9.5 LeastSquares Filters Purdue University（普渡大学9.5 LeastSquares过滤器）.pdfVIP

9.5 Least-Squares Digital Filters • objective is to fit sections of the data to a polynomial using least-squares • choice of polynomial • centering the data and convolution • derivation of a 5-point quadratic filter • Mathcad program to generate filters • noise reduction vs. filter type and length • smoothing Johnson noise • non-smoothing of interference noise • smoothing and simultaneous differentiation 9.5 : 1/12 Objective objective • the noise is known to have a normal pdf, but the signal frequencies are unknown • a least-squares-based technique can be used that fits a section of the measured data to a polynomial • the entire data set is smoothed by fitting successive sections of the data to the same polynomial choice of polynomial • any polynomial can be chosen, however, it should be capable of matching the slope, curvature, inflection points, and positions of maxima and minima • a straight line can match only the slope of the data • a quadratic equation can match the slope and curvature • a cubic equation can match the slope, curvature and inflection points of the data - however, the cubic equation being odd will shift the location of peaks • the quartic equation is the first that possesses all the desired traits 9.5 : 2/12 Operational Considerations centering the filter • least squares filters work by convolution • the filter has an odd number of elements and is centered about the data point being smoothed • for smoothing only the intercept is computed • centering simplifies the calculation by eliminating odd powers of the independent variable • because of the elimination of odd powers, the