A DePIN Architecture for Large Language Model-driven Integrated Sensing and Communications

Integrated Sensing and Communications (ISAC) enables real-time perception and coor- dination in the Internet of Things. However, current ISAC architectures face critical challenges in decentralization, trust, and incentivization. This paper introduces a novel Decentralized Physical Infrastructure Network (DePIN) approach to ISAC, where distributed nodes utilize Large Language Models (LLMs) to analyze multimodal sensor data while earning crypto-token rewards. At the core of this framework, we propose a permissioned blockchain that semantically validates LLM- generated content through oracle contracts and a Proof-of-Code (PoC) consensus mechanism. Each DePIN node leverages an onboard LLM to process sensor data and generate context-aware decisions, evaluated based on semantic quality metrics including coherence, novelty, and factual alignment. We formulate a constrained incentive maximization problem that jointly considers sensing qual ity, LLM inference accuracy, and system cost. We propose a Deep Reinforcement Learning (DRL) algorithm that adaptively optimizes the ISAC and DePIN token incentives across nodes to solve this. Extensive simulations demonstrate that our DePIN-based approach significantly outperforms conventional strategies in maximizing incentive accumulation while minimizing operation costs in dynamic LLM-driven ISAC systems.