Neural Networks for Full‐Duplex Radios

This chapter covers technical aspects of digital self‐interference (SI) cancellation in full‐duplex radios in a self‐contained manner, using both conventional polynomial models and neural networks (NNs). It first describes how the nonlinear effects of various transceiver components can be modeled and derives a comprehensive nonlinear SI model. The chapter then describes various linear and nonlinear cancellation methods, with emphasis on a NN canceler. It analyzes and compares the computational complexity of each approach. For further illustration and comparison, the chapter provides experimental results using measured samples from a hardware testbed which demonstrate that a simple NN‐based nonlinear canceler can match the performance of a state‐of‐the‐art polynomial model for nonlinear cancellation with a significantly lower computational complexity. Finally, the chapter discusses a number of interesting future research directions.