A Deep Neural Network-Based Split-Step Decoding Algorithm in Optical Spatial Modulation System

In order to address the issues of low transmission rate and low spectral efficiency in existing optical spatial modulation, we propose a Deep Neural Network (DNN)-based split-step decoding algorithm for optical spatial pulse position modulation Faster-Than-Nyquist (OSPPM-FTN) scheme to achieve near optimum performance with low-complexity. Additionally, we derive the theoretical upper bound of bit error rate (BER) for the OSPPM-FTN system under the Gamma-Gamma turbulence channel and validate its reliability using Monte Carlo method. Simulation results demonstrate that compared to existing OSPPM at acceleration factor T=0.9, the OSPPM-FTN scheme achieves higher spectral efficiency and transmission rate by 0.17 bit/s/Hz and 0.22 bpcu respectively, albeit with a loss of approximately 1 dB in signal-to-noise ratio (SNR). Moreover, our proposed decoding algorithm exhibits 42% lower computation complexity than the maximum likelihood (ML) algorithm.