This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

The system of translates and the special affine Fourier transform

Md Hasan Ali Biswas Department of Mathematics, Indian Institute of Technology Madras, Chennai - 600036, India mdhasanalibiswas4@gmail.com Frank Filbir Mathematical Imaging and Data Analysis, Helmholtz Center Munich,
Department of Mathematics, Technical University of Munich, Germany
filbir@helmholtz-muenchen.de, frank_filbir@web.de
 and  Radha Ramakrishnan Department of Mathematics, Indian Institute of Technology Madras, Chennai - 600036, India radharam@iitm.ac.in
Abstract.

The translation operator TAT^{A} associated with the special affine Fourier transform (SAFT) A\mathscr{F}_{A} is introduced from harmonic analysis point of view. The analogues of Wendel’s theorem, Wiener theorem, Weiner-Tauberian theorem and Bernstein type inequality in the context of the SAFT are established. The shift invariant space VAV_{A} associated with the special affine Fourier transform is introduced and studied along with sampling problems.

Key words and phrases:
Convolution, Poisson summation formula, shift invariant space, Wendel’s theorem, Zak-transform.
*Corresponding author: Radha Ramakrishnan, email:radharam@iitm.ac.in
2020 Mathematics Subject Classification:
Primary 42A38; Secondary 42A85, 42C15

1. Introduction and background

The special affine Fourier transform (SAFT) was first considered by S. Abe and J. T. Sheridan in [1] for the study of certain operations on optical wave functions. The SAFT is formally defined as

(1.1) Af(ω)=12π|b|f(t)ei2b(at2+2pt2ωt+dω2+2(bqdp)ω)𝑑t,\mathscr{F}_{A}f(\omega)=\frac{1}{\sqrt{2\pi|b|}}\int_{\mathbb{R}}f(t)e^{\frac{i}{2b}(at^{2}+2pt-2\omega t+d\omega^{2}+2(bq-dp)\omega)}dt,

where AA stands for the set {a,b,c,d,p,q}\{a,b,c,d,p,q\} of real parameters which satisfy the relation adbc=1ad-bc=1. The integral transform (1.1) is related to the special affine linear transform of the phase space

(1.2) (tω)=(abcd)(tω)+(pq).\begin{pmatrix}t^{\prime}\\ \omega^{\prime}\end{pmatrix}=\begin{pmatrix}a&b\\ c&d\end{pmatrix}\begin{pmatrix}t\\ \omega\end{pmatrix}+\begin{pmatrix}p\\ q\end{pmatrix}.

Due to the conditions on the parameters a,b,c,da,b,c,d the matrix in (1.2) belongs to the special linear group SL(2,)\mathrm{SL}(2,\mathbb{R}) and the affine transform (1.2) is therefore given by elements from the inhomogeneous linear group

ISL(2,)={(Mv01):MSL(2,),v2}.\mathrm{ISL}(2,\mathbb{R})=\left\{{\footnotesize\left(\begin{array}[]{cc}M&v\\ 0&1\end{array}\right)}:\,M\in\mathrm{SL}(2,\mathbb{R}),\ v\in\mathbb{R}^{2}\right\}.

This justifies the name special affine Fourier transform for (1.1). The action of the group SL(2,)\mathrm{SL}(2,\mathbb{R}) on the time frequency plane and the relation to quadratic Fourier transforms is well studied. We will not go into the details here but refer to the book [11].

A number of important transforms are special cases of the SAFT. For example, A={0,1,1,0,0,0}A=\{0,1,-1,0,0,0\} gives the ordinary Fourier transform and A={0,1,1,0,0,0}A=\{0,-1,1,0,0,0\} its inverse. The parameter set A={cosθ,sinθ,sinθ,cosθ,0,0}A=\{\cos\theta,\sin\theta,-\sin\theta,\cos\theta,0,0\} gives the fractional Fourier transform, and A={1,λ,0,1,0,0}A=\{1,\lambda,0,1,0,0\} produces the Fresnel transform.

In optics, certain one parameter subgroups of the ISL(2,)\mathrm{ISL}(2,\mathbb{R}) are of special interest. Among them are the fractional Fourier transform, the Fresnel transform (also called free space propagation in this context), the hyperbolic transform A={coshθ,sinhθ,sinhθ,coshθ,0,0}A=\{\cosh\theta,\sinh\theta,\sinh\theta,\cosh\theta,0,0\}, the lens transform A={1,0,λ,1,0,0}A=\{1,0,\lambda,1,0,0\}, and the magnification transform A={eβ,0,0,eβ,0,0}A=\{e^{\beta},0,0,e^{-\beta},0,0\}. The latter two cases need a careful analysis for the limit case b0b\to 0 which we will not consider in this paper. We will not try to expound the various connections to optics further but refer to [15] for more details.

In this paper, we consider (1.1) from the point of view of applied harmonic analysis and take it as a signal transform of a (suitable) function. We are mainly interested in studying the principal shift invariant spaces and sampling theorems related to the SAFT. In the classical case, the principal shift invariant space generated by ϕL2()\phi\in L^{2}(\mathbb{R}) is defined as V(ϕ)=span¯{Tkϕ=ϕ(k):k}V(\phi)=\overline{span}\{T_{k}\phi=\phi(\cdot-k):k\in\mathbb{Z}\}. The classical Fourier transform (A={0,1,1,0,0,0}A=\{0,1,-1,0,0,0\} in (1.1)) plays a crucial role for the analysis of such spaces. The crucial point is that the ordinary translation and the classical Fourier transform are intimately related. This is due to the identity eiω(tx)=eiωteiωxe^{i\omega(t-x)}=e^{i\omega t}e^{-i\omega x} which gives, as a consequence the convolution theorem, the relation between translation, modulation, Fourier transform etc. These theorems are used over and over again in Fourier analysis and in the study of shift invariant spaces in particular. It is completely obvious that the ordinary translation does not interact nicely with the SAFT (resp. its complex exponential kernel). Hence working with the SAFT a new concept of a translation is needed. This generalized translation should be linked to the SAFT in an analogous manner as the ordinary translation is linked to the classical Fourier transform. If this is the case, then it seems reasonable to expect that the central theorems (convolution theorem etc.) hold in a similar manner. An idea for the construction of such generalized translation comes from the observation that in the classical setting we have

Txf(t)=1(eiωxf)(t).T_{x}f(t)=\mathscr{F}^{-1}(e^{-i\omega x}\mathscr{F}f)(t).

In this paper, we define a new translation operator TxAT^{A}_{x}, which serves our purpose in the case of SAFT.

In [6] A. Bhandari and A.I. Zayed considered chirp-modulation and used this to obtain a convolution theorem. However, they did not define a generalized translation operator explicitly, and hence did not investigate the consequences of this concept with respect to harmonic analysis. We shall demonstrate that the generalized translation is the suitable concept to obtain analogues of fundamental theorems such as Wendel’s theorem for the multipliers, Wiener theorem and Weiner-Tauberian theorem in connection with the closed ideals of translation invariant spaces in the context of the SAFT. For a study of multipliers and Wendel’s theorem for the Fourier transform we refer to [18], for Wiener theorem and Wiener-Tauberian theorem we refer to [12] and [22]. The novelty of this approach is that apart from these theorems, one can look into the study of multiplier theory, including Hörmander multiplier theorem in the SAFT domain. Moreover, one can define an appropriate modulation operator in connection with the SAFT, using which one can define modulation spaces associated with the SAFT. This in turn, motivates to study multiplier results for the new modulation spaces. (See [7] in this connection.)

Using the new translation operator TxAT^{A}_{x} the AA-shift invariant spaces are defined as VA(ϕ)=span¯{TkAϕ:k}V_{A}(\phi)=\overline{span}\{T^{A}_{k}\phi:k\in\mathbb{Z}\} for an appropriate function ϕL2()\phi\in L^{2}(\mathbb{R}). When ϕ\phi belongs to the Wiener amalgam space W(C,1())W\big{(}C,\ell^{1}(\mathbb{Z})\big{)} the space VA(ϕ)V_{A}(\phi) turns out to be a reproducing kernel Hilbert space. Moreover, we give characterization theorems for the system of translates {TkAϕ:k}\{T^{A}_{k}\phi:k\in\mathbb{Z}\} to be a frame sequence, orthogonal system or a Riesz sequence. If the system of translates is a frame then an important question is about the nature of the dual frame elements. We show that in our setting, the elements of the dual frame of system of AA-translates are also AA-translates of a single function. For a study of shift invariant spaces, system of translates, frames and Riesz basis in the classical case, we refer to Christensen [10].

In the final part of the paper we study the sampling in AA-shift invariant spaces. A fundamental problem in sampling theory is to find, for a certain class of functions, appropriate conditions on a countable sampling set X={xj:jJ}X=\{x_{j}\in\mathbb{R}:j\in J\} under which a given function fVf\in V can be reconstructed uniquely and stably from the samples {f(xj):jJ}\{f(x_{j}):j\in J\}. We refer to the work of Butzer and Stens [8] for a review on sampling theory and its history. When VV is the classical principal shift invariant space with a single generator V(ϕ)V(\phi) or multi-generators V(ϕ1,ϕ2,,ϕr),ϕ,ϕ1,ϕ2,ϕrL2()V(\phi_{1},\phi_{2},...,\phi_{r}),~{}\phi,\phi_{1},\phi_{2},...\phi_{r}\in L^{2}(\mathbb{R}), there is a huge literature available on several interesting problems connected with sampling theory starting from the fundamental Shannon sampling theorem. We cite only a few references in this connection for the reader to get familiarity with this subject matter. (See [2], [3], [4], [5], [9], [13], [14], [17], [19], [20], [24], [25], [26], [27]).

In this paper, similar to Theorem 4.24.2 of [3] and Theorem 2.12.1 of [23], we obtain equivalent conditions for a set XX to be a stable set of sampling for VA(ϕ)V_{A}(\phi) in terms of the operator UU where Ujk=TkAϕ(xj)U_{jk}=T_{k}^{A}\phi(x_{j}), the reproducing kernel and the Zak transform ZAZ_{A}, which we introduce here. We also obtain a sufficient condition for the set of integers to be a stable set of sampling for VA(ϕ).V_{A}(\phi). In the study of non uniform sampling and average sampling, Bernstein type inequalities play an important role. In this paper we obtain an analogue of Bernstein type inequality for VA(ϕ)V_{A}(\phi). However, we do not intend to study non-uniform sampling and average sampling in this paper. We focus on uniform sampling. In particular, when \mathbb{Z} turns out to be a stable set of sampling, we obtain a reconstruction formula and hence a sampling theorem in the sense of L2L^{2} convergence for certain AA-shift invariant spaces VA(ϕ)V_{A}(\phi). Further, under some additional hypotheses on ϕ\phi, we obtain a sampling formula in the sense L2L^{2} convergence and uniform convergence. As corollaries we obtain Shannon sampling theorem in the SAFT domain and sampling theorem for the AA-shift invariant space generated by second order BB-spline.

We organize our paper as follows. In section 22, we define AA-convolution of a measure and a function and chirp modulation CsμC_{s}\mu of a measure μ\mu. Using this and the function ρA(x)=ei2b(ax2+2px)\rho_{A}(x)=e^{\frac{i}{2b}(ax^{2}+2px)}, we obtain a relation between classical translation and AA-translation. We prove an analogue of Wendel’s theorem for the SAFT. In section 33, we study closed ideals in the Banach algebra (L1(),A)(L^{1}(\mathbb{R}),\star_{A}). We obtain analogues of Wiener theorem and Wiener-Tauberian theorem in the context of the SAFT. In section 44, we study AA-shift invariant spaces and their theoretical aspects. Section 55 as well as Section 66 are devoted to sampling theorems in AA-shift invariant spaces. Finally in Section 77, we present a local reconstruction method for sampling in AA-shift invariant spaces along with implementation.

Now we shall provide the necessary terminology and background for this paper.

Let 00\neq\mathcal{H} be a separable Hilbert space.

Definition 1.1.

A sequence {fk:k}\{f_{k}:k\in\mathbb{N}\} of elements in \mathcal{H} is a frame for \mathcal{H} if there exist m,M>0m,M>0 such that

mf2k=1|f,fk|2Mf2,f.m\|f\|^{2}\leq\sum_{k=1}^{\infty}|\langle f,f_{k}\rangle|^{2}\leq M\|f\|^{2},\qquad~{}~{}~{}f\in\mathcal{H}.

The numbers m,Mm,M are called frame bounds. If we have the right hand side inequality for a sequence in \mathcal{H}, then that sequence is called a Bessel sequence.

Definition 1.2.

Let {fk:k}\{f_{k}:k\in\mathbb{N}\} be a Bessel sequence in \mathcal{H}, then the synthesis operator T:2T:\ell^{2}\to\mathcal{H} is defined by

T({ck})=k=1ckfk,{ck}2.T(\{c_{k}\})=\sum_{k=1}^{\infty}c_{k}f_{k},\quad\{c_{k}\}\in\ell^{2}.

The adjoint of TT is given by T(f)={f,fk},fT^{*}(f)=\{\langle f,f_{k}\rangle\},~{}f\in\mathcal{H}, called the analysis operator. Composing TT and TT^{*} we obtain the frame operator

S:,S(f)=k=1f,fkfk.S:\mathcal{H}\to\mathcal{H},~{}S(f)=\sum_{k=1}^{\infty}\langle f,f_{k}\rangle f_{k}.

The operator SS is invertible. Further if {fk:k}\{f_{k}:k\in\mathbb{N}\} is a frame for \mathcal{H}, then {S1fk:k}\{S^{-1}f_{k}:k\in\mathbb{N}\} is also a frame for \mathcal{H} and it is called the canonical dual frame of the frame {fk:k}.\{f_{k}:k\in\mathbb{N}\}.

Definition 1.3.

A sequence {fk:k}\{f_{k}:k\in\mathbb{N}\} in \mathcal{H} is said to be a Riesz basis if there exist a bounded invertible operator TT on \mathcal{H} and an orthonormal basis {uk:k}\{u_{k}:k\in\mathbb{N}\} of \mathcal{H} such that fk=Tuk,k.f_{k}=Tu_{k},~{}\forall~{}k\in\mathbb{N}. The sequence {fk:k}\{f_{k}:k\in\mathbb{N}\} is called a Riesz sequence if it is a Riesz basis for its closed linear span.

Equivalently {fk:k}\{f_{k}:k\in\mathbb{N}\} is a Riesz sequence if there exist m,M>0m,M>0 such that

mc2k=1ckfk2Mc2,for every finite sequence{ck}.m\|c\|^{2}\leq\|\sum_{k=1}^{\infty}c_{k}f_{k}\|^{2}\leq M\|c\|^{2},~{}~{}\text{for every finite sequence}\ \{c_{k}\}.
Definition 1.4.

Let {fk:k}\{f_{k}:k\in\mathbb{N}\} be a Riesz basis for \mathcal{H}. The dual Riesz basis of {fk:k}\{f_{k}:k\in\mathbb{N}\} is the unique sequence {gk:k}\{g_{k}:k\in\mathbb{N}\} in \mathcal{H} satisfying

f=k=1f,gkfk,f.f=\sum_{k=1}^{\infty}\langle f,g_{k}\rangle f_{k},\quad\forall f\in\mathcal{H}.
Definition 1.5.

The Gramian associated with the Bessel sequence {fk:k}\{f_{k}:k\in\mathbb{N}\} is an operator on 2\ell^{2} whose jkthjk^{th} entry in the matrix representation with respect to the canonical orthonormal basis is fk,fj.\langle f_{k},f_{j}\rangle.

It is well known that a sequence {fk:k}\{f_{k}:k\in\mathbb{N}\} is a Riesz sequence if there exist m,M>0m,M>0 such that its Gramian GG satisfy the following inequality:

mc2k|Gck,ck|2Mc2,c={ck}2.m\|c\|^{2}\leq\sum_{k\in\mathbb{N}}|\langle Gc_{k},c_{k}\rangle|^{2}\leq M\|c\|^{2},~{}~{}\forall~{}c=\{c_{k}\}\in\ell^{2}.
Definition 1.6.

A closed subspace VV in L2()L^{2}(\mathbb{R}) is said to be a shift invariant space if fVTkfV,k,fV,f\in V\Rightarrow T_{k}f\in V,~{}\forall k\in\mathbb{Z},~{}f\in V, where Tkf(t)=f(tk).T_{k}f(t)=f(t-k).

In particular, for ϕL2(),V(ϕ)=span¯{Tkϕ:k}\phi\in L^{2}(\mathbb{R}),~{}V(\phi)=\overline{span}\{T_{k}\phi:k\in\mathbb{Z}\} is called the principal shift invariant space.

Definition 1.7.

A set X={xk:k}X=\{x_{k}\in\mathbb{R}:k\in\mathbb{Z}\} is said to be a stable set of sampling for a closed subspace VV of L2()L^{2}(\mathbb{R}) if there exist constants 0<mM<0<m\leq M<\infty such that

mf2k|f(xk)|2Mf2,m\|f\|^{2}\leq\sum_{k\in\mathbb{Z}}|f(x_{k})|^{2}\leq M\|f\|^{2},

for every fV.f\in V.

Definition 1.8.

The Wiener amalgam space W(C,p())W\big{(}C,\ell^{p}(\mathbb{Z})\big{)}, 1p<1\leq p<\infty is defined as

W(C,p())={fC():kmaxx[0,1]|f(x+k)|p<}.W\big{(}C,\ell^{p}(\mathbb{Z})\big{)}=\{f\in C(\mathbb{R}):\sum_{k\in\mathbb{Z}}max_{x\in[0,1]}|f(x+k)|^{p}<\infty\}.

2. The new translation

In this section, we introduce AA-translation operator in connection with the SAFT. Using AA-translation operator, we define AA-convolution of a regular Borel measure and a function. Further, we obtain an analogue of Wendel’s theorem. Towards this end, first we extend the definition of SAFT to the space of all regular Borel measures.

Definition 2.1.

Let fL1()f\in L^{1}(\mathbb{R}). Then the special affine Fourier transform is defined as

(2.1) Af(ω)=12π|b|f(t)ei2b(at2+2pt2tω+dω2+2(bqdp)ω)𝑑t,ω,\mathscr{F}_{A}f(\omega)=\frac{1}{\sqrt{2\pi|b|}}\int_{\mathbb{R}}f(t)e^{\frac{i}{2b}(at^{2}+2pt-2t\omega+d\omega^{2}+2(bq-dp)\omega)}dt,~{}\omega\in\mathbb{R},

where AA stands for the set of six parameters {a,b,c,d,p,q}\{a,b,c,d,p,q\}\subset\mathbb{R} with adbc=1ad-bc=1 and b0b\neq 0.

With the help of the following auxiliary functions

(2.2) ηA(ω)\displaystyle\eta_{A}(\omega) =ei2b(dω2+2(bqdp)ω),\displaystyle=e^{\frac{i}{2b}\left(d\omega^{2}+2(bq-dp)\omega\right)},
(2.3) ρA(t)\displaystyle\rho_{A}(t) =ei2b(at2+2pt),\displaystyle=e^{\frac{i}{2b}(at^{2}+2pt)},

the SAFT can be expressed as

(2.4) Af(ω)=ηA(ω)|b|ρA(f)(ω/b)\mathscr{F}_{A}f(\omega)=\frac{\eta_{A}(\omega)}{\sqrt{|b|}}\rho_{A}(f)(\omega/b)

Since |ηA(ω)|=1=|ρA(t)||\eta_{A}(\omega)|=1=|\rho_{A}(t)| for all t,ωt,\omega\in\mathbb{R}, we immediately see from (2.4) that AfC0()\mathscr{F}_{A}f\in C_{0}(\mathbb{R}) with Af(2π|b|)1/2f1\|\mathscr{F}_{A}f\|_{\infty}\leq(2\pi|b|)^{-1/2}\,\|f\|_{1}. Moreover, (2.4) also shows that A\mathscr{F}_{A} can be extended to L2()L^{2}(\mathbb{R}) and defines a unitary operator on that space. In particular,

Af2=f2.\|\mathscr{F}_{A}f\|_{2}=\|f\|_{2}.

The inverse of A\mathscr{F}_{A} on L2()L^{2}(\mathbb{R}) can also be easily determined using (2.4)

A1f(t)=ρA¯(t)2π|b|f(ω)ηA¯(ω)eiωt/b𝑑ω.\mathscr{F}_{A}^{-1}f(t)=\frac{\overline{\rho_{A}}(t)}{\sqrt{2\pi|b|}}\int_{\mathbb{R}}f(\omega)\overline{\eta_{A}}(\omega)\,e^{i\omega t/b}d\omega.

Finally, (2.4) also provides an extension of the SAFT to M()M(\mathbb{R}), the space of all complex valued bounded regular Borel measures on \mathbb{R}, equipped with the total variation norm. For μM()\mu\in M(\mathbb{R}), we have

Aμ(ω)=12π|b|ei2b(at2+2pt2tω+dω2+2(bqdp)ω)𝑑μ(t),ω,\mathscr{F}_{A}\mu(\omega)=\frac{1}{\sqrt{2\pi|b|}}\int_{\mathbb{R}}e^{\frac{i}{2b}(at^{2}+2pt-2t\omega+d\omega^{2}+2(bq-dp)\omega)}d\mu(t),\,\omega\in\mathbb{R},

with AμCb()\mathscr{F}_{A}\mu\in C_{b}(\mathbb{R}) and Aμ(2π|b|)1/2μM()\|\mathscr{F}_{A}\mu\|_{\infty}\leq(2\pi|b|)^{-1/2}\|\mu\|_{M(\mathbb{R})}.

We now introduce the generalized translation operator associated with SAFT. In order to do so, we fix the following notation

(2.5) Txf(t)\displaystyle T_{x}f(t) =f(tx).\displaystyle=f(t-x).
(2.6) Mωf(t)\displaystyle M_{\omega}f(t) =eiωtf(t).\displaystyle=e^{i\omega t}f(t).
Definition 2.2.

Let xx\in\mathbb{R} and f:f:\mathbb{R}\to\mathbb{R} be a function. Then AA-translation of ff by xx, denoted by TxAfT_{x}^{A}f, is defined as

TxAf(t)=TxMax/bf(t)=eiabx(tx)f(tx),t.T_{x}^{A}f(t)=T_{x}M_{-ax/b}f(t)=e^{-i\frac{a}{b}x(t-x)}f(t-x),~{}t\in\mathbb{R}.

It is easy to see that TxAT_{x}^{A} is norm preserving in all spaces Lp(),1pL^{p}(\mathbb{R}),~{}1\leq p\leq\infty or C0()C_{0}(\mathbb{R}).

We can relate our new translation and the classical translation in the following way, using ρA\rho_{A} and the chirp modulation operator Cab,C_{\frac{a}{b}}, where

(2.7) Csf(t)=eis2t2f(t).C_{s}f(t)=e^{i\frac{s}{2}t^{2}}f(t).
(2.8) CabTxAf=eia2bx2Tx(Cabf)\displaystyle C_{\frac{a}{b}}T_{x}^{A}f=e^{i\frac{a}{2b}x^{2}}T_{x}(C_{\frac{a}{b}}f)
(2.9) ρATxAf=ρA(x)Tx(ρAf).\displaystyle\rho_{A}T_{x}^{A}f=\rho_{A}(x)T_{x}(\rho_{A}f).

In fact,

CabTxAf(t)=eia2bt2eiabx(tx)f(tx)=\displaystyle C_{\frac{a}{b}}T_{x}^{A}f(t)=e^{i\frac{a}{2b}t^{2}}e^{-i\frac{a}{b}x(t-x)}f(t-x)= eia2bx2eia2b(tx)2f(tx)\displaystyle e^{i\frac{a}{2b}x^{2}}e^{i\frac{a}{2b}(t-x)^{2}}f(t-x)
=\displaystyle= eia2bx2Tx(Cabf)(t).\displaystyle e^{i\frac{a}{2b}x^{2}}T_{x}(C_{\frac{a}{b}}f)(t).

Similarly one can show (2.9).

Now, we collect the properties of TxA.T_{x}^{A}.

Proposition 2.3.

We have the following

  • (i)

    TxATyA=eiabxyTx+yA,x,y.T_{x}^{A}T_{y}^{A}=e^{-i\frac{a}{b}xy}T_{x+y}^{A},\quad x,y\in\mathbb{R}.

  • (ii)

    (TxA)=eiabx2TxA,x.(T_{x}^{A})^{*}=e^{-i\frac{a}{b}x^{2}}T_{-x}^{A},\quad x\in\mathbb{R}.

  • (iii)

    Let χω(t)=ρA¯(t)eiωt/b\chi_{\omega}(t)=\overline{\rho_{A}}(t)\,e^{i\omega t/b}. Then TxAχω(t)=χω¯(x)χω(t)T^{A}_{x}\chi_{\omega}(t)=\overline{\chi_{\omega}}(x)\,\chi_{\omega}(t).

  • (iv)

    Let fL1()f\in L^{1}(\mathbb{R}). Then

    (2.10) A(TxAf)(ω)=ρA(x)eixw/bAf(ω),x,ω.\mathscr{F}_{A}(T_{x}^{A}f)(\omega)=\rho_{A}(x)e^{-ixw/b}\mathscr{F}_{A}f(\omega),\quad x,\omega\in\mathbb{R}.
Proof.

The proof is straightforward. ∎

It is interesting to note that the map xTxx\mapsto T_{x} is a group representation, whereas from Proposition 2.3 (i) it follows that xTxAx\mapsto T_{x}^{A} is just a projective representation in general, which shows that the new translation is fundamentally different from that of the classical translation. Proposition 2.3 (iii) is what is known in harmonic analysis as a product formula. The relation (2.10) extends to functions fL2()f\in L^{2}(\mathbb{R}) as well. For those functions we have in particular

(2.11) TxAf=A1(ρA(x)eix/bAf),T_{x}^{A}f=\mathscr{F}_{A}^{-1}(\rho_{A}(x)e^{-ix\cdot/b}\mathscr{F}_{A}f),

where the equality holds in the sense of L2()L^{2}(\mathbb{R}) functions.

Definition 2.4.

Let μM()\mu\in M(\mathbb{R}) and ss\in\mathbb{R}, then CsμC_{s}\mu is defined by d(Csμ)(x)=ei2sx2dμ(x).d(C_{s}\mu)(x)=e^{\frac{i}{2}sx^{2}}d\mu(x).

Clearly, CsμM().C_{s}\mu\in M(\mathbb{R}). Similarly, we can define ρAμ\rho_{A}\mu as d(ρAμ)(x)=ρA(x)dμ(x)d(\rho_{A}\mu)(x)=\rho_{A}(x)d\mu(x).

Using the AA-translation, we define the AA-convolution of μM()\mu\in M(\mathbb{R}) and fL1()f\in L^{1}(\mathbb{R}) as

(2.12) (μAf)(t)=12π|b|TsAf(t)𝑑μ(s).(\mu\star_{A}f)(t)=\frac{1}{\sqrt{2\pi|b|}}\int_{\mathbb{R}}T^{A}_{s}f(t)\,d\mu(s).

The integral in (2.9) can also be viewed as a vector-valued integral as follows.

(2.13) μAf=12π|b|TsAf𝑑μ(s),\mu\star_{A}f=\frac{1}{\sqrt{2\pi|b|}}\int_{\mathbb{R}}T_{s}^{A}fd\mu(s),

where the right hand side is a Bochner integral. The convergence of the integral follows from

TsAf1𝑑μ(s)=f1μ()<.\int_{\mathbb{R}}\|T_{s}^{A}f\|_{1}d\mu(s)=\|f\|_{1}\mu(\mathbb{R})<\infty.

Now we give a relation between classical convolution and AA-convolution of a measure and a function. Consider

(μAf)(x)\displaystyle(\mu\star_{A}f)(x) =12π|b|TsAf(x)𝑑μ(s)\displaystyle=\frac{1}{\sqrt{2\pi|b|}}\int_{\mathbb{R}}T_{s}^{A}f(x)d\mu(s)
=12π|b|eia2bs2Cab1TsCabf(x)𝑑μ(s)\displaystyle=\frac{1}{\sqrt{2\pi|b|}}\int_{\mathbb{R}}e^{i\frac{a}{2b}s^{2}}C_{\frac{a}{b}}^{-1}T_{s}C_{\frac{a}{b}}f(x)d\mu(s)
=eia2bx22π|b|TsCabf(x)eia2bs2𝑑μ(s)\displaystyle=\frac{e^{-i\frac{a}{2b}x^{2}}}{\sqrt{2\pi|b|}}\int_{\mathbb{R}}T_{s}C_{\frac{a}{b}}f(x)e^{i\frac{a}{2b}s^{2}}d\mu(s)
=eia2bx22π|b|TsCabf(x)d(Cabμ)(s)\displaystyle=\frac{e^{-i\frac{a}{2b}x^{2}}}{\sqrt{2\pi|b|}}\int_{\mathbb{R}}T_{s}C_{\frac{a}{b}}f(x)d(C_{\frac{a}{b}}\mu)(s)
=eia2bx2|b|(CabμCabf)(x),\displaystyle=\frac{e^{-i\frac{a}{2b}x^{2}}}{\sqrt{|b|}}(C_{\frac{a}{b}}\mu\star C_{\frac{a}{b}}f)(x),

using (2.8), which in turn implies that

(2.14) Cab(μAf)=1|b|(CabμCabf).C_{\frac{a}{b}}(\mu\star_{A}f)=\frac{1}{\sqrt{|b|}}(C_{\frac{a}{b}}\mu\star C_{\frac{a}{b}}f).

Further, using (2.9), one can show that

(2.15) ρA(μAf)=1|b|(ρAμ)(ρAf).\rho_{A}(\mu\star_{A}f)=\frac{1}{\sqrt{|b|}}(\rho_{A}\mu)\star(\rho_{A}f).

The convolution theorem for the SAFT reads as follows.

Proposition 2.5.

Let μM()\mu\in M(\mathbb{R}) and fL1()f\in L^{1}(\mathbb{R}). Then

(2.16) A(μAf)(ω)=ηA¯(ω)Aμ(ω)Af(ω).\mathscr{F}_{A}(\mu\star_{A}f)(\omega)=\overline{\eta_{A}}(\omega)\mathscr{F}_{A}\mu(\omega)\,\mathscr{F}_{A}f(\omega).
Proof.

The proof is straightforward using the operator ρA\rho_{A}.

In particular, if μ=gdt,\mu=g\,dt, for some gL1()g\in L^{1}(\mathbb{R}) then

A(gAf)(ω)=ηA¯(ω)Ag(ω)Af(ω).\mathscr{F}_{A}(g\star_{A}f)(\omega)=\overline{\eta_{A}}(\omega)\mathscr{F}_{A}g(\omega)\,\mathscr{F}_{A}f(\omega).

See [6] for more details.

The concept of chirp modulation was used in [6] to define a chirp convolution and to get sampling theorems. Although the AA-translation is somehow included implicitly in the definition of the chirp convolution, it has not been used to its full extent in [6] and hence the harmonic analysis of the special affine Fourier transform has not been developed. However we want to make use of our new translation from harmonic analysis point of view. Towards this end we first we prove an analogue of Wendel’s theorem for the SAFT.

We are now in a position to state one of our main results. The following statement is an analogue of Wendel’s theorem in the context of the SAFT.

Theorem 2.6 (Wendel).

Let T:L1()L1()T:L^{1}(\mathbb{R})\to L^{1}(\mathbb{R}) be a bounded linear operator. Then the following statements are equivalent.

  • (i)

    TTsA=TsAT,TT_{s}^{A}=T_{s}^{A}T, for all ss\in\mathbb{R}.

  • (ii)

    T(fAg)=TfAg=fATgT(f\star_{A}g)=Tf\star_{A}g=f\star_{A}Tg, for all f,gL1().f,g\in L^{1}(\mathbb{R}).

  • (iii)

    There exists a unique μM()\mu\in M(\mathbb{R}) such that Tf=μAfTf=\mu\star_{A}f.

  • (iv)

    There exists a unique μM()\mu\in M(\mathbb{R}) such that

    A(Tf)(ω)=ηA¯(ω)Aμ(ω)Af(ω).\mathscr{F}_{A}(Tf)(\omega)=\overline{\eta_{A}}(\omega)\mathscr{F}_{A}\mu(\omega)\mathscr{F}_{A}f(\omega).
  • (v)

    There exists a unique ϕL()\phi\in L^{\infty}(\mathbb{R}) such that A(Tf)(ω)=ϕ(ω)Af(ω).\mathscr{F}_{A}(Tf)(\omega)=\phi(\omega)\mathscr{F}_{A}f(\omega).

Proof.

Let EρAf(t)=ρA(t)f(t)E_{\rho_{A}}f(t)=\rho_{A}(t)f(t) and define T~:L1()L1()\tilde{T}:L^{1}(\mathbb{R})\to L^{1}(\mathbb{R}) by T~=EρATEρA¯\tilde{T}=E_{\rho_{A}}\,T\,E_{\overline{\rho_{A}}}. Then using (2.9) we get

TTxA=TEρA¯EρATxA=\displaystyle T\,T^{A}_{x}=T\,E_{\overline{\rho_{A}}}E_{\rho_{A}}T_{x}^{A}= ρA(x)TEρA¯TxEρA=ρA(x)EρA¯T~TxEρA,\displaystyle\rho_{A}(x)\,T\,E_{\overline{\rho_{A}}}\,T_{x}\,E_{\rho_{A}}=\rho_{A}(x)\,E_{\overline{\rho_{A}}}\,\tilde{T}\,T_{x}\,E_{\rho_{A}},

which shows that TTxA=TxATT\,T^{A}_{x}=T^{A}_{x}\,T iff T~Tx=TxT~\tilde{T}\,T_{x}=T_{x}\,\tilde{T}. Similarly we can show that T(fAg)=TfAgT(f\star_{A}g)=Tf\star_{A}g iff T~(fg)=T~fg\tilde{T}(f\star g)=\tilde{T}f\star g, and Tf=μAfTf=\mu\star_{A}\,f iff T~f=1|b|(EρAμ)f\tilde{T}\,f=\frac{1}{\sqrt{|b|}}(E_{\rho_{A}}\mu)\star f. The equivalence of (i),(ii), and (iii) now follows from Wendel’s theorem in the classical case. That (iii) implies (iv) and (iv) implies (v) is obvious. To show that (i) follows from (v), let fL1()f\in L^{1}(\mathbb{R}). For ss\in\mathbb{R} we have

A(TTsAf)(ω)\displaystyle\mathscr{F}_{A}(T\,T^{A}_{s}\,f)(\omega) =ϕ(ω)A(TsAf)(ω)\displaystyle=\phi(\omega)\,\mathscr{F}_{A}(T^{A}_{s}\,f)(\omega)
=ϕ(ω)ρA(ω)eisω/bAf(ω)\displaystyle=\phi(\omega)\,\rho_{A}(\omega)\,e^{-i\,s\omega/b}\,\mathscr{F}_{A}f(\omega)
=ρA(ω)eisω/bA(Tf)(ω)=A(TsATf)(ω).\displaystyle=\rho_{A}(\omega)\,e^{-i\,s\omega/b}\,\mathscr{F}_{A}(T\,f)(\omega)=\mathscr{F}_{A}(T^{A}_{s}\,T\,f)(\omega).

We end this section by establishing an analogue of the Poisson summation formula for the SAFT and the corresponding AA-translation. From now on, we use the following notation.

I:=[|b|π,|b|π].I:=[-|b|\pi,|b|\pi].
Theorem 2.7.

Let fL1()L2()f\in L^{1}(\mathbb{R})\cap L^{2}(\mathbb{R}). Then the following formula holds.

2π|b|kT2kbπAf(t)e2iabk2π2=nηA¯(n+p)Af(n+p)ei2b(at22nt),t.\sqrt{2\pi|b|}\sum_{k\in\mathbb{Z}}T_{-2kb\pi}^{A}f(t)e^{-2iabk^{2}\pi^{2}}=\sum_{n\in\mathbb{Z}}\overline{\eta_{A}}(n+p)\mathscr{F}_{A}f(n+p)e^{-\frac{i}{2b}(at^{2}-2nt)},\quad t\in\mathbb{R}.

We refer to [6] for the proof.

3. Closed ideals in (L1(),A)(L^{1}(\mathbb{R}),\star_{A})

We have seen that

Cab(fAg)=1|b|(CabfCabg).C_{\frac{a}{b}}(f\star_{A}g)=\frac{1}{\sqrt{|b|}}(C_{\frac{a}{b}}f\star C_{\frac{a}{b}}g).

Thus it is easy to see that gAf1(2π|b|)1/2g1f1\|g\star_{A}f\|_{1}\leq(2\pi|b|)^{-1/2}\|g\|_{1}\,\|f\|_{1}, from which it follows that (L1(),A)(L^{1}(\mathbb{R}),\star_{A}) is a commutative Banach algebra. In this section, we aim to study the closed ideals in (L1(),A)(L^{1}(\mathbb{R}),\star_{A}). Towards this end, first we show that (L1(),A)\big{(}L^{1}(\mathbb{R}),\star_{A}\big{)} possesses a bounded approximate identity as in the classical case (L1(),)(L^{1}(\mathbb{R}),\star).

Theorem 3.1.

The space (L1(),A)(L^{1}(\mathbb{R}),\star_{A}) possesses a bounded approximate identity.

Proof.

Let {gα}\{g_{\alpha}\} be a bounded approximate identity in (L1(),)(L^{1}(\mathbb{R}),\star) and define uα=|b|ρA¯gαu_{\alpha}=\sqrt{|b|}\,\overline{\rho_{A}}\,g_{\alpha}. Then using (2.15), we get

fAuαf1=ρA(fAuα)ρAf1\displaystyle\|f\star_{A}u_{\alpha}-f\|_{1}=\|\rho_{A}(f\star_{A}u_{\alpha})-\rho_{A}f\|_{1} =1|b|((ρAf)(ρAuα))f1\displaystyle=\Big{\|}{\textstyle\frac{1}{\sqrt{|b|}}}\big{(}(\rho_{A}f)\star(\rho_{A}u_{\alpha})\big{)}-f\Big{\|}_{1}
=(ρAf)gαρAf10\displaystyle=\|(\rho_{A}f)\star g_{\alpha}-\rho_{A}f\|_{1}\to 0

as α\alpha\to\infty. Hence {uα}\{u_{\alpha}\} is a bounded approximate identity for (L1(),A)(L^{1}(\mathbb{R}),\star_{A}).

Now, we aim to study AA-translation invariant closed ideals in (L1(),A)(L^{1}(\mathbb{R}),\star_{A}). First we prove the following theorem in this context.

Theorem 3.2.

Let JJ be a closed subspace of L1()L^{1}(\mathbb{R}). Then JJ is an ideal in (L1(),A)(L^{1}(\mathbb{R}),\star_{A}) if and only if it is invariant under AA-translations.

Proof.

Let JJ be an ideal in (L1(),A)(L^{1}(\mathbb{R}),\star_{A}). Let {uα}\{u_{\alpha}\} be an approximate identity in (L1(),A)(L^{1}(\mathbb{R}),\star_{A}). Then for fJf\in J and xx\in\mathbb{R}, we have

TxAf=limαTxA(uαAf)=limαTxAuαAf,T_{x}^{A}f=\lim_{\alpha\to\infty}T_{x}^{A}(u_{\alpha}\star_{A}f)=\lim_{\alpha\to\infty}T_{x}^{A}u_{\alpha}\star_{A}f,

using Theorem 2.6. Since TxAuαAfJT_{x}^{A}u_{\alpha}\star_{A}f\in J, for all α\alpha and JJ is closed, TxAfJT_{x}^{A}f\in J.
Conversely, assume that JJ is invariant under AA-translations. Let fJ,gL1()f\in J,~{}g\in L^{1}(\mathbb{R}). Then viewing fAgf\star_{A}g as a Bochner integral as in (2.13), we can conclude that fAgJf\star_{A}g\in J. ∎

Proposition 3.3.

The collection {gL1():A(g) has compact support}\{g\in L^{1}(\mathbb{R}):\mathscr{F}_{A}(g)\textit{ has compact support}\} is dense in L1()L^{1}(\mathbb{R}).

Proof.

We know that {gL1():g^ has compact support}\{g\in L^{1}(\mathbb{R}):\widehat{g}\textit{ has compact support}\} is dense in L1()L^{1}(\mathbb{R}). Thus, for fL1()f\in L^{1}(\mathbb{R}), there exists gL1()g\in L^{1}(\mathbb{R}) such that g^\widehat{g} has compact support and Cabfg<ϵ\|C_{\frac{a}{b}}f-g\|<\epsilon. This implies that

fCabg=CabfCabCabg<ϵ.\|f-C_{-\frac{a}{b}}g\|=\|C_{\frac{a}{b}}f-C_{\frac{a}{b}}C_{-\frac{a}{b}}g\|<\epsilon.

Further, A(Cabg)\mathscr{F}_{A}(C_{-\frac{a}{b}}g) has compact support as g^\widehat{g} has compact support and

A(Cabg)(ω)=ηA(ω)|b|g^(ωpb),\mathscr{F}_{A}(C_{-\frac{a}{b}}g)(\omega)=\frac{\eta_{A}(\omega)}{\sqrt{|b|}}\widehat{g}(\frac{\omega-p}{b}),

which completes the proof. ∎

Lemma 3.4 (Lemma 4.59 in [12]).

Let fL1()f\in L^{1}(\mathbb{R}) and ω0\omega_{0}\in\mathbb{R}. Then for every ϵ>0\epsilon>0, there exists hL1()h\in L^{1}(\mathbb{R}) with h1<ϵ\|h\|_{1}<\epsilon such that

(f+h)^(ω)=f^(ω0),(f+h)~{}\widehat{}~{}(\omega)=\widehat{f}(\omega_{0}),

for every ω\omega in some neighbourhood of ω0\omega_{0}.

Now we are in a position to state and prove an analogue Weiner’s theorem in connection with the SAFT.

Theorem 3.5.

Let JJ be a closed AA-translation invariant subspace of L1()L^{1}(\mathbb{R}) such that Z(J)=Z(J)=\emptyset, where Z(J):={ω:Af(ω)=0, for all fJ}Z(J):=\{\omega\in\mathbb{R}:\mathscr{F}_{A}f(\omega)=0,~{}\textit{ for all }f\in J\}. Then J=L1()J=L^{1}(\mathbb{R}).

Proof.

In view of Proposition 3.3, it is enough to show that fJf\in J for all fL1()f\in L^{1}(\mathbb{R}) such that Af\mathscr{F}_{A}f has compact support. Let fL1()f\in L^{1}(\mathbb{R}) be such that Af\mathscr{F}_{A}f has compact support. Let K=supp(Af)K=supp(\mathscr{F}_{A}f).
Step 1: In this step we show that for each ω0\omega_{0}\in\mathbb{R}, there exists FJF\in J such that Af(ω)=AF(ω)\mathscr{F}_{A}f(\omega)=\mathscr{F}_{A}F(\omega) in a neighbourhood of ω0\omega_{0}. Since Z(J)=Z(J)=\emptyset, we can choose gJg\in J such that ηA¯(ω0)Ag(ω0)=1\overline{\eta_{A}}(\omega_{0})\mathscr{F}_{A}g(\omega_{0})=1. Then using Lemma 3.4, there exists hL1()h\in L^{1}(\mathbb{R}) with h1<(2π|b|)1/22\|h\|_{1}<\frac{(2\pi|b|)^{1/2}}{2} and

(Cabg)^(ωpb)+(Cabh)^(ωpb)=(Cabg)^(ω0pb),(C_{\frac{a}{b}}g)~{}\widehat{}~{}(\frac{\omega-p}{b})+(C_{\frac{a}{b}}h)~{}\widehat{}~{}(\frac{\omega-p}{b})=(C_{\frac{a}{b}}g)~{}\widehat{}~{}(\frac{\omega_{0}-p}{b}),

in a neighbourhood of ω0\omega_{0}. This implies that

(3.1) Ag(ω)+Ah(ω)=ηA(ω)ηA(ω0)Ag(ω0)=ηA(ω),\mathscr{F}_{A}g(\omega)+\mathscr{F}_{A}h(\omega)=\frac{\eta_{A}(\omega)}{\eta_{A}(\omega_{0})}\mathscr{F}_{A}g(\omega_{0})=\eta_{A}(\omega),

in a neighbourhood of ω0\omega_{0}. Let hn=hAhAAhh_{n}=h\star_{A}h\star_{A}\cdots\star_{A}h (nn-times). Then using the fact that (L1(),A)(L^{1}(\mathbb{R}),\star_{A}) is a Banach algebra, we can show that the series f+nfAhnf+\sum_{n\in\mathbb{N}}f\star_{A}h_{n} converges in L1()L^{1}(\mathbb{R}). Let k=f+nfAhnk=f+\sum_{n\in\mathbb{N}}f\star_{A}h_{n}. Then using convolution theorem, we obtain

Ak(ω)\displaystyle\mathscr{F}_{A}k(\omega) =Af(ω)+nηA¯(ω)nAf(ω)(Ah(ω))n\displaystyle=\mathscr{F}_{A}f(\omega)+\sum_{n\in\mathbb{N}}\overline{\eta_{A}}(\omega)^{n}\mathscr{F}_{A}f(\omega)\big{(}\mathscr{F}_{A}h(\omega)\big{)}^{n}
=Af(ω)11ηA¯(ω)Ah(ω)=Af(ω)ηA¯(ω)Ag(ω),\displaystyle=\mathscr{F}_{A}f(\omega)\frac{1}{1-\overline{\eta_{A}}(\omega)\mathscr{F}_{A}h(\omega)}=\frac{\mathscr{F}_{A}f(\omega)}{\overline{\eta_{A}}(\omega)\mathscr{F}_{A}g(\omega)},

in a neighborhood of ω0\omega_{0}, using (3.1). The second equality in the above equation follows from the fact that |Ah(ω)|1(2π|b|)1/2h1<12|\mathscr{F}_{A}h(\omega)|\leq\frac{1}{(2\pi|b|)^{1/2}}\|h\|_{1}<\frac{1}{2}. Thus

Af(ω)=ηA¯(ω)Ag(ω)Ak(ω)=A(gAk)(ω),\mathscr{F}_{A}f(\omega)=\overline{\eta_{A}}(\omega)\mathscr{F}_{A}g(\omega)\mathscr{F}_{A}k(\omega)=\mathscr{F}_{A}(g\star_{A}k)(\omega),

in a neighbourhood of ω0\omega_{0}. As gAkJg\star_{A}k\in J, by Theorem 3.2, our claim is established.
Step 2: In this step we show that fJf\in J. Appealing to Step 1, for each ω\omega\in\mathbb{R}, choose gωJg_{\omega}\in J such that Af=Agω\mathscr{F}_{A}f=\mathscr{F}_{A}g_{\omega} on a neighborhood UωU_{\omega} of ω\omega. Using compactness of KK, we get U1,U2,UnU_{1},\,U_{2},\,\cdots U_{n} and g1,g2,gnJg_{1},g_{2},\cdots g_{n}\in J such that Ki=1nUiK\subset\cup_{i=1}^{n}U_{i} and Af=Agi\mathscr{F}_{A}f=\mathscr{F}_{A}g_{i} on UiU_{i}. Now choose open WωW_{\omega} such that

{ω}WωWω¯Ui.\{\omega\}\subset W_{\omega}\subset\overline{W_{\omega}}\subset U_{i}.

Again using the compactness of KK, there exist Wω1,Wω2,,WωmW_{\omega_{1}},\,W_{\omega_{2}},\cdots,W_{\omega_{m}} such that Ki=1mWωi¯K\subset\cup_{i=1}^{m}\overline{W_{\omega_{i}}} and Wωi¯Uωi\overline{W_{\omega_{i}}}\subset U_{\omega_{i}}, where ωiUωi{U1,U2,,Un}\omega_{i}\in U_{\omega_{i}}\in\{U_{1},\,U_{2},\cdots,U_{n}\}. Take h1,h2,,hmL1()h_{1},\,h_{2},\cdots,h_{m}\in L^{1}(\mathbb{R}) such that

ηA¯(ω)Ahj(ω)=1 on Wωj¯ and supp(Ahj)Uωj.\overline{\eta_{A}}(\omega)\mathscr{F}_{A}h_{j}(\omega)=1~{}~{}\text{ on }\overline{W_{\omega_{j}}}~{}~{}\text{ and }supp(\mathscr{F}_{A}h_{j})\subset U_{\omega_{j}}.

Then Πj=1m(1ηA¯(ω)Ahj(ω))=0\Pi_{j=1}^{m}(1-\overline{\eta_{A}}(\omega)\mathscr{F}_{A}h_{j}(\omega))=0 on KK. This implies that

Af(ω)=Af(ω)[1Πj=1m(1ηA¯(ω)Ahj(ω))].\mathscr{F}_{A}f(\omega)=\mathscr{F}_{A}f(\omega)[1-\Pi_{j=1}^{m}(1-\overline{\eta_{A}}(\omega)\mathscr{F}_{A}h_{j}(\omega))].

This can be rewritten as f=fAHif=\sum f\star_{A}H_{i}, where HiH_{i} being one of the hjh_{j}’s or their convolutions, and supp(AHi)Uωisupp(\mathscr{F}_{A}H_{i})\subset U_{\omega_{i}}, for some ii. But

A(fAHi)(ω)=ηA¯(ω)Af(ω)AHi(ω)\displaystyle\mathscr{F}_{A}(f\star_{A}H_{i})(\omega)=\overline{\eta_{A}}(\omega)\mathscr{F}_{A}f(\omega)\mathscr{F}_{A}H_{i}(\omega) =ηA¯(ω)Agi(ω)AHi(ω)\displaystyle=\overline{\eta_{A}}(\omega)\mathscr{F}_{A}g_{i}(\omega)\mathscr{F}_{A}H_{i}(\omega)
=A(giAHi)(ω).\displaystyle=\mathscr{F}_{A}(g_{i}\star_{A}H_{i})(\omega).

As giAHiJg_{i}\star_{A}H_{i}\in J, fJf\in J. This completes the proof. ∎

Corollary 3.6.

Let fL1()f\in L^{1}(\mathbb{R}). Then the closed linear span of AA-translates of ff is L1()L^{1}(\mathbb{R}) if and only if Af\mathscr{F}_{A}f never vanishes.

Proof.

Let JJ be the closed linear span of AA-translates of ff. Let Af(ω)0\mathscr{F}_{A}f(\omega)\neq 0 for all ω\omega\in\mathbb{R}. Since JJ is a closed AA-translation invariant subspace, appealing to Wiener’s theorem, we get J=L1()J=L^{1}(\mathbb{R}).
Conversely assume that J=L1()J=L^{1}(\mathbb{R}). Suppose Af(ω0)=0\mathscr{F}_{A}f(\omega_{0})=0 for some ω0\omega_{0}\in\mathbb{R}. Then Ah(ω0)=0\mathscr{F}_{A}h(\omega_{0})=0 for all hspan{TxAf:x}h\in span\{T_{x}^{A}f:x\in\mathbb{R}\}. Since span{TxAf:x}span\{T_{x}^{A}f:x\in\mathbb{R}\} is dense in L1()L^{1}(\mathbb{R}) and Ah(2π|b|)1/2f1\|\mathscr{F}_{A}h\|_{\infty}\leq(2\pi|b|)^{-1/2}\|f\|_{1}, we can conclude that Ah(ω0)=0\mathscr{F}_{A}h(\omega_{0})=0, for all hL1()h\in L^{1}(\mathbb{R}), which is an impossibility. Hence the result follows. ∎

Now, we shall state the analogue of Wiener’s theorem for L2()L^{2}(\mathbb{R}) functions.

Theorem 3.7.

Let fL2()f\in L^{2}(\mathbb{R}). Then the closed linear span of AA-translates of ff is L2()L^{2}(\mathbb{R}) if and only if Af(ω)0\mathscr{F}_{A}f(\omega)\neq 0 a.e.

Proof.

Let M=span¯{TxAf:x}M=\overline{span}\{T_{x}^{A}f:x\in\mathbb{R}\}. Then gMg\in M^{\perp} if and only if TxAf,g=0\langle T_{x}^{A}f,g\rangle=0, for all xx\in\mathbb{R}. For xx\in\mathbb{R}, consider

TxAf,g=A(TxAf),Ag\displaystyle\langle T_{x}^{A}f,g\rangle=\langle\mathscr{F}_{A}(T_{x}^{A}f),\mathscr{F}_{A}g\rangle =ρA(x)eixbωAf(ω)Ag(ω)¯𝑑ω\displaystyle=\int_{\mathbb{R}}\rho_{A}(x)e^{-i\frac{x}{b}\omega}\mathscr{F}_{A}f(\omega)\overline{\mathscr{F}_{A}g(\omega)}d\omega
=2πρA(x)(AfAg¯)^(xb).\displaystyle=\sqrt{2\pi}\rho_{A}(x)(\mathscr{F}_{A}f\overline{\mathscr{F}_{A}g})~{}\widehat{}~{}(\frac{x}{b}).

Thus TxAf,g=0 for all x\langle T_{x}^{A}f,g\rangle=0~{}\text{ for all }x\in\mathbb{R} is equivalent to

(Af(ω)Ag(ω)¯)^(x)=0, for all x,(\mathscr{F}_{A}f(\omega)\overline{\mathscr{F}_{A}g(\omega)})~{}\widehat{}~{}(x)=0,~{}\text{ for all }x\in\mathbb{R},

which is same as Af(ω)Ag(ω)¯=0, a.e. ω.\mathscr{F}_{A}f(\omega)\overline{\mathscr{F}_{A}g(\omega)}=0,\text{ a.e. }\omega\in\mathbb{R}. This shows that M={0}M^{\perp}=\{0\} if and only if Af(ω)0\mathscr{F}_{A}f(\omega)\neq 0 a.e. ω.\omega\in\mathbb{R}.

We conclude this section by establishing an analogue of Wiener-Tauberian theorem. Recall that MyM_{y} denotes the modulation operator which is defined in (2.6).

Theorem 3.8 (Wiener-Tauberian).

Let ϕL(),\phi\in L^{\infty}(\mathbb{R}), hL1()h\in L^{1}(\mathbb{R}) be such that Ah(ω)0\mathscr{F}_{A}h(\omega)\neq 0, for all ω\omega\in\mathbb{R} and

limx±Cab(hAϕ)(x)=mA(Mpbh)(0).\lim_{x\to\pm\infty}C_{\frac{a}{b}}(h\star_{A}\phi)(x)=m\mathscr{F}_{A}(M_{-\frac{p}{b}}h)(0).

Then

limx±Cab(fAϕ)(x)=mA(Mpbf)(0),\lim_{x\to\pm\infty}C_{\frac{a}{b}}(f\star_{A}\phi)(x)=m\mathscr{F}_{A}(M_{-\frac{p}{b}}f)(0),

for all fL1()f\in L^{1}(\mathbb{R}).

Proof.

Since, 0Ah(ω)=ηA(ω)|b|(Cabh)^(ωpb)0\neq\mathscr{F}_{A}h(\omega)=\frac{\eta_{A}(\omega)}{\sqrt{|b|}}(C_{\frac{a}{b}}h)~{}\widehat{}~{}(\frac{\omega-p}{b}), for all ω\omega\in\mathbb{R}, (Cabh)^(ω)0(C_{\frac{a}{b}}h)~{}\widehat{}~{}(\omega)\neq 0, for all ω\omega\in\mathbb{R}. Further,

Cab(hAϕ)(x)=1|b|(CabhCabϕ)(x).C_{\frac{a}{b}}(h\star_{A}\phi)(x)=\frac{1}{\sqrt{|b|}}(C_{\frac{a}{b}}h\star C_{\frac{a}{b}}\phi)(x).

Furthermore,

Af(0)=1|b|(Cabf)^(pb)=1|b|Tpb(Cabf)^(0)\displaystyle\mathscr{F}_{A}f(0)=\frac{1}{\sqrt{|b|}}(C_{\frac{a}{b}}f)~{}\widehat{}~{}(-\frac{p}{b})=\frac{1}{\sqrt{|b|}}T_{\frac{p}{b}}(C_{\frac{a}{b}}f)~{}\widehat{}~{}(0) =1|b|(MpbCabf)^(0)\displaystyle=\frac{1}{\sqrt{|b|}}(M_{\frac{p}{b}}C_{\frac{a}{b}}f)~{}\widehat{}~{}(0)
=1|b|(CabMpbf)^(0),\displaystyle=\frac{1}{\sqrt{|b|}}(C_{\frac{a}{b}}M_{\frac{p}{b}}f)~{}\widehat{}~{}(0),

using CsMy=MyCsC_{s}M_{y}=M_{y}C_{s}. Thus

(3.2) A(Mpbf)(0)=1|b|(Cabf)(0).\mathscr{F}_{A}(M_{-\frac{p}{b}f})(0)=\frac{1}{\sqrt{|b|}}(C_{\frac{a}{b}}f)(0).

Hence

limx±(CabhCabϕ)(x)=m(Cabh)^(0),\lim_{x\to\pm\infty}(C_{\frac{a}{b}}h\star C_{\frac{a}{b}}\phi)(x)=m(C_{\frac{a}{b}}h)~{}\widehat{}~{}(0),

using given hypothesis. Now using the classical Wiener-Tauberian theorem, we get

limx±(fCabϕ)(x)=mf^(0), for all fL1().\lim_{x\to\pm\infty}(f\star C_{\frac{a}{b}}\phi)(x)=m\widehat{f}(0),~{}~{}\text{ for all }f\in L^{1}(\mathbb{R}).

This implies that

limx±(CabfCabϕ)(x)=m(Cabf)^(0), for all fL1().\lim_{x\to\pm\infty}(C_{\frac{a}{b}}f\star C_{\frac{a}{b}}\phi)(x)=m(C_{\frac{a}{b}}f)~{}\widehat{}~{}(0),~{}~{}\text{ for all }f\in L^{1}(\mathbb{R}).

This in turn implies that

limx±Cab(fAϕ)(x)=mA(Mpbf)(0), for all fL1(),\lim_{x\to\pm\infty}C_{\frac{a}{b}}(f\star_{A}\phi)(x)=m\mathscr{F}_{A}(M_{-\frac{p}{b}}f)(0),~{}~{}\text{ for all }f\in L^{1}(\mathbb{R}),

using (3.2). ∎

As a special case, we state the following analogue of the Wiener-Tauberian theorem associated with the fractional Fourier transform.

Corollary 3.9.

Let ϕL()\phi\in L^{\infty}(\mathbb{R}), hL1()h\in L^{1}(\mathbb{R}) be such that θh(ω)0\mathscr{F}_{\theta}h(\omega)\neq 0, for all ω\omega\in\mathbb{R} and

limx±Cθ(hθϕ)(x)=mθh(0).\lim_{x\to\pm\infty}C_{\theta}(h\star_{\theta}\phi)(x)=m\mathscr{F}_{\theta}h(0).

Then

limx±Cθ(fθϕ)(x)=mθf(0),\lim_{x\to\pm\infty}C_{\theta}(f\star_{\theta}\phi)(x)=m\mathscr{F}_{\theta}f(0),

for all fL1()f\in L^{1}(\mathbb{R}).

4. AA-shift invariant spaces

In this section, we aim to study shift invariant spaces associated with the SAFT, called AA-shift invariant spaces, in detail. Recall that the AA-shift invariant space is defined by VA(ϕ)=span¯{TkAϕ:k}V_{A}(\phi)=\overline{span}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} for ϕL2()\phi\in L^{2}(\mathbb{R}). First, we obtain the following result whose proof is similar to that of Theorem 2 in [2] in the classical case.

Theorem 4.1.

Let ϕW(C,1())\phi\in W\big{(}C,\ell^{1}(\mathbb{Z})\big{)} be such that {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} forms a Riesz basis for VA(ϕ)V_{A}(\phi). Then

  • (i)

    VA(ϕ)W(C,2())V_{A}(\phi)\subseteq W\big{(}C,\ell^{2}(\mathbb{Z})\big{)}.

  • (ii)

    If 𝒳={xk:k}\mathcal{X}=\{x_{k}:k\in\mathbb{Z}\} is separated, then there is M>0M>0 such that

    (k|f(xk)|2)1/2Mf,fVA(ϕ).\big{(}\sum_{k\in\mathbb{Z}}|f(x_{k})|^{2}\big{)}^{1/2}\leq M\|f\|,\quad\forall f\in V_{A}(\phi).
Corollary 4.2.

Let ϕW(C,1())\phi\in W(C,\ell^{1}(\mathbb{Z})). Then VA(ϕ)V_{A}(\phi) is a reproducing kernel Hilbert space with the reproducing kernel

K(x,y)=keiabk(xy)ϕ(xk)¯S1ϕ(yk),K(x,y)=\sum_{k\in\mathbb{Z}}e^{i\frac{a}{b}k(x-y)}\overline{\phi(x-k)}S^{-1}\phi(y-k),

where SS is the frame operator for {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\}.

Proof.

Let xx\in\mathbb{R} be fixed. Then, taking 𝒳={x+k:k}\mathcal{X}=\{x+k:k\in\mathbb{Z}\} in the previous theorem we get M>0M>0 such that |f(x)|Mf|f(x)|\leq M\|f\|, for every fVA(ϕ)f\in V_{A}(\phi), which shows that VA(ϕ)V_{A}(\phi) is a RKHS. The reproducing kernel for VA(ϕ)V_{A}(\phi) is K(x,y)=kTkAϕ(x)¯TkAS1ϕ(y)K(x,y)=\sum_{k\in\mathbb{Z}}\overline{T_{k}^{A}\phi(x)}T_{k}^{A}S^{-1}\phi(y). Now, using the definition of TkAT_{k}^{A} our assertion follows. ∎

Theorem 4.3.

If {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is a frame sequence, for ϕL2()\phi\in L^{2}(\mathbb{R}), then the members of its canonical dual frame also are AA-translates of a single function.

Proof.

Let SS be the frame operator for {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\}. First we prove that STkA=TkAS,k.ST_{k}^{A}=T_{k}^{A}S,~{}\forall~{}k\in\mathbb{Z}. Let fVA(ϕ)f\in V_{A}(\phi). Then for kk\in\mathbb{Z}, we have

STkAf\displaystyle ST_{k}^{A}f =kTkAf,TkAϕTkAϕ\displaystyle=\sum_{k^{\prime}\in\mathbb{Z}}\langle T_{k}^{A}f,T_{k^{\prime}}^{A}\phi\rangle T_{k^{\prime}}^{A}\phi
=kf,(TkA)TkAϕTkAϕ\displaystyle=\sum_{k^{\prime}\in\mathbb{Z}}\langle f,(T_{k}^{A})^{*}T_{k^{\prime}}^{A}\phi\rangle T_{k^{\prime}}^{A}\phi
=kf,eiabk2TkATkAϕTkAϕ\displaystyle=\sum_{k^{\prime}\in\mathbb{Z}}\langle f,e^{-i\frac{a}{b}k^{2}}T_{-k}^{A}T_{k^{\prime}}^{A}\phi\rangle T_{k^{\prime}}^{A}\phi
=kf,TkkAϕeiabk(kk)TkAϕ\displaystyle=\sum_{k^{\prime}\in\mathbb{Z}}\langle f,T_{k^{\prime}-k}^{A}\phi\rangle e^{-i\frac{a}{b}k(k^{\prime}-k)}T_{k^{\prime}}^{A}\phi
=f,TAϕTkATAϕ\displaystyle=\sum_{\ell\in\mathbb{Z}}\langle f,T_{\ell}^{A}\phi\rangle T_{k}^{A}T_{\ell}^{A}\phi
=TkASf.\displaystyle=T_{k}^{A}Sf.

Since SS is invertible, TkAf=S1TkASfT_{k}^{A}f=S^{-1}T_{k}^{A}Sf and we have for every hVA(ϕ),h\in V_{A}(\phi),

TkAS1h=S1TkASS1h=S1TkAh.T_{k}^{A}S^{-1}h=S^{-1}T_{k}^{A}SS^{-1}h=S^{-1}T_{k}^{A}h.

Thus, if {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is a frame for VA(ϕ),V_{A}(\phi), then the canonical dual frame {S1TkAϕ:k}\{S^{-1}T_{k}^{A}\phi:k\in\mathbb{Z}\} is given by {TkAS1ϕ:k}.\{T_{k}^{A}S^{-1}\phi:k\in\mathbb{Z}\}. Taking ψ=S1ϕ\psi=S^{-1}\phi, we conclude that the canonical dual frame is also of the form {TkAψ:k}.\{T_{k}^{A}\psi:k\in\mathbb{Z}\}.

Remark 4.4.

If we assume {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is a Riesz basis for VA(ϕ)V_{A}(\phi), then {TkAS1ϕ:k}\{T_{k}^{A}S^{-1}\phi:k\in\mathbb{Z}\} is the dual Riesz basis of {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\}.

Now we obtain a characterization for the system of AA-translates {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} to be a frame sequence in terms of the weight function wϕ(ω):=k|Aϕ(ω+2kbπ)|2w_{\phi}(\omega):=\sum_{k\in\mathbb{Z}}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}.

Theorem 4.5.

Let ϕL2()\phi\in L^{2}(\mathbb{R}). Then the system of AA-translates {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is a frame sequence with bounds m,M>0m,M>0 if and only if

(4.1) m2π|b|χEϕk|Aϕ(ω+2kbπ)|2M2π|b|χEϕ,ωI,\frac{m}{2\pi|b|}\chi_{E_{\phi}}\leq\sum_{k\in\mathbb{Z}}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}\leq\frac{M}{2\pi|b|}\chi_{E_{\phi}},\,~{}~{}\omega\in I,

where Eϕ={ω:wϕ(ω)0}E_{\phi}=\{\omega\in\mathbb{R}:w_{\phi}(\omega)\neq 0\} and I=[|b|π,|b|π]I=[-|b|\pi,|b|\pi].

Proof.

Let {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} be a frame sequence with bounds m,M>0m,M>0. Then

(4.2) mf2k|f,TkAϕ|2Mf2, for all fVA(ϕ).m\|f\|^{2}\leq\sum_{k\in\mathbb{Z}}|\langle f,T_{k}^{A}\phi\rangle|^{2}\leq M\|f\|^{2},~{}~{}\text{ for all }f\in V_{A}(\phi).

Let FF be a finite subset of \mathbb{Z}. Let f=kFckTkAϕVA(ϕ)f=\sum_{k\in F}c_{k}T_{k}^{A}\phi\in V_{A}(\phi). Then Af(ω)=r(ω)Aϕ(ω)\mathscr{F}_{A}f(\omega)=r(\omega)\mathscr{F}_{A}\phi(\omega), where r(ω)=kFckρA(k)eibkωr(\omega)=\sum_{k\in F}c_{k}\rho_{A}(k)e^{-\frac{i}{b}k\omega}. Thus

(4.3) f2=f,f=Af,Af=|r(ω)|2|Aϕ|2𝑑ω=I|r(ω)|2wϕ(ω)𝑑ω.\displaystyle\|f\|^{2}=\langle f,f\rangle=\langle\mathscr{F}_{A}f,\mathscr{F}_{A}f\rangle=\int_{\mathbb{R}}|r(\omega)|^{2}|\mathscr{F}_{A}\phi|^{2}d\omega=\int_{I}|r(\omega)|^{2}w_{\phi}(\omega)d\omega.

Similarly,

f,TkAϕ=Af(ω)A(TkAϕ)(ω)¯𝑑ω\displaystyle\langle f,T_{k}^{A}\phi\rangle=\int_{\mathbb{R}}\mathscr{F}_{A}f(\omega)\overline{\mathscr{F}_{A}(T_{k}^{A}\phi)(\omega)}d\omega =r(ω)|Aϕ(ω)|2ρA(k)¯eibkω𝑑ω\displaystyle=\int_{\mathbb{R}}r(\omega)|\mathscr{F}_{A}\phi(\omega)|^{2}\overline{\rho_{A}(k)}e^{\frac{i}{b}k\omega}d\omega
=Ir(ω)ρA(k)¯wϕ(ω)eibkω𝑑ω\displaystyle=\int_{I}r(\omega)\overline{\rho_{A}(k)}w_{\phi}(\omega)e^{\frac{i}{b}k\omega}d\omega
=2π|b|(r(ω)wϕ(ω))^(k),\displaystyle=\sqrt{2\pi|b|}\big{(}r(\omega)w_{\phi}(\omega)\big{)}~{}\widehat{}~{}(k),

where f^(k)=If(x)ρA(k)¯eibkx𝑑x\widehat{f}(k)=\int_{I}f(x)\overline{\rho_{A}(k)}e^{\frac{i}{b}kx}dx. Hence

(4.4) k|f,TkAϕ|2=2π|b|k|(r(ω)wϕ(ω))^(k)|2=2π|b|I|r(ω)wϕ(ω)|2𝑑ω.\sum_{k\in\mathbb{Z}}|\langle f,T_{k}^{A}\phi\rangle|^{2}=2\pi|b|\sum_{k\in\mathbb{Z}}|\big{(}r(\omega)w_{\phi}(\omega)\big{)}~{}\widehat{}~{}(k)|^{2}=2\pi|b|\int_{I}|r(\omega)w_{\phi}(\omega)|^{2}d\omega.

Now, using (4.2), we get

mI|r(ω)|2wϕ(ω)𝑑ω2π|b|I|r(ω)|2(wϕ(ω))2𝑑ωMI|r(ω)|2wϕ(ω)𝑑ω,m\int_{I}|r(\omega)|^{2}w_{\phi}(\omega)d\omega\leq 2\pi|b|\int_{I}|r(\omega)|^{2}\big{(}w_{\phi}(\omega)\big{)}^{2}d\omega\leq M\int_{I}|r(\omega)|^{2}w_{\phi}(\omega)d\omega,

for every |b||b|-periodic trigonometric polynomial rr. This implies that

m2π|b|wϕ(ω)wϕ(ω)2M2π|b|wϕ(ω),\frac{m}{2\pi|b|}w_{\phi}(\omega)\leq w_{\phi}(\omega)^{2}\leq\frac{M}{2\pi|b|}w_{\phi}(\omega),

a. e. ωI\omega\in I, from which (4.1) follows.
Conversely, assume that (4.1) holds. Then using (4.3), (4.4), we can show that (4.2) holds for all fspan{TkAϕ:k}f\in span\{T_{k}^{A}\phi:k\in\mathbb{Z}\}. Since span{TkAϕ:k}span\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is dense in VA(ϕ)V_{A}(\phi), the proof follows. ∎

In a similar way, we can obtain the characterizations for the system {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} to be a Riesz sequence or an orthonormal system. We state the results without proof. The interested readers can see the proof form [16] and from [6].

Theorem 4.6.

Let ϕL2()\phi\in L^{2}(\mathbb{R}). Then the collection {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is a Riesz basis for VA(ϕ)V_{A}(\phi) if and only if there are m,M>0m,M>0 such that

mk|Aϕ(ω+2kbπ)|2M,m\leq\sum_{k\in\mathbb{Z}}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}\leq M,

for almost all ωI.\omega\in I.

Theorem 4.7.

Let ϕL2()\phi\in L^{2}(\mathbb{R}). Then the collection {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is an orthonormal system in L2()L^{2}(\mathbb{R}) if and only if

(4.5) k|Aϕ(ω+2kbπ)|2=12π|b|,\sum_{k\in\mathbb{Z}}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}=\frac{1}{2\pi|b|},

for almost all ωI.\omega\in I.

5. Sampling in AA-shift invariant spaces

In order to get an equivalent condition for the stable set of sampling in terms of the Zak transform, we first introduce AA-Zak transform.

Definition 5.1.

The AA-Zak transform 𝒵Af\mathscr{Z}_{A}f of a function fL2()f\in L^{2}(\mathbb{R}) is a function on 2\mathbb{R}^{2}, defined as

𝒵Af(t,ω)=η¯A(ω)2π|b|kTkAf(t)ei2b(ak22kω+2pk),t,ω.\mathscr{Z}_{A}f(t,\omega)=\frac{\overline{\eta}_{A}(\omega)}{\sqrt{2\pi|b|}}\sum_{k\in\mathbb{Z}}T_{k}^{A}f(t)e^{-\frac{i}{2b}(ak^{2}-2k\omega+2pk)},~{}t,\omega\in\mathbb{R}.

One can simplify the right hand side and get

𝒵Af(t,ω)=ηA¯(ω)2π|b|kf(tk)ei2b(ak22akt+2kω2pk),fort,ω.\mathscr{Z}_{A}f(t,\omega)=\frac{\overline{\eta_{A}}(\omega)}{\sqrt{2\pi|b|}}\sum_{k\in\mathbb{Z}}f(t-k)e^{\frac{i}{2b}(ak^{2}-2akt+2k\omega-2pk)},\quad\text{for}~{}t,\omega\in\mathbb{R}.
Remark 5.2.

In particular, if we take A={0,1,1,0,0,0}A=\{0,1,-1,0,0,0\}, then AA-Zak transform reduces to the classical Zak transform.

Theorem 5.3.

The AA-Zak transform is an isometry between the spaces L2()L^{2}(\mathbb{R}) and L2([0,1]×I)L^{2}\big{(}[0,1]\times I\big{)}.

See [6] for the proof.

Define an operator T:L2(I)VA(ϕ)T:L^{2}(I)\to V_{A}(\phi) by

TF(x)=kF,EkTkAϕ(x),FL2(I),TF(x)=\sum_{k\in\mathbb{Z}}\langle F,E_{k}\rangle T_{-k}^{A}\phi(x),\quad F\in L^{2}(I),

where Ek(t)=ηA(t)2π|b|ei2b(ak22pk+2kt)E_{k}(t)=\frac{\eta_{A}(t)}{\sqrt{2\pi|b|}}e^{\frac{i}{2b}(ak^{2}-2pk+2kt)}. Clearly {Ek:k}\{E_{k}:k\in\mathbb{Z}\} is an orthonormal basis for L2(I).L^{2}(I).

Suppose {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is a Riesz sequence. Then there are constants m,M>0m,M>0 such that

m(k|F,Ek|2)1/2kF,EkTkAϕL2()M(k|F,Ek|2)1/2,m\big{(}\sum_{k\in\mathbb{Z}}|\langle F,E_{k}\rangle|^{2}\big{)}^{1/2}\leq\|\sum_{k\in\mathbb{Z}}\langle F,E_{k}\rangle T_{-k}^{A}\phi\|_{L^{2}(\mathbb{R})}\leq M\big{(}\sum_{k\in\mathbb{Z}}|\langle F,E_{k}\rangle|^{2}\big{)}^{1/2},

for all FL2(I)F\in L^{2}(I). Since {Ek:k}\{E_{k}:k\in\mathbb{Z}\} is an orthonormal basis for L2(I)L^{2}(I) and FL2(I)F\in L^{2}(I), the above inequality reduces to

mFL2(I)TFL2()MFL2(I),FL2(I).m\|F\|_{L^{2}(I)}\leq\|TF\|_{L^{2}(\mathbb{R})}\leq M\|F\|_{L^{2}(I)},\quad\forall~{}F\in L^{2}(I).

This shows that TT is bounded above and bounded below. By Riesz-Fischer theorem TT is onto. Hence TT is invertible. Moreover, we have

TF(x)\displaystyle TF(x) =kF,EkTkAϕ(x)=F,ηA()2π|b|kei2b(ak22pk+2k)TkAϕ(x)¯\displaystyle=\sum_{k\in\mathbb{Z}}\langle F,E_{k}\rangle T_{-k}^{A}\phi(x)=\big{\langle}F,\frac{\eta_{A}(\cdot)}{\sqrt{2\pi|b|}}\sum_{k\in\mathbb{Z}}e^{\frac{i}{2b}(ak^{2}-2pk+2k\cdot)}\overline{T_{-k}^{A}\phi(x)}\big{\rangle}
(5.1) =F,𝒵Aϕ(x,)¯.\displaystyle=\langle F,\overline{\mathscr{Z}_{A}\phi(x,\cdot)}\rangle.

Now, we are in a position to prove equivalent conditions for stable set of sampling for a shift invariant space VA(ϕ).V_{A}(\phi).

Theorem 5.4.

Assume that VA(ϕ)V_{A}(\phi) is a reproducing kernel Hilbert space, for ϕL2()\phi\in L^{2}(\mathbb{R}), such that {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} forms a Riesz basis for VA(ϕ)V_{A}(\phi). Then the following statements are equivalent.

  • (i)

    The set 𝒳={xj:j}\mathcal{X}=\{x_{j}:j\in\mathbb{Z}\} is a stable set of sampling for VA(ϕ).V_{A}(\phi).

  • (ii)

    There are constants m,M>0m,M>0 such that

    md2()2Ud2()2Md2()2,d2(),m\|d\|_{\ell^{2}(\mathbb{Z})}^{2}\leq\|Ud\|_{\ell^{2}(\mathbb{Z})}^{2}\leq M\|d\|_{\ell^{2}(\mathbb{Z})}^{2},\quad\forall~{}d\in\ell^{2}(\mathbb{Z}),

    where the operator U=(Uj,k)U=(U_{j,k}) is defined by

    Uj,k=TkAϕ(xj)=eiabk(xjk)ϕ(xjk).U_{j,k}=T_{k}^{A}\phi(x_{j})=e^{-i\frac{a}{b}k(x_{j}-k)}\phi(x_{j}-k).
  • (iii)

    The set of reproducing kernels {Kxj:j}\{K_{x_{j}}:j\in\mathbb{Z}\} for VA(ϕ)V_{A}(\phi) is a frame for VA(ϕ)V_{A}(\phi).

  • (iv)

    The set {𝒵Aϕ(xj,):j}\{\mathscr{Z}_{A}\phi(x_{j},\cdot):j\in\mathbb{Z}\} is a frame for L2(I)L^{2}(I).

Proof.

The proof is similar to the classical case. However for the sake of completion, we give the outline of the proof. (i) \Leftrightarrow (ii): If fVA(ϕ)f\in V_{A}(\phi) then there is d={dk}2()d=\{d_{k}\}\in\ell^{2}(\mathbb{Z}) such that f(x)=kdkTkAϕ(x)f(x)=\sum_{k\in\mathbb{Z}}d_{k}T_{k}^{A}\phi(x), and hence f(xj)=kdkTkAϕ(xj)=(Ud)jf(x_{j})=\sum_{k\in\mathbb{Z}}d_{k}T_{k}^{A}\phi(x_{j})=(Ud)_{j}. Since {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is a Riesz basis for VA(ϕ)V_{A}(\phi), the following statements are equivalent.

  • (a)

    There are constants m,M>0m,M>0 such that for every fVA(ϕ)f\in V_{A}(\phi)

    mf2j|f(xj)|2Mf2.m\|f\|^{2}\leq\sum_{j\in\mathbb{Z}}|f(x_{j})|^{2}\leq M\|f\|^{2}.
  • (b)

    There are constants m,M>0m^{\prime},M^{\prime}>0 such that for d2()d\in\ell^{2}(\mathbb{Z})

    md2Ud2Md2.m^{\prime}\|d\|^{2}\leq\|Ud\|^{2}\leq M^{\prime}\|d\|^{2}.

In fact, if (a) holds, then

mf2Ud2Mf2.m\|f\|^{2}\leq\|Ud\|^{2}\leq M\|f\|^{2}.

Now by using the fact that {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is a Riesz sequence, one can find m′′,M′′>0m^{\prime\prime},M^{\prime\prime}>0 such that m′′d2f2M′′d2m^{\prime\prime}\|d\|^{2}\leq\|f\|^{2}\leq M^{\prime\prime}\|d\|^{2}. Thus (b) follows from (a). Similarly one can prove (b) implies (a).
(i)\Leftrightarrow (iii): Let {Kxj:j}\{K_{x_{j}}:j\in\mathbb{Z}\} be the set of reproducing kernels for VA(ϕ)V_{A}(\phi). Then the equivalence follows from the identity f(xj)=f,Kxjf(x_{j})=\langle f,K_{x_{j}}\rangle.
(iii)\Leftrightarrow (iv): Using (5.1) we obtain

j|f(xj)|2=j|f,Kxj|2=j|F,𝒵Aϕ(xj,)¯|2,\sum_{j\in\mathbb{Z}}|f(x_{j})|^{2}=\sum_{j\in\mathbb{Z}}|\langle f,K_{x_{j}}\rangle|^{2}=\sum_{j\in\mathbb{Z}}|\langle F,\overline{\mathscr{Z}_{A}\phi(x_{j},\cdot)}\rangle|^{2},

here TF=fTF=f. Since TT is invertible, we get the equivalence of (iii) and (iv). ∎

Let ϕW(C,1())\phi\in W(C,\ell^{1}(\mathbb{Z})). Then we define the function ϕA\phi_{A}^{\dagger} on the interval II, by

ϕA(ω)=nϕ(n)ρA(n)eibnω.\phi_{A}^{\dagger}(\omega)=\sum_{n\in\mathbb{Z}}\phi(n)\rho_{A}(n)e^{-\frac{i}{b}n\omega}.

From the definition of the AA-Zak transform we obtain 𝒵Aϕ(0,ω)=ηA¯(ω)2π|b|ϕA(ω)\mathscr{Z}_{A}\phi(0,\omega)=\frac{\overline{\eta_{A}}(\omega)}{\sqrt{2\pi|b|}}\phi_{A}^{\dagger}(\omega).

Theorem 5.5.

Let ϕW(C,1())\phi\in W\big{(}C,\ell^{1}(\mathbb{Z})\big{)}. Then the operator U:2()2()U:\ell^{2}(\mathbb{Z})\to\ell^{2}(\mathbb{Z}) defined by Uj,k=TkAϕ(j)U_{j,k}=T_{k}^{A}\phi(j) satisfies the inequalities

ϕA02d2Ud2ϕA2d2,d2(),\|\phi_{A}^{\dagger}\|_{0}^{2}~{}\|d\|^{2}\leq\|Ud\|^{2}\leq\|\phi_{A}^{\dagger}\|_{\infty}^{2}~{}\|d\|^{2},\quad\forall~{}d\in\ell^{2}(\mathbb{Z}),

where ϕA0=infxI|ϕA(x)|,ϕA=supxI|ϕA(x)|.\|\phi_{A}^{\dagger}\|_{0}=\inf_{x\in I}|\phi_{A}^{\dagger}(x)|,~{}\|\phi_{A}^{\dagger}\|_{\infty}=\sup_{x\in I}|\phi_{A}^{\dagger}(x)|.

Proof.

Let d={dn}2()d=\{d_{n}\}\in\ell^{2}(\mathbb{Z}). Then

(Ud)n=mUn,mdm=meibam(nm)ϕ(nm)dm.(Ud)_{n}=\sum_{m\in\mathbb{Z}}U_{n,m}d_{m}=\sum_{m\in\mathbb{Z}}e^{-\frac{i}{b}am(n-m)}\phi(n-m)d_{m}.

Since {12π|b|ρA(n)eibnω:n}\{\frac{1}{\sqrt{2\pi|b|}}\rho_{A}(n)e^{-\frac{i}{b}n\omega}:n\in\mathbb{Z}\} is an orthonormal basis for L2(I)L^{2}(I), we have

2π|b|Ud2\displaystyle 2\pi|b|~{}\|Ud\|^{2} =I|n(Ud)nρA(n)eibnω|2𝑑ω\displaystyle=\int_{I}|\sum_{n\in\mathbb{Z}}(Ud)_{n}\rho_{A}(n)e^{-\frac{i}{b}n\omega}|^{2}d\omega
=I|nmeibam(nm)ϕ(nm)dmρA(n)eibnω|2𝑑ω\displaystyle=\int_{I}|\sum_{n\in\mathbb{Z}}\sum_{m\in\mathbb{Z}}e^{-\frac{i}{b}am(n-m)}\phi(n-m)d_{m}\rho_{A}(n)e^{-\frac{i}{b}n\omega}|^{2}d\omega
=I|nmρA(nm)ϕ(nm)dmρA(m)eibnω|2𝑑ω\displaystyle=\int_{I}|\sum_{n\in\mathbb{Z}}\sum_{m\in\mathbb{Z}}\rho_{A}(n-m)\phi(n-m)d_{m}\rho_{A}(m)e^{-\frac{i}{b}n\omega}|^{2}d\omega
=I|nmρA(n)ϕ(n)dmρA(m)eib(m+n)ω|2𝑑ω\displaystyle=\int_{I}|\sum_{n\in\mathbb{Z}}\sum_{m\in\mathbb{Z}}\rho_{A}(n)\phi(n)d_{m}\rho_{A}(m)e^{-\frac{i}{b}(m+n)\omega}|^{2}d\omega
=I|ϕA(ω)|2|mdmρA(m)eibmω|2𝑑ω.\displaystyle=\int_{I}|\phi_{A}^{\dagger}(\omega)|^{2}|\sum_{m\in\mathbb{Z}}d_{m}\rho_{A}(m)e^{-\frac{i}{b}m\omega}|^{2}d\omega.

This implies

ϕA022π|b|I|mdmρA(m)eibmω|2𝑑ω\displaystyle\frac{\|\phi_{A}^{\dagger}\|_{0}^{2}}{2\pi|b|}\int_{I}|\sum_{m\in\mathbb{Z}}d_{m}\rho_{A}(m)e^{-\frac{i}{b}m\omega}|^{2}d\omega Ud2\displaystyle\leq\|Ud\|^{2}
ϕA22π|b|I|mdmρA(m)eibmω|2𝑑ω\displaystyle\leq\frac{\|\phi_{A}^{\dagger}\|^{2}_{\infty}}{2\pi|b|}\int_{I}|\sum_{m\in\mathbb{Z}}d_{m}\rho_{A}(m)e^{-\frac{i}{b}m\omega}|^{2}d\omega

or equivalently ϕA02m|dm|2Ud2ϕA2m|dm|2\|\phi_{A}^{\dagger}\|_{0}^{2}\sum_{m\in\mathbb{Z}}|d_{m}|^{2}\leq\|Ud\|^{2}\leq\|\phi_{A}^{\dagger}\|_{\infty}^{2}\sum_{m\in\mathbb{Z}}|d_{m}|^{2}, from which the result follows. ∎

As a consequence we obtain the following

Corollary 5.6.

Let ϕW(C,1())\phi\in W(C,\ell^{1}(\mathbb{Z})) be such that {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} forms a Riesz basis for VA(ϕ)V_{A}(\phi) and ϕA(x)0\phi_{A}^{\dagger}(x)\neq 0 for all xIx\in I. Then \mathbb{Z} is a stable set of sampling for VA(ϕ).V_{A}(\phi).

Proof.

Since ϕA(x)0\phi_{A}^{\dagger}(x)\neq 0 for all xIx\in I, it follows from Theorem 5.5 that UU is bounded above and below. Then the assertion follows from Theorem 5.4. ∎

We end this section by proving Bernstein type inequality for VA(ϕ)V_{A}(\phi). Let 𝒜\mathcal{A} denote the class of continuously differentiable functions ϕ\phi such that

  • (i)

    |ϕ(x)|M1|x|0.5+ϵ|\phi(x)|\leq\frac{M_{1}}{|x|^{0.5+\epsilon}} and |ϕ(x)|M2|x|0.5+ϵ|\phi^{\prime}(x)|\leq\frac{M_{2}}{|x|^{0.5+\epsilon}}, for sufficiently large xx, for some M1,M2,ϵ>0M_{1},M_{2},\epsilon>0.

  • (ii)

    ess supωIk(ω+2kbπpb)2|Aϕ(ω+2kbπ)|2<\text{ess sup}_{\omega\in I}\sum_{k\in\mathbb{Z}}(\frac{\omega+2kb\pi-p}{b})^{2}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}<\infty.

Theorem 5.7.

Let ϕ𝒜\phi\in\mathcal{A} be such that {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is a Riesz basis for VA(ϕ)V_{A}(\phi). Then we have the Bernstein type inequality

Bf2Mf2, for all fVA(ϕ),\|Bf\|^{2}\leq M\|f\|^{2},~{}~{}~{}\text{ for all }f\in V_{A}(\phi),

where Bf(x)=f(x)+iaxbf(x)Bf(x)=f^{\prime}(x)+\frac{iax}{b}f(x) and

M=ess supωIk(ω+2kbπpb)2|Aϕ(ω+2kbπ)|2k|Aϕ(ω+2kbπ)|2.M=\text{ess sup}_{\omega\in I}\frac{\sum_{k\in\mathbb{Z}}(\frac{\omega+2kb\pi-p}{b})^{2}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}}{\sum_{k\in\mathbb{Z}}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}}.
Proof.

Let f(x)=kckTkAϕ(x)f(x)=\sum_{k\in\mathbb{Z}}c_{k}T_{k}^{A}\phi(x).
Then f(x)=kckTkAϕ(x)kiakbckTkAϕ(x)f^{\prime}(x)=\sum_{k\in\mathbb{Z}}c_{k}T_{k}^{A}\phi^{\prime}(x)-\sum_{k\in\mathbb{Z}}\frac{iak}{b}c_{k}T_{k}^{A}\phi(x). Since ϕ𝒜\phi\in\mathcal{A}, the above equalities hold pointwise. Thus

Bf(x)\displaystyle Bf(x) =f(x)+iaxbf(x)\displaystyle=f^{\prime}(x)+\frac{iax}{b}f(x)
=kckTkAϕ(x)kiakbckTkAϕ(x)+iaxbkckTkAϕ(x)\displaystyle=\sum_{k\in\mathbb{Z}}c_{k}T_{k}^{A}\phi^{\prime}(x)-\sum_{k\in\mathbb{Z}}\frac{iak}{b}c_{k}T_{k}^{A}\phi(x)+\frac{iax}{b}\sum_{k\in\mathbb{Z}}c_{k}T_{k}^{A}\phi(x)
=kck(ddx(TkAϕ(x))+iaxbTkAϕ(x))\displaystyle=\sum_{k\in\mathbb{Z}}c_{k}\big{(}\frac{d}{dx}(T_{k}^{A}\phi(x))+\frac{iax}{b}T_{k}^{A}\phi(x)\big{)}
=kckBTkAϕ(x).\displaystyle=\sum_{k\in\mathbb{Z}}c_{k}BT_{k}^{A}\phi(x).

Further, Af(ω)=r(ω)Aϕ(ω)\mathscr{F}_{A}f(\omega)=r(\omega)\mathscr{F}_{A}\phi(\omega), where r(ω)=kckρA(k)eibkωr(\omega)=\sum_{k\in\mathbb{Z}}c_{k}\rho_{A}(k)e^{-\frac{i}{b}k\omega}. Now, using A(Bf)(ω)=iωpbAf(ω)\mathscr{F}_{A}(Bf)(\omega)=i\frac{\omega-p}{b}\mathscr{F}_{A}f(\omega) (see Proposition 3.5 in [7]), we obtain

Bf2=A(Bf)2\displaystyle\|Bf\|^{2}=\|\mathscr{F}_{A}(Bf)\|^{2} =A(kckB(TkAϕ))2\displaystyle=\|\mathscr{F}_{A}(\sum_{k\in\mathbb{Z}}c_{k}B(T_{k}^{A}\phi))\|^{2}
=|kckA(BTkAϕ)(ω)|2𝑑ω\displaystyle=\int_{\mathbb{R}}|\sum_{k\in\mathbb{Z}}c_{k}\mathscr{F}_{A}(BT_{k}^{A}\phi)(\omega)|^{2}d\omega
=|kcki(ωp)bA(TkAϕ)(ω)|2𝑑ω\displaystyle=\int_{\mathbb{R}}|\sum_{k\in\mathbb{Z}}c_{k}\frac{i(\omega-p)}{b}\mathscr{F}_{A}(T_{k}^{A}\phi)(\omega)|^{2}d\omega
=|kckρA(k)eibkωωpbAϕ(ω)|2𝑑ω\displaystyle=\int_{\mathbb{R}}|\sum_{k\in\mathbb{Z}}c_{k}\rho_{A}(k)e^{-\frac{i}{b}k\omega}\frac{\omega-p}{b}\mathscr{F}_{A}\phi(\omega)|^{2}d\omega
=|r(ω)|2(ωpb)2|Aϕ(ω)|2𝑑ω\displaystyle=\int_{\mathbb{R}}|r(\omega)|^{2}(\frac{\omega-p}{b})^{2}|\mathscr{F}_{A}\phi(\omega)|^{2}d\omega
=I|r(ω)|2k(ω+2kbπpb)2|Aϕ(ω+2kbπ)|2dω\displaystyle=\int_{I}|r(\omega)|^{2}\sum_{k\in\mathbb{Z}}(\frac{\omega+2kb\pi-p}{b})^{2}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}d\omega
MI|r(ω)|2k|Aϕ(ω+2kbπ)|2dω\displaystyle\leq M\int_{I}|r(\omega)|^{2}\sum_{k\in\mathbb{Z}}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}d\omega
=M|r(ω)|2|Aϕ(ω)|2𝑑ω=Mf2,\displaystyle=M\int_{\mathbb{R}}|r(\omega)|^{2}|\mathscr{F}_{A}\phi(\omega)|^{2}d\omega=M\|f\|^{2},

proving our assertion. ∎

6. Sampling theorems

In this section, our aim is to obtain reconstruction formulae for the functions belonging to certain VA(ϕ)V_{A}(\phi) from integer samples. We prove sampling formulae with L2L^{2} convergence as well as uniform convergence. As a corollary, we obtain the result proved in [6], namely Shannon sampling theorem for the functions which are bandlimited in the SAFT domain.

Theorem 6.1.

Let ϕW(C,1())\phi\in W\big{(}C,\ell^{1}(\mathbb{Z})\big{)} be such that {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} forms a Riesz basis for VA(ϕ)V_{A}(\phi). Then there is a function SVA(ϕ)S\in V_{A}(\phi) such that

(6.1) f(t)=nf(n)TnAS(t),f(t)=\sum_{n\in\mathbb{Z}}f(n)T_{n}^{A}S(t),

for all fVA(ϕ)f\in V_{A}(\phi) if and only if 1/ϕAL2(I).1/\phi_{A}^{\dagger}\in L^{2}(I).

Proof.

Assume that there is a SVA(ϕ)S\in V_{A}(\phi) such that (6.1) holds. Then ϕ(t)=nϕ(n)TnAS(t)\phi(t)=\sum_{n\in\mathbb{Z}}\phi(n)T_{n}^{A}S(t). Taking SAFT on both sides we obtain

(6.2) Aϕ(ω)=nϕ(n)A(TnAS)(ω)=ϕA(ω)AS(ω).\mathscr{F}_{A}\phi(\omega)=\sum_{n\in\mathbb{Z}}\phi(n)\mathscr{F}_{A}(T_{n}^{A}S)(\omega)=\phi_{A}^{\dagger}(\omega)\mathscr{F}_{A}S(\omega).

This implies that supp(Aϕ)supp(ϕA)supp(\mathscr{F}_{A}\phi)\subseteq supp(\phi_{A}^{\dagger}). Since ϕA\phi_{A}^{\dagger} is 2πb2\pi b periodic, we have supp(T2kbπAϕ)supp(ϕA),k.supp(T_{-2kb\pi}\mathscr{F}_{A}\phi)\subseteq supp(\phi_{A}^{\dagger}),~{}\forall~{}k\in\mathbb{Z}.

Since ϕ0\phi\neq 0, we have Aϕ0\mathscr{F}_{A}\phi\neq 0 and hence supp(Aϕ)supp(\mathscr{F}_{A}\phi) is not a set of measure zero. We shall show that ksupp(T2kbπAϕ)=.\bigcup_{k\in\mathbb{Z}}supp(T_{-2kb\pi}\mathscr{F}_{A}\phi)=\mathbb{R}. If not, then there exists a Δ()\Delta(\subseteq\mathbb{R}), a set of positive measure such that Δksupp(T2kbπAϕ)\Delta\subset\mathbb{R}\setminus\bigcup_{k\in\mathbb{Z}}supp(T_{-2kb\pi}\mathscr{F}_{A}\phi) and Aϕ(ω+2kbπ)=0,ωΔ,k.\mathscr{F}_{A}\phi(\omega+2kb\pi)=0,~{}\omega\in\Delta,~{}\forall~{}k. This in turn implies that k|Aϕ(ω+2kbπ)|2=0\sum_{k\in\mathbb{Z}}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}=0 on Δ\Delta. This is a contradiction to our assumption that {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is a Riesz sequence. So ϕA(ω)0\phi_{A}^{\dagger}(\omega)\neq 0 for almost every ω.\omega\in\mathbb{R}. Using (6.2), we get AS(ω)=Aϕ(ω)/ϕA(ω)\mathscr{F}_{A}S(\omega)=\mathscr{F}_{A}\phi(\omega)/\phi_{A}^{\dagger}(\omega) and

|S(t)|2𝑑t=|AS(ω)|2𝑑ω\displaystyle\int_{\mathbb{R}}|S(t)|^{2}dt=\int_{\mathbb{R}}|\mathscr{F}_{A}S(\omega)|^{2}d\omega =|Aϕ(ω)ϕA(ω)|2𝑑ω\displaystyle=\int_{\mathbb{R}}|\frac{\mathscr{F}_{A}\phi(\omega)}{\phi_{A}^{\dagger}(\omega)}|^{2}d\omega
=Ik|Aϕ(ω+2kbπ)|2|ϕA(ω)|2𝑑ω\displaystyle=\int_{I}\frac{\sum_{k\in\mathbb{Z}}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}}{|\phi_{A}^{\dagger}(\omega)|^{2}}d\omega
GϕA0I1|ϕA(ω)|2𝑑ω,\displaystyle\geq\|G_{\phi}^{A}\|_{0}\int_{I}\frac{1}{|\phi_{A}^{\dagger}(\omega)|^{2}}d\omega,

where GϕA0=infωIk|Aϕ(ω+2kbπ)|2\|G_{\phi}^{A}\|_{0}=\inf_{\omega\in I}\sum_{k\in\mathbb{Z}}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}. Since {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} forms a Riesz sequence, GϕA0>0\|G_{\phi}^{A}\|_{0}>0. Consequently 1/ϕAL2(I).1/\phi_{A}^{\dagger}\in L^{2}(I).

Conversely, assume that 1/ϕAL2(I)1/\phi_{A}^{\dagger}\in L^{2}(I). Since {12π|b|ρA(n)eibnω:n}\{\frac{1}{\sqrt{2\pi|b|}}\rho_{A}(n)e^{-\frac{i}{b}n\omega}:n\in\mathbb{Z}\} forms an orthonormal basis for L2(I)L^{2}(I), there is a sequence {cn}2()\{c_{n}\}\in\ell^{2}(\mathbb{Z}) such that

1ϕA(ω)=ncnρA(n)eibnω.\frac{1}{\phi_{A}^{\dagger}(\omega)}=\sum_{n\in\mathbb{Z}}c_{n}\rho_{A}(n)e^{-\frac{i}{b}n\omega}.

Let F(ω)=Aϕ(ω)/ϕA(ω).F(\omega)=\mathscr{F}_{A}\phi(\omega)/\phi_{A}^{\dagger}(\omega). Then

|F(ω)|2𝑑ω\displaystyle\int_{\mathbb{R}}|F(\omega)|^{2}d\omega =|Aϕ(ω)ϕA(ω)|2𝑑ω\displaystyle=\int_{\mathbb{R}}|\frac{\mathscr{F}_{A}\phi(\omega)}{\phi_{A}^{\dagger}(\omega)}|^{2}d\omega
=Ik|Aϕ(ω+2kbπ)|2|ϕA|2𝑑ω\displaystyle=\int_{I}\frac{\sum_{k\in\mathbb{Z}}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}}{|\phi_{A}^{\dagger}|^{2}}d\omega
GϕA1/ϕAL2(I)2,\displaystyle\leq\|G_{\phi}^{A}\|_{\infty}\|1/\phi_{A}^{\dagger}\|_{L^{2}(I)}^{2},

where GϕA=supωIk|Aϕ(ω+2kbπ)|2\|G_{\phi}^{A}\|_{\infty}=\sup_{\omega\in I}\sum_{k\in\mathbb{Z}}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}. Since {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is a Riesz sequence, GϕA<\|G_{\phi}^{A}\|_{\infty}<\infty. Hence FL2()F\in L^{2}(\mathbb{R}). Then there is exactly one SL2()S\in L^{2}(\mathbb{R}) such that AS(ω)=F(ω)\mathscr{F}_{A}S(\omega)=F(\omega). From the definition of ϕA\phi_{A}^{\dagger} we get

AS(ω)=Aϕ(ω)kckρA(k)eibkω=kckA(TkAϕ)(ω),\mathscr{F}_{A}S(\omega)=\mathscr{F}_{A}\phi(\omega)\sum_{k\in\mathbb{Z}}c_{k}\rho_{A}(k)e^{-\frac{i}{b}k\omega}=\sum_{k\in\mathbb{Z}}c_{k}\mathscr{F}_{A}(T_{k}^{A}\phi)(\omega),

which shows that S=kckTkAϕS=\sum_{k\in\mathbb{Z}}c_{k}T_{k}^{A}\phi and hence SVA(ϕ)S\in V_{A}(\phi). Now let fVA(ϕ)f\in V_{A}(\phi) with representation f=kakTkAϕf=\sum_{k\in\mathbb{Z}}a_{k}T_{k}^{A}\phi. Taking SAFT leads to

Af(ω)\displaystyle\mathscr{F}_{A}f(\omega) =kakA(TkAϕ)(ω)\displaystyle=\sum_{k\in\mathbb{Z}}a_{k}\mathscr{F}_{A}(T_{k}^{A}\phi)(\omega)
=kakρA(k)eibkωAϕ(ω)\displaystyle=\sum_{k\in\mathbb{Z}}a_{k}\rho_{A}(k)e^{-\frac{i}{b}k\omega}\mathscr{F}_{A}\phi(\omega)
=AS(ω)ϕA(ω)kakρA(k)eibkω\displaystyle=\mathscr{F}_{A}S(\omega)\phi_{A}^{\dagger}(\omega)\sum_{k\in\mathbb{Z}}a_{k}\rho_{A}(k)e^{-\frac{i}{b}k\omega}
=AS(ω)ϕ()ρA()eibωkakρA(k)eibkω\displaystyle=\mathscr{F}_{A}S(\omega)\sum_{\ell\in\mathbb{Z}}\phi(\ell)\rho_{A}(\ell)e^{-\frac{i}{b}\ell\omega}\sum_{k\in\mathbb{Z}}a_{k}\rho_{A}(k)e^{-\frac{i}{b}k\omega}
=AS(ω)kakϕ()ρA(k)ρA()eib(+k)ω\displaystyle=\mathscr{F}_{A}S(\omega)\sum_{\ell\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}a_{k}\phi(\ell)\rho_{A}(k)\rho_{A}(\ell)e^{-\frac{i}{b}(\ell+k)\omega}
=AS(ω)nkakϕ(nk)ρA(k)ρA(nk)eibnω\displaystyle=\mathscr{F}_{A}S(\omega)\sum_{n\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}a_{k}\phi(n-k)\rho_{A}(k)\rho_{A}(n-k)e^{-\frac{i}{b}n\omega}
=AS(ω)nkakϕ(nk)eibk(nk)ρA(n)eibnω\displaystyle=\mathscr{F}_{A}S(\omega)\sum_{n\in\mathbb{Z}}\sum_{k\in\mathbb{Z}}a_{k}\phi(n-k)e^{-\frac{i}{b}k(n-k)}\rho_{A}(n)e^{-\frac{i}{b}n\omega}
=AS(ω)n(kakTkAϕ(n))ρA(n)eibnω\displaystyle=\mathscr{F}_{A}S(\omega)\sum_{n\in\mathbb{Z}}\big{(}\sum_{k\in\mathbb{Z}}a_{k}T_{k}^{A}\phi(n)\big{)}\rho_{A}(n)e^{-\frac{i}{b}n\omega}
=nf(n)A(TnAS)(ω).\displaystyle=\sum_{n\in\mathbb{Z}}f(n)\mathscr{F}_{A}(T_{n}^{A}S)(\omega).

The last equality finally gives f=nf(n)TnAS.f=\sum_{n\in\mathbb{Z}}f(n)T_{n}^{A}S.

In Theorem 6.1, we obtained a sampling formula for functions belonging to AA-shift invariant spaces with L2L^{2} convergence. Now, our aim is to obtain another version of a sampling theorem where we obtain both L2L^{2} convergence and pointwise convergence of the corresponding reconstruction formula. Towards this end, we prove the following

Lemma 6.2.

Let ϕL2()\phi\in L^{2}(\mathbb{R}). Then the following statements are equivalent.

  • (i)

    For any {ck}2()\{c_{k}\}\in\ell^{2}(\mathbb{Z}), the series of functions kckTkAϕ(t)\sum_{k\in\mathbb{Z}}c_{k}T_{k}^{A}\phi(t) converges to a continuous function.

  • (ii)

    ϕC()\phi\in C(\mathbb{R}) and suptk|ϕ(tk)|2<\sup_{t\in\mathbb{R}}\sum_{k\in\mathbb{Z}}|\phi(t-k)|^{2}<\infty.

Proof.

Since |TkAϕ(t)|=|Tkϕ(t)|=|ϕ(tk)||T_{k}^{A}\phi(t)|=|T_{k}\phi(t)|=|\phi(t-k)|, the proof follows as in Lemma 1 in [27]. ∎

For a sequence {ck}\{c_{k}\}, we define

A({ck})(ω)=12π|b|kc[k]ei2b(ak2+2pk2ωk).\mathcal{F}_{A}(\{c_{k}\})(\omega)=\frac{1}{\sqrt{2\pi|b|}}\sum_{k\in\mathbb{Z}}c[k]e^{\frac{i}{2b}(ak^{2}+2pk-2\omega k)}.

For two sequences {ck}\{c_{k}\} and {dk}\{d_{k}\}, we define

({ck}A{dk})[n]=12π|b|kc[k]TkAd[n].(\{c_{k}\}*_{A}\{d_{k}\})[n]=\frac{1}{\sqrt{2\pi|b|}}\sum_{k\in\mathbb{Z}}c[k]T_{k}^{A}d[n].

Using the fact that {12π|b|ei2b(ak2+2pk2ωk):k}\{\frac{1}{\sqrt{2\pi|b|}}e^{\frac{i}{2b}(ak^{2}+2pk-2\omega k)}:k\in\mathbb{Z}\}, is an orthonormal basis for L2(I)L^{2}(I), we get

A{ck}2=k|c[k]|2.\|\mathcal{F}_{A}\{c_{k}\}\|^{2}=\sum_{k\in\mathbb{Z}}|c[k]|^{2}.

Further, one can show that

A{{ck}A{dk}}(ω)=A({ck})(ω)A({dk})(ω).\mathcal{F}_{A}\{\{c_{k}\}*_{A}\{d_{k}\}\}(\omega)=\mathcal{F}_{A}(\{c_{k}\})(\omega)\mathcal{F}_{A}(\{d_{k}\})(\omega).

Thus

(6.3) I|A({ck})(ω)|2|A({dk})(ω)|2𝑑ω=12π|b|n|kc[k]TkAd[n]|2\int_{I}|\mathcal{F}_{A}(\{c_{k}\})(\omega)|^{2}|\mathcal{F}_{A}(\{d_{k}\})(\omega)|^{2}d\omega=\frac{1}{\sqrt{2\pi|b|}}\sum_{n\in\mathbb{Z}}|\sum_{k\in\mathbb{Z}}c[k]T_{k}^{A}d[n]|^{2}
Theorem 6.3.

Let {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} be a frame sequence for VA(ϕ)V_{A}(\phi). Then the following are equivalent.

  • (i)

    The series kckTkAϕ(t)\sum_{k\in\mathbb{Z}}c_{k}T_{k}^{A}\phi(t) converges to a continuous function for any {ck}2()\{c_{k}\}\in\ell^{2}(\mathbb{Z}) and there exists a frame {TkAψ:k}\{T_{k}^{A}\psi:k\in\mathbb{Z}\} for VA(ϕ)V_{A}(\phi) such that

    (6.4) f(t)=kf(k)TkAψ(t), for all fVA(ϕ),f(t)=\sum_{k\in\mathbb{Z}}f(k)T_{k}^{A}\psi(t),~{}~{}\text{ for all }f\in V_{A}(\phi),

    where the convergence is both in L2()L^{2}(\mathbb{R}) and uniform on \mathbb{R}.

  • (ii)

    ϕC()\phi\in C(\mathbb{R}), k|ϕ(tk)|2\sum_{k\in\mathbb{Z}}|\phi(t-k)|^{2} is bounded on \mathbb{R} and

    (6.5) mχEϕ(ω)|ϕA(ω)|MχEϕ(ω),m\chi_{E_{\phi}}(\omega)\leq|\phi_{A}^{\dagger}(\omega)|\leq M\chi_{E_{\phi}}(\omega),

    for some m,M>0m,M>0, where Eϕ:={ω:wϕ(ω)0}E_{\phi}:=\{\omega\in\mathbb{R}:w_{\phi}(\omega)\neq 0\}, wϕ(ω):=k|Aϕ(ω+2kbπ)|2w_{\phi}(\omega):=\sum_{k\in\mathbb{Z}}|\mathscr{F}_{A}\phi(\omega+2kb\pi)|^{2}.

Proof.

In order to show that (i) implies (ii), it is enough to show that (6.5) holds. Taking f=ϕf=\phi in (6.4), we get

ϕ(t)=kϕ(k)TkAψ(t).\phi(t)=\sum_{k\in\mathbb{Z}}\phi(k)T_{k}^{A}\psi(t).

Taking SAFT on both sides we obtain Aϕ(ω)=ϕA(ω)Aψ(ω)\mathscr{F}_{A}\phi(\omega)=\phi_{A}^{\dagger}(\omega)\mathscr{F}_{A}\psi(\omega). This implies that wϕ(ω)=|ϕA(ω)|2wψ(ω)w_{\phi}(\omega)=|\phi^{\dagger}_{A}(\omega)|^{2}w_{\psi}(\omega), from which it follows that EϕEψE_{\phi}\subset E_{\psi}. Since {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} and {TkAψ:k}\{T_{k}^{A}\psi:k\in\mathbb{Z}\} are frame sequences, there exist m,M>0m,M>0 such that

m|ϕA(ω)|M, a.e. ωEϕ,m\leq|\phi_{A}^{\dagger}(\omega)|\leq M,~{}~{}\text{ a.e. }\omega\in E_{\phi},

using Theorem 4.5. We now show that ϕA(ω)=0\phi_{A}^{\dagger}(\omega)=0 a.e. on IEϕI\setminus E_{\phi}. To see this, take c(ω)=1χEϕ(ω).c(\omega)=1-\chi_{E_{\phi}}(\omega). Then c(ω)=kckρA(k)eibkωc(\omega)=\sum_{k\in\mathbb{Z}}c_{k}\rho_{A}(k)e^{-\frac{i}{b}k\omega}, for some {ck}2()\{c_{k}\}\in\ell^{2}(\mathbb{Z}). Since c(ω)Aϕ(ω)=0c(\omega)\mathscr{F}_{A}\phi(\omega)=0, taking inverse SAFT, we obtain kckTkAϕ(t)=0\sum_{k\in\mathbb{Z}}c_{k}T_{k}^{A}\phi(t)=0, for all tt\in\mathbb{R}. In particular,

{{ck}A{ϕ(k)}}[n]=12π|b|kckTkAϕ(n)=0.\{\{c_{k}\}*_{A}\{\phi(k)\}\}[n]=\frac{1}{\sqrt{2\pi|b|}}\sum_{k\in\mathbb{Z}}c_{k}T_{k}^{A}\phi(n)=0.

Thus using (6.3), we get

0=n|kckTkAϕ(n)|2\displaystyle 0=\sum_{n\in\mathbb{Z}}|\sum_{k\in\mathbb{Z}}c_{k}T_{k}^{A}\phi(n)|^{2} =I|A{ck}(ω)|2|A{ϕ(n)}(ω)|2𝑑ω\displaystyle=\int_{I}|\mathcal{F}_{A}\{c_{k}\}(\omega)|^{2}|\mathcal{F}_{A}\{\phi(n)\}(\omega)|^{2}d\omega
=IEϕ|ϕA(ω)|2𝑑ω,\displaystyle=\int_{I\setminus E_{\phi}}|\phi_{A}^{\dagger}(\omega)|^{2}d\omega,

which proves our claim.
Conversely assume that (ii) holds. Let

Aψ(ω)={1ϕA(ω)Aϕ(ω)ωEϕ0,ωEϕ.\mathscr{F}_{A}\psi(\omega)=\begin{cases}\frac{1}{\phi_{A}^{\dagger}(\omega)}\mathscr{F}_{A}\phi(\omega)&\omega\in E_{\phi}\\ 0,&\omega\notin E_{\phi}.\end{cases}

Then {TkAψ:k}\{T_{k}^{A}\psi:k\in\mathbb{Z}\} is a frame sequence by appealing to Theorem 4.5. Since 1ϕAL2(I)\frac{1}{\phi_{A}^{\dagger}}\in L^{2}(I), it can be easily seen that ψVA(ϕ)\psi\in V_{A}(\phi). With the similar reasoning, we can say that ϕVA(ψ)\phi\in V_{A}(\psi). Thus VA(ϕ)=VA(ψ)V_{A}(\phi)=V_{A}(\psi). Now define,

Aψ~(ω)={ϕA(ω)¯wϕ(ω)Aϕ(ω),ωEϕ0,ωEϕ.\mathscr{F}_{A}\tilde{\psi}(\omega)=\begin{cases}\frac{\overline{\phi_{A}^{\dagger}(\omega)}}{w_{\phi}(\omega)}\mathscr{F}_{A}\phi(\omega),&\omega\in E_{\phi}\\ 0,&\omega\notin E_{\phi}.\end{cases}

We show that {TkAψ~:k}\{T_{k}^{A}\tilde{\psi}:k\in\mathbb{Z}\} is the canonical dual of {TkAψ:k}\{T_{k}^{A}\psi:k\in\mathbb{Z}\}. Let SS be the frame operator associated with the frame {TkAψ:k}\{T_{k}^{A}\psi:k\in\mathbb{Z}\}. Since SS commutes with TkAT_{k}^{A}, for all kk, it is enough to show that Sψ~=ψS\tilde{\psi}=\psi. Consider

A(Sψ~)(ω)\displaystyle\mathscr{F}_{A}(S\tilde{\psi})(\omega) =kψ~,TkAψA(TkAψ)(ω)\displaystyle=\sum_{k\in\mathbb{Z}}\langle\tilde{\psi},T_{k}^{A}\psi\rangle\mathscr{F}_{A}(T_{k}^{A}\psi)(\omega)
=kρA(k)eibkωAψ(ω)Aψ~,A(TkAψ)\displaystyle=\sum_{k\in\mathbb{Z}}\rho_{A}(k)e^{-\frac{i}{b}k\omega}\mathscr{F}_{A}\psi(\omega)\langle\mathscr{F}_{A}\tilde{\psi},\mathscr{F}_{A}(T_{k}^{A}\psi)\rangle
=kρA(k)eibkωAψ(ω)Aψ~(η)A(TkAψ)(η)¯𝑑η\displaystyle=\sum_{k\in\mathbb{Z}}\rho_{A}(k)e^{-\frac{i}{b}k\omega}\mathscr{F}_{A}\psi(\omega)\int_{\mathbb{R}}\mathscr{F}_{A}\tilde{\psi}(\eta)\overline{\mathscr{F}_{A}(T_{k}^{A}\psi)(\eta)}d\eta
=kρA(k)eibkωAψ(ω)Aψ~(η)ρA(k)¯eibkηAψ(η)¯𝑑η\displaystyle=\sum_{k\in\mathbb{Z}}\rho_{A}(k)e^{-\frac{i}{b}k\omega}\mathscr{F}_{A}\psi(\omega)\int_{\mathbb{R}}\mathscr{F}_{A}\tilde{\psi}(\eta)\overline{\rho_{A}(k)}e^{\frac{i}{b}k\eta}\overline{\mathscr{F}_{A}\psi(\eta)}d\eta
=12π|b|keibkωAψ(ω)Eϕeibkη|Aϕ(η)|2wϕ(η)𝑑η\displaystyle=\frac{1}{2\pi|b|}\sum_{k\in\mathbb{Z}}e^{-\frac{i}{b}k\omega}\mathscr{F}_{A}\psi(\omega)\int_{E_{\phi}}e^{\frac{i}{b}k\eta}\frac{|\mathscr{F}_{A}\phi(\eta)|^{2}}{w_{\phi}(\eta)}d\eta
=Aψ(ω)12π|b|keibkωIχEϕ(η)eibkη𝑑η\displaystyle=\mathscr{F}_{A}\psi(\omega)\frac{1}{2\pi|b|}\sum_{k\in\mathbb{Z}}e^{-\frac{i}{b}k\omega}\int_{I}\chi_{E_{\phi}}(\eta)e^{\frac{i}{b}k\eta}d\eta
=Aψ(ω)χEϕ(ω)=Aψ(ω),\displaystyle=\mathscr{F}_{A}\psi(\omega)\chi_{E_{\phi}}(\omega)=\mathscr{F}_{A}\psi(\omega),

which proves our claim. Let fVA(ϕ)f\in V_{A}(\phi). Then Af(ω)=r(ω)Aϕ(ω)\mathscr{F}_{A}f(\omega)=r(\omega)\mathscr{F}_{A}\phi(\omega), where r(ω)=kckρA(k)eibkωr(\omega)=\sum_{k\in\mathbb{Z}}c_{k}\rho_{A}(k)e^{-\frac{i}{b}k\omega}, for some {ck}2()\{c_{k}\}\in\ell^{2}(\mathbb{Z}). For kk\in\mathbb{Z}, consider

f,TkAψ~\displaystyle\langle f,T_{k}^{A}\tilde{\psi}\rangle =Af(ω)A(TkAψ~)(ω)¯𝑑ω\displaystyle=\int_{\mathbb{R}}\mathscr{F}_{A}f(\omega)\overline{\mathscr{F}_{A}(T_{k}^{A}\tilde{\psi})(\omega)}d\omega
=12π|b|Eϕr(ω)|Aϕ(ω)|2ϕA(ω)wϕ(ω)ρA(k)¯eibkω𝑑ω\displaystyle=\frac{1}{2\pi|b|}\int_{E_{\phi}}r(\omega)|\mathscr{F}_{A}\phi(\omega)|^{2}\frac{\phi_{A}^{\dagger}(\omega)}{w_{\phi}(\omega)}\overline{\rho_{A}(k)}e^{\frac{i}{b}k\omega}d\omega
=12π|b|Ir(ω)ϕA(ω)ρA(k)¯eibkω𝑑ω\displaystyle=\frac{1}{2\pi|b|}\int_{I}r(\omega)\phi_{A}^{\dagger}(\omega)\overline{\rho_{A}(k)}e^{\frac{i}{b}k\omega}d\omega
=IA{ck}(ω)A{ϕ(n)}(ω)ρA(k)¯eibkω𝑑ω\displaystyle=\int_{I}\mathcal{F}_{A}\{c_{k}\}(\omega)\mathcal{F}_{A}\{\phi(n)\}(\omega)\overline{\rho_{A}(k)}e^{\frac{i}{b}k\omega}d\omega
=IA{{ck}A{ϕ(n)}}(ω)ρA(k)¯eibkω𝑑ω\displaystyle=\int_{I}\mathcal{F}_{A}\{\{c_{k}\}*_{A}\{\phi(n)\}\}(\omega)\overline{\rho_{A}(k)}e^{\frac{i}{b}k\omega}d\omega
=2π|b|{{ck}A{ϕ(n)}}=f(k).\displaystyle=\sqrt{2\pi|b|}\{\{c_{k}\}*_{A}\{\phi(n)\}\}=f(k).

Hence

f(t)=kf,TkAψ~TkAψ(t)=kf(k)TkA(ψ)(t).f(t)=\sum_{k\in\mathbb{Z}}\langle f,T_{k}^{A}\tilde{\psi}\rangle T_{k}^{A}\psi(t)=\sum_{k\in\mathbb{Z}}f(k)T_{k}^{A}(\psi)(t).

We notice that {f(k)}2()\{f(k)\}\in\ell^{2}(\mathbb{Z}). Thus, in order to show the uniform convergence of the above series, it is enough to show that

k|TkAψ(t)|2=k|ψ(tk)|2M<, for all t,\sum_{k\in\mathbb{Z}}|T_{k}^{A}\psi(t)|^{2}=\sum_{k\in\mathbb{Z}}|\psi(t-k)|^{2}\leq M<\infty,~{}~{}\text{ for all }t\in\mathbb{R},

for some M>0M>0. Since ψVA(ϕ)\psi\in V_{A}(\phi), Aψ(ω)=r(ω)Aϕ(ω)\mathscr{F}_{A}\psi(\omega)=r(\omega)\mathscr{F}_{A}\phi(\omega), where r(ω)=kckρA(k)eibkωr(\omega)=\sum_{k\in\mathbb{Z}}c_{k}\rho_{A}(k)e^{-\frac{i}{b}k\omega}, for some {ck}2()\{c_{k}\}\in\ell^{2}(\mathbb{Z}). Thus wψ(ω)=|r(ω)|2wϕ(ω)w_{\psi}(\omega)=|r(\omega)|^{2}w_{\phi}(\omega). Since {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} and {TkAψ:k}\{T_{k}^{A}\psi:k\in\mathbb{Z}\} are frames for VA(ϕ)V_{A}(\phi), rr is bounded on EϕE_{\phi}. This implies that r~(ω)=r(ω)χEϕ(ω)\tilde{r}(\omega)=r(\omega)\chi_{E_{\phi}}(\omega) is bounded on II. Let r~(ω)=kck~ρA(k)eibkω\tilde{r}(\omega)=\sum_{k\in\mathbb{Z}}\tilde{c_{k}}\rho_{A}(k)e^{-\frac{i}{b}k\omega}, for some {ck}2()\{c_{k}\}\in\ell^{2}(\mathbb{Z}). Since r(ω)Aϕ(ω)=r~(ω)Aϕ(ω)r(\omega)\mathscr{F}_{A}\phi(\omega)=\tilde{r}(\omega)\mathscr{F}_{A}\phi(\omega), ψ(t)=kck~TkAϕ(t)\psi(t)=\sum_{k\in\mathbb{Z}}\tilde{c_{k}}T_{k}^{A}\phi(t). Thus

n|ψ(tn)|2\displaystyle\sum_{n\in\mathbb{Z}}|\psi(t-n)|^{2} =n|kck~TkAϕ(tn)|2\displaystyle=\sum_{n\in\mathbb{Z}}|\sum_{k\in\mathbb{Z}}\tilde{c_{k}}T_{k}^{A}\phi(t-n)|^{2}
=12π|b|I|r~(ω)|2|nϕ(tn)ρA(n)eibnω|2𝑑ω\displaystyle=\frac{1}{2\pi|b|}\int_{I}|\tilde{r}(\omega)|^{2}|\sum_{n\in\mathbb{Z}}\phi(t-n)\rho_{A}(n)e^{-\frac{i}{b}n\omega}|^{2}d\omega
r~(ω)n|ϕ(tn)|2,\displaystyle\leq\|\tilde{r}(\omega)\|_{\infty}\sum_{n\in\mathbb{Z}}|\phi(t-n)|^{2},

proving our assertion. ∎

As a consequence of Theorem 6.1, we obtain Shannon sampling theorem for the SAFT domain by taking ϕ=sinc,\phi=sinc, where sinc(x)={sinπxπx,ifx01,ifx=0.sinc(x)=\begin{cases}\frac{sin\pi x}{\pi x},\quad&\text{if}\,x\neq 0\\ 1,\quad&\text{if}\,x=0\\ \end{cases}. We also write down the sampling theorem when ϕ\phi is taken to be the second order symmetric BB-spline.

Corollary 6.4.

Let ϕ=sinc\phi=sinc and ψ=Cabϕ\psi=C_{-\frac{a}{b}}\phi. Then for every fVA(ψ)f\in V_{A}(\psi), we have the following representation

(6.6) f(t)=kf(k)eia2b(k2t2)sinc(tk),t.f(t)=\sum_{k\in\mathbb{Z}}f(k)e^{i\frac{a}{2b}(k^{2}-t^{2})}sinc(t-k),\quad t\in\mathbb{R}.
Proof.

We have

Aψ(ω)\displaystyle\mathscr{F}_{A}\psi(\omega) =ηA(ω)2π|b|Cabϕ(t)ei2b(at2+2pt2ωt)𝑑t\displaystyle=\frac{\eta_{A}(\omega)}{\sqrt{2\pi|b|}}\int_{\mathbb{R}}C_{-\frac{a}{b}}\phi(t)e^{\frac{i}{2b}(at^{2}+2pt-2\omega t)}dt
=ηA(ω)|b|ϕ^(ωpb)\displaystyle=\frac{\eta_{A}(\omega)}{\sqrt{|b|}}\widehat{\phi}(\frac{\omega-p}{b})
=ηA(ω)2π|b|χ[π,π](ωpb)\displaystyle=\frac{\eta_{A}(\omega)}{\sqrt{2\pi|b|}}\chi_{[-\pi,\pi]}(\frac{\omega-p}{b})
=ηA(ω)2π|b|χI+p(ω),\displaystyle=\frac{\eta_{A}(\omega)}{\sqrt{2\pi|b|}}\chi_{I+p}(\omega),

from which it follows that k|Aψ(ω+2kbπ)|2=12π|b|\sum_{k\in\mathbb{Z}}|\mathscr{F}_{A}\psi(\omega+2kb\pi)|^{2}=\frac{1}{2\pi|b|} for almost all ωI\omega\in I, and this implies that {TkAψ:k}\{T_{k}^{A}\psi:k\in\mathbb{Z}\} is an orthonormal basis for the space VA(ψ)V_{A}(\psi). Moreover, we have ψA(ω)=kCabsinc(k)ρA(k)eibkω=1\psi_{A}^{\dagger}(\omega)=\sum_{k\in\mathbb{Z}}C_{-\frac{a}{b}}sinc(k)\rho_{A}(k)e^{-\frac{i}{b}k\omega}=1. Consequently 1/ψA=1.1/\psi_{A}^{\dagger}=1. So by Theorem 6.1 we have AS(ω)=Aψ(ω),\mathscr{F}_{A}S(\omega)=\mathscr{F}_{A}\psi(\omega), which implies that S=ψ.S=\psi. Hence for fVA(ψ)f\in V_{A}(\psi) we have

f(t)\displaystyle f(t) =kf(k)TkAψ(t)=kf(k)eiabk(tk)eia2b(tk)2sinc(tk)\displaystyle=\sum_{k\in\mathbb{Z}}f(k)T_{k}^{A}\psi(t)=\sum_{k\in\mathbb{Z}}f(k)e^{-i\frac{a}{b}k(t-k)}e^{i\frac{a}{2b}(t-k)^{2}}sinc(t-k)
=kf(k)eia2b(t2k2)sinc(tk).\displaystyle=\sum_{k\in\mathbb{Z}}f(k)e^{-i\frac{a}{2b}(t^{2}-k^{2})}sinc(t-k).

Corollary 6.5.

Let ϕ=sinc\phi=sinc. Let ψ=Cabϕ\psi=C_{-\frac{a}{b}}\phi. Then VA(ψ)=BI+pAV_{A}(\psi)=B_{I+p}^{A}, where BI+pA={fL2():suppA(f)I+p}.B_{I+p}^{A}=\{f\in L^{2}(\mathbb{R}):supp\mathscr{F}_{A}(f)\subseteq I+p\}.

Proof.

It is clear from Corollary 6.4 that VA(ψ)BI+pAV_{A}(\psi)\subseteq B_{I+p}^{A}. Now we shall show that BI+pAB_{I+p}^{A} is a closed subspace of L2()L^{2}(\mathbb{R}) and the orthogonal complement of VA(ψ)V_{A}(\psi) in BI+pAB_{I+p}^{A} is zero, which will prove our assertion. As we have A(Cabf)(ω)=ηA(ω)|b|f^(ωpb)\mathscr{F}_{A}(C_{-\frac{a}{b}}f)(\omega)=\frac{\eta_{A}(\omega)}{\sqrt{|b|}}\hat{f}(\frac{\omega-p}{b}), fCabff\mapsto C_{-\frac{a}{b}}f is an isometry from B[π,π](:={fL2():suppf^[π,π]})B_{[-\pi,\pi]}(:=\{f\in L^{2}(\mathbb{R}):supp\hat{f}\subset[-\pi,\pi]\}) onto BI+pAB_{I+p}^{A}. Therefore BI+pAB_{I+p}^{A} is a closed subspace of L2()L^{2}(\mathbb{R}). In Corollary 6.4 we have seen that {TkAψ:k}\{T_{k}^{A}\psi:k\in\mathbb{Z}\} is an orthonormal basis for VA(ψ)V_{A}(\psi). Consider fBI+pAf\in B_{I+p}^{A} such that f,TkAψ=0,k\langle f,T_{k}^{A}\psi\rangle=0,~{}~{}\forall~{}k\in\mathbb{Z}. Now we shall show that f=0f=0. For all kk\in\mathbb{Z} we have

0=f,TkAψ\displaystyle 0=\langle f,T_{k}^{A}\psi\rangle =A(f),A(TkAψ)\displaystyle=\langle\mathscr{F}_{A}(f),\mathscr{F}_{A}(T_{k}^{A}\psi)\rangle
=I+pA(f)(ω)ρA¯(k)eibkωA(ψ)(ω)¯𝑑ω\displaystyle=\int_{I+p}\mathscr{F}_{A}(f)(\omega)\overline{\rho_{A}}(k)e^{\frac{i}{b}k\omega}\overline{\mathscr{F}_{A}(\psi)(\omega)}d\omega
=ρA¯(k)2π|b|I+pA(f)(ω)ηA(ω)¯eibkω𝑑ω.\displaystyle=\frac{\overline{\rho_{A}}(k)}{\sqrt{2\pi|b|}}\int_{I+p}\mathscr{F}_{A}(f)(\omega)\overline{\eta_{A}(\omega)}e^{\frac{i}{b}k\omega}d\omega.

Using the fact that {12π|b|eibkω:k}\{\frac{1}{\sqrt{2\pi|b|}}e^{-\frac{i}{b}k\omega}:k\in\mathbb{Z}\} is an orthonormal basis for L2(I+p)L^{2}(I+p), we get A(f)(ω)ηA(ω)¯=0\mathscr{F}_{A}(f)(\omega)\overline{\eta_{A}(\omega)}=0 for a.e. ωI+p\omega\in I+p, which in turn implies that f2=A(f)2=0\|f\|_{2}=\|\mathscr{F}_{A}(f)\|_{2}=0, proving our assertion. ∎

As a consequence of Corollary 6.4, Corollary 6.5 can be restated as Shannon sampling theorem for the SAFT domain.

Corollary 6.6.

Let fL2()f\in L^{2}(\mathbb{R}) be such that the supp(Af)I+psupp(\mathscr{F}_{A}f)\subseteq I+p. Then the following sampling formula holds

f(t)=kf(k)eia2b(k2t2)sinc(tk),t.f(t)=\sum_{k\in\mathbb{Z}}f(k)e^{i\frac{a}{2b}(k^{2}-t^{2})}sinc(t-k),\quad t\in\mathbb{R}.
Corollary 6.7.

Let ϕ=χ[0,1]χ[0,1].\phi=\chi_{[0,1]}\star\chi_{[0,1]}. Then for every fVA(ϕ)f\in V_{A}(\phi), we have the following reconstruction formula

f(t)=nf(n)TnAϕ(t),t.f(t)=\sum_{n\in\mathbb{Z}}f(n)T_{n}^{A}\phi(t),\quad t\in\mathbb{R}.
Proof.

Let GG be the Gramian associated with the sequence {TkAϕ:k}.\{T_{k}^{A}\phi:k\in\mathbb{Z}\}. Then Gj,k=TkAϕ,TjAϕG_{j,k}=\langle T_{k}^{A}\phi,T_{j}^{A}\phi\rangle is a tridiagonal operator with all the diagonal elements, di=1d_{i}=1. The elements above the diagonal are given by
uj={b2a2eiabj(eiab(2iba1)(1+2iba)),a01/6a=0,u_{j}=\begin{cases}\frac{b^{2}}{a^{2}}e^{i\frac{a}{b}j}(e^{-i\frac{a}{b}}(\frac{2ib}{a}-1)-(1+\frac{2ib}{a})),&a\neq 0\\ 1/6&a=0,\end{cases}
and the elements below the diagonal are given by
lj={b2a2eiabj(eiab(2iba+1)+(12iba)),a01/6a=0.l_{j}=\begin{cases}-\frac{b^{2}}{a^{2}}e^{-i\frac{a}{b}j}(e^{i\frac{a}{b}}(\frac{2ib}{a}+1)+(1-\frac{2ib}{a})),&a\neq 0\\ 1/6&a=0.\end{cases}
Notice that for a=0,Ga=0,~{}G is strictly diagonally dominant and hence invertible, for a0a\neq 0 |uj|,|lj||u_{j}|,~{}|l_{j}| are dominated by 4b3a3a24b2+1.\frac{4b^{3}}{a^{3}}\sqrt{\frac{a^{2}}{4b^{2}}+1.} Now let a2b=r\frac{a}{2b}=r with |r|1.2|r|\geq 1.2, then GG is strictly diagonally dominant and hence invertible. In other words {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} forms a Riesz basis for VA(ϕ).V_{A}(\phi). Further ϕA=1\phi_{A}^{\dagger}=1 on II, from which the required assertion follows. ∎

7. A local reconstruction method

As in the case of classical shift invariant space we can obtain a local reconstruction method for functions belonging to VA(ϕ)V_{A}(\phi) with continuous generators satisfying polynomial decay from their samples. We state the results without the proof as the proofs follow similar lines. We refer to the works [20], [21].

Proposition 7.1.

Let ϕ\phi be a complex valued continuous function on \mathbb{R} satisfying ϕ(x)=o(1|x|ρ)(ρ>1).\phi(x)=o(\frac{1}{|x|^{\rho}})~{}(\rho>1). Assume that {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is a Riesz basis for VA(ϕ)V_{A}(\phi). Let fVA(ϕ)f\in V_{A}(\phi) and [a,b][a^{\prime},b^{\prime}] be an interval in .\mathbb{R}. Then for a given ϵ>0\epsilon>0 there exist a positive integer MM and a sequence cf={ck}2()c_{f}=\{c_{k}\}\in\ell^{2}(\mathbb{Z}) such that

|f(x)gr(x)|<cf2()ϵNρ2,|f(x)-g_{r}(x)|<\|c_{f}\|_{\ell^{2}(\mathbb{Z})}\frac{\epsilon}{N^{\frac{\rho}{2}}},

for all NMN\geq M, for all x[a,b]x\in[a^{\prime},b^{\prime}] and gr(x)=k[aN+1,b+N1]ckTkAϕg_{r}(x)=\sum_{k\in[a^{\prime}-N+1,b^{\prime}+N-1]}c_{k}T_{k}^{A}\phi. In other words f|[a,b]f|_{[a^{\prime},b^{\prime}]} can be approximately determined by a finite number of coefficients ckc_{k} locally.

Theorem 7.2.

Fix ρ2.\rho\geq 2. Let ϕ\phi be a complex valued continuous function on \mathbb{R} satisfying ϕ(x)=o(1|x|ρ).\phi(x)=o(\frac{1}{|x|^{\rho}}). Assume that {TkAϕ:k}\{T_{k}^{A}\phi:k\in\mathbb{Z}\} is a Riesz basis for VA(ϕ)V_{A}(\phi). Let fVA(ϕ),[a,b]f\in V_{A}(\phi),~{}[a^{\prime},b^{\prime}] be an interval in \mathbb{R} and ϵ>0\epsilon>0. Let MM be a positive integer obtained in Proposition 7.1. Consider those points xjx_{j} in the sample set XX such that xj[a,b]x_{j}\in[a^{\prime},b^{\prime}]. Let (2M+ba1)#XMρ(2M+b^{\prime}-a^{\prime}-1)\leq\#X\leq M^{\rho}, where #X\#X denotes the number of points in XX. Define Uj,k=TkAϕ(xj),1j#X,k[aM+1,b+M1]U_{j,k}=T_{k}^{A}\phi(x_{j}),~{}1\leq j\leq\#X,~{}k\in[a^{\prime}-M+1,b^{\prime}+M-1]\cap\mathbb{Z}. Then there exist grVA(ϕ)g_{r}\in V_{A}(\phi) such that

f|Xgr|Xϵ(1+UU)+𝒪(ϵ2),\|f|_{X}-g_{r}|_{X}\|\leq\epsilon(1+\|U\|~{}\|U^{\dagger}\|)+\mathcal{O}(\epsilon^{2}),

where UU^{\dagger} is the pseudoinverse of UU.

Experimental results

Local reconstruction method in the SAFT domain is implemented using Mathematica. We take a=2,b=3,d=4,p=q=0a=2,~{}b=3,~{}d=4,~{}p=q=0 in the implementation.

We take ϕ=χ[12,12]χ[12,12]\phi=\chi_{[-\frac{1}{2},\frac{1}{2}]}\star\chi_{[-\frac{1}{2},\frac{1}{2}]} and ff, a linear combination of integer AA-translates of ϕ.\phi. Here we reconstruct ff in the interval [10,10][-10,10] from a sample set X[10,10],#X=61,M=10.X\subset[-10,10],~{}\#X=61,~{}M=10. We also plot the SAFT of ff and tabulate error for various values of MM in Table 1.

Refer to caption
(a) Real part
Refer to caption
(b) Imaginary part
Figure 1. Original signal
Refer to caption
(a) Real part
Refer to caption
(b) Imaginary part
Figure 2. Reconstructed signal
Refer to caption
(a) Real part
Refer to caption
(b) Imaginary part
Figure 3. SAFT of the signal in Figure 1
Table 1.
Computation of error
MM Error
1010 5.74792×10145.74792\times 10^{-14}
5050 5.69042×10145.69042\times 10^{-14}
250250 5.50736×10145.50736\times 10^{-14}
400400 4.7852×10144.7852\times 10^{-14}

Acknowledgement

F.F. was partially supported by the Helmholtz Pilot Project Ptychography4.0 ZT-I-0025 and by the project ZT-I-PF-4-024 within the Helmholtz Imaging Platform.

References

  • [1] S. Abe and J. T. Sheridan, Optical operations on wave functions as the Abelian subgroups of the special affine Fourier transformation, Optics Letters 19 (1994), no. 22, 1801–1803.
  • [2] A. Aldroubi and K. Gröchenig, Beurling-Landau-type theorems for non-uniform sampling in shift invariant spline spaces, J. Fourier Anal. Appl. 6 (2000), no. 1, 93–103. MR 1756138
  • [3] A. Aldroubi and K. Gröchenig, Nonuniform sampling and reconstruction in shift-invariant spaces, SIAM Rev. 43 (2001), no. 4, 585–620. MR 1882684
  • [4] A. Aldroubi, Q. Sun, and W.-S. Tang, pp-frames and shift invariant subspaces of LpL^{p}, J. Fourier Anal. Appl. 7 (2001), no. 1, 1–21. MR 1812992
  • [5] A. Aldroubi, Q. Sun, and W.-S. Tang, Nonuniform average sampling and reconstruction in multiply generated shift-invariant spaces, Constr. Approx. 20 (2004), no. 2, 173–189. MR 2036639
  • [6] A. Bhandari and A. I. Zayed, Shift-invariant and sampling spaces associated with the special affine Fourier transform, Appl. Comput. Harmon. Anal. 47 (2019), no. 1, 30–52. MR 3958693
  • [7] M. H. A. Biswas, H. G. Feichtinger, and R. Ramakrishnan, Modulation spaces, multipliers associated with the special affine Fourier transform, Complex Anal. Oper. Theory 16 (2022), no. 6, Paper No. 86, 30. MR 4458766
  • [8] P. L. Butzer and R. L. Stens, Sampling theory for not necessarily band-limited functions: a historical overview, SIAM Rev. 34 (1992), no. 1, 40–53. MR 1156288
  • [9] W. Chen and S. Itoh, A sampling theorem for shift-invariant subspace, IEEE Trans. Signal Process. 46 (1998), no. 10, 2822–2824. MR 1718413
  • [10] O. Christensen, An Introduction to Frames and Riesz Bases, second ed., Applied and Numerical Harmonic Analysis, Birkhäuser/Springer, [Cham], 2016. MR 3495345
  • [11] M. A. de Gosson, Symplectic Methods in Harmonic Analysis and in Mathematical Physics, Pseudo-Differential Operators. Theory and Applications, vol. 7, Birkhäuser/Springer Basel AG, Basel, 2011. MR 2827662
  • [12] G. B. Folland, A Course in Abstract Harmonic Analysis, Studies in Advanced Mathematics, CRC Press, Boca Raton, FL, 1995. MR 1397028
  • [13] H. Führ and J. Xian, Relevant sampling in finitely generated shift-invariant spaces, J. Approx. Theory 240 (2019), 1–15. MR 3926050
  • [14] A. G. García, M. A. Hernández-Medina, and G. Pérez-Villalón, Generalized sampling in shift-invariant spaces with multiple stable generators, J. Math. Anal. Appl. 337 (2008), no. 1, 69–84. MR 2356055
  • [15] J. J. Healy, M. A. Kutay, H. M. Ozaktas, and J. T. Sheridan, Linear Canonical Transforms: Theory and Applications, vol. 198, Springer, 2015.
  • [16] E. Hernández and G. Weiss, A First Course on Wavelets, Studies in Advanced Mathematics, CRC Press, Boca Raton, FL, 1996, With a foreword by Yves Meyer. MR 1408902
  • [17] M. I. Kadets, The exact value of the Paley–Wiener constant, Dokl. Akad. Nauk SSSR 155 (1964), 1253–1254. MR 0162088
  • [18] R. Larsen, An Introduction to the Theory of Multipliers, Springer-Verlag, New York-Heidelberg, 1971, Die Grundlehren der mathematischen Wissenschaften, Band 175. MR 0435738
  • [19] M. Z. Nashed and Q. Sun, Sampling and reconstruction of signals in a reproducing kernel subspace of Lp(d)L^{p}(\mathbb{R}^{d}), J. Funct. Anal. 258 (2010), no. 7, 2422–2452. MR 2584749
  • [20] R. Radha, K. Sarvesh, and S. Sivananthan, Sampling and reconstruction in a shift invariant space with multiple generators, Numer. Funct. Anal. Optim. 40 (2019), no. 4, 365–385. MR 3924210
  • [21] R. Radha and S. Sivananthan, Local reconstruction of a function from a non-uniform sampled data, Appl. Numer. Math. 59 (2009), no. 2, 393–403. MR 2484929
  • [22] W. Rudin, Functional Analysis, second ed., International Series in Pure and Applied Mathematics, McGraw-Hill, Inc., New York, 1991. MR 1157815
  • [23] A. A. Selvan and R. Radha, Sampling and reconstruction in shift-invariant spaces on d\mathbb{R}^{d}, Ann. Mat. Pura Appl. (4) 194 (2015), no. 6, 1683–1706. MR 3419912
  • [24] Q. Sun, Local reconstruction for sampling in shift-invariant spaces, Adv. Comput. Math. 32 (2010), no. 3, 335–352. MR 2595176
  • [25] J. Xian and S. Li, Sampling set conditions in weighted multiply generated shift-invariant spaces and their applications, Appl. Comput. Harmon. Anal. 23 (2007), no. 2, 171–180. MR 2344609
  • [26] Q. Zhang, B. Liu, and R. Li, Dynamical sampling in multiply generated shift-invariant spaces, Appl. Anal. 96 (2017), no. 5, 760–770. MR 3613029
  • [27] X. Zhou and W. Sun, On the sampling theorem for wavelet subspaces, J. Fourier Anal. Appl. 5 (1999), no. 4, 347–354. MR 1700088

Statements and Declarations
1. All authors read and approved the final manuscript. There is no conflict of interest.
2. There is no associated data in the manuscript.