Event Description
Speaker: Gwanmo Ku
Title: Efficient Control Decisions for Resource Allocation in OFDMA Networks
Advisor: Dr. John Walsh
Date: Monday, November 18, 2013
Time: 10:00 a.m.
Location: ECE Conference Room 302, 3rd Floor, Bossone Research Enterprise Center
Abstract
Multiuser OFDMA (Orthogonal Frequency Division Multiple Access) systems have a growing tendency to use resource control technologies such as resource allocation, adaptive modulation and coding (AMC), Hybrid ARQ (HARQ), MIMO, and CQI reporting in order to increase system throughput and reliability. All the technologies mentioned above require a large amount of control information for data transmission. These control signaling overhead take a significant proportion of the today wireless network traffic and often not efficiently encoded. In order to remove the redundant control information, we focus on the distributed lossy source coding technique dealing with the rate distortion function that is an optimum trade-off between a rate and a desired distortion. The primary research issues are thus what the minimum compression rate of control information is, how we can encode the minimum amount of control information, and how we can design an efficient practical resource controller approaching to the fundamental limit.In this proposal, we utilize distributed source coding theory to find the fundamental trade-offs between control (collaboration) information overhead and system performance for resource allocation in OFDMA systems, compare these trade-offs with the amount of overhead and performance achieved by existing resource allocation designs in the 4G cellular standards, and design an improved resource controller which approaches the fundamental trade-off. In particular, our work will establish fundamental limits explicitly describing the trade-offs between the performance of a distributed cooperative resource controller and the amount of collaboration overhead information it must exchange. We will demonstrate that LTE control signaling overhead, at about 32 % of downlink transmission, is still not near to the theoretical performance limit that could be achieved with the same amount of control information. We will also review the existing resource controller design employed in the LTE standard, focusing on the way in which control signals and salient feedback signals for resource allocation and link adaptation are encoded. We then set about, determining the minimum, over all possible resource controller designs, amount of control information that is necessary to obtain a target resource allocation performance. In particular, we develop a simple model for a resource allocation problem by assuming that each user in the system knows their own channel state information, and must compress this information, then send the result to the basestation. The basestation, in turn, upon receipt of this compressed channel state information must make the best resource allocation and link adaptation decisions possible. In order to illustrate our ideas here, we take as our metric the spectral efficiency, but our approach could be modified to include other metrics reflecting fairness and quality of service as well. A key idea utilized in this thesis is to view this overall resource controller design as an instance of a lossy source code for the CEO problem, in which the encoding nodes are the compressors of the channel state, and whose decoding node, the CEO, must estimate the resource allocation that maximizes the spectral efficiency. The loss measured is the reduction in spectral efficiency relative to an omniscient controller having access to all of the channel state in the network. A key issue is how to compute the rate distortion function. In particular, while we will show that, owing to independence of the channel states, there is a nice analytical optimization problem for the rate distortion function describing the fundamental tradeoff between spectral efficiency and the number of bits utilized for the compressed channel state, computing the solution of this optimization must be done numerically. In this vein, a major innovation is a modification of the Blahut-Arimoto algorithm for computing the rate distortion function for the lossy compression of a single source to a hybrid alternating minimization (whose components are two Blahut-Arimoto-like algorithms) which can be shown to converge for the class of CEO problems under investigation. We utilize this technique to calculate the fundamental limits for the resource allocation model we have introduced, utilizing a channel state distribution that is motivated by the LTE standard. Having established fundamental performance limits for the simplified model for resource allocation and link adaptation, we set about designing practical resource controllers to approach these limits.
|
