Performance Evaluation of Autoencoder for Coding and Modulation in Wireless Communications
Jialong Xu, Wei Ren Chen, Bo Ai, Ruisi He, Yujian Li, Junhong Wang, Tutun Juhana, Adit Kurniawan · 2019
The end-to-end autoencoder is a novel and attracting concept to innovate communication system architecture. In its training stage, the end-to-end autoencoder needs differentiable channel models to execute back-propagation algorithm. This shortage impedes the further development of the end-to-end autoencoder. Estimating the channel model by reinforcement learning is a good way to alleviate this problem. However, these training methods increase the difficulty in the training stage. In this paper, we set up a general channel model and employ the measured channel data as input data to alleviate this problem. We compare the performance of the end-to-end autoencoder and Hamming code modulated by Binary Phase Shift Keying under additive white Gaussian noise channel and Rayleigh channel and explore the performance of the end-to-end encoder under the real scenarios. Furthermore, we investigate the performance of the end-to-end autoencoder under mismatched channels. The results demonstrate that the performance of the end-to-end autoencoder is similar with Hamming code and has a significant performance improvement under mismatched channels.