Ellipsoidal Manifold Optimization for Distributed Antenna Beamforming

TL;DR

This paper introduces a novel Riemannian manifold optimization approach for distributed antenna beamforming, reducing problem dimensionality and improving computational efficiency while maintaining optimality.

Contribution

It generalizes the low-dimensional subspace property to per-cluster power constraints and develops a Riemannian conjugate gradient algorithm for efficient beamforming optimization.

Findings

Achieves same local optima as benchmark methods

Significantly higher computational efficiency

Better scalability with more antenna clusters

Abstract

This paper addresses the weighted sum-rate (WSR) maximization problem in a downlink distributed antenna system subject to per-cluster power constraints. This optimization scenario presents significant challenges due to the high dimensionality of beamforming variables in dense antenna deployments and the structural complexity of multiple independent power constraints. To overcome these difficulties, we generalize the low-dimensional subspace property--previously established for sum-power constraints--to the per-cluster power constraint case. We prove that all stationary-point beamformers reside in a reduced subspace spanned by the channel vectors of the corresponding antenna cluster. Leveraging this property, we reformulate the original high-dimensional constrained problem into an unconstrained optimization task over a product of ellipsoidal manifolds, thereby achieving significant dimensionality reduction. We systematically derive the necessary Riemannian geometric structures for this specific manifold, including the tangent space, Riemannian metric, orthogonal projection, retraction, and vector transport. Subsequently, we develop a tailored Riemannian conjugate gradient algorithm to solve the reformulated problem. Numerical simulations demonstrate that the proposed algorithm achieves the same local optima as standard benchmarks, such as the weighted minimum mean square error (WMMSE) method and conventional manifold optimization, but with substantially higher computational efficiency and scalability, particularly as the number of antenna clusters increases.