RSMA-aided Multiuser Gaussian Splatting Transmission

3D Gaussian Splatting (3DGS) has rapidly emerged as a leading technique for photogrammetry and 3D scene representation. However, its extension to streaming and multiuser applications introduces a significant set of challenges, in the form of meeting Quality of Service (QoS) or Quality of Experience (QoE) requirements for users while simultaneously expanding 3DGS to 4D. To combat some of these challenges, we propose a Rate-Splitting Multiple Access (RSMA) aided approach to multi-user streaming of Gaussian Splatting (GS) represented scenes. While traditional 3D and VR scene optimization utilizes tile-based models, we utilize GS’s inherent use of Gaussian primitives as a basis in our formulation. We then develop a Deep Reinforcement Learning (DRL) solution procedure to optimize RS variables, bit rate selection, and beamforming vectors for an optimal user experience. Through simulations, we show that the proposed RSMA-aided multi-user GS transmission scheme outperforms two baselines with considerable gains.