First demonstration of neural sensing and control in a kilometer-scale gravitational wave observatory

Abstract

Suspended optics in gravitational wave (GW) observatories are susceptible to alignment perturbations and, in particular, to slow drifts over time due to variations in temperature and seismic levels. Such misalignments affect the coupling of the incident laser beam into the optical cavities, degrade both circulating power and optomechanical photon squeezing, and thus decrease the astrophysical sensitivity to merging binaries. Traditional alignment techniques involve differential wavefront sensing using multiple quadrant photodiodes, but are often restricted in bandwidth and are limited by the sensing noise. We present the first-ever successful implementation of neural network-based sensing and control at a gravitational wave observatory and demonstrate low-frequency control of the signal recycling mirror at the GEO 600 detector. Alignment information for three critical optics is simultaneously extracted from the interferometric dark port camera images via a CNN-LSTM network architecture and is then used for MIMO control using soft actor-critic-based deep reinforcement learning. Overall sensitivity improvement achieved using our scheme demonstrates deep learning's capabilities as a viable tool for real-time sensing and control for current and next-generation GW interferometers.

A deep learning technique to control the non-linear dynamics of a gravitational-wave interferometer

“…This lock acquisition problem [5,6] has been historically solved with solutions developed case by case that are difficult to scale to more complex systems. Deep Learning techniques have also been recently applied to the steady state linear control of angular motions in gravitational-wave interferometers [7].…”

“…All signals are products of pairs of field amplitudes and complex conjugates. Inspection of equations ( 5)- (7) shows that the fields contain exponential terms that are functions of the MICH and PRCL degrees of freedom, and thus are periodic. The presence of those exponential terms explains the fact that multiple positions, differing by multiples of λ/2, result in the same values for the observed signals, and are therefore indistinguishable and equally suitable as operating point.…”

“…Our work is based on numerical simulations of the interferometer motion, obtained by creating time series of the MICH and PRCL degrees of freedom that mimic what is observed in the real world. Then using the field equations ( 5)- (7) we can produce the optical signals that a model may use as input during real-time operations. The simulation provides a data sampling rate of 2048 Hz.…”