S5 time domain calibration summary

Decided to go with FIR filters this time

Length of sensing FIR 30ms, actuation FIR 2s. H2 has the
complication of very narrow notch filters at the first and second harmonics
of the violin modes;  actuation FIR filter needs to be 4s long for H2.

Initial problems

DARM_CTRL has bad quantization noise problem (see plot).

Means, we need to use DARM_ERR only and use the servo filter.

Hard to make good stable servo filter; errors of a few % a few degrees

Played with servo, realise that part of the problem is that foton is
not printing enough significant figures of the filters (on the foton window)
Find work-around with foton problem and implement better servo;
errors reduced by ~1 %.

Finally

In the band from 40Hz-5000Hz (I made the band bigger than 3kHz at the request of the burst
group) differences between frequency domain calibration and time domain calibration
are as follows:

    Relative magnitude difference    Phase difference
H1          ~< 2.5%                    ~< 4 degrees

H2*         ~< 2%                      ~< 3 degrees

L1          ~< 2%                        ~< 5 degrees

*Additionally H2 has differences that creep to about 10% at the notch frequencies
of the first and second harmonics of the violin modes (only see this in real data, not response plots above).
This is of little or no concern.

Other problems

Set up for running on-line. Discover that code takes much longer
on nodes at Hanford than my desktop (1 hour to calibrate 128s of data!!).

After much screwing around discover that making the sensing FIR filter
shorter is the solution. Make it 15ms (rather than 30ms), up-sample residual signal
by  4 rather than 8.

Implemented shorter sensing filters this morning for H1, H2, L1.
Accuracy almost unaffected.

Eveything is under cvs: calibration/timedomain/runs/S5/