J. Kissel, J. Warner, K. Venkateswara
Based on Rich's and Jeff's SEI log entry 586 and 594, we made a second attempt at sensor correction on ETMX along X and Z on stage 1, which seems to be working better. To judge the performance I have plotted the stage 1 T240 output and also the Oplev Pitch and Yaw output. We used a slighlty modified version of Rich's sensor correction filter described in the above log.
To establish baseline, the first plot shows the Stage 1 T240_X motin (green), the ground STS (blue) and the tilt-subtracted ground super-sensor (red) with no sensor correction applied. I've also shown the coherence between some sensors and the Oplev motion. All degrees of freedom had Ryan's LLO blend filters.
The next file (SensCorrectOnBRSOff) shows the sensor correction turned on for X and Z on stage 1, but with normal gnd_STS, not the tilt-corrected super sensor.
The final file (SensCorrectOnBRSOn) shows the sensor correction with the tilt-subtracted super-sensor for X. Note that the ground motion (blue) is not the same during these data sets, but the relative differences between the lines are important.
Some comments:
1. Based on these plots, it looks like turning sensor correction On, even without tilt-subtraction, improves performance at 0.1-0.5 Hz by factors of 2-5. It's effect below 0.1 Hz is not clear - there may be small tilt amplification. Switching to the tilt-corrected super-sensor slightly improves performance below 0.1 Hz by factors of 2 ish. It is probable that we are limited by tilt-reinjection from the low X blend.
2. We are probably limited by the L4C sensor noise between 0.5 to 1 Hz. By improving the L4C blend, we may be able to get another factor of 2ish at these frequencies.
3. The Oplev motion doesnt show much improvement despite better X performance. The pitch is very sensitive to and probably limited by the RY blend.
Sensor correction for X and Z has been left on overnight, since it may help. It is easy to turn off from the ISI medm screen, if it is affecting performance.
edit: I added the Stage 1 Z performance to the plots. The sensor correction appears to improve z performance by ~10 at the microseism. But there may be more pitch motion at ~10 mHz. Not sure what is causing that.
Tomorrow, we will try HEPI sensor correction which may or may not be better.
edit: I have added another file also showing the Stage 1 RY motion (converted to displacement units), which shows good coherence with X motion confirming tilt-reinjection in X.
J. Warner, K. Venkateswara, H. Paris, J. Kissel Just to add some modeling sauce to Krishna's statements, I attach modeled performance plots comparing Rich's aggressive IIR sensor correction filter (from LHO aLOG 586) against Krishna and Jim's even more aggressive IIR sensor correction filter (from 14561). As Krishna says, we're getting better and better performance out of lowering the corner frequency of the sensor correction filter, made possible with the tilt-corrected ground sensor (despite his modest claims that it's not doing much). Indeed, as we continue to improve the residual ground motion subtraction, we get more evidence as suspected from my modeled performance in SEI aLOG 594, that we are limited by L4C sensor noise from ~0.3 [Hz] to 1 [Hz] (and re-injected RY noise between 0.1 and 0.3 [Hz]). At this point, I have no definitive proof other than a similar shape of the 0.3 [Hz] to 1 [Hz] noise to the model and how it evolves with the latest changes in sensor correction -- but with the improved subtraction, its in some sense "exposing" the L4C noise by removing the limiting residual ground motion. Check pages 1 through 5 for comparisons of the FIR filters, and modeled performance using the Ryan DeRosa blends. Suspecting we can improve the L4C noise limitation by adjusting the T240 / L4C inertial sensor blend cross-over, I asked Hugo and and Jim for some information on how that cross-over is defined in the generic control scripts (knowing full well that Ryan would have chosen something different). In response, they pointed me to Brian Lantz's code for generating this crossover, ${SeiSVN}/seismic/Common/MatlabTools/blend_T240_L4C_111012.m In this function, Brian uses the knowledge of the T240 and L4C sensor noises to "optimize" the cross-over. Assuming this cross-over is better, I (1) Reconstructed the inertial sensor half of Ryan's psuedo-complementary blend filters by adding the existing T240 and L4C filters (grabbed directly from foton using readFilterzpk.m, since there's no matlab representation of these filters) (2) Grabbed Brian's T240 and L4C complementary pair from blend_T240_L4C_111012.m (3) Multiplied Brian's T240/L4C pair by Ryan's inertial sensor blend, such that total inertial sensor blend remains pseudo-complementary to Ryan's displacement sensor blend. (4) Ran through the same model, comparing Ryan's inertial cross-over vs. Brian's inertial cross-over. Blamo! -- if we are indeed limited by L4C noise (confirmed only by eye at this point) -- we can improve the noise from 0.3 to 1 [Hz] by another factor of a few. The filter comparison and modeled improvement is shown on pages 6-8 of the attached. We'll figure out how to actually implement this in foton tomorrow (gulp), so we can demonstrate this live. Plots are and model are produced by ${SeiSVN}/seismic/BSC-ISI/Common/Sensor_Correction_Design_BSC_ISI/design_sensorcorrection_IIR_20141021.m
K. Venkateswara
I had a calibration error in the above plots. I've corrected it and attached the following files:
ST1SCoff.pdf = Stage 1 X sensor correction off.
ST1SCBRSoff.pdf = Stage 1 X sensor correction with just GND_STS, BRS not used.
ST1SCBRSon.pdf = Stage 1 X sensor correction with tilt-subtracted ground sensor.
SCCompare.pdf = Comparison of the three configurations. As ground motion was different during these measurements, this is not a good judge of performance below 0.1 Hz, but is useful above 0.1 Hz.