Joint Self-Interference Cancellation and Data Estimation for OFDM Based Full-Duplex Communication Systems.

In the receiver of a wireless communication system, many tasks are traditionally accomplished by employing model-based approaches. These model-based methods generally provide decent performance, however, in case of model inaccuracies, simplifications, or wrong statistical assumptions, performance degradation is to be expected. The aforementioned issues could be mitigated by using data-driven methods like neural networks (NNs). In this work, we investigate employing NNs for self-interference (SI) cancellation and equalization in full-duplex communication systems. We propose to optimize the corresponding two NNs jointly with the goal of minimizing the bit error ratio performance of the entire chain, instead of optimizing them individually for their corresponding tasks. We show that joint optimization of the NN-based SI canceller and equalizer enhances the performance, and we investigate and discuss why NNs can outperform model-based approaches for the considered system.