Joint Spatial Division and Multiplexing: Opportunistic Beamforming and User Grouping

TL;DR

This paper explores the use of Joint Spatial Division and Multiplexing (JSDM) in massive MIMO systems, proposing user grouping and opportunistic beamforming strategies to improve capacity and reduce feedback overhead in FDD systems.

Contribution

It introduces a novel user grouping algorithm for large antenna systems and analyzes capacity gains with opportunistic beamforming and block diagonalization.

Findings

Opportunistic beamforming with user selection improves capacity in fixed antenna regimes.

Block diagonalization achieves optimal sum-rate scaling with correlated groups.

Probabilistic user scheduling reduces feedback requirements in large systems.

Abstract

Joint Spatial Division and Multiplexing (JSDM) is a recently proposed scheme to enable massive MIMO like gains and simplified system operations for Frequency Division Duplexing (FDD) systems. The key idea lies in partitioning the users into groups with approximately similar covariances, and use a two stage downlink beamforming: a pre-beamformer that depends on the channel covariances and minimizes interference across groups and a multiuser MIMO precoder for the effective channel after pre-beamforming, to counteract interference within a group. We first focus on the regime of a fixed number of antennas and large number of users, and show that opportunistic beamforming with user selection yields significant gain, and thus, channel correlation may yield a capacity improvement over the uncorrelated "isotropic" channel result of Sharif and Hassibi. We prove that in the presence of different correlations among groups, a block diagonalization approach for the design of pre-beamformers achieves the optimal sum-rate scaling. Next, we consider the regime of large number of antennas and users, where user selection does not provide significant gain. Here, we propose a simplified user grouping algorithm to cluster users into groups when the number of antennas becomes very large, in a realistic setting where users are randomly distributed and have different angles of arrival and angular spreads depending on the propagation environment. Our subsequent analysis leads to a probabilistic scheduling algorithm, where users within each group are preselected at random based on probabilities derived from the large system analysis, depending on the fairness criterion. This is advantageous since only the selected users are required to feedback their channel state information (CSIT).