【摘要】 为了更有效地消除超奈奎斯特（Faster-Than-Nyquist, FTN）无线光通信系统中大气湍流效应和码间干扰的影响，进一步提升系统误码性能，在卷积自编码器的基础上，提出了一种注意力和长短时记忆网络（Long Short-Term Memory, LSTM）辅助的残差FTN无线光端到端通信系统。根据FTN信号受相邻信号影响的特点，在系统发射端引入LSTM网络来提取FTN信号特征；同时，将注意力机制引入到接收端来增强其对接收信号的恢复能力；再将残差网络引入到系统中，来获取更高的分类精度，且加速系统的收敛速度。在此基础上，探讨了该系统在不同条件因素影响下的系统最佳训练信噪比。仿真结果证明，改进后的自编码器系统在Mazo限内取得了与奈奎斯特系统相近的误码性能，并且在强湍流信道下，能够明显改善系统“误差下限”的问题，当信噪比为32 dB时将误符号率控制到10^-3以下。更多还原

【Abstract】 In order to effectively eliminate the impact of atmospheric turbulence and inter-symbol interference in faster-than-Nyquist（FTN） optical wireless communication systems, and further enhance the system’s bit error rate（BER） performance, a novel end-to-end FTN optical wireless communication system is proposed by incorporating an attention mechanism and long short-term memory（LSTM） network into a convolutional autoencoder.The LSTM network is introduced at the transmitter to extract FTN signal features considering their dependency on adjacent signals.Meanwhile, the receiver employs an attention mechanism to improve its ability in recovering the received signal.Additionally, a residual network is integrated into the system to achieve higher classification accuracy and accelerate convergence speed.Furthermore, the optimal training signal-to-noise ratio（SNR） under different conditions is discussed.Simulation results demonstrate that the improved system achieves comparable BER performance as Nyquist system within the Mazo limit.Moreover, it effectively addresses the “error floor” issue under strong turbulence channels by controlling the false symbol rate to below 10^-3 when the SNR is 32 dB.更多还原