Aligned CR14 chamber back to 3" disks.
I measured the properties of the beam on the QPD. The total power is 31 uW. The beam shape is not gaussian, since we are seeing the interference of the reflection from the two surfaces:
The X and Y diameters are 1400 and 1300 microns, so I take the average of the two as an estimate of the beam size: 1300 +- 100 um. I also estimated the lever arm length to be 1.03 +- 0.02 m.
This allows me to esitmate the response of the normalized QPD signal to a tilt of the disk surface:
Plugging in the numbers gives a gain of (1900 +- 300) /rad for the normalized signals. I implemented those numbers in the filter banks: now X_NORM and Y_NORM have units of radians, and measure the disk surface angular motion. I also calibrated the SUM channel in microwatts, using the nominal responsivity of 0.45 A/W and the transimpedance of 200k (gain 11.1 uW/V)
Here's teh calibrated spectrum: note that the background noise is much larger than the real one because of the signal jumps.
Turbo pump off and spinning down at 9:37am LT. Pumo completely stopped at 11:15am LT
Openend the chamber and removed the sample at ~11:20am LT
I modified four more QPD boards to implement the new whitening filters detailed in elog 207.
Ordered the clean room (hardware+hepa filters) and vacuum gauges
This morning we installed the clean room curtains and washed them. It turns out that the air filters are supposed to be powered at 277V (?) instead of 115V. So right now the flux is quite low. We are looking into the problem: either replace them with 115V modules or install a small transformer.
We also installed the vacuum chamber on the table and connected all the pumps and gauges. There are no leaks and we could pump down easily the empty chamber. We left for lunch when the pressure was at a few 1e-6 Tor and still going down.
The clean room frame is built and secured to the floor and wall. Panels are being installed on the ceiling and back. Also, the optical table has been leveled.
Ceiling, back and side panels are installed. The air filters have been cabled and connected to the power supply.
Mon May 13 18:37:37 2019
Entered CRIME lab to borrow 4x hair nets and face masks. Can you please advise on what I should order for clean lab equipment? There are more options on techmart than I anticipated. We're in the process of increasing the cleanliness of the SiQ experiment.
There is a buy list of approved clean room supplies posted here https://dcc.ligo.org/LIGO-E1300399. This list is used by designated people to keep clean rooms supplies stock at each site including LIGO labs in Downs, 40m and the CRIME lab. Not sure what lab you are working in and what regulations you have there. Typically we study the list of the approved supplies, figure out what budget can be used for supplies for a particular experiment. Depending on what your project is, you may be able to just take what you need from the existing LIGO stock (I believe there is one for Downs and one for Bridge and 40m) or work with Liz, Bob or Chub on ordering it for your via approved channels.
Some progress on the cleam room: bar fixed to the wall, some more structure built, filters in place. We had to (literally) work around a corner of the low ceiling that we haven't noticed before. More contruction will follow tomorrow. We also had to order some additional parts (more extrusions, brackets, screws, etc...)
The plot below compares the measured Q values of S1600433 right after annealing, and after GariLynn cleaned it. Q values are in general improved a bit after cleaning.
Samples #433 (annealed and cleaned) and #438 (as received from Mark Optics) are now with GariLynn for deep cleaning.
Sample #438 was broken during annealing.
The plot below compares a sample from the first batch and two samples from the second batch. All samples are as received from Mark Optics, no annealing or any other treatment.
Both samples in the second batch show consistently and significantly lower Q values.
GariLynn and I inspected the two samples under the microscope. Surprisingly, the edges and the flats look much better than the samples from the first batch. See elog 148 for an image of a sample from the first batch
Link to IMG_3158.JPG
Link to IMG_3157.JPG
Link to IMG_3156.JPG
Link to IMG_3155.JPG
The PCBs for the QPD circuit and ADC interface are here and look ok. All electronics components are also here (except for the ADC connector which should be ordered separately from Mouser, after confirming that the ADC we're going to use have the same cable as the one we use in the Crackling Noise experiment). The QPD will be shipped on 06/17.
In the last days Jamie installed the patched kernel to run the real time system (RTS) on the new CyMAC. Today (with Jamie's remote advices) I managed to get a IOP (input output process) model compiled and running. There is still no timing input (to be fixed at the beginning of the netx week, I'm presently missing a connector which is on order).
Running the x1iop model gives:
x1iopepics: no process found
Number of ADC cards on bus = 1
Number of DAC16 cards on bus = 1
Number of DAC18 cards on bus = 0
Specified filename iocX1.log does not exist.
x1iopepics X1 IOC Server started
controls@cymac3:/opt/rtcds/tst/x1/scripts$ awg_server Version $Id: awg_server.c 2917 2012-05-22 22:33:39Z alexander.ivanov@LIGO.ORG $
channel_client Version $Id: gdschannel.c 4170 2016-04-05 21:24:46Z jonathan.hanks@LIGO.ORG $
testpoint_server Version $Id: testpoint_server.c 3303 2013-03-05 23:33:45Z alexander.ivanov@LIGO.ORG $
/opt/rtcds/tst/x1/target/gds/bin/awgtpman -s x1iop -4 -l /opt/rtcds/tst/x1/target/gds/awgtpman_logs/x1iop.log started on host cymac3 hostid ffffffffd783d97b
awgtpman Version $Id: awgtpman.c 4170 2016-04-05 21:24:46Z jonathan.hanks@LIGO.ORG $
1) install MATLAB and add to startup.m the following line
2) copied from cymac2 the file /opt/rtcds/rtcds-user-env.sh and changed the content to match the right folders
3) installed readline library which was missing: apt-get install
sudo apt-get install libreadline-dev
4) installed linux headers sudo apt-get install linux-headers-3.2.0-rts-amd64
5) created symbolic link to linux headers:
sudo ln -s /usr/src/linux-headers-3.2.0-rts-common /usr/src/linux
6) changed host name to cymac3
sudo vi hostname
7) created symbolic link to lspci
sudo ln -s /usr/bin/lspci /usr/sbin/lspci
8) created a symbolic link to awgtpman
ln -s /usr/bin/awgtpman /opt/rtcds/tst/x1/target/gds/bin/awgtpman
I got two new ADC and DAC boards from Rolf, with the correct PCIe interface. I installed them into the cymac and checked that the system could boot. The cymac is now sitting in the rack. As requested by Jamie I installed Debian 8.5
I copy here parts of an email from Jamie with instructions on how to run the RTS on the cymac:
I had to change the site/ifo to be "caltech/x3" to avoid EPICS
collisions with other cymacs. You'll need to update your models to
reflect this change (change names, site/ifo parameters in the model,
etc.). So for instance, change x1iop -> x3iop, and update the params
daqd is installed and running, although I haven't really stress tested
it yet (no models/channels/etc). It's running under systemd, which is
the Debian service management system. You can control daqd with the
# systemctl restart daqd
# systemctl stop daqd
# systemctl start daqd
# systemctl status daqd
# journalctl --unit=daqd
controls@cymac3:~ 0$ systemctl status daqd
● daqd.service - Advanced LIGO RTS daqd service
Loaded: loaded (/etc/systemd/system/daqd.service; enabled)
Active: active (running) since Sat 2016-07-09 11:22:38 PDT; 23h ago
Main PID: 20827 (daqd-standiop)
└─20827 /usr/bin/daqd-standiop -c /etc/advligorts/daqdrc
controls@cymac3:~ 0$ sudo journalctl --unit=daqd
Remember to restart daqd ("systemctl restart daqd") after you add/change
I've gotten all the RCG components working, but not without some small
kinks. The RCG currently expects a specific EPICS install, different
than the system install we're using right now. I've hacked a way to
make this work seemlessly for model builds, but it requires sourcing the
following file before *starting* models:
So in the mean time, you can start/stop models via the following:
$ . /opt/rtapps/epics/etc/epics-user-env.sh
We played around a bit with the cymac, in efforts to make things better.
As I see it, the main problems that persist are:
MO03 - edge polished:
Turbo off, QPD centered, before excitation (60 seconds)
PDT: 2016-08-23 08:42:54.514987
PDT UTC: 2016-08-23 15:42:54.514987
UTC GPS: 1156002191.514987
Excitation (white uniform noise, amplitude 5 V)
PDT: 2016-08-23 08:45:01.007626 PDT
UTC: 2016-08-23 15:45:01.007626 UTC
Clean data for ring-down
PDT: 2016-08-23 08:45:46.448949 PDT
UTC: 2016-08-23 15:45:46.448949 UTC
Restarted roughing pump, QPD got misaligned
PDT: 2016-08-23 10:00:29.259345 PDT
UTC: 2016-08-23 17:00:29.259345 UTC
Band-limited noise, +-10Hz around eahc nominal frequency, amplitude scaled based on the inverse of the peak height obtained with white noise. See attached code and plot
from numpy import *
from noise import *
x = loadtxt('/home/controls/Measurements/2016_08_23/mo_02_laserpolished_frequencies.txt')
freqs = x[:,0]
ampl = x[:,1]
bw = 10
bands = map(lambda x: [x - bw, x + bw], freqs)
a = 1 / (ampl/max(ampl))
a[a>50] = 50.
x = multi_band_noise(bands, a, 10, fs=65536)
x = x / 30
Ring down after:
PDT: 2016-08-23 11:07:02.661145 PDT
UTC: 2016-08-23 18:07:02.661145 UTC
I ran a set of COMSOL simulations to determine the dependency of the frequency of each eigenmode on the disk thickness and diameter, within the tolerances. I chose wide ranges: diameter 75.0 +- 0.1 mm and thickness 1.0 +- 0.1 mm, much more than the expected tolerances. It turns out that the frequencies depends almost exactly linearly on both variables: mostly on the thickness and negligibly on the diameter. The following plots shows: the mode shape and frequency (left), the frequency dependency on the two variables (center), the residual of a linear fit and the functional form of the fit (right).
I'm including only the modes that will be measurable by our system (no motion in the center, frequency below 32kHz. Since the disks in my simulation is completely round, I'm showing only one mode for each doublet.
For the optical levers we are going to use the same QPD that are used in aLIGO optical levers (see T1600085 and D1100290): Hamamatsu S5981
Based on the aLIGO design, I put together a design fof the QPD boards, see the first attached PDF file. Some comments:
The stable power supply will be provided by an additional board, which will also interface 8 QPD boards to the ADC connector, see the second attached PDF. The ADC signal grounds can be connected directly to the power supply ground, left floating, or connected with a RC filter, depending on what we find to be the best solution.
Total cost estimated for PCB manufacturing and components, including the QPDs is less than 3k$, for a total of 12 boards (we need 8, plus some spares)
Koji, Rich, and I recently came up with a new QPD design which is better for general lab use than the aLIGO ones (which have a high-noise preamp copied from iLIGO).
This page has the mechanical drawing only, but perhaps Rich can tell us if he's ready to make the first version for you or not. I think you can get by with the old design, but this new one should be lower noise for low light levels.
After a very useful discussion with Rich this morning, I think the circuit based on aLIGO optical levers design should be good for our applications.
It uses a LT1125 as input stage, which has
We expect to send about 5 mW into the disk, getting back a 4% reflection, which would correspond to 200 uW on the QPD. Let's say we lose half of this power in reflections through viewports and such, so we have a total of 100 uW on the QPD, or 25 uW on each quadrant. From the QPD datasheet the repsonse is about 0.4 A/W, so we have a photocurrent of 10 uA. The corresponding shot noise limit is about 1.8e-12 A/rHz.
Using a transimpendace of 200k, the noise at the output of the transimpendance is
So in the worst case the current noise will be about half of the shot noise. This seems good enough.
Disk excited with white uniform noise, amplitude 5 V, for some tens of seconds.
Excitation off at
PDT: 2016-09-13 10:32:23.887615 PDT
UTC: 2016-09-13 17:32:23.887615 UTC
I assembled the disk suspension sytem and installed into the chamber. Although I don't have the magnets and coils, I installed the movable retaining disk, and used it to center the disk.
I first aligned the input laser using the reflection off the black glass, which turns out to be quite bright and very well visible. Tomorrow I'm going to measure how much power we have in the black glass.
The reflection from the disk is slighlty separated from the reflection from the black glass, so I can block it using an iris.
At 6:50pm I closed the chamber and started the roughing pump. At 7:05pm pressure was below 1 Tor so I started the turbo pump. When leaving pressure is about 1.6e-5 Tor.
It's possible to build an analytical model of the resonant frequencies of a simple thin disk. For example, see J. Sound and Vibrations 188, 685 (1995), section 2
The solutions are given in term of Bessel functions:
where J is a Bessel function of the first kind, and I a modified Bessel function of the first kind, a is the disk radius.
The coefficient Cmn and the eigenvalue can be found as solution of the following two equations
Then the eigenfrequencies are given by
where rho is the material density, h the disk thickness and D the flexural rigidity
where E is the material Young's modulus and nu the Poisson's ratio.
From all those results we can conclude that the frequency scaling with respect to disk radius and thickness are very simple:
Also, the frequencies scales as sqrt(E/rho)
The dependency on the Poisson's ratio is more complex since nu is involved in the eigenfrequency equation shown above.
Unfortunately the thin disk model does not exactly match the COMSOL results: deviations of few tens of Hz are present, probaly due to the thin disk approximation. The COMSOL model is more accurate to match the experimental frequencies.
However, I checked that the eigenfrequencies predicted by COMSOL also scales as predicted with thickness and radius.
Using the measurements on the six samplex we got from Mark Optics, after annealing, I was able to tune the COMSOL model to fit all measured frequencies within 6 Hz. I chose to change the disk thickness (since diameter and Young'r modulus are degenerate) and the Poisson's ratio.
Here is an example of the difference between the measured and modeled frequencies:
The table below summarizes the best fit for each of the disks
Since the material is the same, I would expect the Poisson's ratio to be constant. So for future modeling I'm using the average of the values above: 0.166
A preliminary design of the ESD board is available on the DCC: D1600214
We first measured the distance of the ESD from the disk in the test chamber (CR0). We had to remove the retaining ring to have reliable measurements
So initially the distance between disk and ESD is 1.22 mm
We re-aligned the optical setup to a horizontal reference, and moved down the ESD as much as we could. It's not completely clear if the ESD is touching the disk. We'll see after pump down. The new distance from the top of the ESD to the mounting plate is about 11.80 mm, so we should have moved the ESD 0.5mm closer to the disk.
Pump down started at ~1:30pm
The plots below compare the SNR and peak amplitude of all excited modes, in the new and old configuration. The new confgiuration is worse than the old one. This is unexpected, since the distance between ESD and disk is smaller.
However, yesterday we found out that setting the ESD so close to the disk is very tricky, and we might have some touching.
Additionally, the measured Q values of all modes are signfiicantly lower (by factors of >3), so it seems there is some additional friction. The mode frequencies are still compatible with the expected values, so it's unlikely that the ESD is touching the disk. One possible explanation for the worse Q can be residual gas damping in the area between the ESD and the disk: basically the gas moelcules that are left in the enclosed region between disk and ESD can create a viscous damping, which gets larger when the distance gets smaller [PhysRevLett.103.140601, arxiv:0907.5375]. I'll try to do some computations later today.
I made a COMSOL model that can compute the distribution of elastic energy for each mode, dividing it into:
Then I used the measured Q values for the MO_101 disk and tried to see if I could reproduce it with the energy distribution. The first plot here shows that the loss angle of the disk (inverse of the Q) has a trend that is already quite well reproduced by the ratio on edge energy over total energy:
In particular the edge energy distribution is enough to explain the splitting of the modes in families. This fit is obtained assuming that the edge losses are uniform along the entire edge, and frequency independent. If we assume a "thickness" of the edge of the order of 1 micron, the loss angle is about 3.5e-3, which seems resonable to me since the edge is not polished.
Then I tried to improve the fit by adding also bulk, shear and surface losses. It turns out that shear is not very important, while bulk and surface are almost degenerate. The following plot shows a fit using only edge and surface losses:
The result is improved, expecially for the modes with lower loss angle. Again, assuming a surface thickness of 1 micron, the main surfaces have a loss angle of 1.3e-5, while the edge is 2.3e-3.
Including all possible losses gives a fit which is basically as good as the one above:
However, the parameters I got are a bit differentL: the surface losses are reduced to zero, while bulk dominates with a loss angle of 1.4e-4, and shear is not relevant.
In conclusion, I think the only clear message is that the Q of our disks are indeed limited by the edge. The remaining differences are difficult to ascribe to a paritcular source. Since th disks are thin, I tend to ascribe them to the surface, which would imply that we are far from being able to see the bulk/shear losses. If I use only edge and surface losses, I found as expected that the polished main surfaces have much lower loss angle by a factor 200 or so.
S1600439 has been measured as received (before annealing, elog 137) and after annealing (elog 144).
Q values are significantly increased for almost all modes, see the plot below for a comparison. Only modes with low Q are not improved.
The same set of samples described in the previous entry have been annealed at 500C for 9 hours. Then the loss angles have been measured again.
The plot below shows the measured loss angle for all modes and all samples. After annealing all loss angles are significantly decreased, and they also show an increasing trend with frequency. As before, the blue points are the measurement points (averages of 8 ring-downs each) and the error bars are computed from the statistical error of the measurments. The red line shows the average of the loss angles for frequencies below 15 kHz, weigthed with the data points uncertainties. The red shaded area shows the 95% confidence interval of the mean.
If we plot the frequency-averaged loss angle as a function of the serial number, we see that there isn't much of a spread in the values:
We can again plot the loss angle as a function of the process variables. There are three main parameters that are changed in the deposition: the assist beam voltage, the assist beam current and the content of oxygen in the assist beam. The plots below show the losses as a function of those parameters. The x axis changes in each of the four panels, and for each plot, the color code is linked to one of the process variables:
This time I can't see much of a trend anywhere in those plots.
Since the loss angles show a clear increasing trend with frequency, instead of computing the mean value, I fit each dataset with a linear dependency on the frequency. To improve the fit I restricted the computations only to frequencies below 12 kHz. The results are shown below
The following plot shows the fitted loss angle at 1 kHz, as a function of the serial number. There is more spread in the results than when using the simple average:
And again, the dependency of the loss angle at 1 kHz on the process parameters:
The lowest loss angle is obtaine on sample S1600525, which was deposited without oxygen, low current and low voltage. But it's also the one sample that was deposited in a precedent separate run, and annealed twice at 500C.
A set of substrates have been coated by the Colorado State University Fort Collins group, with ~500 nm tantala and various ion assist beam parameters. Here's a table summarizing the depositions parameter, by Le Yang
main ion source voltage / V
main ion source current / mA
main ion source Ar flow / sccm
target oxygen flow / sccm
assist ion source voltage / V
assist ion source current / mA
assist ion source gas/sccm
thickness / nm
The plot below shows the measured loss angle for all modes of all samples, before annealing. The error bars for the datapoints are from the 95% confidence intervals computed from 8 measurements each. The red line is the average value over frequencies, and the shaded red area gives the 95% confidence interval of the mean value. The loss angle seems reasonably indipendent of frequency.
The following pot then shows the averaged loss angle as a function of the serial number, for reference
There are three main parameters that are changed in the deposition: the assist beam voltage, the assist beam current and the content of oxygen in the assist beam. The plots below show the losses as a function of those parameters. The x axis changes in each of the four panels, and for each plot, the color code is linked to one of the process variables:
Quoting Le Yang and Carmen Menoni
The disks coated at Montreal are hold with three small clips. Therefore there are three small regions close to the edge that are not coated. See the picture below to see one of the samples with the clips.
To check the effect of the clip fingerprints on the dilution factor, I set up a COMSOL simulation. For simplicity, I started with only two small clips as shown below:
The result is that they have a very small effect. The first plot below compares the dilution factor (energy in the coating over total energy) with and without the fingerprints:
Another way to look at it is given below: the plot shows the percentage difference in the computed dilution factor. It's always smaller than 1%, so completely neglegible. In conclusion: we don't need to model the clips.
I ran a series of COMSOL simulations to compute the dilution factors of a coated disk with the dimensions we are currently using (75mm diameter, 1mm thick, 1um of coating).
The mesh is generated as follows:
The plots below shows the effect on the dilution factor convergence of the three parameters above. It turns out that the size of the surface trinagular mesh is the most relevant parameter, followed by hte number of layers in the coating. Instead, the number of layers in the substrate is not particularly relevant.
Using the already installed high voltage feedthrough, I cabled one of the electrostatic actuators (1mm gap between electrodes) and installed it into the chamber. One of the electrodes is connected to the feedthrough cenral pin, the other is grounded on the bottom of the chamber.
The electrostatic actuator is mounted at about 1 mm above the disk, see pictures.
As a preliminary test, I checked that switching the HV amplifier on and off with about 1.5kV produces a visible motion (~2-3 mm) of the optical lever beam. So the actuator is working.
I was looking at some past trend data and discovered that EPICS values were not written to the frames. I added the following two lines to /opt/rtcds/tst/x3/target/fb/master to fix this:
Now EPICS values are saved to frames, but they are all zero! I noticed that we always had the same problems with the cymac2 too.
So for the moment being I set up daqd to save X_NORM_IN1 and Y_NORM_IN1 at 32 Hz. In this way I can monitor the QPD centering.
Apparently, there was a mismatch in the configuration, and DAQD was adding a wonderful 16 Hz comb all over the spectrum.
I stopped the processes, but couldn't restart x3cr1. It turned out that I can't save a channel to frames with a sampling frequency lower than 256 Hz. I changed the model, recompiled and restarted. Now the 16 Hz is gone.
Installed the etched disk: using manually the centering ring allowed me to get the beam on the QPD. A couple of taps to the disk were enough to get the beam centered.
Pump down started at 8:52am
The pressure is at abour 3e-6 Torr. I centered the QPD and started an excitation. The HV amplifier manual states that the driver can source both positive and negative voltage, so this time I didn't add any offset, but simply drove with 1000 V peak to peak. After the excitation the QPD was slightly miscentered in X and I had to manually recenter it.
Good data starting from
PDT: 2016-09-20 16:38:09.330642 PDT
UTC: 2016-09-20 23:38:09.330642 UTC
NOTE: it's a good idea to take a look at both the X and Y signals for each mode. Some of them look stronger in Y than in X. So far I only used X.
New excitation (2000V) at about 8:06am. Had to recenter the QPD again after the excitation.
Engaged the 500Hz high pass filter on the ESD filter bank. New excitation ended at 8:11am. Amplitude 1000 V. Recentered the QPD at 8:11:35am