| Start Date: | 5/21/2014 | Start Time: | 10:00 AM |
| End Date: | 5/21/2014 | End Time: | 11:30 AM |
|
Event Description
Title: Efficient Control Decisions for Resource Allocation in OFDMA Networks
Advisor: Dr. John Walsh
Date: Wednesday, May 21, 2014
Time: 10:00 a.m.
Location: ECE Conference Room 302, 3rd Floor, Bossone Research Enterprise Center
Abstract
This thesis designs efficient control signaling for resource allocation in OFDMA networks, with special attention given to improving the resource controller in the LTE standard. We are interested in two aspects of resource controller design, the amount of control information a resource controller utilizes, and the performance, for instance the data spectral efficiency, it attains. Our overall aim is to understand the fundamental trade-off between these two quantities, and to learn how to design resource controllers that approach this trade-off. To get a sense of the state of the art in resource controller design, we first investigate the resource controller in the LTE standard, evaluating the amount of control information it requires. After this, we set about determining the fundamental trade-off between the amount of control information and the spectral efficiency by modeling the problem using information theory. To do this, we model the problem of resource allocation as a distributed source-coding problem called the CEO problem. Under this model, we are able to obtain an expression describing the fundamental limit for the trade-off between control information and performance as an instance of a rate distortion function. This expression involves an optimization problem, which is in some cases non-convex, whose solution gives the rate distortion function. Although for the resource allocation models under investigation there is no closed form expression for the solution to this optimization problem, we derive a novel adaptation of the Blahut-Arimoto algorithm to the CEO model with independent sources, and use it to numerically calculate the rate distortion function. Our novel algorithm always converges, and is locally convergent to the rate-distortion function global optimum, but, depending on the initialization, may not converge to the global optimum. For this reason, initialization is a key factor when using this algorithm. We show how to derive, for the minimum possible distortion for some problems, the minimum rate using graph entropy. This provides a global optimum for one value of a Lagrange multiplier, at which the developed algorithm for nearby Lagrange multipliers can be initialized to find their global optima. In this manner, we provide a method, consisting of the Blahut-Arimoto algorithm adaptation and a series of initializations, which, provided the change in Lagrange multipliers is small enough, is capable of numerically finding the rate distortion function and region for the CEO problem with independent sources. Finally, we build practical distributed quantizers that yield control signaling encodings for a resource controller that approaches the fundamental overhead performance trade-off limit, and in this manner improves upon the resource controller in the LTE standard. |
|
Location:
ECE Conference Room 302, 3rd Floor, Bossone Research Enterprise Center |
Audience:
Current StudentsFacultyStaffGraduate Students |
|
