Augmentation of Self-Interference Cancellation for Full-Duplex Using NARX Neural Networks

A self-interference cancellation augmentation technique based on a NARX (Nonlinear Autoregressive Exogenous) network model is implemented and evaluated on an OFDM-based full-duplex system testbed operating at 2.4 GHz. In a comparison with the state-of-the-art polynomial models, our experimental results demonstrate the significant computational efficiency of the proposed NARX model. Specifically, the NARX model with one hidden layer reduces computations by 83.3% while achieving the same cancellation level within a bandwidth of 2 MHz.