Free Space Optics Communications for Low-Power Handheld Mobile Devices

This research demonstrates a machine learning (ML) approach to array-based free-space optical (FSO) communication using mobile devices. Modern warfighters need non-radio frequency (RF) communication methods to eliminate the risks associated with RF communication, such as detection, eavesdropping, and jamming. FSO communications promises tremendous throughput among other advantages, such as low-probability of intercept/detect and resistance to jamming. However, atmospheric conditions significantly reduce achieved performance by introducing fading and noise on the channel. To increase channel resilience and throughput, we employ spatial codes using an array of lasers at the transmitter and train several ML models on the channel alphabet to provide efficient decoding at the receiver. We compare the performance of a Single Shot Detection (SSD) MobileNet model with a You-Only-Look-Once model during the training process, and we demonstrate data transfer over a proof-of-concept system using the trained SSD MobileNet model. We detail the hardware and software implementation for the proof-of-concept, which uses handheld mobile devices and an array of low-cost, low-power lasers. Future experimentation is planned to incorporate forward-error correction and testing over greater distances under realistic conditions.