Design of DBN based Demodulator in Underwater Wireless Optical Communications

In this work, we present a machine learning based demodulation method for underwater wireless optical communication (UWOC) systems. Specifically, the demodulator is designed based on deep belief network (DBN) network. The DBN demodulator extracts features layer-by-layer through three restricted Boltzmann machines. To make the method more practical, we build a flexible end-to-end UWOC experimental platform to obtain a real data set of signals from ten different modulation methods. The extensive experiments are conducted and our results show that the DBN demodulator enjoys a better performance than other benchmarks.