Toward Scalable Centralized Network-Assisted Full-Duplex Cell-Free Massive MIMO Networks

Abstract

Cell-free massive MIMO (CF-mMIMO) is a key candidate technology for 6G networks, enabling more uniform service quality and higher spectral efficiency (SE) through the coordination of distributed access points (APs). Conventional half-duplex (HD) time-division duplexing (TDD) statically partitions uplink (UL) and downlink (DL) resources, limiting adaptability to traffic asymmetry and reducing resource utilization. Dynamic duplexing approaches, such as network-assisted full-duplex (NAFD) (also known as dynamic TDD (DTDD)), can overcome this limitation but typically rely on computationally intensive optimization procedures that do not scale to the ultra-dense deployments envisioned for future wireless systems. This paper proposes a scalable centralized NAFD-enabled CF-mMIMO framework that combines interference-aware linear precoding and combining with scalable AP scheduling adapted to traffic asymmetry and spatial user heterogeneity. The proposed signal processing schemes preserve scalability while achieving performance very close to that of their non-scalable MMSE counterparts. Numerical results show that the DL performance of the scalable and non-scalable implementations is nearly indistinguishable, while the UL differences remain minor. Even under conservative cross-link interference (CLI) suppression assumptions, the top 10% of users can almost double their DL spectral efficiency compared with conventional HD-TDD operation, whereas the bottom 10% of UL users experience an improvement from approximately 4 to 5.25 bit/s/Hz. In the DL, however, the worst users may experience noticeable performance degradation despite the substantial gains achieved by the best users. Although with diminishing returns, these performance metrics can be improved by increasing the number of serving APs and/or antennas per AP. The results also identify the CLI mitigation levels required for NAFD operation to outperform HD-TDD and show that DL CLI is the dominant limiting factor. Overall, the findings demonstrate that scalable scheduling, interference-aware processing, and effective CLI suppression are key enablers for practical NAFD operation in ultra-dense CF-mMIMO networks.