Trade-offs in Decentralized Multi-Antenna Architectures: The WAX Decomposition

TL;DR

This paper introduces the WAX decomposition, a novel matrix factorization, to analyze the trade-offs between decentralization level and interconnection bandwidth in multi-antenna base station architectures, balancing complexity and efficiency.

Contribution

It proposes a flexible framework for centralized and decentralized architectures and introduces the WAX decomposition to explore their operational limits.

Findings

WAX decomposition enables information-lossless processing in decentralized architectures.

The framework quantifies the trade-offs between decentralization and interconnection bandwidth.

Operational limits of decentralized processing are characterized using the WAX decomposition.

Abstract

Current research on multi-antenna architectures is trending towards increasing the amount of antennas in the base stations (BSs) so as to increase the spectral efficiency. As a result, the interconnection bandwidth and computational complexity required to process the data using centralized architectures is becoming prohibitively high. Decentralized architectures can reduce these requirements by pre-processing the data before it arrives at a central processing unit (CPU). However, performing decentralized processing introduces also cost in complexity/interconnection bandwidth at the antenna end which is in general being ignored. This paper aims at studying the interplay between level of decentralization and the associated complexity/interconnection bandwidth requirement at the antenna end. To do so, we propose a general framework for centralized/decentralized architectures that can explore said interplay by adjusting some system parameters, namely the number of connections to the CPU (level of decentralization), and the number of multiplications/outputs per antenna (complexity/interconnection bandwidth). We define a novel matrix decomposition, the WAX decomposition, that allows information-lossless processing within our proposed framework, and we use it to obtain the operational limits of the interplay under study. We also look into some of the limitations of the WAX decomposition.