Degradation detection of wireless IP links based on local stationary binomial distribution models

A degradation detection problem of link quality in a long-distance 2.4 GHz wireless system is discussed. The time series to be monitored is periodic and non-stationary. The decision algorithm for degradation is difficult to define, and methods based on conventional traffic theory are not useful for IP link quality. Thus we should introduce some kind of intelligent data analysis technique. The authors propose to apply an AI-based method which solves a similar problem in a commercial switching telephone and ISDN network. The method partitions a target time-series into local stationary segments. Optimization of partitioning is based on the minimal Akaike (1974) information criterion principle. The technique called sequential probability ratio test is also applied to make efficient decisions about degradation. Thus experiments to apply our proposed method to this domain are conducted with wireless systems at a real field. The result shows the AI-based method is also effective for the degradation detection of wireless IP links.