Author: Haleema Mehmood

Publisher:

Published: 2015

Total Pages:

ISBN-13:

This dissertation addresses non-convex utility-maximization problems in multiuser, multicarrier communications systems. Utility maximization is an effective tool for communication system design. Utility functions are used to translate user or design preferences to optimization objectives. Physical-layer resource allocation in a communication system is then based on maximizing the utility of the system. This dissertations formulates a utility-maximization problem with a rate region constraint on the rate vectors. It considers various discrete optimization problems for multiuser, multicarrier communications systems. It presents algorithms that find utility-maximizing rate tuples for concave and non-concave utility functions over rate regions of Gaussian vector multiuser channels. The non-concave functions considered are staircase and sigmoidal utility functions that are widely used utility models for multimedia applications. Using a branch-and-bound method, a sequence of bounds on the optimal objective function value is obtained that converges to the global maximum sum of utilities. At each step, the algorithm solves a concave subproblem for multiuser power allocation using dual decomposition. For multicarrier systems, further decomposition of the Lagrangian across the subcarriers provides a low complexity method for utility maximization. The second part of this dissertation presents the concept of revenue potential as a tool for Internet service providers for revenue-based power allocation. Broadband Internet service providers are for-profit companies. From their perspective, utility of a multiuser channel is the total price they can charge their customers for provision of services. Broadband price functions are generally staircase functions of advertised data rates. The problem of finding the revenue potential of a DSL binder is a utility-maximization problem with a staircase utility function. The problem for the crosstalking DSLs is solved using the branch-and-bound method. A serious crosstalking problem arises in mixed deployments of vectored DSLs, sometimes known as G. Vector, and legacy DSLs. Such deployments require rate control on unvectored lines to realize vectoring gains. Revenue potential is used to determine the optimal-revenue rate-limits for unvectored lines. A strategy that limits the rates offered to the unvectored customers is shown to increase long-term revenue for the ISP. The third part of this dissertation presents a spiderweb plotting technique for multidimensional rate regions. This technique is useful for visualizing subsets of high-dimensional rate regions on a two-dimensional plot. Utility maximization is an alternative to building such rate regions and returns a single useful point instead of multiple data points. Both techniques are useful in different applied contexts. The last part of this dissertation considers a problem of joint user-clustering and bit-allocation for coaxial cable systems. The problem arises in next-generation coaxial cable systems, sometimes known as DOCSIS 3.1, with multicarrier modulation and adaptive bit loading. Users need to be grouped together and assigned bit profiles to maximize spectral efficiency. The problem is a discrete, non-convex, utility-maximization problem. A greedy, coordinate-ascent algorithm is presented to find clustering solutions to the utility-maximization problem. A comparison with optimal solutions found by exhaustive search shows that the algorithm gives close to optimal performance.