Deep Unfolding-Based Joint Beamforming and Detection Design for Ambient Backscatter Communications With IRS

Abstract

In this letter, we investigate a novel ambient backscatter communication (AmBC) system in which an intelligent reflecting surface (IRS) acts as a passive transmitter to communicate with a reader. We are interested in the joint design of the IRS reflecting beamforming and symbol detection to minimize the bit error rate (BER). However, the problem is challenging to be solved optimally, since the BER is related to the clustering-based detector without a concrete close-form expression and the IRS reflecting unit modulus constraint is non-convex. To solve this issue, we propose a novel deep unfolding neural network (DUNN) combining data-driven and model-driven for passive reflecting beamforming design and symbol detection, which is learned to approximate the BER model from the training samples and the unit modulus constraint is satisfied by treating the optimization variables as network parameters. Numerical results demonstrate that the proposed scheme has superior performance of detection.