Study on a Deep Learning Based Demodulation for the Optical M-ary/SS System

The conventional optical M-ary spread spectrum (M-ary/SS) system requires a correlator with all sequences and and maximum likelihood estimation as the demodulator. The correlator needs to be optimally configured according to the sequences. In this paper, we propose a new optical M-ary/SS system that replaces the demodulator with a deep learning based demodulator. This approach allows for a simplified receiver design, independent of the sequences used, by the learning process through a deep neural network. As a result, a maximum gain of about 0.7 dB is achieved in bit error rate at 10⁻⁵ by the proposed system.