A Flexible Framework for Polynomial-Time Resource Allocation in Streaming Multiflow Wireless Networks

Abstract

We present a wide-area, multiflow ad-hoc network model leveraging information-theoretic rate control, emphasizing interfering rather than colliding transmissions. We seek to allocate resources in this network by optimizing scheduling, routing and power control to solve the max-min throughput problem for all flows involved. In general, our time-slotted, fully-interfering model leads to an NP-hard problem. Further, the rate-control element of the mixed-integer program results in a non-convex problem in the continuous domain. The complexity of the joint problem makes an optimal solution prohibitively difficult to find, leading us to propose a two-pronged approach to determine a near-optimal resource allocation. First, we propose a novel decomposition technique, breaking the joint problem into a sequence of more tractable subproblems. Second, we present a data structure serving multiple purposes: it compactly represents network conditions as they evolve with time, and also serves as the basis for our cubic-time dynamic programming algorithms used to solve and catalog the subproblems. The result is a schedule, route, and power allocation for all data frames involved. We demonstrate the performance of our techniques on the max-min throughput problem, while also showing that they are sufficiently general to apply to a wide variety of optimality criteria in which decisions over transmission schedules and packet routing must be made.