Using Neural Network Classifier of Packet Loss Causes to Improve TCP Congestion Control over Ad Hoc Networks

A backpropagation neural network cIassifier is used for distinguishing network congestion and link error over ad hoc networks. TCP considers a 11 packet losses as network congestion, and reacts to such congestion by reducing its data rate. In ad hoc networks, there are many packet losses are due to Iink errors, but not due to network congestion; and then TCP often reduces its data rate mistakenly, cannot keep a reasonable data rate. Then we introduce an improvement TCP with a packet loss classifier, TCP-BP algorithm. The result of our simulation shows that the TCP-BP algorithm is superior to Vegas and TCP Westwood algorithm in TCP throughput.