I used the netgpib package that Yoichi Aso wrote (mostly to teach myself how to use it). As an example I measured the noise on the LT7812 (+12 V) voltage regulator in the UPDH box (#1437).
With these command lines:
SPSR785.py -f 10k -i crocetta -a 10 -b 10kHz -n 800 -v 30 --avgmode RMS --ic1 AC --ig1 Float -w Hanning
SPSR785.py -f 100k -i crocetta -a 10 -b 100kHz -n 800 -v 30 --avgmode RMS --ic1 AC --ig1 Float -w Hanning
I get two files with different BWs. It sets up and grabs the data from the SR785 easily. There are similar scripts for some of the other GPIB devices we have.
Not really low noise, is it?
I deleted all the pictures from the lab camera, there were 1100 of them and it was slow. I have local copies of everything, let me know if you need anything.
Can I go to a transimpedance of 300k (3x the resistance) and divide the capacitor by 3? So long as I am not AC coupling (yet), and my noise is below my DAQ noise, I dont need to do separate amplification, as the DAQ noise will probably set my SNR even if I up the transimpedance resistor. This is more of a op amp feedback question.
If I understand correctly, the capacitor is there to keep the op amp's feedback from going unstable, but I am not really sure how to model the capacitance of the op amp (in an envelope manner).
I have finished rebuilding the Mach Zehnder. There is some serious scatter/glow from various places, but there was before as well. I am dumping all the obvious beams on blades. Pictures incoming.
I am coloring 3-5" in height on the inside of the box with sharpie, in an attempt to absorb the scatter around beam height. I have no idea if this change in color actually corresponds to changing the reflectivity for 1064, but I made an attempt at a blackbody anyways.
Full post + pix to come
I'm sure that its more precise to just scan the counter propagating beam with the first direction locked (or not). In that case, you get to directly read the frequency off of the thing driving the AOM. No cal required.
Whoa, that seems high! In the post I made about this new setup, I estimated the finesse to be ~550 assuming 50 ppm losses and no degradation in R for the Y1S mirrors in our cleanest of clean labs. Of course, the value we are using for the PZT gain may be off by a factor of 2 or more, so we'll know better once we do what you're saying with the sidebands. For completeness, what was the drive signal to the PZT for this measurement, 10 Vpp triangle @ 100 Hz?
I scanned the cavity to measure the FWHM of the transmitted carrier. The plot is attached. I have converted from volts to frequency using 2.5MHz/V as the piezo response. This number might be entirely made up, and I will actually try to calibrate this tomorrow using the method that Frank explained (using the sidebands as a frequency reference).
The FWHM from this measurement was 79704Hz. I measured the cavity length and got 3.175m, but this is probably accurate to around +-4cm. It is really hard to measure right up to the mirror, particularly when they're in different heights of mirror mount. Having said that, Frank thinks the error in the voltage measured by the scope could be up around 10% so that will probably be the dominant error in our finesse measurement.
This gives a finesse of 1185 and a FSR of 94.4MHz.
After some hunting, I found the correct schematic for this Universal PDH box. It is v1 of the D0901351 variety. The 'version numbers' that Alberto refers to in the Wiki are his own made up numbers and don't correspond to the actual numbers on the schematics.
Attached is the LISO model results of what we can get with ~10 R and C changes. I found that luckily the zero in the U7 stage, as stuffed, happens to cancel the cavity pole frequency of the gyro! Seems too lucky to be true, but that's what the numbers say. I await for Alastair to measure the cavity finesse and bandwidth for real in the morning.
In this new PDH circuit design, I've made the boost stage have a high frequency gain of -1 instead of -0.01. The boost frequency is getting moved up to ~1-2 kHz. It will be a true integrator below this frequency.
The OP27 input stage will have a gain of 40 instead of 2. This is good for noise. By using a 25 Ohm resistor on the input, this will also balance the current in the 2 legs of the OP27 and reduce some offset.
We will also take away the 50 Ohm resistor at the OP27 (U1) input. The proper place to terminate the IF port of the mixer is at the IF port and NOT after the IF low pass filter. Those mini-circuits mixers are not 50 Ohms above the bandpass frequencies. That makes the high harmonics shoot backwards into the mixer and make it angry.
According to my LCR meter, the NPRO PZT capacitance is 8.5 nF (as measured from the front panel of the controller). So its to small to be used to make the dominant pole in the servo. We'll keep the pole in the front of the circuit.
Soldering to follow in the morning...
Unimportant EMI - ignore and press on at full speed.
I built the attached circuit twice on a 3M solderless protoboard, and had a noise discreprency between my two channels when trying to measure the PD dark noise.
I plugged the PD into a scope, and saw the following difference (second picture). These ringdowns occur whether or not my PD is on.
What is this op amp behavior, and what makes it worse?
The scale is the same for both traces on the scope.
i've replaced the temporary hard disk for the full frames by a new, larger one. As nobody wants data from the busted one i will send it to the manufacturer for an exchange now...
Woe. The noise of the 50 Ohm resistor is ~1 nV/rHz. The gain of your opamp is 100k/50 = 2000. So the output noise should be at least 2 uV/rHz.
Its better to diagnose the noise by not having the resistor there and/or to put the PD there. I have also added the LT1792 to the LISO SVN.
I tried to make a simple current to voltage convert for my Green PDs, and measure the noise. Hilarity did not ensue.
I made the following connections for an LT1792 (I also tried a LT1012).
| ___ 100k_____|__
| |\ |
Gnd-50Ohm-BNC----|- \ _______|___BNC____ to SR785
I stuffed this into my existing hydra board, and pulled out all the other components. My Noise looked WAY TOO HIGH
I was getting well over a uV/rtHz up to 800 Hz. This does not add up.
I also tried:
I am curious about:
There really has to be something stupid I am doing...
I was able to get the cavity to stay locked in the unmodulated (CCW) direction by increasing the gain of the slow channel. The issues we were having this morning were caused by the ~25 kHz resonance of the PZT, which must have reached unity gain when we put enough PDH gain in. The added slow gain seems to be enough at low frequency to keep the thing locked, but if I try to up the fast gain the thing starts to oscillate. We will have to modify the servo to accommodate the PZT spike to make the lock more robust.
Once having locked the original direction, I steered the double-passed beam from the AOM into the CW direction of the cavity and aligned it to isolate a 00 mode at the output. It's not quite perfect, as the REFL signal is still clipped a bit out of the faraday isolator, but it's enough to couple a decent amount of power into the cavity, and the clipping will likely disappear when we put (any) modematching optics in this path.
The AOM seems to be working very well. I was able to ramp the frequency tuning voltage into the driver and see the different modes scan through on the CCD at the output. I increased the voltage from 7.8 V (which corresponded to 45 MHz) to 8.77 V (which must be very close to the desired FSR/2 of 47.5 MHz), and saw a strong 00 at the output when the CCW beam was on resonance. Hence, the CW beam is not strictly on resonance, but it is "locked" a fixed frequency very close to one FSR from the CCW beam, and we see something close to resonance at the output as a result.
The input power in each direction is close to the same (as cursorily seen on an IR card), and the emergent power at the transmitted end of the CW path is roughly half of that on the CCW path. Given NO modematching in the CW direction, this must mean that we are fairly close to resonance in that direction.
Tomorrow, we will begin doing the following:
Here is a picture of both directions simultaneously outputting a reasonable 00 with a macroscopic frequency shift ~1 FSR:
Here are plots for the noise of my op amps (in Volts).
I compared 8 op amps, these are the two winners I came up with, AD743 clearly wins. I also included (for later reference) the noise of the ISS PDs I am using.
Op amps I looked at:
This list was made out of Frank's head based on what he thinks we already have. The AD743 seems to give a SNR of 1k throughout my measurement band, which was my somewhat arbitrary requirement (e.g. based on what I got optically when I used prisms, but I don't know how much my old layout was limiting me).
We were able to resolve the AOM double-pass issue by enlarging the beam going into it. We had originally tried using the smallest practical waist (~ 70 um) because the jitter suppression scheme only works perfectly--i.e. beam ROC = mirror ROC with the mirror exactly R away from the AOM--in the limit where zR --> 0, or w0 = 0. It appears that this caused the beam to diverge too quickly within the AOM, which is pretty long. We have changed the waist size to ~200 um at the expense of moving the mirror in to z = 15.5 cm from the AOM (as opposed to z = R = 30 cm as it was before). We are now able to isolate a decent-looking double-pass beam, which we verified by measuring its power. Hopefully the loss in jitter suppression will be insignificant as we only expect the AOM to shift by hundreds of Hz or so, corresponding to a very small angular displacement.
After this progress this morning, I tried to realign the cavity and get the original unmodulated direction to lock, but I am having quite some difficulty. I can align it and isolate the 00 mode pretty well, and there appears to be a decent amount of power emerging at the transmitted end on resonance (as seen on IR card), but the thing just doesn't want to lock. I have tried to adjust the RF phase to no avail. I suspect it has something to do with a polarization issue with the BS I am using to screen some power from the REFL beam before the PD, but I can't figure it out. I will try again tomorrow morning.
Once this direction locks again, we can feed in the modulated beam around the other direction, and we should have all the necessary equipment to set up the second loop. I spoke with Frank today, and I believe we will need a Marconi to set up a PLL at the output to do the frequency readout. We will figure this out when it comes up, I guess.
Of course, one might ask why I rebuilt the Mach Zehnder readout...
Here are plots of the various PDs with one arm blocked and the HEPA on (I lated checked that the HEPA being off didn't really fix the coherence). Things to note:
In conclusion, I think I was geometry limited before (using prisms), and I am rebuilding the readout with an SNR of 1000. This is more than 3, so its better
Some more updates:
Update on rebuilding:
I shrank the Mach Zehnder by 2" per side to accommodate putting the PD readout inside THE BOX. It is still tight, but I think I can come up with a way to fit everything.
The Mach Zehnder is somewhat realigned, and somewhat mode matched (the fringes arent obviously two different modes anymore). Once I put PDs in, I can fine tune the alignment easily.
I am in the process of rebuilding the Mach Zehnder with the following goal:
I want the entire experiment to live inside one box, readout included. To this end I needed to compactify the Mach Zehnder somewhat.
In realigning the ovens, I looked closer at the crystal which was giving me problems before (the exact same behavior). When I look through it, I can see what looks like damage through the center of the crystal. This is consistent with the crazy behavior I get where aligning through the center of the crystal produces terrible beam profiles.
| // |
| // <--- Crack? |
I will proceed with the experiment, having to go with some funny alignment.
which days do we need to copy? It's a lot more data as expected, 64gb per day (!) so plz let me know which day is required to copy. We should time it well as the framebuilder will delete the old data almost instantly to free space for tyhe new data. So if we delete all current data it depends on how much of the old stuff we copy how long we have to access the data before its deleted automatically
Alastair and I have been working on getting the AOM double-pass setup going. We have borrowed a curved Y1 from Peter to use as a retroreflector while we wait for the one we ordered. We are using a Tektronix FG through the 2-W MiniCircuits RF amp on the shelf to drive the AOM at the moment, and that appears to work decently well. We are running into issues with alignment, however.
Right now, we can get the 1st-order beam isolated with an iris and reflected back along the same path back into the AOM, but what comes back out isn't very clean. What we should see is the singly diffracted beam that gets reflected straight back through the AOM (without a second diffraction), and then the doubly diffracted beam--what we want--which should be a little dimmer and should lie exactly along the input beam. We tried lowering the AOM drive power until only one beam was left (the single-pass beam), but this never happened; all we got was something that looked like two dim beams close together or one very elliptical one.
We think it is because we don't have any way of adjusting the orientation of the AOM at the moment. Alastair has drawn up a new mount which will allow us to put a tilt stage on top. In the meantime, we will try to get something remotely reasonable out and attempt to lock the cavity with that.
Yes, the mount for the AOM sucks ass. I drew up a new one last night that will incorporate one of the New Focus 9082 alignment stages. It's with the machine shop now so we should have it towards the end of next week. I also ordered the alignment stage from Newportfocusnew (as I will now be calling them).
That's the 9082 stage arrived now.
the new disk for the framebuilder arrived. So i try to move the data from the last days before the crash to the disk which is currently installed. As i was asked to copy only the last view days the space on this smaller disk is sufficient. If everything is done i will partition the new, larger disk and copy everything to there. Is there any time i can't shut down the frontend? Any plans for working with the RT stuff so far? If not, i'll try to do this late today or tomorrow afternoon, depending on the work on the cavity stuff...
I will be using a Thorlabs DET110 with something like a 5 or 10k load resistor, and an SR560 for gain to replace the really crappy thorlabs PD for green.
Now I really wish we had made those low noise DCPDs...I guess that you can steal the ISS PDs for now. Just move them into the appropriate place and use the ISS box to power them and read them out. We can just continue to use some Thorlabs diode + SR560 as an ISS for now.
One of my switchable gain Si detectors appears to have particularly bad noise perfomance.
I compared the dark Voltage noise of my Thorlabs PDs. I am barely below the ADC noise on one of my PDs, and it onlyu gets worse at lower frequencies, so for Beta, on my current setup, the PD noise is limiting me below about 5 Hz.
Added a measurement:
I made a rough measurement of the mode shape of the laser used in the fiber setup. I need to go back in to get more points and estimate the error, but to first approximation the beam waist is about 1550 microns. The data I was able to take is included below (the x axis is in cm, the y in microns), but I will go back and take more data tomorrow morning.
It's also nice to put the curve that you're fitting with, the fitting parameters, and to extend the plotted curve to show the waist itself
More to come.
I made analog measurements of the PD dark noise, and for the most part it seems below the noise downstream of the AA box.
BAFFLING BEHAVIOR SEE PLOT TITLED MZWTF:
When I block one arm, I have decent coherence above 1 Hz. When I block the other, the same is true. When I unblock both arms, and let the Mach Zehnder interfere, my Coherence takes a dive at 1-2 Hz. I DO NOT HAVE ANY IDEA HOW THIS IS POSSIBLE.
THE INDIVIDUAL TIME SERIES LOOK LIKE I EXPECT, AND THE MACH ZEHNDER IS STILL ALIGNED ENOUGH TO INTERFERE.
I started chasing my PD noise, and I found extra noise after the input to the AA box, on two channels, one of which I was using in my Mach Zehnder setup.
I also added some slow channels to the DAQ.
These are name as follows in the DAQ:
It seems that one of my photodiodes was bad. PD4 (the IR phodotiode that was incoherent with all the others) was swapped for a PD of the same model from Thorlabs, and the mystery incoherence disappeared.
I still am only at the 95-99% coherence level at the DAQ below 8 Hz, but before PD4 was only 70% Coherent with all the other PDs, and it was the only PD that had this behavior.
I added a plot of what I am working with rright now. DC1 and DC2 are green, DC3 and DC4 are IR. The Spectra are taken with 1 arm blocked. The pink curve is my ADC noise.
I have updated the 40m SVN with all the code needed to run the MZ subtraction. The data files are about 250 megs total. I will upload them to the elog if that is acceptable, if not I will upload the downsampled data (which should be about 16 megs).
I put the transfer function inside the subtraction function (because I was doing it wrong before). Now f_domainsubtraction.m should be useful as a general frequency domain subtraction tool.
as it turned out that one of the hard disks failed so i had to replace it. The device contains all "full frame" data, nothing else. Unfortunately i didn't have a spare disk of that size so i replaced it by a smaller one. So the new volume is empty and no past (full) data is available. Trend data is OK.
So if nobody needs the full data from the last couple of weeks i will send the disk back to WD to get it replaced. If the data is/was important we can wait until i get a new disk of the same size (i ordered one today). If this shows up the next two days or so we could try to copy most of the full data to the new one, but only if really required as it takes ~6h or so to duplicate the disk and i would like to avoid setting everything up. So if anyone needs the old full data plz let me know. If i don't here something within the next two days i will send the broken disk back to WD. Again: any TREND data is good, only FULL data is broken.
No data I have taken needs to be recovered. If it costs us *very little effort/money, I would like to have the last few days, but I can retake it all too.
I just ordered a replacement curved mirror for double passing the AOM. The one we have at the moment is for 45P, and we really need 0 degrees incidence. It is listed as in stock and I've put down for quick delivery. It will be a Y1-1025-0-0.30CC
I've taken a beamscan for the path where the AOM is going to be installed. This is just so that we can accurately pick the lens and curved mirror that are going in here. There will be a second mode matching telescope after this set of optics before the CW beam goes into the cavity.
1) Matlab figure of the fit
2) the text file of the beam scan (1st column is z distance from last turning mirror (m), 2nd column is x-diameter (um), 3rd column is y-diameter (um))
3) Photo showing the location on the bench of the 'last turning mirror' prior to the AOM that the z values are taken relative to (in other words the mirror is position z=0)
From the fit the values for the waist radius and location are:
x 267+- 4um, at -0.324 +- 0.009 m
y 356 +- 3 um, at -0.277 +- 0.004m
We will fit the optics for the AOM using the average of these which is a waist of 311.5um at 0.3005m
0.0365 1086.9 940.5
0.0465 1106.3 964.7
0.0565 1123.1 981
0.0665 1145.5 982.4
0.0765 1163.1 983.7
0.0865 1189.6 990.5
0.0965 1208.5 985.5
0.1065 1231.7 1020.2
0.1165 1237.3 1026
0.1265 1266.2 1047.7
0.1365 1286.2 1050.4
0.1465 1316.3 1069.1
0.1565 1334.7 1079.6
0.1665 1353.4 1110.8
0.1765 1384.1 1116.8
0.1865 1388.2 1126.8
0.1965 1425.9 1139.6
0.2065 1448.8 1152.5
0.2165 1468.6 1165.5
0.2265 1488.1 1182.4
0.2365 1492.2 1196
0.2465 1531 1213.6
0.2565 1549.2 1244.2
0.2665 1574.1 1264.1
0.2765 1598.3 1281.4
0.2865 1646.2 1305.3
0.2965 1647.7 1308.3
0.3065 1687.9 1313.5
0.3165 1718.3 1340.8
0.3265 1716.3 1342.9
0.3365 1739 1370.1
0.3465 1783.7 1385.8
0.3565 1793.2 1399.6
0.3665 1798.4 1410.8
0.3765 1822.6 1423.6
0.3865 1860.1 1441.3
0.3965 1882.4 1463.6
0.4065 1901.1 1479.6
0.4165 1916.5 1493.2
0.4265 1941.4 1510.1
0.4365 2013.7 1543.2
0.4465 2040.4 1551.2
0.4565 2079.9 1570.1
0.4665 2090.1 1597.4
0.4765 2126.7 1605.5
0.4865 2157 1629.6
0.4965 2183 1634.9
0.5065 2201.1 1643.7
0.5165 2230.6 1674.6
0.5265 2276.1 1687
0.5365 2271 1703.2
we had a disk failure on the weekend on one of the harddisks. So fb0 will be down until we replaced that disk...
I have added all my subtraction code to the SVN, and will try to keep it under version control. It should also now be comprehensible to anyone who wants to read it. It is in this directory on the 40m svn:
I have included a .png diagram of the code.
Do you detrend x1-x4 before you calculate phi532 and phi1064? I thought detrending changes the DC values and thus leads wrong results for the phases.
I have diagrammed my MZ subtraction algorithm. I am redoing my code now that I have realized a few things so that another person can read it. I will update the guide with the associated MATLAB code names, and continue to put everything in the SVN.
x1,x2,x3,x4 are the time series of my PD's during a segment of free running MZ data.
n1,...,n4 is a (detrended) time series of my noise in each PD due to some source.
oops, I do not. Changing diagram to reflect this.
Koji and I tried to chase down some incoherence in my intenstiy noise measurement.
Things wrong that we found:
THE PLOT (SUMMARY)
We now have the mount for the AOM. Zach and I went looking for a suitable curved mirror to double pass the AOM and found a 30cm 45P at the 40m. I'm guessing it's not ideal since we're using normal incidence but we'll probably have to order another one.
Zach had thought out the jitter side of things, and in order to ideally get rid of this we want the distance from the AOM to match the radius of curvature of the mirror. Then we need to set the waist size so the wavefront curvature matches the mirror.
In order to do this we need Zr<<Radius of mirror. In other words we want a small waist in the AOM. Somewhere around 70um should allow us to have the waist on the way back to be about 1mm different in position from on the way through if we're using the 30cm mirror.
The inner loop of the mach-zehnder is setup and crudely aligned. The next step is to finish the alignment, lock the small mach-zehnder and then build the bigger one around it.
Also: I am still baffled by why I can't subtract better where there is such a high coherence at low frequency, it's above 0.99, yet I only seem to be able to get a 1.2ish order subtraction. My frequency domain results are about the same as the Wiener filtering results generated by the code: MZwino in the svn.
You should use your knowledge of mathematics to make an estimate for the amount of coherent signal based upon the coherence estimate. Then you can plot the predicted subtraction factor v. the actual subtraction.
We need a fisheye scheme so's that we can stick a camera underneath the shelf and take a picture of the whole gyro side of the table. Ideas?