Deep Agentic AI Framework for Adaptive Wireless Resource Optimization in Semiconductor-Assisted 6G Communication Networks

Abstract

  Next-generation telecommunications are increasingly overlapping with various application domains, most notably vehicular systems, the Internet of Things, augmented and virtual reality, and robotic swarms. These overlaps underpin the need for different levels of autonomous operation—from the user perspective—while posing unique challenges for the underlying communications systems, which must still meet the operational objectives of each individual domain. Suggested solutions exploit advancements spanning mobile networks, cloud-edge computing infrastructures, algorithms for wireless resource allocation, and the full automation of system operations. Within this context, agentic AI alters the traditional separation of aircraft control planes—mainly dedicated to perception, planning, and acting—by integrating the development of sufficient high-fidelity capability. High levels of task automation supersede human engagement and enable system stewardship, provided the AI are developed, deployed, and managed under appropriate principles.

Even so, support for most domain requirements remains oversimplified, with operator assistance and resource provisioning and allocation invariably under human skill. Consequently, a new layer of cognitive automation spanning perception, decision making, and operation is being gradually integrated into systems via AI dedicated to such tasks. The framework governs a self-aware general-purpose cognitive loop that officiates the request and supply of contextualized information and control actions within the context of data acquisition and potentially also for planning and learning adopted by lower-level AI. Because its candidates reside within the operational ecosystem and their autonomy level can also vary from system supervision to collaborative support, the cognitive loop naturally lends itself to agentic AI. Beyond theoretical considerations, the approach has been applied to the dynamic scheduling of integrated sensing and communications resources