Feasibility of 6G Sub-Band Full-Duplex: Cancellation Methods and Performance Results

In this paper, we examine artificial neural network (ANN) approaches to cancel self-interference (SI) in sub-band full-duplex (SBFD). SBFD operation divides the un-paired communication channel to uplink (UL) and downlink (DL) parts, which facilitates lowered latency and improved flexibility of resources based on the network requirements. Though the UL and DL parts do not overlap in their exact frequency resources, in the presence of nonlinear power amplifier (PA) in the base station (BS) transmitter (TX), the spectral regrowth around the DL carriers will overlap with the UL part, causing severe performance loss due to the SI. To combat this, the SI needs to be cancelled at the receiver (RX). For this, we formulate and compare several ANN approaches, which model the joint effects of the PA and the linear SI channel, both contributing to the exact SI waveform. Accompanied with RF measurements conducted with real-life hardware at 3.5 GHz, the notable performance gains obtained by employing the ANN approaches are evidenced and shown.