Machine-learning-enhanced, phase-based multi-tone continuous-wave lidar

Abstract

Lidar technologies have been investigated and commercialized for various applications such as autonomous driving and aerial vehicles. The pulsed time of flight and frequency-modulated continuous-wave lidars are the two common lidar technologies that dominate. As an alternative to the available lidars, we developed the phase-based multi-tone continuouswave (PB-MTCW) technology that can perform single-shot simultaneous ranging and velocimetry measurements with a high resolution at distances far beyond the coherence length of a CW laser, without employing any form of sweeping. The proposed technique utilizes relative phase accumulations at phase-locked RF sidebands to identify the range of the target after a heterodyne detection of the beating of the echo signal with an unmodulated CW optical local oscillator (LO). Upto-date, we demonstrated that the PB-MTCW lidar could perform ranging ×500 beyond the coherence length of the laser with <1cm precision. Here, we implement machine learning (ML) algorithms to the PB-MTCW architecture to improve the ranging resolution, as well as to provide a solution to multi-target reflections using tone-amplitude variations. We used four different training schemes by utilizing the acquired RF tones and phases from simulation results, experimental results, and their combinations in a convolutional neural network model. We demonstrate that the ML algorithm yields an average mean square error of ~0.3mm compared to the actual target distance, hence enhancing the ranging resolution of PB-MTCW lidar. It is also shown that the ML algorithm can distinguish multiple targets in the same line of sight with a 98%±0.7% success rate depending on the targets' reflectance and distances.