As more devices, sensors, and services become interconnected, the amount of data that is generated and must be consumed continues to surge. To truly bridge our physical and digital worlds, all this information needs to be first transmitted and processed in real time. This data ecosystem powers everything, ranging from smart city infrastructures relying on continuous sensing and environmental monitoring, to distributed learning systems that train Artificial Intelligence (AI) models across mobile devices, and real-time control loops for robotics and industrial automation. As these capabilities mature, they pave the way toward more demanding and immersive 6G-era applications, such as autonomous vehicle platooning, drone-swarm control for logistics and inspection, and interactive eXtended Reality (XR) experiences, to name just a few. However, these emerging services not only require massive data ingestion and processing but also impose strict demands on latency, reliability, and computational efficiency, highlighting the need for new architectures that seamlessly integrate communication and computation across the network.
Typically, communication and computation are treated as separate operations, each utilizing the radio and distributed computing resources across the network. Multiple mobile devices and sensors share the wireless medium using multiple access schemes, which assign distinct time or frequency resources to each device to avoid interference. More advanced schemes allow multiple devices to transmit over the same time-frequency resources by distinguishing their signals based on the received power or other characteristics, leveraging sophisticated signal processing at the receiver to separate and decode overlapping transmissions. Once each signal is successfully decoded, the data is processed according to the application. This may include computing statistics, as in environmental monitoring, training AI models for tasks such as pattern recognition, or other decisions to implement control strategies.
What is over-the-air computation?
Over-the-air computation (or AirComp in short) has emerged as a key advancement in wireless data aggregation by unifying communication and computation into a single integrated operation. Signals transmitted simultaneously by multiple devices over the same time-frequency resources naturally superimpose in the air, following the waveform superposition property of wireless channels. A typical superposition of wireless signals corresponds to the sum of individual transmitted waveforms. However, by applying appropriate preprocessing during message encoding, it is possible to realize other types of aggregation functions at the receiver. These functions correspond to the class of nomographic functions, including but not limited to averages, weighted sums, geometric means, and max/min operations. In this way, the signal observed at the receiver directly represents the desired computed outcome performed “over the air”, bypassing complex and time-consuming signal decoding and separation, while enabling efficient and scalable aggregation from massive numbers of devices over limited spectrum resources.
Why is over-the-air computation relevant to 6G?
Over-the-air computation is aligned with the vision of 6G networks, where communication is no longer viewed as merely a data service, but as a native enabler of distributed intelligence. Many of the emerging 6G applications introduced earlier, such as sensing in smart cities and federated learning across massive number of devices, do not require the recovery of individual data streams. Instead, they rely on aggregate information, for example global model updates or collective measurements. By computing these functions directly in the wireless medium, over-the-air computation drastically reduces latency and signaling overhead, while enabling larger-scale deployments of interconnected devices. Therefore, over-the-air computation constitutes an advancement in multiple access schemes, designed to inherently support the requirements and characteristics of 6G applications.
What are the challenges of over-the-air computation?
Since over-the-air computation relies on superposing signals from multiple devices, any distortions introduced by the wireless channel, such as fading, noise, or signal misalignment, can impact the accuracy of the computed result. This discrepancy between the desired and the received computation can be effectively quantified using the Mean Squared Error (MSE) metric or by evaluating application-level performance indicators, such as the accuracy of a distributed or federated learning algorithm. To minimize these errors and improve the application metrics, careful control of the wireless propagation environment and intelligent allocation of radio resources must take place.
How do we approach this in the 6G-LEADER project?
In the 6G-LEADER project, we target to harness the full potential of over-the-air computation scheme to specifically support demanding AI-based applications. To achieve this, we develop intelligent resource allocation strategies to control the transmission timing of devices, their transmission power, and the receive beamforming at the receiver’s end. The ultimate goal is to minimize the MSE of the over-the-air aggregated results while jointly optimizing network-wide objectives such as energy and spectral efficiency. By integrating highly reconfigurable RadioFrequency (RF) structures, such as Reconfigurable Intelligent Surfaces (RIS), we further manage to manipulate the wireless propagation environment, improving signal alignment and the reliability of over-the-air aggregation. Our resource allocation solutions are developed in the form of xApps, designed for realistic integration into O-RAN 6G network implementations and evaluation over distributed learning applications. By leveraging the near-real-time control loop of O-RAN deployments, the developed xApps are poised to dynamically provide radio configurations that enhance over-the-air computation performance. This marks a first step toward efficient, scalable, and fast AI-native 6G applications.
