End-to-End Learning for Chromatic Dispersion Compensation in Optical Fiber Communication

Abstract

In this Letter, we investigate the chromatic dispersion compensation problem in optical fiber communication. An end-to-end autoencoder (AE) is proposed to replace the transceiver of the traditional intensity modulation direct detection system. To deal with the obstructed gradient return problem in end-to-end transmission, we introduce a generative adversarial network to simulate the channel transmission process and employ a square-law detector for incoherent detection to reduce the complexity. Simulation results show that the BER of the proposed system can be significantly cut down compared with the conventional electric domain compensation algorithms.