Utility-based power control for cooperative dynamic spectrum access networks.

Abstract

Dynamic spectrum access (DSA) is a paradigm proposed by the Federal Communications Commission aiming at efficient management of the available wireless network resources. The present thesis deals with resource allocation in cooperative DSA networks. Two models are considered: (a) open sharing networks (spectrum commons), possibly deployed over unlicensed bands; and (b) networks with primary users, and secondary users who access the spectrum upon paying a fee to the primary. In both types of DSA networks, collaborating terminals adhere to diverse (maximum and minimum) quality-of-service (QoS) constraints in order to effect hierarchies between primary and secondary users or to prevent abusive utilization of the available spectrum in an open sharing model. The focus is on peer-to-peer networks with co-channel interference in both single- and multi-channel settings. Utilities that are functions of the signal-tointerference- plus-noise-ratio (SINR) are employed as QoS metrics. By adjusting their transmit power, users can mitigate the generated interference and also meet the QoS requirements. A novel formulation accounting for heterogeneous QoS requirements as well as maximum interference constraints is obtained after introducing a suitable relaxation and recasting a constrained sum-utility maximization as a convex optimization problem. The optimality of the relaxation is established under general conditions. Based on this relaxation, an algorithm for optimal power control that is amenable to distributed implementation is developed, and its convergence is established. The algorithm relies on gradient-based iterations to find saddle points of the Lagrangian function associated with the constrained convex optimization problem. In the context of power control, errors may be introduced in the gradient vectors as a result of the distributed implementation of the algorithm. The effects of these errors are studied in a general setting. To this end, two running averages (ergodic sequences) of the iterates generated by the algorithm are formed, each with complementary strengths. Under the main assumptions of problem convexity and error boundedness, bounds on the constraint violation and the suboptimality per iteration index are derived. Numerical tests verify the analytical claims and demonstrate performance gains relative to existing schemes.