NAOJ GW Elog Logbook 3.2
The transfer functions of AA loop was measured with swipe sine method and reported in elog2230. But the signal at high frequency was not quite clear. Therefore, Eleonora suggested me to measure with fft method. The excitations were uniform noise and sent to each degree of freedom, their amplitude are
| Input_yaw | 3000 |
| Input_pitch | 10000 |
| End_yaw | 2000 |
| End_pitch | 2000 |
Note that if the excitation is not large enough, the coherence between these two signals will be exactly 1 at almost all the frequencies.
The measurement result is shown in the attached figures.
The air conditioner was switch to warm up mode last week. After that, I found green injection direction was changed a lot. The green beam references on the PR/BS chamber are fine (as shown in the figure 1). But the green beam was quite far from good height on the first target in the arm (as shown in the figure 2 and 3).
Therefore, I checked the PR/BS oplev signals (as shown in the figure 4). From this figure, we see that mainly PR pitch was moved a lot. Additionally, BS pitch also moved in the direction. This movement is so large, so I used picomotor to recover PR position. After recovering PR pitch, BS was moved up with oplev. With good alignment, the green beam on the first target is shown in the figure 5 and 6. However, after locking FC, the GR DC tra was found to be only around 3000.
I checked green power before AOM, it was only about 30mW, but the nominal value should be 52mW. So I changed SHG temperature from 3.096 to 3.068. Note that 3.068 makes SHG have a local maximum green output, which gives 52mW before AOM. The AOM modulation depth is 5.5dBm, which gives 14.5mW injected into FC.
Besides, recently we have SHG output oscillation problem. This is due to the gain of SHG servo is smaller than 2.25. After I put the SHG servo gain back to 2.25, oscillation disappeared.
[Aritomi, Yuhang]
We measured CCFC RF noise (before demodulation after amplification).
Setting of spectrum analyzer: RBW: 300kHz,VBW: 10kHz
We used 50:50 BS to pickoff CCFC error signal and pump green power is 42 mW.
Note that CCFC and CCFC FC unlock were measured with max hold while others are normal setting. I think that's the reason why the floor noise level in CCFC and CCFC unlock are a bit higher than TAMA PD noise level.
What I did
In order to confirm the molecular layer formation at cryogenic temperature, I started another measurement with modified spacer.
The spacer has a slit to let residual gas molecules impinge on the folding mirror.
The incident angle is about 22.5 deg.
I installed it and pumped down.
Then I did some measurements to characterize the mirror property.
I turned on the cryocooler at 16:08.
Results
The finesse of the cavity was about 22000, which is better than before.
It seems that the beam hit the defect before.
At the defect, several tens ppm optical loss exists in folding mirror.
The frequency shift between p- and s-pol was about 20 MHz which is almost consistent with the theoretical calculation.
This measurement enables characterization of coatings.
Next step
I will monitor the finesse and splitting frequency for one or more weeks.
I monitored PR/BS oplev signal, INPUT/END AA correction signal, SHG transmission, Filter Cavity GR transmission, Filter cavity IR transmission, Filter cavity IR detuning and BS pointing for 200 minutes. The result is shown in the attached figure.
If we check the shape/peaks of FC IR detuning and BS pointing pitch, the correlation between them is quite obvious.
However, no clear correlation was found between PR/BS motion and INPUT/END correction, which could be due to the motion is always making effect with the combination of two mirrors.
In this measurement, we assumed that green power is 56mW and generated squeezing is 21 dB, but nonlinear gain was not optimized in this measurement and actual nonlinear gain (or generated squeezing) should be lower. I assume that generated squeezing is 16dB in this measurement and fitted the FDS measurement again (attached picture). In this case, homodyne angle changes from 0 deg to 90 deg and detuning is between 89-98 Hz. The detuning fluctuation might be better with CCFC.
[Aritomi, Yuhang]
We tried to measure FDS with CCFC with 42mW green. MZ offset is 4.5 and p pol PLL is 135 MHz for 42mW green. For 18mW green, MZ offset should be 4.2 and OD0.2 should be placed in pump injection path.
SR560 for CCFC: gain is 1000 and filter is 0.1Hz 1st order LPF.
We found that we can change the squeezing angle with CCFC, but with squeezing quadrature, the shot noise level is -123.8 dBm which is about 10 dB anti squeezing (attached picture). It seems CCFC makes the squeezing level worse a lot.
As it was found that AA loop is introducing locking accuracy problem (elog2227), and I also tried to reduce the coupling from y/p to l(the improvement is mainly the coupling from y to l) (elog2229).
After that I did the comparison in the following three different situations:
1. No AA loop but local control engaged for input/end mirrors.
2. With the same AA loop (compared with elog2227)
3. With AA loop but the gain of yaw control was increased by a factor of 10 for end mirror, the gain of end/input mirror pitch were increased by a factor of 2.
The comparison is shown in the attached figure 1.
1.According to the transfer function of INPUT/END pitch (figure 2, 3), it seems the resonance around 8 and 9Hz is feed back a lot. But according to the transfer function of INPUT/END yaw (figure 4, 5), it seems the resonance around 1.45Hz is only obvious in the orange line.
2. By checking the noise spectrum of each mirror motion, the new peak in orange line around 6.5Hz should come from the Pitch/Yaw of PR.
3. By checking the noise spectrum, the orange ~3Hz peak's reduction may be due to END mirror pitch is controlled better. The orange ~5Hz peak's reduction maybe due to INPUT mirror pitch is controlled better.
Note that: the green and blue lines were measured within almost 5mins. But the organge line was measured almost 4hours later.
The setting of AA loop is as follows:
|
|
Poles |
Zeros |
Gain |
|
END_pit |
0, 20(Q=1) |
2(Q=1) |
2 |
|
END_yaw |
0, 20(Q=1) |
2(Q=1) |
20 |
|
INPUT_pit |
0, 20(Q=1) |
2(Q=1) |
0.3 |
|
INPUT_yaw |
0, 20(Q=1) |
1.3(Q=1) |
2 |
|
|
Corresponding filters are shown in attached figure 1 to 4. |
||
Corresponding filters transfer functions are shown in the attached figures 1-4 (order: end_pit, end_yaw, input_pit, input_yaw).
With this loop, I also measured open-loop transfer functions. They are shown in attached figure 5-8 (order: end_pit, end_yaw, input_pit, input_yaw). They have unity gain frequency shown in the following table.
|
END_pit |
0.64Hz |
|
END_yaw |
6.98Hz |
|
INPUT_pit |
0.17Hz |
|
INPUT_yaw |
0.43Hz |
By looking at the spectrum of filter cavity lock correction signal, at the same time, driving pitch/yaw of INPUT/END mirror, the coupling to length could be seen as a peak. By adjusting driving matrix, this coupling could be minimized. This was done for INPUT y/p and END y/p. For each DOF of cavity, the excitation was examined with different driving matrix.
The excitation and spectrum of one test is shown in attached figure 1 and 2. By testing many different driving matrix, I did following modification to reduce coupling to length.
INPUT mirror H4: change from -1 to -0.5
END mirror H1: change from 1 to 0.7
This modification may help to solve the problem of worse locking accuracy caused by AA loop, as reported in elog2227.
Aritomi and Yuhang
The AA loop helps to stabilize both GR and IR beam alignment (elog 2226), but is currently introducing length fluctuation which results in a worse locking accuracy (elog 2227).
With AA loop closed, FDS is measured with detuning of ~550Hz to check if detuning is also stable for FDS measurement. (the whole set-up is the same with the one we used in this Feb)
All the measurement was done in a time scale of 20 minutes. The result is shown in the attached figure.
1. The fit result of detuning is changing within 12Hz.
2. The maximum measured squeezing is about 2.5dB.
3. FC could change anti-squeezing(~7dB) into squeezing(~1dB).
[Aritomi, Yuhang]
We measured IR locking accuracy with/without AA (attached picture). IR lock accuracy with AA is 6 Hz and it is larger below 10Hz compared with no AA. Since the control bandwidth of AA is below 1Hz, mirror resonance above 1Hz is not damped with AA.
Eleonora (remote) and Yuhang
We found GR_tra becomes very stable after engaging AA/pointing/z_corr loops. Before checking FDS, the BAB was tested today. This kind of test was done long time ago (elog2049), we bascially follow the same method.
1. I put BAB on resonance. However, I notice later that PLL was unstable and unlocked at some point. Apart from that, the IR transmission and detuning was stable.
The second trend and minute trend data is shown in the attached figure 1 and 2.
We could see that GR_tra was not stable, which is due to SHG is not stable at the beginning.
The calibration is done with the method in elog2035, the calibration is shown in attached figure 3.
2. I put BAB in a detuned case. The second trend and minute trend data is shown in attached figure 4 and 5.
We could see that both IR_tra and IR_demo are stable in the detuned case.
Eleonora (remotely) and Yuhang
It was suspected that green transmission power changes due to different green beam hitting position on transmission camera (elog2220). Therefore, we decide to fix the green transmission direction. Eleonora has already designed the feedback loop to control the green transmission direction. So I just used that loop.
1. I tried to close the loop with gain of 1. Then I get result of attached figure 1.
From this test, we could see that the loop works. But it seems the gain is too small.
2. I increase the gain to 50 for pitch and 10 for yaw. Then I get result of attached figure 2 and 3.
We could see that green transmission is kept stably.
But from the last entry, it seems to be better to feedback to PR.
Eleonora (remotely) and Yuhang
As suggested by Matteo, we should check oplev singal of INPUT/END mirrors. Besides, as suggested by Raffaele, the control bandwidth of AA can be evaluated by checking how long time loop takes to go back to be aligned from misalignment. Therefore, I did the following check:
1. I lock FC and close AA loop. The locking strategy for AA is: 1.lock with low pass filter 2.switch off low pass filter 3.switch on integrator.
2. By adjusting BS, beam moves up on camera. Then I get oplev error signal as attached figure 1.
By looking at INPUT pitch oplev signal, we could see that AA loop takes almost 4 min to recover the pitch misalignment. Therefore, the control bandwidth should be very small for AA loop pitch. Therefore, it maybe better to increase the gain.
3. I monitor PR/BS/INPUT/END mirrors oplev signal and sum/diff signal of FC_tra_GR PSD for about 40 min. They are shown in attached figure 2, 3 and 4.
According to elog1874, PR/BS oplev signals have similiar calibration factors. Therefore, the attached figure 2 shows each oplev signal in a similar scale, and we could see that PR pitch moved more than others during half an hour. From attached figure 3, we could see that INPUT/END pitch moved accordingly. For yaw direction, there is also correspondence between PR/BS and INPUT/END. But END yaw is always osillating with a frequency around 0.05Hz (time scale around 20 sec). From the attached figure 4, we could see the effect of beam motion in pitch direction. The diff singal from PSD kept chaning in pitch direction (which is consistent with the motion seen from camera). Although the AA loop kept the alignment between beam axis and cavity axis, the green transmission power is not kept.
4. I increase gain of AA loop by 100 times for both INPUT and END mirror.
5. I move BS in pitch direction again. Then the 4 mirrors oplev signals and transmission PSD signals are shown in the attached figure 5, 6 and 7.
From attached figure 5, we could easily see the movement I did for BS pitch. From attached figure 6, we could see that INPU/END pitch follows the change of BS. Especially, after increasing the gain of AA loop in pitch, the recovery of alignment becomes quite rapid. This means the control bandwidth should be larger now. From attached figure 7, we see the beam motion on PSD. But it is a bit strange that there is also signal in yaw direction.
The large motion of mirror after closing AA loop is due to the large gain. This large gain was causing an over shoot. By adding a ramp time of 10min, the lock becomes better. As shown in the figure 2, there is still over shoot, but mirror comes back to the good position within two minutes due to the small increase of servo output.
You can see the ramp time is 600 seconds in the figure 1.
Aritomi and Yuhang
Last week, we closed Auto-alignment (AA) loop and monitor FC_tra for half an hour. As shown in the attached figure 1, both FC_tra_GR and FC_tra_IR drifts.
With the AA loop closed, I locked FC while transmitted GR hits on different positions of Camera_tra (different alignment). As shown in the attached figure2, FC_tra_GR changes according to different positions (alignment). Therefore, although the incident beam axis aligns with the cavity axis by the AA loop, there is different cavity behavior for GR_tra for different alignment which is causing the long time drift of FC_tra. Probably the situation will be even more different for IR, which should be checked soon. For example, it is similar to the FC optical losses for IR is different for different alignment conditions.
Can you plot the oplev signals for input and end mirror at the same time of the plot you attached in the entry?
Aritomi and Yuhang
All the loop used integrator with gain of 0.1. The pitch loop has both 10Hz low pass filter as well. These filters were chosen a bit randomly. They will be improved later.
With these filters, the loop was closed. And the alignment of filter cavity was kept as shown in the attached figure. We could see that the green power goes to maximum after the loop is closed and stays stable. Besides, all the error singal goes to zero and oscillates around.
However, there is still problem. Everytime when the loop is closed, there is a very large motion of mirror. We need to solve this problem. Apart from this, the loop works well.
Can you plot the oplev signals for input and end mirror at the same time of the plot you attached in the entry?
The large motion of mirror after closing AA loop is due to the large gain. This large gain was causing an over shoot. By adding a ramp time of 10min, the lock becomes better. As shown in the figure 2, there is still over shoot, but mirror comes back to the good position within two minutes due to the small increase of servo output.
You can see the ramp time is 600 seconds in the figure 1.
Aritomi and Yuhang
The transfer function of each control loop was measured. The measurement was done with the excitation and the channel before and after that.
The control loop for yaw has unity gain frequency around 0.2Hz.
But the control loop for pitch has unity gain frequency less than 0.1Hz.
