Spectrum Pricing for Cognitive Radio

This thesis examines how the price paid by the end users via an auction model can be used in regulating and controlling the admission process given a dynamic spectrum access and a heterogeneous small cell network. The performance of the system is judged by the energy consumed, the system throughput and the delay.
A first price auction model with a reserve price is designed to take into consideration the signal to noise ratio of the users by introducing a novel tax and subsidy scheme called the green payments. Furthermore, the use of multiple bidding process and an admittance threshold, known as the probability of being among the highest bidders, helps in further reducing the energy consumed and improves the system throughput. A utility function is also found useful in determining the satisfaction of the users and in formulating a theoretical model for the admission process.
Bid learning performance using Linear Reinforcement learning, Q learning, and Bayesian learning is compared and the results show that Bayesian learning converges faster because it incorporates prior information. It is shown that incorporating a price based utility function into the punishment or the reward weighting factor can help the learning process to converge at the optimal bidding price. A game model is formulated to allow all users in the system to learn depending on their priority. This enables users to learn different parameters such as the best offered bid price and the appropriate time to participate in the auction process. Results show that provided all the users take part in the learning process, a Nash Equilibrium can be established. The energy and the delay associated with the auction process are also further reduced when all the users are learning the different parameters.