An Experimental Investigation of SIMO, MIMO, Interference-Alignment (IA) and Coordinated Multi-Point (CoMP)

TL;DR

This paper experimentally compares interference alignment (IA) and coordinated multi-point (CoMP) transmission in a multi-antenna system, revealing CoMP's superior performance at higher backhaul costs and highlighting the impact of hardware imperfections on real-world results.

Contribution

It provides the first experimental implementation and comparison of IA and CoMP with detailed channel measurements and models dirty-RF effects impacting performance.

Findings

CoMP outperforms IA and single-user MIMO in measured scenarios.

Theoretical performance estimates are significantly higher than actual measurements.

Dirty-RF effects like phase noise reduce real-world system performance.

Abstract

In this paper we present experimental implementations of interference alignment (IA) and coordinated multi-point transmission (CoMP). We provide results for a system with three base-stations and three mobile-stations all having two antennas. We further employ OFDM modulation, with high-order constellations, and measure many positions both line-of-sight and non-line-of-sight under interference limited conditions. We find the CoMP system to perform better than IA at the cost of a higher back-haul capacity requirement. During the measurements we also logged the channel estimates for off-line processing. We use these channel estimates to calculate the performance under ideal conditions. The performance estimates obtained this way is substantially higher than what is actually observed in the end-to-end transmissions---in particular in the CoMP case where the theoretical performance is very high. We find the reason for this discrepancy to be the impact of dirty-RF effects such as phase-noise and non-linearities. We are able to model the dirty-RF effects to some extent. These models can be used to simulate more complex systems and still account for the dirty-RF effects (e.g., systems with tens of mobiles and base-stations). Both IA and CoMP perform better than reference implementations of single-user SIMO and MIMO in our measurements.