Beaconless adaptive optics for atmospheric laser propagation with multi-plane convolutional neural network

We report on the development of a machine learning method to emulate the action of adaptive optics for atmospheric laser propagation in the absence of a dedicated beacon laser. Our technique measures scattered-light intensity profiles recorded in image planes nearly conjugate to the object plane and uses a convolutional neural network to correlate those images to a phase profile, which produces an improved figure of merit. This technique offers the possibility of emulating phase-conjugate adaptive optics in laser-propagation applications where a scattering surface exists at the object plane, but a beacon is unavailable. We quantify the improvement in the figure of merit relative to a beam that has been tip-tilt corrected to remain centered in the object plane.