Neural Network-Based Digital Predistortion and Self-Interference Cancellation in a Quadrature Balanced Full Duplex Transmitter

Abstract

This work presents a neural network (NN) implementation of a digital self-interference cancellation (SIC) filter and a digital predistortion (DPD) linearizer in a quadrature balanced full duplex (FD) transceiver front-end. A quantitative description of the NNs design and functionality is laid out. The proposed algorithms were evaluated in measurements using a discrete-component quadrature balanced RF front-end and a 20 MHz 802.11ac WiFi signal with 10 dB peak-to-average power ratio (PAPR) around the center frequency of 2.4 GHz. At 13 dBm average transmit (TX) power, the NN-DPD corrects TX error vector magnitude (EVM) by 13 dB to the value of -41.5 dB. Total TX-RX isolation of 50 dB is demonstrated in the RF domain, out of which 20 dB is contributed by the passive TX-RX isolation and 30 dB by active TX leakage suppression using the NN SIC filter.