Empirical distributions of statistics of correlation analysis for protected information CDMA-systems

• Main Author: Сергій Леонідович Волков
• Format: Article
• Language: English
• Published: PC Technology Center 2013-09-01
• Series: Tehnologìčnij Audit ta Rezervi Virobnictva
• Subjects: distribution; statistics; correlation analysis; system, measurement; signature; synthesis.
• Online Access: http://journals.uran.ua/tarp/article/view/18417
• ISSN: 2226-3780; 2312-8372

It is noted in the paper that the multivariate normal distribution is not always a proper model to describe the actually observed signatures. For a number of statistics that are used in testing the hypotheses regarding the observed multi-dimensional variables it is shown that in the case of laws that differ from the multivariate normal distribution in a fairly wide range, significant changes in marginal distributions of statistics do not occur. Empirical distributions of statistics data are well described by the limiting laws obtained in the classical correlation analysis on the assumption of normality of the observed vector. It is shown that based on the correlation analysis results the conclusions can be made on the existence and nature of functional dependence or the preference of the particular regression model for description of the object under the study. The necessity to have the ability to model pseudo-random vectors under the laws with controlled deviation from the multivariate normal distribution is noted in order to trace the changes in distributions of the studied statistics of the correlation analysis. The results extend the scope of correct use of the methods of classical correlation analysis in applications.