Overview
This manuscript presents a unified system architecture and an evaluation methodology specifically designed for AI-native Integrated Sensing and Communication (ISAC). AI-native ISAC is defined as ISAC where learning-based agents dynamically adapt sensing, communication, and actuation policies under conditions of uncertainty. The proposed methodology aims to provide a structured approach for evaluating systems in which radio sensing, data transmission, and control actions are coupled within a singular closed-loop framework.
Research Context
Integrated Sensing and Communication (ISAC) involves the integration of radio sensing, data transmission, and control actions into a single closed-loop system. When artificial intelligence (AI)-driven policies are introduced, these systems adapt sensing and communication online across varied sensing tasks and objectives. This adaptation means that end-to-end performance is influenced by multiple factors, including waveform characteristics, channel conditions, inference latency, environmental uncertainty, dynamic changes, and hardware non-idealities. These factors collectively lead to trade-offs among sensing accuracy, communication reliability, and resource overhead within such systems.
Approach
The research formalizes the design space pertinent to closed-loop ISAC. It proposes a three-stage validation pipeline for the evaluation of AI-native ISAC systems. The initial stage involves bounds and feasibility analysis. The subsequent stage utilizes high-fidelity digital-twin simulation. The final stage incorporates preliminary over-the-air validation.
To further standardize evaluation, the methodology includes a minimal reporting checklist. This checklist is designed to establish a link between technical Key Performance Indicators (KPIs) and application-level Key Value Indicators (KVIs). Examples of technical KPIs mentioned include data rate, SINR (Signal-to-Interference-plus-Noise Ratio), target detection, parameter estimation, track quality, localization error, outage, latency, overhead, and energy per decision. These technical KPIs are connected to application-level KVIs such as availability and mission effectiveness.
Two specific instantiations are used to illustrate the application of this methodology. These include Unmanned Aerial Vehicle (UAV)-based outdoor coverage extensions and Reconfigurable Intelligent Surface (RIS)-enabled indoor coverage extensions. These instantiations serve as examples for structuring reproducible baselines and generating comparable evidence across diverse deployments, with the objective of bridging the gap between theoretical ISAC gains and claims regarding deployment-ready performance.
Findings
- The research defines AI-native ISAC as ISAC where learning-based agents adapt sensing, communication, and actuation policies online under uncertainty.
- It formalizes the design space of closed-loop ISAC.
- A three-stage validation pipeline is proposed: bounds and feasibility analysis, high-fidelity digital-twin simulation, and preliminary over-the-air validation.
- A minimal reporting checklist links technical KPIs (e.g., data rate, SINR, target detection, localization error, energy per decision) to application-level KVIs (e.g., availability, mission effectiveness).
- Representative instantiations, such as UAV-based outdoor and RIS-enabled indoor coverage extensions, are used to demonstrate structuring reproducible baselines and comparable evidence for heterogeneous deployments.