Learning-Based Power Allocation for Per-Beam Timing Advance Enabled Asynchronous Cell-Free Massive MIMO

Abstract

Asynchronous transmission in cell-free massive multiple-input multiple-output (CF-mMIMO) systems gives rise to delay-induced inter-carrier interference (ICI) and inter-symbol interference (ISI), which can significantly reduce spectral efficiency (SE). Although per-beam timing advance (PBTA) enables beam-level timing compensation for user-specific desired transmissions, residual asynchronous interference generally remains due to mismatched timing references across different links. This paper investigates user-level power allocation for PBTA-enabled asynchronous CF-mMIMO systems. To further suppress the residual coupled interference after beam-domain timing alignment, a twin delayed deep deterministic policy gradient (TD3)-based power-allocation scheme is developed to exploit channel and interference information for user-level power optimization. Numerical results show that the proposed method consistently outperforms conventional asynchronous baselines and maintains clear performance gains over a wide range of asynchronous conditions and antenna configurations.