AI & TELECOMMUNICATIONS

Agentic AI: The Autonomous Paradigm Shift

Agentic AI represents a paradigm shift, moving beyond conventional AI's predefined tasks to enable continuous perception-reasoning-action loops. It achieves autonomous context interpretation, goal-driven decision-making, and feedback-based policy improvement with minimal human intervention.

Key characteristics include: Autonomy (independent reasoning and action), Memory and Adaptability (retains critical information, learns from historical knowledge), and Explicit Reasoning & Agent Coordination (transparent decision processes, integration of specialized agents and external tools).

The evolution of AI agents has progressed from symbolic/rule-based, through ML-based and LLM-based, culminating in Agentic AI's ability to perform complex, deliberative planning and execution by integrating diverse frameworks like LLMs and DRL.

Integrated Sensing & Communication Systems

Integrated Sensing and Communication (ISAC) is a cornerstone for 6G networks, enhancing spectrum efficiency by unifying sensing and communication functions. The paper identifies two primary architectures:

• Radar-Communication Coexistence (RCC): Features physically separate sensing and communication systems, offering high flexibility but lower resource utilization and complex interference management.
• Dual-Functional Radar and Communication (DFRC): Integrates both functionalities into a single hardware platform, boosting resource utilization but posing challenges in optimizing distinct signal characteristics simultaneously.

Agentic AI plays a critical role in these systems by providing intelligent processing and autonomous operation, ensuring adaptability in dynamic and complex wireless environments, as illustrated by its core loop:

Optimizing ISAC: Traditional vs. Agentic AI

Traditional ISAC optimization methods face limitations in dynamic and complex environments. Agentic AI and Generative AI offer advanced solutions:

Agentic ISAC Framework: Design & Validation

The paper proposes a novel Agentic ISAC framework that integrates DRL, Generative AI (GenAI), and Large Language Models (LLMs). This framework operates through a continuous perception-reasoning-action loop, enhanced by:

• LLM-based Reward Function Design: Automates reward function creation for DRL based on system context, mitigating manual design challenges and learning biases.
• Transformer-based MoE: Provides robust reasoning and planning by capturing temporal dependencies and aggregating insights from multiple specialized experts.
• GenAI Integration: Improves environment state analysis and signal enhancement for DRL.

A case study involving a dual-functional BS-enabled sensing and communication system demonstrates the framework's effectiveness in maximizing communication rate and minimizing the Cramér-Rao bound (CRB) for position estimation.

Future Frontiers of Agentic ISAC

To further advance Agentic AI in ISAC systems, the research outlines several promising directions:

• Secure Agentic AI Frameworks: Essential for ensuring data integrity, confidentiality, and tamper resistance of knowledge bases, given that erroneous data can lead to system failures. Incorporating blockchain and differential privacy methods will be key.
• Lightweight Agentic AI Frameworks: Addressing the resource-intensive nature of integrating DRL, GenAI, and LLMs. Developing efficient frameworks for resource-constrained ISAC applications is crucial for usability and deployment.
• Cross-Domain Agentic AI Frameworks: Integrating knowledge and reasoning mechanisms across different domains to improve decision-making. This involves cross-domain information fusion and transfer mechanisms for unified planning in complex environments.

These directions aim to enhance the robustness, efficiency, and generalization capabilities of Agentic AI-based ISAC systems, paving the way for more intelligent and autonomous future networks.