Federated Learning Composite Ad Hoc Interference Maps

Efficient resource allocation is essential for ad hoc wireless networks and is becoming increasingly crucial for massive Machine Type Communication (mMTC). Resource allocation typically requires interference maps maintained either centrally or by each device. Centralized interference maps, used for Access Point (AP)-assisted resource allocation, require the interference between each device to be measured to capture the global interference map, a mammoth combinatorial task considering massive deployments. Unfortunately, the channel estimation overhead is a significant obstacle due to the limited channel coherence duration. On the other hand, while distributed localized interference maps do not incur a high overhead to construct, they only consider local conditions. As such, localized interference maps tend to be suboptimal. This work introduces a novel Federated Learning (FL) approach to interference map generation. Our method enables AP-autonomous resource allocation and is less complex than traditional interference map generation techniques. As a result, it yields a higher per device sum rate than achievable with localized interference mans.