An entropy-switched adaptive smoothing approach for time series data

This paper describes a method of removing noise from time series data records whilst preserving salient features of short duration, such as sharp transitions and significant peaks. A practical example is drawn from fault-current testing of circuit breakers, for which the scheme was originally designed. It is demonstrated that the clarity of signal traces can be improved while preserving important transient features. However, the approach is generic and based upon the entropy gradient detection method used in image processing. Local entropy is used as a criterion for selecting the degree of smoothing required, so that features of interest can be preserved. Algorithm modularity allows ready adaptation for specific needs.