Congestion control of multimedia stream for ATM multiplexers using neural networks

A new method to the problem of congestion control arising at the user network interface (UNI) of ATM networks is proposed in this paper, based on neural networks. In the proposed method, the coding rate for multimedia stream and the corresponding user percentage are taken as the controller output to adjust the cells' arrival rate to the multiplexer buffer. Compared with the previous methods where the encoding rate is regulated in a body for all multimedia streams, the proposed method not only minimize the cell loss rate but also guarantee the quality of multimedia streams into the multiplexer buffer. Furthermore, two controller structures are given. In the first controller structure, the variables in the output layers of neural networks include the continuous coding rate and the corresponding user percentage, and the continuous coding rate is quantized to be discrete value. In the second controller structure, the neural network output only include the continuous coding rate, and discrete encoding rate and the corresponding user percentage are calculated by a given formula. Simulations with 150 voice sources into the multiplexer buffer and simulations with 30 video sources into the multiplexer buffer show that the proposed method is effective in controlling congestion and maintaining the quality of the traffic.