. The ideal gas law does not allow it to go to zero.

RTFM. When are you supposed to spin up the turbo?

[Alastair, Giordon, Zach]

Alastair showed us how to safely remove the lid of the bell jar with the hoist. We inspected the inside to see how we might set up our first measurement. His fiber setup is still in there, and we decided that as a first go, we will just weld the first sample we received from Gregg to the end of the fiber (after we've shortened it appropriately). Part of the reasoning is that this (3" x 0.25") sample looks as though it's already been welded before:

The setup inside the vacuum system will be:

• Sample suspension
  • Intermediate fused silica mass suspended via fiber from the steel plates toward the top of the chamber
  • Our disc suspended from the intermediate mass via fiber
• ESD
  • There is a network of teflon bars on threaded rods, which was designed to be adjustable
  • We will drill one of these rods so that I can screw the ESD onto it and adjust it to the right height/horizontal position
• HV connection
  • There is already a proper HV feedthrough from the HV amplifier, and we have in-vac HV wire on spools
  • I will talk to Margot or someone to find out the best way to solder this wire to the (+) end of the ESD
  • Using another piece of wire, we will connect the (-) end of the ESD to the vacuum (earth) via a bolt or something.

This is very nice because we will have a minimal amount of equipment inside the vacuum chamber. We may decide that we want to build the fancy cat's cradle (nodal) support, or something else, but this is the fastest way to start measuring something.

It should also be mentioned that this single-point welded support has been a big problem for interferometric setups because of wild torsional motion. We are hoping that this will be a higher-order effect with our transmissive setup. We'll see.

Here is the hour of truth (I think). I ran simulations of wavelets. These are not anymore characterized by a specific frequency, but by a corner frequency. The spectra of these wavelets almost look like a pendulum transfer function, where the resonance frequency now has the meaning of a corner frequency. The width of the peak at the corner frequency depends on the width of the wavelets. These wavelets propagate (without dispersion) from somewhere at some time into and out of the grid. There are always 12 wavelets at four different corner frequencies (same as for the other waves in my previous posts). The NN now has the following time series:

You can see that from time to time a stronger wavelet would pass by and lead to a pulse like excitation of the NN. Now, the first news is that the achieved subtraction factor drops significant compared to the stationary cases (plane waves and spherical waves):

And the 4*pi, 10 seismometer spiral dropped below an average factor of 0.88. But I promised to introduce an absolute figure to quantify subtraction performance. What I am now doing is to subtract the filtered array NN estimation from the real NN and take its standard deviation. The standard deviation of the residual NN should not be larger than the standard deviation of the other noise that is part of the TM displacement. In addition to NN, I add a 1e-16 stddev noise to the TM motion. Here is the absolute filter performance:

As you can see, subtraction still works sufficiently well! I am now pretty much puzzled since I did not expect this at all. Ok, subtraction factors decreased a lot, but they are still good enough. REMINDER: I am using a SINGLE-TAP (multi input channel) Wiener filter to do the subtraction. It is amazing. Ideas to make the problem even more complex and to challenge the filter even more are welcome.

Whitened specgram please to help see small differences. Also your noisy / quiet spectra look like they are contaminated at the low frequencies by the leakage from low frequencies due to the FFT window being too short. Perhaps retry with a 8 second FFT window to see if the effect is different.

I tried highpass the mich signal with corner frequency at ~ 15 Hz and then take the spectrogram:

When whitened, the spectrogram looks pretty stationary. When comparing the noisy and quiet spectra I used 8 second FFT window, we don't see any significant difference:

The 15 ~ 55 Hz non-stationary noise seen in Elog 1745 is indeed a leakage from low frequencies due to FFT window being too short, as Rana suggested. 

At about 11:25 LT I checked that no measurement was running, disabled the autolocker, broke the lock, restarted the x1kr2 model to add the epics channel X1:KR2-BBOARD_LOCK_AUTOLOCKER_REQUEST that will be used in the next version of the autolocker script.

At 11:30 LT the autolocker was up and running again, the IFO is locked again.