Abstract
We construct a family of frames describing the norm and seminorm of the space \(H^s(\mathbb {R}^d)\). We also characterise Besov spaces modeled on \(L^2(\mathbb {R}^d)\). Our work is inspired by the discrete orthonormal Stockwell transform introduced by R.G. Stockwell, which provides a time-frequency localised version of the Fourier basis of \(L^2([0,1])\). This approach is a hybrid between Gabor and Wavelet frames. We construct explicit and computable examples of these frames, discussing their properties and comparing them with the existing literature.
Similar content being viewed by others
References
Battisti, U., Riba, L.: Window-dependent bases for efficient representations of the Stockwell transform. Appl. Comput. Harmonic Anal. 40(2), 292–320 (2016)
Battisti, U., Berra, M.: Explicit examples of \(L^2\)-frames associated to \(\alpha \)-partitioning. arXiv:1509.01437
Bagher-Ebadian, H., Siddiqui, F., Liu, C., Movsas, B., Chetty, I.J.: On the impact of smoothing and noise on robustness of CT and CBCT radiomics features for patients with head and neck cancers. Med. Phys. 44(5), 1755–1770 (2017)
Berra, M., de Hoop, M.V., Romero, J.L.: A multi-scale Gaussian beam parametrix for the wave equation: the Dirichlet boundary value problem. arXiv:1705.00337
Borup, L., Nielsen, M.: Frame decomposition of decomposition spaces. J. Fourier Anal. Appl. 13(1), 39–70 (2007)
Cabrelli, C., Mosquera, C.A., Paternostro, V.: Linear combinations of frame generators in systems of translates. J. Math. Anal. Appl. 413(2), 776–788 (2014)
Candès, E.J., Donoho, D.L.: New tight frames of curvelets and optimal representations of objects with piecewise \(C^2\) singularities. Commun. Pure Appl. Math. 57(2), 219–266 (2004)
Canuto, C., Tabacco, A.: Multilevel decompositions of functional spaces. J. Fourier Anal. Appl. 3(6), 715–742 (1997)
Christensen, O.: An Introduction to Frames and Riesz Bases. Applied and Numerical Harmonic Analysis. Birkhäuser Boston, Inc., Boston (2016)
Daubechies, I.: Ten Lectures on Wavelets, vol. 61 of CBMS-NSF Regional Conference Series in Applied Mathematics. Society for Industrial and Applied Mathematics (SIAM), Philadelphia (1992)
Drabycz, S., Stockwell, R.G., Mitchell, J.R.: Image texture characterization using the discrete orthonormal S-transform. J. Digit. Imaging 22(6), 696–708 (2008)
Feichtinger, H.G.: Atomic characterizations of modulation spaces through Gabor-type representations. Rocky Mt. J. Math. 19(1), 113–125 (1989). Constructive Function Theory—86 Conference (Edmonton, AB, 1986)
Feichtinger, H.G., Gröbner, P.: Banach spaces of distributions defined by decomposition methods. I. Math. Nachr. 123, 97–120 (1985)
Fornasier, M.: Banach frames for \(\alpha \)-modulation spaces. Appl. Comput. Harmonic Anal. 22(2), 157–175 (2007)
Grobner, P.: Banachraeume Glatter Funktionen und Zerlegungsmethoden. ProQuest LLC, Ann Arbor, MI (1992). Thesis (Dr.natw.)—Technische Universitaet Wien (Austria)
Gröchenig, K.: Foundations of Time-Frequency Analysis. Applied and Numerical Harmonic Analysis. Birkhäuser Boston, Inc., Boston (2001)
Guo, Q., Molahajloo, S., Wong, M.W.: Phases of modified Stockwell transforms and instantaneous frequencies. J. Math. Phys. 51(5), 05210-11 (2010)
Jansen, M.: Non-equispaced B-spline wavelets. Int. J. Wavelets Multiresolut. Inf. Process. 14(6), 1650056 (2016). (35)
Lions, J.-L., Peetre, J.: Sur une classe d’espaces d’interpolation. Inst. Hautes Études Sci. Publ. Math. 19, 5–68 (1964)
Mallat, S.: A Wavelet Tour of Signal Processing. Academic Press, New York (1999)
Ottosen, E.S., Nielsen, M.: A characterization of sparse nonstationary Gabor expansions. J. Fourier Anal. Appl. (2017). https://doi.org/10.1007/s00041-017-9546-6
Pridham, G., Steenwijk, M.D., Geurts, J.J.G., Zhang, Y.: A discrete polar Stockwell transform for enhanced characterization of tissue structure using MRI. Magn. Reson. Med. 44, 1755–1770 (2017)
Riba, L., Wong, M.W.: Continuous inversion formulas for multi-dimensional Stockwell transforms. Math. Model. Nat. Phenom. 8(1), 215–229 (2013)
Starck, J.-L., Candès, E.J., Donoho, D.L.: The curvelet transform for image denoising. IEEE Trans. Image Process. 11(6), 670–684 (2002)
Stockwell, R.G.: A basis for efficient representation of the S-transform. Digit. Signal Process. 17, 371–393 (2007)
Stockwell, R.G., Mansinha, L., Lowe, R.P.: Localization of the complex spectrum: the S transform. IEEE Trans. Signal Process. 44, 998–1001 (1996)
Voigtlaender, F.: Structured, compactly supported Banach frame decompositions of decomposition spaces (2016). arXiv:1612.08772
Wang, Y., Orchard, J.: Fast discrete orthonormal Stockwell transform. SIAM J. Sci. Comput. 31(5), 4000–4012 (2009)
Wong, M.W., Zhu, H.: A characterization of Stockwell spectra. In: Modern Trends in Pseudo-Differential Operators, vol. 172 of Operator Theory: Advances and Applications, pp. 251–257. Birkhäuser, Basel (2007)
Yan, Y., Zhu, H.: The generalization of discrete Stockwell transforms. In: 9th European Signal Processing Conference, pp. 1209–1213 (2011)
Zhu, H., Goodyear, B.G., Lauzon, M.L., Brown, R.A., et al.: A new local multiscale fourier analysis for medical imaging. Med. Phys. 30(6), 1134–1141 (2003)
Acknowledgements
We thank Fabio Nicola and Sandra Saliani for useful discussions on the subject. We also acknowledge the anonymous referee who helped improving the quality of the paper. We acknowledge that the present research has been partially supported by MIUR grant Dipartimenti di Eccellenza 2018-2022.
Author information
Authors and Affiliations
Corresponding author
Appendices
Technical results
We show some technical result needed to prove the frame property.
Lemma A.1
Let \(s\ge 0\) and \(\varphi \) such that \(\varPhi _{j,k}\)—cf. (10)—satisfies
for some \(\alpha \). Then
Proof
Inequality (38) is trivially verified when \(\omega \in I_{j,k}\) since \(\omega \asymp 2^j\).
Assume \(\omega \notin I_{j,k}\), then we immediately notice that by hypothesis (37)
Now, there exists \(\overline{\omega }\in I_{j,k}\) such that \(d(\omega ,I_{j,k}) = |\omega - \overline{\omega }|\), moreover \(|\overline{\omega }|\asymp 2^j\) because the partition is admissible. Hence, by triangular inequality
\(\square \)
With the same argument, one can show the following result.
Lemma A.2
Let \(s\ge 0\) and \(\varphi \) such that for all \(j \in {\mathbb {N}}\), \(\varPhi _{j,k}\)—cf. (10)—satisfies
for some \(\alpha >0\). Then
for all \(j \in {\mathbb {N}}\).
As a consequence of Lemma A.1, we have the following result.
Lemma A.3
Assume the hypothesis (37) of Lemma A.1 above and also that \(\alpha > d/2\). Then, there exists \( b_s \in \mathbb {R}\) such that
Proof
Given \(\omega \in \mathbb {R}^d\), since the partition is admissible, there exists a finite collection of indices \(\varGamma _F, |\varGamma _F|<W_1\) such that \(\omega \in I_{j,k}, \{j,k\}\in \varGamma _F\). Assume for the moment \(W_1 = 1\) and \(\omega \in I_{\overline{j},\overline{k}}\). First, we want to estimate
for \(j=\overline{j}\). Clearly, \(\omega \asymp 2^{\overline{j}}\), hence \(\frac{2^{2js}}{(1+|\omega |)^{2s}} |\varPhi _{\overline{j},k}(\omega )|^2\lesssim 1\). Moreover, since the number of possible directions is bounded by \(C_K\),
With this majorisation, we can treat also the cases of the adjacent coronae, i.e. for \(j = {\overline{j}\pm 1}\). If \(j> \overline{j}+ 1\) or \(j < \overline{j}- 1\), then \(\mathrm {d}(\omega , I_{j,k})\gtrsim 2^j\), and the result follows as above. Hence, if \(W_1=1\), the result follows easily by the requirement on \(\alpha \). Indeed, using Lemma A.1, one gets
which is clearly bounded. Now we can repeat this argument a finite number of time if \(W_1>1\) and the result follows. \(\square \)
Lemma A.4
Let \(s\ge 0\) and \(\varphi \) such that \(\varPhi _{j,k}(\omega )\) satisfies hypothesis (37). Define
then for \(\nu \in (0,1]\) and all j, k the system of functions \( \left\{ \widetilde{\varphi _{j,k,\lambda }}(t) \right\} _{\lambda \in \nu {\mathbb {Z}}} \) is a Bessel sequence uniformly in j, k, that is
with \(C_{\nu }\) independent on j, k.
Proof
A well known result (see again e.g. [9, Thm 9.2.5, p.206]) states that the Bessel property (45) of \(\left\{ \widetilde{\varphi _{j,k,\lambda }}(t) \right\} _{\lambda \in \nu {\mathbb {Z}}}\), for fixed j, is equivalent to the following condition
By definition
Using the hypothesis (37) and relation (38), we can write
where the second inequality follows from our assumption on the radius. Again, by our hypothesis on \(\alpha \), the sum in (46) is convergent, and uniformly bounded with respect to j. \(\square \)
Lemma A.5
Let \(\varPhi _{j,k}\), \(\widetilde{\varphi _{j,k,\lambda }}\) as in Lemma A.4 and \(E_{j,k}\) as in (13).
If \(\nu \in (0,1]\) and \( {\text {supp}}\widehat{f} \cap E_{j,k} =\emptyset ,\) then
with \(C_{\nu }\) as in Lemma A.4, therefore independent on j, k.
Proof
Since \(\hat{f}(\omega )=0\) if \(\omega \in E_{j,k}\)
Therefore, using the same property of Lemma A.4, (47) is equivalent to prove that
where \(m_{\gamma ,j,\nu } = \left( \gamma -m \right) \frac{2^j}{\nu }\). Since \(\nu \le 1 \), for each j, k, there exist a finite number of consecutive indices m such that \(\left( \gamma -m \right) \frac{2^j}{\nu } \in E_{j,k}\) for some \(\gamma \in [0,1]\). We set
We notice that \(M_{j,\nu }\) is uniformly bounded with respect to j, by the properties of the partitioning and by the definition of \(E_{j,k}\).
If \(m\in M_{j,\nu }\) and \(m_{\gamma ,j,\nu }\in E_{j,k}\), then
Otherwise \(\chi _{\mathbb {R}{\setminus } E_{j,k}} \left( m_{\gamma ,j,\nu } \right) =1\) and, using Lemma A.1,
Hence, (48) is bounded by
The second term in the equation above may be bounded as follows
Then the assertion follows as in Lemma A.4. \(\square \)
Lemma A.6
Let \(s\ge 0\) and \(\varphi _{\bullet },\varphi \) be a system of functions such that there exists \(a>0\) such that, for all \(\omega \in \mathbb {R}^d\) then
and the constant a does not depend on j, k. Then
with a constant \(C_s\) which depends on s only and is uniform with respect to \(\omega \).
Proof
For \(s=0\), the statement is trivial while for general s, notice that
while if \(\omega \in I_{\bullet }\), then \((1+|\omega |)\asymp 1\). \(\square \)
Remark A.1
Inequality (50) could be used as hypothesis on the window function weaker then ours. Since it is quite cumbersome to be checked, we prefer to work with a more transparent assumption.
We state now the counterpart of Lemmas A.4 and A.5 in the framework of seminorm discretisation.
Lemma A.7
Let \(s\ge 0\) and \(\varphi \) such that \(\varPhi _{j,k}(\omega )\) satisfies hypothesis (39). We define
then, for all \(\nu \in (0,1]\) and j, k the system of functions \( \left\{ \widetilde{\varphi _{j,k,\lambda }}(t) \right\} _{\lambda \in \nu {\mathbb {Z}}} \) is a Bessel sequence uniformly in j, k, that is
with \(C_\nu \) independent on j, k.
Lemma A.8
Let \(\varPhi _{j,k}\), \(\widetilde{\varphi _{j,k,\lambda }}\) as in Lemma A.7 and \(E_{j,k}\) as in (13) for \(j\in {\mathbb {N}}\) and as
for negative integers. If \(\nu \in (0,1]\) and \( {\text {supp}}\widehat{f} \cap E_{j,k} =\emptyset , \) then
with \(C_\nu \) independent on j, k.
Proof of Lemmas A.7, A.8
If \(j\in {\mathbb {N}}\) the proofs for is the same of Lemmas A.4 and A.5. In order to prove Lemma A.7 for \(-j\) with \(j\in {\mathbb {N}}{\setminus } \left\{ 0 \right\} \), we use the following relation
and the fact that \(\left\| {\text {D}}_{2^{-2j}} f \right\| _{L^2(\mathbb {R}^d)}= \left\| f \right\| _{L^2(\mathbb {R}^d)}\). For Lemma A.8 notice also that \( {\text {supp}}\widehat{f} \cap E_{-j,k}=\emptyset \) implies \( {\text {supp}}\widehat{{\text {D}}_{2^{-2j}} f}\cap E_{j,k}=\emptyset . \) \(\square \)
Interpolation techniques
We define a multi-resolution partition using the notation of [8]. Then, we adapt the interpolation result to our specific case.
Set \(V=L^2(\mathbb {R}), Z = H^1(\mathbb {R})\) and let
where \(\left\{ \varphi _{j,k,\lambda }\mid j, k, \lambda \in \varDelta \right\} \) is the frame defined as in (27). We also define the related projectors
where \(\varphi ^D_{j,k,\lambda }\) is the dual window. By definition
and also \(V_j \subseteq H^s(\mathbb {R}^d), j\in {\mathbb {Z}}\). Finally, we notice that due to our definition
Lemma B.9
Let \(r\in {\mathbb {Z}}\), then the following inequalities hold.
and
Proof
This follows immediately from Theorem 3.4 and a few observations. We notice that we can write the Sobolev seminorm as
where \(v^{(r)}\) is the rth derivative of v. We also notice that if \(\left\{ \varphi _{j,k,\lambda } \right\} _{j,k,\lambda \in \varGamma }\) is a frame for \(H^r\), then \(\left\{ 2^{-jr}\varphi _{j,k,\lambda }^{(r)} \right\} _{j,k,\lambda \in \varGamma }\) is a frame for \(L^2\). Indeed,
and the frequency window
clearly satisfies the requirements of Definition 3.3 and thus yields a frame, as claimed.
Hence, we can represent the rth derivative of the function v as
Finally, using the minimal property of the frame coefficients, see [16, Proposition 5.1.4.],
as desired.
To prove Jackson’s inequality, notice that, using again the minimal property of the frame coefficients,
and this concludes the proof. \(\square \)
We now review Theorem 1 in [8] adapting it to our framework.
Theorem B.1
Consider the spaces \(V_j\) and the associated partitions \(P_j\) that satisfies (57) and Jackson–Bernstein inequalities. Then, for any \(0<\alpha <1, q>1\) one has
and a norm for this space is \(\left\| \cdot \right\| = \left\| \cdot \right\| _2 + \left| \cdot \right| _{\alpha ,q}\).
Proof
The hypothesis of [8, Theorem 1] are fulfilled, hence the result follows. \(\square \)
Remark 9
The same result holds if we interchange the role of the window function and its dual.
Rights and permissions
About this article
Cite this article
Battisti, U., Berra, M. & Tabacco, A. Stockwell-like frames for Sobolev spaces. J. Pseudo-Differ. Oper. Appl. 9, 701–734 (2018). https://doi.org/10.1007/s11868-018-0259-7
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11868-018-0259-7