On Discrete Polynomial Least-Squares Approximation in Moving Time Windows
Erich Fuchs · Birkhäuser Basel eBooks · 1999
The paper presents a modification algorithm for least-squares approximations in moving time windows for equidistant discrete data. Properties of discrete shift and difference operators are used to derive fast square-rootand division-free algorithms, especially suitable for implementation on DSPs, for some particular discrete weights. It can be shown that a modification of an equally weighted least-squares fit needs not more than 5 n + 2 multiplications and 5 n - 2 additions for polynomial degrees n ≤ 5.