Multicarrier Signal Unfolding for 25% Capacity Improvement in Optical Fiber Transmission

Spectrally efficient frequency division multiplexing (SEFDM) can achieve improved spectral efficiency than orthogonal frequency division multiplexing (OFDM) waveform by narrowing subcarrier spacing. However, the resulting inter-carrier interference (ICI) limits the performance of simple detectors such as matched filter (MF). Traditional iterative detection (ID) has competitive detection performance via cancelling interference iteratively and comparing symbols with manually defined thresholds. Unfolding the ID detector results in a stacked architecture similar to a multi-layer neural network (NN). This architecture allows the combination of NN and the ID process in each iteration, where the ID process is mathematically deterministic. This work proposes an unfolding detector which adjusts an available symbol threshold mapping result via NN processing, potentially giving us the optimal threshold mapping scheme. Simulation results show that the proposed unfolding detector achieves better error performance on SEFDM signal detection than traditional detectors, across all examined optical fiber transmission distances. Furthermore, it exhibits close error performance to OFDM with a $25 \%$ spectral efficiency gain.