On Modeling And Congestion Control For Wireless Networks

The TCP protocol can provide reliable end-to-end communication services and protect the communication performance on the Internet, which makes it be the most widely used protocol. Wireless networks are able to meet needs of connecting the Internet anytime and anywhere, so they have been widely concerned in recent years. To networks, congestion control is a core issue all the time, especially in wireless networks. Because the original TCP was designed for a low rate of loss of packets in wired networks, and it assumed that the lost packets is only due to congestion. If extending the original algorithm of congestion control in TCP to wireless networks in a straightforward manner, it will lead to low performances. Otherwise, the original model of congestion control used in wired networks couldn’t meet the dynamic characteristics of wireless networks. Therefore, how to improve TCP performances under wireless networks is very important.At first, the thesis analyses reasons of congestion, introduces the procedure of TCP congestion control, and elaborates characteristics of wireless networks. Then introduces the original TCP congestion control into wireless networks, analyses performances of TCP, and indicates shortages and improves algorithms in wireless networks.Because the assumption that congestion is the only reason of the lost packets is not suitable for wireless networks, the thesis pulls in the mechanism of Explicit Congestion Notify to get rid of the effects of the lost packets, which is not caused by congestion. The mechanism can make the model of wireless networks more reasonable. Finding out the original model dosen’t suit wireless networks through researchs, the bandwidth estimation algorithm is introduced into the thesis. But the original bandwidth estimation algorithm was affected by data burst, etc. which makes the available bandwidth jitter and can not adjust the bandwidth adaptively according to interval time of ACKs. The thesis puts forward the improvement on those two problems using the index approximation and the exponential moving average filter, which could get the available bandwidth more accurately. Based on the mechanism of Explicit Congestion Notify and the bandwidth estimation algorithm, the thesis improves the original model to meet the dynamic characteristic. This thesis designs the single neural adaptive PI controller for the improved model, and then carries out the simulation for the length of queue. The simulation results indicate that the effect of the controller is effective.