Shiftable multiscale transform by Simoncelli et all, proof proposition 1.

39 Views Asked by At

I'm reading through enter link description here this paper. And I'm trying to understand the proof of the proposition 1, which states:

The transformation defined by $$ y[n] = \int_{0}^{2\pi} dx h(n\Delta_x - x)f(x) $$ is shiftable if and only if there's a set of interpolation functions, $b_n(x_0)$ that satisfy the matrix equation $$ \begin{pmatrix} e^{jx_0k_0} \\ e^{jx_0k_1} \\ \vdots \\ e^{jx_0k_{M-1}} \end{pmatrix} = \begin{pmatrix} 1 & e^{j\Delta_x k_0} & e^{j 2 \Delta_x k_0} & & e^{j(N-1)\Delta_x k_0} \\ 1 & e^{j\Delta_x k_1} & e^{j 2 \Delta_x k_1} & \ldots & e^{j(N-1)\Delta_x k_1} \\ & \vdots & & \\ 1 & e^{j\Delta_x k_{M-1}} & e^{j 2 \Delta_x k_{M-1}} & & e^{j(N-1)\Delta_x k_{M-1}} \end{pmatrix} \begin{pmatrix} b_0(x_0) \\ b_1(x_0) \\ \vdots \\ b_{N-1}(x_0) \\ \end{pmatrix} \;, \forall x_0 $$

The paper proves such proposition. And I understand the proof. However the form of the linear system above doesn't seem to me depends from the $h$, and therefore to me it seems that for any $h$ we have a shiftable transformation.

PS. Shiftability means that for all $x_0$ we have

$$ \int_{0}^{2\pi} dx h(x_0 - x)f(x) = \sum_{j=0}^{N-1} b_j(x_0) y[n] $$

Namely convolution can be expressed as interpolation of $N$ samples. Notation in general of the paper is not entirely clear, for example Fourier series of $h(x)$ is denoted by $H(k)$, with a different variable, while to me the variable should be the same.

1

There are 1 best solutions below

1
On

The statement of the proposition continues after presenting the linear system. It continues by saying: "where $\{k_0, k_1, \dots, k_{M-1}\}$ is the set of frequencies for which $H(k)$, the Fourier series of $h(x)$, is nonzero". I guess that gives your dependence on $h$.