40m QIL Cryo_Lab CTN SUS_Lab TCS_Lab OMC_Lab CRIME_Lab FEA ENG_Labs OptContFac Mariner WBEEShop
  PSL, Page 8 of 52  Not logged in ELOG logo
ID Date Author Type Category Subject
  2257   Thu Nov 22 12:49:35 2018 anchalDailyProgressTempCtrl 

The configuration run:

I ran overnight PIDAutoTune with following parameters:
RelayAmplitude = 0.75 W
Offset = 0.25 W (So Differential Heater switches between -0.5W to 1W)
Setpoint = 34 MHz
Initial Differential Heater (Actuator) = 0 W
Common Heater = 0.5 W
Runtime = 15 hrs
This gave the following results:
Critical Gain, Kc = -12.27416
Critical Time Period = 84.3s (Note the reduction in time from the last measurement)
PID constants: Kp = -2.4548, Ki = -0.05824, Kd = -68.98078


Conclusions and next steps over the long weekend:

Today, I manually brought the frequency to near 27.3 MHz and switched on the PID with the estimated parameters. It looked like it was doing a good job except for momentary jumps in the actuation. This might be readout error or simply that we need some sort of LPF on the actuation calculation. Even 0.1s actuation on South heater causes a big problem because of huge change in current and because the south heater is a much more effective heater than north.

Just to check my hypothesis, I modified the PIDLocker_beta.py code to create a condition that actuation will be changed only if it is changing by less than 0.5 or if it has been asked to change in last two timesteps. I see that this successful in avoiding the harsh jumps. Now I'm running it to see how it works with an initial error of about 5 MHz. It is much more than I would like to start this PID (for reasons stated below) but I need to go for the weekend now.


Thoughts:

  • In my opinion, since the PID parameters turned out to be so huge, they will properly work in small deviation from setpoint only. Definitely within ~50 MHz of setpoint which was maximum amplitude during Relay tuning.
  • This also suggests that we probably cannot create an all regime working PID with our actuators. We simply need more ability to actuate if current cooling times are going to be constant for a faster convergence to the desired setpoint from far off place.
  • I and Awade were discussing this yesterday. If we ant convergence to 27.34 MHz from initial points as far away as ~500-900 MHz, we need some artificially intelligent code.
  • Not too intelligent though. In awade's words "we need something dumber than neural networks (because of the time involved in making and training them) and smarter than a PID".
  • I'm thinking of a code which looks at higher derivatives than first, maybe up to the third derivative.
  • If we get a code like this, which we can train during night times it might be a good thing to have to save about (6hrs to a day) while bringing the frequency to the desired point. I'll keep this at a lower priority for now though as we have other pressing issues to work on.
  2256   Wed Nov 21 19:33:23 2018 awadeMiscElectronics EquipmentSR785 (SN46950)

The main SR785 (Serial number 46950) has a dead screen. 

I power cycled the unit and also checked that the screen brightness wasn't set to zero.

Getting it serviced at Stanford Research Systems can be expensive.  In the past they have just sent the CRT and it is easy to replace ourselves.  I'll put in a request for a quote and orgainize for the screen to be replaced later this week.

  2255   Wed Nov 21 17:54:00 2018 anchalDailyProgressTempCtrlNotes about Cavity Heater PID

Last night I ran PIDAutoTune.py for Cavity heaters.


Breif description of this code:
PIDAutoTune.py : Originally written by awade, this code implements Relay method to find PID constants for a  general PID setup with EPICS channel. For now, one needs to make some changes in the code itself for different setups, but I'm trying to make it more general slowly. Relay Methos involves in actuating in a hard step function manner. The code simply switched back and forth between +h and -h (h is called Relay Amplitude ahead) according to which side of setpoint the process is. The system ideally reaches an oscillatory state and the amplitude and time period of these oscillations provide means to estimating the PID constants. Refer 'Relay-based PID Tuning' by D. Williams for more details.


So after running this code for the first time, I understood the direction (or sign of Kp) required. Using these and with a more settled pre-cavity beat frequency, I ran the code again with the following parameters:
Relay Amplitude = 1.0 W
Initial Differential Heating (Actuator) = 0.0 W
Common Heating = 0.5 W
Setpoint = 500 MHz
This gave the following results:
Critical Gain, Kc = 0.107
Critical Time Period = 1663 s (Note the high oscillation time)
PID constants: Kp = 0.0214, Ki = 2,576e-5, Kd = 11.87


Manual feedback:

So the above values obviously didn't work like any other first result. But I found important insights from them (listed below). Then I just manually tried to bring the frequency down to the region of interest (around 27 MHz) and this manual tweaking also gave me some intelligence about these cavity heaters (listed below). I was able to reach around 34 MHz with an almost stable beatnote.


Conclusions from this run:

  • The heaters of the two cavities are not equally effective. The South Cavity heater is almost twice as better as the North one. So because of this, with our present PID which uses differential heating, we can not do this very well. So I will probably make a provision for setting an offset from 0 around which the differential heating should be actuating.

  • The PIDlocker needs to be adapted with lasers crossing each other in frequency. This basically changes the required direction of actuation. We need to put in something which detects if the lasers are flipped so that the sign of PID constants is changed. For this, I'll put in a new channel for changing signs of the constants and modify PID locker accordingly.

  • Because of the long critical time period, the Kd constant will probably need to be much more than the allowed region in HOPR and LOPR. I have already made it +-50 in the database file.


Current measurement:

I'm running a new PIDAutoTuner measurement:
RelayAmplitude = 0.75 W
Offset = 0.25 W (So Differential Heater switches between -0.5W to 1W)
Setpoint = 34 MHz
Initial Differential Heater (Actuator) = 0 W
Common Heater = 0.5 W
Runtime = 15 hrs
I'm hoping I'll get more accurate data this time as the frequency is very stable with at most 100 kHz fluctuations pk-pk.

 

  2254   Tue Nov 20 17:31:13 2018 anchalDailyProgressOtherPrecav Beatnote Freq counter and PID optimization

I learned today from Andrew on how to setup the pre-cavity beat note frequency counter and use it for cavity heater PID. Now, I need to figure out good PID parameters to start heating up cavities in a controlled fashion. For now, I have kept the differential heating off to keep the cavities stable.

But then, I realized that the Slow PIDs in FSS aren't functioning. Whenever I switched them on, it would take the SLOWOUT so far away that the frequency lock will break. So to fix this, I read more about PID tuning and figured that the pre-defined PID constants for Slow PID might be wrong. I ran the PIDAutoTuner (which I have changed a little bit, see git commit) which gave very different values for these constants. Further, the newly found values were even worse. So I started looking into the PID code itself and when I was checking the error signal, the obvious truth revealed. The FASTMON channels which are used as the process variable in the SLOW PID were not being updated. Then I realised a whole batch of channels are stuck to 0.

So I restarted the channels by restarting the AcromagBoot.conf service on the acromag computer. This suddenly made everything SUPER UNSTABLE. Even with locks and autolockers switched off, the laser frequency was just dancing and giving a light show out of the PMC! While trying to fix this, I also realized that the variable gains in the PMC controls have been reset to 0 in some cycle of restarting services. So I found the right values again by monitoring the error signal.

Finally, I had to restart the acromag computer as it was running 2 different instances of EPICS channel causing the above mashup.

After everything got settled back, I ran the PIDAutotuner code and it updated the PID constants values in the SLOW PID. The setpoint for both south and north SLOW PIDs is 1V. I have updated default values in .db files.
 


PMC Variable Gain value:

South: 30 dB (No oscillations happen upto here, so no harm in keeping it here unless we find any wacky behavior)
North: 22 dB (Oscillations start at 24 dB, I'll increase the margin if need be)


SLOW PID parameters:

  K_p K_i K_d
South -0.00040 -0.00017 -0.00062
North -0.00041 -0.00015 -0.00076

My take:

I personally feel the SLOW PIDs are just fine with a long range of values unless we have a better measure of what is bad and what is good, particularly, along with the cavity heater PID. Because SLOW PIDs for frequency are important only when we are bringing the cavities to 26.5 MHz apart. Once they are there, it doesn't matter much. I'll think more about this.


Near Future Steps:

Once I set the cavities on course to reach 26.5 MHz, I'll start finishing the code to take weekly noise budgets. Then I'll work on the weekdays to improve something in the experiment and check with noise budget taken during the weekends. I'll start with optimizing lens positions and then installing ISS.
 

  2253   Sun Nov 18 18:50:39 2018 anchalSummarySouth CavitySouth Path up and running again

The South PMC is installed and the South FSS is running again. Attached is the beam path profile from ala mode used for the south path.
While aligning this beam into the cavity (which is a very difficult task apparently), I disturbed the north path by mistake. So I had to align both paths from scratch again.

Few notes for aligning beam into cavities in future:

  • The only way to know if the beam is reflecting off the mirror, not the enclosure or something else inside the tank, is to bend and look through an IR viewer. So my first step would be to do this and ensure the beam is at least hitting the mirror. This step requires the removal of the thermal shield.
  • After this, it takes a while to actually start seeing the reflection from the mirror. I bend to see into the cavity to see if the reflection from the mirror is hitting something else and change angles accordingly.
  • At this stage, once we get the reflection, we very nicely just match it with the incident beam. BUT, make sure you are not seeing anything in the transmission. It is possible that the beam is simply going through the spacer. A way to check if the transmitted light is through the cavity or spacer is by changing laser frequency (Slow input to laser) slightly.
  • Once we get the coincident reflection with no transmission, scan the slow input about 0.75 V to see if any of the modes are excited. If you get lucky, some lower order mode gets excited to give some information about how wrong the incident beam is.
  • From this point onwards, it is just walking of beam all over the mirror of the cavity and scanning laser frequency to get any glimpse of the fundamental mode.
  • Once the fundamental mode is caught, we actuate the fast input to the laser with ~3 Vpp sinusoidal signal at around 4 Hz. Adjust the periscope to get some tangible transmission and then just do the usual optimization by walking the beam.

Important things to remember:

  • Make the autolock condition of PMC less hard so that it remains locked with the fluctuating laser frequency.
  • Every time you are moving a knob, double check if you are moving the right one. This is how I fucked up a perfectly good alignment on the North side.

Also, I have made correct length cables for FSS LO delays for both paths and both cavities are getting locked nicely. At North cavity, I see ~70% mode matching and at South cavity ~60%. These were measured with laser power meter.

Future steps are to align ouput of the cavities onto Beat note detector and start measuring noisebudget weekly. After that, we'll concentrate on increasing mode matching and reducing noise sources in the paths.

Attachment 1: SouthOptimizedPath.pdf
SouthOptimizedPath.pdf
  2252   Mon Nov 12 11:53:50 2018 anchalNotesopticChanged half waveplate angle after South EAOM

For some reason after the weekend, I found that the laser power on the south path decreased by about 300 uW. On checking power with a laser power meter, I found that a PBS, right after South EAOM and a half waveplate, is dumping more than half of the power as it is vertically polarized. The half waveplate is supposed to rotate the vertical polarization from EOAM into horizontal for cleaning by the PBS. I have rotated this waveplate to get maximum output from PBS.

Still, the loss at this stage seems much more in comparison to North Path.
Loss at North Path after EAOM  due to horizontal polarization at output = (354 uW)/(2431 uW) x 100 = 14.56%
Loss at South Path after EAOM due to horizontal polarization at output = (545 uW)/(1957 Uw) x 100 = 27.85%
I have checked that input polarization is correctly vertical for both EAOMs.(Only about ~1% light horizontally polarized for both paths at this point). Also, both EAOMs (New Focus 4104) are currently terminated with 50 Ohm terminators at their modulation ports.

 

  2251   Tue Nov 6 13:41:04 2018 anchalSummaryPMCPMC installed on South Path

Changes:

  • The PMC on the south path has been installed with the PMC servo card on the rack.

  • I added sliders for setting max, min and step size of autolocker scanning in PMC interface medm windows.

  • I used one PLCX-25.4-103.0-UV-1064 and three steering mirrors in front of EOM to mode match with PMC.

  • I laid down new cables for PMC reflection RFPD signals and EOM.

  • I removed the power supply providing 9V to PMC Servo Cards and made a voltage regulator box with LM7809CT which converts 3-pin 18V to two BNC 9V.

  • We are now using external mini-circuits in-line mixers (ZFM-3-S+) and low pass filters (SLP-1.9+) instead of using onboard mixer and filter on PMC servo card and injecting signal through FP1Test input.

  • I mounted the inline hanging mixers and splitters on the rack with LO delay lines inside the rack so there are no hanging parts now.

  • I changed requirements in ALConfig_NPMC.ini and ALConfig_SPMC.ini so that we can inject the signal from FP1Test and autolocker still functions.

  • The HV supply of PMC servo boards was found to be around 180V. We changed it to the designed 160V.

  • Attaching few pics for above changes.

Few numbers:

South PMC Mode matching ~ 70%
Modulation frequency = 14.75 MHz
Modulation Index \approx 0.142 rad
 

Few notes:

  • When aligning to PMC, it is a good idea to keep a white screen infront of transmission and work in dark with IR viewer to get an idea of what is happening.
  • A camera on the back side of PMC also helps when sufficient light starts building up in the cavity.
  • While finding optimum LO delay using the SRS delay box DB64, use the rated 2.5 ns extra lag written in the datasheet.
  • Once maximum swing in PDH error signal is found, check locking once and see if the transmission is good or not. You might be locking to the sideband as the servo locks to one kind of slope only. If you find feeble transmission, just add T/2 delay in the delay box where T is the time period of modulation frequency. This lands you to near right spot.
  • While soldering an SMA connector to the cable, put the heat shrink in first and past it to the cable before proceeding. I made this mistake few times.
Attachment 1: South_PMC_Alignment.jpg
South_PMC_Alignment.jpg
Attachment 2: NorthAndSouthPMCServoCardsonRack.jpg
NorthAndSouthPMCServoCardsonRack.jpg
Attachment 3: 9VConverter.jpg
9VConverter.jpg
Attachment 4: FGMixersLPFandSplitters.jpg
FGMixersLPFandSplitters.jpg
  2250   Fri Oct 19 17:05:10 2018 anchalSummaryNoiseBudgetSummary of present noise budget

Attached is the latest full noise budget we have. I generated it by rerunning current noise budget notebook. The aim of this post is to create a checkpoint of where we are, how the analysis was done and what we are going to do in the future.

Current Measurements:
There are 5 measurements of ASD of the beat note currently. 1 set (dark red) is from un-dated data (committed as ported from 40mSVN). Others have dates marked on them. These data sets are the spectrum of beat note frequency. No post-processing is done on these measurements and they are plotted as measured.

Coating Brownian Noise (Green):

  • This is calculated using Harry et al. (2001) Eq 21 (written slightly differently here with only coating contribution)
                                              S_x^{\text{(cBr)}}(f) = \frac{2 k_\text{B} T}{\pi^2 f} \frac{d \phi_\text{c}}{w^2 E_\text{s}^2 E_\text{c}(1-\sigma_\text{c}^2)} \left[E_\text{c}^2 (1+\sigma_\text{s})^2 (1-2\sigma_\text{s})^2 + E_\text{s}^2 (1+\sigma_\text{c})^2(1-2\sigma_\text{c})\right]
  • Here, d: Coating Thickness  =  4.6806 \pm 0.0004 \mu m
              w: Beam spot size on mirrors = 215.4 \pm 0.5 \mu m
              \phi_c: Coating Dielectric Loss Angle assuming \phi_c = \phi_{||} = \phi_{\perp} = (2.41 \pm 0.2)\times 10^{-5}
              E_s: Substrate's Young modulus = (72 \pm 1)\times 10^9 Pa
              E_c: Coating Young modulus =  (100 \pm 20)\times 10^9 Pa
               \sigma_s: Substrate's Poisson ratio = 0.170 \pm 0.001
               \sigma_c: Coating Poisson Ratio = 0.311 \pm 0.062
  • Noise from above formula is in displacement. It is converted into frequency noise using the conversion factor for the cavity, f_{conv} = \frac{c}{L \lambda} = (7.65 \pm 0.05)\times 10^{15} Hz/m.
  • It is the highest estimated source of noise in relevant frequency ranges.
  • This is an oversimplified model and we are working on incorporating analysis due to Hong et al. (2013) for calculating this noise.

Coating Thermo-Optic Noise (Pink):

  • This is calculated using analysis given in Evans et al. (2008). Particularly, Eq 4:
                                              S^{\Delta z}_{TO} = S^{\Delta T}_{TO}(\bar{\alpha_c}d - \bar{\beta}\lambda -\bar{\alpha_s}d\frac{C_c}{C_s})
  • Here, S^{\Delta z}_{TO}: Noise spectral density of displacement noise due to thermo-optic noises.

                    S^{\Delta T}_{TO}: Profile-weighted temperature fluctuation PSD, calculated using:
                                                                S_T(f) = \frac{2^{3/2} k_\text{B} T^2}{\pi\kappa_\text{s}w}M(f/f_\text{T})     with          M(\Omega) = \Re\left[\int\limits_0^\infty\! \mathrm{d}u\, \frac{u\ \mathrm{e}^{-u^2/2}}{\left(u^2-\mathrm{i}\Omega\right)^{1/2}}\right]      
                       \bar{\alpha_c}: Effective thermoelastic coefficient for coating, calculated using Evans et a. (2008) Eq A1 and A4:
                                                                \bar{\alpha_k} = \alpha_k \frac{1+\sigma_s}{1-\sigma_k}\left [ \frac{1+\sigma_k}{1+\sigma_s} + (1-2\sigma_s)\frac{E_k}{E_s} \right ]        and        \bar{\alpha_c} = \sum^{N}_{k=1} \bar{\alpha_k} \frac{d_k}{d}
                                          Here,  \alpha_k : Thermoelastic coefficient for the k-th layer. \alpha_{AlGaAs} = (5.24 \pm 0.52) \times 10^{-6} K^{-1} and \alpha_{GaAs} = (5.97 \pm 0.6) \times 10^{-6} K^{-1}
                                                     d_k : Thickness of each layer. This is read from coatingLayers.csv which contains optimized. There are 57 layers in total.
                                                    Rest variables are the same as previously stated.
                      \bar{\beta}: Effective thermo-refractive coefficient for coating, calculated using Evans et a. (2008) Eq B3-B8. This is bit tedious calculations so I'll leave out the details.
                      \lambda : Wavelength of beam light = 1064 nm
                     C_c: Effective heat capacity per unit volume of the coating,C_c = \sum^N_{k=1} C_k \frac{d_k}{d} where C_k is the heat capacity of the kth layer.
                                                                        C_{AlGaAs} = (1.698 \pm 0.001)\times 10^6 \frac{J}{Km^3}        and       C_{GaAs} = (1.75 \pm 0.09)\times 10^6 \frac{J}{Km^3}
                      C_s : Heat capacity per unit volume of substrate = (1.6 \pm 0.1)\times 10^6 \frac{J}{Km^3}
                      Rest variables are as stated before.

  • Same as before, the calculated noise is displacement noise and is converted into frequency noise.
  • This noise has been reduced significantly by making thermoelastic and thermorefractive noise cancel each other Its further analysis is at low priority as it is well below other noise sources.

Substrate Brownian Noise (Yellow):

  • This is calculated using Cole et al. (2013) Eq.1.:
                                              S_x^{\text{(sBr)}}(f) = \frac{2 k_\text{B} T}{\pi^{3/2} f} \frac{1-\sigma_\text{s}^2}{w E_\text{s}}\phi_\text{s}
  • Here all variables are as stated before.
  • Note: This is less by a factor of 2 from expression in the paper because the expressions in papers are for 2 mirrors together. We in our calculation take into account the number of mirrors separately.
  • This noise source is the third highest source of noise in relevant frequency ranges. It is hard to make any change other than changing substrate material itself to reduce this noise source.
  • This is a much better-understood source of noise and there are no proposed changes in its analysis for now.

Substrate Thermo Elastic Noise (Sky Blue):

  • This is calculated using Somiya et al (2010) Eq 3 and 8: (Analytical expression for the theory by Cerdonio et al. (2001))
                                S_x^{\text{(subTE)}}(f) = \frac{4 k_\text{B} T^2}{\pi^{1/2}} \frac{\alpha_\text{s}^2 (1+\sigma_\text{s})^2 w}{\kappa_\text{s}} J(f/f_\text{T})       where    J(\Omega) = -\operatorname{Re}\left\{\frac{\mathrm{e}^{\mathrm{i}\Omega/2}}{\Omega^2} (1 - \mathrm{i}\Omega)\, (\operatorname{Erfcom}\!{\left[\frac{\Omega^{1/2}(1+\mathrm{i})}{2}\right]})\right\} + \frac{1}{\Omega^2} - \frac{1}{(\pi\Omega^3)^{1/2}}   
  • Here, \kappa_s : Substrate Heat Conductivity = 1.38 \pm 0.2 W/(K m)
              f_\text{T}: Thermal relaxation frequency calculated by Cerdonio et al. (2001) Eq 9, f_T = \frac{1}{2\pi}\frac{\kappa_s}{C w^2}
              
  • Note: There is a difference of $\sqrt{2}$ factor from Somiya and Cerdonio's expression because w in Somiya et al. and r0 are related as $r_0 = w/\sqrt{2}$ as explained in Black et al. PRL 93, 241101 (2004).
  • This noise source is the highest noise source up to 90 Hz and is second to Coating Brownian Noise in the above frequencies.
  • There no proposed changes in the analysis for this source for now.

PDH Shot Noise (Orange):

  • This is calculates using (in W^2/Hz):
                                                              S_P^\text{(PDHshot)} = 2h\nu P_0 \left[J_0(\Gamma)^2 (1-\eta) +3 J_1(\Gamma)^2\right]
  • Here, \Gamma: PDH modulation index of FSS = 0.2 \pm 0.005 \, rad
              \eta: Cavity Visibility = 0.35 \pm 0.05
            P_0: Incident power on FSS EOM = 3 \pm 0.2 \, mW
  • This is then converted into frequency noise spectral density by using PDH Slope (Also accounting for cavity pole):
                                                                S_f^\text{(PDHshot)} = S_f^\text{(PDHshot)} \left(\frac{1 + \frac{f}{f_p}}{\Gamma'} \right)^2
  • Here, f_p : Cavity Pole = 136 \pm 9 kHz
              \Gamma^' :  PDH slope = 8.85 \pm 0.86 mW/Hz
  • This is a fairly low noise source and is a straightforward calculation. No updates required here other than remeasuring \Gamma and \eta .

PLL Oscillation Noise (Grey):

  • This is directly measured and the data on the attached graph is from 2012.
  • This data should be taken again and if possible should be included in automated scripts.

PLL Readout Noise (Purple):

  • This is calculated with what is referred to as "Tara's Magic Number":
    S^{f}_{PLL} = (f\times0.0207\times5.04\times 10^{-5})^2
  • I have no idea how this formula emerged. It would be best to measure the noise from PLL again and dissect and measure all independent noise sources again.

Seismic Noise (Black):

  • This was measured last in 2011 October with a seismometer on the table.
  • This doesn't affect much our region of interest but should be updated as we are 7 years in the future now.

Photothermal Noise (Brown):

  • This is the coupling mechanism of intensity noise into frequency noise.
  • To calculate this, following conversion from incident power noise to photothermal noise is used:
                                           S^{photoThermal}_f(f) = |H(f)|^2 P_{abs}^2 S_{RIN}
  • Here,  H(f) :  Photothermal Transfer function in (m/W)
               P_{abs}: Power absorbed by mirror calculated by P_{abs} = \alpha_c \frac{P_0 \mathcal{F}}{\pi}
                              Here, \alpha_c : Coating absorptivity = (6 \pm 1)\times 10^{-6}
                                        P_0: Incident Power
                                         \mathcal{F}: Cavity finesse = 15000 \pm 1000
                S_{RIN}: Relative Intensity Noise Spectral Density
  • Relative Intensity Noise is measured directly. For the plot attached, RIN was measured only on the south path and north path's RIN was assumed same.
  • Photothermal Transfer Function is calculated due to Farsi et al. (2012) Eq. A51:
                              H(f) = H_\text{c}(f) + H_\text{s}(f) + H_\text{tr}(f)
  • Here, H_\text{c}(f) : is contribution through the coating, calculated using Eq. A44 (Details excluded)
               H_\text{s}(f): is contribution through the substrate, calculated using Eq. A45 (Details excluded)
               H_\text{tr}(f): is contribution through the thermorefractive process, calculated using Eq. A49 (Details excluded)
  • This noise, it turns out is quite low in the relevant frequency range. No further updates are required in the analysis.
  • The RIN needs to be measured again though for both the paths.

Residual NPRO Noise (Green):

  • This calculation requires free-running frequency noise of NPRO. It is assumed to be following which is taken from Wilke et al. Opt. Lett. 25, 14 1019-1021 (2000):
                                              \sqrt{S_{\nu}(f)} = (10^4 \text{ Hz/Hz}^{-1/2})\times(1 \text{ Hz}/f) .
  • Open Loop Transfer Function of FSS was measured in 2004 (see PSL:1504). The measurements are fitted with model transfer functions to get zeros and poles which are then used to estimate OLTF at frequency points of interest.
  • The assumed NPRO free-running noise is then suppressed by these OLTFs (north and south) and added in quadrature.
  • We need to measure the OLTFs again and if possible, measure the true free-running frequency noise of the NPROs.

Near future plan:

We should be able to finish aligning South Path with the newly installed PMC by end of this week. Then if everything works fine, we should start taking beat note measurements every week and focus on updating old data in the estimate plots.
 

Attachment 1: 20181019_154643noiseBudget.pdf
20181019_154643noiseBudget.pdf
  2249   Fri Oct 12 14:41:17 2018 awadeSummaryEnvironmentTemperature logging PSL lab (Sep-Oct 2018)

Temperature data in the ATF (QIL) lab has been collected for a month now.  Data is sampled at 0.1 s intervals and saved on WS2. I'm not attaching the full data here (its about 4 Gb/month in csv format) but the minute trend is included along side the plots below.

Plotted below is the hour trend from September 11th up until October 13th 2018* (UTC).  

The Thermostate trace shows temperature as measured at the AC control sensor.  EnvMon is the themerature at the center of the table (sensor is mounted to table directly). 

 

*Note that these sensors were never calibrated for their absolute offset (but they should be pretty close).  Drift of the transducer box is unknown.  It uses a LT1128 op amp to convert AD590 sense current into voltage.  This is not the right choices.  Also the resistors are the lower grade radial type thick film, which is also not the best choice for immunity to circuit induced drift.

Attachment 1: 20181011_PSLLabTemp_Sep9toOct11_AbsUncal_HourTrend.pdf
20181011_PSLLabTemp_Sep9toOct11_AbsUncal_HourTrend.pdf
Attachment 2: MinuteTrend_Sep_PSLLabTempData.tar.gz
  2248   Thu Oct 11 12:02:46 2018 anchalSummaryPMCSouth PMC servo card ready to use

I have mounted south PMC servo card on a 1U chassis with two acromag cards. The card should be ready to use.

Changes to Servo Card:
I switched resistor R44 with 39k (as we did in North PMC card) to create a 10Hz pole with PZT capacitance. The updated schematic is attached.

Acromag XT1541-000 Configuration:
Attached .zip file contains configuration file and the screenshot of configuration info.
IP Address: 10.0.1.49
Subnet Mask: 255.255.255.0
Gateway Address: 10.0.1.1
Host Name: XT1541-000-SPMC-DAC

XT1541-000-SPMC-DAC Pin Connections
Signal Name Acromag Card Pin Servo Card Pin Notes
PMCSW1 DIO 0 P1-6A Connected through resistor divider (see option D, attachment, PSL:2058 and PSL:2070)
R1 = 3k || 4.3k ||150k \Omega = 1746.55 \Omega
R2 = 3.3k || 3.9k || 470k \Omega = 1780.73 \Omega
This gives Voff = 0.66 V , Von = 4.5 V with Vexc = 9 V on the DAC and Rpullup = 10k  \Omega
PMCSW2 DIO 1 P1-7A Connected through resistor divider (see option D, attachment, PSL:2058 and PSL:2070)
R1 = 3k || 4.3k ||150k \Omega = 1746.55 \Omega
R2 = 3.3k || 3.9k || 470k \Omega = 1780.73 \Omega
This gives Voff = 0.66 V , Von = 4.5 V with Vexc = 9 V on the DAC and Rpullup = 10k  \Omega
BLANKING DIO 2 P1-9A Connected through resistor divider (see option D, attachment, PSL:2058 and PSL:2070)
R1 = 270 || 390 ||4.7k \Omega = 154.31 \Omega
R2 = 300 || 330 || 390k \Omega = 157.08 \Omega
This gives Voff = 0.66 V , Von = 4.5 V with Vexc = 9 V on the DAC and Rpullup = 1k  \Omega
MGAIN2 OUT 00 P1-4A -
INOFFSET2 OUT 01 P1-5A -
PMCRAMP OUT 02 P1-6A -

Acromag XT1221-000 Configuration:
Attached .zip file contains configuration file and the screenshot of configuration info.
IP Address: 10.0.1.48
Subnet Mask: 255.255.255.0
Gateway Address: 10.0.1.1
Host Name: XT1221-000-SPMC-ADC
All channels are configured for range \pm10 V with averaging factor set to 200.

XT1221-000-SPMC-ADC Pin Connections
Signal Name Acromag Card Pin Servo Card Pin Notes
LODET2 IN0+ P1-1A RTN and IN0- connected to GND
PMCERR IN1+ P1-2A RTN and IN1- connected to GND
PMCOUT IN2+ P1-3A RTN and IN2- connected to GND

 

Attachment 1: D980352-E01_South_PMC-Mods-AG-10-11-18.pdf
D980352-E01_South_PMC-Mods-AG-10-11-18.pdf
Attachment 2: SouthPMCBoard.jpg
SouthPMCBoard.jpg
Attachment 3: South_PMC_Acromag_Configuration.zip
  2247   Mon Oct 8 15:10:17 2018 anchalSummaryPDTransimpedance and Dark Noise measurement of all RFPDs in PSL

tl;dr:

Transimpedance measurements of all RFPDs in PSL were taken at the setup at 40m. All measurement results and data are in the git folder link below. Dark Noise measurement was also attempted but it turned out that the measurement is limited by Agilent 4395A's noise floor and hence this measurement needs to be done again using low noise preamplifiers.

Data:
Git folder containing data, plots, results and jupyter notebook.
https://git.ligo.org/cit-ctnlab/ctn_labdata/tree/master/data/20181005_CTN_Lab_All_RFPD_TF_Data

Measurement Method:
I used the transimpedance measurement setup at 40m to take these measurements. It includes a laser current driver which is modulated through a bias-tee. This, in turn, gives an amplitude modulated laser. The laser is then sent through 2 focusing lenses to fall on a beamsplitter. One output of this beam splitter is read by a reference photodiode whose transimpedance is known to good accuracy. The other output is read by the 'RFPD under test'. The DC outputs of the RFPDs are read through an oscilloscope with1 MOhm DC coupling and is averaged over 10s. The RF outputs of the reference photodiode (B) and RFPD under test (A) are connected to input ports B and A of an Agilent 4395A network analyzer. The length of these RF output cables is the same. The RF source from Agilent 4395A is split using an RF splitter with one half feeding back to R port and the other half feeding into the bias-tee connected with laser current driver. The splitting of the source wasn't necessary but is mentioned here as it was done for some other reason which is not used in the analysis. Then 25 separate runs of transfer function A/B are measured in 4395A in the range 10 kHz to 1 MHz with 801 points at 10kHz IFBW (saved as SNxxx_LF_*.txt) and 25 separate runs in range 1MHz to 100 MHz with 801 points at 30kHz IFBW (saved as SNxxx_HF_*.txt). These runs are then stitched together in the analysis to create 25 datafiles (saved as SNxxx_TF_*.txt).
From this point onwards, analysis mentioned in repo iris in file iris.py is used. I didn't explore any other methods of analyzing data and calculating transimpedance as it seemed that Craig had already worked on this and figured out a good method for doing so. I'll describe this analysis in my own words, however, the best description can be obtained from Craig himself. Each of the 25 transfer function is first converted into transimpedance with physical units. Following formula is used for doing so:

                                                                                                                                                               Z_{AC,PD} = Z_{AC,Ref} R_{PC} e^{-i \delta \phi}} T_{meas}
Here, 
ZAC,PD : Calculated RF Transimpedance
Z AC,Ref : Known RF transimpedance of reference photodiode
RPC : Photocurrent ratio at DC of reference PD to RFPD under test. This is calculated as:
                                       R_{PC} = \frac{\frac{V_{DC,Ref}}{Z_{DC,Ref}}}{\frac{V_{DC,PD}}{Z_{DC,PD}}}
\delta \phi: Extra phase delay in RFPD under test's light path due to the extra distance of light travel (\Delta L). This is calculated as:
                                       \delta \phi = \frac{\Delta L f}{c}
Tmeas : Measured transfer function A/B from Agilent 4395A

This formula can be derived easily by first taking the ratio of light powers on the two photodiodes at DC and then taking ratio at AC and using photocurrent ratio at DC to substitute for unknown values like power attenuation, the fraction of power falling on PD and ratio of responsivity of the two photodiodes. Note that this circumvents any differences due to the different focusing of light on photodiodes, losses due to mirror and beamsplitter and different responsivities of the two photodiodes. The extra path length (\Delta L) is measured using a ruler. The uncertainty in this measurement and all other measurements are carried forward and described later.
After this conversion, the median values of real and imaginary parts of the complex transimpedance is taken out of the 25 traces separately at each frequency point. These median values are used to create a complex median transimpedance representing the estimate of the analysis. Covariance of the real and imaginary parts of the 25 transfer function is used to create a covariance matrix at each frequency point. This is transformed into covariance of magnitude and phase basis. From this, the variance in magnitude and phase of the transimpedance is extracted. This variance is added with other noise sources (Uncertainties in DC voltage levels and known DC and AC transimpedances) in quadrature to get an estimate of the uncertainty of transimpedance at each frequency point.
Apart from transimpedance, dark noise measurement was also attempted. I took the spectrum of RF out when a beam dump is placed in front of the RFPD under test. The spectrum is taken from 10kHz to 100 MHz with 801 points and 10 kHz bandwidth. To estimate noise of 4395A, I took the same spectrum with input port A shorted. The two spectrums are then subtracted in quadrature to get an estimate of the dark noise. However, it is found that these measurements are in fact limited by the noise of 4395A itself. So we have decided to take dark noise measurements separately with low noise preamplifiers later. The data files of this attempt are still in the data folder with the analysis in the jupyter notebook.

Uncertainty handling:

The uncertainty in transfer function measurement is estimated using the method described above. Othe sources of uncertainties are the following:

  • RFPD DC transimpedance: These are read off from the resistor values in the actual circuit. I have assumed 2% uncertainty in each resistor value in the DC transimpedance circuit. Since transimpedance is created through 3 resistors, the uncertainty is 2\sqrt{3}% in the calculated transimpedance.
  • Reference Photodiode: Reference Photodiode's marked dc and ac transimpedance are assumed to have 1% error in them. This number came in a discussion with Gautum.
  • DC Voltage level: Uncertainty in this value was simply eye-balled during the measurement. Without any amplitude modulation, I just saw how much the 10s averaged value fluctuate and used at as the uncertainty.
  • Extra Path Length: Since extra path length of light was measured quite crudely with a ruler, I took uncertainty in this measurement as 1 cm. 

Plots and results:
In the attached folder, plots for each RFPD are there. The *TI_Four_Squares.pdf plots show median transimpedance with each TF sweep also plotted in red on left. The right 2 boxes are residuals of TF sweep from the median transimpedance calculated.
Further, for comparison, RFPD counterparts of north and south paths taking reflection measurements from the cavity and PMC are plotted together.
All the characteristic results are tabulated in CTN_Lab_All_RFPD_Characteristics.xlsx file in the data folder.

PSL All RFPD Characterisitics
SN Position Peak Freq. (MHz) TI at Peak (V/A) Notch Freq. (MHz) TI at Notch (V/A) DC TI (V/A)
SN009 North Cavity Refl 36.036 2594 +/- 191 73.109 77.8 +/- 9.2 2020 +/- 68
SN010 South Cavity Refl 36.67 1529 +/- 99 74.237 76.8 +/- 8.5 2222 +/- 75
SN020 North PMC Refl 21.692 2891 +/- 248 42.413 137.6 +/- 12.0 2222 +/- 75
SN001 South PMC Refl 15.107 14430 +/- 1256 29.402 558.0 +/ 37.2 2222 +/- 75
SN101 Beat Note 27.34 1178 +/- 77 No Notch No Notch 2222 +/- 75

Conclusions:

  • All RFPDs have roughly the intended peaks in their transimpedance and the notches are close to 2\omega value.
  • North and South Cavity Relfection PDs have an almost similar response. This is good to have equivalent paths. So I think we solved the problem which initiated this whole project.
  • North and South PMC Reflection PDs are quite different primarily dude to different frequencies at which they will be operated at. The low peak frequency in South path has resulted in a very high transimpedance at the peak. But since PMCs only affect the mode shape, this should not affect the equivalency of the two paths.

Edit Tue Oct 9 09:54:40 2018 (awade): Narrowed table width to prevent horizontal scrolling.


Edit Wed May 15 19:48:20 2019:

See https://nodus.ligo.caltech.edu:30889/ATFWiki/doku.php?id=main:experiments:psl:rfpd for the latest changes in RFPD.

Attachment 1: All_RFPD_TI_and_Dark_Noise_Measurements.pdf
All_RFPD_TI_and_Dark_Noise_Measurements.pdf All_RFPD_TI_and_Dark_Noise_Measurements.pdf All_RFPD_TI_and_Dark_Noise_Measurements.pdf All_RFPD_TI_and_Dark_Noise_Measurements.pdf All_RFPD_TI_and_Dark_Noise_Measurements.pdf All_RFPD_TI_and_Dark_Noise_Measurements.pdf All_RFPD_TI_and_Dark_Noise_Measurements.pdf
  2246   Mon Oct 8 10:22:09 2018 awadeSummaryRFSome notes on reflectionless RF filters for demodulation electronics

Edit Thu Oct 25 14:57:57 2018 (awade): Note that the LISO models here have wrong units.  They are supposed to be in units of power but were not converted from V/V.  Things listed as dB are correct but multiply the exponent of magnitude plots by two. This will be corrected in future posts.


I came across two interesting papers by Matt Morgan et al. on the subject of reflectionless RF filters.   

It seems like there wasn't a lot of thought put into optimal RF filter designs before this point that properly terminated the stop band elements. The Morgan filters are essentially diplexing filters arraignments, that has been done before. However, in the case of these reflectionless filters, there are additional symmetries that lower the inductor and capacitor values, reduce the requirements on component Q and widen the component tolerance requirements in general. The matched component design means that all circuit parts can be drawn from the same batch, improving the manufacture variation errors and also reducing the sensitivity to temperature. Together this means that we can more readly make custum filters with off-the-shelf discreet components with more confidence they will preform as designed and be relativity imune to thermal variations.

Comparisons to the equivalent Butterworth and Chebyshev designs suggest that it performs just slightly worst than Chebyshev but with a better complex gain slope, this means flatter phase in the passband and better stability of the filter for mismatch with component variation. It seems like these reflectionless filter structures are more directly comparable to the inverse Chebyshev filters or the elliptic (Cauer filter) topologies. These have minimal or no pass band ripple at the expense of slower roll off in the stop band some stronger stop band dips.  The bonus is that they are properly terminated at all frequencies and absorb all stopband by design.

The shape of the pass band response is much closer to the current elliptical filter installed in that there is a strong dip designed to coincide with the PDH modulation frequency. The dip corresponds to the first pole of the filter.  As documented previously in PSL: 2238 the stop band attenuation of the TTFSS elliptical filters is not much better than 14 dB. However, the attentuation at the critical modulation frequency is 58 dB and that is what really matters.  The problem with the current TTFSS RF demodulation design is that it does not have proper dumping the the stop band, much of the rejected ω and 2ω frequency components are reflected strait back into the mixer.  

The naive wisdom is that we add a terminating resistor at the input of the RF LP filter to quell backreflections to the mixer.  This is probably good enough for many configurations where we just don't care that much.  However, 50 Ω to ground isn't strictly impedance matched accross the whole frequency band and messes up the impedance in the passband. What we want is something that optimally dumps the higher order RF terms and presents the mixer with a well defined impedance at its output* accross the whole band: that way the mixer is loaded properly to its design specs.

Below is an schematic of a first order reflectionless filter with component values selected to achieve the pole dip at 36 MHz.

Here the optimal choice for passive component values is:

L = \frac{Z_0}{\omega_\textrm{pole}}

C = \frac{1}{Z_0\omega_\textrm{pole}}

R = Z_0

where Z_0 is the input terminating impedance of the filter.  It is also implicitly assumed that the output terminating impedance is matched to this Z_0 value.

I modeled the worst case filter design variations by performing a Monty Carlo simulation of the ideal circuit with LISO.  I Assumed a standard deviation of compoent values of 5.0% (this is what the spec sheets claim and set each component to normal random sample for each component about its ideal design value. Below is a plot of 500 samples (thin low alfa lines) along with computed median and 1σ band. In reality the component variation will be much less as they are drawn from the same batch and the manufactures probably leave a margin of error in the absolute value of components that they can deliver to.

First order reflectionless filter modeled with Monty-Carlo random error of components 5% std each component.

What the model shows is that the tolerance to component variation is probably ok.  It doesn't show the impact of component Q on filter performace (especially about the deep notch around the pole is what we are really interested in).  I'm working on getting PySpice to simulate this with the coilcraft spice models.

Here also is the input impedance of the filter as a function of frequency:

This seems less good, as can be seen from the actuall mismatch for individual samples.  The real variation of inductors and caps drawn from the same batch probably isn't this bad.

 


*Its not clear to me what the output impedance of a naked mixer is supposed to be. In the TTFSS design a 22 Ω resistor is in series with a MAX333A switching chip before the RF filter (these 333A's have ~ 30 Ω series resistance): so it seems like that design assume that you need to add resistance in serieis to get to 50 Ω termination.  

Also attached is the notebook used to compute the plots in this post using pyliso.

 

Edit Thu Feb 21 19:30:48 2019 (awade): fixing unrendered laxex

Attachment 1: Reflectionless-RF-filter-36-MHz.pdf
Reflectionless-RF-filter-36-MHz.pdf
Attachment 2: 2018-10-08_plot_MCliso_ReflLess1storderLPTF.pdf
2018-10-08_plot_MCliso_ReflLess1storderLPTF.pdf
Attachment 3: 2018-10-08_plot_MCliso_ReflLess1storderLPInputImp.pdf
2018-10-08_plot_MCliso_ReflLess1storderLPInputImp.pdf
Attachment 4: RFDemodFilters-elogversion.ipynb.zip
  2245   Fri Oct 5 12:42:29 2018 awadeDailyProgressPDPMC reflection PD DC path fix

The DC path of the north PMC resonant reflection photodetector was not showing any signal.

Some work on the North PMC resonant reflection phododetector:

  • Drop in voltage accross R4 is 0.9 V, its 22 Ω so that would be a current of 40 mA: way too much current when dark.
  • I looked inside and there was a lot of blobbed solder joints R4 and C8, C33 was also not fully contacted on the ground side. I wicked all the excess solder off and retouched all the joints that didn't look great. These modifications didn't fix the problem
  • Checked voltage supplies and DC levels in the DC path.  The first OP27 (U5) is showing 1 V on the input but 2 V on the output.  This should be a buffering stage: the op amp is busted. I removed U5 and replaced with new.  Pulled of the pin5 pad but this is not used anyway.  The dark PD doesn't show any voltage drop accross R4 anymore, this is good.
  • Modified the gain of the final DC output stage OP27 (U6).  It is now R16=1 kΩ, R11=10 Ω so gain of that stage is x101 and the total DC transimpedance gain is 2222 Ω.  
Attachment 1: NorthPMC_refl_rfpd.pdf
NorthPMC_refl_rfpd.pdf
  2244   Mon Oct 1 23:44:59 2018 ranaSummaryFSSCharacteristics of new RFPDs in FSS

Please explain in much more detail how the RF transimpedance calibration was done.

Quote:

 In analysis, 2000 Ohms of DC transimpedance is assumed for TFs.

  2243   Mon Oct 1 17:38:00 2018 awade and anchalDailyProgressFSSSouth Path misaligned and mismatched; time for PMC

After putting in the new EOM driver at 37 MHz, we have been unable to mode match into the south cavity. After beam profiling the path, we found that since we changed the EOM and changed the path, the cylindrical lens is no longer cleaning the incoming mode properly. We found ellipticity of 0.88 and 1.16 before and after the EOM. As we anyways had plans for putting in the mode cleaner on the south path as well, we have decided to do it now when the incoming path to cavity is anyways badly misaligned and mismatched.
Steps for putting in mode cleaner:
0) Profile the beam of existing marks to be able to revert back if required.
1) Remove the cylindrical lens and associated optics and profile the beam after the ISS amplitude modulation.
2) Install 14.75 MHz resonant EOM with waveplates and the mode cleaner and mode match into it.
3) Install the PMC reflection 14.75 MHz RFPD
4) Estimate and/or profile the beam from the mode cleaner and adjust the lens for a clean mode matching with the South Cavity.

  2242   Thu Sep 27 10:58:58 2018 awade and anchalDailyProgressFSSFirst FSS lock 36 MHz north path

We hooked up the 36 MHz crystal source to the TTFSS boxes and EOM driver yesterday and scanned the north cavity to see the error signal.  We got a bunch of noisy junk around each resonance rather than a clean PDH function. This post is diagnosing what the issue is.

Some setup details

The the frequency source is a preamplified OCXO crystal sources (see PSL:2235, for info and links) that outputs about +17 dBm out of the LO SMA and +26 dBm out of the EOM driver SMA. 

We attenuated the LO output with a 10 dB minicircuits attenuator (VAT-10+) to bring the power down to +7dBm into the TTFSS field box.  The TTFSS field box RF board (LIGO-D0901894) then amplifies the signal up to +17 dBm to be used in a level 17 mixer for demodulating the photodetector signal. 

The EOM drivers need about -3 dBm to produce a modulation depth of about 0.3.  The EOM output of the OXCO box was attenuated with a 9 dB attenuator (VAT-9+) followed by 20 dB attenuator (VAT-20+) to give a measured -3 dBm at 36 MHz (north).  

The new photodetector RF transfer functions are detailed in PSL:2240 and the tuned EOM drivers TF are posted in PSL:2236.   All these units are now powered from the same ±18 V power strip, eliminating a bunch of independent  bench top power supplies that were previously being used.  The benefit of this is now they all share a common ground referenced to the same central grounding point at the rack (in a tree like network) and that we are removing flaky banana plug connections everywhere in the experiment (bananas are bad PSL:1705, PSL:1759, PSL:1760, PSL:1843, 40m:505240m:5098).  Previously many elements in the FSS electronics were not grounded properly.  

Testing RF demodulation

A 37 MHz +7 dBm sine wave was injected into the TTFSS field box LO port and a +0 dBm signal was injected at 37.000010 MHz (10 Hz offset) to see if we were actually getting a signal.  We got a 656 mV peak to peak sine wave out of the mixer monitor port of the FSS box (50 Ω terminated). From the internals of the TTFSS box (see LIGO- D0901894) the ADT1-1 transformer couplers have an insertion loss of 0.31 dB @ 40 MHz and the level-17 JMS-1H has a conversion loss of order 5.90 dB (middle of its band).  The RF notched filter has a pass band insertion loss of 0.45 dB (see previous PSL post).  Before reading out at the monitor port the signal is amplified by U3 on LIGO-D040105 by 3.16 (+ 5.0 dB).  This should give a final Vpp of 0.978 V.  So we are seeing less power conversion than we expected. 

It should be noted that the LO path in the FSS board has a +10 dB amplifier followed by a ADC-10-4 coupler with an insertion loss of order 0.74 dB.  It could be that the mixer is not seeing the full +17 dBm and this is why the conversion looks a little low.

Other than this slightly lower than expected conversion this RF demodulation stage seems to be working as expected.

EOM driver saturating

The EOM driver was receiving -3 dBm of power.   However as it turns out with the x 155 gain the output measured at the RF monitor port of the EOM driver was order 440 mV, or 1.02 rad which is way too much for now.  The attenuation was adjusted so the power into the EOM driver was  -6 dBm, giving 310 mVpp at the monitor port.

A ground loop issue with the North laser controler

I plugged the RF output of the PMC reflection PD into a spectrum analyzer.  As we scan accross the resonance I see a pair of 500 kHz peaks either side of the fundamental mode.  When I lock the PMC and look at the RF out of the refcav reflection PD I see the same pair of 437 kHz sidebands around the 21.5 MHz peak.  I see a similar thing close to DC with a  447 kHz line and 877 kHz.  These are the same features mixed up to 21.5 MHz.  

When I disconnect the PZT in of the north laser controler these peaks go away.  It looks like we have some kind of ground loop issue.  

 

  2241   Wed Sep 26 17:17:18 2018 anchalNotesPDCaution about RF Cage in resonant RFPDs

The RF Cage that goes on top of the circuit in resonant RFPDs is grounded. It helps in avoiding interferance, however, if some component, most likely the tunable inductors touch this cage when they are closed, it creates parasitic inductance and/or capacitance. I observed this happening in one of our RFPDs where the resonant frequency would drop significantly as well as the peak magnitude once the cage is put on. I circumvented this problem by applying a layer of insulation tape on the inside of the RF cage cover. This helps avoiding any parasitic contact with the gnd.
Issued in Public Interest.

  2240   Wed Sep 26 16:59:17 2018 anchalSummaryFSSCharacteristics of new RFPDs in FSS

We installed new RFPDs with components s.t. they are tuned near 35.5 MHz. We tuned the inductor L1 and capacitor C40 to bring the peaks and notches to desired values. Attached are TFs and dark noise measurements of these two photodiodes. The transfer function measurements are adjusted with the transfer function of the cables up to the photodiodes. The photodiodes were blocked with a beam dump during measurements and are supplied with +-18 V through the new power strip under the table. In the analysis, 2000 Ohms of DC transimpedance is assumed for TFs. Note: This was a wrong assumption. So do not trust the y-axis. This is a transfer function measurement through test-in ports which could be wrong.
Following are some measured numbers:

South Path:
SN 010
RFPD tuned to 37 MHz.
Notes:
Inductor L1 changed to 146-06J08SL
+5V regulator LM309H replaced with new one.
Measured Peak at 36.882 MHz
Measured Notch at 74.237 MHz

North Path:
SN 009
RFPD tuned to 36 MHz.
Measured Peak at 36.247 MHz
Measured Notch at 72.356 MHz

Conclusions:
Except for South RFPD being low Q (as it was written on it). the two RFPDs have similar response and same amount of dark noise. This should reduce the contributions from south path in the excess noise measured in the last BN measurement.

Attachment 1: NorthandSouth_35500kHz_RFPD_Measurements.pdf
NorthandSouth_35500kHz_RFPD_Measurements.pdf NorthandSouth_35500kHz_RFPD_Measurements.pdf
  2239   Mon Sep 24 12:53:26 2018 awade and anchalSummaryPMCSummarization of efforts to fix PMC Servo problems: external mixer and LP filter

[awade and anchal]

ANCHAL ADD UPDATED TF PLOTS AND PART NUMBERS

For the PMC electronics: this flattening of the TF above 400 Hz was because the on board LP filtering in the RF demodulation stage (LP filter after the mixer) was letting too much RF through in the pass band.  This is a design floor of this particular board that a 1/f^2 filter (made with discrete components) was used instead of a all-in-one 1/f^4 chip.  With too much RF going into the LT1028 there is some kind of saturation problem which causes a leveling off above 400 Hz when taking the TF.

To fix this we put an external minicircuits mixer (model?) followed by a 1/f^4 minicircuits (model?) filter with a corner of 1.9 MHz -- with 50 Ω termination -- into the FP1test point (J3).  At the modulation frequency of 21.5 MHz this should give an attenuation of order -42 dB, much better than before.

[INSERT OLTF]

With the demodulation performed with eternal components the open loop transfer function behaved as expected with the 10 Hz pole setup by the output resistor and PZT working all the way out to very high frequency.  As a result the loop is now stable and working as expected.

 

Quote:

I and Andrew measured transfer functions of each stage of PMC Servo box using Agilent 4395A and HP 41800A probe. Attached are the measured results overplotted with LISO analysis. It seems the PMC Servo box is functioning as it is designed for. Note that in the plots, at higher frequencies there are flat or increasing magnitude regions. These are just because of measurement reaching noise floor. We check independently with sweeps at increased power for higher frequencies and the curves match LISO analysis more or less.
Next, we connected the PZT to the servo box and measured the transfer function with PZT. Again this came out as expected with a good one pole low pass filter behavior as the circuit is designed for. The transfer function (5th plot in the file) keeps going small with a linear slope well up to 10 MHz (measured till this point). So everything checks out fine with he PMC servo box.
But as we lock the servo and measure the closed loop transfer function of the system, we see a strange flat region above 400Hz. Even after the low pass filter with PZT, in a closed loop, it looks like something is actively canceling the pole above 400Hz. But there is nothing else active in the loop. The 6th plot is the measured open loop transfer function (calculated from closed loop transfer function). Note that this flatness is not noise floor or saturation of any kind. We verified this by changing source power and the frequency response at output changes proportionately to this source power change (eg if source power is changed from 5mV to 50mV, the frequency response at the output at 500Hz changes from 4.5mVpkpk to 45mVpkpk), maintaining the flat behavior above 400Hz.
The open loop transfer(G)  function was calculated from closed loop transfer function (C) using this formula:
                                                G = 1 + (1.216)/C
where 1.216 is the measured flat response of the buffer in between the test input (FP2Test) and test output (FP3Test).
So we would like if someone can give some suggestions with this.
Addition: Attaching up-to-date schematic

Quote:

This circuit just doesn't do what it says it should do.  Need to inject waveform at FP1test and probe at each stage.  Then compare against LISO model.  If something is busted we need to know at which point we are getting this extra zero in the response. There is a lot not great about this particular board but it should just have a flat response above 488 Hz.

I don't think AD602 is in the LISO library.  Should be able to add it as some kind of hack op amp with fixed gain, 100nV/rtHz of noise (with 10 Hz corner) and some appropriate current noise with corner of 1 kHz.  Maybe check the AD602 datasheet.

We need to clean this up this week or do something drastic like replacing electronics with minicircuits or removing MC altogether.  For now we need to move onto solving bigger problems like the ISS, thermal stability, PLL readout noise 

Also, on the ISS do you now have a prototype working servo circuit and photodetector?

Quote:

We earlier found that the intended LPF isn't working so we thought of this external LPF idea. So I checked today the PZT input with LCR meter to see if it is in good condition or not. It gave C = 406.4nF, L=60.6 mH and R = 31kOhm. From the spec sheet, the C value looks 20% below the rated value but the spec value has uncertainity of +- 15%, so maybe our PZT is still good.

With the measured values, I calculated again (fitted using LISO) what good value of output resistor would make it closest to 10Hz pole. The value came out to be 37.1k Ohm. I have replaced this output resistor with 39k Ohm now. I'm attaching updated schematic for future reference.

keywords for search: PMC North Driver Board Schematic

Quote:

should probably put back the 30k resistor. It makes a low pass filter with the PZT capacitance (not the cable capacitance). This is to prevent shorting of the HV drive.

 

 

 

 

  2238   Mon Sep 24 12:49:20 2018 awadeDailyProgressFSSModifying TTFSS box LP elliptical filters

After we improved the performance of the PMC locking -- that was degraded because of excessive RF getting past the demodulation stage, see PSL:2228) -- I though I'd have a look at the TTFSS boxes.  It turns out that the FSS has a elliptical LP with a notch tuned to 21 MHz.  This had not been modified since the box was made, meaning that for the period of 14.75 MHz modulation it was sub-optimal.

Previous state of TTFSS box's RF LP filtering

I modeled the TF from the output of the mixer (U2 in D0901894-v1)  to the input of the servo (U3 in LIGO-D040105 rev C).  The model is compared below with the measured TF injected from Test1 to TP1 on the servo board.  As plotted below, for both north and south paths, it is in ok agreement to within the tolerances of the components. Note that one discrepancy with how I measured the TF is that I injected into Test1 without 22 Ω in series.  This doesn't seem to make a big difference.  The bump at 50 MHz in the LISO model is an artifact of the AD829 but doesn't seem to turn up in the measured TF.

 

Modifying the Elliptical filter

The original circuit is illustrated below:

Schematic modeling RF LP filter in TTFSS box (modeling the virtual ground of the
ad829 op amp as ground)

Here a basic LP filter is formed between the C13 and C14 and L3.  The combination of C12 and L3 form the notch.

For the best LP filter at the modulation frequency we want to lower the first (lowest) poles of the filter to as low as possible without lowering the required capacitance of C12 so low that parasitic capacitance is a concern for the notch.  I've attached a notebook below, this includes some pyliso and widget sliders that where values can be varied to find the best combination. Also plotted below is a contour plot of the notch frequency as a function of the inductor (L3) and the notching capacitor in parallel (C12), this is also in the notebook.

Notch frequency as a function of inductor and capacitor choices.  Green point marks the closest whole values that give 36 MHz, the blue box is the ±5% uncertainity in the values for both of these components. Thermal drift is so small that box for this error is too small to display.

For 36 MHz a good combination looks like L3 = 750 nH, C13 = C14 = 220 pF, C12 = 26 pF (green point labeled in above plot).  Absolute tolerance of these components are 5% with a thermal drift of 30 ppm/K for the ceramic capacitors and +25 to 125 ppm/K for the ceramic core inductor (see see cap spec sheet and coil craft ceramic core spec sheet for 1206cs series). A box in the above plot shows the ±5% error level on components showing the range of potential error in the notch frequency due to manufacture variations.

The South TTFSS RF board was modified with L3 changed from 1200 nF to 750 nF and C12 being changed from 47 pF to a 18 + 1.5 + 1.5 + 1.0 pF = 22.0 pF stack (i.e. parallel).  These values were initially guessed and then trial and error was used to swap out small value caps until the notch was centered on 36 MHz. The bottom of the dip was an attenuation of -58 dB (plotted below)

The North TTFSS RF board was modified with L3 changed from 1200 nF to 750 nF and C12 being changed from 47 pF to a 18 + 3.9 + 1.5 + 1.0 pF = 24.4 pF in parallel. This gave a notch at 37 MHz with a dip that was also -58 dB. The measured transfer function is plotted below.

This is a high(ish) Q narrow peak so we don't want it to drift too much. As mentioned above the thermal drift is 30 ppm/K for caps and 125 ppm/K for the inductor (worst case).  The LISO model indicates that this corresponds to a drift of 0.5 kHz/K for the cap and 2.2 kHz/K for the inductor: this is a tolerable margin of error. Thermal drift, even for large 10's K variations in box temperature, is not a concern.

The LISO model indicated that we could squeeze some more attenuation by shifting C13 and C14 from 220 pF to a larger value.  But when I tried 1 nF (for both) I found that there was some weird 1/f roll up below 200 kHz when I measured the TF (rather than flat). I got the same even when I tried 440 pF.   Maybe its some impedance matching thing and the resistors need to be modified; I wanted to get the FSS back up and working so I didn't look further into this. C13 and C14 were left at 220 pF for both North and South TTFSS boxes.

It also looks like the real measured TF have a broader notch than the LISO model.  There is probably some stuff that isn't modeled in the simple LISO model.  This relaxes the requirments for the notch tuning a little.

TTFSS field boxes were reinstalled on the table, ready for the demodulation phase to be tuned for the new frequencies.

---

Data is commited into https://git.ligo.org/cit-ctnlab/ctn_labdata/tree/master/data/20180925_FSS_EllipticFilterRetuned_TF and zipped and attached below.

Attachment 1: 20180921_plot_TFNorthTFFSSRFFilter.pdf
20180921_plot_TFNorthTFFSSRFFilter.pdf
Attachment 2: 20180921_plot_TFSouthTFFSSRFFilter.pdf
20180921_plot_TFSouthTFFSSRFFilter.pdf
Attachment 3: TTFSS-EllipticalLPFilter.pdf
TTFSS-EllipticalLPFilter.pdf
Attachment 4: 20180921_plot_EllipicalLPFilter_NotchFreqContour.pdf
20180921_plot_EllipicalLPFilter_NotchFreqContour.pdf
Attachment 5: 20180921_plot_SouthModifiedSouthRFLPEllipticModifiedSep.pdf
20180921_plot_SouthModifiedSouthRFLPEllipticModifiedSep.pdf
Attachment 6: 20180921_plot_NorthModifiedSouthRFLPEllipticModifiedSep.pdf
20180921_plot_NorthModifiedSouthRFLPEllipticModifiedSep.pdf
Attachment 7: 20180921_FSS_EllipticFilterTF.zip
  2237   Fri Sep 21 18:10:53 2018 ranaDailyProgressBEATBack to beat

I don't understand what that means. Please provide 10x more details on how the measurement was made.

Also, clearly one of these traces is not like the other. What does that mean ???

Quote:

Attached are transfer function measurements of the North and South Cavity Reflection RFPD (14.75MHz resonant RFPD) along with dark noise around 14.75 MHz.

Edit:[09/14/2018, 16:12] Changed plots to physical units. Used 2k Transimpedance for Bode Plot and 2.5kHz bandwidth (801 points in 2MHz) for noise plots.

 

  2236   Fri Sep 21 11:24:42 2018 anchal and awadeDailyProgressFSSEOM Drivers TF

We have replaced the resonant 14.75MHz EOM from south path with a broadband EOM (Newport 4004). We soldered two D1200794-v3 EOM driver boards and tuned them to 36 Mhz and 37 MHz to be used with the new crystal oscillators. Tuning these driver circuits is bit challenging. First of all, this driver board needs some modification to have footprints matching footprints with available inductors and capacitors. It would be better to have components in 1206 footprint as all of our other boards use this size. We used 143-15J12SL (Green) Shielded coilcraft inductor for tuning this circuit. We also replaced C6 with 1.2pF as 0.5pF wasn't available. But it didn't affect the tuning range drastically.Secondly, it would be nice if the board is designed to fit inside a compact metal box for RF isolation.

Attached are measured transfer functions of these driver circuits with EOMs attached. The transfer function is monitored from RF_mon port which is supposed to have coupling ratio of 155 as per the dcc document.

Attachment 1: North_And_south_EOM_Driver_TF_Measurements.pdf
North_And_south_EOM_Driver_TF_Measurements.pdf
  2235   Wed Sep 19 20:11:24 2018 awade, anchalDailyProgressFSSInstalling OCXO preamp 36 MHz + 37 MHz

[awade, anchal]

TL:DR The OCXO preamp seemed to have a very high current draw, looking at the specs of the components we should expect 1.08 A during warm up and 0.74 A thereafter.

It is working as expected.

---

Anchal installed a new 18VDC power strip under the table for suppling various electronics from a single (common grounded source).  I plugged the OCXO preamp (LIGO-D1600008), for 36 MHz and 37 MHz RF generation into the strip and found that the 18V supply hit its current limit.

Components inside box are pictured below.  The Wenzel crystal oscillators have a 5 W draw in the first 5 minutes as they heat up (for their 15 V supply).  In total the initial current draw was 1.1 A between the pair of independent frequency sources.  This load also included voltage regulators and amplifier units which should be about 200 mW per channel. All up this is a requirement of 1.08 A during warm up and 0.74 A thereafter.

After 5 minutes in the current draw had dropped back to 0.8 A. So this was about right. I bumped up the current limit on the Sorensen supplies.

Attachment 1: IMG_4220.JPG
IMG_4220.JPG
  2234   Mon Sep 17 14:08:56 2018 awadeMiscComputersWS1 CPU fan replaced

Found that WS1 had died again.  CPU fan seized up again.

Replaced and computer restarted fine. WS1 up and running normally.

Quote:

WS1, the main computer in the PSL lab died last night.  On reboot bios screen says that CPU fan died.  

I opened the computer up and had a look.  Bearings on main CPU fan were a bit stiff.  I wiggled them a bit and it now spins, with some noise, when booted.  

I'll order a replacement CPU fan KDE1209PTVX 12V, 7.0 W and replace pronto.

Main CPU fan
Main CPU fan, bearings nearing end of life.

---

Also there was no backup of this computer.  Almost all the stuff is kept in Git version control, but we should get these computers back on scripted backups.

 

---

Edit Sat Sep 8 18:28:19 2018: Rebooted and restarted frequency counter. Beat note has drifted off to 750 MHz.  Will take a while to bring it back in.

Fan has been reordered, will arive Friday next week.

 

  2233   Fri Sep 14 00:57:14 2018 ranaDailyProgressBEATBack to beat

please replace TF and noise plots with ones that have physical units on the y-axis: V/A for the Bode plot and W/rtHz for the noise plot

  2232   Wed Sep 12 16:27:19 2018 anchalDailyProgressBEATBack to beat

Attached are transfer function measurements of the North and South Cavity Reflection RFPD (14.75MHz resonant RFPD) along with dark noise around 14.75 MHz.

The transfer functions are measured by injecting into the test in port and reading out from RF port at -15dBm source power. The noise spectra are measured by shorting the test in port and taking spectrum from RF port when the detectors are on. In both measurements, the photodiode is blocked with a beam dump.
These measurements were required because of the conclusions made in PSL:2230. Indeed as suspected, the south path resonant RFPD measuring reflection of the cavity at 14.75 MHz has a ~100 times weaker response than the north counterpart as seen in the attached plots. Since the dark noise of south RFPD is about half of the noise of north RFPD (see plot 2), it suggests that south RFPD circuit itself is not working properly and is not amplifying the signal enough. Andrew mentioned that he and Craig saw this earlier and decided to shift FSS to higher frequencies with crystal oscillators. We have the OCOX preamp for 36 MHz and 37 Mhz ready to go with RFPDs at 35.5 MHz that can be tuned to these frequencies. So future steps are to replace the RFPDs with these 35.5MHz ones and tuning them to 36 MHz and 37 MHz and putting in broadband EOM driven by resonant EOM drivers at these frequencies. See future posts for updates on these steps.

Edit:[09/14/2018, 16:12] Changed plots to physical units. Used 2k Transimpedance for Bode Plot and 2.5kHz bandwidth (801 points in 2MHz) for noise plots.

Edit:[09/22/2018, 10:12] Added how measurements were taken, the reason for them and some conclusions. I'm getting into the third year now!

Attachment 1: NorthandSouth_14750kHz_RFPD_Measurements.pdf
NorthandSouth_14750kHz_RFPD_Measurements.pdf NorthandSouth_14750kHz_RFPD_Measurements.pdf
  2231   Sat Sep 8 17:34:54 2018 awadeMiscComputersWS1 CPU fan dead

WS1, the main computer in the PSL lab died last night.  On reboot bios screen says that CPU fan died.  

I opened the computer up and had a look.  Bearings on main CPU fan were a bit stiff.  I wiggled them a bit and it now spins, with some noise, when booted.  

I'll order a replacement CPU fan KDE1209PTVX 12V, 7.0 W and replace pronto.

Main CPU fan
Main CPU fan, bearings nearing end of life.

---

Also there was no backup of this computer.  Almost all the stuff is kept in Git version control, but we should get these computers back on scripted backups.

 

---

Edit Sat Sep 8 18:28:19 2018: Rebooted and restarted frequency counter. Beat note has drifted off to 750 MHz.  Will take a while to bring it back in.

Fan has been reordered, will arive Friday next week.

  2230   Fri Sep 7 23:23:00 2018 awade, anchalDailyProgressBEATBack to beat

[awade, anchal​]

Today we go the beat to converge on the 26 MHz resonance of the resonant transmission beat detector.  Anchal and I took a TF of the PLL and found the UGF to be 40 kHz, sufficient for a BN spectrum.  Beat not power was about -1 dBm.  

Sorry no beat spectrum, we lost lock with the south path autolocker turned off and the cavity heat PID didn't realize and knocked the BN way off.  Going to take another 12 hours for it to settle down again.

It looks like the spectrum was about 0.1 Hz at its lowest point (>3 kHz).  We saw that there was a very prominent narrow peak at about 2.8 kHz.  We had a look at the mixer monitor signals and saw that the narrow peak was coming from the south path FSS.  After doing a cross spectrum between the beat and the north and south FSS it became apparent that a significant portion of our noise is coming from the south FSS somehow.  We need to track down exactly what would be coupling in a 2.8 kHz narrow peak and also the broader noise of the south FSS.  

I suspect​ it might be a photodetector​ issue.  We had excess noise is one path relative to the other before and swapped detectors and field boxes.  We never quite diagnosed it but it could well be that the 14.75 MHz tuned detector that is now in the south path is responsible.  The first thing to check is the TF by injecting into the test port.  Also the dark noise should be checked around 14.75 MHz.  The optical TF can also be taken with the Jenny rig at the 40m.  Now might be a good time to switch modulation frequencies to 36 MHz and 37 MHz.  These are fine for HOM spacing and we have the Wenzel crystals ready to go.  We also have two detectors tuned to 35.5 MHz sitting on the shelf and BB EOM drivers that just need stuffing onto boards.  This might take Anchal a week to do, he seems to be good with electronics.

Another unresolved problem is the residual AM from the 14.75 MHz phase modulators.  I was never really able to reduce this down and keep it down.  Thermal or alignment drift seemed to make it really hard to minimize.  It could be bad alignment though the crystal or thermal drift.  They have insulating hats on but they still are less than optimal.  An ISS will do a bit to suppress noise opened up by this effect but we would like to solve it properly.

  2229   Thu Aug 30 12:01:37 2018 ranaSummaryPMCSummarization of efforts to fix PMC Servo problems

post up-to-date schematic

Update Thu Aug 30 15:13:48 2018: now added to elog 2228

  2228   Wed Aug 29 12:01:54 2018 anchalSummaryPMCSummarization of efforts to fix PMC Servo problems

I and Andrew measured transfer functions of each stage of PMC Servo box using Agilent 4395A and HP 41800A probe. Attached are the measured results overplotted with LISO analysis. It seems the PMC Servo box is functioning as it is designed for. Note that in the plots, at higher frequencies there are flat or increasing magnitude regions. These are just because of measurement reaching noise floor. We check independently with sweeps at increased power for higher frequencies and the curves match LISO analysis more or less.
Next, we connected the PZT to the servo box and measured the transfer function with PZT. Again this came out as expected with a good one pole low pass filter behavior as the circuit is designed for. The transfer function (5th plot in the file) keeps going small with a linear slope well up to 10 MHz (measured till this point). So everything checks out fine with he PMC servo box.
But as we lock the servo and measure the closed loop transfer function of the system, we see a strange flat region above 400Hz. Even after the low pass filter with PZT, in a closed loop, it looks like something is actively canceling the pole above 400Hz. But there is nothing else active in the loop. The 6th plot is the measured open loop transfer function (calculated from closed loop transfer function). Note that this flatness is not noise floor or saturation of any kind. We verified this by changing source power and the frequency response at output changes proportionately to this source power change (eg if source power is changed from 5mV to 50mV, the frequency response at the output at 500Hz changes from 4.5mVpkpk to 45mVpkpk), maintaining the flat behavior above 400Hz.
The open loop transfer(G)  function was calculated from closed loop transfer function (C) using this formula:
                                                G = 1 + (1.216)/C
where 1.216 is the measured flat response of the buffer in between the test input (FP2Test) and test output (FP3Test).
So we would like if someone can give some suggestions with this.
Addition: Attaching up-to-date schematic

Quote:

This circuit just doesn't do what it says it should do.  Need to inject waveform at FP1test and probe at each stage.  Then compare against LISO model.  If something is busted we need to know at which point we are getting this extra zero in the response. There is a lot not great about this particular board but it should just have a flat response above 488 Hz.

I don't think AD602 is in the LISO library.  Should be able to add it as some kind of hack op amp with fixed gain, 100nV/rtHz of noise (with 10 Hz corner) and some appropriate current noise with corner of 1 kHz.  Maybe check the AD602 datasheet.

We need to clean this up this week or do something drastic like replacing electronics with minicircuits or removing MC altogether.  For now we need to move onto solving bigger problems like the ISS, thermal stability, PLL readout noise 

Also, on the ISS do you now have a prototype working servo circuit and photodetector?

Quote:

We earlier found that the intended LPF isn't working so we thought of this external LPF idea. So I checked today the PZT input with LCR meter to see if it is in good condition or not. It gave C = 406.4nF, L=60.6 mH and R = 31kOhm. From the spec sheet, the C value looks 20% below the rated value but the spec value has uncertainity of +- 15%, so maybe our PZT is still good.

With the measured values, I calculated again (fitted using LISO) what good value of output resistor would make it closest to 10Hz pole. The value came out to be 37.1k Ohm. I have replaced this output resistor with 39k Ohm now. I'm attaching updated schematic for future reference.

keywords for search: PMC North Driver Board Schematic

Quote:

should probably put back the 30k resistor. It makes a low pass filter with the PZT capacitance (not the cable capacitance). This is to prevent shorting of the HV drive.

 

 

 

Attachment 1: D980352-AGupta20180712Mods.pdf
D980352-AGupta20180712Mods.pdf
Attachment 2: PMC_Servo_LISO_Analysis_Vs_Measured.pdf
PMC_Servo_LISO_Analysis_Vs_Measured.pdf PMC_Servo_LISO_Analysis_Vs_Measured.pdf PMC_Servo_LISO_Analysis_Vs_Measured.pdf PMC_Servo_LISO_Analysis_Vs_Measured.pdf PMC_Servo_LISO_Analysis_Vs_Measured.pdf PMC_Servo_LISO_Analysis_Vs_Measured.pdf
  2227   Mon Aug 13 17:41:39 2018 anchalSummaryISSDCC link of finished design

The board design is finished and all the files have been uploaded on dcc (LIGO-D1800214-v1). PCB Layout, circuit schematic, Gerber files, LISO analysis, front panel and full BOM are attached. There are two copies of ISS circuits on the single board and they will be mounted on rack with a 2U front panel.

Notes:
1) The Noise performance analysis is present in the LISO_and_LTSpice_Files folder.
2) The present values in stage 4 of the circuit has cavity pole neutralization for a pole frequency of ~40kHz. To make this different, change the capacitor C22 and C53 accordingly.
3) The resistors R9 and R14 in Stage 0 can be populated to give an overall gain to the circuit.

  2226   Mon Aug 13 17:36:39 2018 anchalSummaryPDPhotodiode Transimpedance Amplification Circuit

Redesigned D1400384-v2 is ready and uploaded on dcc. PCB Layout, circuit schematic, Gerber files, LISO analysis, front panel and full BOM are attached.
The circuit is kept the same except for replacing LT1128 with OPA827 in the transimpedance amplifier stages and adding few OMIT and ZERO resistances to have more options for using the circuit. Following are the notes (also present as ReadMe.txt in the .zip file in dcc):

Notes:
1) To not have whitening option, do not populate IC5, U8, U9 and their peripheral resistors and capacitors and put a 0 Ohm resistor (Jumper) at R20 and R26

2) To have fixed transimpedance, do not populate IC8 and its peripheral resistors and capacitors and populate R7 and R22 with the desired transimpedance.

3) In case of switchable transimpedance, R11=R15=100 and R14=R28=10k are the two options in present design. These can also be changed according to choice. Mark the front panel accordingly with the choices.

4) For low transimpedances, it is better to replace U5 and U7 with LT1128 instead.

  2225   Mon Aug 6 15:02:57 2018 anchalDailyProgressISSFinal Design for ISS Servo v3.2

Yes I realized this later. AD743, unfortunately, was the best choice for us. Its replacement ADA4627 from Analog Devices is worse than OPA827 which is the next best choice according to the LISO simulation. So we'll go ahead with OPA827 in the Stage 1. I'll put the final schematic with board design soon.

Quote:

AD743 is discontinued.  We have a few in stock, and there are still a few SOIC-16 version that you can buy. 

But better to design with a op amp that is replaceable into the future.

Quote:

I checked with some replacements of opamp at stage 1 to reduce the noise due to current noise. It seems AD743 is the best balanced opamp for this location. Attached is the circuit schematic of this final version v3.2 , with LISO analysis and also the analysis done to reach to this opamp.

 

 

  2224   Sat Aug 4 21:37:55 2018 johannesDailyProgressISSFinal Design for ISS Servo v3.2

These are some "low-noise" FET input OpAmps which are all still alive:

  • MAX4475
  • AD8655
  • LT1792

not exact substitutes for AD743 but lower noise of all alternatives that I've come across so far.

 

Quote:

AD743 is discontinued.  We have a few in stock, and there are still a few SOIC-16 version that you can buy. 

But better to design with a op amp that is replaceable into the future.

Quote:

I checked with some replacements of opamp at stage 1 to reduce the noise due to current noise. It seems AD743 is the best balanced opamp for this location. Attached is the circuit schematic of this final version v3.2 , with LISO analysis and also the analysis done to reach to this opamp.

 

 

  2223   Wed Aug 1 17:31:19 2018 awadeDailyProgressISSFinal Design for ISS Servo v3.2

AD743 is discontinued.  We have a few in stock, and there are still a few SOIC-16 version that you can buy. 

But better to design with a op amp that is replaceable into the future.

Quote:

I checked with some replacements of opamp at stage 1 to reduce the noise due to current noise. It seems AD743 is the best balanced opamp for this location. Attached is the circuit schematic of this final version v3.2 , with LISO analysis and also the analysis done to reach to this opamp.

 

  2222   Wed Aug 1 11:16:22 2018 anchalDailyProgressISSFinal Design for ISS Servo v3.2

I checked with some replacements of opamp at stage 1 to reduce the noise due to current noise. It seems AD743 is the best balanced opamp for this location. Attached is the circuit schematic of this final version v3.2 , with LISO analysis and also the analysis done to reach to this opamp.

Attachment 1: Cryo_ISS3_2schematic.pdf
Cryo_ISS3_2schematic.pdf
Attachment 2: Cryo_ISS3_2_LISO_Analysis.pdf
Cryo_ISS3_2_LISO_Analysis.pdf Cryo_ISS3_2_LISO_Analysis.pdf
Attachment 3: Cryo_ISS3_1_ISS1_Optimization.pdf
Cryo_ISS3_1_ISS1_Optimization.pdf Cryo_ISS3_1_ISS1_Optimization.pdf
  2221   Mon Jul 30 18:58:27 2018 anchalDailyProgressISSMinor version improvement in design.

It seems difficult to implement our AC coupled circuit in D1300694_v1 board which is DC coupled.

Meanwhile, I was looking into improving noise performance of present design ISS3, and it seems we can get some improvement directly by including a buffer at the input (to have good input impedance) and reduce resistance values in stage 1 to lower the impact of the current noise of first stage amplifier. This design ISS3_1 and its LISO analysis are attached.

Attachment 1: Cryo_ISS3_1schematic.pdf
Cryo_ISS3_1schematic.pdf
Attachment 2: Cryo_ISS3_1_LISO_Analysis.pdf
Cryo_ISS3_1_LISO_Analysis.pdf
  2220   Mon Jul 30 16:54:50 2018 anchalDailyProgressISSUnderstanding the ISS box we found

I was given an ISS box which was made in 2013. It is numbered D1300694-v1 and the schematics are present in DCC. I tried to go through the schematic to get an understanding of it and mark the differences of the real circuit on the schematic. PFA a commented schematic which marks the measured component values and shows what is not present on this circuit.

Breif Description of this circuit and what it intended to do:
The circuit is DC coupled and the PD signal first goes through an instrumentation amplifier with about 1.33 gain. Then a 5V reference is subtracted from this signal (so the servo do DC stabalization to keep PD signal at 5V).
Then the signal goes through 3 stages of filtering and amplification.
Stage 1: (Always On) Is a simple gain of 6.36
Stage 2: Is a 554 Hz Zero and 10kHz pole with DC gain of 18 (Called Boost 1)
Stage 3: Is a 0 Hz Pole and 4.57MHz zero with the open loop gain of AD829 at DC. (Called Boost 2)
The Boost 1 and 2 are supposed to start only when rms value of PD signal is below a certain threshold. Currently, this functionality is not working due to a problem in RMS-to-DC converter as specified in elog 40m/9332.

I will see if I can find a way to hack this circuit to do what we want to do without making any irreversible changes.

Attachment 1: D1300694-v1_Commented_AG.pdf
D1300694-v1_Commented_AG.pdf D1300694-v1_Commented_AG.pdf D1300694-v1_Commented_AG.pdf D1300694-v1_Commented_AG.pdf D1300694-v1_Commented_AG.pdf D1300694-v1_Commented_AG.pdf
  2219   Mon Jul 30 13:28:37 2018 anchalDailyProgressISSNoise analysis of soldered circuit

Just replotting the measured noise with LISO prediction for comparison.

Attachment 1: ISS_Noise_Analysis_Meas_vs_LISO.pdf
ISS_Noise_Analysis_Meas_vs_LISO.pdf
  2218   Fri Jul 27 12:29:43 2018 ranaDailyProgressISSNoise analysis of soldered circuit

for any circuit where power noise is measured, the units should be input referred current noise, not voltage noise. So you have to include a mode of the photodetector circuit.

  2217   Thu Jul 26 18:30:51 2018 anchalDailyProgressISSNoise analysis of soldered circuit

I soldered the servo circuit on a prototype board. The transfer function of the soldered board is matching very well with the analytical calculation. We found a few more corrections required in the circuit relating to the voltage offset compensation resistors. Then I did a complete noise analysis of the circuit using SR785. Attached are the results. First I measured the transfer function of each individual stage by sourcing the input of the circuit and taking the ratio of the output to the input of each stage separately. Then, I used 2 different methods to calculate input referred noise at different stages of the circuit (which are soldered together).
Method 1: I connected a 50 Ohm terminator at the input of the stage and measured output noise of the stage. Then I divided it by the transfer function of the stage.
Method 2: I only measured output noise at the output of each stage. And calculated the input referred noise by dividing the output noise by transfer function and subtracting the excess noise from the previous stage at its input.
In the process, I have written a module which can be used by others too to fit transfer function data and calculate input referred noise in general using SR785. I would put this code on github soon.
It seems that as the LISO analysis shows, most of the noise below 20-30 Hz is due to the amplification of input current noise of the first stage amplifier. Next steps are to figure out a better opamp if any to reduce noise in this region. We are well below the shot noise for 2mW 1550 nm laser above 30 Hz.

Attachment 1: ISS_Noise_Analysis_Method1.pdf
ISS_Noise_Analysis_Method1.pdf ISS_Noise_Analysis_Method1.pdf ISS_Noise_Analysis_Method1.pdf ISS_Noise_Analysis_Method1.pdf ISS_Noise_Analysis_Method1.pdf ISS_Noise_Analysis_Method1.pdf ISS_Noise_Analysis_Method1.pdf ISS_Noise_Analysis_Method1.pdf
Attachment 2: ISS_Noise_Analysis_Method2.pdf
ISS_Noise_Analysis_Method2.pdf ISS_Noise_Analysis_Method2.pdf ISS_Noise_Analysis_Method2.pdf ISS_Noise_Analysis_Method2.pdf ISS_Noise_Analysis_Method2.pdf ISS_Noise_Analysis_Method2.pdf ISS_Noise_Analysis_Method2.pdf ISS_Noise_Analysis_Method2.pdf
  2216   Tue Jul 24 14:28:08 2018 johannesDailyProgressISSIntroduced zero to cancel cavity pole. Mostly final circuit.

Modified the design to make stage three passive again and moved stage 5 (addition of offset) to stage 4. PFA the circuit schematic.

Also attached are LISO analysis of the circuit.

The first file is the transfer function of the feedback circuit alone with its input impedance.

The second file is the open loop transfer function of the feedback circuit with AOM and cavity, the input referred noise and the opamps' stability in closed loop operation.

P.S. The unstable region shown in figure one of the second file is calculated as the region where |KP + 1|<1 as per the phase of calculated KP, where KP is the open loop gain function.

Quote:

By making stage 3 active you may be adding a lot of OpAmp noise, since you're amplifying it by a factor of 100 at low frequencies. It sits behind the first two gain stages so it might not matter for the input referred noise? But I don't think it's good practice. What was wrong with having that stage passive? A gain stage like this, which doesn't provide gains higher than 1 and actually attenuates over a broad range is usually better kept passive because of amplifier noise.

Did you check the circuit for stability? The AD829 in stage 3 is driving a pretty low resistance as load. Not sure if LTSpice can identify such problems.

 

Attachment 1: Cryo_ISS3_AOMCavity_Schematic.pdf
Cryo_ISS3_AOMCavity_Schematic.pdf
Attachment 2: Cryo_ISS3_LISO_Analysis.pdf
Cryo_ISS3_LISO_Analysis.pdf Cryo_ISS3_LISO_Analysis.pdf
Attachment 3: CRYO_ISS3AOMCavityLisoAnalysis.pdf
CRYO_ISS3AOMCavityLisoAnalysis.pdf CRYO_ISS3AOMCavityLisoAnalysis.pdf CRYO_ISS3AOMCavityLisoAnalysis.pdf
  2215   Tue Jul 24 13:47:48 2018 anchalDailyProgressISSIntroduced zero to cancel cavity pole. Mostly final circuit.

Here is LISO analysis on the version 2 of this circuit. I'll take Johannes' advice and remove the active filter on stage 3 and move the addition to stage 4 in the next version.

The first file is the transfer function and the input impedance of just the feedback circuit (the 5 stages).

The second file is open loop transfer function of feedback circuit with modeled AOM and Cavity, the input referred noise and the opamps'stability in the closed loop.

Attachment 1: Cryo_ISS2_LISO_Analysis.pdf
Cryo_ISS2_LISO_Analysis.pdf Cryo_ISS2_LISO_Analysis.pdf
Attachment 2: CRYO_ISS2AOMCavityLisoAnalysis.pdf
CRYO_ISS2AOMCavityLisoAnalysis.pdf CRYO_ISS2AOMCavityLisoAnalysis.pdf CRYO_ISS2AOMCavityLisoAnalysis.pdf
  2214   Tue Jul 24 11:04:37 2018 johannesDailyProgressISSIntroduced zero to cancel cavity pole. Mostly final circuit.

By making stage 3 active you may be adding a lot of OpAmp noise, since you're amplifying it by a factor of 100 at low frequencies. It sits behind the first two gain stages so it might not matter for the input referred noise? But I don't think it's good practice. What was wrong with having that stage passive? A gain stage like this, which doesn't provide gains higher than 1 and actually attenuates over a broad range is usually better kept passive because of amplifier noise.

Did you check the circuit for stability? The AD829 in stage 3 is driving a pretty low resistance as load. Not sure if LTSpice can identify such problems.

  2213   Mon Jul 23 18:07:14 2018 anchalDailyProgressISSIntroduced zero to cancel cavity pole. Mostly final circuit.

After discussion with Johannes, I added a zero near the cavity pole to neutralize its effect at higher frequencies. I also replaced the passive high frequency boost with an active one in stage 3. Attached is the new circuit schematic (Mostly final one as it looks good). Also attached are open loop transfer function from LTSpice.
Next steps are to do noise analysis and individual opamp stability analysis in LISO one last time.

Attachment 1: CRYO_ISS2LoopGainTF.pdf
CRYO_ISS2LoopGainTF.pdf
Attachment 2: Cryo_ISS2_AOMCavity_Schematic.pdf
Cryo_ISS2_AOMCavity_Schematic.pdf
Attachment 3: Cryo_ISS2_AOMCavity_LTSpiceAnalysis.zip
  2212   Sat Jul 21 14:53:16 2018 awadeDailyProgressPMCNew modifications to the PMC north board

This circuit just doesn't do what it says it should do.  Need to inject waveform at FP1test and probe at each stage.  Then compare against LISO model.  If something is busted we need to know at which point we are getting this extra zero in the response. There is a lot not great about this particular board but it should just have a flat response above 488 Hz.

I don't think AD602 is in the LISO library.  Should be able to add it as some kind of hack op amp with fixed gain, 100nV/rtHz of noise (with 10 Hz corner) and some appropriate current noise with corner of 1 kHz.  Maybe check the AD602 datasheet.

We need to clean this up this week or do something drastic like replacing electronics with minicircuits or removing MC altogether.  For now we need to move onto solving bigger problems like the ISS, thermal stability, PLL readout noise 

Also, on the ISS do you now have a prototype working servo circuit and photodetector?

Quote:

We earlier found that the intended LPF isn't working so we thought of this external LPF idea. So I checked today the PZT input with LCR meter to see if it is in good condition or not. It gave C = 406.4nF, L=60.6 mH and R = 31kOhm. From the spec sheet, the C value looks 20% below the rated value but the spec value has uncertainity of +- 15%, so maybe our PZT is still good.

With the measured values, I calculated again (fitted using LISO) what good value of output resistor would make it closest to 10Hz pole. The value came out to be 37.1k Ohm. I have replaced this output resistor with 39k Ohm now. I'm attaching updated schematic for future reference.

keywords for search: PMC North Driver Board Schematic

Quote:

should probably put back the 30k resistor. It makes a low pass filter with the PZT capacitance (not the cable capacitance). This is to prevent shorting of the HV drive.

 

 

  2211   Thu Jul 12 15:18:26 2018 anchalDailyProgressPMCNew modifications to the PMC north board

We earlier found that the intended LPF isn't working so we thought of this external LPF idea. So I checked today the PZT input with LCR meter to see if it is in good condition or not. It gave C = 406.4nF, L=60.6 mH and R = 31kOhm 17MOhm. From the spec sheet, the C value looks 20% below the rated value but the spec value has uncertainity of +- 15%, so maybe our PZT is still good.

With the measured values, I calculated again (fitted using LISO) what good value of output resistor would make it closest to 10Hz pole. The value came out to be 37.1k Ohm. I have replaced this output resistor with 39k Ohm now. I'm attaching updated schematic for future reference.

keywords for search: PMC North Driver Board Schematic

Quote:

should probably put back the 30k resistor. It makes a low pass filter with the PZT capacitance (not the cable capacitance). This is to prevent shorting of the HV drive.

 

Attachment 1: D980352-AGupta20180712Mods.pdf
D980352-AGupta20180712Mods.pdf
  2210   Mon Jul 9 16:15:06 2018 ranaDailyProgressPMCNew modifications to the PMC north board

should probably put back the 30k resistor. It makes a low pass filter with the PZT capacitance (not the cable capacitance). This is to prevent shorting of the HV drive.

  2209   Fri Jul 6 23:43:54 2018 anchalDailyProgressPMCNew modifications to the PMC north board

Using buzzer, we found that the north path PMC responds strongly to buzzing at 484Hz. This was coincidentally the zero of one of the filters inside PMC feedback circuit. We tried to suppress the output with a 10Hz pole passive LPF at the output but it didn't work. On opening the box, we found that the actual circuit has different values and different circuit topology than assumed. In particular, the resistor at the output was found to be 30.9k instead of 1k. This I removed and shorted as we would use a more robust external LPF which does not depend on the capacitance of the cable. Secondly, the circuit topology of the filter which has a pole at 2Hz and zero at 482Hz was different than our schematic. It was a poor design with weird values. I implemented liso fitting in python wrapper and used it to optimize the circuit. Input file and jupyter notebook of optimization is attached. Also, last time Andrew changed R9 to 100 Ohms from 10 Ohms and was skeptic about the opamps stability with it. I ran a stability diagnostic for this part of the circuit as well. Input file U2Buffer.fil and jupyter notebook are attached.

Quote:

A few more changes to the PMC north board. Hopefully this is the last.

Changing series resistance at the output

There was some limit to the gain before the PMC loop became unstable.  I realized that I hadn't implemented the 10 Hz pole at the output, that could be the reason.  The resistor in series with the output was only 1 kΩ.  The Noliac NAC2124 PZT on the PMC cavity has a specified capacitance of 510 nF and about 5 meters of BNC cabling should have a capacitance of 0.5 nF (for 101pF/m).  This should give a pole of about 312 Hz, we want to bring this down.  I replaced the 1 kΩ series resistance at the HV output with a 30.9 kΩ to bring to pole down to 10 Hz.  

Schematic is updated below.

Removing short from PD RF input

For whatever reason the AC coupling resistor had been deliberately shorted on the PD RF input to the board (C3).  I removed the short.  It is AC coupled again.

 

Altering the Acromag excitation voltage and divider resistors

I realized running the Acromags off 6 V is a little awkward.  Most power supplies come in other standard values like 5 V, 7.5 V, 9V etc.  I've raised the excitation of the binary channels to 9 V and made some changes to the divider resistors. Resistors on SW1 and SW2 binary engages were changed to equal values of 1790 Ω for R1 and R2 (see option D, attachment, PSL:2058). On the blanking pin used to activate the AD602, equal values for R1 and R2 of 154 Ω were used.  These resistor choices give a high voltage of 4.5 V and a low of 0.66 V.  There is also a margin of error if the power supply used goes a little over voltage.

I've ordered standard barrel 2.1x5.5 mm panel Jacks so that the unit can be powered from a standard 9V plug pack.

 

Attachment 1: LowPrefFilter.fil
r R11 7.96k IN n1
r R12 79.6k n1 OUT
c C11 0.1u n1 n4
r R13 3301 n4 OUT
op LT1028 LT1028 GND n1 OUT
Attachment 2: PMCLowPrefFilterOpt.ipynb
{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
... 706 more lines ...
Attachment 3: U2Buffer.fil
r R10 49.9 IN GND
r R4 220k GND n3
r R5 220k n3 n2
c C9 2u n3 gnd
r R9 1000 n2 gnd
r R2 9.09k n2 OUT
op U2 LT1028 IN n2 OUT
Attachment 4: U2Stability.ipynb
{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
... 515 more lines ...
Attachment 5: D980352-A_AWade20180201Mods_--AGupta20180707Mods.pdf
D980352-A_AWade20180201Mods_--AGupta20180707Mods.pdf
  2208   Fri Jun 22 00:35:31 2018 awadeHowToComputersRunning parallel tasks on Caltech LIGO cluster

Shruti and I are running various training routines for the machine learning and non-linear controls.  It can be hard to guess the best learning rates, random action injection rates and other hyperparameters of the NN and tensorflow optimizations.  Although the best approach is to work intuitively on simple examples and then scale up, the optimization and rates of learning can be a little opaque.  At some stage we will want to throw a bunch of computing power at systematically narrowing down what works and what doesn't.

We basically want to spin up a bunch of training trials to test a range of hyperparameters without having to wait a full day for turn around for each iteration through a list of values. Running tensorflow based training on GPU might offer speedup on each step but won't necessarily help if it isn't a well parallelized problem. Its not clear to me that, for instance, the baselines deepq minibatching will work faster if we simultaneously draw samples from the buffer and do the gradient decent in parallel with between-graph replication.  At the end of each training episode the outcomes of each separate minibatch gradients have to be combined (that seems non-trivial) and then redistributed across the GPU (which sounds like it will have some hefty overhead as we scale up).  Managing this kind of parallelizing seems too far down the rabbit-hole of optimization science for our investigations.

I've been poking the people over at the jupyterhub LIGO chat channel about running parallelized clusters from notebooks.  LIGO is now running python notebooks on the LDG at http://jupyter.ligo.caltech.edu (and test server http://jupyter2.ligo.caltech.edu).  These can now launch a cluster of n nodes directly from the jupyter gui and we can use the ipyparallel python module to run parallel tasks directly from jupyter.  The only problem is that it ships with a generic virtualenv for the python3 kernel that doesn't include our gym or baselines environments from OpenAI.  We've also made modifications to these packages making them even more propritory.  Furthurmore, there is a problem with ipyparallel clusters, we've found that they won't launch worker engines unless the version of python is exactly the same. The juputer notebook kernals that we are using are python 3.5.4 and the workers are something like python 3.6

As a workaround we can launch our own ipcluster cluster on the ldas-pcdev14 headnode (or ldas-pcdev5) and connect easly directly from jupyter notebooks. ipyparallel manages all of the scheduling and we can launch over 20 learning runs simultaneously and/or schedule a longer list to run. This is relativlty easy to do and doesn't involve much hackery. 

I've got this working from within a python notebook (attached) and have documented the steps needed to get it running.  The actual worker nodes work just a little slower (15% slower maybe) than our Macbook Pros.  The advanage is that we now have more scope to make a bunch of parrallel trials and to detach from those instances.

Edit (awade) Sun Jun 24 00:24:20 2018: fixed issue where tensorflow graph is kept somehow as a global variable by baselines.

 

Attachment 1: MinWorkingExample_baselines_with_ipyparallel.ipynb.zip
ELOG V3.1.3-