Orthogonal frequency division multiplexing/offset quadrature amplitude modulation (OFDM/OQAM) is a promising modulation candidate for passive optical network (PON) due to its high flexibility, better time–frequency focusing characteristic, great resistance to the inter symbol and inter carrier interferences and higher spectrum efficiency compared to OFDM. However, the intrinsic imaginary interference together with linear and nonlinear distortions make it more difficult to recover the transmitted OFDM/OQAM signal at the receiver side. To mitigate the transmission impairments, a modified convolutional neural network (CNN) is utilized to learn the channel state information and the constellation demapping mechanism for OFDM/OQAM-PON. The distorted received signals are equalized implicitly to obtain the transmitted binary bits directly. The simulation results show that the CNN based receiver (Rx) can compensate the linear and nonlinear distortions more effectively compared to traditional pilot-based Rx, especially for the high order modulation long-reach PON.

正交频分复用/偏移正交幅度调制（OFDM / OQAM）具有无与伦比的灵活性，更好的时频聚焦特性，对符号间和载波间干扰的强大抵抗力，是无源光网络（PON）的有希望的调制候选者。与OFDM相比，频谱效率更高。但是，固有的虚部干扰以及线性和非线性失真使得在接收器端恢复发送的OFDM / OQAM信号更加困难。为了减轻传输损伤，使用了改进的卷积神经网络（CNN）来学习信道状态信息和OFDM / OQAM-PON的星座解映射机制。失真的接收信号被隐式均衡，以直接获得发送的二进制位。