Gabor frames with no Gabor duals

Abstract.

Let $H$ be an infinite-dimensional Hilbert space and let $\{x_{n}\}_{n\in\mathbb{N}}$ be a frame for $H$. We say that a sequence $\{y_{n}\}_{n\in\mathbb{N}}$ is an alternative dual of $\{x_{n}\}_{n\in\mathbb{N}}$ if

$$x=\sum_{n=1}^{\infty}\langle x,y_{n}\rangle x_{n}\quad\text{for all }x\in H,$$

with the convergence of the series in the norm of $H.$ We say that a sequence $\{z_{n}\}_{n\in\mathbb{N}}$ is a pseudo-synthesis dual of $\{x_{n}\}_{n\in\mathbb{N}}$ if

$$x=\sum_{n=1}^{\infty}\langle x,x_{n}\rangle z_{n}\quad\text{for all }x\in H,$$

with the convergence of the series in the norm of $H.$ We prove that, at every level of smoothness of the generator, there exist Gabor frames for $L^{2}(\mathbb{R}^{d})$ with infinitely many alternative duals (and pseudo-synthesis duals) such that none of them is a Gabor system.

1. introduction

Let $g\in L^{2}(\mathbb{R}^{d})$ be a nonzero function and let $\Lambda=\{(a_{n},b_{n})\}_{n\in\mathbb{N}}$ be a subset of $\mathbb{R}^{2d}.$ The Gabor system associated with $\Lambda$ and $g$, denoted by ${\mathcal{G}}(g,\Lambda)$, is the sequence $\{M_{b_{n}}T_{a_{n}}g\}_{n\in\mathbb{N}}$ consisting of discrete set of time-frequency shifts of $g$. Here for fixed $x,\xi\in\mathbb{R}$, $T_{x}$ is the translation operator defined by

$$T_{x}\colon L^{2}(\mathbb{R}^{d})\rightarrow L^{2}(\mathbb{R}^{d}),\quad(T_{x}g)(t)\coloneq g(t-x),$$

and $M_{\xi}$ is the modulation operator defined by

$$M_{\xi}\colon L^{2}(\mathbb{R}^{d})\rightarrow L^{2}(\mathbb{R}^{d}),\quad(M_{\xi}g)(t)\coloneq e^{2\pi i\xi t}g(t).$$

We say that ${\mathcal{G}}(g,\Lambda)$ is a Gabor frame if there exist some positive constants $A\leq B$, called frame bounds, such that the frame inequality

(1.1) $$A\,\|f\|^{2}\,\leq\,\sum_{n\in\mathbb{N}}\bigl|\langle f,M_{b_{n}}T_{a_{n}}g\rangle\bigr|^{2}\,\leq\,B\,\|f\|^{2},$$

is satisfied for all $f\in L^{2}(\mathbb{R}^{d})$. It is known that for every Gabor frame there exists at least one dual sequence $\{y_{n}\}_{n\in\mathbb{N}}\subseteq L^{2}(\mathbb{R}^{d})$ such that

(1.2) $$f\,=\,\sum_{n\in\mathbb{N}}\langle f,y_{n}\rangle M_{b_{n}}T_{a_{n}}g$$

and

(1.3) $$f\,=\,\sum_{n\in\mathbb{N}}\langle f,M_{b_{n}}T_{a_{n}}g\rangle y_{n},$$

for all $f\in L^{2}(\mathbb{R}^{d})$ with the convergence of the series in the norm of $L^{2}(\mathbb{R}^{d}).$ However, if a Gabor frame has unbounded amount of redundant elements, then there exists at least one sequence $\{y_{n}\}_{n\in\mathbb{N}}\subseteq H$ such that Equation (1.2) holds for all $f\in L^{2}(\mathbb{R}^{d})$, whereas Equation (1.3) fails for at least some $f_{0}\in L^{2}(\mathbb{R}^{d}).$ Likewise, there exists at least one sequence $\{z_{n}\}_{n\in\mathbb{N}}\subseteq H$ such that Equation (1.3) holds for all $f\in L^{2}(\mathbb{R}^{d})$, but Equation (1.2) fails for at least some $f_{0}\in L^{2}(\mathbb{R}^{d})$ (See [7, Corollary 3.1]). Here we say that an element of a frame is redundant if removing that element does not destroy the completeness of the frame. Consequently, the notions arising from Equation (1.2) and (1.3) do not coincide in general. Following the terminology in [7], we say that a sequence is an alternative dual of ${\mathcal{G}}(g,\Lambda)$ is it satisfies Equation (1.2) for all $f\in L^{2}(\mathbb{R}^{d})$ and a pseudo-synthesis dual of ${\mathcal{G}}(g,\Lambda)$ if it satisfies Equation (1.3) for all $f\in L^{2}(\mathbb{R}^{d})$.

The form and properties of an alternative dual (respectively pseudo-synthesis dual) has a significant impact on the convergence of Equation (1.2) (respectively Equation (1.3)). In fact, one can even infer properties of a Gabor frame from properties of its alternative duals (or pseudo-synthesis duals). For references along this direction, we refer the reader to [9],[1],[10] and [7]. As a result, determining the conditions under which an alternative dual (or pseudo-synthesis dual) inherits the structure or properties of its associated Gabor frame has long been a central question. However, whenever a Gabor frame contains even a single redundant element, the collection of its associated dual sequences expands from a singleton to an infinite family ([2, Theorem 6.3.7]), which makes this type of problem a nontrivial task. In [7], Heil and the author resolved several questions in this direction. For example, they showed that a Gabor frame possesses finitely many redundant elements if and only if every associated alternative dual (and pseudo-synthesis dual) is itself frame. They also established that every Gabor frame has the exact same number of redundant elements as each of its alternative duals (and pseudo-synthesis duals). Another closely related open problem is to determine under what conditions a dual sequence associated with a Gabor frame must itself be a Gabor system. Using perturbation techniques ([2, Theorem 6.3.7]), it is not too hard to construct a Gabor frame that admits an alternative dual (and a pseudo-synthesis dual) that fails to be a Gabor system. This settles the case in which the Gabor frame has a unique dual sequence. It is therefore natural to ask whether there exists a Gabor frame with infinitely many alternative duals (and pseudo-synthesis duals) such that none of them is a Gabor system. In this paper, we explicitly construct a Gabor frame, containing redundant elements, for which neither its alternative dual nor its pseudo-synthesis dual is a Gabor system. We remark that in the case where $\Lambda=a\mathbb{Z}\times b\mathbb{Z}$ for some $a,b>0$, it is known that ${\mathcal{G}}(g,\Lambda)$ admits at least one alternative dual that is also a pseudo-synthesis dual and forms a Gabor frame.

2. Main Result

We begin with two lemmas.

Lemma 2.1.

Fix $a>0$ and fix any irrational real number $\alpha$. For each $n\in\mathbb{Z}$ let

$$c_{n}=\int_{0}^{a}e^{2\pi ia^{-1}(\alpha-n)x}\,dx.$$

Then the following statements hold.

1. There exists some constant $C$ such that $\frac{\text{Im}(c_{n})}{\text{Re}(c_{n})}=C$ for all $n\in\mathbb{Z}.$

2. $\lim_{n\rightarrow\infty}\frac{|c_{n}|^{2}}{\text{Re}(c_{n})}=0.$

Proof.

(a) For each $n\in\mathbb{Z}$ we compute

(2.1) $$\begin{split}c_{n}\,=\,\frac{a}{2\pi i(\alpha-n)}(e^{2\pi i\alpha}-1)\,=\,\frac{ae^{i\pi\alpha}}{2\pi i(\alpha-n)}\bigl(2i\sin(\alpha\pi)\bigr)\,=\,\frac{ae^{i\pi\alpha}\sin(\alpha\pi)}{\pi(\alpha-n)}.\end{split}$$

Therefore, we have $\text{Re}(c_{n})\,=\,\frac{a\cos(\alpha\pi)\sin(\alpha\pi)}{\pi(\alpha-n)}$ and $\text{Im}(c_{n})\,=\,\frac{a\sin^{2}(\alpha\pi)}{\pi(\alpha-n)}.$ It then follows that

$$\frac{\text{Im}(c_{n})}{\text{Re}(c_{n})}\,=\,\tan(\alpha\pi)\quad\text{for all }n\in\mathbb{Z}.$$

(b) Note that since $(c_{n})_{n\in\mathbb{Z}}\in\ell^{2}(\mathbb{Z})$, $c_{n}\rightarrow 0$ as $n\rightarrow\infty.$ It follows that

$$\dfrac{|c_{n}|^{2}}{\text{Re}(c_{n})}\,=\,\dfrac{(\text{Re}(c_{n}))^{2}+(\text{Im}(c_{n}))^{2}}{\text{Re}(c_{n})}\,=\,\text{Re}(c_{n})+\tan(\alpha\pi)\text{Im}(c_{n})\rightarrow 0\penalty\ \text{as }n\rightarrow\infty.$$

∎

Lemma 2.2.

Fix $a>0.$ Let $\theta(x)$ be any real-valued function defined on $[0,a].$ For almost every $x\in[0,a]$ the sequence of functions $\bigl\{\cos\bigl(2\pi a^{-1}kx+\theta(x)\bigr)\bigr\}_{k\in\mathbb{N}}$ does not converge to $0$ as $k\rightarrow\infty.$

Proof.

Let $E=[0,a]\setminus a\mathbb{Q}$. Note that $|E|=a$ and $a^{-1}x\in\mathbb{Q}^{c}$ for all $x\in E.$ Fix $x\in E$. By Kronecker’s approximation theorem (for example, see [8]), $\{ka^{-1}x\penalty\ \text{mod }1\,|\,k\in\mathbb{N}\}$ is dense in $[0,1]$ for every $x\in E.$ Consequently, $\bigl\{\bigl(ka^{-1}x+\frac{\theta(x)}{2\pi}\bigr)\penalty\ \text{mod }1\,|\,k\in\mathbb{N}\bigr\}$ is dense in $[0,1]$ for every $x\in E.$ If $\cos\bigl(2\pi a^{-1}kx+\theta(x)\bigr)$ were to converge to $0$ as $k\rightarrow\infty,$ then for sufficiently large $N$, $\bigl\{\bigl(ka^{-1}x+\frac{\theta(x)}{2\pi}\bigr)\penalty\ \text{mod }1\,|\,k\geq N\bigr\}$ would have to be concentrated on two small neighborhoods of $1/4$ and $3/4,$ which is a contradiction. ∎

We are now ready to prove our main result. For simplicity, we present the construction in the one-dimensional setting. However, higher-dimensional analogue can be obtained by straightforward modifications. Recall that a Gabor frame is a Riesz basis if there exists a unique associated alternative dual that is also a pseudo-synthesis dual such that both Equation (1.2) and (1.3) hold for all $f\in L^{2}(\mathbb{R}).$

Theorem 2.3.

There exist some $g\in C^{\infty}(\mathbb{R})\cap L^{2}(\mathbb{R})$ and some countable subset $\Lambda\subset\mathbb{R}^{2}$ such that the following statements hold.

1. ${\mathcal{G}}(g,\Lambda)$ is a Gabor frame for $L^{2}(\mathbb{R})$,

2. ${\mathcal{G}}(g,\Lambda)$ has infinitely many alternative duals and pseudo-synthesis duals,

3. No alternative dual or pseudo-synthesis dual of ${\mathcal{G}}(g,\Lambda)$ is a Gabor system.

Proof.

Let $a>0$ and let $h\in L^{2}(\mathbb{R})$ be supported on $[0,a]$, satisfying $C\leq|h|^{2}\leq D$ almost everywhere on $[0,a]$ for some positive constants $C,D.$ By Painless Nonorthogonal Expansions ([4, Theorem 1], see also [5, Theorem 11.4]), $\{M_{a^{-1}k}T_{an}h\}_{n,k\in\mathbb{Z}}$ is a Gabor frame for $L^{2}(\mathbb{R})$ with frame bounds $aC$ and $aD.$ Moreover, $\{M_{a^{-1}k}T_{an}h\}_{n,k\in\mathbb{Z}}$ is a Riesz basis with the unique alternative dual, which is also a pseudo-synthesis dual, $\{M_{a^{-1}k}T_{an}h^{-1}\}_{n,k\in\mathbb{Z}}.$ Now let $g=\widehat{h}$ be the Fourier transform of $h$. By taking the Fourier transform, we see that ${\mathcal{G}}(g,a^{-1}\mathbb{Z}\times a\mathbb{Z})$ is a Gabor frame for $L^{2}(\mathbb{R})$. Let $\alpha\in a^{-1}\mathbb{Q}^{c}$ be an arbitrary irrational number and let $\Lambda=(a^{-1}\mathbb{Z}\cup\{\alpha\})\times a\mathbb{Z}$. Clearly, ${\mathcal{G}}(g,\Lambda)$ remains a Gabor frame. Moreover, ${\mathcal{G}}(g,\Lambda)$ satisfies statement (b) by [2, Theorem 6.3.7]. We will show that every alternative dual and every pseudo-synthesis dual of ${\mathcal{G}}(g,\Lambda)$ fails to be a Gabor system. Let $\Lambda^{\prime}=a\mathbb{Z}\times(a^{-1}\mathbb{Z}\cup\{\alpha\}).$ It is then equivalent to show that no alternative dual or pseudo-synthesis dual of ${\mathcal{G}}(h,\Lambda^{\prime})$ can be written in the form $\{T_{\beta}M_{\gamma}\phi\}_{(\beta,\gamma)\in\Gamma}$ for any $\phi\in L^{2}(\mathbb{R})$ and any subset $\Gamma\subseteq\mathbb{R}^{2}$.

Case 1 Alternative Duals$\colon$ Let $\{h_{n,k}\}_{n,k\in\mathbb{Z}}\cup\{h_{\alpha}\}\subseteq L^{2}(\mathbb{R})$ be an alternative dual of ${\mathcal{G}}(h,\Lambda^{\prime})$. That is,

(2.2) $$f\,=\,\sum_{n,k\in\mathbb{Z}}\langle f,h_{n,k}\rangle M_{a^{-1}k}T_{an}h+\langle f,h_{\alpha}\rangle M_{\alpha}h\quad\text{for all }f\in L^{2}(\mathbb{R}),$$

with the convergence of the series in the norm of $L^{2}(\mathbb{R}).$ Fix $n\in\mathbb{Z}\setminus\{0\}$ and let $f\in L^{2}(\mathbb{R})$ be any function that is supported on $[na,(n+1)a]$. Then we have

(2.3) $$f\,=\,P_{n}(f)=P_{n}\bigl(\sum_{n,k\in\mathbb{Z}}\langle f,h_{n,k}\rangle M_{a^{-1}k}T_{an}h+\langle f,h_{\alpha}\rangle M_{\alpha}h\bigr)=\sum_{k\in\mathbb{Z}}\langle f,h_{n,k}\rangle M_{a^{-1}k}T_{an}h,$$

where $P_{n}$ denotes the orthogonal projection onto $L^{2}([na,(n+1)a].$ By the uniqueness of the alternative dual of $\{M_{a^{-1}k}T_{an}h\}_{n,k\in\mathbb{Z}}$, it follows that

(2.4) $$h_{n,k}\,=\,M_{a^{-1}k}T_{an}h\,=\,T_{an}M_{a^{-1}k}h^{-1}\quad\text{a.e. for any }(n,k)\in(\mathbb{Z}\setminus\{0\})\times\mathbb{Z}.$$

For the case $n=0$, since $\alpha\notin a^{-1}\mathbb{Z}$, there exists an infinite sequence of nonzero scalars $(c_{k})_{k\in\mathbb{Z}}\in\ell^{2}(\mathbb{Z})$ such that $M_{\alpha}h=\sum_{k\in\mathbb{Z}}c_{k}M_{a^{-1}k}h$. Then for $f\in L^{2}(\mathbb{R})$ that is supported on $[0,a]$ we have

(2.5) $$\begin{split}f\,=\,\sum_{k\in\mathbb{Z}}\langle f,h_{0,k}\rangle M_{a^{-1}k}h+\langle f,h_{\alpha}\rangle M_{\alpha}h&\,=\,\sum_{k\in\mathbb{Z}}\langle f,h_{0,k}\rangle M_{a^{-1}k}h+\sum_{k\in\mathbb{Z}}\langle f,h_{\alpha}\rangle c_{k}M_{a^{-1}k}h\\ &\,=\,\sum_{k\in\mathbb{Z}}\langle f,h_{0,k}+\overline{c_{k}}h_{\alpha}\rangle M_{a^{-1}k}h.\end{split}$$

Likewise, we have $h_{0,k}+\overline{c_{k}}h_{\alpha}=M_{a^{-1}k}h^{-1}$, and hence

$$h_{0,k}=M_{a^{-1}k}h^{-1}-\overline{c_{k}}h_{\alpha}\quad\text{a.e. for all }k\in\mathbb{Z}.$$

Now, suppose to the contrary that there exist some $\phi\in L^{2}(\mathbb{R})$ and some countable subset $\Gamma=\{(\beta_{n,k},\gamma_{n,k})\}\cup\{(\beta_{\alpha},\gamma_{\alpha})\}$ in $\mathbb{R}^{2}$ such that

(2.6) $$h_{\alpha}\,=\,T_{\beta_{\alpha}}M_{\gamma_{\alpha}}\phi\quad\text{and}\quad h_{n,k}\,=\,T_{\beta_{n,k}}M_{\gamma_{n,k}}\phi\quad\text{for any }n,k\in\mathbb{Z}.$$

Consequently, we have $|h_{n,k}|=|T_{\beta_{n,k}}\phi|$ for any $n,k\in\mathbb{Z}.$ Therefore, the diameters of the supports of $h_{n,k}$ are equal for all $n,k\in\mathbb{Z}$. Moreover, for any $n,k,n_{1},k_{1}\in\mathbb{Z}$, one can always obtain $|h_{n,k}|$ by applying certain translation to $|h_{n_{1},k_{1}}|$. By Equation (2.4), $h_{n,k}$ is supported on $[na,(n+1)a]$ for any $(n,k)\in(\mathbb{Z}\setminus\{0\})\times\mathbb{Z}$. It follows that for any $k\in\mathbb{Z}$

(2.7) $$|h^{-1}|\,=\,|T_{-an}h_{n,k}|=|h_{0,k}|=|M_{a^{-1}k}h^{-1}-\overline{c_{k}}h_{\alpha}|\quad\text{a.e.}$$

By taking the square, we obtain that for any $k\in\mathbb{Z}$

(2.8) $$|h^{-1}|^{2}=|h^{-1}|^{2}-2\text{Re}\bigl(e^{2\pi ia^{-1}kx}h^{-1}c_{k}\overline{h_{\alpha}}\bigr)+|c_{k}h_{\alpha}|^{2}.$$

Equivalently,

(2.9) $$2\text{Re}\bigl(e^{2\pi ia^{-1}kx}h^{-1}\overline{h_{\alpha}}c_{k}\bigr)\,=\,|c_{k}h_{\alpha}|^{2}.$$

Let $\theta_{1}(x)$ and $\theta_{2}(x)$ be real-valued functions such that $h^{-1}=|h^{-1}|e^{i\theta_{1}}$ and $\overline{h_{\alpha}}=|h_{\alpha}|e^{i\theta_{2}}$. We then rewrite the left-hand side of Equation (2.9) as

(2.10) $$\begin{split}2\text{Re}\bigl(e^{2\pi ia^{-1}kx}h^{-1}\overline{h_{\alpha}}c_{k}\bigr)&\,=\,2|h^{-1}||h_{\alpha}|\text{Re}(e^{i(2\pi a^{-1}kx+\theta_{1}+\theta_{2})}c_{k})\\ &\,=\,2|h^{-1}||h_{\alpha}|\bigl(\text{Re}(c_{k})\cos(\psi_{k}(x))-\text{Im}(c_{k})\sin(\psi_{k}(x))\bigr),\end{split}$$

where for each $k\in\mathbb{Z}$ we define $\psi_{k}(x)=2\pi a^{-1}kx+\theta_{1}(x)+\theta_{2}(x).$ Since $|h_{\alpha}(x)|$ is nonzero almost everywhere on $[0,a]$, by Equation (2.9) we further obtain

(2.11) $$\cos(\psi_{k}(x))-\frac{\text{Im}(c_{k})}{\text{Re}(c_{k})}\sin(\psi_{k}(x))=\frac{|c_{k}|^{2}}{2\text{Re}(c_{k})}|h_{\alpha}(x)||h(x)|\quad\text{a.e. on }[0,a].$$

By Lemma 2.1(b), the right-hand side of Equation (2.11) converges to $0$ as $k$ tends to $\infty$ for almost every $x\in[0,a]$. Thus, it remains to show that the left-hand side does not converge to $0$ on a set of positive measure of $[0,a].$ By Lemma 2.1(a), $\frac{\text{Im}(c_{k})}{\text{Re}(c_{k})}=C$ for some nonzero constant $C.$ It then follows that

(2.12) $$\begin{split}\cos(\psi_{k}(x))-\frac{\text{Im}(c_{k})}{\text{Re}(c_{k})}\sin(\psi_{k}(x))&\,=\,\cos(\psi_{k}(x))-C\sin(\psi_{k}(x))\\ &\,=\,\sqrt{1+C^{2}}\cos\bigl(\psi_{k}(x)-\eta\bigr),\end{split}$$

where $\eta=\tan^{-1}(C).$ By Lemma 2.2(b), we see that for almost every $x\in[0,a]$ the left-hand side of Equation (2.11) does not converge to $0$ as $k\rightarrow\infty.$

Case 2 Pseudo-synthesis Duals$\colon$ Let $\{h_{n,k}\}_{n,k\in\mathbb{Z}}\cup\{h_{\alpha}\}\subseteq L^{2}(\mathbb{R})$ be a pseudo-synthesis dual of ${\mathcal{G}}(h,\Lambda^{\prime})$. That is,

(2.13) $$f\,=\,\sum_{n,k\in\mathbb{Z}}\langle f,M_{a^{-1}k}T_{an}h\rangle h_{n,k}+\langle h,M_{\alpha}h\rangle h_{\alpha}\quad\text{for all }f\in L^{2}(\mathbb{R}),$$

with the convergence of the series in the norm of $L^{2}(\mathbb{R}).$ Fix $n\in\mathbb{Z}\setminus\{0\}$ and let $f\in L^{2}(\mathbb{R})$ be any function that is supported on $[na,(n+1)a]$. Then we have

(2.14) $$f\,=\,P_{n}(f)=\bigl(\sum_{n,k\in\mathbb{Z}}\langle P_{n}f,M_{a^{-1}k}T_{an}h\rangle h_{n,k}+\langle P_{n}f,M_{\alpha}h\rangle h_{\alpha}\bigr)=\sum_{k\in\mathbb{Z}}\langle f,M_{a^{-1}k}T_{an}h\rangle h_{n,k}.$$

By the uniqueness of the pseudo-synthesis dual of $\{M_{a^{-1}k}T_{an}h\}_{n,k\in\mathbb{Z}}$, it follows that

$$h_{n,k}\,=\,M_{a^{-1}k}T_{an}h\,=\,T_{an}M_{a^{-1}k}h^{-1}\quad\text{a.e. for any }(n,k)\in(\mathbb{Z}\setminus\{0\})\times\mathbb{Z}.$$

The rest of the proof follows a similar argument to that in Case $1$. ∎

Finally, we close this paper with a few remarks.

Remark 2.4.

(a) The generator of the Gabor frame in Theorem 2.3 is not only smooth, but also extends to an entire function of exponential type at most $2\pi a.$ ([11])

(b) One might suspect that a potential reason why a Gabor frame fails to possess an alternative dual or a pseudo-synthesis dual that is itself a Gabor system is closely related to the “high density” of the associated set of time-frequency shifts. A classical notion that have been frequently used to quantify the density of a set in this context is the Beurling density. The lower and upper Beurling density associated with a countable subset $\Lambda\subseteq\mathbb{R}^{d}$ are defined by

(2.15) $$D^{-}(\Lambda)=\liminf_{r\rightarrow\infty}\inf_{x\in\mathbb{R}^{d}}\frac{\#\bigl(\Lambda\cap B_{r}(x)\bigr)}{|B_{r}(x)|}\quad\text{and}\quad D^{+}(\Lambda)=\limsup_{r\rightarrow\infty}\sup_{x\in\mathbb{R}^{d}}\frac{\#\bigl(\Lambda\cap B_{r}(x)\bigr)}{|B_{r}(x)|},$$

respectively. If $D^{-}(\Lambda)=D^{+}(\Lambda)$, then we say $\Lambda$ has a uniform Beurling density (see [6] for more introductions to Beurling density). A straightforward computation shows that the associated set of time-frequency shifts in Theorem 2.3 has a uniform Beurling density $1.$ On the other hand, it is known that if ${\mathcal{G}}(g,\Lambda)$ forms a frame for $L^{2}(\mathbb{R}^{d})$, then $D^{-}(\Lambda)\geq 1$ ([3]). Thus, the set of time-frequency shifts associated with the Gabor frame constructed in Theorem 2.3 already achieves the minimal possible uniform Beurling density for a Gabor frame in this sense.

(c) The smoothness of the generator is not the critical issue that prevent a Gabor frame from admitting alternative duals or pseudo-synthesis duals that are Gabor systems. Let ${\mathcal{G}}(h,\Lambda^{\prime})$ be the same Gabor frame specified in the proof of Theorem 2.3. Assume that there exists some alternative dual or pseudo-synthesis dual that can be written in the form $\{T_{\beta}M_{\gamma}\phi\}_{(\beta,\gamma)\in\Gamma}$ for some $\phi\in L^{2}(\mathbb{R})$ and some subset $\Gamma\subseteq\mathbb{R}^{2}$. With slight modifications to the proof of Case 1 of Theorem 2.3, one can show that there exists some $f\in L^{2}(\mathbb{R})$ that is compactly supported and still satisfies Theorem 2.3 (a)–(c).