Massive MIMO Antenna Selection: Asymptotic Upper Capacity Bound and Partial CSI

TL;DR

This paper derives an asymptotic upper capacity bound for massive MIMO antenna selection with partial CSI, demonstrating that acquiring limited CSI can achieve most of the capacity, and proposes an efficient adaptive selection algorithm.

Contribution

It introduces a low-complexity asymptotic capacity bound for massive MIMO antenna selection with partial CSI and proposes an adaptive algorithm based on the Pareto principle.

Findings

The asymptotic capacity bound closely matches numerical simulations.

Most of the channel capacity can be achieved with limited CSI acquisition.

The proposed adaptive algorithm significantly reduces CSI and computational requirements.

Abstract

Antenna selection (AS) is regarded as the key promising technology to reduce hardware cost but keep relatively high spectral efficiency in multi-antenna systems. By selecting a subset of antennas to transceive messages, AS greatly alleviates the requirement on Radio Frequency (RF) chains. This paper studies receive antenna selection in massive multiple-input multiple-output (MIMO) systems. The receiver, equipped with a large-scale antenna array whose size is much larger than that of the transmitter, selects a subset of antennas to receive messages. A low-complexity asymptotic approximated upper capacity bound is derived in the limit of massive MIMO systems over independent and identical distributed (i.i.d.) Rayleigh flat fading channel, assuming that the channel side information (CSI) is only available at the receiver. Furthermore, numerical simulations are provided to demonstrate the approximation precision of the asymptotic results and the tightness of the capacity bound. Besides the asymptotic analysis of the upper bound, more discussions on the ergodic capacity of the antenna selection systems are exhibited. By defining the number of corresponding rows in the channel matrix as the amount of acquired CSI, the relationship between the achievable channel capacity and the amount of acquired CSI is investigated. Our findings indicate that this relationship approximately follows the Pareto principle, i.e., most of the capacity can be achieved by acquiring a small portion of full CSI. Finally, on the basis of this observed law, an adaptive AS algorithm is proposed, which can achieve most of the transmission rate but requires much less CSI and computation complexity compared to state-of-the-art methods.