Learnable Semi-Blind Receiver Design for Phase Noise Impaired OFDM Systems

Phase noise (PN) is a non-negligible detrimental factor for orthogonal frequency division multiplexing (OFDM) systems in high-frequency communication systems. In this paper, we focus on a joint channel estimation, PN compensation, and data detection (JCPD) problem and propose two intelligent semi-blind receivers for slow and fast time-varying channels respectively. In particular, a low-complexity JCPD algorithm based on the alternating direction method of multipliers (ADMM) and the minorization-maximization (MM) techniques is proposed, whose complexity per iteration is similar to the fast Fourier transform (FFT). Then, we unfold the iterations into model-driven networks by introducing trainable parameters, which avoids cumbersome parameter search and also results in faster convergence. Furthermore, for the fast time-varying channel subject to the inter-carrier interference (ICI) from both the channel and the PN, we propose an effective JCPD algorithm and also a low-complexity version by adopting the MM technique, along with the corresponding model-driven networks. Simulation results validate that our proposed JCPD algorithms for both slow and fast time-varying channels significantly outperform the existing algorithms with affordable complexities.