XZero: On Practical Cross-Technology Interference-Nulling for LTE-U/WiFi Coexistence

TL;DR

XZero is a practical system enabling LTE-U to nullify interference towards WiFi nodes without channel state information, using a tree-based search for efficient null beam configuration, significantly reducing interference and reconfiguration delay.

Contribution

XZero introduces a novel, practical interference nulling method for LTE-U/WiFi coexistence that operates without channel state information, using a feedback-driven tree search approach.

Findings

Achieves 15.7 dB interference reduction at WiFi nodes.

Reconfigures null beams in under a second, 10x faster than exhaustive search.

Proven effective in large indoor testbed with SDR and commodity hardware.

Abstract

LTE-U/WiFi coexistence can be significantly improved by placing so-called coexistence gaps in space through cross-technology interference-nulling (CTIN) from LTE-U BS towards WiFi nodes. Such coordinated co-existence scheme requires, for the exchange of control messages, a cross-technology control channel (CTC) between LTE-U and WiFi networks which was presented recently. However, it is unclear how a practical CTIN operates in the absence of channel state information which is needed for CTIN but cannot be obtained from the CTC. We present XZero, the first practical CTIN system that is able to quickly find the suitable precoding configuration used for interference nulling without having to search the whole space of angular orientations. XZero performs a tree-based search to find the direction for the null beam(s) by exploiting the feedback received from the WiFi AP on the tested null directions. We have implemented a prototype of XZero using SDR platform for LTE-U and commodity hardware for WiFi and evaluated its performance in a large indoor testbed. Evaluation results reveal on average a reduction by 15.7 dB in interference-to-noise ratio at the nulled WiFi nodes when using a ULA with four antennas. Moreover, XZero has a sub-second reconfiguration delay which is up to 10x smaller as compared to naive exhaustive linear search.