Upper Confidence Bound learning approach for real HF measurements

New strategies based on cognitive radio are being discussed to make a more efficient use of the HF band. Multiple users transmit in this band with a worldwide coverage but having multiple collisions with other HF stations. The use of the Upper Confidence Bound (UCB) algorithm is proposed in this work to provide them with a dynamic spectrum access mitigating mutual interference. Based on reinforcement learning, it is used to select the best channel of a wideband HF transceiver in terms of availability. The feasibility of this proposal is demonstrated with real measurements of amateur contests in the HF band. To the best of the authors' knowledge, this is one of the few works on learning with real HF measurements.