Approximation of Internet Traffic using Robust Wavelet Neural Networks

Abstract

Robust Hurst parameter estimation of traffic data traces tops the bill of nowadays problems of the field of traffic engineering. Almost every going approach fits up the goal of as far as possible precise H parameter estimation; however this option is not as indispensable as approximate estimation of boundaries of H parameter if traffic demonstrates long range dependence. Constantly this is satisfactory condition for defining adequate traffic engineering operations. In this paper we verify a possibility of robust wavelet based H parameter estimation algorithm with ulterior traffic classification, which is based on wavelet transform of fractional Brownian motion synthesized data traces, and forthcoming wavelet coefficient clustering and operating with neural network learning capabilities. In this paper algorithm is described. Experimental data are depicted and future research subjects are pointed. Ill. 5, bibl. 14 (in English; summaries in English, Russian and Lithuanian).