On Active Queue Management Algorithms For TCP Networks Based On Control Theory

With the rapid increase of network scale and the continouous growth of network users and applications, the network congestion has become an important problem. The router-based active queue management mechanism, which combines with the TCP protocol congestion control, is the main approach to solve the problem of congestion control for the TCP networks at present. So the research on active queue management algorithm has become an active issue in the field of the TCP networks.The TCP networks can be considered as a nonlinear dynamic feedback control system. So much more effective results can be achieved from the perspective of control theory. Based on the analysis of the active queue management mechanism, the dissertation presents some active queue management algorithms by means of the methods in control theory. The main research works and conclusions are as follows:For the problem of congestion control in TCP networks, an adaptive sliding mode control algorithm is presented based on the RBF neural network. Considered the upper bound of the system uncertainties may not be easily obtained, the RBF neural network is used to learn the uppler bound of system uncertainties in order to avoid achieving the the upper bound of system uncertainties in advance. And the output of the RBF neural network is used to compensate the upper bound of system uncertainties, so that the effects of the system uncertainties can be eliminated. The RBF neural network is used to design an adaptive sliding mode controller which not only ensure the existence of the sliding mode on the surface and asymptotic stability of the systems, but also eliminate the effects of the system uncertainties. Simulation results verify the favorable stability and robustness of the algorithm.For the TCP networks in the presence of highly variable network parameters and UDP flows, two nonlinear adaptive sliding mode control algorithms are presented. Because the upper bound of the system uncertainties could not easily be obtained, an adaptive sliding mode controller is presented to the upper bound of the system uncertainties. However, the chattering could not be eliminated effectively with this algorithm because of the use of sign function. So, another adaptive sliding mode controller is presented to directly adapt the system uncertainties. Simulation results demonstrate that the adaptive sliding mode controller for the system uncertainties is superior to that for the upper bound, and the adaptive sliding mode controller for the upper bound is superior to PI controller.In order to improve the robustness of the TCP networks, a nonlinear active queue management algorithm is presented based on adaptive global sliding mode control. The global sliding mode control is used to eliminate the arrival stage of sliding mode control and ensure the robustness of network systems in the whole control process. The RBF network is used to directly approximate the lumped uncertainties of the systems so that the estimation error is effectively reduced. Because the sign function and the saturation function are not used, not only the system chattering is effectively eliminated, but also the system response is smoother. Simulation results show that the proposed algorithm has good robustness and fast system response.For the TCP networks in the presence of more abrupt network load and time-varying round trip time, an active queue management algorithm is presented based on RBF neural networks and sliding mode control. Since network system parameters are unknown and time-varying, the RBF neural network is used to approximate the network system parameters so that the active queue management algorithm is easily implemented. The network system parameters are well estimated by updating the RBF neural network weights according to Lyapunov theory. By using the output of the RBF neural network as the sliding mode controller parameters, an active queue management algorithm was designed to guarantee the network system is asymptotically stable. Simulation results show that the proposed algorithm has fast system response and steady queue length as well as better robustness under various network conditions.For the problem of finite-time congestion control in TCP networks in the presence of uncertainty and the disturbances of the unresponsive flows, two adaptive finite-time active queue management algorithms are presented. First, an adaptive finite-time active queue management algorithm is proposed based on terminal sliding mode and RBF neural network. The algorithm uses RBF network to estimate the lumped uncertainty of the system so that the knowledge of the bound of the lumped uncertainty is not required. An adaptive terminal sliding mode controller is designed based on the output of the RBF neural network. However, the computation of this algorithm is more complex, which not only reduces the executive efficiency of routers but also makes the deployment in routers difficult. So, another adaptive finite-time active queue management algorithm is proposed based on backstepping method. With the help of the algorithm, the instantaneous queue length can converge to the desired queue length in finite time. Simulation results demonstrate that the algorithm has better finite-time stability and robustness.Concerning with the congestion control for TCP networks in the presence of input constraint, parameter uncertainties and UDP flow disturbances, two active queue management algorithms are proposed based on adaptive sliding mode control. First, an adaptive queue management algorithm is presented based on the conventional sliding mode control. Although the algorithm has better robustness against the system parameter variation and UDP flow disburtances, it can only guarantee the system is asymptotic. So, another active queue management algorithm is proposed based on terminal sliding mode control in order to improve the convergent performance of the TCP network systems. It makes the system obtain better finite-time stability. Simulation results verify the effectiveness of the proposed algorithm.Lastly, the summary of the whole dissertation is given and the research directions in future are put forward.