Abstract
This thesis investigates the use of zero-shot artificial intelligence (AI) agents for autonomous cross-protocol waveform selection in dynamic radio frequency (RF) environments. Building on the principles of cognitive radio, this work explores how an intelligent radio system can observe channel conditions, reason through RF engineering trade-offs, and adapt its waveform configuration to satisfy user-defined communication objectives. The proposed framework uses agentic AI capabilities, including tool use, memory, and complex reasoning, to enable an autonomous radio agent with waveform generation and selection capabilities. The agent interprets user intent, evaluates measured RF channel conditions, and selects appropriate waveform configurations across heterogeneous wireless protocols without requiring domain-specific training or user-provided RF expertise. The results show that the agent can reason through typical RF engineering decisions based on channel conditions and user objectives. It can evaluate trade-offs involving robustness, throughput, bandwidth availability, interference, and link reliability to support effective waveform selection in dynamic RF environments. Overall, this research demonstrates the potential of zero-shot AI agents to serve as the reasoning engine or “brain” of future intelligent radio systems. By moving beyond rigid rule-based adaptation, the proposed framework advances cognitive radio toward more autonomous, flexible, and goal-directed operation across heterogeneous wireless protocols.