Vector Similarity Search-Based MCS Selection for Iterative Signal Detection in Massive Multi-User MIMO-OFDM Systems

This paper proposes a novel vector similarity search (VSS)-based modulation and coding scheme (MCS) selection for massive multi-user multiple-input multiple-output orthogonal frequency division multiplexing (MU-MIMO-OFDM) systems that employ uplink multi-user detection (MUD) based on iterative signal estimation. To maximize the uplink throughput of MU-MIMO-OFDM systems, it is necessary to assign a carefully selected MCS to each user according to an accurate prediction of the mutual information (MI) that can be achieved with MUD based on the knowledge of the estimated channel state information (CSI). However, since the detection accuracy of iterative MUDs, such as expectation propagation (EP), varies depending on the convergence characteristics, it is challenging to analytically predict the achievable MI in the presence of MUD. To address this difficulty, we propose a novel method for predicting the achievable MI by creating a vector database (VDB) offline that stores feature vectors (keys) computed from CSI and the actual MI (values) achieved with iterative MUD, and then searching this VDB online using approximate nearest neighbors (ANN) search, which enables VSS at ultra-high speed. Simulation results show that the MCS selection based on the proposed MI prediction achieves higher uplink throughput than the conventional schemes in MU-MIMO-OFDM systems using the EP-based MUD.