That is strange.  I had also forgotten that the 40m wiki isn't hosted at Caltech.  I still can't get access to either of them.

Seems fine, now...

 The ATF wiki is down.  So is the 40m one.  The elogs are still running and I can ssh into Nodus which is where the ATF wiki now lives.  I'm having a bit more of a look at this now.

 Below is a comparison of the measured input-referred noise of PDH box #1437 with the estimated noise from LISO. It appears that the measured noise is substantially higher below around a kHz. The only thing not included in the model (besides shielding capacitors, etc.) is the AD8336 variable gain stage, but unless this is a limiting noise source it should not contribute to the input-referred noise. The LISO model output is provided below, as well.

 I've ordered the new router.  Joe tells us that the 40m router is a Linksys BEFSX41  , however this model doesn't seem to be available anywhere (unless you count secondhand ones on amazon....) even on the linksys website.  Instead I have ordered the next one up in the range which appears from the specs to be the same router but with 8ports.  It is a BEFSR41.  Should be with us in approx 2-3days as it was in stock.

 [Zach, Koji]

Last week, we did some work towards calibrating the actuation gains of the VCO and NPRO PZT, which was documented in Koji's elog posts. This post summarizes the method.

First, a sine wave with a nominal amplitude of 10 mVpp @ 1 kHz was injected into the PZT sweep of the primary PDH box, and the peaks in the primary error and feedback signals were recorded, as well as that in the secondary error signal (which was being held in the linear response region by the main servo):

• Primary error: 685.2410 uVrms
• Primary feedback: 98.7438 uVrms
• Secondary error: 64.9701 uVrms

Note that the error signals were being read out at INMON, so they are multiplied up by ~42.

Next, the same drive signal was connected to the Tektronix VCO external FM input, with a nominal deviation setting of 100 kHz/V and a carrier frequency of 47.5 MHz (as usual). The output was then connected to the RF analyzer to study its frequency structure and determine the modulation depth.

The structure of an FM signal in frequency space consists of a carrier and a series of sidebands at integer numbers of the modulation frequency from the carrier, whose powers are determined by the bessel expansion. I.e., the nth sideband will be at a frequency f_c + n*f_mod with a power P_n = ( J_n(Γ)/J₀(Γ) )² * P_c, where Γ = Δf / f_mod is the modulation depth and 2*Δf is the peak-to-peak amplitude of the frequency modulation.

The powers of the sidebands were measured (in dBm) and divided by the carrier power, and this array of values was fit to the appropriate array of bessel function ratios to obtain a modulation index and therefore an FM amplitude. Below are two plots, one for the 1-kHz excitation mentioned above and another for the same test done with a 100-Hz excitation with the same amplitude, which is more impressive and used more points.

The inferred peak-to-peak FM deviation was thus ~993 * 2 = 1986 Hz. Since the FG displays the voltage for a 50-ohm load, this value is reasonable as the external FM input has an impedance of 10kohm. So, 0.01 mVpp (*2 for display error) * 10K/(10K + 50) * 100 kHz/V ~ 1990 Hzpp.

This signal was put into the AOM and elicited a peak of 247.258 Vrms in the secondary loop error signal. Since the AOM is double passed, the true optical deviation was 2 * 1986 = 3972 Hzpp. This implies that, when it was put into the PZT sweep input, the same injection caused a deviation

(64.9701 uVrms / 247.258 uVrms) * 3972 Hzpp = 1043 Hzpp.

Using the signal strength in the PZT feedback path from above, we can infer a PZT actuation gain of

G_PZT = 1043 Hzpp / (98.7438 uVrms * 2*sqrt(2)) = 3.734 MHz/V

 Frank and I had to play some games with openmotif/grace/medm to get things going again. Basically, the EPEL repo should be added to all of the linux machines using rpm.

Then, we must install grace from the EPEL and this will also install the correct 64 bit openmotif which is used by MEDM (Motif Editor and Display Manager). As of now, dataviewer

and DTT and MEDM are working on most machines.

I changed over the user/group ownership of the /frames subdirectories to controls for both fb0 and fb1. Lets see if this is correct.

I also updated the firmware on the linsys router, turned off uPnP, and also turned off DHCP (as a test). I've been noticing that what happens

lately is that the connection to the outside world just goes away sometimes and then comes back after a few minutes. This along with

the fact that our bandwidth to our own LIGO network is only 600 kB/s makes me think that the issue is on Larry's side of the table.

 Below is a plot of one particular gyro signal channel along with various electronics noise contributions to the gyro signal in general. The traces are:

1. The secondary error signal with the secondary loop open (but the primary closed, so that the error signal is kept within the linear regime). This gives a linear output proportional to the measured frequency mismatch between the two cavity directions, and is thus a form of gyro signal. Typically, we would lock the second loop and read this out at the feedback point, but it is useful to do this way as an intermediate measure when trying to characterize the loops.
2. Noise at the INMON of the primary box with the laser off, divided by 42 to obtain the level at the true servo input.
3. Same as (2), but for the secondary loop.
4. Noise at the INMON of the primary box with a 50-ohm terminator in place of the PD (i.e. noise in the demod electronics).
5. Output noise of the primary PDH box with the input terminated, divided by the estimated analytical transfer function of the box with G = 0.5 (as when the noise spectrum was measured).

All of these signals were then divided by the cavity response in [V/Hz]---which are all the same except for in (3)---and then multiplied by (lambda P / 4A) to obtain spectra in units of (rad/s)/rHz. The input-referred noise of PDH 2 is not shown, but given the same output level would result in roughly 10x the noise contribution to the gyro signal due to the roughly 10x lower optical response of the cavity in the secondary direction.

now 300mJ (20W and 15ms). Beam diameter still 1mm.

I keep increasing the pulse duration until i see first damage of this already heavily used diode to see where the damage threshold is, then use go back a little bit an use a new, unused diode to see if i can measure changes in parameters.

after ~1.5 the netbook stopped working again. Will switch to the old IBM notebook tomorrow. Jan is using it right now as a backup computer.

Below is a comparison of the "unsuppressed" primary loop frequency noise with the free-running NPRO noise taken from Rana's thesis (p.80). The primary error signal was multiplied by the OLTF to obtain the unsuppressed voltage noise at the servo input, and this was then divided by the cavity response quoted in the previous elog post to obtain the unsuppressed noise in Hz. The plot shows that the gyro cavity displacement noise is ~10-100x higher than the free-running laser noise up to a few kHz.

 By using the measured NPRO PZT actuation gain of ~4 MHz/V, along with an analytical model of the PDH box and primary OLTF, I estimated a cavity response of roughly 5.6 x 10^-8 V/Hz. Using this, I calibrated both the primary error signal and feedback signal into units of frequency to see that they were consistent. The calibrations are

• Feedback signal: multiply by the PZT actuation gain in [Hz/V] and divide by the OLTF (to get the suppressed noise)
• Error signal: divide by the cavity response in [V/Hz]

The plot below demonstrates that they are self-consistent. Note that this is only done up to 10 kHz because I have not made an analytical model of the notch filter yet.

As is the standard mandated by CDS, I have converted the UID for controls to 1001 and the groupID of the controls group to 1001 and made the user controls be a member of the controls group only.

I had done this on ws1/ws2 before and there were naturally inconsistencies with it and fb1. I have just now done it fb1 and then fixed up many files by doing:

chgrp -R controls *; chown -R controls *

in /cvs/cds/caltech/ and in /opt/

There will certainly be short term negative consequences of this action, but its better we do it sooner than later. I'll have to get some Seifert consultation to finish fixing fb0.

netbook was frozen the second time this weekend. Software installations didn't change since a week (since i'm using it for the pulsing experiment).
We have seen this before when using it for the wincam or the particle counter, but very rarely.
Now it happend twice within 24h, once after ~23000 pulses, second time after ~6000 (3h of operation after reboot)

Will restart the measurements...

The network interface for ws1 was failing to work for some reason. I tried the usual Linux forums for advice but most of it was useless.

Finally one of them suggested that there was a warm power up v. cold power up issue with the Realtek 8169 Network card chipset.

So I turned off the machine and then reformatted the disk and did a fresh install. The network is now working (allowing this elog access).

But then...I realized that (someone) gave me a 32-bit install DVD and so I'm now wiping it and starting over.

It turned out that even 10000 pulses of 100mJ (20W, 5ms) are not enough to cause any damage to the diode. Still using the very first diode until i find the threshold.
Then i switch to a new one and turn down the power a little bit.

After testing an old, completely dead diode with a smaller beam i saw damage after 50 pulses having 100mJ and higher (no bias, just bare diode).
I didn't measure the size of the beam at that point (will do that tomorrow) but that indicated that the actual power density on the DUT is still about a factor of 5 to 10 smaller as required for surface damage.
So i increased the the pulse energy from 100mJ to 250mJ to see what happens to the diode over night. As the power maximum is reached by now i had to increase the pulse length (10ms)

The picture below shows the damage to the diode. The two smaller damages are from 100mJ pulses, the larger one 200mJ (50 pulses each).
One can see that the smaller spot size with 100mJ energy barely damages the surface (the actual damage is the small spot in the center, the surrounding part is damaged by mechanical stress indicated by cracks), however the 200mJ pulse creates more damage. The diameter of the active area is 3mm, so the size of the damaged area is ~200um.

attached the schematic of the current setup and a screenshot of the program.

did some test runs the last couple of days to see at which power level something happens to the device performance.
Current devices are the old, already very noisy 3mm GTRAN diodes.
Devices are mounted in OMC heatsink using prototype of OMC invac preamp (100R fixed resistance) and the higher bias voltage of 12V (originally 5V).
DC current on PD before pulsing is adjusted to 30mA, pulse comes on top of that. No additional current limit for bias voltage. Spot size ~1mm.

I measure the following device parameters before each new cycle

• dark current noise spectrum (1Hz to 100kHz)
• I/V-curve to 20V reverse bias
• Impedance, then fitting series resistance and capacitance
• implemented (but not recorded so far): shunt impedance, calculated from slope of I/V-curve from 10mV to -10mV bias (that's how it is measured)

time for each cycle: about 28min.

I started with the following parameters about a week ago

• peak power: 1W
  
• pulse duration: 1ms
  
• number of pulses/cycle: 1000
  
• time between pulses (for diode cooling off): 1s
  
• number of cycles: 50

As nothing changed i increased the peak power step by step until i reached the following parameters today:

• peak power: 20W
  
• pulse duration: 2ms
  
• (rest the same)

so far no visible changes in measured parameters. Also didn't visually damage a device so far. So if i don't see anything tomorrow i will try to physically damage one diode to see where the damage threshold is.
I probably don't see changes in noise performance as the excess noise of these 3mm diodes is about three orders of magnitude higher as the good EG&G ones. Unfortunately we sent all diodes to the sites.

 I have noticed that the elog is taking excruciatingly long to load today. Is this happening to anyone else?

I've been working on getting the model to switch the boost on and off based on the transmission pd signal.  So far I have the software part working but not the hardware.

In terms of software, initially I tried modifying the epics atf1.db file.  I added  a calc channel that compared the trans_pd value to a user input channel from the MEDM screen.  I could get this to switch, but got stuck at the point where I tried to write this back out through the DAC.  While you can use an 'ao' channel in epics to do this, we want instead to write out through the front end.  I set up an epics channel that could write out through the front end, by putting it into the model but got stuck trying to work out how to write the calc channel to the channel that is connected to the DAC.  Making them the same channel didn't work and I couldn't find a way to transfer a value from a calc channel to an 'ai' channel.

Instead I got it to work by putting doing the comparison in the Simulink model.  The trans_pd epics value is compared to a user input MEDM value, and the binary output of this then toggles a switch that sets DAC output 10 high or low.  One nice thing about this is that it doesn't require any manual editing of the .db file.

There is one problem with connecting this up right now.  Frank thinks the DAC output is probably differential (we measure 5v out on the boost controller channel when connected to a scope, but 10v on a multimeter), and if we connect it to a grounded piece of equipment (such as the boost input on the PDH box) we will be shorting the output.  We should take the DAC card out of the rack and check that this is the case.  If so then it seems we need a differential input on the boost switch and everything else that we are controlling from the DAC (such as the slow input on the laser).

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.

 Here is a transfer function from the IN to INMON ports of the primary PDH box (#1437). It is pretty constant up to about 10kHz, where it begins to fall off. The phase lag at 10kHz is ~3 degrees. This is consistent with Koji's LISO modeling of the stage, and it indicates that cascading OP27s may not be a suitable practice for our application.

 It is actually an SLP, not BLP that we have in the primary PDH setup (though I assume this just refers to the connector type, and not anything with the circuitry). In any case, we care about the low-frequency response of the filter at least as much as the high frequency attenuation. Here is an audio transfer function showing that the filter is essentially a straight wire in our frequency band of interest.

Koji pointed out that the PD noise at DC is irrelevant to the demod noise. I have repeated the measurement the correct way. All of these spectra were generated by looking at the "INMON" port of the primary PDH box. They were then divided by the meaning of life (42) to reflect the true level at the "IN" port. The traces are:

1. 50-ohm terminator in place of the PD (demod noise only)
2. PD connected but shutter closed (demod noise + PD dark noise)
3. PD connected, 100 mW incident power, EOM off (demod noise + PD dark noise + shot noise)
4. PD connected, 100 mW incident power, EOM on (demod + PD dark + shot + RFAM from EOM)
5. Measurement noise floor for (1) and (2)
6. Measurement noise floor for (3) and (4)

Two things of note:

• There is a large amount of RFAM noise (yellow trace) that needs to be minimized. This might explain a lot of our noise coupling.
• 100 mW is not sufficient for shot noise to surpass the dark PD noise. This is the highest power I was able to put on the diode downstream of the initial beam split even when putting all of the light through one of the two paths.

I agree with Koji's assertion that our first priority should be replacing the REFL PDs we have in place at the moment. We will likely go back to the PDA255 for the primary loop while we wait on the custom RFPDs.

I've made a second MEDM screen for the gyro.  It is more of an indicator screen showing which signals we are aquiring from which part of the gyro.  It has a few charts that we can modify as we think of things that we want displayed (it shows just photodiode DC values and actuation signals at the moment).

I added a matrix to the TEST channels and incorporated that into the C2ATF_GYROMAIN.adl

I've spent a while routing cables.  We now have all the correct channels attached to the ADC and I've labelled them at the gyro end.  I ordered a couple more patch bays so that each bench can have two.  Frank is going to pass on the info for the BNC cables that are used to connect the patch bays - they were custom lengths and he has the original quote.  We'll buy enough new ones that we can route 2*16 channels for both benches.

while searching for optical properties of Tantala thin films i found the following web page to do all kinds of conversion and calculations. So check it out ...

http://www.luxpop.com

1126

- The input gain of 32.5dB=42 is good.

It seems that the amplifier starts loosing gain above 10kHz. I wonder how much phase delay we get at 10kHz.
==> We need the phase measurement too.
This may indicate that op27 is not adequately fast for our application.
A quick calculation with LISO shows the phase delay of 3deg at 10kHz.

r r1 25 gnd nm
r r2 1k nm no
op op1 op27 np nm no

uinput np 0

# uncomment this for the transfer function
#uoutput no:db:deg 
# uncomment this for the noise
noisy all
noise no sum all

gnuterm pdf

freq log 100 1M 400

1127

We need the TF measurement with phase.

1128

Measurement noise floor is too high. You could use the input monitor (OP27 G=41~42) as a preamp.
The above LISO simulation showed 3.3nVrtHz. This is already pretty good preamp. 

I don't understand the dark PD noise. Is it demodulated signal? or DC noise (which is not relevant to us)?

The PD demodulated dark noise of 30nV/rtHz does not agree with our measurement last week (1uV/rtHz => 24nV/rtHz).
Probably the noise level of the FFT is affecting the measurement. USE any PREAMP

We just need the noise levels of the input monitor output in the following configuration:
- The cable terminated (Noise of the demodulation system)
- The PD connected with no light (above + noise of the PD)
- The PD connected with nominal amount of DC light - no cavity / no modulation (+shot noise)
- The PD connected with nominal amount of DC light - no cavity / with modulation (+modulation RFAM noise)

1129

The result is incomprehensible.

Use DC photocurrent (I_DC) and output noise level (V_out) for the horiz and vert axes, respectively.
For V_out, pick the noise level appropriately at an uncontaminated frequency.
Fit this curve by the model V_out = g_det sqrt(I_det + I_DC), where the parameters g_det and I_det are the equivalent transimpedance
and the equivalent noise current. g_det is a kind of gain, and I_det gives you how much DC current do you need to make the shot noise
and the system noise comperable. (i.e. You must put more light than I_det)

And the I_det is the order of the sub-mA for the low noise PD. You need to put ~10mA (= 500mW) to really test the PD shot noise.

AndI am afraid that this is not the demodulated noise.

1130

What to do

> Remeasure the cavity response (V/Hz).

No.

What we need is the precise PZT calibration and precise OPTF modeling.

Once the fast PZT is calibrated and the openloop TF is measured, we can infer the applied frequency disturbance.
Then we can easily deduce the cavity response (V/Hz).

MENU on Wednesday

1. Re-confirm the validity of the Tektronix VCO's purported external FM deviation setting.

Zach has the previous measurement but the value depends on how much output impedance the external source has.
So we test the frequency shift with the excitation we used for the IFO calibration.

2. Try to find two nice equivalent PD and characterize them.

3. Koji measures the circuit TFs while Zach tries to obtain noise plots.

4. Realignment, open loop TF measurement, CDS hooking up, etc... if possible.

1.  Remeasure the cavity response (V/Hz). I thought I had this from when I measured the finesse, but I realized that I did this with the transmitted power and not the error signal, so I don't have the information I want. Once I have this, I can do several things, including:
   • Convert all the spectra Koji and I measured last week into gyro signal units (rad/s/rHz)
   • Divide the primary error spectrum by the OLTF of the loop to determine the unsuppressed frequency noise level and compare it with the expected free-running laser noise
2. Remeasure the PDH box TF using the SR785, then
   
   • Put the new plot on the wiki
   • Refer the measured input-shorted PDH box output noise to the input to determine its contribution to frequency noise, compare this with LISO
3. Re-confirm the validity of the Tektronix VCO's purported external FM deviation setting. I did this for DC frequency shifts using a calibrator in ELOG 1090, and yesterday I double-checked it by putting the frequency-modulated signal into the RF analyzer to see that the sidebands were at the correct frequencies, but I realized that I was looking around 95 MHz instead of 47.5 MHz, so I was really looking at sidebands on the carrier harmonic. Either way, the frequency spacing was exactly the 1kHz it was supposed to be, but I will make the correct plot and put it here.
4. Connect the decided-upon signals to CDS so that we can systematically monitor them at any point in the future. Alastair has put the finishing touches on our new-and-improved front end code, so this should be ready to go.
5. Do a thorough realignment of the cavity in both directions and record the improvement in transmission, then make new measurements of the error signals and OLTF. I was originally going to do this first, but I thought it might be wise to do all the other junk I have done first while Alastair finished up with the front end. 

 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.

 NOTE: The plot below is somewhere between misleading and wrong. See reply.

Attached is a plot showing the following (all of these were taken with the laser shutter closed in order to focus on electronics noise):

• Voltage noise out of the mixer with the PD connected to it
• Voltage noise out of the mixer with a 50-ohm terminator in place of the PD
• Noise directly from the PD. Note: the raw data for this was divided by two to account for the change in transimpedance gain when going from Hi-Z (Agilent) to 50-ohm (mixer)
• Measurement noise floor

The "mixer output" here is actually the output of the lowpass filter after it.

The noise from the demodulation seems negligible in comparison to the dark PD noise, with the exception of a broad peak at around 49 kHz that appears to come from the mixer.

 Attached is the attenuation plot for the BLP-1.9+ we are using after the mixer from the spec sheet.

 I have updated the schematic for PDH box #1437 (primary loop) and put it in the gyro wiki. I have not been able to take a new transfer function with the Agilent, as it seems to be too noisy for some reason. I will use the SR785 (which has a higher input range) to make one ASAP. In the meantime, the general shape can be found in ELOG 1080 (this TF would be fine but the gain setting was not recorded at the time).

I also measured the change in gain as a function of the dial setting, which is linear in dB (as it should be given the specifications of the AD8336 variable gain stage it is connected to). I did this by injecting a sine wave @ 1kHz with the FG and measuring the amplitude at the output. For some reason, the results I got seemed unreasonably high (~112 dB with G = 0.5, whereas the TF linked above shows ~40 dB @ 1kHz, which would be unattainable for any gain setting given my measurement). It is possible that I was driving with a higher amplitude than I thought; I will repeat the measurement tomorrow).

Finally, I verified that there is a factor of ~40 gain from the INPUT to INPUT_MON ports on the PDH box, as seen below.

I've rebuilt the C2ATF front-end and renamed all the channels for the gyro as well as changing the topology.  I started a list of channels on the wiki here.

I have made one new medm screen for us to use to start with.  It is called C2ATF_GYROMAIN.adl  (see screenshot below).

Most of the channels now just come in and go to a filter so that we can aquire them.  The laser piezo actuation signal does go back out again through channel 9 so we can act on the slow feedback.  I have added 4 general channels that come in and go back out, just in case there are things we wish to do that I haven't thought of yet.

The front-end is running happily for the moment.

The 35W laser is on again.

Moved all equipment from EE lab back to the ATF and currently doing some tests. 
All the equipment next to the 35W laser table (near lab corner)  is in use  (automated measurements). 

Plz don't touch those things

 I am wiping the floppy disk entitled "Alastair's Agilent traces 1", which I know that several of us have used to transfer data from time to time. I have temporarily backed it up on my hard drive, but if no one claims any data I will delete it.

 Don't ask me George, I have no idea.  We now have a huge door sized white board though, and a sort of fake wall covering up the door.  I'm just assuming that at some point the old one will be gone since it no longer has a wall to live on.  Any more questions?

[Zach / Koji]

Other things to do:

Assemble the noise budget plot

• Put every possible thing into the noise budget plot
• Propose other noise measurements

Tuning of the secondary loop

• Take the thorough alignment of the beams
• Record the improvement of the transmited power by the alignment improvement.
• Adjust the openloop gain of the secondary loop
• Measure the new error / feedback signals.

CDS preparation

- At least the following stuffs for each loop are needed to be connected to the CDS system in order to make the characterization continuously:
(ADC)

• The error signal (or input monitor)
• TP2 or TP6 (= the signal before the summing point)
• output monitor (= the signal after the summing point)
• Reflection DC
• Transmission DC
• and Gyro beat signal

(DAC)

• PZT sweep input

- Consider the whitening filters to amplify the signal above the ADC noise level.

Circuits

- Confirm the circuits and make the latest circuits diagram.

- Measure the latest filter TF. If we already have it, put the numerical data / the liso model somewhere.

- Measure the input equivalent noise of the filter. 
i.e. short the input of the filter box measure the outupt noise of the box (or the notch). Divide it with the TF to convert it
to the input equivalent noise.

- Measure the demodulator noise levels.

- Shot noise characterization: put DC light onto the PDs and measure the demodulator noise with various DC powers.

It is done for now.

[Zach / Koji]

Calibration of the laser PZT drive

Method:

- Lock the primary cavity.
- Inject the sinusoidal signal from DS345 to the PZT sweep input. The frequency was 1kHz. The amplitude display showed 10mVpp(@50Ohm).
- Look at the peak in the PZT feedback and the error signal.
- Also look at the error of the secondary loop. (*)

- Disconnect VCO feedback and connect DS345 to it. This gives us the calibration signal with known amplitude.
- Look at the error signal of the secondary loop. Compare it with the above (*)

Result:

The result of the signal injection into the primary loop

- The primary loop error showed the peak amplitude of 685.2410 uVrms
- The primary loop feedback showed the peak amplitude of 98.7438 uVrms
- The secondary loop error showed the peak amplitude of 64.9701 uVrms

We put the same signal into the VCO control. (Note: It is not necessary to be the same amplitude.)

- The secondary loop error showed the peak amplitude of 247.258 uVrms

Strangely the primary error signal also showed 0.808uVrms peak, which should not appear in the primary error.

Thought:

Quick calculation

- The response of the VCO driver is 100kHz/V. Assume this has 50Ohm input (TO BE CONFIRMED). The VCO drive has 10k input impedance. => 10K/(10K+50) = 0.995
- This means we shook the VCO frequency by 100kHz/V * 10mVpp * 2 * 0.995 = 2.0kHzpp
- The actual AOM changes the laser frequency twice because of its double pass configuration. ==> 4.0kHzpp ...(a)

- (a) yield 247uVrms, while the PZT injection did 685uVrms 64.97uVrms. ==> 2.8 times larger. 4.0kHz x 0.26 = 1.04kHzpp ...(b)

- PZT actuation of 99uVrms (=280uVpp) caused this (b). ==> 3.7MHz/V

To Do:

• Confirm the above logic
  • How much was the VCO frequency actually swung?
    Calibrate the VCO driver again by putting the same signal into the VCO and look at the VCO output with the RF analyzer.
• Calibrate the primary and secondary error signals. How much V/Hz do they have?
• What cause the strange signal coupling from the secondary cavity to the primary cavity?
  • Is this optical, or electorical?
  • Shake the VCO freq and observe the error signal of the primary. Block the beam of the secondary at various places.
    (i.e. just before the faraday, just after the AOM, before the AOM etc...) Try to understand what is happening?
  • Is this coupling harmful?

[Zach / Koji]

Measurement of the secondary (CW) cavity error signal.

Method:

- Lock the primary cavity.
- Tune the VCO frequency such that we can transmit the secondary (CW) beam to the transmission CCD. Tune the error signal to zero.
- Leave the secondary loop open. Look at the error signal of it in order to determine whether it is in the linear range or not. (Yes it was)
- Measure the error signal while the loop is open or closed.

Result:

- The control bandwidth seems to be ~100Hz. The control gain seems to be ~3.
- The error signal seems to be completely dominated by the dark noise of the detection system (= PD+demodulator).

Thought:

- The gain and the bandwidth are way too low to obtain the gyro signal from the VCO feedback
- If you look at the structures above 100Hz, the optical gain is about ten times smaller than that of the primary loop.
  What the heck is and the bandwidth are way too low to obtain the gyro signal from the VCO feedback
- Note that the signal is supposed to be amplified by a factor of +41. This means the dark noise floor level is ~25nV/rtHz.

To Do:

• Calibrate these spectra in Hz/rtHz.
  • How much is the optical gain in V/Hz or V/m.
  • Overlay calibrated spectra of the primary and secondary loops.
  • There looks excess noise below 10Hz. We must reveal what it is through the measurements.
• This error signal is the gyro output signal as far as there is no feedback.
  Calibrate this signal to (rad/s)/rtHz (or rad/Hz). Put this measurement on the noise budget plot.
• Convert the dark noise measurement into the noise budget of the Gyro signal.
• Measure the control loop gain. Investigate the shape of the loop up to 100kHz.
   
• Measure the shot noise level
  • How much DC power do we typically have? (Actually we don't have the record during this measurement, but we can measure it again)
• Measure low frequency spectra with CDS
• How to improve the noise floor
  • Understand why the optical gain is so low.
  • Is this noise level disturbing in terms of the gyro requirement? How about in the low frequency?
  • How much is the demodulator noise? Put a 50-ohm terminator on the cable instead of the PD. Then measure the same signal.
  • How much is the gain of the input stage. It is supposed to be +41, but we are not sure.
    => Think about the noise level at the demodulator output.
  • How much is the noise level of the PD measured by an RF analyzer? Is it consistent with the above analysis?
  • Do we need a new resonant RF PD? How much should we improve the noise level with the new PD? And how much noise does the new one actually have?
• Where the VCO frequency noise comes? ==> Measurement of the VCO noise.

[Zach / Koji]

Measurement of the primary (CCW) cavity error / feedback signal.

Method:

- Lock the cavity. Measure the spectrum of the input monitor output.
- Close the laser shutter. Measure the same spectrum.

- Measure the feedback signal after the notch filter.

Result:

- The global shape of the error signal is a kind of flat. Though it's spiky due to mechanical resonances.
  Below 10Hz it got smoother but the actual shape is not obvious because of the rough resolution.
- At around 100Hz, the error is below the dark noise. This means the out-of-loop stability does not go below the dark noise level.

Thought:

- The broad peak at around 10Hz is the servo bump.
- Note that the signal is supposed to be amplified by a factor of +41. This means the dark noise floor level is ~25nV/rtHz.

To Do:

• Calibrate these spectra in Hz/rtHz.
  • How much is the optical gain in V/Hz or V/m.
• Convert this measurement into the noise budget of the Gyro signal.
• Make the unsuppressed spectrum
  • => How much cavity length fluctuation does the error signal feel if there is no feedback and the sensor is infinitely linear?
  • This requires the model of the open loop TF
  • Compare the consistency with the same quantity derived from the feedback signal
  • Compare the unsuppressed frequency noise spectra with the free running laser freq noise (see rana's phd thesis P.80).
  • Convert the spectrum into the displacement noise (m/rtHz) and compare with the displacement on the table.
• Measure the shot noise level
  • How much DC power do we typically have? (Actually we don't have the record during this measurement, but we can measure it again)
• Measure low frequency spectra with CDS
• How to improve the noise floor
  • Is this noise level disturbing in terms of the gyro requirement? How about in the low frequency?
  • How much is the demodulator noise? Put a 50-ohm terminator on the cable instead of the PD. Then measure the same signal.
  • How much is the gain of the input stage. It is supposed to be +41, but we are not sure.
    => Think about the noise level at the demodulator output.
  • How much is the noise level of the PD measured by an RF analyzer? Is it consistent with the above analysis?
  • Do we need a new resonant RF PD? How much should we improve the noise level with the new PD? And how much noise does the new one actually have?

[Zach / Koji]

Measurement of the primary (CCW) cavity openloop TF.

Method:

- Inject the actuation signal from the PZT sweep input
- Take after-the-sum signal and before-the-sum signal from TP6 and the output mon, respectively.

Result:

- UGF: 10kHz (this is OK)
- Phase margin: 14deg (this is too small)

Thought:

- The phase compensation between 5k to 20k is not enough.
- The phase delay above 10k seems too big. Possibly come from not-so-fast opamps (op27)?

To Do:

• Make the model of the openloop tf in alll frequency range (1mHz-100kHz).
  • The cavity cut-off has already been measured fc=98kHz (elog 9062)
  • Measure and characterize the low pass (fc=1.9MHz) just after the demodulator
  • Update the circuit diagram to show the latest condition
  • Make the filter transfer function with dial gain of 2.5
  • Check the gain dependence on the gain dial
  • Measure and characterize the notch filter (utilize earlier measurements?)
  • What is the calibration of the laser fast PZT and the transfer function?
  • Is the low frequency boost useful? (The boost may eat up our phase margin and make the servo unstable.)
  • How to construct the model of the temperature loop?
• Design and implement better phase compensation.
• Make a new custom servo circuit.

 Why did we switch whiteboards?

Why would we let them throw this one away?

RIP Old Whiteboard

I found the old white board and got a photo of the info before the goons who removed it can throw it away.

As a tribute to the old whiteboard, I think that Aidan in particular will appreciate this

About two days before this elog post (I am a bad grad student), Koji and I took some more measurements in a continuing effort to fully characterize the CCW (main) loop.

Some things we did:

• Open-loop transfer function
• Error signal spectra
• Calibration using a 1-kHz sine injection

The last two were also done for the CW (AOM) loop, with the second also done with only the CCW loop closed to see the "free-running differential-mode" noise. We noticed at least two interesting things:

• The CW noise suppression was far from stellar at low frequencies and non-existent above a relatively low frequency (which escapes me at the moment)
• The 1-kHz calibration line was partially visible in the CCW error spectrum when the excitation was on the AOM VCO. This indicates either electrical cross-coupling or more likely backscatter noise at the cavity input (which surely got worse after we put in windows so that theta = 0 deg). On that note, we may eventually find it necessary to use a different modulation frequency for each of the two directions, but it's certainly not clear that that will be necessary yet.

Since the new front end is not quite up to speed (and due to bandwidth limitations), all of this was done using the Agilent. Due to Koji's far-superior Agilent prowess, he did most of the button-pushing, and he will give me the data once he offloads it so that I can process it.

I found someone working without safety glasses yesterday evening. I told him he couldn't work without them and he said he was done for the day, but he asked me to see that the lasers were turned off so that he could work early this morning. I spoke with Frank and Koji about it and we decided that we would turn the lasers off to be safe but let the people above us know what happened. We wrote that the lasers were off on the whiteboard in the changing room.

The front-end code is working again now on FB0.  Many of the folders were owned by root because of some previous badly written documentation - these have all now been chown'ed into submission.  The main lesson learned is: DON'T EVER MAKE THE MODEL AS ROOT!

We now have a really simple route to re-making the model when we have any changes.  This is documented on the ATF Wiki here. Please don't deviate from this process ever.

Dataviewer now works on WS2 (though doesn't work from mainmenu because it doesn't start up grace).  The gyro MEDM screens aren't working right now because I changed some of the parts names in the model.  I also plan on hijacking a few more channels for the gyro before remaking the model again.  There is still an issue with the user accounts on FB0 which needs fixed at some point in the future.

Most importantly we now have a Rana approved color scheme for the terminal shell on WS2.

if the boxes of the main screen are white the daqd process will not start. It automatically stops if the frontend code isn't running as it is configured to wait for the right DCU to be up (in that case the only DCU in that system IS the frontend, so it can't be started). Any messages while compiling the stuff? Any errors?  My suggestion: save the current simulink model in a save place, take the last, known to be working version and do all the compiling and see if everything starts. If yes you have a bug in your model, if no the way you compile and start the stuff is wrong.

Once the frontend is running, if the daqd still isn't starting try checking the logfile of the daqd process. it's in the /fb directory. Usually it tells you exactly what's wrong...