Deep Reinforcement Learning for Energy Efficiency Maximization in RSMA-IRS-Assisted ISAC System

The combination of rate splitting multiple access (RSMA) and integrated sensing and communication (ISAC) has recently played a constructive role in various emerging applications associated with sixth-generation networks. However, in complex urban environments, the network performance may be severely restricted by transmission blockages. With the help of intelligent reflecting surfaces (IRS), this paper leverages a virtual line-of-sight link to guarantee the qualityof-service (QoS). First, a three-dimensional (3D) geometry-based stochastic channel model (GBSM) is developed to characterize the IRS-empowered ISAC networks with RSMA. Based on the proposed channel model, we formulate an energy efficiency (EE) maximization problem, subject to transceiver beamforming constraints, IRS phase shift constraints, and QoS constraints. The impact of some important system parameters on the EE is numerically investigated using the proximal policy optimization (PPO) method. One key observation of our results is that the system EE degrades significantly at higher frequencies even for double-Rayleigh fading channels, which may make it difficult for the subsequent sensing link establishment.