Deep learning-based power amplifier linearization in OFDM systems with unknown channel state information

Abstract

This paper presents an end-to-end deep learning-based approach for orthogonal frequency-division multiplexing (OFDM) communication systems impaired by nonlinear power amplifiers (PAs) and channel fading. The PA nonlinearity is modeled using the modified Rapp model, and simulations are performed on a 64-subcarrier OFDM system with a cyclic prefix (CP) of 8 and 16-quadrature amplitude modulation (16-QAM). The proposed autoencoder-based OFDM–PA (AE-OFDM-PA) system jointly optimizes the transmitter and receiver through end-to-end learning, enabling simultaneous compensation of both PA nonlinearities and channel distortions without requiring explicit channel state information (CSI) estimation. Instead, the model leverages embedded pilot sequences to learn the implicit CSI representation directly from data, allowing the receiver to correct amplitude and phase distortions adaptively. Simulation results demonstrate that AE-OFDM-PA significantly outperforms conventional OFDM and OFDM-PA systems, achieving over 70× block error rate (BLER) improvement compared with the uncompensated OFDM-PA system at an input back-off (IBO) of 3 dB. Furthermore, the proposed method achieves approximately 11.5 dB adjacent channel leakage ratio (ACLR) improvement over the classical memory polynomial digital predistortion (DPD) technique, while slightly reducing the peak-to-average power ratio (PAPR). Overall, AE-OFDM-PA provides a robust, spectrally efficient, and low-complexity solution for nonlinear and fading environments with unknown or varying CSI.