Spatial Intercell Interference Cancellation with CSI Training and Feedback

TL;DR

This paper analyzes practical intercell interference cancellation (ICIC) in a 2-cell network, optimizing training and feedback strategies to enhance throughput and adapt to varying SNR conditions, transforming interference-limited networks into noise-limited ones.

Contribution

It introduces an adaptive ICIC scheme with optimized training and feedback resource allocation considering practical CSI feedback models and channel conditions.

Findings

ICIC can significantly improve throughput even with low-rate feedback.

Optimal power and feedback resource allocation depend on feedback type and SNR.

Adaptive switching between beamforming and ICIC enhances performance.

Abstract

We investigate intercell interference cancellation (ICIC) with a practical downlink training and uplink channel state information (CSI) feedback model. The average downlink throughput for such a 2-cell network is derived. The user location has a strong effect on the signal-to-interference ratio (SIR) and the channel estimation error. This motivates adaptively switching between traditional (single-cell) beamforming and ICIC at low signal-to-noise ratio (SNR) where ICIC is preferred only with low SIR and accurate channel estimation, and the use of ICIC with optimized training and feedback at high SNR. For a given channel coherence time and fixed training and feedback overheads, we develop optimal data vs. pilot power allocation for CSI training as well as optimal feedback resource allocation to feed back CSI of different channels. Both analog and finite-rate digital feedback are considered. With analog feedback, the training power optimization provides a more significant performance gain than feedback optimization; while conversely for digital feedback, performance is more sensitive to the feedback bit allocation than the training power optimization. We show that even with low-rate feedback and standard training, ICIC can transform an interference-limited cellular network into a noise-limited one.