Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Programmable and GPU-Accelerated Edge Inference for Real-Time ISAC on NVIDIA ARC-OTA

Davide Villa, Mauro Belgiovine, Nicholas Hedberg, Michele Polese, Chris Dick, Tommaso Melodia

TL;DR

Problem: Enabling real-time ISAC in software-defined, edge RAN under bandwidth constraints. Approach: A GPU-accelerated, plug-and-play dApp framework on NVIDIA ARC-OTA with a concrete cuSense indoor localization dApp, leveraging E3 interfaces and shared-memory telemetry. Findings: Demonstrates sub-millisecond CSI extraction, end-to-end dApp latencies in the low milliseconds, and sub-meter localization accuracy across unseen runs, all within a production-grade stack. Significance: Provides an open-source reference design for AI-native RANs and ISAC applications, accelerating development of real-time, GPU-driven sensing on edge networks.

Abstract

The transition of cellular networks to (i) software-based systems on commodity hardware and (ii) platforms for services beyond connectivity introduces critical system-level challenges. As sensing emerges as a key feature toward 6G standardization, supporting Integrated Sensing and Communication (ISAC) with limited bandwidth and piggybacking on communication signals, while maintaining high reliability and performance, remains a fundamental challenge. In this paper, we provide two key contributions. First, we present a programmable, plug-and-play framework for processing PHY/MAC signals through real-time, GPU-accelerated Artificial Intelligence (AI) applications on the edge Radio Access Network (RAN) infrastructure. Building on the Open RAN dApp architecture, the framework interfaces with a GPU-accelerated gNB based on NVIDIA ARC-OTA, feeding PHY/MAC data to custom AI logic with latency under 0.5 ms for complex channel state information extraction. Second, we demonstrate the framework's capabilities through cuSense, an indoor localization dApp that consumes uplink DMRS channel estimates, removes static multipath components, and runs a neural network to infer the position of a moving person. Evaluated on a 3GPP-compliant 5G NR deployment, cuSense achieves a mean localization error of 77 cm, with 75% of predictions falling within 1 meter. This is without dedicated sensing hardware or modifications to the RAN stack or signals. We plan to release both the framework and cuSense pipelines as open source, providing a reference design for future AI-native RANs and ISAC applications.

Programmable and GPU-Accelerated Edge Inference for Real-Time ISAC on NVIDIA ARC-OTA

TL;DR

Problem: Enabling real-time ISAC in software-defined, edge RAN under bandwidth constraints. Approach: A GPU-accelerated, plug-and-play dApp framework on NVIDIA ARC-OTA with a concrete cuSense indoor localization dApp, leveraging E3 interfaces and shared-memory telemetry. Findings: Demonstrates sub-millisecond CSI extraction, end-to-end dApp latencies in the low milliseconds, and sub-meter localization accuracy across unseen runs, all within a production-grade stack. Significance: Provides an open-source reference design for AI-native RANs and ISAC applications, accelerating development of real-time, GPU-driven sensing on edge networks.

Abstract

The transition of cellular networks to (i) software-based systems on commodity hardware and (ii) platforms for services beyond connectivity introduces critical system-level challenges. As sensing emerges as a key feature toward 6G standardization, supporting Integrated Sensing and Communication (ISAC) with limited bandwidth and piggybacking on communication signals, while maintaining high reliability and performance, remains a fundamental challenge. In this paper, we provide two key contributions. First, we present a programmable, plug-and-play framework for processing PHY/MAC signals through real-time, GPU-accelerated Artificial Intelligence (AI) applications on the edge Radio Access Network (RAN) infrastructure. Building on the Open RAN dApp architecture, the framework interfaces with a GPU-accelerated gNB based on NVIDIA ARC-OTA, feeding PHY/MAC data to custom AI logic with latency under 0.5 ms for complex channel state information extraction. Second, we demonstrate the framework's capabilities through cuSense, an indoor localization dApp that consumes uplink DMRS channel estimates, removes static multipath components, and runs a neural network to infer the position of a moving person. Evaluated on a 3GPP-compliant 5G NR deployment, cuSense achieves a mean localization error of 77 cm, with 75% of predictions falling within 1 meter. This is without dedicated sensing hardware or modifications to the RAN stack or signals. We plan to release both the framework and cuSense pipelines as open source, providing a reference design for future AI-native RANs and ISAC applications.

Paper Structure

This paper contains 19 sections, 10 equations, 14 figures, 3 tables.

Table of Contents

  1. Introduction
  2. dApp Background and Motivation
  3. Key Challenges and Requirements
  4. GPU-accelerated dApp Framework Design
  5. High-Level Architecture
  6. Real-Time ADL and Shared Memory
  7. E3 Agent and E3 Manager
  8. dApp Reference Architecture
  9. Profiling dApp Framework Performance
  10. ARC-OTA Experimental Setup
  11. Framework Benchmarks
  12. cuSense: Integrating ISAC in a dApp
  13. Overview and Goals
  14. Uplink CSI Processing Pipeline
  15. cuSense Experimental Evaluation
  16. ...and 4 more sections

Figures (14)

  • Figure 1: Overview of our work at the intersection of GPU-acceleration, O-RAN dApps, and ISAC use cases.
  • Figure 2: NVIDIA ARC-OTA dApp Integration Architecture.
  • Figure 3: Aerial Data Lake ping-pong mechanism and shared memory structure.
  • Figure 4: Data path integration between Aerial L1, Real-time ADL, shared-memory, and the E3 Agent. The steps (Op. 1–4) match the operations in Table \ref{['tab:datapath']}.
  • Figure 5: Support for multiple nodes, E3 Agents, and dApps in the same gNB.
  • ...and 9 more figures