Nonlinear Interference Alignment in a One-dimensional Space

TL;DR

This paper introduces a novel nonlinear interference alignment method called Interference Dissolution (ID) that effectively partitions a one-dimensional space into two fractional dimensions, achieving near-capacity performance across all SNR levels.

Contribution

The paper proposes a new nonlinear interference alignment technique, ID, which outperforms existing schemes by achieving near-capacity rates at finite SNR in one-dimensional channels.

Findings

ID achieves a rate of two symbols per channel use.

ID provides 0.5 DoF per symbol.

Sum achievable rate is at most one bit away from capacity.

Abstract

Real interference alignment is efficient in breaking-up a one-dimensional space over time-invariant channels into fractional dimensions. As such, multiple symbols can be simultaneously transmitted with fractional degrees-of-freedom (DoF). Of particular interest is when the one dimensional space is partitioned into two fractional dimensions. In such scenario, the interfering signals are confined to one sub-space and the intended signal is confined to the other sub-space. Existing real interference alignment schemes achieve near-capacity performance at high SNR for time-invariant channels. However, such techniques yield poor achievable rate at finite SNR, which is of interest from a practical point of view. In this paper, we propose a radically novel nonlinear interference alignment technique, which we refer to as Interference Dissolution (ID). ID allows to break-up a one-dimensional space into two fractional dimensions while achieving near-capacity performance for the entire SNR range. This is achieved by aligning signals by signals, as opposed to aligning signals by the channel. We introduce ID by considering a time-invariant MISO channel. This channel has a one-dimensional space and offers one DoF. We show that, by breaking-up the one dimensional space into two sub-spaces, ID achieves a rate of two symbols per channel use while providing $\frac{1}{2}$ DoF for each symbol. We analyze the performance of the proposed ID scheme in terms of the achievable rate and the symbol error rate. In characterizing the achievable rate of ID for the entire SNR range, we prove that, assuming Gaussian signals, the sum achievable rate is at most one bit away from the capacity. We present numerical examples to validate the theoretical analysis. We also compare the performance of ID in terms of the achievable rate performance to that of existing schemes and demonstrate ID's superiority.