Opportunistic scheduling and resource allocation among heterogeneous users in wireless networks
MetadataShow full item record
This dissertation studies and proposes new methods to perform opportunistic scheduling in different scenarios for centralized wireless networks. We first study the performance of measurement-based opportunistic scheduling strategies in practical scenarios where users’ heterogenous capacity distributions are unknown. We make the case for using maximum quantile scheduling, i.e., scheduling a user whose current rate is in the highest quantile relative to its current empirical distribution. Under fast fading, we prove a bound on the relative penalty associated with such empirical estimates, showing that the number of independent samples need only grow linearly with the number of active users. Furthermore, we show several desirable properties of maximum quantile scheduling for the saturated regime, and give bounds on performance under the dynamic regime. Next we propose a novel class of opportunistic scheduling disciplines to handle mixes of real-time and best effort traffic at a base station. The objective is to support probabilistic service rate guarantees to real-time sessions while still achieving opportunistic throughput gains across users and traffic types. Under fast fading and maximum quantile scheduling, we are able to show a stochastic lower bound on the service a real-time session would receive. Such bounds are critical to enabling predictable quality of service and thus the development of complementary resource management and admission control strategies. Our simulation results show that the scheme can achieve more than 90% of the maximum system throughput capacity while satisfying the QoS requirements for real-time traffic. Finally, we propose methods to reduce the feedback overhead for users’ channel state information needed for opportunistic scheduling at a base station. We first propose a contention based scheme known as ‘static splitting’ for a best effort traffic only scenario. Next we consider reducing feedback overhead in a system supporting a mixture of best effort and real-time traffic. We argue that one needs to combine contention based schemes with polling subsets of users to reduce the amount of feedback needed to exploit opportunism, and yet meet real-time users’ QoS guarantees. Based on this argument we propose a joint polling and opportunistic scheduling (JPOS) scheme.