Multi-dimensional linear canonical transform with applications to sampling and multiplicative filtering (Q2146524)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Multi-dimensional linear canonical transform with applications to sampling and multiplicative filtering |
scientific article |
Statements
Multi-dimensional linear canonical transform with applications to sampling and multiplicative filtering (English)
0 references
16 June 2022
0 references
Given a \(2n\times 2n\) symplectic matrix \(M=\left(\begin{matrix} A & B\\ C & D\end{matrix}\right)\), that is, \(M^T J M=J\) where \(J=\left(\begin{matrix} 0 & I_n\\ -I_n & 0\end{matrix}\right)\), the (multidimensional) linear canonical transform (LCT) \(\mathcal{L}_M\) of \(f\in L^2(\mathbb{R}^n)\) is \[ \mathcal{L}_M[f](w)=\int_{\mathbb{R}^n} f(x) K_M(x,w)\, dx \] where (with \(\Omega(B,n)=(2\pi)^{-n/2} |\det{B}|^{-1/2}\) and \(|\det{B}|\neq 0\)) \[ K_M(x,w)=\Omega(B,n)\exp\left\{\frac{i}{2}(w^T DB^{-1}w-2w^TB^{-T}x+x^TB^{-1}Ax)\right\}\, . \] The Fourier transform corresponds to \(M=J\). The LCT satisfies the Parseval property \(\langle \mathcal{L}_M[f], \, \mathcal{L}_M[g]\rangle=\langle f,\, g\rangle\). The authors define a linear canonical convolution by \[ (f \circledast_M g)(t) =\int_{\mathbb{R}^n} f(x) g(t-x) \exp\left\{\frac{i}{2}(x^T B^{-1} A(x-t) + (x-t)^T B^{-1}A x )\right\}\, dx \] and prove (Thm.~2.1) that \[ \mathcal{L}_M[(f \circledast_M g)](w)=\frac{\exp\left\{-i(w^T DB^{-1}w)/2\right\}}{\Omega(B,n)} \mathcal{L}_M[(f )](w) \mathcal{L}_M[(g )](w)\, . \] A corresponding result (Thm.~2.3) is proved for what the authors call the linear canonical correlation, which is the same as the convolution but with \(g\) replaced by \(\bar{g}\) in the corresponding convolution/transform integral (and with \(\mathcal{L}_M[(g )]\) replaced by \(\overline{\mathcal{L}_M[(g )]}\) in the corresponding product formula). An analogue of the Heisenberg uncertainty inequality is proved (Thm.~3.1): \(\Delta_{f}^2\Delta_{\mathcal{L}_M[(f )]}^2\geq \frac{n^2}{4} |\det{B}|^2\) where \(\Delta_f^2=\frac{\|(x-x_0) f(x)\|^2}{\|f\|^2}\) with \(x_0=\|xf\|^2/\|f\|^2\). The inequality is an identity when \(f\) is a multiple of \(\exp\left(-\frac{ic x^TB^{-1}A x+2|x|^2}{2c}\right)\) for some \(c>0\). A version of Beckner's logarithmic (entropic) uncertainty inequality is also proved (Thm.~3.2), namely \[ \int \ln |w| |\mathcal{L}_M[f](w)|^2\, dw +\int \ln |x| |f(x)|^2\, dx\geq \left[\frac{\Gamma'(n/4)}{\Gamma(n/4)}+\ln |\det{B}|-\ln\pi\right]\|f\|^2\, . \] Perhaps the main result (Thm.~4.1) is a sampling formula analogous to Shannon's sampling theorem in the context of the LCT. One assumes that \(f\in L^2(\mathbb{R}^n)\) is bandlimited in the LCT domain to a region \(\mathcal{D}\) that is optimally contained in a regular polyhedral region denoted by \(\tilde{\mathbb{U}}(2E)\). Then \(f\) can be expressed in terms of its samples \(f(Nk)\) via \begin{multline*} f(x)=\frac{\det{N}}{(2\pi)^n|\det{B}|}\exp\left\{-\frac{i}{2}(x^T B^{-1}Ax \right\} \sum_{k\in\mathbb{Z}^n} f(Nk)\\ \times \left[ \exp\left\{-\frac{i}{2}((Nk)^T B^{-1}A(Nk) \right\} \left\{ \int_{\tilde{\mathbb{U}}(2E)}\exp\left\{-i((B^{-1}w)^T)(x-Nk) \right\} dw\right\} \right] \end{multline*} provided \(EK=\pi BN^{-T}\) for some non-singular integer matrix \(K\). Here \(\mathbb{U}(N)=\{Nx: x\in [0,1)^n\}\) for a nonsingular \(n\times n\) matrix \(N\), and \(\tilde{\mathbb{U}}(2N)\) is a translation of \(\mathbb{U}(2N)\). An ample discussion of the lattice coset geometry that underlies the result is provided and the result is illustrated for specific choices of \(M\). The authors assert that the approach here is the first that takes full advantage of the \(n(2n+1)\) dimensionality of the symplectic group.
0 references
linear canonical transform
0 references
convolution
0 references
symplectic matrix
0 references
sampling
0 references
multiplicative filtering
0 references
uncertainty principle
0 references
0 references
0 references
0 references
0 references