Deep Autoencoder-based Z-Interference Channels
Xinliang Zhang, Mojtaba Vaezi · 2023
A deep autoencoder (DAE)-based communication over the two-user Z-interference channel (ZIC) is introduced in this paper. The proposed DAE-ZIC is designed to minimize the bit error rate (BER) in the presence of interference by jointly optimizing the encoders and decoders. Effectively, this is an end-to-end communication that designs new constellations for the ZIC. Normalization layers are embedded in the proposed DAE design to realize an average power constraint so that there are no regular shape restrictions on the constellation symbols. We compare the performance of the DAE-ZIC with two baseline methods, which are ZIC with regular and rotated constellations. Simulation results show a significant gain in BER reduction. On average, in weak, moderate, and strong regimes, 31%–75% BER improvement is achieved compared to the best existing methods.