Opportunistic channel access in cognitive radio networks (CRNs) under an unknown environment is gradually receiving a great deal of attention. This article studies the basic problem of decentralized secondary users (SUs) performing multiple channel sensing and access in CRN, when sensing is not perfect. The channel availability information is unknown and must be estimated and learned through big-data samples from the wireless channels by SUs. Both the independent identical distribution (i.i.d.) channel model and the Markov channel model are considered. In the i.i.d. model, the availability of each channel is modeled as an i.i.d. process, while in the Markov model, the availability of each channel is set as a Markov chain with an unknown probability transition matrix. If multiple SUs access to the identical channel, collision will occur and none of SUs gets a reward. Learning loss, which is also referred to as regret, is thus inevitable. To handle with the sampling data on large scale, we formulate the channel sensing and access process as a multi-armed bandit problem (MABP), based on which big-data driven online algorithms are proposed. The theoretical analysis and simulations prove that the regret of our algorithms is both logarithmic in finite time and asymptotically.
|Number of pages||13|
|Journal||IEEE Transactions on Emerging Topics in Computational Intelligence|
|Publication status||Published - Feb 2021|
- logarithmic regret
- multi-armed bandit problem
- Multi-user channel sensing and access