Dynamic spectrum management architecture and algorithms for the future mobile Internet

This thesis presents an investigation of network assisted dynamic spectrum access techniques intended for use with emerging unlicensed, white space, and cognitive radio bands. Dynamic spectrum access (DSA) is motivated by the rapid proliferation of wireless devices which are expected to increase to the order to tens of billions by 2020. The dramatic increase in radio density by as much as 3-4 orders of magnitude relative to today's baseline implies the need for fundamentally new techniques that are both highly efficient and highly scalable. This thesis contributes towards that goal and studies wireless co-existence techniques in the `age of the Internet' - leveraging ubiquitous network connectivity of wireless devices to enable spectrum co-existence through distributed collaboration. The first part of the thesis describes the evolution of the mobile Internet, its relation to DSA techniques, and architectural solutions for better supporting current and future mobile Internet use-cases. Through this exercise, the need for network-level collaboration for improving the effectiveness of DSA techniques is shown. In the next two parts of the thesis, two specific applications of such an inter-network cooperation technique are presented - (i) Client-access point (AP) association optimization, and (ii) Channel selection. For the first application, the problem of connecting clients to APs is formulated as a non-linear integer program, and then the effect of inter-network cooperation is shown on the performance of the optimal solution. Large scale simulations with multiple overlapping networks, each consisting of 15-35 access points and 50-250 clients in a 0.5x0.5 sq.km show an average of 150\% improvement in random deployments and up to 7x improvements in clustered deployments for the least-performing client throughputs. In the channel selection application, a new scalable and accurate model for estimating the throughput of a Wi-Fi AP under arbitrary interference graphs is first shown. Based on this model, a graph based channel selection correction-phase is proposed, which can be appended to any centralized channel assignment scheme for performance improvement. Simulations with 100-500 APs/sq.km in homogeneous and mixed settings, corresponding to all APs adhering to same or different channel assignment schemes respectively, show ~$30% improvement in the number of starved APs. Further, in the case of mixed deployments, a key finding is made - as the percentage of centrally managed APs in a region is increased in comparison to simple residential APs, the performance of existing managed APs goes down due to decrease in the room for improvement. Results from a series of eight-node experiments on the ORBIT radio testbed are given for further validation of the simulation outcomes. Having shown the potential gains from cooperation in terms of client-AP association and channel selection, the final part of the thesis outlines the techniques through which such forms of cooperation can be practically implemented. In particular, a specific set of software-defined network (SDN) extensions for wireless control are described in the context of a dense Wi-Fi scenario with multiple network operators. Experimental results from a real-time proof of concept prototype using radio nodes on the ORBIT testbed are given. The results for a small two network scenario validate the proposed inter-network coordination protocol and demonstrate useful performance gains as density increases.Ph. D.Includes bibliographical referencesby Akash Bai