Robust AQM Algorithm Of TCP Flow Model Of Time-varying Uncertainty

Continuous development of network technology makes the network congestion get worse. Active Queue Management (AQM) is the most effective method to solve the congestion control, which is valued by many researchers. In order to solve the impact of network parameter uncertainties and disturbances (mainly non-TCP flows and short life cycle formed TCP streams), this paper proposes several robust AQM algorithms, which combined with robust control theory and puts forward the following three algorithms:(1) Considering the short-term burst stream occupies the actual link bandwidth as the external disturbance while considering the time delay and parameter uncertainties, and then use a state-observer to obtain the immeasurable TCP window size combined with H-infinity robust control theory to design a state observer discrete H-infinity robust AQM algorithm. The simulation results show that, compared with the classical algorithm, the queue of the algorithm can stabilize in the near ideal value and better performance considering the disturbance and parameter uncertainties.(2) For dynamically computer network system, considering time delay, parameter uncertainty, modeling error and the impact of the actual structure deterministic of the system with nonlinear perturbations. The unmodeled was built as a one-dimensional wiener process, the system is described as a time-delay systems with nonlinear perturbations and norm-bounded uncertain parameters, combined with H-infinity robust control theory to design a stochastic robust AQM algorithm with nonlinear perturbation. Simulation results show that the algorithm could meet a given performance within the dynamically network environment.(3) Taking into account parameters uncertainties and disturbances that exist in the actual system (mainly non-TCP flows and short life cycle formed TCP streams), designing a disturbance observer to estimate the impact of disturbance, we proposed a robust AQM controller based on disturbance observer. By constructing a suitable Lyapunov functions, we solve LMI to get the controller. Simulation results show that this method can make the queue length converge rapidly to near the desired length and has a strong robustness to the load disturbances and the changes of the parameters.