Tonight, I did some characterization of the Photline fiber-coupled amplitude modulators we will use for our experiment (MXAN-LN-10 --- datasheet attached nope google it yourself). These are electro-optic devices that work by using an internal mach-zehnder to convert phase modulation into amplitude modulation.
The test setup for all measurements was the same. I used the exact configuration that I have been using for the beat (see CRYO:1182), but I simply blocked one laser, so that only one beam was hitting the 1811 PD. The amplitude modulators were inserted (one at a time) between the East laser and its output coupler.
The first thing I did was to investigate the insertion loss of the modulators. We chose the low-loss option, which just meant that the company hand-selected modulators with loss of < 3dB (= 50% power transmission).
I didn't go crazy with precision here, because systematics with fiber coupling can easily prevent a measurement to better than a few percent (an example of this: I installed a 1-meter patch fiber between the laser and the output coupler, instead of the modulator, and I actually saw a slight increase in output power vs. the case with the laser going straight to the output coupler… go figure).
In both cases, I measured very nearly 50% reduction in power (at the top of the MZ fringe---see below) vs. the case with no modulator. So, these things have a loss very close to 3 dB, as advertised. An important thing to point out is that we will need to bias these away from maximum transmission to get a linear PM -> AM coupling, so the real power reduction in our setup will be more than 50%.
These modulators have an SMA-connectorized "RF" input, as well as two bare pins connected to a separate set of "DC" electrodes (they also have two more pins connected to the cathode and anode of an internal PD, presumably at the other MZ output port, which is kind of cool). As far as I can tell, the RF input is also DC coupled, only it is 50-ohm terminated.
I did a DC sweep of both electrodes from 0-10 V while measuring the output power:
(The RF applied voltage range is lower due to sagging from the 50-ohm load).
Fitting these curves, I determined the following Vpis:
These are consistent with the numbers listed on the datasheet.
Next I measured the actuation transfer functions ([RIN/V]) from 1 Hz to 100 MHz, driving the RF input while applying a mid-fringe bias to the DC input, and using
Note the dead zone from 50-500 kHz---this was by accident, as I forgot to check the low-frequency resolution of the RF measurement. I will redo this sometime.
Here are the results:
The response very flat, and roughly what is expected from the DC sweep:
(1/P0) * dP/dV|mid-fringe = pi/Vpi ~ 0.5 ( = -6 dB).
I upped the sample rate of the x1cry model to 64K, in the following way:
The only tricky part is the last step. Changing the sample rate requires the filter coefficients to be updated, so they still match the filter designs. But when you open the filter file in Foton, it does the opposite: updates the designs so they match the old, incorrect coefficients. Since x1cry had only a few filters defined, I went through the file and reverted the designs by hand. (Newer versions of Foton would let you automate this step.)
I spent some time tonight measuring the free-running laser beat noise in various ways. Recall that, as of yesterday, I had tried setting up a couple analog PLLs to no avail and I didn't trust the spectrum I was getting from the Zurich PLL. So, I wanted to measure it another way to see if I could corroborate.
First, eye candy:
Now, an explanation of the various measurements.
I-Q demodulation method
This is a method I have used with some success in measuring the Marconi noise in its quietest state (with no modulation and therefore no means of feedback---see ATF:1877). It is done in the following way:
The main complication here is that, as you can see in the plot, the high-frequency RMS of the beat is several tens of kHz, which means you still have to sample at a high rate to get what you need. The best acquisition scheme I could think of was the Zurich box, which can do 460 kS/s. Still, to take meaningful data, I had to very carefully tune the laser beat to the Marconi LO and then quickly engage acquisition before the (wildly fluctuating) IF signals drifted above the Nyquist frequency (around one second of data was used to make this trace).
That said, the result doesn't look that crazy, and in fact it agrees very well with the DFD measurement that was carried out in a completely different way (see below).
Delay-line frequency discriminator (DFD) method
This is the usual scheme where one mixes a signal with a time-delayed version of itself to create dispersion. What I did:
This method worked swimmingly and reproduced exactly the result I found using the I-Q scheme. The noise floor (cyan in the plot) was measured by sending a quiet Marconi sine wave of the same amplitude and frequency as the beat through the pipeline.
Zurich PLL method
This method is incredibly straightforward. Simply plug the beat (ensuring it's < 1 Vrms and under 50 MHz) into the Zurich box and lock the internal PLL by pressing "ON" on the screen. Route the PLL control signal to one of the front panel outputs and choose the scale factor in V/Hz. I chose the same number as I measured for the DFD (including the SR560 gain) for ease of comparison on the spectrum analyzer.
I'm not sure what to believe. One would think the Zurich PLL is the most trustworthy, but a) I still am bothered by the time-domain behavior I see in the PLL control signal when I adjust the laser beat while watching it, and b) I've generated two nearly identical spectra that differ from it using completely different schemes from measurement to FFT.
All that said, I think the excess noise (and thanks to Dmass for saving me time by pointing this out) is just coming from the ThorLabs drivers, so this should be redone when we have our low-noise ones.
If the "locked indicator" light is not green on the Zurich (first tab, under "Reference", then what you get out is junk (e.g. you have unlocked the lock in, and i hasn't re-acquired yet) - you can do this by kicking it too hard with a frequency shift, which would be easy to do if you were slewing laser frequency, as the coefficients of the laser [Hz/mA] is so big. When the lock in loses the signal, you have to manually re-lock it (toggle off and on the button which has the mouseover text: "enable the fixed center frequency mode of the PLL"). You can get something which sort of looks like a PLL signal which has terrible noise and weird glitchy response when the lock in isn't locked in.
Your instinct to look for slewing at the PLL control point is correct, and a sign that the state of the PLL is healthy/unhealthy
Yes, I noticed this effect. I'm talking about immediately after acquiring---or re-aquiring---PLL lock. I did this several times at different beat frequencies to see what effect it had on the noise (the spectrum changed considerably, which is another bad sign).
On Monday, after I did some inventory of all the parts we have received from various companies, Dmass helped me mount the RIO lasers into their mounts so that I could get started with the optical setup. We cleaned the surfaces with methanol, applied a small layer of silver thermal compound, and then screwed them in.
I then borrowed the following to run the lasers:
After finding the right cables, I powered up the lasers and verified the P-I curve for each as listed on the spec sheets.
I then built a quick (temporary) optical beat setup, combining the two beams on an 1811. I had the temperatures (actually, thermistor resistances) set to what was listed as the testing set point on the datasheet, and as soon as I overlapped the beams and focused them onto the PD, there was already a strong ~50 MHz optical beat.
I have spent some time since then trying to lock various kinds of PLLs, both to interrogate the free-running frequency noise and to get used to controlling the lasers. Some things I've tried:
The first two were not helped by the fairly basic loop shaping afforded by attenuators and an SR560.
I think my next step will be to simply use the I-Q demodulation method (like I did to measure the no-FM Marconi noise in ATF:1877) to measure the frequency noise. I'll compare that to what I get with the Zurich PLL.
(I realized that we should probably use the CRYO elog rather than the SUS one, so I've reposted this here).
Today, we unpacked the IR Labs cryostat that will be the centerpiece of the Cryo SUS experiment.
Everything was more or less in order, except that the baseplate does not have any outward extensions with which to mount the cryostat to the table. Also, the holes for the screws holding the baseplate to the barrel are not countersunk. So, as of right now, the entire cryostat sits on these screws' caps, which is not ideal. We need to either a.) get a new baseplate made up with some wings on it and countersinking for the screws, or b.) work out another way to hold and mount the cryostat (for example, we might want some soft isolating material there anyway, though it will come at the expense of alignment drift).
I followed the instructions and removed the strange anodized heat shield bottom plate that comes with it during shipping, replacing it with the usual one and then resealing the chamber. As directed, I also pumped out the air again---the charcoal getter is not supposed to be exposed to atmosphere for long periods of time.
Photothermal (and other absorption effects) actually measured.
Cavity loss at 300K is 10ppm for west cavity. Fit looks very good, and jives with our understanding of the transfer functions. More to follow.
The 4395A network analyzer in the Cryo lab takes ages to save data to floppy. Someone had hooked up a Prologix GPIB-ethernet adapter to it, but it wasn't working on our network. I set it up as follows:
So, it acts like it has the static address 10.0.5.222 while on the cryo network -- but you can still take it to other networks and use it without monkeying with the settings.
Python scripts copied over from the 40m are installed on gaston, in the directory ~controls/netgpibdata. The AG4395A.py script was tested and works (much faster than the floppy drive).
While testing, Nic and I found and disabled a rogue DHCP server running on the framebuilder.
I think that if you see a signal in the demodulated PDH error point with the cavity unlocked, that this must be RF AM on the light.
What mechanism would produce this much AM? It can't be made through common path modulation of the carrier and sidebands. It must be an etalon formed somewhere between the EOM and the input mirror of the suspended cavity. Could be windows / viewports; this can be tested with the ND filter insertion technique we discussed on Wednesday.
The 4395 saved data in dBm/Hz while displaying Vrms/rtHz, so I had to figure out the conversion factor, and found the following useful table:
The Zurich Instruments HF2 has a very nice built in PLL feature. You give it a sine wave up to 50MHz and it will lock a PLL on it, and it will give the control signal at one of the outputs.
It has a nice PLL design interface (first attachment) (the design can be compared to the measured closed loop gain in the second attachment, this doesn’t include the marconi frequency modulation calibration). You tell it the bandwidth you want and it can internally set it’s PID to achieve that, and it will complain if you are asking it for too much. It claims to be able to get a 50kHz UGF, but I was only reliably able to get like 20kHz.
So right off the bat, this won’t be what we want if our goal is a very high BW PLL to suppress whatever nonlinear noise mechanism Rana and Dmass are worried about.
However, if 20kHz BW is enough, then this might be a pretty nice PLL to use. With a 1MHz modulation range, 20kHz PLL BW, and a 20MHz carrier, the noise is given in the last attachment. (below 1mHz/rtHz up to 1kHz, then starts to rise like f until it hits the PLL bandwidth, and it rolls off again.) The red trace is the spectrum analyzer noise.
data and scripts are all here.
PDH sensing noise:
New lower sensing noise with addition of transformers between LO and RF in both paths.
Cavity unlocked (detuned with temp) noise shows the scatter bump - unsure if this is interesting or should be totally obvious
The LB input noise gets to a minimum of 40 nV/rtHz if we turn the gain 100% up at high freq. This is 4x over what they claim
(shown it is 50 nV/rtHz)
Data is all on svn and will be put into noisebudget update
Looked at noisebudget on SVN and some of the measurements are dated / no loner applicable:
Take and compare:
ISS - what Koji has done with ISS/FSS would be good to copy - sum in DC offset at end of ISS box -
Stefan - input to last stage - change + pin to offset
Check out Nic's old elog after recp't of NEOS drivers - want [ V_rf / V_aud ]
Noise to back of AOM driver (SR785 -> SR560), measure RIN, measure beat, see beat increase
Check out Evan's median subtraction algo (to estimate coupling of RIN) - this takes median of spectra
RIN on transmission - coherent or not?
ND filter before PBS - increase power by X, decrease using ND filter
Isolators good? These are untested-ish - do we want translation stages for these?
Coherent white noise added into error point?
High freq noise at monitor PDs + low PD bandwidth (what PDs are we using / do we see the high RIN on the beat PD)?
Birefringence in coatings leading to multiple peaks with overlapping PDH error signals?
Get FTB-1-1 minicircuits transformers - skip RF grounding for now
New PLL scheme - mix beat down into low band
Other servo boxes if no zurich: Old uPDH box? PDHv1? Misc NIM servo boards?
AD829 xover with SR560 for PLL?
PLOT Look at PLL RMS when beat, make PLL rms go down
Talked to Rana about the increase in sensing noise due to DC grounding in the PDH demod electronics (we were plugging the Gold PD directly into the mixer- he confirmed that we want transformers here)
He agreed that we want to not make monstrously sized ground loops and should use transformers for DC isolation like so:
I borrowed a pair of "balanced to unbalanced" 1:1 transformers in the PSL/CTN lab and stuck them in the readout chain - the noise went down.
I tried all four permutations of balanced / unbalanced on both the LO and RF inputs of the mixer - I saw no difference in noise between these.
There is no grounding done at RF in these mixers (I opened and looked). I have seen grounding put into the transformers in other demod chains.
Slightly downstream of this I (previously) made another change in the setup: here is a picture of what I added between the LB1005 box and the driver:
The LB1005 is a P-I with crossover at 30 kHz (near the cav pole)
The RC filter between the driver and LB acts as 40 dB of low frequency boost
The following measurements are being done this week:
Photothermal transfer function at 120K
Wilcoxen seismometer (accelerometer?) added to setup
Cold seismic noise (boiling noise)
Cold temp noise
Cold beat w/ Weiner subtraction of PDH** / intensity / seismic
**PDH error signal f-domain Weiner subtraction @ low gain regime
We resurrected the intensity loops on Friday. They almost work.
Why is it difficult?
Does it work?
We still have the strangely large intensity noise in both paths.
The lowest transmitted RIN we have measured in lab is:
A plot of the [RIN/Hz] transfer function as a function of offset from cavity resonance is below -
Maximum possible value is 2.5e-5 RIN/Hz
When we have tuned offsets to try to make this go away (in the past), we have not been crazy offset. I think a reasonable estimate based on this, and transmitted power optimization is < 0.5 RIN/Hz
If we had a monstrous offset (~40 kHz), the frequency noise required to produce this is ~1Hz/rtHz, which is well in excess of our previous best beat
=> the intensity noise measured at transmission is not frequency noise + offset
Nic and I are turning on the ISS and taking a beat tomorrow, to see if Intensity noise -> frequency noise coupling is currently limiting (photothermal)
Would there be any way to get white jitter noise which couples into intracavity power but not into the input PDs?
We discovered that we can effect the intensity noise with the half-wave plate right before the cryostat input window - the L/4 plate next to it has no effect.
I would suspect laser feedback**, but measured frequency noise at transmission, and laser output intensity noise are both too low, unless it could somehow be laser feedback induced jitter which looked white up to high frequency?
** because it is so easy to mode hop the lasers even with Faraday's at their output
I was artificially limiting my PLL range by directly driving the Marconi with the IF output of the mixer - this restricted it to ~500mV range.
The input range of the Marconi is (measured to be) +/- 2V (~ 8x higher than the mixer output I was using. Free-ish lunch here.
We might have to wait until we're cold to have low enough drift to use the Marconi in any truly low noise way, but I believe it's no longer limiting at even room temperatures. Beat noise spectra tomorrow.
I replicated Koji's impedance measurement setup for those-who-cannot-afford-an-impedance-mesaurement-kit (a la this elog), but did not see any large dips in the impedance from mechanical features. I guess this is not so surprising, as the coupling of acoustic modes into impedance isn't necessarily large.
After discussing with Chris, we decided to misalign the polarisation into the EOM and measure the transfer function of Volts -> RIN.
Attached are plots of [RIN/V] through the EOM - the magnitude fluctuates with the polarization misalignment, but the phase remained consistent.
Using this as a proxy for [rad / V], it is unclear which one would be easier to use as a high frequency path - The West EOM has a much higher Q single resonance, and larger phase excursion, but we might be able to correct for it electronically, whereas the East EOM has larger more frequent phase excursions at lower frequency.
REGARDLESS, both of these EOMs are in use for generating sidebands, and if we want to have a high frequency phase correction path, we need to purchase two more EOMs**
** we explored using a bias T at the laser to switch to laser current mod produced sidebands to free up the EOMs and save $$, but this made the laser noise go up and we were unable to track down the culprit over a couple days. We also began to explore making a circuit which had resonant gain and DC coupling so that we could use one EOM for both paths, but we found little enough evidence that this was efficient (but it is interesting), and thus abandoned it.
I tried to implement the ISS servo today for the east path with little to no success
It it not clear to me that the servo was engineered with the actual driver in mind
The AOM driver is only linear over a narrow input range: [156mV to 450 mV] (can be + or -)
The ISS board seems to go to not have this offset built in, and it is not clear if the excitation inputs would be an appropriate place to add this.
I played around with using terminators on the input and output signals to try to force the DC value at the output of the ISS board to the middle of the linear range, but could not get the loop to do any actual intensity noise suppression for any switching state on the front
I punt to Nic, interim king of ISS servo board, following the passing of Prince Chaz, rightful heir to the throne, to determine what the correct way to sum an offset into the output of the board is.
It's not pretty, but the servo will create the output offset internally because of the low frequency boost and integrator. When you close the loop, it will push the driver into the linear range.
Noise floor is marconi noise in PLL (see elog:1049)
Explored anomalous intensity noise with Chris. Were able to get sizable decrease in the noise by tuning the L/2 plates on the input beam.
No change to beat spectrum level.
With both temperature loops engaged, beat drift rate is 400Hz/sec.
RMS frequency noise from PDH loops is ~200Hzrms - mostly from Fourier frequencies 100kHz to 500kHz
Might be able to get away with 2kHz Marconi range if we take care of the drift rate => this problem might disappear @ 120K because of the zero crossing
Would need a 400 Hz marconi range (assuming no other noise pops up) to see coating thermal noise @ 1kHz
What noise/range did the DFD box from the 40m end up hitting?
The AOM driver is only linear over a narrow input range: [156mV to 450 mV] (can be + or -)
I looked at the sensing noise (which goes WAY up when the ground of the IF port of the mixer touches the ground of the servo box - from 13 nV to 100 nV)
Sensing noise, input referred servo box noise, and in loop err signal noise are ALL below the current beat limit ( ~ 7e-2 Hz/rtHz around 1kHz ) by ~5x
We should figure out what this noise floor is.
I set the beat back up and locked both loops with new LB1005 + zero-pole between LB and driver topology
Tuned and measured coupling, RAM offset, and modulation depends for both paths.
Locked both paths and tuned gain to minimize error signal RMS
Measured error point of both paths w/ SR785 (see below) - divided by error slope for calibration
Both signals ~100 nVrms/rtHz at floor
Set PLL back up - used level 10 mixer, with amp on beat signal so that beat RF level = +6 dB
Measured beat at input to marconi for a couple different ranges
PLL *not* divided out:
** PLL UGF(s)***
East in loop error signal too high - coherent with transmitted intensity noise - incoherent with input intensity noise - above measured input referred noise of LB1005.
Control / Error < measured transfer function between these two places
More info as I investigate.
Locked with new gain distribution (zero pole pair between output of servo and driver) - UGF of 200kHz achieved (tuned by minimizing RMS of error signal),
error signal RMS = 52 mVrms
Input referred noise of LB1005 measured in this loop state and calibrated
Noise low enough to merit operation in this state ~ 20x lower than previous best ASD measurement in Hz/rtHz
Finally measured the current noise coming out of the drivers using the "Rai Weiss Low Noise preamp"
Measurement limited by johnson noise of load resistor (20 Ohms)
Good measurement up to ~200 kHz - subtracted the noise floor in quadrature (hence noisy above 200 kHz)
Looking at elog:1126, we can see this is smaller than the free running laser noise (which is ~30 Hz/rtHz down to 10 kHz, and goes up as 1/sqrt(f) below that)
The peak around 100 kHz makes the free running noise increase by ~10% (in units of ASD) - it corresponds to ~3 Hz/rtHz laser noise
LB1005 input noise is naturally 42 nV/rtHz at high freq - to get 10 nV/rtHz at high frequency if we add a x4 preamp, and have the gain knob set to +7.0 - the noise starts going up around 1 kHz
The LB box had excessively nigh input noise compared to its spec (42 nV/rtHz minimum measured vs 10 nV/rtHz minimum spec).
I checked out both the 40m and the OMC lab LB boxes to see if one was broken, and Chris
I tested the input referred noise as measured by the output in a variety of states, including the input offset on/off/at different values and the integrator on/off (had to build a low UGF loop and look at out of band noise for this)
Chris & I opened one up to look at it, and it looks like the 40nV noise is after one of the first couple op-amps in the circuit
As an additional clue, Stefan had complained that the LB1005 box noise was too high (interpretation of this data point left as an exercise to the studious elog reader)
The current "i_mon" path is totally worthless and only monitors our DC current level
I propose the following change to the current driver board in order to repurpose the current monitor path into an actual buffered control signal monitor (lower left quadrant of schematic)
This gives us a buffered 50 ohm output for monitoring the control signal and taking transfer functions / looking at noise without modifying our loading (now more relevant because of our passive network to modify gain)
The AD8671 should be fine with a 100 ohm load (50+50) - the RMS of the control signal is small at that point (~7mVrms), and the temperature loops should keep the DC value in check
The AD8671 datasheet (here) says it can be shorted indefinitely, so the brief current saturation during lock acquisition should be ok (the DC values during lock are ~5 V, which is 50 mA with a 100 ohm load)
As the options for TTFSS seemed to be:
Ways we could use the TTFSS:
1) Turn down the gain knobs, remove both feedback path poles, shape loop with passive network at output
This is the same route we were going with the LB1005 box - there is no reason to expect the noise will be good if we attenuate and multiply this much (AD8602s?) at the input - would have to change board topology
If we want to use a passive zero-pole at the output of the box, we toss AC range - if we use the AC range requirements set in elog:1150, and make a pole at 40 kHz, we only have 45 dB of loop gain at DC, which is a bit low for comfort
2) Turn down gain knobs, remove *some* feedback path poles, use pockel's cell path, shape accordingly for high frequency x-over
Need to buy 2 more EOMs and add them into path - would still need to rework PZT path -> current driver because shape / gain are so different
3) Find place to add bypass path in parallel like we did in early generation of PDH board (make P-I x-over with a DC coupled bypass path) - do some whitening (gain/range distribution) with passive network between driver and servo board
This is getting into overly hacky territory - not comfortable cutting up the 40m board, and the TTFSS is too far from what we want to justify buying one to mod it
4) Use its phase corrector path while using some other servo for the laser current (this seems very reasonable) - would either use the LB1005 or new generation of PDH boards built in collaboration with W.Z.K.
This seems very reasonable
We need a better understanding of the LB1005 noise / range / AC range requirements of the servo - see elog:1156
Dmass and I borrowed a spare TTFSS and interface chassis from the 40m lab, and got them working in the shop.
We briefly flirted with using the PZT path as part of our frequency servo, but we were soon disenchanted by the circuit's lack of flexibility. For the time being, we still need the option of running a standalone frequency loop, with high bandwidth. The PZT path is ill suited to this. It puts two poles near 40 kHz, where we'd want a zero to compensate the cavity pole. And its low frequency pole is implemented as a passive low-pass on the output, cutting down the range at high frequency. [Dmass can elaborate on the AC range requirements for this loop.]
Those features are there to make the crossover with the EOM path, while coping with a set of issues our laser's frequency actuator doesn't have: the resonances and capacitive load of a PZT. We could hack around them -- but we'd be hacking a board that has no generic or switchable filter stages, a step backward from the status quo (LB1005 or PDH board). We suspect better alternatives are out there.
We did get far enough along to check the PZT path TF against the TTFSS matlab model (available here). The TF shape was predicted well, but the model had an extra 14dB of gain not found in our measurements. We found out why: the model didn't account for the dividers formed by R7 and R34 with the 100 ohm input impedance of the AD602s. A corrected version is attached to this entry.
% [hOL, hEOM, hPZT] = getTFs(param)
% Example (parameter version 3):
% [hOL, hCross, hEOM, hPZT, info] = getTTFSS(paramTTFSS(3, f), f);
% zplotlog(f, hOL);
function [hOL, hCross, hEOM, hPZT, info] = getTTFSS(param, f)
f = f(:);
% param = paramTTFSS(ver)
function param = paramTTFSScryo(ver, f)
if nargin < 1
ver = 0;
if nargin > 1
f = f(:);
I couldn't find a plot of post demodulation sensing noise for either of the gold PDH PDs
Here is the sensing noise after the LPF for the east gold PD + mixer chain (output of mixer has a 50 ohm terminator T-ed into it, and uses a minicircuits 5 MHz LPF)
The RMS Voltage noise is somewhat high
If I want to calculate the offset from cavity resonance due to the RMS sensing noise (i.e. the ASD shown below) should I divide out the cavity pole or not? aka: pick algorithm 1 or 2 below:
1) Divide ASD (Vrms/rtHz) by plot 3 in elog:1126, then by plot 2, then calculate the cumulative rms (this seems like it might be wrong because it would give me a high frequency slope of f before I tookt he RMS, and white sensing noise would swamp the rms if we include enough bandwidth)
2) Divide ASD (Vrms/rtHz) by plot 3 in elog:1126, then by the DC value of plot 2, then take the cumulative RMS (these calibrations commute since they are both frequency independent factors, and could just be multiplied by the RMS shown below in red to give the offset in Hz)
This is the drawing I am sending around. The idea would be to get a company to make an adapter so that we can have CVI optics for 1550 nm and Viton in the outside window.
Made the above to decrease the PDH board's (currently LB1005) input referred noise without losing so much range that we cannot lock / stay locked
The blue box does: the following when plugged into the current driver:
Swapping the box in and out did exactly what I expect (allowed me to put 20 dB more gain into the LB1005 box)
I didn't get all the gain increase I was expecting - I was only able to increase the gain of the LB by 8 dB; it seems the gain of the loop changed by +10 dB between Tuesday and now.
There is +10 dB extra gain somewhere outside the LBbox compared to the gain on Tuesday
This leaves +7 dB unaccounted for, and no known 50 ohm mistakes to solve the mystery
This is mildly disconcerting, but not horrifying. We should periodically check to see if the plant + sensing gain is drifting around. If it stays at this value, I will just assume I made some mistake in loading (maybe an extra 50 ohm terminator T'ed in somehow)
The attenuator idea works.
The Input referred noise of the LB1005 is still pretty high and shouldn't be hard to beat with the TTFSS board - as soon as we prove this is the case we should get 2 TTFSS boards.
After talking with Rana:
We should be able to lower the AC gain without lowering the DC gain (and thus sacrificing our range) by:
changing R18 in the laser driver to 47 ohms
adding a pomona box between the LB box and the driver with a 450 ohm resistor in series and a 47 uF capacitor in parallel.
I have some nice 47 uF caps in hand.
Summary of East Actuation Range Requirements:
We didn't actually know how much range we needed for the PDH boards
Changed driver back to 1mA/V
Realigned east path 00 mode
Tuned input power (as measured at Refl PD) to 1mW
Tuned waveplates and got 0.78 coupling
Swept cavity via driver input (10 Vpp = 10 mA now that I reverted to old resistor values)
Checked demod phase by symmetry of sideband error signals - was nearly correct
Plant = 2/fcav*sqrt(coupling)*gamma*Pin = 1.8e-8 W/Hz
RFPD = 600 V/W (trusting elog:1126)
I used the LB1005 to lock the loop (t was very easy) - gain knob = 4.25
The error signal RMS was less than 1/3 the peak to peak of the swept out discriminant
I looked at the control signal at various time scales on a scope and monitored its drift rate and RMS:
I quickly checked the loop gain with the SR785 to make sure we weren't in some abnormally low gain state - the gain @100kHz was 10dB, and it was falling smoothly as 1/f up to that point
I then checked the range of the control signal, using the error signal spectrum as measured on a SR785 and the transmitted power to diagnose the relative health of the loop as the control signal drifted
I then tried to measure the locking transients (see attached pictures)
The large smooth part of the control signal transient is presumably due to clunkiness in the knob on the front of the ITC510 (with it's absurdly large Kelvin/Radians coefficient)
New drivers or a cymac based temperature knob via the temp tune input on the back of the ITC510 could solve this. This needs to be addressed if we want to unload ANY gain from the current driver, or we won't be able to lock via the temperature knob, since these control signal swings take up most of the actuation range measured above
I don't know what determines the amplitude of the discontinuous jump at the moment of lock acquisition - is this inherent to the system in some way, or can we make this smaller? They are sizable compared to the range, which bodes poorly for unloading actuator strength
If we can crush the transients somehow we can unload gain from the laser driver and win in the noise budget
Driver input range much larger than output range of servo board in any condition
LB1005 noise too high to be useful
Simulated current driver in LTspice - it has ridiculous input range because of its topology - (we would need to drive HV at the input to blow it)
We can keep turning down the gain by increasing the resistors all the way up to 100k (10 uA/V) and only lose by the output range of the PDH servo board times the gain of the driver
I unloaded this gain (the 10x) into the PDH servo board did a quick check to answer the "how much can we get out of the LB1005 box?" question
Looking at elog:1126, the generous error signal slope is 8.3e4 Hz/V
This is about 20x the predicted coating thermal noise level
I believe this is coming from the output, and we could win more by unloading more gain into the servo board, but past labwork tells me we need ~1MHz range, so that's out.
current driver coefficient changed from 1 mA/V to 0.084 mA/V for both paths
I tried to modify the current driver to turn down its [mA/V] coefficient so that we didn't have such an absurd Hz/V actuator coefficient
I changed R14,15,16,17,19,20,21,22,23 from 1K to 11K on both paths (based on a previous conversation with Rich)
I put a 1 Ohm load across the output of the driver, and measured the response through the driver (both with a 4395 and a Scope)
As expected, the loading (50 Ohms or Hi Z) made little difference since it is in parallel with 1 Ohm
As desired, the response got much lower
Old response: 1mA/V
New response: 0.084 mA/V
I am unsure exactly if/how I have modified the range of the current driver with this change, so I emailed Rich and asked for the spice model we have used for this previously rather than repeat that work
DCC link to current driver for the curious/forgetful/human (here)
I think it should be fine as shown. The first couple of modes are the actual translational ones.
The Mode 8 shape, to me, looks like one where the little mass is moving by itself, but the space and support are not. As long as this is not coincident with the spacer body modes it shouldn't matter too much.
You might try looking into using some SS, but I don't imagine there's too much difference. If the shop has one of the kinds of Spring Steel in stock, that's a good option for flexures (since we're not too worried about UHV specs):
COMSOL model might be incorrectly estimating frequencies because dummy load acted as inertial mass, should just model cavity on supports.
Do we use gravity? Things which adding gravtity to COMSOL gives us:
Simulating the cavity on supports and looking at the ratio of strain energy density in the cavity vs the supports answers:
"If we are isolating the cavity well enough so that most of the damping at the body mode comes from the support and the cavity itself, then the strain energy density ratio times the mechanical loss factors will predict the correct Q"
However, if we are not isolating the cavity sufficiently well, and some of the energy/cycle is leaking out into the outlying structure (or the mechanical screw joints between structures), then COMSOL will predict an incorrect Q. We want to answer in some quantifiable fashion:
"How well are we isolating the body mode and supporting a high Q with our support scheme?"
We could just add mode layers of support structure to the model, but we don't believe we can correctly model the screw joints. I proposed the following recipe to do it all in 1 simulation
I investigated the following geometry in COMSOL:
The square load the top is an ultra dense near infinite young's modulus material (user specified) meant to simulate how the cavity interacts with the support node
This is what it looks like under load (as usual, COMSOL's deformation is not-to-scale):
I changed around the width of the material left, the thickness of the cut, and the height of the cut in MATLAB, to generate the attached plots
The mode shapes are as follows:
If this accurately simulates the eigenfrequency of the cavity under load, it seems we will be in trouble because of mode 8 - none of the recomended cuts pushed this to an eigenfrequency which was lower than the body mode, which seems to imply DC translational force coupling.
If we believe the simulation, we can:
Choose different material (want high density, lowish young's modulus, high yield strain)
Choose different geometry (likely something for a new generation of the experiment/student)
The cavity drum mode on the PEEK supports looked like this:
I constrained the bottom half of the support points, and found the eigenfrequencies'
The drum mode (pictured above) is at f = 37190 Hz
N.B. I didnt have the windows or vent hole in the model, but the frequency agreed fairly well with experiment and evans previous (support free) result.
I don't have a good number for loss of the PEEK parts, so I am unsure if this is wholly responsible for our low Q. It would be interesting to plug in the loss numbers (by exporting the 3D shape to MATLAB and following the strain energy density ratio times loss formula)
Another thing we didn't previously have good numbers for was contact area.
These are analytically solved problems (see Wikipedia's contact mechanics page)
I calculated the relevant contact areas to be:
Next up: Silicon cavity with new Al 6061 supports COMSOL model
Area of most confusion: the Russians commented that the best way to support a high Q mode was to put something very soft (e.g. gold foil) between the cavity and the support node. I am unsure how to properly model this is COMSOL / properly say what it will do.
The question: if I take a rigid body (Si cavity), place it on supports which have little/no mechanical response below the mode we're interested in (drum mode), and slip a thin layer of gold between those two, what happens to the transfer function between the cavity and the support?
We were holding the cavities in a very dumb way. We will now hold them in a less dumb way.
The original cavity supports look like this:
The axis of the support is parallel to the axis of the cavity, which greatly overconstrained how we were holding the cavity (4 line contacts)
The new cavity support looks like this
and is optimally constrained (4 point contacts on the bottom of a cylinder)
Since this is what the cavity sits on, it would be bad if there were accidental resonances at the cavity body mode frequency.
Here is a plot of the eigenfrequencies of each support:
The modes around 0 Hz are all artifacts of the COMSOL simulation (translational body modes from not being constrained)
The PEEK support has a mode relatively close to the cavity body mode:
The equivalent mode in the 6061 aluminum support is:
For reference, the lowest frequency mode in the new standoffs is:
And seems very decoupled from longitudinal motion of the cavity.
Relevant material properties:
We want to replace the cryostat windows with fused silica (IR grade) windows with a Viton seal on both sides. We will get the windows coated by CVI (or someone) for 1550 AR on both sides. Need the existing hole pattern in order to continue.
Some useful links:
From MDC Webpage:
Viewport Sealing Methods
Can't use current modulation based sidebands in "as-is" setup - there were HUGE (5%) line harmonic (60, 120, 180) dips in transmission which got worse as we started gain peaking.
When I switched to using the Pockels cell for PM sidebands, here were no visible dips in transmission visible on the scope when triggered on the AC line.
If we want an actuator for the phase corrector path, we either need to hunt for and find/mitigate the source of this ground loop, buy another Pockels cell, or add a DC summing path into the resonant sideband circuit
I borrowed a TTFSS from the 40m lab to see if its HV stage may be useful for the cryo phase correction path. Attached is the transfer function of the HV board, measured from TP1 to TP4 -- looks similar to what we previously implemented on the PDH board. (TP4 should be -66dB below the HV output path in this measurement, due to the voltage divider formed by R51 and the input impedance of the analyzer.)