In this paper, an intensity modulation and direct detection (IM-DD) discrete multi-tone (DMT) interconnection system employing polar coded and deep learning-assisted forward error correction decoding probabilistic shaping 16 quadrature amplitude modulation (PS-16QAM) is studied. By employing many-to-one (MTO) based PS to achieve signal Gaussian distribution model, the proposed optical DMT system is with the advantages of shaping overhead free profile, and optimized polar coded modulation architecture. To overcome the ambiguous problem for the overlapping symbol decision in PS, a deep learning assisted forward error correction decoding is proposed for belief propagation (BP) based polar decoding. The computational complexity of deep learning assisted polar decoding is superior to the original BP decoding one, and with faster rate convergence and better optical transmission performance. Simulation results in a 100Gb/s optical DMT transmission system present that, the deep learning assisted polar coded PS-16QAM signal achieves 0.38-dB superior receiver power sensitivity compared with conventional polar coded system over 10-km standard single mode fiber transmission.
|