A deep learning-based adaptive receiver for full-duplex systems

To accommodate the growth of data traffic of 6G and beyond networks, achieving a significant improvement in spectrum efficiency is inevitable. Full-duplex systems are very promising since they have the potential of increasing the spectral efficiency compared to half-duplex systems. The main challenge facing the deployment of full-duplex systems is self-interference. Conventional receivers include series of processing blocks that recover the desired signal after removing the effect of self-interference. In the presence of noise and fading channels, the received signal becomes distorted and the recovery is thus challenging. Deep learning algorithms have shown great success in efficient parameter estimation, as well as adaptive decision making. In this study, a deep learning-based full-duplex receiver, namely FDDR, is proposed to replace the receiver's entire information recovery process of the full-duplex system rather than optimizing each processing module of the receiver separately. Simulation results show that FDDR approaches the same Bit Error Rate performance as a conventional receiver that uses a Kalman Filter as a channel estimator with 80% less complexity. The Bit Error Rate performance of FDDR is also tested in the case of Doppler frequency mismatch between the training and testing phases, multiple self-interference reflections and cyclic prefix free scenarios and shows superior performance.