Joint beamforming and movement design for agentic AI assisted ISAC system

Integrated Sensing and Communication (ISAC) designs often overlook power amplifier (PA) nonlinearity and user mobility. To address these gaps, we study an ISAC system model incorporating memoryless PA nonlinearity and agent mobility, where agents proactively move toward the base station to improve channels at a locomotion energy cost. This model opens a new direction for self-managing CR-IoT infrastructure in space in hardware adaptation, topology adaptation, and autonomous management. We formulate a max-min energy efficiency (EE) optimization problem subject to transmit power, movement bounds, and hardware distortion constraints. To solve this non-convex fractional program, we propose a two-layer dinkelbach-block coordinate descent (BCD)-successive convex approximation (SCA) algorithm framework to jointly optimize beamforming, sensing waveforms, and movement distances. Additionally, a large language model (LLM) agent dynamically adjusts operational parameters based on network context for cognitive adaptability. Simulations demonstrate that our framework substantially outperforms conventional static and heuristic baselines. Specifically, it yields up to 40% sum-rate gains in dense user scenarios and 30% EE improvements under moderate power budgets. These advantages are particularly pronounced under strong PA nonlinearity, confirming that jointly optimizing mobility and hardware awareness yields robust performance against severe signal distortion. Finally, beam pattern analysis validates the flexible reconfigurability of transmit waveforms.