A Cascade Deep Neural Network Approach for Compensation of RF Hardware Impairments in Wireless Transmitters

Abstract

Radio frequency (RF) hardware impairments can affect the transmit and receive signals in several ways, including deterioration of signal integrity, loss of power/spectral efficiency, and system performance degradation. While substantial efforts have been invested in modeling and mitigating the nonlinear effect of power amplifiers (PAs), accurate modeling of the entire RF hardware component chain is missing. This paper addresses the various imperfections and impairments in RF transceivers, particularly within analog circuits, such as digital-to-analog-converter (DAC) nonlinear distortion, in-phase/quadrature-phase (IQ) imbalance, and PA nonlinearity. We introduce a Cascade Neural Network Digital Predistortion (Cascade-NNDPD) model to compensate for these impairments. The proposed model employs a two-stage neural network approach: the first stage utilizes a phase normalized time-delay neural network, termed PNTDNN for PA nonlinearities, while the second stage deploys an additional network (MLP, LSTM, BiLSTM, or GRU) to address remaining distortions. Our results demonstrate the potential of Cascade-NNDPD design in mitigating RF hardware impairments, thus enhancing the performance and reliability of wireless communications.