Estimating loss rates in an integrated services network by neural networks

Loss rate is one of the most important quality of service (QoS) requirements in a packet communication network carrying multimedia traffic. This paper presents a method for estimating loss rates as a function of a feature vector, x, based on a maximum likelihood principle. Two backpropagation networks, a multi-layer perceptron (MLP) network and a radial basis function (RBF) network, are applied to samples of simulated M/M/1 queue data, as well as to samples of simulated combinations of constant bit rate (CBR) and Poisson traffic data, at different x. Our results show that a) this method is superior to earlier methods since it properly treats the samples, eliminating nuisance parameters, b) it gives a direct measure of QoS that can be used in higher level communication network decision functions, and c) RBF networks give more accurate loss rate estimates than MLP networks do