Maximum likelihood estimation of ATM traffic model parameters

Parameter estimation occurs in many instances of traffic characterization associated with single source or aggregate traffic at the input or output of ATM multiplexers. The Markov modulated Poisson process for aggregated bursty sources, and the Markov modulated autoregressive process for variable bit rate encoded video sources are just two examples. The maximum likelihood estimation approach is a natural way to obtain parameters of these statistical models, which may involve incomplete data or a mixture of density functions. Iterative computation based on the expectation maximization algorithm can provide an attractive way of achieving maximum likelihood estimates of the parameters in these situations. We use the expectation maximization algorithm and show how this approach can be used to estimate the parameters of a Markov modulated autoregressive model for video traffic sources. Numerical results are discussed for such a model using a sequence of VBR video frames as observed data