A Module to Enhance the Generalization Ability of End-to-End Deep Learning Systems in Optical Fiber Communications

AnAdapter module is developed to increase the generalization capability of the end-to-end learning system for optical fiber communication systems. The Adapter has an interpretable structure and can be inserted and used without changing the original signal processing structure. The Adapter module can improve the system generalization capability and achieve the correct demapping within the transmission distance fluctuation range of $\pm$100 km and the power fluctuation range of $\pm$0.5 dBm. In addition, using an Adapter can improve the performance of optical communication by modifying the equalization algorithm without altering the structure of the existing transmission system.