High availability motion sensor with nonlinear interferometry and AI

Robin Matha

Laser “self-mixing” is (in principle) a well-established sensing approach based on interference between optical waves taking place inside the laser generating the light. The physical principle is sound and the first-principle physical modelling is rather straightforward: monochromatic light emitted by a laser device hits a target and (after some propagation) re-enters the laser, altering its operation point. By monitoring the operation point of the laser, one should be able to reconstruct the displacement of the target along the direction of propagation of the laser light. In spite of this apparent simplicity, reconstructing the displacement from the measured signal has proven much more challenging than expected (a first roadblock), and uncontrollable variations in the reflectivity of the target (the second roadblock) further hinder reconstruction efforts. Here we show that an adequately trained neural network can reconstruct the displacement of the target with excellent accuracy (even across major changes in the experimental apparatus), removing the first roadblock. In addition, we also analyse a multi-sensor configuration which, equipped with a neural network for signal analysis, is able to circumvent the second roadblock and provides the first high-availability motion sensor based on self-mixing interferometry.

While these results have an interest on their own since they solve the two major problems of self-mixing interferometry, we believe they also show the great potential of hybrid approaches leveraging nonlinear photonics and AI for real world sensing tasks.

Robin Matha studied photonic and optronic system engineering at Polytech Orsay. He was intern as assistant engineer at Macquarie University on fiber laser integration and later at ONERA in Saclay simulating the effect of speckle on laser self-mixing interferometry. He graduated in spring 2021 and since then he has been preparing a PhD at ONERA and Institut de Physique de Nice on self-mixing and neural networks.