End-to-end performance of a CNN-AE based faster-than-Nyquist rate free space optical communication system
The performance of faster-than-Nyquist(FTN)free space optical(FSO)communication system is significantly affected by inter-symbol interference(ISI).To address this issue,an end-to-end communication system leveraging a convo-lutional neural network autoencoder(CNN-AE)has been proposed to eliminate the influence of ISI and facilitate signal re-covery.Additionally,given the non-orthogonal characteristics of FTN signals,an alternating training algorithm has been em-ployed.This algorithm trains the transmitter and receiver weights separately,addressing the issue of data size mismatch be-tween model input and output during supervised training.Based on this,the bit error rate(BER)performance in Gamma-Gamma atmospheric turbulence channel was analyzed.Simulation results demonstrate that compared with traditional systems using maximum likelihood sequence estimation(MLSE),the proposed system achieves superior BER performance under various conditions.Specifically,when the acceleration factor falls within the Mazo limit,the proposed system effectively e-liminates the complex mixed ISI caused by FTN shaping and atmospheric turbulence channel,resulting in BER performance comparable to that of orthogonal transmission systems.