Study Of Congestion Control Algorithm And Stability Analysis For Data Service

This thesis addresses the design and stability analysis with time delay for data service in communication networks. First, A modified linear quadratic gauss (MLQG) congestion controller with the use of frequency domain filter and the fractal auto regressive integral moving average (FARIMA) model is given for long range dependent networks. The MLQG controller can bring queue length of the bottleneck node to the desired value in the presence of round trip time (RTT) and self-similar traffic. The use of high order neural network (HONN) is further investigated to approximate the unknown nonlinear model of queuing dynamics of a single buffer. The convergence with RTT of HONN based explicit rate (ER) congestion control and adaptive law can be guaranteed using Lyapunov stability arguments. Simulations validate that HONN based ER congestion control can decrease cell loss with delay and unknown available capacity. Also, this thesis exploits the robustness of a PI congestion controller in ATM networks with respect to time-varying virtual connections. In the context of Internet protocol (IP) network supporting transmission control protocol, the principle of variable structure (VS) control is used to design active queue management (AQM) algorithm and to analyze stability. To alleviate the chattering problem in VS control, fuzzy sliding mode (FSM) AQM with the least fuzzy rules is presented to achieve the satisfactory performance in terms of stabilized queue length, high utilization and low packet loss ratio. Local stability with RTT for FSM AQM is also considered. Moreover the global stability results of optimization congestion control under different network topology are given using Lyapunov-Razumikhin method. At last with the aim of downlink radio resource allocation in code division multiple access (CDMA) system, an optimization model is given based on economics utility functions. With the dynamic pricing algorithm joint power allocation, the downlink throughput can be maximized within the power constraint. Global stability with distributed action delays for this algorithm can be guaranteed by means of Lyapunov-Razumikhin arguments. Simulation results demonstrate that dynamic pricing outperforms the peer congestion price algorithm in throughput maximization.