Deep-learning based Cooperative Spectrum Prediction for Cognitive Networks

摘要

The concept of dynamic spectrum access under the basic principles of Cognitive Radio (CR) networks is considered to alleviate the current inefficient use of radio spectrum. Relying on the presumed knowledge of the radio environment, CR opportunistically allows secondary users to access licensed spectrum bands when not in use by their respective owners (primary users), in a non-interference manner. As identifying and exploiting the available spectrums is time and energy consuming, different spectrum prediction methodologies are used in order to provide meaningful data about current and future spectrum usage in specific time and location. However, local spectrum prediction done from a single SU`s perspective can be unreliable and prone to error under harsh channel conditions. Where applicable, cooperative spectrum prediction in multi-user environment has the potential to overcome the limitations of local spectrum prediction accuracy. In this paper, cooperative spectrum prediction with neural network predictors is investigated. Considering heterogeneous CR network, soft cooperative fusion is implemented to exploit the spatial dependency of spectrum measurement data and provide occupancy predictions for an SU in near proximity range without it conducting spectrum sensing.