Application of Machine Learning to Signal Detection in Underwater Wireless Optical Communication Links

We consider the application of a machine-learning (ML)-based method to the demodulation of the received signal in underwater wireless optical communication (UWOC) links. This approach is justified when the underwater optical channel is subject to strong variations due to various phenomena such as pointing errors and turbulences, which directly impact the received optical power, requiring accurate and agile channel estimation. The investigated ML method is based on the well-known
$K$
-nearest neighbors (KNN). We demonstrate excellent link performance for different types of modulation schemes even under high data rates and low received optical powers, for instance, achieving effective bit rates of 2.96 and 2.54 Gbps using 16-QAM and 32-QAM modulation schemes, respectively, at a received optical power of -16.4 dBm. We also discuss the implementation aspects of the proposed approach, including its computational complexity.