Active Queue Management Algorithm Based On Control Theory

The Internet is a time-varying, highly nonlinear system with large time lag, classical AQM schemes, due to their nonadaptive property, are incapable of controlling this kind of system, which brings out the necessity of designing AQM algorithms that are adaptive to network condition changes. In this dissertation, we propose methods which adapt controller parameters based on status of system read from control error, thus acheives adaptive control without performing indentification of the controlled system which involves great cost. The main research results are as follows:(1) An AQM algorithm based on PSD controller is proposed. PSD has a PID-like structure, it uses properties of error signal to indentify key parameters of the control loop, which are used to dynamically changing controller parameters, and thus becomes adaptive. Compared to other adaptive schemes, PSD has low computational complexity.(2) An AQM scheme based on Adaptive Weight Cascade PID controller is proposed. The output of AWCPID is composed of two parts:rate based and queue based packet dropping/marking probability. The ratio of the two parts is dynamically adapted according to the offsets of rate and queue length from the targets. An adaptive reference packets arrival rate, which is adapted based on network condition, is adopted to give the controller adaptivity. This algorithm deals with the immediate cause and cumulative effect of congestion efficiently.(3) An AQM scheme based on Variable Structure PID controller is proposed. The parameters of Variable Structure PID controller is adapted in such way that when the error of process is big, strengthen the influence of propotional part and weaken the influence of intergral and differential part, vice versa. This algorithm can speed up the response of system, avoid integral saturation phenomenon and also has excellent steady status performance.Simulation results show that the proposed algorithms are effective in mantaining queue lengths to be closely around the targets and are robust against parameter changes of the controlled TCP/AQM system.