Spotted a ton of accumulating overflows on the DAQ - specifically the output for DMass's PZT on his PMC.
It looks, from the EPICS screen, like that channel is trying to output a lower number of counts (~ -90000) than the DAQ can reasonably drive. (+/- 20V = 64K counts). We should try and set upper and lower limits on the channels. I'm not sure why this isn't set automatically in the RCG.
Also made a note on the ATF wiki ...
Seems to be a non-fatal error though.
I had left some of my output filters off on accident from doing a cavity sweep. This should be fixed.
Came in and found the PMC_PZT output trying to deliver an obscenely large number of counts to the DAC.
Not really sure where the issue is coming from - looks like the filters in the attached module. Anyway, just set a limit on what that module can output for the time being. Will leave this one for DMass to figure out.
Frank was talking about rebooting the frame builder last night. That shouldn't have affected the front end though. But if we do want to reboot the front end will all our settings be saved and restored automatically?
[Alastair, Koji, Zach]
As we discussed doing, we removed the PBS that was used to split the power between beams and put a 50/50 power splitter in its place:
The thinking was that doing this would minimize the AM coupling from polarization rotation in the EOM. After realigning the entire experiment, we observed no improvement in the low-frequency noise spectrum (compare with this post):
It does seem like there is a reduction from 10 mHz to 100 mHz, but it is tough to tell whether or not this is meaningful given the FFT bandwidth. There remains the possibility of true AM from the EOM itself, though it seems unlikely that it would be stronger than that from the rotation. I suppose we should measure it either way.
Another not-extremely-likely case is that the phase noise from the two oscillators is the same below 1 Hz. This means that their absolute noise levels (V/rHz) would have to be different by a factor of two to accommodate the fact that the AOM oscillator noise couples in twice as strongly from the double-pass. Since the phase noise tends to go up with carrier frequency, and since the AOM carrier is half that of the PLL, this isn't out of the question. We are planning to beat our two Marconis together to see if the low-frequency noise is enough to explain the gyro noise. Frank and Tara's data suggest that it stays fairly flat at lower frequencies, but they don't have a measurement at the low frequencies we're concerned with.
I received the PBSs from Tower Optical today, but it appears that the wave plates will come in a different shipment...
I have attached some pictures of the PCB required to make digital measurement for the seismometer.
InAsSb PD QE Test
The relationship between the spot radius and the apparent QE (EQE) was measured.
1) The spot size was checked with DataRay Beam'R2. The beam scanner was mounted on the post with a micrometer stage in the longitudinal direction. (Attachment1 upper plot)
It was confirmed that the beam is focused down to ~22um. The incident power was about 0.9mW.
2) The InAsSb detector (Sb3513A2) was mounted on the PD holder and then mounted on the stage+post. The photocurrent was amplified by a FEMTO's transimpedance amp (V/A=1e3Ohm). The dark current and the total photocurrent were measured at each measurement point with the beam aligned to the PD every time. The estimated EQEs were plotted in the lower plot of the attachment.
Note that P2, P3, and P6 elements have the size of (500um)^2, (750um)^2, and (1000um)^2, respectively.
The absolute longitudinal position of the sensor was of course slightly different from the position of the beam scanner. So the horizontal axis of the plots was arbitrary adjuted based on the symmetry.
The remarkable feature is that the QE goes down with small spot size. This is suggesting a nonlinear loss mechanism such as recombination loss when the carrier density is high.
With the present incident power, the beam size of 100um is optimal for all the element sizes. For the larger elements, a bigger beam size seems still fine.
The next step is to estimate the clipping loss and the saturation threshold with the Gaussian beam model.
Clipping and saturation were investigated by the semi-analytical model. In the analysis, the waist radius of 20um at the micrometer position of 8mm is used.
Firstly, the clipping loss was just geometrically calculated. Here the saturation issue was completely ignored.ã€€The elements P6, P3, and P2 have the sizes of (500um)^2, (750um)^2m, and (1000um)^2, respectively. However, these numbers could not explain the clipping loss observed at the large spot sizes. Instead, empirically the effective sizes of (350um)^2, (610um)^2, and (860um)^2 were given to match the measurement and the calculation. This is equivalent to have 70um of an insensitive band at each edge of an element (Attachment 1). These effective element sizes are used for the calculation throughout this elog entry.
2) Saturation modeling
To incorporate the saturation effect, set a threshold power density. i.e. When the power density exceeds the threshold, the power density is truncated to this threshold. (Hard saturation)
Resulting loss was estimated using numerical integration using Mathematica. When the threshold power density was set to be 0.85W/mm^2, the drop of QE was approximately matched at the waist (Attachment 2). However, this did not explain the observed much-earlier saturation at the lower density. This suggests that the saturation is not such hard.
In order to estimate the threshold power density, look at the beam size where the first saturation starts. The earlier sagging of the QE was represented by the threshold density of 0.1W/mm^2. (Attachment 3)
The QE of the (500um)^2 element has been tested with a half-power (0.51mW) instead of 0.92mW.
It is clear that the central dip depth is reduced by the lower power density.
I took some noise measurements for the doubling noise experiment. Here are the individual PD Spectra, and the Calculated Phase Spectra (no subtraction spectrum yet)
Dark Noise is with a razor blade dump in front of the Mach Zehnder input.
Intensity Noise is a blade dump in one arm one arm (and multiplied by two).
I have three data segments (time series) for each of the four photodetectors. PD1_data, PD1_dark, PD1_intensity, etc. For the dark noise, I block the input beam to the Mach Zehnder. For the intensity noise, I block one arm.
I blocked one arm and measured the incident power on each photodiode, as well as the power in the reflection off it.
Numbers are ~5%
I tuned up our first TRANS PD, S/N 03, that Alastair was working on with the J-Laser tonight. Here is the TF:
As before for the REFL diodes, I measured the output node of the diode with a probe so that I could tune the readout notch to 100 MHz. I had to change C2 to 33 pF to get 100 MHz in range with the tunable cap (C4), which I changed to a 1.5-50 pF one. Here is a full-span plot and a zoom around the notch:
I think it looks ratty at high frequencies because we are starting to see interference effects from the O(1 m) length cables.
While I'm still waiting for the proper connector for the vacuum feedthru of the IRLabs cryostat, I have connected to the Dsub9/15 split cable to another Dsub9 connector so that I can test the cooling of the InAsSb sensor in air. Also, the 2004nm laser, a fiber-coupled faraday isolator, and 90:10 beam splitter was moved to the cryostat table and fixed on a black al breadboard. [Attachment 1]
The InAsSb TEC was controlled by the TEC controller of ITC-50. I didn't change the PID parameters of the controller but the temperature nicely setteled to the setpoint. The sensor has a 2.2kOhm thermister. And the max current for the TEC was unknown. The TEC driver had the current limiter of 0.3A and it was not changed for now. With this current limit, the thermistor resistance of 10Kohm was realized. This corresponds to the temperature of about -20degC. According to the data sheet given by Alex, the resistance/temperature conversion is given by the formula
1/T = 7.755e-4 + 3.425e-4*log(R)+1.611e-13*log(R)^3
To satisfy the curiosity, the dark current of a (500um)^2 element was measured between -250K and -300K. At -254K, the dark current went down to the level of 40uA (1/15 of the one at the room temp). For the measurement, the bias voltage was set to be 0.5 and 0.6V. However, it was dependent on the diode current. (Probably the bias circuit has the output impedance). This should be replaced by something else.
The quantities we want to measure as a function of the temperature:
- Temperature: 2.2k thermister resistance / 100ohm platinum RTD
- QE (Illuminating output / Dark output / Reference voltage / Reference dark output)
- Dark current (vs V_bias) -> Manual measurement or use a source meter
- Dark noise (PSD) 100kHz, 12.8k, 1.6kHz, 100Hz
1) I've brought another TEC driver fro the PD temp control. This unit was borrowed from the 2um ECDL setup. Eventually, we need to return this to ECDL. (Attachment 1)
The PID loop of the TEC control works. But it is not well optimized yet. If you change the target temp too quickly, the TEC out seemed oscillating. Watch the TEC out carefully and change the temp setpoint slowly.
So far I have tried to cool the thermister up to 30kOhm (~232K) and I_TEC was 0.33A. I did not try further. I felt it was better to cool the PD base for further trial.
2) A part of the alignment study, the beam is aligned to A2P6. Also, the lens position was investigated, and I decided to move the lens ~1 inch away from the window. (Attachment 2)
In fact, this allowed us to insert the power meter between the lens and the window.
The QEs were measured at 293K, 239K, 232K, and 293K again. The cooling was provided by the PD TEC. At each temperature, the incident power was changed from 30uW to 1mW to see the dependence of the QE on the incident power to check the possible saturation.
The QE was 79~81% (the window T=96.6% was already compensated). I'm not 100% sure this 1% variation in the plateau is real or due to insufficient calibration of the REF PD.
The REF PD was calibrated at 1mW at 100mA injection current to the laser.
No obvious saturation was observed.
We can cool the PD with LN2 and we should make a careful alignment of the beam at each temperature.
I see the following periodically in the trend of my photodetectors. This is data point number on the x axis (sample rate = 4096) and counts on the y axis. The signal goes from -3500 to -9300 as the Mach Zehdner goes through fringes. I don't know what the jumps to zero are.
Here is the latest schematic for the PDs along with the board layout. I'm going to check over the routing one last time but it will probably require checking by someone with more RF experience too. The pcb drawing doesn't show all the features in the pdf. It uses split internal planes to distribute the power for the diode bias and also the +/-5v for the opamp. I've kept a full ground plane as the first one down, so there is a continuous ground plane directly underneath the tracks on the surface.
There are quite a few rule violations that I've just noticed when running the design rule checker. They're mostly clearance issues between the silk screen layer and various pads, but I want to get rid of them all at this stage.
The PD design is on the svn under gyro_electronics.
Okay, so most of these violations seem trivial. I've gone into the rule manager and set it up to use the same rules as Rich had on the other RFPD board and all the trivial violations have now gone. I've fixed all of them except one, which is a maximum hole size. We are using 195mil for the holes for the stand-offs. I'll need to check if there is a reason for that being there (ie is there some limit from the manufacturer) before I remove the rule or change the hole size.
I've added pads for the 1mm Perkin-Elmer photodiode. I've just done this manually rather than adding a new part footprint. The PD mounting point now looks like this screenshot.
For info I've attached the pdf for the Perkin-Elmer photodiodes from their website and also another pdf that I found that shows the footprints.
I bypassed Techmart completely and ordered the boards myself. The company has given us a shipping date of 1st Feb, and I've asked for next day delivery. I have emailed Gina to ask her to cancel the techmart order.
I submitted the PD board design for manufacture today. I didn't know what was the best way to do this, so I've put it in through Techmart by attaching the .zip file. I'll call the company in a couple of days to check that they received the order. It should take 4days to manufacture so we should have them soon. I'll update the svn folder with the current design and put the files on the DCC.
The PD boxes arrived today, and with the exception of a few minor issues they look great!! Here are a few shots. Here one box is shown assembled and mounted to the brass base and insulator that Alastair designed. The dimensions for these were chosen so that the PD is at 4" from the table.
The issues were/are as follows:
[Alastair, Steve, Zach]
I finalized the dimensions of the RFPD boxes this morning. This afternoon, Alastair and I went over them and made sure they jived with the base dimensions.
Steve let me use his P-card to order them. The grand total was $700 ($175/box).
Lead time is 5 days from order to shipping, then UPS ground.
We put in the PD in transmission, and here is our lovely signal. We took the beam in transmission and put it through a beamsplitter and then a lens and onto the photodiode and aligned it by moving each beam one at a time and maximizing the DC signal from the PD. We then plugged the AC signal into a spectrum analyzer, and we can see the beat signal at 95MHz.
Let's get some overhead photos of the table with lots of the parts and beampaths labeled please. It will help in understanding these results so far.
For example, based on the peak height in the spectrum analyzer and the transmission diode's transimpedance gain, what is the contrast ratio for the two beams?
And let's get that signal demodulated with a Rubidium Marconi and fed into the DAQ! We should be able to get some long term data with it.
Let's get some overhead photos of the table with lots of the parts and beampaths labeled please. It will help in understanding these results so far.
The peak on the spectrum analyser is at -68dBV. Converting this to volts (10^(-68/20)) we get 0.00040V
The transimpedance gain for the New Focus 1811 is 40V/mA and the response is approximately 0.75A/W. This means we have about 10^-9 amps of photocurrent, and 0.013uW of laser power.
The laser power on the photodiode is 16uW, so we have a contrast of about 0.08%......which doesn't seem like a lot. Admittedly our beams are different sizes coming out of the cavity and they are going through just one lens to focus down onto the PD, so perhaps this is where most of the contrast is lost. The beams are approximately the same power. We can increase the power on the PD a little as well.
Still trying to find the source of the low frequency noise, and I think we might have. Last week, we noticed that the AOM actuation signal had a similar shape even when the loop was disconnected (i.e. when modulation was turned off on the AOM VCO). Looking aft of the servo with the loop open doesn't really give anything meaningful without taking the loop shape into account, but it got us thinking that we need to carefully look at all the elements in the loop.
To do some more hunting, I decided to look at the gyro signal as measured by the open-loop secondary loop error signal, with the AOM frequency tuned to resonance manually. This is fine because the FSR does not shift nearly enough to push the secondary loop outside the linear region. I calibrated the response by sweeping the AOM at 1 Hz and measuring the slope of the error signal.
Here is a plot of some observations:
Clearly, the PD noise is a big problem. I looked at the time series (and thought I saved it, but I didn't---sorry), and what it looks like is that the low frequency junk is caused by random, discrete, loud events as opposed to stationary noisiness. You can see as the FFT averages that it starts out as relatively white noise at the HF floor level, and then all of the sudden jumps up when a glitch happens. Judging by the terminated noise level, I think this is originating in the PD circuitry.
Of course, the HF floor is too high, as well. This is probably because the secondary loop optical gain is somehow much lower than the level we had before. This could be the result of misalignments and/or power changes as we have been adjusting the waveplates to minimize RAM. The fact that this noise comes from the PD explains why the LF noise we have seen changes slightly from time to time, as the CW optical gain changes slightly from measurement to measurement.
The HF noise floor of the PDs was consistent with the LISO estimate (and the original design), so if we achieve the design optical gain it should not be a problem. The glitch behavior IS, however, and we have to sort this out immediately.
Having done some analog testing today, I'm convinced that the glitch-like noise from the PDs is not the ONLY issue. As shown in the previous post, I observe enough LF dark noise to explain the LF gyro noise if I wait for glitching to happen. However, the general LF hump shape is evident with only one average once the light is allowed onto the PD.
The best candidate is still some sort of RF noise. We need to work on the EOM stabilization, but I am also currently investigating some pickup crosstalk. With the secondary PD blocked, if I scan the primary servo I still see a PDH response in the secondary error signal, which should not happen. I've checked that the PD cables aren't switched, and blocking the primary PD makes both error signals go to zero. working on this now...
Today, in the process of designing a intensity control servo I took some noise readings of the PDs and also characterized how the modulation function in the Marconu FG works, as it would this would be essential in the process of desigining a efficient feedback loop.
Firstly, the alignment of PDs was off, so I had to correct them, which took me more time then it should have, but I finally got it done. Next, I moved on to taking noise readings to measure the 'free running' laser noise, but as it was pointed out to me, this plot in itself is meaningless until the power at which it was taken is specified or in other words the quantity of use would be which is also called the Relative Intensity Noise(RIN). What I measured was the Voltage noise in the PDs for a fixed Bias voltage( set using the power control of the laser). I calculated them for different bias voltages and the plot is shown below, the wierd shape is because I tried to splice three dfferent spans and it did ot combine as smoothly as I expected. I took noise at 3 different spans(100Hz, 1kHz and 50kHz) and combined them using splice.m program written by Koji. The change in the voltage noise with the bias voltage can be seen very evidently . We could then use the voltage noise at 1V as a measure of RIN . Again the plot looks wierd becasue of the splice function I used, I will post a better plot when I take another set of readings, in my next log. This plot tells us approximately how much of RIN is there and how much suppression would be needed in our servo. Additionally, I also measured the dark noise but they really wierd after splicing, so I will post those as well in my next log.
The marconi RF function generator I am using has a modulation input which can be used to control the power going into the AOM which would inturn control the power in the main beam, this is my plan in implementing the intensity feedback. So, I studied the AOM and how it responds to imput modulations of different kinds, this is what I learnt:-
The DAQ system in the ATF lab has not been yet setup completely and as mentioned in the previous eLog, we decided to go ahead and build those circuits by hand, as labaorious as they were we were finally able to get a few servo designed, simulated, characterised and running. In this eLog, I describe the servos we built.
The first servo was the cavity locking servo, as mentioned in a previous post that the SR560 used has very very low output rails(-4V to +4V) and hence can hardly keep the cavity locked in response to the laser drifting for a few minutes. We implemented this as a filter on a solder board, with rails of -15V to +15V but this wasnt enough, to hold the cavity locked for more than 10mins. We needed some very high voltages! So we put to use the piezo driver in the ATF lab.
Our initial servo was a simple one pole active RC filter with a cutoff of ~10Hz and a DC gain of 100. This worked and kept the cavity locked, but it unlocked itself after a 10mins or so. Now, when we implemented the piezo driver, we could keep the cavity locked for much much longer times(~40 mins) but it was only marginally stable and showed some features of instability. This was because the piezo driver itself has a low pass characterstic with a pole of ~8Hz and a gain of 20 and this was making the feedback loop unstable, because now we have slope after 10Hz and at around unity loop gain frequency ~300Hz the phase margin was very poor ( probably 10's of degrees) this made the loop unstable.
Solution was simple- add a zero and push up the phase margin! This was done with a zero at around 500Hz and an additional flat gain stage was put in with a gain of 30, this was to further increase the UGF and push it beyond ~3-5 Khz. This was first simulated and the built and tested.
The schematic of the setup is given below along with the components values(note: I have simply modeled the piezo driver's response as non-inverting low pass filter,this does not refer to the PZT actuator's response but just the high voltage piezo driver's response)
The following images are comparison of TF simulated on LISO and measured TF for each of the stages and the combined TF as well. But the TF with the piezo could not be taken yet, so just the LISO result is shown. Some comments on them:-
The measured response is almost a perfect match for the simulated LISO response with the gain differing by 0.02dB. So the flat gain is working as expected with a wide bandwidth of about ~20kHz.
2) The following is the TF with the second boost stage included:-
The TF is as expected. With the zero almost exactly at 500Hz.
3) The full cavity servo simulated TF:-
The features are exactly as predicted, a steep slope from the two poles and a flattening effect by the zero after 500Hz, this combined with a flat gain stage pushes the UGF to almost 7kHz, which is more than sufficient for our purpose and the phase margin has also drastically improved. Also find images of the attached circuit.
The next step was to analyze the and design the intensity servo. For this first the free running laser noise was measured this was around , which is very bad. We wanted to reduce this common mode noise to shot noise limit, which would require almost 5 orders of supression. Also we would want to setup a stable offset as if its forced to zer we wont have any power from laser at all. We are in the process of designing this and we should be able achieve this in a few days.
Today, I implemented the intensity servo and characterised it using the voltage injection method to calculate its loop gain. I also compared it with the simulations for the same that I performed on LISO. I have attached the figures below. I also completed the circuit I made to measure the differential signal of the PDs(its just a voltage regulator setup with a AD620 with a gain of 100), but I could not characterize or take measurements using that. I will do that tomorrow.
1) RIN with and without feedback( both in-loop and out-of loop). The gain for this loop was set at 200, this was to compensate the -16.5dB loss in the remaining setup(AOM, RF function generator etc).
2) Open loop gain Transfer function- simulated and measured.
I am still stuck on how to measure the difference in the reponse of the two transimpedence amplifiers on the photodiode readout board, I have a few ideas, I am assessing their validity.
something seems bogus here...the suppression at 1 Hz is much more than the open loop gain. usually this is mathematically impossible...
I did not note this inconsistency. I will take new measurements. I am wondering what error on my part could have caused this to happen.
As mentioned in a previous eLog, due the inconsistency observed, I decided to measure the Free running RIN and supressed RIN again. I present the results below. A few things are bothering me though:
1)The peak at 25kHz has been supressed! How? UGF is around 10kHz!
2) The supression is more almost 2 orders(40dB) at 100Hz. Expected supression about 30dB.
The splicing has resulted a sudden jump at 100Hz, but I think thats nothing to be worried about, as expected we get an a little more than 1 order of suppression, in the bandwidth of interest from 3Hz-10kHz. Also, I think I figured out what went wrong last time. I had the spectrum analyzer in the DC coupled mode when I was measuring the free running RIN of the laser, where as for the suppressed case I was using AC coupled mode of spectrum analyzer, I think that could possibly be the reason for the inconsistency observed in my previous measurement.
I finished stuffing the first RFPD board today, so that we can begin with the testing. Unfortunately, it seems to have failed its first test, as something shorted when I powered it up. Rich and I traced it back to one of the 15 V regulators, and we are going to use that as a starting place tomorrow.
There was nothing obviously wrong with the board as checked by measuring resistances between terminals that should not be shorted (and vice versa), so we hope that there was just a bad regulator.
Besides that, the board looks sweet!
I traced the problem back to the -15 V regulator. There appears to be 25 ohms between the output terminal and ground.
I checked each of the capacitors between these two nets (i.e. I removed them and then measured their capacitances) and everything checked out.
Finally, I removed all components between these nets. The 25 ohms remained. I did other things like removing nearby diodes and also the 9-pin D-sub, all to no avail.
I checked one of unstuffed PCBs and it did not have the same issue. I started stuffing it (just the power management), and it was able to regulate to +/-15 V with no problem. This suggests that there is something wrong with the first board internally.
Tomorrow, I will finish stuffing the second board with fresh parts. I will check for shorts along the way to see if I can catch the demon in the act.
screen -RAad autorun
On Tuesday 6/7 we met to discuss next steps for getting PD testing up and running in CAML (cryo auxiliary mariner lab). Here is a rough 2-month plan:
Step 0 [1 week]: Get CAML workstation running and confirm connection to EPICS/CDS.
- Organize BNC cables coming in from QIL and connect to the PD testing table.
Step 1 [1 week]: Outline desired measurements and sketch diagram of electronics/components.
- Decide how we want these component organized; i.e. how many boxes, number of inputs/outputs per box.
Step 2 [2 weeks]: Implement desired setup.
Step 3 [2 weeks]: Replicate QE measurements on JPL PDs
- Tweak setup as necessary
Step 4 [2 weeks]: Test MCT detectors
Rich removed the window from PD on Friday. The basics steps for the removal are the following:
0. PD in a socket, it helps;
1. In the beginning you open the screws on the cutter up so they just hold the photodiode;
2. Make a light little line first at desired height and make sure that it is a circle and not a helix;
3. You do not want to go start right away full force on it, you want to make tiny little incremental cuts. Eventually it just falls apart;
Today we were able to get the servo running which we described in the previous eLog. The servo locked beautifully and the instabilities other wise observed without the zero we specifically added was gone. There was small change that had to be made, the servo had the wrong sign which we very took care of. So, just for the sake of completeness, I am attaching the TF of the inverted and the non- inverted servo. Also, I am attaching the corrected schematic. I coudn't take a snap of how the lock was before and after we introduced the our servo( I will make sure its in the next log).
Addititonally, we assembled koji style beam dumps onto our setup, but the alignment was such that we couldn't place the third beam dump for the reflection coming from the bea splitter. So we may have to realign the beam splitter and then re-align the PDs as well. Its not as tedious as it sounds and we should be done with it tomorrow.
The only thing that would now remain would be to design and implement the servo for intensity suppression, what makes it ore difficult is that we need to include a very quiet reference in the circuit and we are brainstroming over it. Also, find attached a image of the actual servo we built. Also, I have attached the LISO simulation code.
#---------------Frequency tracking control servo---------------
#stage-1- non-inverting gain stage,G=30
r r1 1k n1 gnd
r r2 32.8k n1 n2
op u1 lt1128 nin n1 n2
#stage-2, Low pass filter pole=10Hz, zero=500Hz
r r4 2k n2 n3
Below is a plot of the measured intensity noise (calibrated to W/rHz using the responsivity and transimpedance from the datasheet) of the PDA10A we are currently using for the primary lock loop, for various levels of incident power. In doing so I realized that this is a Si detector with only 0.02 A/W responsivity @ 1064 nm, which is not ideal for our setup. There are fast InGaAs diodes that we can use instead.
Below ~200 uW, the spectrum seems to be dominated by the dark noise of the PD. Above this level, the broadband noise floor goes up approximately as sqrt(P), but at a level much higher than the predicted shot noise level. For comparison, the expected shot noise level for P = 500 uW is about 1.4 x 10-11 W/rHz, which is below even the measurement noise floor. So, there is some power-dependent noise source that is limiting us here once we've surpassed the dark noise. Perhaps this is a consequence of using a detector with such poor quantum efficiency for 1064nm. There also seems to be some lower-frequency 1/f noise that increases with incident power.
Below is a plot of some noise measurements on the PDA255 with which we have replaced the old PDA10A as the primary REFL diode. The traces are:
I realized that last time I did this I simply turned off the oscillator to take the no-EOM data. This is wrong because it also turns off the LO to the mixer, so the PD signal is not getting mixed down. This time, I physically disconnected the EOM to take this trace.
The plot is not very illuminating, except in that it shows that the noise in the demodulation setup is dominated by the dark noise of the PD. As seen in a previous post, the dark PD noise level is above but comparable to the input noise of the PDH box above ~1Hz. Below this---in our frequency band of highest interest---the PDH input noise dominates at the moment. The same linked plot shows that the theoretical noise estimate for the PDH input is substantially lower than the measured noise at low frequencies, so the PDs will certainly limit us if we get it to this level.
It goes without saying that both the final PDs and the final servo should contribute much less noise than these do.
You ought to measure the PDA255 noise with the RF analyzer and compare it to the spec and the thermal noise you expect from such a diode.
Is the demodulator noise of the PDH box good enough for our purposes? i.e. will we reap the benefit of a good PD or not?
Below is a plot of the measured gain of a 1-kHz sine wave through PDH box #1437 with a varying gain knob setting. The variable gain stage (AD8336) is linear in dB with a slope of roughly 5 dB per unit. This allows us to generalize the measured TF at, say, G = 0.5 to any gain setting. Alastair's measurement in ELOG 1080 is consistent with a gain setting of 0.5, so if we can get his data we will have a generally applicable TF.
As I outlined in this post, the primary PDH box needed some modifying if we are to see any significant improvement from the new setup. I made these changes:
The input-referred noise is now at the level explained by the LISO model---unlike it was before---though this calls for a funny story:
When I plotted the noise and divided by the transfer function I measured, I got a noise level that was exactly 10x higher than predicted by LISO. After quite a while of scratching my head and checking my code, I realized that I mistakenly changed the gain of the wrong stage (despite writing down the right one in my notes), causing the noise from the nasty LF356 switchable stage to have 10x the influence. I will correct it and re-measure tomorrow, but the important thing is that we have removed the noisy 8336 from near the input.
Here is a picture of the modified portion of the board and a LISO prediction of the noise when I've corrected the mistake. You can compare with the plot in the link above, but this corresponds to a true low-frequency voltage noise betterment of 20 dB at low frequency, which adds to the 40 dB improvement in the servo's contribution from the expected optical gain increase. This actually puts the servo noise below the aLIGO requirement between about 100-350 mHz.
I corrected the mistakes to the PDH box. The LISO estimate of the noise from the previous post is no longer quite right, however, as I had to reduce the gain at different points than I had originally intended (originally I was going to do at least part of the reduction at the output stage, but this proves difficult because doing so without being very careful results in a different gain for the inverted and non-inverted modes the way the output stage is designed). The result is that the high-frequency noise is higher than anticipated. Taking the increase in optical gain into consideration, though, the high-frequency contribution to the gyro noise is still lower than before.
Here is a transfer function, showing an overall gain decrease of about 35 dB from the previous case (this post, about halfway down):
Here is the input-referred noise spectrum, along with the LISO estimate for this circuit. The excess low-frequency noise is now absent, but the noise at high frequencies is higher than estimated. It looks as though this might actually be the noise of the spectrum analyzer (the output noise level here is on the order of 20 nV/rHz, and though I had it auto-ranging it could have gotten hung up). I will check this in the AM. Either way, there is a big improvement in broadband and the servo's contribution to the gyro noise is below requirement in a big chunk of our operational band (see previous post).
The input noise of the spectrum analyzer at the lowest range is ~20 nV/rHz. Subtracting this from the measured output noise spectrum of the PDH box before referring to the input, I get a noise spectrum that matches closely the estimated noise from LISO. This box is ready to go. Note that simply subtracting the noise off the way I did does not lead to a large systematic error because the true servo noise is of just about the same level.