As mixers need to be operated in saturation mode, I temporarily take the amplifier channel used for IRMC demodulation to amplify the signal from DDS. DDS provides about -6dBm signals. With the mentioned amplifier, LO was amplified to about 5dBm. At the same time, the RF signal was about -15dBm.

When we scan the CCFC phase with a sine wave, the demodulated signals will deviate from sine wave if the demodulation process has problems. So I did this test with 5dBm LO (shown in attached figure 1) for different RF power (-15dBm, -9dBm, -6dBm and -3dBm). These tests are in attached figures 2 to 5. All these figures seem to provide good shape demodulated signals (sinusoidal). From these figures, we could also see that the pk-pk signal also increases with the increase of RF power almost linearly (115mV, 206mV, 288mV, 380mV).

I also checked the CCFC error signals for these cases(figure 6,7,8). They are consistent with the error signals we found in elog2308. And apparently, better SNR is achieved with -3dBm RF power.

(We could add 12dB+12dB+3dB attenuator for the -3dBm signal to simulate a factor of ~25 decreases of CCSB power)

I fitted the measured CCFC error signal by fitting the CC detuning (with respect to carrier), demodulation phase, and starting time (first plot). In this plot, misalignment effect is not considered.

In the second plot, I added the misalignment effect in theoretical curve by fixing the mode matching to 94%.

Matteo and Yuhang

The suppression of filter cavity length noise provides stable detuning, which is vital for the production of frequency dependent squeezing. The CCFC control loop is designed to achieve this goal.

To understand better how CCFC control works, several characterization works have been done recently. They are listed as follows:

1. Figure 1 shows many length error signals and noise curves. The addition of CCFC error signal introduces length noise for GR loop at low frequency. This is validated by figure 3 and 4. The GR+IR error signal doesn't change because the filter cavity length change doesn't change.

2. Figure 2 shows correction signals. For the correction signals send to the main laser or end mirror, they are the same whether there is CCFC or not. This is consistent with the unchange of IR+GR error signal.

3. Figures 3 and 4 show FC GR TRA/REF DC spectrums. CCFC causes the GR length noise increase, which translates into intensity noise. 

In elog2231 and elog2267, a worse locking accuracy was found to be caused by AA.

Today I compared the FC_IR_TRA while AA is on or off. It seems AA induced noise increase doesn't have the same shape with FC length noise (but similar).

This noise increase is clearly visible but could be well suppressed if CCFC lock is implemented.

12dB attenuator was added for RF signal (before the 32dB amplifier)

12dB attenuation was applied to LO signal (DAC current control was reduced from 1/2(-12dBm) to 1/8(-24dBm))

Current RF amp: -15dBm

Current LO amp: -24dBm

[Aritomi, Yuhang, Matteo]

We found that we still had saturation problem of CCFC RF and LO so we reduced them.

Then we measured CCFC error signal with different CCFC demodulation phase (Pic. 1). AOM FM freq is 300 mHz and deviation is 2kHz, so AOM scan speed for IR is 4kHz/(5/3 s)/2 = 1.2 kHz/s. CCFC amplitude for normalization is 28 mV. The calibration factor of CCFC error signal is determined by fitting the blue curve around 0, which is 1191 Hz/V.

We locked CCFC with 70 deg and 250 deg CCFC demodulation phase (both are I phase, but sign is opposite) and compared the locking accuracy with CCFC lock (Pic. 2). We found that CCFC locking accuracy with 250 deg is smaller than 70 deg above 1kHz. Changing CCFC demod by 180 deg means that CCSB on resonance and off resonance are swapped. CCSB noise on resonance is filtered out by cavity pole while the noise of other CCSB is not. If noise of upper/lower CCSB are different, this noise difference can happen.

IR filter is 500 gain and 30 Hz low pass filter.

Anyway now CCFC locking accuracy is below 1Hz if the calibration factor is correct. Strange thing is that locking accuracy above 10kHz is much better than BAB locking accuracy with green lock.

Elog2300 described optimization for CCFC error signal. To characterize better these error signals, I put measured CCFC error signal as follows.

Figure 1 is CCFC error signal at different demodulation phase, after modematching optimization.

Figure 2 is CCFC error signal at different demodulation phase, after mixer optimization.

Matteo and Yuhang

As reported in elog2300, we optimized mode matching and mixer. We obtained a larger CCFC error signal after that. Then we used it to lock the filter cavity length for IR. Control loop information is summarized as follows:

1. Gain of CCFC loop: 50
2. Corner frequency of CCFC loop: 30Hz (one order low pass)
3. Error signal shape: figure 1
4. Open loop transfer function (only CCFC part): figure 2 (40mVpk-pk excitation used)
5. Error signal spectrum (loop on/off): figure 3.
6. Calibration for error signal: AOM speed (4000Hz/2.5s)(figure 4) divided by error signal slope around zero (75mV/43.7ms)(figure 1 and 5) divided by 2 (AOM scan green to IR) : 4000*43.7/2.5/75/2 = 466 Hz/V

We could see that the CCFC method stabilized length noise for IR below ~1kHz. The IR length noise reached 2.3Hz after closing the CCFC loop. Compared with the old case, we could see that the main difference in IR length noise is around 1kHz~10kHz. The reason for this difference is still unknown. But if the CCFC error signal can go back to the old case, the CCFC loop can reduce IR length noise to less than 1Hz.

By changing CCFC demodulation phase, CCFC error signal offset should change in a sinusoidal way. I checked this after the optimization of mixer. The result is in attached figure 1.

I checked the histogram of CCFC and CC1 error signal. This check is after the mixer optimization.

We could see that strange behavior of histogram reported in elog2302 disappeared.

Matteo and Yuhang

As reported in elog2289 and elog2302, the demodulated signal from mixer ZX05-1L-S+ has strange behaviors, such as not exactly sinusoidal or strange data distribution. We realized these issues but we didn't know what is the reason.

On 2020/22/07, we checked two quadrature-phase signals of CCFC error signal while CC1 phase is scanned more than 2pi. While checking, we found these two quadrature-phase signals were not the same. Attached figure 1 shows these two signals.From this figure, the quadrature-phase signal is quite similar with sinusoidal shape while the in-phase one is quite linear between each maximum and minimum. After observing this difference, we start to investigate what is the difference between these two channels.

Comparison of these two channels:

1. The RF signals come from the same PD, the LOs come from the same channel of DDS3

2. LO signal is splitted by ZMSCQ-2-90, RF signal is splitted by ZFDC-10-1-S+

3. They use the same mixer ZX05-1L-S+ and the same low pass filter SLP-1.9+

The splitting of LO makes one LO ~11dB smaller than the other one (The splitting of LO should give identical output. However, there is difference due to frequency issue.). The splitting of RF makes one RF ~10dB smaller than the other one. (RF signal is about -3dBm before splitting)

In the end, we found the problem comes from LO. We were using ~-6dBm LO, which is smaller than the datasheet requirement. However, in practice, this mixer needs even smaller LO (-12dBm LO is used now).

When CC1 is locked and the filter cavity is detuned, the CCFC error signal only shows an offset. This field should be identical with CC1 error signal if offset is not considered. Before the optimization of the mixer, we checked the histogram of this offset. From the attached figure, we could see that this histogram has some problems (no data located in the center). It could come from an oscillation of this signal.

We should recheck it after the optimization of the mixer. 

The filter cavity GR lock's OLTF may differ with the filter cavity GR+IR lock's OLTF at low frequency. Therefore, we start to investigate GR OLTF's low-frequency part.

In the attached figure, there are four measurements. Their legends are listed in the sequence of time on 2020/12/07. We could see that:

1. All measurement shows flat gain at low frequency, which is different from what we expect.

2. Morning and evening measurements' magnitude are quite different at low frequency. The reason for this difference is still unknown.

3. Measurement phases are different with/without SR560 (just passing through without gain/filters). We could see that the phase margin is better if SR560 is used.

Matteo and Yuhang

Based on Aritomi-san's code, I add the degradation from mode-matching to the CCFC error signal. The simulation result is in attached figure 1. From this simulation, worse mode-matching makes CCFC error signal degrade around resonance. But mode-matching doesn't affect the CCFC error signal's offset.

Based on this simulation, we sent BAB to the filter cavity and checked the mode-matching was about 0.75. We found the IR drift happened only in the yaw direction. After optimizing yaw, mode matching increased to about 0.9. When we checked the CCFC error signal's pk-pk value, we found some issues with this signal's demodulation. After optimizing the mixer, we saw an even better CCFC error signal. The comparison of CCFC error signals before and after optimization is in attached figure 2.

I compared the mm-optimized/mm-original CCFC error signal's minimum. In the simulation, the ratio is 0.64. While in measurement, it is 0.58.

We took a spare PSD and replaced the old one for PR Oplev. The spectrum of PSD was measured and shown in figure 1. We can see that the new PSD has higher noise than the reference. Apart from that, the new PSD also shows different peaks, which needs to be further examined.

TAMA PSD for PR pitch show excess noise again, the situation is shown in the figure 1.

I attached OLTF of CCFC and green lock. Note that I flipped the sign of measured data to match the measurement and theory. The measured phase is not consistent with theory.

In elog1727, I tuned CC PLL frequency from the fitting of CC separation frequency and CC PLL frequency, but the error of the fitting parameters is quite large with respect to optimal CC separation frequency 108 Hz. So this method is not precise to decide the correct detuning.

As written in elog2294, current CCFC error signal is not consistent with theoretical plot with optimal detuning, but instead it is similar to the theoretical plot with 25 Hz detuning.

If the current detuning is 25 Hz, we have to change the detuning by 29 Hz to obtain optimal detuning 54 Hz. Using the formula in elog1727, the CC PLL frequency has to be changed by 2*29 Hz/1.907605 =  30.41 Hz. Since the current CC PLL frequency is 6.99704303 MHz, optimal CC PLL frequency should be either 6.99707344 MHz or 6.99701262 MHz. I checked both cases by looking at CCFC error signal and found that 6.99701262 MHz is correct one (In DDS, 6.99701253 MHz was set). 

Here is the new CC PLL setting. I saved this setting as 20201126_dds3_CCFC.

Attached plot shows CCFC error signal with different CCFC demodulation phase. Amplitude of the CCFC error signal is normalized with 83mV which is the amplitude of CCFC error signal when CCSB are off resonance of FC and CC1 is scanned.

Now the shape of CCFC error signal is similar to theoretical plot. In addtion to that, zero crossing point of blue curve in second plot is around 58Hz which is almost optimal detuning.

CCFC error signal with 25 Hz detuning is very similar to the measurement.

Actually, the errors of the fitting parameters are -1907605 +/- 36859 and 13347486 +/- 257882. This error is quite large with respect to 108 Hz. We need to fine tune CC PLL frequency by looking at CCFC error signal.