AI based on frequency slicing deep neural network for underwater visible light communication

Abstract

In this paper, we propose a low-complexity frequency slicing deep neural network (FSDNN) for wide-band signal post-equalization in a 1.2 m underwater visible light communication system. FSDNN and deep neural network (DNN) outperform the least mean square equalizer. Then, by splitting the received signal into two parallel signals using a digital low-pass filter and a high-pass filter, we demonstrate that the FSDNN significantly reduces the complexity of the traditional DNN post-equalizer. Moreover, the complexity of the FSDNN decreases considerably to 11.15% compared with the conventional DNN for a 2.7 Gbit/s wide-band transmitted signal with a similar bit error ratio performance.