Pilot Resource Management for Channel Estimation Error Reduction in Digital Twin-Assisted User-Centric Cell-Free Massive MIMO Networks

This paper addresses channel estimation error (CEE) mitigation in user-centric cell-free (CF) massive multiple-input multiple-output (mMIMO) systems under pilot contamination (PC). To suppress the PC, we formulate an optimization problem to minimize the normalized CEE through the joint treatment of the pilot power allocation and the pilot assignment. However, the NP-hard nature of the problem makes finding a global optimal solution computationally infeasible. To tackle this challenge, we decompose the problem into two sub-problems: the fractional programming (FP) technique is employed for pilot power allocation, and a multi-agent reinforcement learning (RL) approach is applied for pilot assignment. The RL-based scheme, however, involves iterative action generation and reward computation, leading to significant overhead from repeated information exchanges. To mitigate this issue, we leverage a digital twin (DT) of the CF mMIMO system, utilizing its virtual environment and contextual awareness to optimize CEE reduction with minimal overhead. By employing closed-form expressions, the proposed schemes achieve computational efficiency without reliance on complex numerical solvers. Finally, we solve the original NP-hard problem via the block-coordinate descent and sequential optimization methods. Numerical evaluations demonstrate that the proposed FP-based pilot power allocation scheme improves the 95%-likely spectral efficiency (SE) by up to 15% and reduces the average pilot transmission power by up to 83% compared to existing methods. Furthermore, the pilot assignment scheme enhances the average SE performance by up to 7.4% over state-of-the-art pilot assignment approaches. These results validate the effectiveness of our proposed framework in reducing CEE, leading to enhanced improvement in system performance in the presence of PC.