Non-Linear Digital Self-Interference Cancellation for In-Band Full-Duplex Radios Using Neural Networks

摘要

Full-duplex systems require very strong self-interference cancellation inorder to operate correctly and a significant portion of the self-interferencesignal is due to non-linear effects created by various transceiver impairments.As such, linear cancellation alone is usually not sufficient and sophisticatednon-linear cancellation algorithms have been proposed in the literature. Inthis work, we investigate the use of a neural network as an alternative to thetraditional non-linear cancellation method that is based on polynomial basisfunctions. Measurement results from a full-duplex testbed demonstrate that asmall and simple feedforward neural network canceller works exceptionally well,as it can match the performance of the polynomial non-linear canceller withpotentially significantly lower computational complexity.