This study introduces an end-to-end communication learning system based on a generative adversarial network (GAN) after discussing the advantages of laser link communication. This improves the real-time and global optimization of the communication system. Moreover, this study introduces the Wasserstein GAN to resolve mode collapse and training instability in the training and application of a traditional GAN. Finally, the Wasserstein GAN is applied to the end-to-end communication system, and the experimental results show that the Wasserstein GAN can effectively simulate an additive Gaussian white noise channel and a lognormal channel, thus avoiding the training instability and mode collapse of the traditional GAN.