The structure of shift-modulation invariant spaces: the rational case (Q880092)

The author develops fundamental aspects of the structure of subspaces \(V\) of \(L^2(\mathbb{R}^n)\) that are invariant under a lattice of shifts on the one hand and under integer modulations \(M_k f(x)=e^{2\pi i x\cdot k}f(x)\) on the other. By a lattice \(\Lambda\), one means the image of \(\mathbb{Z}^n\) under an invertible matrix \(P\) having integer entries. Its dual lattice is \(\Lambda^\ast=\{\gamma\in\mathbb{R}^n: \gamma\cdot \Lambda\subset \mathbb{Z}^n\}\). The quotient \(\Lambda^\ast/\mathbb{Z}^n\) has \(p=| \text{det}P|\) coset representatives \(d_1,\dots,d_p\) and induces an action \([k]\circ (z_1,\dots,z_p)=(z_{\nu(1)},\dots,z_{\nu(p)})\), where \(\nu\) is the unique permutation of \(\mathbb{Z}_p\) such that \([k]+[d_i]=[d_{\nu(i)}]\). One denotes by \(L^2_{\Lambda}(\mathbb{T}^n,\mathbb{C}^p)=\{f\in L^2(\mathbb{T}^n,\mathbb{C}^p): f(x+k)=[k]\circ f(x),\;k\in \Lambda^\ast\}\). Letting \(I_n\) and \(I_{\Lambda^\ast}\) denote fundamental domains for \(\mathbb{Z}^n\) and \(\Lambda^\ast\), respectively, one can define a vector Zak transform \(\mathcal{Z}f(x,\xi)=(Zf(x,\xi+d_1),\dots,Zf(\xi+d_p))\), where \(Zf(x,\xi)=\sum_{k\in\mathbb{Z}^n}f(x-k)e^{2\pi ik\cdot\xi}\). The shift-modulation invariant (SMI) subspace is then expressed in terms of a measurable range function \(J\) that assigns to each \((x,\xi)\in \mathbb{R}^n\times\mathbb{R}^n\) a linear subspace \(J(x,\xi)\) of \(\mathbb{C}^p\) such that \([k]\circ J(x)=J(x+k)\), \(k\in \Lambda^\ast\). The first main theorem states that there is, indeed, a one-to-one correspondence between SMI spaces \(V\), defined as closed spans of countable unions \(e^{2\pi ix\cdot k}\varphi_\alpha(x+\lambda)\), where \(k\in \mathbb{Z}^n\), \(\lambda\in \Lambda\), and \(\alpha\) ranges over some countable set \(\mathcal{A}\) of generators, and measurable range functions \(J\) defined on \(I_n\times I_{\Lambda^\ast}\) and mapping \((x,\xi)\) to \(J(x,\xi)=\text{span}\{\mathcal{Z}\varphi_\alpha(x,\xi):\alpha\in\mathcal{A}\}\). If one defines the dimension function \(\dim_V(x,\xi)=\dim J(x,\xi)\), then the second main result can be stated as follows. Two SMI spaces \(V\) and \(W\) are unitarily equivalent if and only if their dimension functions agree almost everywhere on \(I_n\times I_{\Lambda^\ast}\). The third main result gives a more practical decomposition of SMI spaces in terms of finitely generated spaces, namely, \[ V=\bigoplus_{i=1}^p S(\varphi_i,\Lambda), \] where each \(\varphi_i\) is a principal generator -- meaning the SMI space with the single generator \(\varphi_i\) has dimension function bounded by one -- and such that the sets where the dimension functions are nonzero are nested (if \(\dim S(\varphi_{i},\Lambda)>0\), then \(\dim S(\varphi_{i+1},\Lambda)>0\)). The last few results address Gabor theory and the relationship between Gabor frames and basis properties for SMI spaces and for the finite-dimensional range spaces \(J(x,\xi)\). The main result here essentially provides a correspondence between Gabor frames, bases, etc., and their properties for generators of the SMI spaces into corresponding properties for corresponding generators of the range spaces. The correspondence is expressed in terms of the Zak transform.