A neural network controller for congestion control in ATM multiplexers

This paper presents and adaptive approach to the problem of congestion control arising at the User-to Network Interface (UNI) of an ATM multiplexer. We view the ATM multiplexer as a non-linear stochastic system whose dynamics are ill-defined. Real-time measurements of the arrival rate process and the queueing process, are used to identify, and minimize congestion episodes. The performance of the system is evaluated using a performance-index function which is a quantative measure of “how well” the system is performing. A three-layers backpropagation neural network controller generates a signal that attempts to minimize congestion without degrading the quality of the traffic. During periods of buffer over-load the control signal, adaptively, modulates the arrival process such that its peak-rate is throttled-down. As soon as congestion is terminated, the control signal is adjusted such that the coding rates are restored back to their original values. Adaptability is achieved by continuously adjusting the weights of the neural network controller such that the performance of the system, measured by its performance index function, is maximized over a certain optimization period. The performance index function is defined in terms of two main objectives: (1) to minimize the cell loss rate (CLR), i.e., minimize congestion episodes, and (2) to maintain the quality of the video/audio traffic by maintaining its original source coding rate. The neural network learning process can be viewed as a specialized form of reinforcement learning in the sense that the control signal is reinforced if it tends to maximize the performance index function. Performance evaluation results prove that this approach is effective in controlling congestion while maintaining the quality of the traffic.