Enabling Programmable Inference and ISAC at the 6GR Edge with dApps

Abstract

The convergence of communication, sensing, and Artificial Intelligence (AI) in the Radio Access Network (RAN) offers compelling economic advantages through shared spectrum and infrastructure. How can inference and sensing be integrated in the RAN infrastructure at a system level? Current abstractions in O-RAN and 3GPP lack the interfaces and capabilities to support (i) a dynamic life cycle for inference and Integrated Sensing and Communication (ISAC) algorithms, whose requirements and sensing targets may change over time and across sites; (ii) pipelines for AI-driven ISAC, which need complex data flows, training, and testing; (iii) dynamic device and stack configuration to balance trade-offs between connectivity, sensing, and inference services. This paper analyzes the role of a programmable, software-driven, open RAN in enabling the intelligent edge for 5G and 6G systems. We identify real-time user-plane data exposure, open interfaces for plug-and-play inference and ISAC models, closed-loop control, and AI pipelines as elements that evolutions of the O-RAN architecture can uniquely provide. Specifically, we describe how dApps - a real-time, user-plane extension of O-RAN - and a hierarchy of controllers enable real-time AI inference and ISAC. Experimental results on an open-source RAN testbed demonstrate the value of exposing I/Q samples and real-time RAN telemetry to dApps for sensing applications.

Figures (5)

• Figure 1: System-level challenges toward enabling programmable and inference within the 6GR architecture.
• Figure 2: Hierarchical sensing architecture. dApps provide local, real-time sensing at the via the E3 interface for I/Q, , and data. Multiple dApps run in parallel for different sensing topologies (monostatic, bistatic, ranging, spectrum). The Near/Non-RT coordinates multi-node fusion and network optimization through sensing xApps and rApps. Processed results are exposed to core network sensing s and external APIs.
• Figure 3: life-cycle management framework. Centralized training produces validated models published to the Model Catalog. The Orchestrator in the performs intent-driven algorithm/site matching and automated deployment. Monitoring feeds enable continuous adaptation and retraining.
• Figure 4: -based sensing accuracy vs. data movement overhead for full-duplex monostatic sensing. A drone target at 500 m range, 25 m/s velocity, $-20$ dBsm . Bandwidth (10--100 MHz) and slot duration (0.25--1.0 ms) increase jointly and are represented by the sensing overhead rate in the x axis.
• Figure 5: CDF of range estimation error from an -based testbed. A subspace method running as a dApp on full data accessed via the E3 interface substantially outperforms the scalar peak-detection pipeline used in current positioning, especially with fewer observations ($M=20$) gangula2025rtt.