A Constrained Quadratic Programming Technique for Data-Adaptive Design of Decorrelation Filters

Abstract

AbstractSignals from sensors with high sampling rates are often highly correlated. For the decorrelation of such data, which is often applied for the efficient estimation of parametric data models, discrete filters have proven to be both highly flexible and numerically efficient. Standard filter techniques are, however, often not suitable for eliminating strong local fluctuations or trends present in the noise spectral density. Therefore we propose a constrained least-squares filter design method. The spectral features to be filtered out are specified through inequality constraints regarding the noise spectral density. To solve for the optimal filter parameters under such inequality constraints, we review and apply the Active Set Method, a quadratic programming technique. Results are validated by statistical tests. The proposed filter design algorithm is applied to GOCE gradiometer signals to analyze its numerical behaviour and efficiency for a realistic and complex application.KeywordsActive Set MethodAdjustment with inequality constraintsDecorrelationFilter