Learning from Peers at the Wireless Edge

摘要: The last mile connection is dominated by wireless links where heterogeneous nodes share the limited and already crowded electromagnetic spectrum. Current contention based decentralized access system reactive in nature to mitigate interference. In this paper, we propose use neural networks learn predict spectrum availability a collaborative manner such that its can be predicted with high accuracy maximize minimize interference between simultaneous links. Edge have wide range of sensing computation capabilities, while often using different operator networks, who might reluctant their models. Hence, introduce peer Federated Learning model, local model trained on results each node shared among peers create global model. need for base station or point act as centralized parameter server replaced empowering edge aggregators models minimizing communication overhead transmission. We generate channel data, which used train Simulation both show over 95% predicting opportunities various network topology.