Over the Air Computation for Distributed Power Control Using SDR Testbed

Interference management is a critical challenge in future wireless networks, requiring scalable solutions beyond centralized approaches. In this work, we propose and experimentally validate an over-the-air computation framework for distributed power control on a Software-Defined Radio (SDR) testbed. Using the graph neural networks (GNNs) paradigm, our approach embeds message-passing operations within standard 3GPP Synchronization Signal Blocks (SSBs) to efficiently exchange Neural Network Information (NNI) along with known channel state information (CSI). By aggregating signals directly in the wireless medium, the communication overhead related to message passing is reduced, while the inherent synchronization capabilities of SSBs ensure precise timing and coordination. Experimental results show that the resulting power-control policy delivers throughput gains, underscoring its practical utility for 6 G networks and subnetworks.