Trade-Offs in Decentralized Multi-Antenna Architectures: Sparse Combining Modules for WAX Decomposition

TL;DR

This paper explores decentralized multi-antenna architectures for base stations, focusing on the WAX decomposition to optimize processing trade-offs, and introduces sparse combining modules for efficient hardware implementation.

Contribution

It provides explicit constructions of sparse linear combining modules for the WAX decomposition, enabling decentralized processing with arbitrary levels of decentralization.

Findings

Developed new sparse combining module constructions.

Facilitated decentralized calculation of WAX decomposition.

Enhanced hardware efficiency for multi-antenna processing.

Abstract

With the increase in the number of antennas at base stations (BSs), centralized multi-antenna architectures have encountered scalability problems from excessive interconnection bandwidth to the central processing unit (CPU), as well as increased processing complexity. Thus, research efforts have been directed towards finding decentralized receiver architectures where a part of the processing is performed at the antenna end (or close to it). A recent paper put forth an information-lossless trade-off between level of decentralization (inputs to CPU) and decentralized processing complexity (multiplications per antenna). This trade-off was obtained by studying a newly defined matrix decomposition--the WAX decomposition--which is directly related to the information-lossless processing that should to be applied in a general framework to exploit the trade-off. {The general framework consists of three stages: a set of decentralized filters, a linear combining module, and a processing matrix applied at the CPU; these three stages are linear transformations which can be identified with the three constituent matrices of the WAX decomposition. The previous work was unable to provide explicit constructions for linear combining modules which are valid for WAX decomposition, while it remarked the importance of these modules being sparse with 1s and 0s so they could be efficiently implemented using hardware accelerators.} In this work we present a number of constructions, as well as possible variations of them, for effectively defining linear combining modules which can be used in the WAX decomposition. Furthermore, we show how these structures facilitate decentralized calculation of the WAX decomposition for applying information-lossless processing in architectures with an arbitrary level of decentralization.