a new model for error clustering in telephone circuits - e-strongmstrong.pdfVIP

J. M. Berger B. Mandelbrot A New Model for Error Clustering in Telephone Circuits Abstract: This paper proposes a new mathematical model to describe the distribution of the occurrence of errors in data transmission on telephone lines. Wesuggest: a) that the statistics of telephone errors can be described in terms of an error probability depending solely on the time elapsed since the last occurrence of an error; b) that the distribution of inter-error intervals can be well approximated by a law of Pareto of exponent less than one; the relative number of errors and the equivocation tend, therefore, to zero a5 the length of the message is increased. The validity of thoseconcepts is demonstrated with the aid of ex- perimental data obtained from the Germantelephonenetwork.Further consequences,refinements, and uses of the model are described in the body of the paper. 1. Introduction: The competingmodels The model that we shall propose for describingthe occur- fromgood to bad states and viceversa. The bad state rence of errors is best describedby contrasting it with two gives riseto the cluster of errors, and the transition proba- earlier alternatives. Thefirst is the binary symmetric chan- bilityischosen to account for the occurrence of these nel without memory. In that case, the transmission of any clusters. This second type of modelobviouslygivesrise given bit is not influenced by the