Signal Detection for Ambient Backscatter Communications Using Unsupervised Learning

Abstract

Ambient backscatter communication (AmBC) has attracted much attention recently due to its low-cost and low power consumption in connecting billions of devices for future Internet-of-Things (IoT). Signal detection in AmBC can be a challenging issue because of the difficulty in estimating relevant channels, and the spectrum sharing nature of the AmBC system. In this paper, a novel machine learning inspired signal detection method is proposed for AmBC systems. This method exploits the features of the received signals directly and groups them into clusters through unsupervised learning. Furthermore, labeled bits from tag are transmitted for cluster-bit mapping to assist signal detection without estimating the channel coefficients and the noise power. Two detection approaches are proposed for the cases that the spreading gain N ≫ 1 and N = 1, where the features follow different mixture distributions. The detection thresholds are derived to optimize the detection performance using the learned parameters. Finally, extensive simulation results are provided to verify the performance of the proposed schemes.