Attention Mechanism Aided Signal Detection in Backscatter Communications With Insufficient Training Data

In backscatter communications (BackCom), the tag signal is subject to both double channel fading and direct link interference (DLI). With DLI being typically much stronger than the backscattered signal, the detection of the tag signal becomes challenging. To address this issue, we propose a novel deep learning (DL) model known as the backscattered signal recovery network (BSRnet). BSRnet is designed to recover the backscattered signal even in the presence of a strong DLI at the reader side. Given the potential scarcity of training data, BSRnet incorporates an innovative attention mechanism. This mechanism guides the model to focus on and optimize crucial learnable parameters. Distinct from traditional DL models, BSRnet can circumvent bottlenecks created by zero-padding and sampling operations and mitigate issues related to the distribution mismatch. Simulation results highlight the superior performance of BRSnet in comparison to existing models, confirming its advanced design strategy. Furthermore, the importance of addressing these bottlenecks is underscored through two ablation studies, validating the effectiveness of the attention mechanism.