Team Deep Neural Networks for Interference Channels
Paul de Kerret, David Gesbert, Maurizio Filippone · 2018
In this paper, we propose to use Deep Neural Networks (DNNs) to solve so-called Team Decision (TD) problems, in which decentralized Decision Makers (DMs) aim at maximizing a common utility on the basis of locally available Channel State Information (CSI) without any additional communication or iteration. In the proposed configuration -coined Team DNNs (T-DNNs)-, the decision at each DM is approximated using a DNN and the weights of all DNNs are jointly trained, even though the implementation remains fundamentally decentralized. Turning to a practical application, the problem of decentralized link scheduling in Interference Channels (IC) is reformulated as a TD problem so that the T-DNNs approach can be applied. After adequate training, the scheduling obtained using the T-DNNs flexibly adapts to the decentralized CSI configuration to outperform other scheduling algorithms, thus proposing a novel efficient solution to a problem that has remained elusive for years.