Noise-dampening AI improves gravitational wave detection

Experts at the Laser Interferometer Gravitational-Wave Observatory (LIGO) and Google DeepMind have trained an artificial intelligence program to dampen pesky background vibrations which drown out signals from the mergers of binary neutron stars and potential intermediate-mass black holes.

“We were already at the forefront of innovation, making the most precise measurements in the world, but with AI we can boost LIGO’s performance to detect bigger black holes,” says Rana Adhikari, a professor of physics at California Institute of Technology (Caltech), USA.

“This technology will help us not only improve LIGO but also to build next-generation, even bigger gravitational-wave detectors.”

LIGO’s interferometers, located in the US states of Louisiana and Washington, detect gravitational waves – undulations in space-time which ripple out from colliding cosmic bodies.

Each arm of the L-shaped facilities contains a 4km-long vacuum tube in which lasers are bounced back and forth with the help of massive mirrors suspended at each end.

As gravitational waves reach Earth, they stretch and contract these arms by almost imperceptible amounts. LIGO’s laser system detects these sub-atomic scale distortions, with most signals captured in the 30 to 2000 Hz range.

Engineers work hard to reduce background noises which can drown out these signals, including unwanted movements in LIGO’s mirrors caused by the motion of the ocean kilometres away.

“It’s as if the LIGO detectors are sitting at the beach,” says Christopher Wipf, a gravitational-wave interferometer research scientist at Caltech.

“Water is sloshing around on Earth, and the ocean waves create these very low-frequency, slow vibrations that both LIGO facilities are severely disturbed by.” 

The solution to the problem works much like noise-cancelling headphones, Wipf explains.

“Imagine you are sitting on the beach with noise-cancelling headphones. A microphone picks up the ocean sounds, and then a controller sends a signal to your speaker to counteract the wave noise,” he says.

“This is similar to how we control ocean and other seismic ground-shaking noise at LIGO.”

But cancelling out this background noise introduces a new, higher frequency quiver in the mirrors.

“If you have ever listened to these headphones in a quiet area, you might hear a faint hiss. The microphone has its own intrinsic noise. This self-inflicted noise is what we want to get rid of in LIGO.”

Unfortunately, this self-inflicted hiss lies in a key frequency range – 10 to 30 Hertz – where gravitational waves caused by the mergers of hypothesised intermediate-mass black holes might occur.

Improved low-frequency sensitivity of gravitational wave observatories would also make it easier for LIGO to detect black holes with eccentric (oblong) orbits.

“The 10- to 30-Hz band is also important for the early (premerger) detection of binary neutron stars (BNSs), potentially doubling the warning time, which would enable real-time observation of neutron star collisions, the subsequent creation of heavy elements, and the birth of black holes,” write the authors of a new study presenting the research in the journal Science.

LIGO researchers approached DeepMind to develop an AI to control this noise better than the traditional feedback control system. The new controller, Deep Loop Shaping, was trained using a ‘reinforcement learning’ approach.

“This method requires a lot of training,” Adhikari says.

“We supplied the training data, and Google DeepMind ran the simulations. Basically, they were running dozens of simulated LIGOs in parallel. You can think of the training as playing a game. You get points for reducing the noise and dinged for increasing it. The successful ‘players’ keep going to try to win the game of LIGO. The result is beautiful –the algorithm works to suppress mirror noise.”

Tests at the LIGO Livingston Observatory in Louisiana showed Deep Loop Shaping controlled noise in the 10 to 30Hz band more than 30 times better than the standard controller, according to a

The team plans to conduct longer duration tests before implementing the AI on LIGO systems.