More detailed discussion about improvement of the coilout noise is in DAC wiki.

Though we don't need drastic (e.g. 10 times) improvement of the coilout noise, improvement as a factor of 2 might become important for our situation.

Ushiba-kun pointed out that DAC noise projection was slightly smaller than the real sensitivity in the case of disengaging the de-whitening filters.
So I compared with the sensitivity in klog#31708, klog#31710, klog#31712, and klog#31716 which are the cases of disengaging de-whitening filters for TM of ITMX, ITMY, ETMX, and ETMY as shown in Fig.1.
My estimation is the case of disengaging TM de-whitening for all 4 Type-As, so it must be larger than the cases of disengaging de-whitening filter for only one Type-A suspension.

In my estimation, actuator efficiency was regarded as one super-actuator in stead of each coil which was same manner as the front-end calibration. This means actuator efficiency was applied for a quadratic sum of output noise of HPCD. For estimating the projection from SUMOUT_L to displacement, it's reasonable way. On the other hand, DAC noise and HPCD noise is injected from each coil path where ACTALIGN matrix and coil balancing gain don't work. We noticed this fact and concluded that we need to estimate the DAC noise contribution by estimating SUMOUT equivalent value once and then we can compute the displacement equivalent value. In other words, we need to inverse matrix of EUL2COIL and ACTALIGN. Of course, another way is measuring the transfer function from COILDRV_VMON to displacement which was resumed in recent model update (klog#31857).

I re-estimated the DAC noise contribution and showed it as a black curve in Fig.2. Because it's a projection for the case of disengaging de-whitening filters for all 4 test masses, it must be larger than one in the case of disengaging de-whitening only for one test mass. But it still same level. So it still remains a mystery of a factor of 2. Anyway, DAC noise contribution seems to be larger than one estimated in the previous post. So it might be more serious for the target sensitivity at that time. So we may need a validation not only for my calculation but also for the measured data.

By the way, some tests of disengaging de-whitening filters were done just after the lock and this situation was kept only a few minutes as shown in Fig.3. LSC_LOCK went to OBSERVING around t=-3 (middle panel) and de-whitening was engaged around t=0 (bottom panel). So it's difficult to ensure that IFO was enough calmed down or not.

Though I evaluated the relation between the sensitivity and DAC+HPCD noise with the DARM spectrum with disabling de-whitening filters measured in klog#31708, klog#31710, klog#31712, and klog#31716 in the previous work,
there was some concern about the reliability of data because the de-whitening filter was disengaged only during a few minutes just after the IFO lock.
So I re-evaluated the relation between the sensitivity and DAC+HPCD noise with measured data in klog#31519 which was longer duration data.
Though a result that projection is smaller than measured data isn't solved yet, we should start to consider about preparation of more weak actuators including a use of LPCD even for the most optimistic case.

-----
Gray curve in Fig.1 represents the DARM sensitivity with de-whitening filters as a reference. Lightgreen, cyan, green, and blue curves represent a DARM noise without de-whitening on ETMX, ETMY, ITMX, and ITMY, respectively. A noise excess as a factor of 3-5 appeared by disengaging de-whitening filters. Black curve (shown only 55-105Hz) represents a quadratic sum of lightgreen, cyan, green, and blue curves which can be roughly regarded as a DAC noise contribution in the case of disengaging de-whitening filter on all 4 Type-A suspensions. Note that because noise excess by DAC is only a factor of 3-5, contribution from another noise such as thermal noise is double-counted in the black curve and it seems to induce an overestimate as 10-15%. Red curve represents the DAC noise projection in the case of disengaging de-whitening filters for all 4 Type-A suspensions. Magenta curve is 1.8 times larger than the red curve, where a factor of 1.8 is eye-fit to adjust to the black curve.

Though I haven't identified the reason of 1.8 times mismatch between the red curve and black curve, we can assume some cases. The most pessimistic case is that using the transfer function from SUMOUT_L to displacement for DAC noise which are injected from each coil paths is improper. In this case, all noise projections (with and without de-whitening) are underestimated as a factor of 1.8. The most optimistic case is that DAC noise or input noise of HPCD installed in mine is larger than one in the circuit room e.g. due to the electrical environment. In this case, DAC noise projection is underestimated as a factor of 1.8 but HPCD noise projections with engaging 2 or 3 stages of de-whitening filters are correct.

Figure 2 shows the DAC+HPCD noise projection in the most pessimistic and optimistic cases. Gray and black curve represent current sensitivity and sensitivity 0.25 times smaller than current one. In the most pessimistic case, engaging 1 or 2 stages of de-whitening filter (cyan or lightgreen curve) is not enough for achieving the black curve. DAC noise contribution with engaging 1 stage of de-whitening (cyan curve) is not negligible even for current sensitivity (gray curve). Estimated contribution of cyan curve to gray curve is ~15%. In the most optimistic case, engaging 1 stage of de-whitening filter is not enough for achieving the black curve and it's not negligible even for current sensitivity. Though the DAC noise contribution with 2 stages of de-whitening filter (green curve) can be negligible for current sensitivity, it's not negligible for achieving the black curve. Estimated contribution from green curve to black curve is ~25%.

As I already reported such as klog#31941, engaging 3 stages of de-whitening filters doesn't make a improvement on the sensitivity from the case of engaging 2 stages because output noise of coil driver becomes a limitation in this case (see also Fig.3, there is no improvement between green and purple curves representing 2 and 3 DWs, respectively). So this is not a effective way to escape from DAC+HPCD noise. Though we need to more investigation for the mismatch as a factor of 1.8 from the view point of making noise budget, we should start to consider a use of LPCD or a preparation of more weak actuators from the view point of improving the sensitivity.