Toward Energy-Aware Multimodal Beam Prediction in 6G V2I Networks: A Survey and Conceptual Framework

The evolution toward Sixth Generation (6G) wireless networks introduces a paradigm shift in Vehicle-to-Infrastructure (V2I) communications, wherein ultra-reliable millimeter-wave (mmWave) connectivity must be sustained despite severe blockage dynamics and stringent energy constraints. While recent advances in multimodal beam prediction leverage auxiliary sensors such as LiDAR, Radar, cameras, and GPS to anticipate channel conditions and reduce beam training overhead, the energy cost of high-fidelity sensing modalities remains largely unquantified in the literature. This survey presents a detailed analysis of the accuracy-energy trade-off in multimodal beam prediction systems, synthesizing hardware power profiles from empirical datasheets and establishing a unified optimization framework that treats sensor activation as a constrained resource allocation problem. Through comparative evaluation of sensing modalities, we demonstrate that LiDAR-based prediction incurs a 7.5× power penalty relative to Radar while yielding marginal accuracy improvements in Line-of-Sight scenarios. We propose a hierarchical adaptive fusion architecture employing confidence-based early exit mechanisms to dynamically gate sensor activation, thereby minimizing average power consumption without compromising link reliability. The analysis contextualizes energy-aware beam management within emerging Green AI principles and outlines a research agenda for sustainable 6G networks, including Integrated Sensing and Communication (ISAC) and neuromorphic sensing paradigms.