Reconfigurable Intelligent Surface for Green Edge Inference

Reconfigurable intelligent surface (RIS) as an emerging cost-effective technology can enhance the spectral- and energy-efficiency of wireless networks. In this article, we consider an RIS-aided green edge inference system, where the inference tasks generated from resource-constrained mobile devices (MDs) are uploaded to and cooperatively performed at multiple resource-enhanced base stations (BSs). Taking into account both the computation and uplink/downlink transmit power consumption, we formulate an overall network power consumption minimization problem, which calls for the joint design of the set of tasks performed by each BS, uplink/downlink beamforming vectors of BSs, transmit power of MDs, and uplink/downlink phase-shift matrices at the RIS. However, the resulting combinatorial optimization problem is nonconvex and highly intractable. We tackle the challenge of combinatorial variables by exploiting the group sparsity structure of the beamforming vectors. Moreover, a block-structured optimization with mixed
$\ell _{1,2}$
-norm and difference-of-convex-functions (DC) based three-stage framework is proposed to solve the problem, where the mixed
$\ell _{1,2}$
-norm and DC techniques are adopted to induce the group sparsity structure and handle the nonconvex rank-one constraint, respectively. Simulations demonstrate the supreme performance gain of deploying an RIS and confirm the effectiveness of the proposed algorithm over the baseline algorithms in reducing the overall network power consumption.