Abstract:

HPCD noise might be reduced by a factor of 5-6 by just changing the order of the engagement of dewhitening filter.
It should be checked with the test bench or real circuit.

Detail:

To investigate the noise from coil driver outputs, I checked the noise spectra of coil drivers.
First, I noticed the high power coil driver (HPCD) output noise is reduced a lot when chaging the dewhitening filter (DWF) settings from 2 to 3 stages while the noise reduction is small from 0 to 1 and 1 to 2.
So, I checked why this noise reduction happens.
This time, I used the data on the JGWdoc (JGW-S1807833-v1) for the calculation.

According to the circuit diagram (JGW-D1503503-v4), 16k-Ohm resistances at the dewhitening filter stages could limit the output noise of the HPCD.
Figure 1 shows the noise spectrum of HPCD outputs without DWF (blue) and with 3-stage DWFs (orange).
Green line in the graph shows the thermal noise of 16k-Ohm resistances at the DWFs (squrare root of square sum of 3 resistances).
Red line in the graph represents the thermal noise of 16k-Ohm resistance with single DWF because resistance at the final stage is reduced only by the lase DWF.

Since DWFs are engaged from those close to the input, final 16k-Ohm thermal noise is reduced only when all DWFs are on.
Probably that is why the noise spectrum measured in JGW-S1807833-v1 is reduced al lot only when changing DWF setting from 2 to 3.
So, if the order of engagement of DWFs is changed, HPCD output noise could be reduced.
It should be tested.

Images attached to this report

31771_1732593678_Screenshot from 2024-11-26 13-01-05.png

Comments to this report:

takahiro.yamamoto - 16:18 Tuesday 26 November 2024 (31774)

If it's really effective to reduce DAC+HPCD noise, C-code of coil driver BIO controller should be modified.
I modified C-code and check the improvement on the standalone system before a new code would be installed to KAGRA system.
In conclusion, Ushiba-kun's prediction seems to be correct and there was an improvement in output noise.
And also hacked C-code works fine. So we can install the new code in the next maintenance day.

-----
For measuring the output noise of HPCD, input port of HPCD was shorted and output port was connected to the whitening filter on which 36dB gain and 2 stages of whitening were engaged. Black curve in Fig.1 shows the ADC, AA and whitening filter noise with 36dB gain and 2 stages of whitening. This noise seems to be limited by the noise of whitening filter not ADC (see also klog#31763).

Output noise of coil driver without any de-whitening filters are shown as dark green curve. Brown, light green, and blue curve represent the output noise with one stage of de-whitening at the 1st, the 2nd, and the 3rd stage of de-whitening respectively. We can see the improvement by engaging a more latter stage of whitening filters. In the case of engaging the 3rd stage, whitening filter noise limits this measurement. So we couldn't know actual noise level in the case of engaging the 3rd stage. Anyway some improvement can be seen by engaging the 3rd stage instead of the 1st or the 2nd stage. So Ushiba-kun's suggestion seems to be effectively to reduce the coil driver noise.

In the cases of engaging 2 stages of de-whitening are shown in Fig.2. Again, we can see that the 3rd stage filter being engaged is important for noise reduction e.g. blue or green curves vs. brown curve.

In order to engage a latter stage first, I modified the C-code for the coil driver BIO controller. This hack seems to work fine as shown in Fig.3 (disengage all de-whitening), Fig.4 (engaging 1 stage of de-whitening), Fig.5 (engaging 2 stages of de-whitening) and Fig.6 (engaging all 3 stages of de-whitening). These 4 states which are used during the lock acquisition are assigned as 1-4 on the state request numbers. For various future test, 5-8 on the state request numbers are also defined in C-code. List of states are as follows.
1: LP1:OFF, LP2:OFF, LP3:OFF 2: LP1:OFF, LP2:OFF, LP3: ON 3: LP1:OFF, LP2: ON, LP3: ON 4: LP1: ON, LP2: ON, LP3: ON 5: LP1:OFF, LP2: ON, LP3:OFF 6: LP1: ON, LP2:OFF, LP3:OFF 7: LP1: ON, LP2:OFF, LP3: ON 8: LP1: ON, LP2: ON, LP3:OFF

Images attached to this comment

takahiro.yamamoto - 18:23 Friday 29 November 2024 (31863)

After the model update to apply a new BIO controller (see also klog#31838 and klog#31857), it's working as expected.

For example, 1 and 2 stages of de-whitening filter for ETMX_MN_H and ETMX_MN_V, respectively.
Before this update, former stages were engaged at first as shown in the attachment.
On the other hand, latter stages are now engaged at first.

It's difficult to recover DC lock with today's weather situation.
So an effect on the noise level can be checked in next week.

Images attached to this comment