Scalable Subchannel Assignment for Cell-Free Massive MIMO Systems under Non-Uniform User Deployment

초록

Cell-free massive multiple-input multiple-output (CF-mMIMO) leverages many distributed access points (APs) that jointly serve user equipments (UEs) on the same time–frequency resource, yielding macro-diversity and spatial-multiplexing gains when APs substantially outnumber UEs (M \gg K). In practice, however, UE locations are often highly non-uniform, and dense local regions frequently arise where many nearby UEs compete for the same favorable APs. This hotspot-induced overlap increases co-channel interference, tightens pilot reuse, and reduces the desired-signal power per UE, thereby degrading network throughput. At the same time, subchannel scheduling in such dense regions must remain scalable, since waiting for full instantaneous channel state information (CSI) or repeated network-wide refinement can incur substantial signaling and computational overhead. To address this problem, we propose an LSFC-only pre-pilot subchannel scheduling framework for CF-mMIMO under non-uniform UE deployment, which constructs an interference-oriented assignment metric from large-scale fading coefficients (LSFCs) and separates strongly coupled UEs across subchannels before instantaneous CSI becomes available. We further show that the dominant within-subchannel interference components admit a conservative upper bound that increases with an LSFC-overlap surrogate, thereby relating the proposed metric to the interference structure reflected in the scheduling objective. Using this metric, the proposed scheduler determines the total number of UE–subchannel connections and then performs low-complexity subchannel index assignment, thereby controlling both the amount and placement of subchannel reuse. The scheduler operates before pilot signaling using only LSFCs, without requiring instantaneous CSI or additional signaling at the scheduling stage. Simulation results show that the proposed scheme improves the average downlink rate by 54.72% over benchmark methods in non-uniform dense user scenarios.

키워드