A Learnable Multi-Layer Structure for Joint Tx/Rx IQIs and CFO Compensation in FDE Systems

Abstract

This paper proposes radio frequency (RF) impairments compensation schemes for frequency domain equalization (FDE) systems, including single-carrier cyclic-prefix (SCCP) and orthogonal frequency division multiplexing (OFDM). Our approach addresses impairments caused by transmitter (Tx) and receiver (Rx) in-phase/quadrature-phase imperfections (IQIs) as well as carrier frequency offset (CFO) in SCCP and OFDM systems. We introduce a distortion compensation approach with a multi-layer structure that effectively compensates distortions caused by Tx/Rx IQIs, CFO, and frequency selective channel at each layer. Based on this approach, we develop a learnable multi-layer filter in a model-driven deep learning framework, where all weights of the multi-layer filter are determined solely from the final layer output of the filter in a data-driven manner. Simulation results demonstrate that the proposed learnable multi-layer filter achieves comparable bit error rate (BER) performance to its non-adaptive counterpart that assumes perfect channel state information, while outperforming existing adaptive methods in terms of convergence speed and robustness in the online adaptation scenarios with limited training data. Furthermore, we present a technique to estimate the distortion parameters from the trained filter weights, potentially enabling real-time system monitoring. The proposed methodology with the multi-layer structure is promising for enhancing reliability and performance across various wireless communication configurations in the presence of hardware impairments, particularly in scenarios where accurate channel and impairment information is unavailable.