H. Ahmadi, Y.H. Chew, N. Reyhani, C.C. Chai, L.A. DaSilva, Learning solutions for auction-based dynamic spectrum access in multicarrier systems, Computer Networks, Volume 67, 4 July 2014, Pages 60-73, ISSN 1389-1286, http://dx.doi.org/10.1016/j.comnet.2014.03.026.
Abstract:
In this work, we address competition among autonomous cognitive radios (CRs) which are competing for frequency bands, and model their interactions. We design auction mechanisms with and without an entry fee and define utility functions based on the total achieved capacity per unit price. We equip the CRs with a learning algorithm in order to bid more efficiently. Our considered network consists of a central spectrum moderator (CSM) and a number of competing CR pairs. The CSM auctions the available spectrum bands, and each CR bids for them in order to transmit its data. A CR bids for a subcarrier at a value proportional to the achievable capacity for it to transmit on that subcarrier, and the CR is free to dynamically assign portions of its transmit power to available subcarriers in order to maximize its achievable capacity per unit of price. We propose a Dirichlet process-based and a Gaussian process regression-based learning algorithms which make use of the outcomes of the past auctions to learn the bidding behavior of the competing CRs. By learning the bidding behavior of competing CRs, a CR can improve its bidding efficiency by concentrating its transmit power on the subcarriers that it has a higher chance of getting access to. Simulation results show that the proposed nonparametric learning algorithms achieve significantly higher utilities for the CRs than the myopic approach, and both the Dirichlet and the Gaussian processes provide roughly the same level of improvement.