Cluster-Based Cell-Free Massive MIMO Systems: A Novel Framework to Enhance Spectral Efficiency with Low Complexity

Abstract The issue of diminished spectral efficiency (SE) of the downlink (DL) transmission in distributed cell-free massive MIMO (CF-mMIMO) systems poses a significant challenge in terms of user equipment (UE) performance when compared to their centralized CF-mMIMO counterparts. The primary root cause of this issue can be attributed to the reduced efficacy of distributed precoders, which are devised using local channel state information (CSI) in distributed systems. This reduced efficacy becomes particularly pronounced in terms of interference mitigation when compared to centralized precoders. To address this issue, this paper proposes a novel architectural framework for CF-mMIMO systems, referred to herein as the "cluster-based structure." Within this innovative structure, a hybrid amalgamation of centralized and distributed configurations is employed, complemented by the introduction of a unique cluster arrangement for the access points (APs) within the network. In this design, the CSI of APs within each cluster is collectively shared within a local processor unit. Consequently, by harnessing this enhanced repository of local channel information, local precoders are formulated, which facilitate more effective interference mitigation with reduced computational complexity compared to the centralized approach. This approach ultimately results in a significantly augmented SE when contrasted with the distributed architecture. In this paper, the DL signal-to-interference-plus-noise ratio (SINR) of the UEs in this new architecture is derived analytically, and two precoders, namely maximum ratio (MR) and minimum mean square error (MMSE), are proposed for the new architecture. Furthermore, an elucidation of the computational complexity associated with the MMSE precoder in the context of the cluster-based framework will be presented, drawing a comparative analysis with both centralized and distributed structural configurations. The simulation results unequivocally demonstrate that within the cluster-based framework, the optimal SE for the network is attained when utilizing four clusters in conjunction with the MMSE precoding technique, leading to a notable reduction in computational complexity exceeding \(85%\). Importantly, this approach surpasses the SE performance of the centralized structure.