1. ILIAS – WG1 Hierarchical suspension control G.Losurdo INFN Firenze

2. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Virgo Superattenuator PASSIVE isolator Designed with 3 points of actuation: –Inverted pendulum –Marionette –Recoil mass Local controls –Inertial damping of internal modes (IP) –Pre-alignment/damping of the payload modes (optical levers on marionette/mirror) Global control: locking correction distributed hierarchically over the three actuation points

3. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Digital Controls DSP 1: sends correction for inertial damping and tide control to IP actuators DSP 2: sends correction for local controls/AA/locking to marionette/RM actuators

4. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Control electronics Digital electronics (16 bit ADC – DSP – 20 bit DAC) DSP characteristics: –Clock frequency 60 MHz –2 poles/2 zeroes filter in 330 nsec –3x3 matrix-vector product in 1  sec –Max. sampling freq. 160 kHz (used at 10 kHz) –Frequency accuracy at 10 kHz:  f=2.5  Hz

5. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Inverted pendulum Gravity as antispring: low resonant frequency Pre-isolation effect Low control forces: –f 0 =40 mHz, m=1 ton, l=6 m, x =1 cm  F = 0.6 N !! Ideal as control platform: soft actuation

6. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Sensors/Actuators Inertial sensors: –DC-100 Hz bandwidth –Equivalent displacement sensitivity: better than 10 -11 m/rt(Hz) Displacement sensors: –Used for DC-0.1 Hz control –Sensitivity: 10 -8 m/rt(Hz) –Linear range: few cm Coil magnet actuators: –Linear range: few cm

7. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Control strategy From a MIMO to 3 SISO systems:diagonalization with respect to IP modes

8. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Inertial Damping performance Fringe signal Inverted pendulum motion 24 hrs Rms over long periods  1  m d/dt(L 2 - L 1 ) ~ 0.25  m/s

9. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Low frequency position control is needed because: –Inertial sensors do not provide DC error signal –Inertial sensors response at f<40 mHz can be spoiled by tilt Problem: blend the sensors –dominating the tilt effect –minimizing the seismic noise re-injection –Simplyfing the control strategy Blending the sensors Accel. LVDT Highpass Lowpass + Highpass + Lowpass = 1

10. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino The seismic noise filtering depends on L(s) The loop design is independent on the L(s) cutoff

11. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Local control setup Optical levers read both the mirror and the marionette Marionette control allows larger bandwidth

12. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Marionette error signal allows a bandwidth of 2-3 Hz Uncontrolled resonance (1.2 Hz) exists: needs blending with mirror error signal

13. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Hierarchical control Required locking accuracy:  L  10 -12 m Tidal strain over 3 km:  L  10 -4 m The required dynamic range can be covered by two stages. The third one helps for bandwidth/noise issues… Tide/drifts compensation Control of the resonances Widening the bandwidth…

14. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino SA local sensing Mirror PSD Marionette PSD F7 LVDTs IP LVDTs/ACC

15. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino SA sensing IP and F7 diagonalized with respect to VRS (P.Ruggi, S.Braccini, F.Frasconi) P.Ruggi

16. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino RM actuation RM actuators can compensate up to 100  m (in high power/high noise configuration) Tidal strain can be larger Locking is lost Power in the cavityIP positionCorrection to mirror microns

17. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Tide Control Re-allocation of the low frequency (<10 mHz) correction to the IP Cavity transmission Correction to the mirrorSuspension point position 24 h

18. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino C4 run Tide control: data vs prediction

19. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino 16 mHz – the problem Main rotational mode of the SA Long decay time, large elongation. Hard to be controlled from the marionette 1 2 3 4 7 1 Braccini, Vicerè

20. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino 16 mHz – solution Damp it off using F7 actuators!

21. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino F7 control Hardware/software for F7 control implemented on the NE tower 16 mHz resonance control activated Used either with or without LC Possibility to control other “dangerous” SA modes to be studied Open loop gain correction error signal

22. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Locking from the RM: noise Reference mass actuators dynamics: 100  m DAC noise: 300 nV/Hz 1/2 A.Gennai

23. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Standard design DAC noise: 300 nV/sqrt(Hz) (17.5 effective bits) Coil Driver noise: 80 nV/sqrt(Hz)

24. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Noise Reduction To reduce the DAC noise we should insert a resistor in series with coil driver. To get closer to VIRGO specs, the resistor value should be about 500 ohms. Bigger values could be used if force will be enough to keep the cavities locked. The resistor limits the maximum force we can apply and therefore makes lock acquisition very difficult (impossible?)

25. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino New Solution We supply the required additional force for lock acquisition with a transconductive amplifier. During lock acquisition phase only DAC 1 is used. During linear phase DAC 1 output set to zero and DAC 2 is used to keep the lock.

26. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Basic Equations Lock Acquisition: g 1 = 1, g 2 = 0 Linear Regime:g 1 = 0, g 2 = 75 Note: coil pole shifted above 20 kHz

27. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino High power – low noise switch switch

28. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Noise figures DAC noise expected (?) @100 Hz: 3 10 -16 m/Hz 1/2 Virgo design sensitivity @100 Hz: 2 10 -19 m/Hz 1/2 Required noise reduction @100 Hz: 1500 Using tide control allows to reduce the required correction by a factor 100 Re-allocating locking to the marionette in the SA resonance region should provide the residual attenuation

29. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Correction to the mirror in C4 Marionette control with 3 Hz bandwidth allows to reduce the correction by 50 (V p = 2 mV) Coil driver gain could be reduced by a factor 5000  50 To be re-allocated to marionette V p =0.1 V

30. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Mechanics of the last stages Complicated dynamics, important couplings…

31. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Use of SA simulation SA simulation has been important to design the marionette control strategy: –Tuning of SIESTA to reproduce the measured TF –Use of tuned simulation to estimate and subtract the couplings due to the sensing –Calculation of a filter to compensate for  x marionette motion induced by longitudinal forces SA mode tuning: I.Fiori, A.Vicerè LC tuning: S.Avino, E.Calloni, I.Fiori

32. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Marionette TF matrix FzFz F Tx F Ty z Tx Ty I.Fiori Strongly coupled system!

33. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Using 4 coils to move the marionette along z: reduce the z-  x coupling Good data-simulation agreement I.Fiori, A.Gennai I.Fiori, S.Avino

34. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Mechanics The two mechanical TF are different –For the structure around 1 Hz –For the asymptotical slope 1/f 2 1/f 4

35. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino “Modified” marionette Adding two zeroes makes the marionette TF “very similar” to the RM one 1/f 2

36. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino 1st scheme: composed lock ACQ Advantage: simpler, no need of transition Drawback: marionette control bandwidth limited by higher ITF noise (no SSFS) PHD Locking compensator L(s)(s+s 0 ) 2 H(s) In the GC In the DSP zCorr cavity power RM correctionmarionette correction

37. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino 2 nd scheme: re-allocation Advantage: allows wider marionette bandwidth To be tested with AA and SSFS PHD Anti- Ramp Ramp 10s Locking compensator L(s)(s+s 0 ) 2 H(s) In the GC In the DSP zCorr cavity power RM correction marionette correction

38. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Filters To blend the two systems use usual strategy L(s) = 3 rd order low pass filter, H(s) = 1-L(s)

39. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Hierarchical control The north cavity has been locked by distributing the forces over the three SA stages: the controllability of the SA has been demostrated 1997 ! microns DC-0.01 Hz 0.01-1 Hz 1-50 Hz

40. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Performance with no AA/SSFS Mirror displacement correction over the two stages

41. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino C4 data - extrapolation C4 data (noisy stretch) have been filtered with current hierarchical control TFs to predict the correction one expects on the RM when SSFS is ON Expected zCorr rms = 3 mV. L.Holloway

42. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino One should consider peak values of zCorr instead of rms. In C4, over 18 hrs: zCorr peak ~ 10 zCorr rms With hierarchical control in the present configuration one can assume: zCorr peak ~ 30 mV Therefore, the coil driver gain can already be reduced by ~ 300 We are not far from Virgo sensitivity… New promising design is being tested in MATLAB (L.Holloway) Peak correctionrms correction C4 data

43. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Interaction with angular control The alignment/power fluctuations are larger when hierarchical control is ON This is a concern: to be tested with AA Larger statistics needed, analysis going on Standard locking Hierarchical locking

44. ILIAS-WG1– July 7th, 2004 G.Losurdo – INFN Firenze-Urbino Next steps Test hierarchical locking with linear alignment Widen the bandwidth of marionette control Switch to low noise coil driver after re-allocation