Decentralized spectrum management through user coordination

Abstract

Wireless devices are becoming ubiquitous - they are a vital component in our daily life. Unfortunately, the deployment and expansion of new wireless technologies is being slowed down or even blocked by the inefficient access of radio spectrum. Historical (and current) spectrum allocation policies assign a fixed spectrum band to each wireless technology. Over the time, this static assignment results in an artificial "spectrum scarcity," including over-allocation and under-utilization of licensed bands, and an increasingly crowded unlicensed band [1]. To address spectrum scarcity and realize the potential of radio spectrum, we need new mechanisms to dynamically distribute spectrum among competing wireless devices according to their demand and usage. Dynamic spectrum allocation is feasible since new generation of wireless devices can quickly adjust their radio transmission frequencies with a wide range of spectrum, enabled by recent advance in Cognitive Radio hardware [2, 3]. While maximizing spectrum utilization is the primary goal of dynamic spectrum systems, a good allocation scheme needs to provide fairness across users. Dynamic spectrum management is challenging, particularly in large-scale wireless networks. Due to the phenomenon of radio interference [4], allocation of spectrum exhibits a form of externality. A user seizing spectrum without coordinating with others can cause harmful interference with its surrounding neighbors, and thus reducing spectrum utilization and degrading others' performance. Therefore, spectrum allocation needs to address the constraint of radio interference, which makes the problem NP-hard [4, 5]. There are multiple complimentary ways to address the problem of spectrum allocation, using approximation algorithms. The widely used approach is centralized approximations. Given a fixed topology, prior work [5, 6] reduces the allocation problem into a conventional graph coloring problem or its variants. A central manager obtains the conflict topology that specifies interference constraints among users [4], and performs coloring algorithms to derive a conflict-free spectrum assignment that intends to maximize a system utility. While operating based on global network knowledge, good centralized algorithms like graph-coloring face high complexity cost, and hence are not efficient in dynamic or large-scale networks. In particular, a topology-optimized allocation algorithm begins with no prior information, and assigns each user a close-to-optimal assignment. When network topology or spectrum availability change, the network needs to completely recompute spectrum assignments for all users after each change, resulting in high computational and communication overhead. This costly operation needs to be repeated frequently to maintain spectrum utilization and fairness. In addition, centralized algorithms require the existence of a centralized server. In this chapter, we consider a decentralized approach to spectrum allocation where instead of relying on any central servers, users perform local coordinations to modify their spectrum usage to approach a new conflict free spectrum assignment that maximizes system utility. In addition to being low-cost, this approach provides quick adaptation to topology variations. When network dynamics occur, our approach starts from the previous spectrum allocation, and performs a limited number of computations to arrive at a new solution to the new topology and spectrum availability. The rest of the chapter is organized as follows. We begin in Sect. 12.2 by defining the spectrum allocation problem and the system utility functions. Next, we propose a local coordination framework in Sect. 12.3 and develop specific strategies to improve system utilization and fairness in Sect. 12.4. We then in Sect. 12.5 provide a set of theoretical analysis to evaluate system utility and algorithm complexity. Next in Sect. 12.6, we conduct experiments to evaluate the performance of bargaining strategy and validate the theoretical lower bound. We summarize related work in Sect. 12.7, discuss implications and future directions before we conclude the chapter. © 2007 Springer Science+Business Media, LLC.