Machine Learning for Ambient Backscatter Channel Estimation and Signal Detection: Opportunities and Challenges

As a promising low-power connection paradigm in the ubiquitous Internet of Things (IoT), ambient backscatter communication (AmBC) collects energy from ambient radio frequency (RF) signals while using them as carrier signals, which brings ultra-low power consumption and deployment cost. However, it has not been widely applied in practice because of its difficulties in weak signal detection. To overcome these difficulties, machine learning (ML)-based methods have been highlighted recently. ML methods can achieve accurate signal processing under a low receive signal-to-interference-plus-noise ratio (SINR) in unpredictable interference communication scenarios, benefiting from their outstanding inference and classification tools. In this survey, a brief review of AmBC is first introduced and the four-fold signal-receiving challenges of AmBC are discussed. After that, two key signal processing technologies, i.e., AmBC channel estimation and AmBC signal detection, are emphasized. The representative ML-based methods of AmBC channel estimation and AmBC signal detection are summarized, following their advantages and disadvantages. Finally, some valuable research directions on this topic are introduced to guide future research.