Chance constrained stochastic optimization in wireless communication system
David Shui Wing Hui · 2014
Providing performance reliability in uncertain environments is an important task, yet a crucial challenge, in many engineering problems. Particularly for today’s high speed wireless communication systems (WCSs), obtaining accurate channel state information at the transmitter (CSIT) for scheduler design is a difficult task; hence robust design is of critical concern. In this thesis, we study how to achieve various design objectives in such adverse scenarios by modeling different Chance Constrained Stochastic Optimization (CCSO) problems; in particular, we introduce three novel problems and their corresponding solutions as follows. First we study throughput optimization under heterogeneous delay constraints via our proposed simple queueing theoretical formula for rate equivalence of delay performance. Second we provide distributive implementation and feedback reduction techniques via a statistical tool called extreme value theory. While the first two problems are studied under the classical error distribution model approach, our third contribution proposes a new CCSO framework, which allows one to get rid of the Cumulative Distribution Function (CDF) assumption of the uncertainty. Our approach further provides significant performance enhancement over classical CCSO by introducing an information-adaptive procedure that distinguishes uncertainty into useful information and noise, instead of the single type of uncertainty presented in the existing literature. We focus on demonstrating the applicability of these three solutions through a widely deployed example of a WCS called Orthogonal Frequency Division Multiple Access (OFDMA) system. In particular, we are the first to provide an optimal joint-subcarrier encoding design, with a jointly optimal subcarrier and power allocation which satisfies the outage constraint. Furthermore, both in simulation and theory, we demonstrate the performance enhancements of our schemes over existing CSIT-error inconsiderate schemes, and their capability to provide the aforementioned novel functionality: providing delay constraints satisfaction, distributive design with low feedback overhead, and online design in the absence of CSIT error statistics, without incurring extra complexity, and with convergence proof for the proposed algorithms.