B-spline enhanced time-spectrum analysis

A new technique for the time-spectrum analysis of non-stationary signals is presented. The proposed technique smoothly fits a system's time-varying spectral coefficients using the combined methods of Fourier analysis and B-splines. The resulting algorithm is efficient and generally effective. Algorithm assumptions and limitations are identified; performance is explored using simulated data. Provided certain conditions are met, the algorithm degenerates into the well-known cases of the simple and averaged periodograms. Methods are presented to calculate knot spacing based on the frequency and geometric properties of the ensuing time-spectrum curve. Near real-time capabilities are also discussed. Finally, the method is compared with other time-spectrum analysis techniques such as the evolutionary periodogram (EP).