Low-Complexity Digital Predistortion for Fully-Connected Hybrid Beamforming MIMO Systems with Memory Depth Adaptation

This paper proposes a low-complexity digital predistortion (DPD) method for fully-connected hybrid beamforming (FC-HBF) systems. By analyzing the coupling relationship between the memory effects and nonlinear orders, a memory depth adaptive adjustment strategy is introduced to reduce model complexity by constraining the memory depth of high-order basis functions. Theoretical analysis demonstrates that the long-memory effects of high-order nonlinear terms contribute weakly and can be pruned without significant performance degradation. Simulation results on a 4-stream 256-element FC-HBF system show that the proposed model reduces parameters by 60% compared to the traditional multi-variable polynomial (MVP) model, and its linearization performance even surpasses that of the MVP model. For individual beam signal, the normalized mean square error (NMSE) improvement can even reach up to 1.5dB.