The paper is organized as follows: In Section 3 we discuss operator algebras of time-frequency shifts for a separated set and we show that there is a correspondence between symmetry of the separated set in ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  and the structure of the corresponding operator algebra. In this context we review some well-known results on projective representations and we define the adjoint set of an arbitrary separated set in ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  and discuss its relation to the structure of the commutant of the operator algebra.

In Section 4 we define noncommutative tori and Feichtinger's algebra.

In this context we show that  1 is a Hilbert $C^{*}(\Lambda )$  -valued inner product for $f,g$  in $S_0\left({\mathbb{R}}^d\right)$  using functorial properties of $S_0\left({\mathbb{R}}^d\right)$  . Furthermore we introduce the Short-Time Fourier Transform and its properties for functions in Feichtinger's algebra $S_0\left({\mathbb{R}}^d\right)$  .

In Section 5 we prove our main theorem, by showing that Feichtinger's algebra is a bimodule with respect to the pair of $C^{*}$  -algebras $C^{*}(\Lambda )$  and $C^{*}\left({\Lambda }^0\right))$  . We discuss the notion of strong Morita equivalence in this setting and its relation to Gabor expansions, Gabor frame operators, $\Lambda$  -invariant operators for a lattice $\Lambda$  in ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  . In the proof of Rieffel's associativity condition we observe its equivalence to the Fundamental Identity of Gabor Analysis and the existence of Janssen representation for Gabor frame operators, [Rief88] . Our presentation follows [Rief88] , where he used the Poisson's summation formula for symplectic Fourier transform with respect to a lattice and its adjoint lattice in the proof of the Fundamental Identiy of Gabor Analysis. Furthermore, we present the Gröchenig/Leinert [GL04] result on Wiener's lemma for twisted convolution in the context of Connes' construction of projective modules over $C^{*}(\Lambda )$  .

In Section 6 we apply our results on strong Morita equivalence of the $C^{*}$  -algebras $\left(C^{*}\right(\Lambda ),C^{*}({\Lambda }^0\left)\right)$  to the structure of Gabor frames.

Especially we derive the Wexler-Raz biorthogonality principle for the existence of dual windows for Gabor frames from the relation between traces on the Morita equivalent $C^{*}$  -algebras on $C^{*}(\Lambda )$  and $C^{*}\left({\Lambda }^0\right)$  , respectively.

3 Operator Algebras of Time-Frequency Shifts

The most general framework in Gabor analysis builds reconstruction formulas for functions on ${\mathbb{R}}^d$  from a set of time-frequency shifts $\mathcal{G}=\left\{\pi \right(X_j):X_j\in A\}$  for a countable subset $A=\{X_j=(x_j,{\omega }_j):j\in J\}$  with ${inf}_{j,k}|X_j-X_k|>\delta >0$  in the time-frequency plane ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  . We will refer to this setting as non-uniform Gabor systems.

First we derive some results about the structure of the operator algebra of a non-uniform Gabor system $\mathcal{G}$  from general principles. First the commutant ${\mathcal{G}}^{\prime }$  of all bounded operators on $L^2\left({\mathbb{R}}^d\right)$  commuting with every operator in $\mathcal{A}$  is a unital Banach algebra of time-frequency shifts with respect to operator composition and the operator norm, because $\mathcal{A}$  is a subset of unitary (bounded) operators on $L^2\left({\mathbb{R}}^d\right)$  . Furthermore, if $\mathcal{G}$  is generated by a set $A$  which is symmetric with respect to the origin, then our operator algebra $\mathcal{G}$  is invariant under taking adjoints, moreover $\mathcal{G}$  is actually a $C^{*}$  -algebra of time-frequency shifts acting on $L^2\left({\mathbb{R}}^d\right)$  . We have the following chains between $\mathcal{G}$  and its $n$  -th commutant ${\mathcal{G}}^{\left(n\right)}$  :

$$\begin{array}{ccc}\mathcal{G}& \subset & {\mathcal{G}}^{{}^{\prime \prime }}={\mathcal{G}}^{\left(4\right)}=\cdots =\end{array}$$

$$\begin{array}{ccc}& {\mathcal{G}}^{{}^{\prime }}& ={\mathcal{G}}^{\left(3\right)}={\mathcal{G}}^{\left(5\right)}=\cdots .\end{array}$$

From now one, let our set of time-frequency shifts $A$  be a lattice $\Lambda$  of ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  then $\mathcal{G}={\mathcal{G}}^{{}^{\prime \prime }}$  . The associated Gabor system is called to be regular. These observations indicate a strong relationship between the symmetry of the set $A$  and the structure of the corresponding operator algebra $\mathcal{G}$  and its commutant ${\mathcal{G}}^{{}^{\prime }}$  . In the following we discuss the connection between the set $A$  of points in ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  and the set $A^0$  of points, which generate the commutant ${\mathcal{G}}^{{}^{\prime }}$  . This investigation yields a beautiful structure of regular Gabor systems, which is the very reason for the existence of all of our subsequent results. Therefore, we treat the relation between $A$  and $A^0$  more concretely.

Our presentation of the relation between $A$  and $A^0$  relies on the properties of projective representations of the time-frequency plane ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  . Later we will need some results about bicharacters and 2-cocycles associated to our projective representations. Therefore, we now recall their definitions and some of their basic properties. For a similar treatment of projective representation, see [DV04] .

Let $G$  be a locally compact abelian group and $\mathbb{T}$  the multiplicative group of complex numbers of absolute value $1$  . A map $\beta$  of $G\times G$  with values in $\mathbb{T}$  is a multiplier or (2-)cocycle if it satisfies for all $x,y,z\in G$  :

$$\begin{array}{ccc}& & \beta (x,0)=\beta (0,x)=1,\end{array}$$

$$\begin{array}{ccc}& & \beta (x+y,z)\beta (x,y)=\beta (x,y+z)\beta (y,z).\end{array}$$

Two cocycles $\beta$  and ${\beta }^{\prime }$  are called equivalent or cohomologous if there is a Borel map $c$  of $G$  into $\mathbb{T}$  , such that for all $x,y\in G$  $${\beta }^{\prime }(x,y)=\beta (x,y)\frac{c(x+y)}{c\left(x\right)c\left(y\right)}$$  A projective representation $\pi$  of $G$  is a map of $G$  into the unitary group of a Hilbert space $\mathcal{\mathscr{H}}$  such that for a cocycle $\beta$  $$\pi \left(x\right)\pi \left(y\right)=\beta (x,y)\pi (x+y),\pi \left(0\right)=1.$$  The map $X=(x,\omega )\mapsto \pi (x,\omega )=M_{\omega }{T}_x$  for $X\in {\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  into $\mathcal{U}\left(L^2\right({\mathbb{R}}^d\left)\right)$  is a projective representation of the time-frequency plane with cocycle ${\beta }^{\prime }(X,Y)=e^{2\pi iy\cdot \omega }$  for $X=(x,\omega )$  and $Y=(y,\eta )$  .

This projective representation of the time-frequency plane is equivalent to a projective representation $\pi$  with $\underline{\text{the}}$  symplectic bicharacter $\beta (X,Y)=e^{2\pi i(y\omega -x\eta )}$  via the map $c:(x,\omega )\mapsto e^{2\pi ix\cdot \omega }$  for $X=(x,\omega )$  and $Y=(y,\eta )$  in ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  .

Recall that a bicharacter of a locally compact abelian group $G$  is continuous map $b$  of $G\times G$  into $\mathbb{T}$  , which is a character in each argument.

Any such $b$  induces a morphism $\gamma ={\gamma }_b$  of $G$  into $\widehat{G}$  by $$\langle x,{\gamma }_b\left(y\right)\rangle =b(x,y).$$  A bicharacter is called antisymmetric if it satisfies for all $x,y\in G$  $$b(x,y)b(y,x)=1,b(x,x)=1$$  and symplectic if it is alternating and ${\gamma }_b$  is an isomorphism.

In the study of projective representations $\pi$  of $G$  with cocyle $\beta$  symplectic bicharacters of $G$  appear naturally in the commutation rule

$$\begin{array}{c}\pi \left(x\right)\pi \left(y\right)\pi \left(x{)}^{-1}\pi \right(y{)}^{-1}=b(x,y)I\end{array}$$

(2)

with $b(x,y)=\frac{\beta (x,y)}{\beta (y,x)}$  for $x,y\in G$  .

For our projective representation by time-frequency shifts of ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  we recover the bicharacter $\rho (X,Y)=e^{2\pi i(y\omega -\eta x)}$  for $X=(x,\omega )$  and $Y=(y,\eta )$  . Due to the close relation of the projective representation of ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  by time-frequency shifts $\pi (x,\omega )=M_{\omega }{T}_x$  with Heisenberg's commutation relation the bicharacter $\rho (X,Y)=e^{2\pi i(y\omega -\eta x)}$  is called the Heisenberg cocycle for ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  .

The commutation relation ( 2 ) motivates the following definition.

Definition 3.1. Let $A$  be a subset of $G$  and $b$  a bicharacter of $G$  . Then the adjoint set $A_b^0$  of $A$  with respect to $b$  is given by $$A_b^0=\{x\in G:b(x,a)=1\text{for all}\mathtt{}\mathtt{}a\in A\}.$$ 

The adjoint of a set $A$  is a closed subgroup of $G$  by continuity of the character $b$  . Furthermore a subgroup $A$  of $G$  is called to be isotropic for $b$  if $b{|}_{A\times A}\equiv 1$  , or equivalently $A\subset A_b^0$  . We call a group $A$  of $G$  maximal isotropic for $b$  if $A=A_b^0$  , which are again closed subgroups of $G$  .

We are now in position to answer our original question on the relation between a set $A$  generating a non-uniform Gabor system $\mathcal{G}$  and the set $A^0$  generating the commutant ${\mathcal{G}}^{{}^{\prime }}$  .

$$\begin{array}{ccc}A& \subset & A_b^{00}=A_b^{\left(4\right)}=\cdots =\end{array}$$

$$\begin{array}{ccc}& A_b^0& =A_b^{\left(3\right)}=A_b^{\left(5\right)}=\cdots ,\end{array}$$

where $A_b^{\left(n\right)}$  denotes the $n$  th adjoint of $A$  .

The above chains of relations are the set analogous of  2 and therefore, the map $A\mapsto A_b^{\left(0\right)}$  is the desired correspondence.

For the projective representation of ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  by time-frequency shifts $\pi (x,\omega )=M_{\omega }{T}_x$  with the Heisenberg cocycle $\rho$  we obtain the well-known adjoint group ${\Lambda }^0$  of a regular Gabor system generated by a lattice $\Lambda$  of ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  . In this case the maximal isotropic lattice of ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  is the standard (von Neumann) lattice ${\mathbb{Z}}^d\times {\mathbb{Z}}^d$  .

Let us mention that the operator algebra $\mathcal{G}=\left\{\pi \right(\lambda ):\lambda \in \Lambda \}$  is a commutative group of time-frequency shifts if and only if $\Lambda$  is isotropic, i.e. $\Lambda \subset {\Lambda }^0$  .

It is an important fact that one can interpret the Heisenberg cocycle as follows: $\rho$  is a character of ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  , and every character of ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  is of the form

$$\begin{array}{c}X\mapsto \rho (X,X^{\prime })\text{for some}{X}^{\prime }\in {\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d.\end{array}$$

(3)

This induces an isomorphism between ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  and its dual group ${\widehat{\mathbb{R}}}^d\times {\mathbb{R}}^d$  .

If $\Lambda$  is a lattice in ${\mathbb{R}}^{\times }{\widehat{\mathbb{R}}}^d$  , then every character of $\Lambda$  extends to a character of ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  and therefore every character of $\Lambda$  is of the form

$$\begin{array}{c}\lambda \mapsto \rho (\lambda ,Y),\lambda \in \Lambda \end{array}$$

(4)

for some $Y\in {\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  , where $Y$  needs not to be unique. The homomorphism from ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  to $\widehat{\Lambda }$  has as kernel the adjoint lattice

$$\begin{array}{c}{\Lambda }^0=\{Y\in {\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d|\rho (\lambda ,Y)=1\text{for all}\lambda \in \Lambda \}.\end{array}$$

(5)

As a consequence we have, that the adjoint set of a lattice has the structure of a lattice. The skew-bicharacter $\rho$  of ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  gives a Fourier transform ${\widehat{F}}^s$  on the time-frequency plane. We call

$$\begin{array}{ccc}{\widehat{F}}^s\left(Y\right)& =& \int {\int }_{{\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d}\rho (Y,X)F\left(X\right)dX\end{array}$$

$$\begin{array}{ccc}& =& \int {\int }_{{\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d}{e}^{2\pi i(y\cdot \omega -x\cdot \eta )}F(x,\omega )dxd\omega \end{array}$$

the symplectic Fourier transform of $F\in L^2({\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d)$  , since it is induced by the symplectic form $\Omega$  on ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  . The symplectic Fourier transform will be essential in our proof of the Fundamental Identity of Gabor Analysis.

4 Noncommutative Tori and Feichtinger's Algebra

A noncommutative $2d$  -torus ${\mathcal{A}}_{\Theta }$  is the universal $C^{*}$  -algebra generated by $2d$  unitaries $U_1,...,U_{2d}$  subject to the commutation relations $$U_j{U}_k=e^{2\pi i{\theta }_{jk}}{U}_k{U}_j,k,j=1,...,2d$$  for a skew symmetric matrix $\Theta =\left({\theta }_{jk}\right)$  with real entries.

We regard $\Theta$  as a real skew-bilinear form on ${\mathbb{Z}}^{2d}$  , with entries given by $\Theta (e_j,e_k)={\theta }_{jk}$  . Then a noncommutative $2d$  -torus ${\mathcal{A}}_{\Theta }$  is the twisted group $C^{*}$  -algebra $C^{*}(\mathbb{Z},\beta )$  , where $\beta :{\mathbb{Z}}^{2d}\times {\mathbb{Z}}^{2d}\to \mathbb{T}$  is a 2-cocycle such that $$\beta (\lambda ,{\lambda }^{\prime })\overline{\beta ({\lambda }^{\prime },\lambda )}=e^{2\pi i\Theta (\lambda ,{\lambda }^{\prime })}\text{for}\lambda ,{\lambda }^{\prime }\in {\mathbb{Z}}^{2d}.$$  We remark that a noncommutative $2$  -torus $A_{\theta }$  is often called rotation algebra. The commutation rules for the two unitary operators $U$  and $V$  generating $A_{\theta }$  read as $$UV=e^{2\pi i\theta }VU,$$  for a real number $\theta$  . Let $\theta =\alpha \beta$  then the $C^{*}$  -algebra generated by time-frequency shifts $\pi (\alpha k,\beta l)=M_{\beta l}{T}_{\alpha k}$  for $k,l\in {\mathbb{Z}}^d$  is a representation of ${\mathcal{A}}_{\theta }$  on $L^2\left({\mathbb{R}}^d\right)$  .

In this paper we want to treat the general case of $2d$  -noncommutative tori or equivalently $C^{*}$  -algebras $C^{*}(\Lambda ,\beta )$  of time-frequency shifts $\pi \left(\lambda \right)$  for a lattice $\Lambda$  in ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  with $$\pi \left(\lambda \right)\pi \left({\lambda }^{\prime }\right)=\beta (\lambda ,{\lambda }^{\prime })\pi (\lambda +{\lambda }^{\prime }),\lambda ,{\lambda }^{\prime }\in \Lambda .$$  Therefore, an element of $C^{*}(\Lambda ,\beta )$  is given by $${\sum }_{\lambda \in \Lambda }{a}_{\lambda }\pi \left(\lambda \right)$$  for an arbitrary complex-valued sequence $\mathbf{a}=(a_{\lambda }{)}_{\lambda \in \Lambda }$  .

Moreover, this representation is faithful on $L^2\left({\mathbb{R}}^d\right)$  , which is of great importance in our proofs. One consequence, is that it is sufficient to establish statements for a dense subspace of $L^2\left({\mathbb{R}}^d\right)$  . For an operator algebraic proof see [Rief88] and in [?]  a proof is given using time-frequency methods.

The choice of a sequence spaces on $\Lambda$  induces on the noncommutative torus an additional structure. The space $\mathcal{S}(\Lambda )$  of sequences on $\Lambda$  with decay faster than the inverse of any polynomial yields a smooth structure on the algebra of functions on $C^{*}(\Lambda ,\beta )$  , i.e. $${\mathcal{A}}_{\Lambda }^{\infty }=\{A\in \mathcal{ℬ}\left(L^2\left({\mathbb{R}}^d\right)\right):A={\sum }_{\lambda }{a}_{\lambda }\pi (\lambda ),\mathbf{a}=(a_{\lambda }{)}_{\lambda \in \Lambda }\in \mathcal{S}(\Lambda )\}.$$  In the present paper we introduce another structure on $C^{*}(\Lambda ,\beta )$  .

Namely,

$$\begin{array}{c}{\mathcal{A}}_{\Lambda }^1=\{A\in \mathcal{ℬ}\left(L^2\left({\mathbb{R}}^d\right)\right):A={\sum }_{\lambda }{a}_{\lambda }\pi (\lambda ),\mathbf{a}=(a_{\lambda }{)}_{\lambda \in \Lambda }\in {\ell }^1(\Lambda )\}.\end{array}$$

(6)

The commutation rules for the time-frequency shifts $\pi \left(\lambda \right)$  give $C^{*}(\Lambda ,\beta )$  the following structure:

• (1) Let $A_1={\sum }_{\lambda \in \Lambda }{a}_1\left(\lambda \right)\pi \left(\lambda \right)$  and $A_2={\sum }_{\lambda \in \Lambda }{a}_2\left(\lambda \right)\pi \left(\lambda \right)$  for ${\mathbf{a}}_1,{\mathbf{a}}_2\in {\ell }^1(\Lambda )$  then the product of $A_1$  and $A_2$  is given by $$A_1\cdot A_2={\sum }_{\lambda \in \Lambda }{\mathbf{a}}_1{♮}_{\Lambda }{\mathbf{a}}_2\left(\lambda \right)\pi \left(\lambda \right),$$  where $${\mathbf{a}}_1{♮}_{\Lambda }{\mathbf{a}}_2\left(\lambda \right)={\sum }_{\mu \in \Lambda }{a}_1\left(\mu \right)a_2(\lambda -\mu )\beta (\mu ,\lambda -\mu )$$  denotes twisted convolution of ${\mathbf{a}}_1$  and ${\mathbf{a}}_2$  and ${\mathbf{a}}_1{♮}_{\Lambda }{\mathbf{a}}_2\left(\lambda \right)$  is again in $\in {\ell }^1(\Lambda )$  .
• (2) Let $A={\sum }_{\lambda \in \Lambda }a\left(\lambda \right)\pi \left(\lambda \right)$  for $\mathbf{a}\in {\ell }^1(\Lambda )$  then involution $$A^{*}={\sum }_{\lambda \in \Lambda }a\left(\lambda {)}^{*}\pi \right(\lambda )$$  induces an involution on ${\ell }^1(\Lambda )$  :

  $$\begin{array}{c}a(\lambda {)}^{*}=\beta (\lambda ,\lambda )\overline{a(-\lambda )}.\end{array}$$

  (7)

Therefore, we have that

Proposition 4.1. $\left({\ell }^1(\Lambda ),{♮}_{\Lambda },*\right)$  is an involutive Banach algebra.

For the construction of a Hilbert $C^{*}(\Lambda ,\beta )$  -module $V$  we are looking for a time-frequency homogenous Banach space, with properties similar to $\mathcal{S}\left({\mathbb{R}}^d\right)$  . In his seminal paper [F81] Feichtinger has introduced such a space, nowadays called Feichtinger's algebra and denoted by $S_0\left({\mathbb{R}}^d\right)$  .

Hence, we recall the definition of Feichtinger's algebra and some of its basic properties. For a detailed discussion we refer the reader to [F81, FK98, Gr01] . There are many characterizations of Feichtinger's algebra $S_0\left({\mathbb{R}}^d\right)$  . The connection between Feichtinger's algebra and Rieffel's work is most transparent in terms of time-frequency analysis.

Time-frequency representations contain information about the content of time and frequency in a signal $f$  . For our further investigations we restrict our considerations to the Short Time Fourier Transform (STFT). The STFT of a function $f$  with respect to a window function $g$  is defined for $f,g\in L^2\left({\mathbb{R}}^d\right)$  as

$$\begin{array}{c}{V}_{g}f(x,\omega )={\int }_{\mathbb{R}}f\left(t\right)\overline{g(t-x)}{e}^{-2\pi it\cdot \omega }dt,(x,\omega )\in {\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d\end{array}$$

(8)

or equivalently as

$$\begin{array}{c}{V}_{g}f(x,\omega )=\langle f,M_{\omega }{T}_{x}g\rangle ,\text{for}(x,\omega )\in {\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d.\end{array}$$

(9)

Feichtinger's algebra $S_0\left({\mathbb{R}}^d\right)$  is defined as follows

$$\begin{array}{c}{S}_0\left({\mathbb{R}}^d\right)=\{f\in L^2({\mathbb{R}}^d)|\int {\int }_{{\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d}\left|V_{\phi }f\right(x,\omega \left)\right|dxd\omega <\infty \},\end{array}$$

(10)

where $\phi \left(x\right)=2^{-d/4}{e}^{-\pi x\cdot x}$  is a Gaussian and its norm is defined by $$\parallel f{\parallel }_{S_0}=\int {\int }_{{\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d}\left|V_{\phi }f\right(x,\omega \left)\right|dxd\omega .$$  Any other non-zero Schwartz function $g\in \mathcal{S}\left({\mathbb{R}}^d\right)$  , instead of the Gaussian $\phi$  , yields the same space and an equivalent norm for $S_0(\mathbb{R})$  .

Remark 4.2. Feichtinger's algebra is a particular example of a class of Banach spaces, the so-called modulation spaces, which Feichtinger defined via integrability and decay conditions on the STFT over ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  (cf. [F03] ). They have been recognized as the correct class of function spaces for questions in time-frequency analysis, especially Gabor analysis, [FG89a, FG97] .

Theorem 4.3. $S_0\left({\mathbb{R}}^d\right)$  is a Banach algebra under pointwise multiplication.

By this definition of $S_0(\mathbb{R})$  elementary properties of the STFT yield invariance properties of $S_0(\mathbb{R})$  . In the following lemma we state two well-known facts about STFT.

Lemma 4.4. Let $f,g\in L^2\left({\mathbb{R}}^d\right)$  and $(u,\eta )\in {\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  . Then

• (1) Covariance Property of the STFT $$V_g\left(\pi \right(u,\eta \left)f\right)g(x,\omega )=e^{2\pi iu\cdot (\omega -\eta )}{V}_{g}f(x-u,\omega -\eta ).$$ 
• (2) $$V_{g}f(x,\omega )=e^{-2\pi ix\cdot \omega }{V}_{\hat{g}}\hat{f}(\omega ,-x).$$ 

• Proof.
  • (1) The covariance property of the STFT is a consequence of the of the commutation relation $$T_x{M}_{\omega }=e^{-2\pi ix\cdot \omega }{M}_{\omega }{T}_x,(x,\omega )\in {\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d,$$  furthermore one has by definition of the STFT:

    $$\begin{array}{ccc}{V}_g\left(\pi \right(u,\eta \left)f\right)g(x,\omega )& =& \langle M_{\eta }{T}_{u}f,M_{\omega }{T}_{x}g\rangle \end{array}$$

    $$\begin{array}{ccc}& =& \langle f,T_{-u}{M}_{-\eta }{M}_{\omega }{T}_{x}g\rangle \end{array}$$

    $$\begin{array}{ccc}& =& e^{2\pi iu\cdot (\omega -\eta )}{V}_{g}f(x-u,\omega -\eta ).\end{array}$$

  • (2) The formula expresses, that for the STFT the Fourier transform yields a rotation of the time-frequency plane by an angle of ${90}^{\circ }$  . Its another manifestation of the fact that STFT contains information of $f$  and $\hat{f}$  .
  □

Many properties of Feichtinger's algebra $S_0\left({\mathbb{R}}^d\right)$  are elementary consequences of properties of the STFT. The following theorem may be seen as a realization of this principle, where we show translation invariance and Fourier invariance of $S_0\left({\mathbb{R}}^d\right)$  from Lemma ( 4.4 ).

Theorem 4.5. Let $f\in S_0\left({\mathbb{R}}^d\right)$  and $(u,\eta )\in {\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  . Then

• (1) $\pi (u,\eta )f\in S_0\left({\mathbb{R}}^d\right)$  and $\parallel f{\parallel }_{S_0}=\parallel \pi (u,\eta )f{\parallel }_{S_0}.$ 
• (2) $\hat{f}\in S_0\left({\mathbb{R}}^d\right)$  and $\parallel \hat{f}{\parallel }_{S_0}=\parallel f{\parallel }_{S_0}.$ 

• Proof.
  • (1) By definition of $S_0\left({\mathbb{R}}^d\right)$  we have by the invariance of the Gaussian $\phi$  under Fourier transform that

    $$\begin{array}{ccc}\parallel f{\parallel }_{S_0}& =& \int {\int }_{{\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d}\left|V_{\phi }\right(M_{\eta }{T}_{u}f\left)\right(x,\omega \left)\right|dxd\omega \end{array}$$

    $$\begin{array}{ccc}& =& \int {\int }_{{\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d}\left|V_{\phi }f\right(x-u,\omega -\eta \left)\right|dxd\omega .\end{array}$$

  • (2) For the invariance under the Fourier transform we use (2) of Lemma  4.4 , and that the definition of $S_0\left({\mathbb{R}}^d\right)$  is independent of the window $g\in \mathcal{S}\left({\mathbb{R}}^d\right)$  and that different windows yield equivalent norms for $S_0\left({\mathbb{R}}^d\right)$  :

    $$\begin{array}{ccc}\parallel \hat{f}{\parallel }_{S_0\left({\mathbb{R}}^d\right)}& =& \int {\int }_{{\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d}\left|V_{\phi }\hat{f}\right(x,\omega \left)\right|dxd\omega \end{array}$$

    $$\begin{array}{ccc}& \le & C\int {\int }_{{\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d}\left|V_{\hat{\phi }}\hat{f}\right(x,\omega \left)\right|dxd\omega \end{array}$$

    $$\begin{array}{ccc}& =& C\int {\int }_{{\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d}\left|V_{\phi }f\right(-\omega ,x\left)\right|dxd\omega \end{array}$$

    $$\begin{array}{ccc}& =& C\int {\int }_{{\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d}\left|V_{\phi }f\right(x,\omega \left)\right|dxd\omega =\parallel f{\parallel }_{S_0}.\end{array}$$

  □

Later in our treatment of Rieffel's associativity condition for $\left(C^{*}(\Lambda ),C^{*}\left({\Lambda }^0\right)\right)$  we shall need the following properties of $S_0\left({\mathbb{R}}^d\right)$  .

Theorem 4.6. Let $f,g$  in $S_0\left({\mathbb{R}}^d\right)$  then $$V_{g}f\in S_0({\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d).$$ 

For a proof of this statement we refer the reader to [FK98] , but it also follows from the functorial properties and minimality of $S_0\left({\mathbb{R}}^d\right)$  in [?] .

Let $F$  be a function on the time-frequency plane ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  then the sampling operator for a lattice $\Lambda$  in ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  is defined as follows $${\mathbf{R}}_{\Lambda }:F\mapsto {\left(F\left(\lambda \right)\right)}_{\lambda \in \Lambda }.$$ 

Remark 4.7. We will also write occasionally $F{|}_{\Lambda }$  instead of ${\mathbf{R}}_{\Lambda }F$  .

Theorem 4.8. Let $\Lambda$  be a lattice in ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  and $F\in S_0({\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d)$  then

$$\begin{array}{c}{\mathbf{R}}_{\Lambda }F\in S_0(\Lambda )={\ell }^1(\Lambda )\end{array}$$

(11)

and ${\mathbf{R}}_{\Lambda }$  is bounded on $S_0(\Lambda )$  .

As a consequence of the last theorem we obtain:

Corollary 4.9. Let $f,g\in S_0\left({\mathbb{R}}^d\right)$  then $V_{g}f{|}_{\Lambda }\in {\ell }^1(\Lambda )$  , i.e.

$$\begin{array}{c}{\sum }_{\lambda \in \Lambda }|\langle f,\pi (\lambda )g\rangle |<\infty .\end{array}$$

(12)

Remark 4.10. For $\Lambda$  discrete $S_0(\Lambda )={\ell }^1(\Lambda )$  .

We now define a left action of $C^{*}(\Lambda ,\beta )$  on $S_0\left({\mathbb{R}}^d\right)$  by a Gabor expansion for the window $g$  and the lattice $\Lambda \in {\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  : $$\mathbf{a}g={\sum }_{\lambda \in \Lambda }a\left(\lambda \right)\pi \left(\lambda \right)g,\mathbf{a}=\left(a_{\lambda }\right)\in S_0(\Lambda ,\beta ).$$  The invariance of $S_0\left({\mathbb{R}}^d\right)$  under time-frequency shifts implies that this action is well-defined on $S_0\left({\mathbb{R}}^d\right)$  .

Proposition 4.11. Let $\mathbf{a}\in S_0(\Lambda ,\beta )$  and $g\in S_0\left({\mathbb{R}}^d\right)$  , then $$\parallel {\sum }_{\lambda \in \Lambda }a\left(\lambda \right)\pi \left(\lambda \right)g{\parallel }_{S_0}\le \parallel \mathbf{a}{\parallel }_1\parallel g{\parallel }_{S_0}.$$ 

• Proof.

  $$\begin{array}{ccc}\parallel {\sum }_{\lambda \in \Lambda }a\left(\lambda \right)\pi \left(\lambda \right)g{\parallel }_{S_0}& \le & {\sum }_{\lambda \in \Lambda }\left|a\right(\lambda \left)\right|\parallel \pi \left(\lambda \right)g{\parallel }_{S_0}\end{array}$$

  $$\begin{array}{ccc}& =& {\sum }_{\lambda \in \Lambda }\left|a\right(\lambda \left)\right|\parallel g{\parallel }_{S_0}\end{array}$$

  $$\begin{array}{ccc}& =& \parallel \mathbf{a}{\parallel }_1\parallel g{\parallel }_{S_0}.\end{array}$$

  □

Corollary  4.9 and Proposition  4.11 yield that for $f,g\in S_0\left({\mathbb{R}}^d\right)$  the left action of $$\langle f,g{\rangle }_{\Lambda }=V_{g}f{|}_{\Lambda }$$  is well-defined on $S_0\left({\mathbb{R}}^d\right)$  . In Gabor analysis the mapping of $\langle f,g{\rangle }_{\Lambda }\mapsto \langle f,g{\rangle }_{\Lambda }g$  is called a Gabor type frame operator with window $g$  and lattice $\Lambda$  , denoted by $$S_{g,\Lambda }f={\sum }_{\lambda \in \Lambda }\langle f,\pi \left(\lambda \right)g\rangle \pi \left(\lambda \right)g.$$  The above discussion shows that the Gabor type frame operator $S_{g,\Lambda }$  is continuous on $S_0\left({\mathbb{R}}^d\right)$  for $f,g\in S_0\left({\mathbb{R}}^d\right)$  . In Section 6 we present some consequences for the reconstruction of square-integrable functions $f\in L^2\left({\mathbb{R}}^d\right)$  and is Rieffel's central observation was that

$$\begin{array}{c}\langle f,g{\rangle }_{\mathcal{A}}={\sum }_{\lambda \in \Lambda }\langle f,\pi \left(\lambda \right)g\rangle \pi \left(\lambda \right),f,g\in \mathcal{S}\left({\mathbb{R}}^d\right)\end{array}$$

(13)

is a $C^{*}$  -valued innerproduct for $\mathcal{A}=C^{*}(\Lambda ,\beta )$  . In the sequel we prove that  13 defines a $C^{*}$  -valued innerproduct for $C^{*}(\Lambda ,\beta )$  for $f,g$  in Feichtinger's algebra $S_0\left({\mathbb{R}}^d\right)$  .

First we show that  13 is compatible with the action of $S_0(\Lambda ,\beta )$  on $S_0\left({\mathbb{R}}^d\right)$  . More precisely, we prove the following proposition.

Proposition 4.12. Let $f,g\in S_0\left({\mathbb{R}}^d\right)$  and let $\mathbf{a}\in S_0(\Lambda ,\beta )$  . Then $$\langle \mathbf{a}f,g{\rangle }_{\Lambda }=\mathbf{a}{♮}_{\Lambda }\langle f,g{\rangle }_{\Lambda }.$$ 

• Proof. For $\lambda \in \Lambda$  we have

  $$\begin{array}{ccc}\langle \mathbf{a}f,\pi \left(\lambda \right)g\rangle & =& {\sum }_{{\lambda }^{\prime }\in \Lambda }a\left({\lambda }^{\prime }\right)\langle \pi \left({\lambda }^{\prime }\right)f,\pi \left(\lambda \right)g\rangle \end{array}$$

  $$\begin{array}{ccc}& =& {\sum }_{{\lambda }^{\prime }\in \Lambda }a\left({\lambda }^{\prime }\right)\langle f,{\pi }^{*}\left({\lambda }^{\prime }\right)\pi \left(\lambda \right)g\rangle \end{array}$$

  $$\begin{array}{ccc}& =& {\sum }_{{\lambda }^{\prime }\in \Lambda }a\left({\lambda }^{\prime }\right)\langle f,\pi (\lambda -{\lambda }^{\prime })g\rangle \beta ({\lambda }^{\prime },\lambda -{\lambda }^{\prime })\end{array}$$

  $$\begin{array}{ccc}& =& \mathbf{a}{♮}_{\Lambda }\langle f,\pi \left(\lambda \right)g\rangle ,\end{array}$$

  since $\pi \left({\lambda }^{\prime }\right)\pi (\lambda -{\lambda }^{\prime })=\beta ({\lambda }^{\prime },\lambda -{\lambda }^{\prime })\pi \left(\lambda \right)$  . □

We now come to the statement of one of our main theorems.

Theorem 4.13. Feichtinger's algebra $S_0\left({\mathbb{R}}^d\right)$  is a left Hilbert $C^{*}(\Lambda ,\beta )$  -module with respect to the inner product, given for $f,g$  in $S_0\left({\mathbb{R}}^d\right)$  by $$\langle f,g{\rangle }_{\mathcal{A}}={\sum }_{\lambda \in \Lambda }{V}_{g}f\left(\lambda \right)\pi \left(\lambda \right).$$ 

The proof of Theorem  4.13 is postponed now and will be given after the discussion of the Fundamental Identity of Gabor Analysis, because the positivity of the innerproduct is a direct consequence of FIGA. In Section  5 we derive FIGA from an identity for products of STFT by an application of Poisson's summation formula for the symplectic Fourier transform, which requires the adjoint lattice ${\Lambda }^0$  of the lattice $\Lambda$  in ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  . In Section  5 we study the structure of $C^{*}(\Lambda )$  and define a $C^{*}\left({\Lambda }^0\right)$  -valued innerproduct. The proof of Rieffel's associativity condition is an elementary reformulation of the FIGA. Therefore, Section  5 is the natural place for our presentation of the FIGA. At the end of this section we present generalizations of some notions of Hilbert spaces to Hilbert $C^{*}$  -modules.

Let $V$  be a Hilbert $\mathcal{A}$  -module then a Hilbert module map from $V$  to $V$  is a linear map $T:V\to V$  that respects the module action:

$T(\mathbf{a}f)=\mathbf{a}\mathbf{}\left(T\right(f\left)\right)$  for $\mathbf{a}\in \mathcal{A}$  and $f\in V$  . The adjoint of an operator on a Hilbert space plays a central role in the study of operators on Hilbert spaces and of operator algebras, such as $C^{*}$  -algebras or von Neumann algebras of operators. The following definition gives a generalization of adjoints to Hilbert $\mathcal{A}$  -module.

Definition 4.14. Let $V$  be a Hilbert $\mathcal{A}$  -module. A map $T:V\to V$  is adjointable if there exists a map $T^{*}:V\to V$  satisfying $$\langle Tf,g{\rangle }_{\mathcal{A}}=\langle f,T^{*}g{\rangle }_{\mathcal{A}}$$  for all $f,g$  in $\mathcal{A}$  . The map $T^{*}$  is called the adjoint of $T$  . We denote the set of all adjointable maps by $\mathfrak{L}\left(V\right)$  and the set of all bounded module maps in $V$  by $\mathfrak{B}\left(V\right)$  .

An elementary consequence of the definitions is the following facts about adjointable maps.

• (1) Let $T$  be in $\mathfrak{L}\left(V\right)$  , then its adjoint is unique and adjointable with $T^{**}=T$  .
• (2) Let $T,S$  be in $\mathfrak{L}\left(V\right)$  , then $ST\in \mathfrak{L}\left(V\right)$  with $(ST{)}^{*}=T^{*}{S}^{*}$  .
• (3) $\mathfrak{L}\left(V\right)$  equipped with the operator norm $\parallel T\parallel =sup\{\parallel Tx\parallel :\parallel x\parallel \le 1\}$  is a $C^{*}$  -algebra.
• (4) $\mathfrak{B}\left(V\right)$  equipped with the operator norm $\parallel T\parallel =sup\{\parallel Tx\parallel :\parallel x\parallel \le 1\}$  is a Banach algebra.

In the case of $S_0\left({\mathbb{R}}^d\right)$  as $C^{*}(\Lambda ,\beta )$  -module the adjointable maps are those operators $T:S_0\left({\mathbb{R}}^d\right)\to S_0\left({\mathbb{R}}^d\right)$  where $T^{*}$  commutes with all time-frequency shifts $\left\{\pi \right(\lambda ):\lambda \in \Lambda \}$  . By definition of the $C^{*}(\Lambda ,\beta )$  -innerproduct we have

$$\begin{array}{ccc}\langle Tf,g{\rangle }_{\mathcal{A}}& =& {\sum }_{\lambda \in \Lambda }\langle Tf,\pi \left(\lambda \right)g\rangle \pi \left(\lambda \right)\end{array}$$

$$\begin{array}{ccc}& =& {\sum }_{\lambda \in \Lambda }\langle f,T^{*}\pi \left(\lambda \right)g\rangle \pi \left(\lambda \right)\end{array}$$

$$\begin{array}{ccc}& =& {\sum }_{\lambda \in \Lambda }\langle f,\pi \left(\lambda \right)T^{*}g\rangle \pi \left(\lambda \right)\end{array}$$

$$\begin{array}{ccc}& =& \langle f,Tg{\rangle }_{\mathcal{A}}.\end{array}$$

In [FK98] Feichtinger and Kozek treated selfadjoint operators on $S_0\left({\mathbb{R}}^d\right)$  , which commute with $\left\{\pi \right(\lambda ):\lambda \in \Lambda \}$  . They called those operators $\Lambda$  -invariant. The set of all selfadjoint adjointable operators of $\mathfrak{L}\left(S_0\right({\mathbb{R}}^d\left)\right)$  is an ideal in $\mathfrak{L}\left(S_0\right({\mathbb{R}}^d\left)\right)$  .

The notion of finite rank operators and of compact module operators is of great relevance in the construction of Morita equivalences between $C^{*}$  -algebras, see Section  5 .

Definition 4.15. Let $f,g$  be elements of a Hilbert $\mathcal{A}$  -module $V$  . Then a rank one operator $K_{f,g}:V\to V$  is defined by $$K_{f,g}h:=\langle f,h{\rangle }_{\mathcal{A}}g.$$  The set of compact Hilbert module operators on $V$  is the closed subspace of $\mathfrak{L}\left(V\right)$  generated by the rank one maps $K_{f,g}$  .

We denote the set of compact Hilbert module operators by $\mathcal{K}\left(V\right)=\overline{\{K_{f,g}:f,g\in V\}}$  .

Remark 4.16. A compact Hilbert module operator is not necessarily a bounded operator on $V$  , but for Hilbert $\mathbb{C}$  -modules $\mathcal{\mathscr{H}}$  the notion specializes to the definition of a compact operator on $\mathcal{\mathscr{H}}$  .

The following proposition gives some elementary facts about compact Hilbert module operators.

Proposition 4.17. Let $f,g\in V$  and $T\in \mathcal{ℒ}\left(V\right)$  . Then one has

• (1) $K_{f,g}$  is adjointable and $K_{f,g}^{*}=K_{g,f}$  .
• (2) $T{K}_{f,g}=K_{Tf,g}$  .
• (3) $K_{f,g}T=K_{f,T^{*}g}$  .
• (4) $\parallel K_{f,g}\parallel \le \parallel f\parallel \parallel g\parallel$  .

A direct consequence of the previous observations is the following statement.

Proposition 4.18. Let $V$  be a Hilbert $\mathcal{A}$  -module, $\mathfrak{K}\left(V\right)$  is a closed ideal in $\mathfrak{L}\left(V\right)$  .

Now we investigate the set of rank one module operators for our Hilbert $C^{*}(\Lambda ,\beta )$  module $S_0\left({\mathbb{R}}^d\right)$  . By definition a rank one module operator is given by

$$\begin{array}{ccc}{K}_{g,f}\gamma & =& \langle f,\gamma {\rangle }_{\mathcal{A}}g\end{array}$$

$$\begin{array}{ccc}& =& {\sum }_{\lambda \in \Lambda }\langle f,\pi \left(\lambda \right)\gamma \rangle \pi \left(\lambda \right)g=S_{g,\gamma ,\Lambda }f,\end{array}$$

for $f,g,\gamma \in S_0\left({\mathbb{R}}^d\right)$  . The operator $S_{g,\gamma ,\Lambda }$  is called a Gabor frame operator with analysis window $\gamma$  and synthesis window $g$  for a lattice $\Lambda$  .

Therefore, a finite rank module operator is a finite sum of Gabor frame operators, a so-called multi-window Gabor frame operator. Furthermore, a rank one module operator $S_{g,\gamma ,\Lambda }$  is an adjointable operator, i.e. it is a $\Lambda$  -invariant operator. This elementary fact has far reaching consequences, see Section  6 .

5 Feichtinger's Algebra as Bimodule for $C^{*}(\Lambda )$  and $C^{*}\left({\Lambda }^0\right)$ 

In Section 3 we discussed the relation between an operator algebra of time-frequency shifts generated by a lattice $\Lambda$  in ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  and its commutant. In this section we continue discussion in the light of Morita equivalence of $C^{*}$  -algebras.

The adjoint lattice of $\Lambda$  in ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  was defined as the set of all points $X=(x,\omega )$  in ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  such that $\rho (\lambda ,X)=1$  , which by the commutation relation of time-frequency shifts  2 is equivalent to $${\Lambda }^0=\{{\lambda }^0\in {\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d:\pi (\lambda \left)\pi \right({\lambda }^0)=\pi ({\lambda }^0\left)\pi \right(\lambda )\text{for all}\lambda \in \Lambda \}.$$  Therefore, the set of all bounded operators on $L^2\left({\mathbb{R}}^d\right)$  commuting with elements from $C^{*}(\Lambda ,\beta )$  is the $C^{*}$  -algebra generated by time-frequency shifts $\pi \left({\lambda }^0\right)$  for ${\lambda }^0$  in ${\Lambda }^0$  .

In Section 4 we have defined a left action of $C^{*}(\Lambda ,\beta )$  on $S_0\left({\mathbb{R}}^d\right)$  .

Now $S_0\left({\mathbb{R}}^d\right)$  has the structure of a bimodule, where the right action is induced by the opposite algebra of $C^{*}({\Lambda }^0,\beta )$  . Following Rieffel in [Rief88] , $C^{*}({\Lambda }^0,\beta )$  can be generated by ${\pi }^{*}\left({\lambda }^0\right)$  acting on the left on $S_0\left({\mathbb{R}}^d\right)$  , which commutes with the right action of $C^{*}(\Lambda ,\beta )$  on $S_0\left({\mathbb{R}}^d\right)$  .

Therefore, the opposite algebra of $C^{*}(\Lambda ,\beta {)}^{\text{opp}}$  is generated by ${\pi }^{*}\left({\lambda }^0\right)$  with $\overline{\beta }(X,Y)=\overline{\beta (X,Y)}$  for $X=(x,\omega )$  and $Y=(y,\eta )$  as cocycle, i.e.

$C^{*}({\Lambda }^0,\overline{\beta })$  . By definition the opposite algebra of $C^{*}(\Lambda ,\beta )$  gives a right action on $S_0\left({\mathbb{R}}^d\right)$  by a Gabor expansion with respect to the lattice ${\Lambda }^0$  $$g\mathbf{b}=|\Lambda {|}^{-1}{\sum }_{{\lambda }^0\in {\Lambda }^0}b({\lambda }^0\left){\pi }^{*}\right({\lambda }^0)g,g\in S_0({\mathbb{R}}^d),\mathbf{b}\in S_0({\Lambda }^0,\overline{\beta }).$$  Note, that cohomologous cocycles yield isomorphic $C^{*}$  -algebras. By a reasoning similar to the one used in Section 4 for the left action $C^{*}(\Lambda ,\beta )$  we obtain that the right action is well-defined on $S_0\left({\mathbb{R}}^d\right)$  .

Before $S_0\left({\mathbb{R}}^d\right)$  is given the structure of a right $C^{*}({\Lambda }^0,\overline{\beta })$  we state the Fundamental Identity of Gabor analysis, because it is essential in our construction of the bimodule $S_0\left({\mathbb{R}}^d\right)$  for $C^{*}(\Lambda ,\beta )$  and $C^{*}({\Lambda }^0,\overline{\beta })$  .

Theorem 5.1 (FIGA). Let $f_1,g_1,f_2,g_2\in S_0\left({\mathbb{R}}^d\right)$  , then $${\sum }_{\lambda \in \Lambda }{V}_{g_1}{f}_1\left(\lambda \right)\overline{V_{g_2}{f}_2\left(\lambda \right)}=|\Lambda {|}^{-1}{\sum }_{{\lambda }^0\in {\Lambda }^0}{V}_{g_1}{g}_2({\lambda }^0)\overline{V_{f_1}{f}_2\left({\lambda }^0\right)}$$ 

In [Rief88] Rieffel proved FIGA for Schwartz functions $f_1,f_2,g_1,g_2$  in $\mathcal{S}\left(G\right)$  for an elementary locally compact abelian group $G$  . In [TO95] Tolmieri and Orr proved a special case of Rieffel's result for functions on $\mathbb{R}$  in their study of Gabor frames. Later, Janssen continued the work of Tolmieri/Orr and introduced a representation of Gabor frame operators, Janssen's representation [Jan95] . In his proof of the Morita equivalence of $C^{*}(\Lambda ,\beta )$  and $C^{*}({\Lambda }^0,\overline{\beta })$  Rieffel had derived Janssen's representation of a Gabor frame operator.

Following Rieffel we use Poisson summation formula for symplectic Fourier transform in the proof of FIGA. The following theorem states the Poisson summation formula for the symplectic Fourier transform, see Section 2 for the definition.

Theorem 5.2. Let $F\in S_0({\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d)$  then

$$\begin{array}{c}{\sum }_{\lambda \in \Lambda }F\left(\lambda \right)=|\Lambda {|}^{-1}{\sum }_{{\lambda }^0\in {\Lambda }^0}{\widehat{F}}^s({\lambda }^0)\end{array}$$

(14)

holds pointwise and with absolute convergence of both sums.

• Proof. (FIGA)If $f,g\in S_0\left({\mathbb{R}}^d\right)$  we have that $V_{g}f\in S_0({\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d)$  . Then $F=V_{g_1}{f}_1\overline{V_{g_2}{f}_2}$  is in $S_0({\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d)$  , because $S_0({\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d)$  is a Banach algebra under multiplication. Poisson's summation formula for $F$  yields FIGA. Therefore, we compute the symplectic Fourier transform of $F$  .

  $$\begin{array}{ccc}{\widehat{F}}^s\left(Y\right)& =& \int {\int }_{{\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d}{V}_{g_1}{f}_1\left(X\right)\overline{V_{g_2}{f}_2\left(X\right)}\rho (Y,X)dX\end{array}$$

  $$\begin{array}{ccc}& =& \int {\int }_{{\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d}\langle \pi \left(Y\right)f_1,\pi \left(Y\right)\pi \left(X\right)g_1\rangle \overline{\langle f_2,\pi \left(X\right)g_2\rangle }\rho (X,Y)dX\end{array}$$

  $$\begin{array}{ccc}& =& \int {\int }_{{\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d}\langle \pi \left(Y\right)f_1,\pi \left(X\right)\pi \left(Y\right)g_1\rangle \overline{\langle f_2,\pi \left(X\right)g_2\rangle }\rho (X,Y)dX\end{array}$$

  $$\begin{array}{ccc}& =& \langle f_1,\pi \left(Y\right)f_2\rangle \overline{\langle g_1,\pi \left(Y\right)g_2\rangle },\end{array}$$

  where in the last step we used Moyal's formula. □

As a first application of FIGA we finish the proof of Theorem  4.13 by showing the positivity of

$$\begin{array}{c}\langle f,f{\rangle }_{\mathcal{A}}={\sum }_{\lambda \in \Lambda }\langle f,\pi \left(\lambda \right)f\rangle \pi \left(\lambda \right)\end{array}$$

(15)

as an operator on $L^2\left({\mathbb{R}}^d\right)$  .

Proposition 5.3. Let $f\in S_0\left({\mathbb{R}}^d\right)$  then $\langle f,f{\rangle }_{\mathcal{A}}$  is a positive element of $C^{*}(\Lambda ,\beta )$  .

• Proof. The representation of time-frequency shifts of $C^{*}(\Lambda ,\beta )$  is faithful, therefore, it suffices to establish positivity for a dense subspace of $L^2\left({\mathbb{R}}^d\right)$  . Of course we choose $S_0\left({\mathbb{R}}^d\right)$  as dense subspace. Let $g\in S_0\left({\mathbb{R}}^d\right)$ 

  $$\begin{array}{ccc}\langle \langle f,f{\rangle }_{\Lambda }g,g\rangle & =& \langle {\sum }_{\lambda \in \Lambda }\langle f,\pi \left(\lambda \right)f\rangle \pi \left(\lambda \right)g,g\rangle \end{array}$$

  $$\begin{array}{ccc}& =& {\sum }_{\lambda \in \Lambda }\langle f,\pi \left(\lambda \right)f\rangle \overline{\langle g,\pi \left(\lambda \right)g\rangle }\end{array}$$

  $$\begin{array}{ccc}& =& {\sum }_{{\lambda }^0\in {\Lambda }^0}\langle f,\pi \left({\lambda }^0\right)g\rangle \overline{\langle f,\pi \left({\lambda }^0\right)g\rangle }\ge 0.\end{array}$$

  □

In an analogous manner as in our discussion of Theorem  4.13 we get that the right action of $C^{*}({\Lambda }^0,\overline{\beta })$  with properly defined ${♮}_{{\Lambda }^0}$  and involution $*$  defines a right Hilbert $C^{*}({\Lambda }^0,\overline{\beta })$  -module structure on $S_0\left({\mathbb{R}}^d\right)$  with respect to the $\mathcal{ℬ}:=C^{*}({\Lambda }^0,\overline{\beta })$  -innerproduct $$\langle f,g{\rangle }_{\mathcal{ℬ}}:=|\Lambda {|}^{-1}{\sum }_{{\lambda }^0\in {\Lambda }^0}\langle \pi ({\lambda }^0)g,f\rangle \pi ({\lambda }^0),f,g\in S_0({\mathbb{R}}^d).$$  Two $C^{*}$  -module structures $\left(\mathcal{A},\langle .,.{\rangle }_{\mathcal{A}}\right)$  and $\left(\mathcal{ℬ},\langle .,.{\rangle }_{\mathcal{ℬ}}\right)$  on a bimodule $V$  are compatible if

$$\begin{array}{c}\langle f,g{\rangle }_{\mathcal{A}}h=f\langle g,h{\rangle }_{\mathcal{ℬ}},\text{for all}f,g,h\in V.\end{array}$$

(16)

Some authors call  16  Rieffel's associativity condition for $\langle .,.{\rangle }_{\mathcal{A}}$  and $\langle .,.{\rangle }_{\mathcal{ℬ}}$  .

In our setting Rieffel's associativity condition expresses Janssen's representation of a Gabor frame operator $S_{g,\gamma }$  for a window $g,\gamma \in S_0\left({\mathbb{R}}^d\right)$  .

Theorem 5.4. Let $\mathcal{A}=C^{*}(\Lambda ,\beta )$  and $\mathcal{ℬ}=C^{*}({\Lambda }^0,\overline{\beta })$  with the above defined actions and innerproducts $\langle .,.{\rangle }_{\mathcal{A}}$  and $\langle .,.{\rangle }_{\mathcal{ℬ}}$  , respectively. Then $$S_{g,\gamma ,\Lambda }f=|\Lambda {|}^{-1}{S}_{f,\gamma ,{\Lambda }^0}g$$  for all $f,g,\gamma \in S_0\left({\mathbb{R}}^d\right)$  .

• Proof. As in the proof of positivity of $\langle f,f{\rangle }_{\mathcal{A}}$  for $f\in S_0\left({\mathbb{R}}^d\right)$  it suffices to show that for all $\gamma ,h\in S_0\left({\mathbb{R}}^d\right)$  $$\langle S_{g,\gamma ,\Lambda }f,h\rangle =|\Lambda {|}^{-1}\langle S_{f,\gamma ,{\Lambda }^0}g,h\rangle .$$ 

  $$\begin{array}{ccc}\langle \langle f,g{\rangle }_{\mathcal{A}}\gamma ,h\rangle & =& {\sum }_{\lambda \in \Lambda }\langle f,\pi \left(\lambda \right)g\rangle \overline{\langle h,\pi \left(\lambda \right)\gamma \rangle }\end{array}$$

  $$\begin{array}{ccc}& \stackrel{FIGA}{=}& |\Lambda {|}^{-1}{\sum }_{{\lambda }^0\in {\Lambda }^0}\langle f,\pi ({\lambda }^0)h\rangle \overline{\langle g,\pi \left({\lambda }^0\right)\gamma \rangle }\end{array}$$

  $$\begin{array}{ccc}& =& \langle f\langle g,\gamma {\rangle }_{\mathcal{ℬ}},h\rangle .\end{array}$$

  □

A Hilbert $C^{*}$  -module $V$  over $\mathcal{A}$  is called full when the collection $\{\langle f,g{\rangle }_{\mathcal{A}}:f,g\in V\}$  is dense in $V$  .

Definition 5.5. Two $C^{*}$  -algebras $\mathcal{A}$  and $\mathcal{ℬ}$  are strongly Morita equivalent if there exists a full Hilbert $C^{*}$  -module $V$  over $\mathcal{ℬ}$  such that $\mathcal{ℬ}\simeq \mathcal{K}(V,\mathcal{A})$  .

Remark 5.6. We denote by $\mathcal{K}(V,\mathcal{A})$  the $C^{*}$  -algebra of compact Hilbert $\mathcal{A}$  operators $K_{f,g}^{\mathcal{A}}$  .

The Morita equivalence of $C^{*}(\Lambda ,\beta )$  and $C^{*}({\Lambda }^0,\overline{\beta })$  is a consequence of the following theorem.

Theorem 5.7. Let $S_0\left({\mathbb{R}}^d\right)$  be given a bimodule structure as defined above. Let $\mathcal{A}=C^{*}(\Lambda ,\beta )$  and $\mathcal{ℬ}=C^{*}({\Lambda }^0,\overline{\beta })$  then

• (1) $\{\langle f,g{\rangle }_{\mathcal{A}}:f,g\in S_0({\mathbb{R}}^d\left)\right\}$  is dense in $\mathcal{A}$  , i.e. $S_0\left({\mathbb{R}}^d\right)$  is a full Hilbert $\mathcal{A}$  -module.
• (2) $\{\langle f,g{\rangle }_{\mathcal{ℬ}}:f,g\in S_0({\mathbb{R}}^d\left)\right\}$  is dense in $\mathcal{ℬ}$  , i.e. $S_0\left({\mathbb{R}}^d\right)$  is a full Hilbert $\mathcal{ℬ}$  -module.
• (3) For all $f\in S_0\left({\mathbb{R}}^d\right)$  and $A\in \mathcal{A}$  , we have $$\langle fA,fA{\rangle }_{\mathcal{A}}\le \parallel A{\parallel }^2\langle f,f{\rangle }_{\mathcal{A}},$$  i.e. boundedness of the right action.
• (4) For all $f\in S_0\left({\mathbb{R}}^d\right)$  and $B\in \mathcal{ℬ}$  , we have $$\langle Bf,Bf{\rangle }_{\mathcal{ℬ}}\le \parallel B{\parallel }^2\langle f,f{\rangle }_{\mathcal{ℬ}},$$  i.e. boundedness of the left action.

implies that $S_0\left({\mathbb{R}}^d\right)$  is an equivalence bimodule $(\mathcal{A},\mathcal{ℬ})$  with norm $\parallel f\parallel :=\langle f,f{\rangle }_{\mathcal{A}}^{1/2}$  .

• Proof. Our proof follows Rieffel's approach, see [Rief88] .
  • (1) The linear span of the range of $\langle .,.{\rangle }_{\mathcal{A}}$  is an ideal in $\mathcal{A}$  .
    Then the norm closure $I$  of this linear span is an ideal in $\mathcal{A}$  .

    Furthermore $I$  is invariant under modulation and because $\pi \left(\lambda \right)$  is a faithful representation of $\mathcal{A}$  , we get the desired conclusion.
  • (2) By similar arguments as in $\left(1\right)$  .
  • (3) It suffices to verify the inequality for a dense subspace of $L^2\left({\mathbb{R}}^d\right)$  . Let $h\in S_0\left({\mathbb{R}}^d\right)$  and $A\in C^{*}(\Lambda ,\beta )$  , then

    $$\begin{array}{ccc}\langle h\langle Af,Af{\rangle }_{\mathcal{A}},h\rangle & =& \langle \langle h,Af{\rangle }_{\mathcal{ℬ}}Af,h\rangle \end{array}$$

    $$\begin{array}{ccc}& =& \langle Af,\langle Af,h{\rangle }_{\mathcal{ℬ}}h\rangle \end{array}$$

    $$\begin{array}{ccc}& =& \langle Af,Af\langle h,h{\rangle }_{\mathcal{A}}\rangle \end{array}$$

    $$\begin{array}{ccc}& =& \langle A(f\langle h,h{\rangle }_{\mathcal{A}}{)}^{1/2},A(f\langle h,h{\rangle }_{\mathcal{A}}{)}^{1/2}\rangle \end{array}$$

    $$\begin{array}{ccc}& \le & \parallel A{\parallel }^2\langle f,f\langle h,h{\rangle }_{\mathcal{A}}\rangle \end{array}$$

    $$\begin{array}{ccc}& =& \parallel A{\parallel }^2\langle h\langle f,f{\rangle }_{\mathcal{A}},h\rangle \end{array}$$

    holds for all $f$  in $S_0\left({\mathbb{R}}^d\right)$  . A standard density argument yields the desired result.
  • (4) By similar arguments as in $\left(3\right)$  .
  □

Corollary 5.8. $C^{*}(\Lambda ,\beta )$  and $C^{*}({\Lambda }^0,\overline{\beta })$  are strongly Morita equivalent.

By definition a projective module $V$  is isomorphic to a direct summand of a free module ${\mathcal{A}}^n$  with standard basis $\left\{e_j\right\}$  , i.e. there is a self-adjoint $n\times n$  -matrix $P$  with entries in $\mathcal{A}$  which is a projection, such that $V=P{\mathcal{A}}^n$  . Rieffel proved that if $\mathcal{A}$  and $\mathcal{ℬ}$  are unital $C^{*}$  -algebras and if $V$  is a $(\mathcal{ℬ},\mathcal{A})$  -equivalence bimodule, then $V$  is a projective right $\mathcal{ℬ}$  -module, and a projective left $\mathcal{A}$  -module. Furthermore, $\mathcal{A}$  is equivalent to the $C^{*}$  -algebra $\mathcal{K}(V,\mathcal{ℬ})$  of compact Hilbert $\mathcal{ℬ}$  -module operators.

In particular, let $\mathcal{ℬ}=C^{*}({\Lambda }^0,\overline{\beta })$  and let $V$  denote the right $\mathcal{A}$  -module obtained by completing $S_0\left({\mathbb{R}}^d\right)$  as described earlier. Then, we have:

Theorem 5.9. Feichtinger's algebra $S_0\left({\mathbb{R}}^d\right)$  is a finitely generated projective $\mathcal{ℬ}$  -module and $\mathcal{K}\left(S_0\right({\mathbb{R}}^d),\mathcal{ℬ})$  is equivalent to $C^{*}(\Lambda ,\beta )$  .

In [Rief81a] Rieffel made the observation that finitely generated projective $C^{*}$  -modules possess a reconstruction formula in terms of a tight module frame, which is a generalization of the familiar notion of tight frames for Hilbert spaces. In a subsequent paper we discuss the connection between tight module frames for $C^{*}(\Lambda ,\beta )$  and the characterization of $S_0\left({\mathbb{R}}^d\right)$  with multi-window Gabor frames.

We finish this section with a connection between Connes' construction of projective $\mathcal{A}$  -modules in [Con81] and the recent result of Gröchenig and Leinert on Wiener's lemma for twisted convolution, [GL04] . Let ${\mathcal{A}}_0$  be the subalgebra of absolutely convergent series of $\mathcal{A}$  , i.e. $${\mathcal{A}}_{\Lambda }^1=\{A\in L^2({\mathbb{R}}^d):A={\sum }_{\lambda \in \Lambda }a(\lambda \left)\pi \right(\lambda ),\mathbf{a}\in S_0(\Lambda \left)\right\}$$  with norm $\parallel A{\parallel }_{{\mathcal{A}}_{\Lambda }}^1=\parallel \mathbf{a}{\parallel }_1$  . Then the main result of Gröchenig/Leinert is the following noncommutative version of Wiener's lemma.

Theorem 5.10. If $A\in {\mathcal{A}}_0$  is invertible in $C^{*}(\Lambda ,\beta )$  , then $A^{-1}\in {\mathcal{A}}_0$  .

We present an important result by Connes, which due to the complexity of our assumptions is a little bit lengthy.

Theorem 5.11 (Connes). Let $\mathcal{A}$  be a unital $C^{*}$  -algebra, let $V$  be a projective right $\mathcal{A}$  -module with $\mathcal{A}$  -valued inner product, and let $B=\mathcal{K}(\mathcal{A},V)$  . Then $\mathcal{ℬ}$  is a $C^{*}$  -algebra and $V$  has a corresponding $\mathcal{ℬ}$  -valued innerproduct. Let ${\mathcal{A}}_0$  and ${\mathcal{ℬ}}_0$  be dense $*$  -subalgebras of $\mathcal{A}$  and $\mathcal{ℬ}$  respectively containing the identity elements, and let $V_0$  be a dense subspace of $V$  which is closed under the actions of ${\mathcal{A}}_0$  and ${\mathcal{ℬ}}_0$  , and such that the restrictions to $V_0$  of the innerproduct have values in ${\mathcal{A}}_0$  and ${\mathcal{ℬ}}_0$  , respectively. If ${\mathcal{ℬ}}_0$  has the property that any element of ${\mathcal{ℬ}}_0$  which is invertible in $\mathcal{ℬ}$  has its inverse in ${\mathcal{ℬ}}_0$  , then $V_0$  is a finitely generated projective right ${\mathcal{A}}_0$  -module.

We now apply this construction to the above given equivalence bimodule $S_0\left({\mathbb{R}}^d\right)$  and the lattice $\Lambda$  in ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  and $A=C^{*}({\Lambda }^0,\overline{\beta })$  and ${\mathcal{ℬ}}_0=\{B\in L^2({\mathbb{R}}^d):B={\sum }_{{\lambda }^0\in {\Lambda }^0}b({\lambda }^0\left)\pi \right({\lambda }^0),\mathbf{b}\in S_0({\Lambda }^0\left)\right\}$  then

Theorem 5.12. $S_0\left({\mathbb{R}}^d\right)$  is a finitely generated projective right ${\mathcal{A}}_0$  -module.

For the Schwartz space $\mathcal{S}\left({\mathbb{R}}^d\right)$  Connes called this a beautiful result and of great importance in noncommutative analysis.

6 Application to Gabor Analysis: Biorthogonality Relation of Wexler-Raz

Recently, Gabor frames have been applied in various fields of mathematics, electrical engineering and signal analysis, see [FS98, FS03] . In this section we give a first glimpse of the usefulness of Rieffel's work on strong Morita equivalence of $C^{*}$  -algebras generated by time-frequency shifts.

Let $\Lambda$  be a lattice in ${\mathbb{R}}^d\times {\widehat{\mathbb{R}}}^d$  and $g\in L^2\left({\mathbb{R}}^d\right)$  then a Gabor system $\mathcal{G}(g,\Lambda ):=\left\{\pi \right(\lambda )g:\lambda \in \Lambda \}$  for a Gabor atom $g\in L^2\left({\mathbb{R}}^d\right)$  is a Gabor frame if there are finite positive reals $A,B$  such that $$A\parallel f{\parallel }^2\le {\sum }_{\lambda \in \Lambda }|\langle f,\pi (\lambda )g\rangle {|}^2\le B\parallel f{\parallel }^2,\text{for all}f\in L^2({\mathbb{R}}^d).$$  This is equivalent to invertibility and boundedness of the Gabor frame operator $$S_{g,\Lambda }f={\sum }_{\lambda \in \Lambda }\langle f,\pi \left(\lambda \right)g\rangle \pi \left(\lambda \right)g,\text{for all}f\in L^2\left({\mathbb{R}}^d\right).$$  As a consequence of the invertibility of $S_{g,\Lambda }$  we have the following reconstruction formulas for $f\in L^2\left({\mathbb{R}}^d\right)$ 

$$\begin{array}{c}f=(S_{g,\Lambda }{)}^{-1}{S}_{g,\Lambda }f={\sum }_{\lambda \in \Lambda }\langle f,\pi (\lambda )g\rangle \pi (\lambda \left)\right(S_{g,\Lambda }{)}^{-1}g\end{array}$$

(17)

or $$f=S_{g,\Lambda }(S_{g,\Lambda }{)}^{-1}f={\sum }_{\lambda \in \Lambda }\langle f,\pi (\lambda \left)\right(S_{g,\Lambda }{)}^{-1}g\rangle \pi \left(\lambda \right)g.$$  The coefficients in reconstruction formulas  17 and  6 are not unique, because in general time-frequency shifts $\pi \left(\lambda \right)$  and $\pi \left({\lambda }^{\prime }\right)$  are not linearly independent for $\lambda ,{\lambda }^{\prime }\in \Lambda$  . Therefore, many researchers have investigated the set of all possible dual windows $\gamma$  such that $S_{g,\gamma }=I$  . Of special importance is the function ${\gamma }_0:=(S_{g,\Lambda }{)}^{-1}g$  , the canonical dual window.

There are many characterizations of ${\gamma }_0$  in the set of all dual windows.

The Gabor frame operator $S_{g,\Lambda }$  commutes with time-frequency shifts $\left\{\pi \right(\lambda ):\lambda \in \Lambda \}$  , therefore, the dual Gabor frame $\left\{\pi \right(\lambda ){\gamma }_0:\lambda \in \Lambda \}$  has the structure of a Gabor frame. This observation and ( 17 ) for $(S_{g,\Lambda }{)}^{-1}f$  yields that the inverse frame operator of a frame $\mathcal{G}(\Lambda ,g)$  is given by

$$\begin{array}{c}(S_{g,\Lambda }{)}^{-1}f=S_{{\gamma }_0,\Lambda }f={\sum }_{\lambda \in \Lambda }\langle f,\pi (\lambda ){\gamma }_0\rangle \pi (\lambda ){\gamma }_0.\end{array}$$

(18)

Gröchenig and Leinert were motivated by a practical question on the quality of the canonical dual window $S_{g,\Lambda }^{-1}g$  of a Gabor frame $\mathcal{G}(g,\Lambda )$  generated by a Gabor atom $g$  in $S_0\left({\mathbb{R}}^d\right)$  . They established that Feichtinger's algebra is a good class of Gabor atoms. Namely,

Theorem 6.1. Let $\mathcal{G}(g,\Lambda )$  be a Gabor frame generated by $g\in S_0\left({\mathbb{R}}^d\right)$  then the canonical dual window ${\gamma }_0=S_{g,\Lambda }^{-1}g$  is in $S_0\left({\mathbb{R}}^d\right)$  .

• Proof. Let $g\in S_0\left({\mathbb{R}}^d\right)$  then by assumption $S_{g,\Lambda }$  is invertible in $C^{*}(\Lambda ,\beta )$  , but Theorem  5.10 implies that $S_{g,\Lambda }^{-1}$  is an element of ${\mathcal{A}}_{\Lambda }^1$  , i.e. ${\gamma }_0\in S_0\left({\mathbb{R}}^d\right)$  . □

In [Jan95] Janssen had proved that for a Gabor frame $\mathcal{G}(g,\alpha \mathbb{Z}\times \beta \mathbb{Z})$  generated by a Schwartz function $g$  the canonical dual window is also a Schwartz function under the restriction that $\alpha ,\beta \in \mathbb{Q}$  . Janssen had conjectured that his result is also valid for irrational lattice constants $\alpha ,\beta$  . We remark that a resolution of Janssen's conjecture is a corollary of Connes result that $\mathcal{S}\left({\mathbb{R}}^d\right)$  is closed under holomorphic functional calculus, [Con81] and [Rief88] .

In [WR90] Wexler/Raz characterized the set of all dual atoms with the structure of a Gabor frame for Gabor expansions on finite abelian groups. Their work had been extended to the continuous setting independently by Daubechies, H.L. Landau and Z. Landau in [DLL90] , by Janssen in [Jan95] and by Ron and Shen [RS93] ,[RS97] . In the work on this problem the so-called Janssen representation of a Gabor frame operator was introduced in [Jan95] . Also Feichtinger and Zimmermann considered this topic and found the minimal assumptions for the validity of Wexler-Raz's biorthogonality relation and Janssen's representation, [FZ98] . In [FK98] and [FZ98] Feichtinger and his collaborators introduced the notion of the adjoint lattice for elementary locally compact abelian groups, which Rieffel already used in his construction of equivalence bimodules between noncommutative tori, [Rief88] . In this section we derive the result of Wexler-Raz from the Morita equivalence of ${\mathbb{C}}^{*}(\Lambda ,\beta )$  and $C^{*}({\Lambda }^0,\overline{\beta })$  and the relation between the canonical traces ${\tau }_{\mathcal{A}}$  and ${\tau }_{\mathcal{ℬ}}$  , respectively.

One of the early successes of operator algebras was the classification of all commutative $C^{*}$  -algebras by Gelfand as the involutive complex-valued continuous functions over a compact space. Riesz's representation theorem for positive linear functionals of involutive complex-valued continuous functions over a compact space $X$  yields to an extension of the Lebesgue integral. Therefore, integration of continuous functions over a compact space is considered as a trace on a commutative $C^{*}$  -algebra.

Therefore, traces on general $C^{*}$  -algebras are the natural framework for non-commutative Radon measure theory.

The existence of canonical traces on $\mathcal{A}=C^{*}(\Lambda ,\beta )$  and $\mathcal{ℬ}=C^{*}({\Lambda }^0,\overline{\beta })$  is one of the pleasant properties of noncommutative tori.

First we recall that a trace ${\tau }_{\mathcal{C}}$  on a $C^{*}$  -algebra $\mathcal{C}$  is a linear functional satisfying

$$\begin{array}{ccc}& & \tau \left(I\right)=1,\text{for the identity operator}I\text{of}\mathcal{C},\end{array}$$

$$\begin{array}{ccc}& & \tau \left(AB\right)=\tau \left(BA\right),\text{for all}A,B\in \mathcal{C},\end{array}$$

$$\begin{array}{ccc}& & \tau \left(A^{*}A\right)>0\text{for all nonzero}A\text{in}\mathcal{C}.\end{array}$$

In the case of $\mathcal{A}$  a normalized trace ${\tau }_{\mathcal{A}}$  is given by $${\tau }_{\mathcal{A}}(\langle f,g{\rangle }_{\mathcal{A}})=\langle f,g\rangle ,f,g\in S_0\left({\mathbb{R}}^d\right),$$  and for $\mathcal{ℬ}$  the canonical trace ${\tau }_{\mathcal{ℬ}}$  is normalized by $${\tau }_{\mathcal{ℬ}}(\langle f,g{\rangle }_{\mathcal{ℬ}})=|\Lambda {|}^{-1}\langle f,g\rangle ,f,g\in S_0({\mathbb{R}}^d),$$  which follows from Morita equivalence of $\mathcal{A}$  and $\mathcal{ℬ}$  . This fact can be considered as a noncommutative Poisson summation formula

$$\begin{array}{c}{\tau }_{\mathcal{A}}(\langle f,g{\rangle }_{\mathcal{A}})=|\Lambda {|}^{-1}{\tau }_{\mathcal{ℬ}}(\langle f,g{\rangle }_{\mathcal{ℬ}}).\end{array}$$

(19)

Our restriction in the following theorem to $g,\gamma \in S_0(\mathbb{R})$  is just for convenience. We refer to Gröchenig's excellent survey [Gr01] of Gabor analysis for the general case of $g,\gamma \in L^2\left({\mathbb{R}}^d\right)$  .

Theorem 6.2 (Wexler-Raz). The following conditions are equivalent:

• (1) $S_{g,\gamma }=I$  is a normalized tight frame for $L^2\left({\mathbb{R}}^d\right)$  .
• (2) $|\Lambda {|}^{-1}\langle \gamma ,\pi ({\lambda }^0)g\rangle ={\delta }_{\lambda ,0}$  .

• Proof. $\left(2\right)\Rightarrow \left(1\right):$  Follows from the fact that the identity of $\mathcal{A}$  is $I={\delta }_{\lambda ,0}$  , where ${\delta }_{i,k}$  is the Kronecker delta. Therefore, by assumption $${\tau }_{\mathcal{ℬ}}\left(S_{g,\gamma }\right)={\tau }_{\mathcal{ℬ}}\left(I\right)={\delta }_{\lambda ,0}$$  and by application of  19 we get $${\tau }_{\mathcal{ℬ}}\left(I\right)=|\Lambda {|}^{-1}{\tau }_{\mathcal{A}}(\langle g,\gamma \rangle )=|\Lambda {|}^{-1}\langle \gamma ,g\rangle .$$  The implication $\left(1\right)\Rightarrow \left(2\right)$  is trivial in the light of Rieffel's associativity condition. □

Corollary 6.3. For $g,\gamma \in S_0\left({\mathbb{R}}^d\right)$  dual functions the two Gabor systems $\mathcal{G}(g,{\Lambda }^0)$  and $\mathcal{G}(\gamma ,{\Lambda }^0)$  are biorthogonal to each other on $L^2\left({\mathbb{R}}^d\right)$  .

The proof is an elementary reformulation of Theorem  6.2 .

Corollary 6.4. A Gabor system $\mathcal{G}(g,\Lambda )$  is a tight frame if and only if $\mathcal{G}(\gamma ,{\Lambda }^0)$  is an orthonormal system with frame bound $A=\parallel \Lambda {\parallel }^{-1}\parallel g{\parallel }^2$  .

The statement is well-known, see [Gr01] for the elementary proof.

Corollary 6.5. Let $g_1,...,g_n,{\gamma }_1,...,{\gamma }_n\in S_0\left({\mathbb{R}}^d\right)$  , then for the multi-window Gabor frame $S={\sum }_{i=1}^n{S}_{g_i,{\gamma }_i}$  the following are equivalent:

• (1) $S_{g_1,{\gamma }_1}+\cdots +S_{g_1,{\gamma }_1}=I$  .
• (2) $|\Lambda {|}^{-1}\left(\langle {\gamma }_1,\pi \left({\lambda }^0\right)g_1\rangle +\cdots +\langle {\gamma }_1,\pi \left({\lambda }^0\right)g_1\rangle \right)={\delta }_{\lambda ,0}$  .

The proof follows the same reasoning as for a single Gabor frame.

7 Conclusions

In the last decade operator algebra techniques have been of minor interest in Gabor analysis. But in [DLL90] ,[Jan95] and [GL04] deep results about Gabor frames were obtained with the help of operator algebras.

We included our approach to the Wexler-Raz biorthogonality principle as an indication for the usefulness of Morita equivalence in Gabor analysis. In the following we list some topics, where our approach gives new insight, too.

• (1) The original motivation for our study of Rieffel's results about Morita equivalence was the density result. There are different approaches to this important theorem [DLL90] ,[FK98] and [Be04] , which at the first sight seem unrelated. In [Lu04a] we show that all these approaches cover different aspects of Morita equivalence between $C^{*}$  -algebras generated by time-frequency shifts with respect to a lattice in the time-frequency plane.
• (2) Our interpretation of Rieffel's construction of equivalencebimodules for noncommutative tori in the notions of Gabor analysis enables us to answer the question posed by Manin on the connection between his quantum theta functions and the quantum theta vectors of Schwarz, see [Lu04b] .

Other applications of Rieffel' setting yield new results on Feichtinger' conjecture and on the sturcture of multi-window Gabor frames, which is part of our current research.

Acknowledgement: This investigations are part of the authors Ph.D. thesis under the supervision of H.G. Feichtinger, whom I want to thank for many helpful discussions and remarks on the content of the present paper. Finally, I want to express my gratitude to E. Matusiak and B. Scharinger for their careful reading of the manuscript.

References

1. I. Bekka, Square integrable representations, von Neumann algebras and an application to Gabor analysis, J. Four. Anal. Appl., 10(4):325-349, 2004.
2. H. Bursztyn and A. Weinstein, Poisson Geometry and Morita equivalence, arXiv:math. SG/0402347 v2, 2004.
3. P. Casazza, M. Lammers Bracket Products for Weyl-Heisenberg frames, in [FS03], 71-98, 2003.
4. A. Connes, $C^{◆}$  -algebras et gèomètrie différentielle, C.R. Ac. Sci. Paris, t.290:599-604, 1980.
5. I. Daubechies, The wavelet transform, time-frequency localization and signal analysis, IEEE Trans. Inform.Theory, 35:961-1005, 1990.
6. I. Daubechies, Models for the irreducible representations of a Heisenberg group, Infin. Dimens. Anal. Quantum Probab. Relat. Top., 7(4):527546, 2004.
7. I. Daubechies, H.J. Landau, Z. Landau, Gabor time-frequency lattices and the Wexler-Raz idenity,J. Four. Anal. Appl., 1(4):437-478, 1995.
8. H.G. Feichtinger, On a New Segal Algebra, Monatsh. Math., 92:269-289, 1981.
9. H.G. Feichtinger, K. Gröchenig, Banach spaces related to integrable group representations and their atomic decompositions. I., J. Funct. Anal., 86(2):307-340, 1989.
10. H.G. Feichtinger, K. Gröchenig, Gabor frames and time-frequency analysis og distributions, J. Funct. Anal., 146(2):464-495, 1997.
11. H.G. Feichtinger, K. Gröchenig, Banach spaces related to integrable group representations and their atomic decompositions. II., Monatsh. Math., 108(2-3): 129-148, 1989.
12. H.G. Feichtinger and W. Kozek, Quantization of TF–lattice invariant operators on elementary LCA groups, in [FS98], pp. 233–266.
13. H.G. Feichinger and T. Strohmer, Gabor Analysis and Algorithms: Theory and Applications, Birkhäuser, Boston, 1998.
14. H.G. Feichinger and T. Strohmer, Advances in Gabor Analysis, Birkhäuser, Boston, 2003.
15. H.G. Feichtinger. Modulation spaces of locally compact Abelian groups. In R.Radha, editor, Proc. Internat. Conf. on Wavelets and Applications, pages 1–56, Chennai, January 2002, 2003.
16. H.G. Feichinger and T. Strohmer, Advances in Gabor Analysis, Birkhäuser, Boston, 2003.
17. H.G. Feichtinger and G. Zimmermann, A Banach space of test functions for Gabor analysis, in [FS98], pp.123–170.
18. D. Gabor, Theory of communication, J.IEEE(London), 93(III):429-457, 1946.
19. K. Gröchenig, Foundations of Time-Frequency Analysis, Birkhäuser, Boston, 2001.
20. K. Gröchenig and M. Leinert, Wiener's lemma for twisted convolution and Gabor frames, J. Amer. Math., 1:1-17, 2004.
21. A.J.E. Janssen, Duality and biorthogonality for Weyl-Heisenberg frames, J. Four. Anal. Appl., 1(4):403-436, 1995.
22. I. Kaplansky, Modules over operator algebras, Trans. Amer. Math. Soc., 75:839-858, 1953.
23. F. Luef, The density theorm in Gabor analysis and Morita equivalence, in preparation.
24. F. Luef, On quantum theta functions and quantum theta vectors, in preparation.
25. K. Morita, Duality for modules and its application to the theory of rings with minimum condition, Sci. Rep. Tokyo Kyoiku Daigaku Sect. A 6:83-142, 1958.
26. J. Packer and M.A. Rieffel, Wavelet filter functions, the matrix completion problem, and projective modules over $C\left({\mathbb{T}}^n\right)$  , J. Four. Anal. Appl., 9:2 , 101 116, 2003.
27. J. Packer and M.A. Rieffel, Projective Multi-Resolution Analysis for $L^2\left({\mathbb{R}}^2\right)$  , J. Four. Anal. Appl., 10:5 , 439 464, 2004.
28. W. L. Paschke, Inner product modules over $B^{*}$  -algebras, Trans. Amer. Math. Soc. 182: 443-468 (1973).
29. H. Reiter, Metaplectic groups and Segal algebras, Lect. Notes Math., 1382, Berlin, Springer Verlag 1989.
30. M.A. Rieffel, Morita equivalence for $C^{*}$  -algebras and $W^{*}$  -algebras, J. Pure Appl. Alg., 5: 51-96, 1974.
31. M.A. Rieffel, Induced representations of $C^{*}$  -algebras, Adv. Math., 13:176-257, 1974.
32. M.A. Rieffel, Strong Morita equivalence of certain transformation group $C^{◆}$  -algebras, Math. Annalen, 222:7-23, 1976.
33. M.A. Rieffel, $C^{*}$  -algebras associated with irrational rotations, Pac. J. Math., 4:415-429, 1981.
34. M.A. Rieffel, Von Neumann algebras associated with pairs of lattices in Lie groups, Math. Ann., 257(4):403-418, 1981.
35. M.A. Rieffel, Morita equivalence for operator algebras, Proc. Symp. Pure Math.,93:285-298, 1982.
36. M.A. Rieffel, Projective modules over higher-dimensional noncommutative tori, Can. J. Math., 40:257-388, 1988.
37. A. Ron, Z. Shen, Weyl-Heisenberg frames and stable bases, Talk at Oberwolfach conference on Approx. Theory,1993.
38. A. Ron, Z. Shen, Weyl-Heisenberg frames and Riesz bases in $L^2\left({\mathbb{R}}^d\right)$  , Duke Math. J., 89(2):237-282, 1997.
39. M.A. Rieffel, A. Schwarz, Morita equivalence of multidimensional noncommutative tori, Int. J. Math., 10:289-299, 1999.
40. R. Tolmieri, R. Orr, Poisson Summation, the ambiguity function and the theory of Weyl-Heisenberg frames, J. Four. Anal. Appl., 1(3):233-247, 1995.
41. X. Tang,A. Weinstein, Quantization and Morita equivalence for constant Dirac structures on tori, e-print, arXiv:math.QA/0305413, 2003.
42. J. Wexler, S. Raz, Discrete Gabor Expansions, Signal Processing, 21(3):207-221, 1990.
43. P.J. Wood, Wavelets and Hilbert Modules, J. Four. Anal. Appl., 10:6 , 573 598, 2004.

Fakultat fur Mathematik, Nordbergstrasse 15, 1090 Wien, Austria E-mail address : Franz.Luef@univie.ac.at