Signal auto smoothing

The classical approaches of noise reduction do not adapt well to the variety of signals that can be encountered for different analysis methods (chromatography, XRD, Raman ...). The parameterization of these methods is often complex, with many parameters not very explicit or not intuitive and to be adapted according to the case to be processed. They are also very difficult to use routinely. To make these tools operational, they must adapt easily to many scenarios, and especially they have to be simple to parameterize: a noise reduction approach will be difficult to use in practice if the parameters are too numerous or difficult to adjust, even if this method surpasses all others in terms of the quality of the results provided.

Our strategy at IFPEN has therefore been to develop new algorithms adapting locally to a wide variety of cases, while being the simplest to parameterize, that is to say with few parameters, and easily interpretable parameters for non-expert domain. The chosen approach was to take inspiration from a representation of the signal seen as a relief with bumps and valleys, approach easily explainable to a non-expert of the signal processing, and for that we turned to the morphology mathematical.

Our noise reduction algorithm uses a variety of noise reduction filters whose setting related to the size of the computing neighborhood adapts automatically and locally to the signal. Conventional linear filtering techniques express the smoothed value at a point as a linear combination of the values of the samples located in the interval around the considered point. This filtering can be interpreted in terms of scale corresponding intuitively to the length and the weighting applied. The choice of the right filtering scale is vital in order to preserve as much as possible the local properties of the signal that we wish to analyze. For more details, refer to the patent indicated below.