Correlation-Aware Greedy User Grouping for Zero-Forcing Precoding in Massive MIMO Downlink Systems

Abstract

Massive multiple-input multiple-output (MIMO) systems achieve high spectral efficiency through spatial multiplexing, where zero-forcing (ZF) precoding is widely adopted due to its effective interference suppression and moderate computational complexity. However, the performance of ZF precoding is highly dependent on user grouping, since strong intra-group channel correlation results in ill-conditioned Gram matrices and severe noise enhancement. Existing grouping strategies often require complicated optimization procedures or direct evaluation of precoding performance, leading to increased computational complexity. This paper proposes CorrGreedy, a low-complexity correlation-aware user grouping framework for group-wise ZF precoding under fixed-size grouping constraints. The proposed method introduces a correlation-based surrogate objective by minimizing the off-diagonal energy of normalized group Gram matrices, and employs a swap-based greedy refinement algorithm to improve user partitions without repeated ZF precoder evaluations. Theoretical analysis establishes the relationship between the proposed correlation metric, Gram matrix conditioning, and ZF noise enhancement. Extensive simulations under spatially correlated Massive MIMO channels demonstrate that CorrGreedy significantly improves spectral efficiency, bit error rate performance, and energy efficiency compared with conventional ZF, MMSE, clustering-based grouping, and existing ZF-oriented user grouping schemes. Moreover, convergence and complexity evaluations verify the effectiveness and scalability of the proposed algorithm, while statistical analysis confirms that the proposed correlation metric accurately reflects ZF performance variation.