Online deep learning based energy efficient optimization for IRS-assisted eMBB and URLLC services

Abstract

In this paper, we consider a downlink multiple-input single-output (MISO) system for enhanced mobile broadband (eMBB) and ultra-reliable and low-latency communication (URLLC) services assisted by intelligent reflecting surface (IRS). We formulate an optimization problem which aims at maximizing energy efficiency (EE) by jointly optimizing the beamforming vectors at the BS, IRS reflection coefficients matrix and resource allocation while satisfying the requirement of quality of service (QoS). The EE maximization problem is non-convex and contains various constraints. In this paper, a primal–dual online deep learning (PDODL) algorithm is proposed to tackle this issue. Specifically, we transform the original problem into the primal–dual problem by integrating constraints into the objective function. Then, online deep learning (DL) algorithm is adopted where the optimization variables are viewed as the network parameters which are further expressed in the form of unconstrained counterparts. The PDODL algorithm takes the objective function of the primal–dual problem as loss function because its decrease through network training is equal to the solving process of the optimization problem. Simulation results verify the effectiveness of the proposed algorithm and IRS can significantly improve the EE performance of the system.