ScholarWorks@성균관대학교

초록

Channel estimation in backscatter communication (BackCom) systems remains a significant challenge due to the battery-free device nature, which often fail to respond to traditional pilot signals. While prior works have predominantly relied on simulation-based evaluations with ideal reception assumptions, such methods fall short in practical scenarios where pilot reachability and BackCom devices’ response cannot be guaranteed. In this work, we present the fully implemented and evaluated channel estimation framework for battery-free BackCom systems using a real-world testbed. We introduce Directional Random Phase Perturbation, a novel pilot beamforming technique that combines directional scanning with randomized phase perturbation to induce backscatter responses even in environments where pilot signal reachability is uncertain. We further propose a neural channel estimation model based on self-attention. The model is carefully designed to integrate both pilot signal features and Least Square-based (LS) initial estimates as inputs and to explicitly decompose the channel into transmit and receive components for improved estimation accuracy. In addition, we introduce task-specific data augmentation methods, including random phase shifting and antenna pseudo-flipping, to enhance the model under limited training data. Evaluated on the real testbed, our framework improves beamforming gain by up to 73.8% and achieves 28.8% higher data transmission success rates compared to conventional methods.

키워드