Dynamic resource allocations in wireless data communications.

Abstract

The newly emerged wireless data services pose new challenges to the scheduling and resource allocations including power and bandwidth. This dissertation is focused on these issues with the objective of developing guiding principles for new protocol designs to optimize the wireless network performance. The resource allocations are formulated as optimizations based on the channel information. We adopt the utility functions to quantify users' preference between conflicting objectives such as high throughput and low energy consumption. If the instantaneous channel measurements are available, a distributed power control algorithm is designed to optimize the CDMA system performance with the constraint that each user is guaranteed a payoff no less than that in the Nash Equilibrium of the non-cooperative power control game where each user maximizes its own payoff. The bandwidth scarcity makes dynamic optimization with incomplete channel information very appealing. Markov chain is shown by our statistical analysis as a generally reasonable dynamic model of the fading/shadowing channel at the packet/frame level. Therefore, we set up the dynamic optimization problems as the Partially Observable Markov Decision Processes (POMDP). Despite the widely known difficulties with POMDP, we have obtained the optimal transmission scheduling and power control. The above schemes can be extended to include queuing considerations. Markov-related traffic models can be easily accommodated. The queuing design with the self-similar traffic is still in its infancy. Our work starts from the real-time forecast of the traffic volume. We have invented a low-complexity adaptive predictor in which traffic volumes are aggregated properly so that a long history is captured by a small number of samples. The current result has its own significance in statistics and queuing design in wired networks.