Stochastic system identification for ATM network traffic models: A time domain approach

In our paper we discuss a new time domain approach to the traffic identification problem for ATM networks. The Markov modulated Poisson process is identified in two steps. By applying a nonnegative least squares algorithm we obtain in a very fast way a description of the first order statistics of the data. This first order characterisation includes also an estimate of the model order. Consequently, we are able to identify a Markov modulated Poisson process without a priori knowledge of the model order. The identification of the second order statistics is based on unconstrained optimisation algorithms.