Control of data networks: Models, stability and controllers.

摘要

In modern communication networks, packet switching is the switching method of choice, having obvious advantages. One drawback is the occurrence of delays varying from one packet to another. Trying to connect a controller and a plant through such a network is a non-trivial task, since the time-variant delay can easily lead to either instability or poor performance. The proposed work addresses these problems by constructing models for systems with time-variant communication delays, and studies the stability, and design of controllers for several classes of control systems.; In this work, the initial focus is on the problem of finding appropriate and sufficiently accurate models for the resulting feedback system. This process is broken down into finding models for the communication links between controller and plant, and finding models for the interface between the network and a plant/controller. The arising models are all time-variant delay discrete time systems, in which the time delay is bounded and uncertain. Several common congestion control systems will be presented and modeled.; Chapter 3 is dedicated to the problem of stability of the arising feedback system. The first contribution in this area is the development of an algorithm that dramatically reduces the complexity of testing a necessary and sufficient stability condition for time-variant delay systems. The second contribution is a set of simple to cheek, sufficient stability conditions. A study that compares the time-variant vs time-invariant stability condition is also performed. Finally, it is shown that buffer set point control with a stable equilibrium is an elusive goal in most networking environments.; Finally, the focus is switched to the design of an adequate controller for this type of system. It is shown that a typical time invariant controller cannot provide good performance; therefore, a Smith predictor based time-variant controller will be employed.; While the main results of this work are in the area of congestion control, this research is applicable to areas as diverse as automotive control, manufacturing systems, tele-operation, tele-robotics, load balancing in distributed computing, sensor networks, and many others.