A neural network model for traffic management in broadband networks

Asynchronous Transfer Mode (ATM) Broadband networks support a wide range of multimedia traffic (e.g. voice, video, image, and data). This research presents a novel framework for traffic management at the cell level using Neural Networks (NNs). Our new approach incorporates bank of NNs for traffic characterization and description, another NN system for traffic enforcement, and finally a reinforcement learning-based NN to provide a rate-based feedback access control. The NN traffic description method presents a novel approach to characterize and model the multimedia traffic. A bank of backpropagation NNs is used to characterize and predict the time bit-rate variations of the multimedia packet's arrival process. The NN traffic enforcement system includes two policing mechanisms: one is the "Neural Network Traffic Enforcement Mechanism (NNTEM)", the second is the "Reinforcement Learning Neural Network Controller". Both mechanisms do not rely upon the policing of simple parameters such as mean bit-rate, peak bit-rate, or burst duration, but rather an elaborate and very accurate policing of all higher-order moments via the probability density function (pdf) of the traffic. The rate-based feedback control is applied at the access node of the network and is implemented by the reinforcement learning method which, also, ensures an optimal control approach. The algorithm utilizes a feedback control signal to throttle the peak bit-rate of the arrival stochastic process to the input statistical multiplexer. The feedback control signal is produced (using the NN controller) such that the system performance is maximized. The system performance is defined in terms of the buffer overflow and the coding rate of the input source(s). The results of our new approach show that it is extremely effective in controlling and managing the ATM multimedia traffic when compared to existing methods such as Leaky Bucket and other window-type mechanisms.