Adaptive Local Combining with Decentralized Decoding for Distributed Massive MIMO

Abstract: A major bottleneck in uplink distributed massive multiple-input multiple-output networks is the sub-optimal performance of local combining schemes, coupled with high fronthaul load and computational cost inherent in centralized large scale fading decoding (LSFD) architectures. This paper introduces a decentralized decoding architecture that fundamentally breaks from the conventional LSFD, by allowing each AP calculates interference-suppressing local weights independently and applies them to its data estimates before transmission. Furthermore, two generalized local zero-forcing (ZF) framework, generalized partial full-pilot ZF (G-PFZF) and generalized protected weak PFZF (G-PWPFZF), are introduced, where each access point (AP) adaptively and independently determines its combining strategy through a local sum spectral efficiency optimization that classifies user equipments (UEs) as strong or weak using only local information, eliminating the fixed thresholds used in PFZF and PWPFZF. To further enhance scalability, pilot-dependent combining vectors instead of user-dependent ones are introduced and are shared among users with the same pilot. The corresponding closed-form spectral efficiency expressions are derived. Numerical results show that the proposed generalized schemes consistently outperform fixed-threshold counterparts, while the introduction of local weights yields lower overhead and computation costs with minimal performance penalty compared to them.