• Denis A. Khotimsky, Measurement and Estimation of Heterogeneous ATM Traffic: The rate-based flow control adopted by the ATM Forum as a target framework of congestion prevention and management relies on having an accurate picture of the actual network traffic. Measurements may be interpreted as samples from a stochastic process. An estimation procedure is required to derive useful network parameters from them. This dissertation focuses on defining and analyzing the models which allow for identification of the heterogeneous traffic components and estimation of their rates.

    The study of the characteristic features of ATM traffic leads to the abstraction of the heterogeneous process, a doubly-stochastic process which contains a hidden state and state-dependent observable output. As the complexity of the classic approach which involves the hidden Markov chain methods does not fit the high-speed ATM environment, an alternative technique based on exponential smoothing is proposed and analyzed. The main results include formulating the necessary and sufficient conditions of asymptotic unbiasedness, a sufficient condition for exponential smoothing to produce the linear minimum variance estimate, as well as sufficient conditions for the existence of a unique value of the decay parameter that minimizes the variance.

    The analysis of the distribution function of the heterogeneous process estimate involves the study of the limiting distributions of the random sums of the non-identically distributed independent random variables and leads to the development of the heterogeneous component identification algorithm. The censoring procedure identifies the intervals around the modes of the estimate's density function, the peak density intervals, and uses them to sample of the individual components of the heterogeneous process and to determine their moments.

    The primary contribution of this work is in the area of estimation theory (the error analysis of exponential smoothing estimate in the heterogeneous process environment) and network engineering (development of the censoring algorithm for the component rate identification). (August, 1996)