RSMA-Assisted Integrated Sensing and Communications: A Survey and a Two-Stage Energy Efficiency Optimization Framework

This paper studies the energy efficiency (EE) optimization of rate-splitting multiple access (RSMA) enabled integrated sensing and communications (ISAC) systems under imperfect channel state information at the transmitter (CSIT). Existing approaches typically optimize precoders directly from uncertain CSIT, relying on worst-case or statistical treatments that do not improve the channel estimates. To this end, we develop a two-stage framework that combines data-driven channel refinement with model-based optimization for a monostatic multiple-input single-output (MISO) RSMA-ISAC platform. A convolutional block attention module (CBAM) based network is first employed to refine the channel estimates by capturing inter-user correlations and spatial characteristics, yielding representations whose residual error is lower than the available imperfect estimates. Using these refined channels as input, the non-convex EE maximization problem, subject to communication rate, transmit power limit and a sensing performance requirement expressed through the Cramér–Rao bound (CRB) constraint, is then solved through a successive convex approximation (SCA) procedure. Extensive numerical results across different antenna sizes, channel-error SNRs, and circuit power levels show that the proposed refinement stage significantly improves EE compared with raw imperfect-CSIT design, worst-case robust optimization, and a weighted-sum-rate WMMSE (WSR-WMMSE) baseline. The comparison with an independently implemented transformer-based refinement network further shows that the proposed CBAM-based refiner achieves higher EE, confirming its effectiveness. The framework maintains high sensing performance and exhibits stronger resilience to channel uncertainty than conventional space-division multiple access (SDMA)-ISAC schemes under the same refinement pipeline.