Listen-After-Collision Mechanism for Dynamic Spectrum Access Using Deep Q-Network With an Improved Thompson Sampling Algorithm
IEEE Internet of Things Journal
Dynamic spectrum access (DSA) is a key technology in cognitive radios, where secondary users (SUs) opportunistically access spectral holes of primary users (PUs) (i.e., channels unoccupied by PUs). The existing DSA schemes often use the listen-before-talk (LBT) mechanism to avoid transmission collisions with PUs. However, LBT-based schemes may not be able to achieve high utilization efficiency of spectral holes as SUs need to perform spectrum sensing over multiple spectrum holes heavily. To address this issue, in this work, we propose a new mechanism called listen-after-collision (LAC), where an SU accesses a channel of PUs without spectrum sensing, and it performs spectrum sensing only after a transmission collision occurs. Moreover, a deep Q-network with an improved Thompson sampling algorithm (DQN-iTSA) is proposed to predict both the availabilities and the time-lengths of spectral holes to avoid unacceptable transmission collisions, and also to determine the order of the channels for sensing by jointly considering the characteristics of spectral holes and the channel qualities of SU transmissions. Extensive simulation results are provided to demonstrate the superior performance of DQN-iTSA, which show that DQN-iTSA achieves the highest throughput among the compared methods.
Open Access Status
This publication is not available as open access