Abstract
Because filter implementation rule for IP_ACCINF was not unified on four TypeA suspensions, we discussed a new rule and prepared (not applied yet) a new filter implementation.
The rule we decided:
- FM1: individual factor (CompGain)
- FM2: count -> volt (cnt2V)
- FM3: volt -> um/s (V2um/s)
- FM4: Inverse Geophone + DCcut (InvGeo+HP)
Details
Current filter implementation is not unified among TypeA suspensions as shown in Fig.1.
For ETMX, because all gain factors are merged, it's difficult to understand the meaning of each value (see also Fig.2.)
At first, I prepared filter modules in new rule without changing filter shape.
Red and black curves in Fig.3 represent the filters in new rule and original filters, respectively.
Detailed implementation is shown in Fig.4.
By the way, the reason why Inverse Geophone and DCcut are merged is to avoid the numerical rounding error.
The filter shape of DCcut is also different between ITMX and ETMX as shown in Fig.5, and we decided to adopt the idea on ITMX.
So I also implemented new design of DCcut for ETMX. New design is shown as blue curve in Fig.3.
I will test these filters tomorrow.
Just some comments.
I understand you changed this to avoid the overflowing problem. But having DC cut at 0.01 Hz is not the best configuration especially when we're blending at 0.08 Hz for L and T. There's too much phase distortion around the blending. It could lead to a strange super sensor response and could lead to all sorts of uncertainties down the road.
Yes, your concern about phase margin around the blending frequency is correct.
Actually I and Ushiba-kun are now considering to change the corner frequency of DCcut as more lower frequency.
Corner frequency at 0.01Hz is now just a copy of the ITMX implementation.
On ITMX, 0.01Hz DCcut filters has been used since several months ago.
Phase margin of ITMX is roughly 15deg. smaller than one of original filters of ETMX.
So we should re-consider to the corner frequency of DCcut both ITMX and ETMX.
Current filters are red curve in Fig.1 of klog#18383
(not Blue curve which is a new DC cut filter yet.)
Fig.1 shows the current filter bank.
On FM9, I also prepared a new Inverse Geophone + DCcut filters which is represents blue curves in Fig.1 of klog#18383.
Geophone spectra didn't change before and after replacing filters (See also Fig.2)
Finally I confirmed ALIGNED is now available.
As Ushiba-kun reported as klog#18425, 0Hz pole without feedback control often induces numerical rounding errrors.
So we decided to avoid using 0Hz pole which can be removed.
e.g. Though TF shapes of Example-A and B shown below are same, effects of numerical rounding errors are quite different even if both 'zpk' and 'butter' are contained one FM module.
Eight panels on the left hand side in Fig.1 shows filter implementation of ACCINF_{H1,H2,H3,H4}.
Orange and blue curves are original ETMX implementation, same implementation as ITMX, respectively.
Current implementation (FB screen is shown in Fig.2) is represented as red curves.
Overall gain is adjusted as gain at 100Hz becomes 0dB for InvGeo+HP (FM4).
And also, conversion filters from velocity to displacement at FM1 of ACCBLEND are changed form 0Hz pole to 0.1mHz.
(See also 6 penels on the right hand side of Fig.1 and Fig.3)
-----
Examples:
A: butter("HighPass",4,0.01) zpk([-0.3255+i*0.99827;-0.3255-i*0.99827],[0;0],-0.155038,"f")
B: zpk([-0.3255+i*0.99827;-0.3255-i*0.99827],[0;0],-0.155038,"f") butter("HighPass",4,0.01)
