Wireless communication networks develop rapidly, featuring excessive demand for performance improvements. This thesis studies performance and fairness issues of opportunistic schedulers, widely used in modern wireless standards like HSDPA and HDR. Opportunistic schedulers take into account the condition of the wireless link in order to select a terminal for service during a timeslot, by allocating to it all the available network resources. The main attribute of this type of scheduling is that the average transmission rate does depend on the number of active terminals, in the form of a multiuser diversity gain which acts as a multiplicative factor on the average rate.The above characteristics yield to a State Dependent Processor Sharing (SDPS) model, which in its general form considered here, features separate classes of terminals and a generic form of the state space, as a result of the admission control scheme. From a mathematical point of view, we manage to provide the distribution of the sojourn time for an SDPS system, which is expressed through the exponential of an appropriate M-matrix. The form allows the characterization of the asymptotic decay rate of the sojourn time’s distribution tail and provides easily computable upper and lower bounds.
Furthermore, the conducted research studies time-scale separation phenomena, by separating the classes of terminals into a slow and a fast group, according to their average transmission rate. By leveraging on the nearly completely decomposable structure of the full system, we arrive in accurate approximations that offer conceptual and computational simplicity. Additionally, we consider some special important cases of the service rate function and/or the admission control scheme, which allow to determine the effect of the presence of slow class terminals on the performance of the terminals belonging in the fast group of classes.
The results and conclusions of the generalized problem have been applied in the practical case of a channel-aware wireless network. A comparison of two basic opportunistic schedulers with respect to performance and fairness issues follows. We result in several conclusions regarding the value and the suitability of each scheduler, the appropriate form of the service pool and the specification of the worst case scenario related to the fast class terminals. Moreover, a new scheduling scheme is proposed, combining the nice properties oh the previous two schedulers. The theoretical findings are compared and verified numerically.