Design approximation problems for linear-phase nonrecursive digital filters (Q696890)

In connection with the design of linear-phase non-recursive digital filters, the following four important approximation problems occur: (a) The (linear) problem of frequency response approximation: (b) The (nonlinear) problem of desired magnitude response approximation; (c) The simultaneous approximation of desired magnitude response and phase response; and (d) The simultaneous approximation of desired magnitude response and group delay response on proper subsets of the design frequency domain \(D\) (usually, a subset of \([0,\pi])\). For linear-phase filters, the above-mentioned four types of approximation problems reduce to one real linear minimum norm problem. The main purpose of this paper is to investigate systematically, from the mathematical point of view, this linear minimum norm problem for all \(L^p\) and \(l^p\) \((p\)-norms). Section 2 of the paper discusses the existence and uniqueness of a solution to the minimum norm problem for all \(p\)-norms, in relation to the assumptions of the second algorithm of Remez to the four mentioned approximation problems. Special attention is paid to the characterization of the obtained solution by the alternation condition in the case of the maximum norm \((p=\infty)\) and for the absence of constraints. Section 3 of the paper derives results on the algorithm convergence (guaranteed for every design frequency closed domain \(D)\) and on the rate of convergence of the sequence of approximation errors.