Simultaneous Nonlinear Self-Interference Cancellation and Signal of Interest Recovery Using Dual Input Deep Neural Network in New Radio Access Networks

Abstract

An efficient method for simultaneous nonlinear self-interference (SI) cancellation and signal-of-interest (SOI) recovery is proposed and experimentally verified in a mmWave over fiber testbed, based on a specially designed dual-input deep neural network (DI-DNN). The mmWave band has been adopted in the U.S. to provide 5G wireless access services due to its wide-bandwidth, light-of-sight propagation, and inherent compatibility with small cell architecture. To further meet the uprising bandwidth-demanding services, the in-band full-duplex (IBFD) scheme is introduced to magnify the mmWave channel capacity for the 5G and beyond wireless communications. Fortunately, the mmWave band is more realistic to implement IBFD compared with the conventional sub-6GHz band, as it operates as a highly directional beam and shorter transmission distance resulting in a higher SOI to SI ratio. In our proposed solution, a DI-DNN is, for the first time, proposed and implemented to simultaneously realize the SI cancellation and the SOI recovery. Moreover, we firstly mitigate the nonlinearity arising from the transmitters of both the SI and SOI, as well as the nonlinear crosstalk from the SI to the SOI after detection by a nonlinear receiver. The demonstration of the DI-DNN joint with the experimental implementation in a mmWave over fiber system, promise spectral efficient, high-capacity signal transport in the 5G and beyond wireless communications.