Experimental demonstration of an OFDM-UWOC system using a direct decoding FC-DNN-based receiver

In this paper, a novel fully-connected deep neural network (FC-DNN)-based receiver is proposed and experimentally demonstrated in an underwater wireless optical communication (UWOC) system, with the objective of direct decoding orthogonal frequency division multiplexing (OFDM) signals via deep learning methods. Unlike existing schemes, which treat deep neural networks as a single equalizer or a symbol classifier that requires pre-processed labels, the receiver we proposed can achieve direct output from received OFDM signals to the desired bit stream without any redundant pre-processing and post-processing. The feasibility of the proposed receiver has been validated by changing the transmission data rate and the received optical power of experimental setups. Furthermore, our experiments indicate that the proposed direct decoding FC-DNN-based receiver can perform as well as the typical conventional bulky multi-structure receiver without using pilots and cyclic prefixes (CPs) in OFDM-UWOC systems.