Robust MIMO Over-the-Air Computation for Distributed Estimation

As industrial networks transition toward the Industry 5.0 paradigm, the integration of sensing, communication, and computation into a unified, task-oriented framework becomes essential for supporting resilient digital twins and human-robot collaboration. Over-the-Air Computation (AirComp) has emerged as a vital technology for low-latency data aggregation in industrial IoT. However, the inherent reliance of AirComp on the arithmetic mean makes it critically vulnerable to outliers generated by sensor failures or Byzantine behavior. This paper proposes a robust MIMO-AirComp framework developed around the Geometric Median. By leveraging the spatial dimensions provided by MIMO architectures, we adapt the iterative Weiszfeld algorithm for simultaneous analog transmission over the wireless multiple-access channel. We demonstrate that by transmitting weighted measurement vectors, the fusion center obtains a resilient global estimate even when nearly half the network is compromised. Our methodology includes a detailed analysis of the pre-processing requirements at the sensors and the beamforming strategies at the receiver. Simulation results confirm that our robust estimator provides a significant increase in resilience against data-poisoning phenomena compared to state-of-the-art AirComp techniques, with negligible overhead in benign conditions.