A statistical method for assessing network stability using the Chow test

Abstract

A statistical method is proposed for the assessment of stability in noise monitoring networks. The technique makes use of a variation of the Chow test applied between multiple measurement nodes placed at different locations and its novelty lies in the way it utilises a simple statistical test based on linear regression to uncover complex issues that can be difficult to expose otherwise. Measurements collected by a noise monitoring network deployed in the center of Pisa are used to demonstrate the capabilities and limitations of the test. It is shown that even in urban environments, where great soundscape variations are exhibited, accurate and robust results can be produced regardless of the proximity of the compared sensors as long as they are located in acoustically similar environments. Also it is shown that variations of the same method can be applied for self-testing on data collected by single stations. Finally it is presented that the versatility of the test makes it suitable for detection of various types of issues that can occur in real life network implementations; from slow drifts away from calibration, to severe, abrupt failures and noise floor shifts.