A nonlinear approach to signal processing by means of vector measure orthogonal functions

Other special orthogonal polynomials and functions (33C47) Approximation by arbitrary nonlinear expressions; widths and entropy (41A46) Fourier series in special orthogonal functions (Legendre polynomials, Walsh functions, etc.) (42C10) Vector-valued measures and integration (46G10)

Review text

Let \(X\) be a Banach space, \(\Sigma\) a \(\sigma\)-algebra of subsets of a set \(\Omega\) and \(\lambda:\Sigma\to X\) a (\(\sigma\)-additive) vector measure. A sequence \((f_n)\) of real-valued \(\lambda\)-square integrable functions is called \(\lambda\)-orthogonal if \(\int f_n^2\,d\lambda \neq 0\) for all \(n\) and \(\int f_nf_k\,d\lambda=0\) for all \(n\neq k\). One considers also a control measure \(\mu:\Sigma\to \mathbb R\) for \(\lambda\) (i.e., \(\lambda\ll \mu\)) such that \(L_1(\lambda)\subset L_1(\mu)\). By Rybakov's theorem, such a measure always exists and is of the form \(\mu=|\langle \lambda,x_0'\rangle|\) for some \(x_0'\in X'\) (the dual of \(X\)). Another essential tool used in the paper is a Bochner integrable function \(\phi:\Sigma\to X'\) such that \(\lambda_{\phi(\omega)}(A):=\langle\lambda,\phi(\omega)\rangle(A)\), \(A\in \Sigma\), gives a positive measure (called parametric measure) that controls \(\mu \) for every \(\omega\in\Omega\). The nature of this function \(\phi\) is analysed and one shows that it can be pointwise approximated by simple functions in the norm of the Bochner space \(L_1(\lambda)\). Within this context, one can define the pointwise dependent Fourier coefficients of a function \(g\in L_2(\lambda)\) by \(\alpha^\phi_k(\omega)=\left(\int_\Omega gf_k\,d\lambda_{\phi(\omega)}\right)\big/\left(\int_\Omega f_k^2\,d\lambda_{\phi(\omega)}\right)\), \(k\in \mathbb N\). The authors study the convergence of the corresponding Fourier series with functional coefficients, given by \(\sum_{k=1}^\infty \alpha^\phi_k(\omega)f_k(\omega)\) for \(\mu\)-almost all \(\omega\in\Omega\), defining a convenient approximation procedure for signal processing involving time dependence of the measure. They also define a suitable functional structure and an error and prove a best approximation theorem with respect to this error. As a motivation, some classical examples by Stieltjes and Nevanlinna, related to the moment problem, are included.

Recommendations

Cites work

Cited in

(7)

This page was built for publication: A nonlinear approach to signal processing by means of vector measure orthogonal functions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q355143)