Transmission scheduling of periodic real-time traffic in wireless networks

Abstract

An increasing number of applications rely on wireless networks for distributing information. Communicating time-sensitive data such as position, video, voice and telemetry, can be particularly challenging in wireless networks due to packet losses. In this thesis, we consider a single-hop wireless network, in which a base station is sending time-sensitive data packets to a set of clients. Our goal is to study transmission scheduling strategies for real-time traffic. Even though this problem has been explored in the literature, we present novel results that provide useful insight into the optimal scheduling problem. Previous work considered the problem of maximizing the throughput of networks with instantaneous feedback and without feedback. We address the general case of delayed feedback. Delayed feedback is particularly important for communication systems in which the round trip delay is much greater than the packet transmission time, and it has a significant impact on the scheduling decisions and network performance. In addition, we consider the case of clients receiving multiple parallel packet flows with heterogeneous deadlines. It is a well-known result that the Shortest Time to Extinction (STE) policy optimizes the throughput in wired networks. In this thesis, we establish a class of wireless networks for which the STE policy is throughput-optimal, i.e. minimizes the expected number of packets that expire due to the deadlines. Finally, we study the wireless network from the perspective of the Age of Information (AoI). This recently proposed performance metric represents the freshness of the information at the clients. We use Dynamic Programming to formulate and solve the problem of characterizing the scheduling policy that minimizes the AoI. The AoI metric is compared with throughput, and insights are drawn from numerical results. Simulations suggest that AoI-optimal policies are always throughput-optimal, while the converse is not true.