Delay-differentiated scheduling in a wireless network

This paper studies scheduling in a time-varying channel where different users have different delay-requirements. The scheduling problem is formulated as a stochastic dynamic programming problem, taking into account user-specific delay penalties and the time-varying channel state. It is shown that a scheduling policy closely resembling the exponential rule optimizes the throughput-delay trade-off in a single-user model, assuming the channel is not time-correlated. Previous work has proved the throughput-optimality of the exponential rule using fluid limit technique; in this paper the delay-throughout-optimality of a similar rule is shown using dynamic optimization with delay-penalties. In a multiple-access channel, heuristic approximations to the optimal single-user scheduling rule are studied numerically both in a time-correlated and in a memory-less environment.