Time-Aware Graph Neural Network for Asynchronous Multi-Station Integrated Sensing and Communications Fusion in Open...

Area of Science:

• Wireless communication networks
• Graph neural networks
• Signal processing

Background:

• Open RAN (O-RAN) Near-RT RIC faces challenges with out-of-order telemetry due to network jitter.
• Traditional spatial fusion methods are invalidated by this temporal scrambling.
• Asynchronous sensing reports require new reliability metrics.

Purpose of the Study:

• To introduce Age-of-Sensing (AoS) as a dynamic reliability metric for asynchronous sensing data.
• To develop an AoS-aware graph neural network (GNN) paradigm for robust sensing fusion.
• To present Time-Aware Fusion (TA-Fusion) for prioritizing fresh telemetry and suppressing stale data.

Main Methods:

• Developed an AoS-aware GNN paradigm incorporating sensing freshness into graph-based fusion.
• Introduced a TA-Gate mechanism within TA-Fusion to recalibrate node trust before graph aggregation.
• Utilized a standardized O-RAN benchmark for performance evaluation.

Main Results:

• TA-Fusion achieved a root mean square error (RMSE) of 12.22 m.
• Demonstrated a 21.7% reduction in Mean Absolute Error (MAE) compared to the AoS-aware GNN baseline.
• Maintained robustness in extreme jitter scenarios, outperforming traditional linear methods.

Conclusions:

• The proposed AoS-aware GNN paradigm offers a resilient spatial foundation for delay-critical Integrated Sensing and Communication (ISAC) services.
• TA-Fusion provides real-time feasibility for 6G orchestration under substantial network jitter.
• The framework ensures consistent error bounds across diverse base station geometries without recalibration.