Abstract
This paper studies the recovery of a superposition of point sources from noisy bandlimited data. In the fewest possible words, we only have information about the spectrum of an object in the low-frequency band [−f lo,f lo] and seek to obtain a higher resolution estimate by extrapolating the spectrum up to a frequency f hi>f lo. We show that as long as the sources are separated by 2/f lo, solving a simple convex program produces a stable estimate in the sense that the approximation error between the higher-resolution reconstruction and the truth is proportional to the noise level times the square of the super-resolution factor (SRF) f hi/f lo.
Similar content being viewed by others
Notes
To be precise, the theorem holds for any feasible point \(\tilde{x}\) obeying \(\| \tilde {x} \| _{\mathrm{TV}}\leq\| x \| _{\mathrm{TV}}\); this set is not empty since it contains x.
References
Batenkov, D., Yomdin, Y.: On the accuracy of solving confluent Prony systems. SIAM J. Appl. Math. 73(1), 134–154 (2013)
Betzig, E., Patterson, G.H., Sougrat, R., Lindwasser, O.W., Olenych, S., Bonifacino, J.S., Davidson, M.W., Lippincott-Schwartz, J., Hess, H.F.: Imaging intracellular fluorescent proteins at nanometer resolution. Science 313(5793), 1642–1645 (2006)
Bhaskar, B.N., Tang, G., Recht, B.: Atomic norm denoising with applications to line spectral estimation. Preprint
Bienvenu, G.: Influence of the spatial coherence of the background noise on high resolution passive methods. In: Proceedings of the International Conference on Acoustics, Speech and Signal Processing, vol. 4, pp. 306–309 (1979)
Blu, T., Dragotti, P., Vetterli, M., Marziliano, P., Coulot, L.: Sparse sampling of signal innovations. IEEE Signal Process. Mag. 25(2), 31–40 (2008)
Candès, E.J., Fernandez-Granda, C.: Towards a mathematical theory of super-resolution. Commun. Pure Appl. Math. (2013, to appear). doi:10.1002/cpa.21455
Chi, Y., Scharf, L.L., Pezeshki, A., Calderbank, A.R.: Sensitivity to basis mismatch in compressed sensing. IEEE Trans. Signal Process. 59(5), 2182–2195 (2011)
Clergeot, H., Tressens, S., Ouamri, A.: Performance of high resolution frequencies estimation methods compared to the Cramér-Rao bounds. IEEE Trans. Acoust. Speech Signal Process. 37(11), 1703–1720 (1989)
Donoho, D.L.: Superresolution via sparsity constraints. SIAM J. Math. Anal. 23(5), 1309–1331 (1992)
Duarte, M.F., Baraniuk, R.G.: Spectral compressive sensing. Appl. Comput. Harmon. Anal. 35(1), 111–129 (2013)
Dumitrescu, B.: Positive Trigonometric Polynomials and Signal Processing Applications. Springer, Berlin (2007)
Fannjiang, A., Liao, W.: Coherence-pattern guided compressive sensing with unresolved grids. SIAM J. Imaging Sci. 5, 179 (2012)
Fannjiang, A.C.: The MUSIC algorithm for sparse objects: a compressed sensing analysis. Inverse Probl. 27(3), 035013 (2011)
Grant, M., Boyd, S.: CVX: Matlab software for disciplined convex programming, version 1.21 (2011). http://cvxr.com/cvx
Greenspan, H.: Super-resolution in medical imaging. Comput. J. 52, 43–63 (2009)
Harris, T.D., Grober, R.D., Trautman, J.K., Betzig, E.: Super-resolution imaging spectroscopy. Appl. Spectrosc. 48(1), 14–21 (1994)
Helstrom, C.: The detection and resolution of optical signals. IEEE Trans. Inf. Theory 10(4), 275–287 (1964)
Hess, S.T., Girirajan, T.P., Mason, M.D.: Ultra-high resolution imaging by fluorescence photoactivation localization microscopy. Biophys. J. 91(11), 4258 (2006)
Hu, L., Shi, Z., Zhou, J., Fu, Q.: Compressed sensing of complex sinusoids: an approach based on dictionary refinement. IEEE Trans. Signal Process. 60(7), 3809–3822 (2012)
Huang, B., Bates, M., Zhuang, X.: Super-resolution fluorescence microscopy. Annu. Rev. Biochem. 78, 993–1016 (2009)
Itakura, F.: Line spectrum representation of linear predictor coefficients of speech signals. J. Acoust. Soc. Am. 57(S1), S35 (1975)
Kennedy, J., Israel, O., Frenkel, A., Bar-Shalom, R., Azhari, H.: Super-resolution in PET imaging. IEEE Trans. Med. Imaging 25(2), 137–147 (2006)
Khaidukov, V., Landa, E., Moser, T.J.: Diffraction imaging by focusing-defocusing: an outlook on seismic superresolution. Geophysics 69(6), 1478–1490 (2004)
Laurent, B., Massart, P.: Adaptive estimation of a quadratic functional by model selection. Ann. Stat. 28(5), 1053–1302 (1992)
Lindberg, J.: Mathematical concepts of optical superresolution. J. Opt. 14(8), 083001 (2012)
Makovoz, D., Marleau, F.R.: Point source extraction with MOPEX. Publ. Astron. Soc. Pac. 117(836), 1113–1128 (2005)
McCutchen, C.W.: Superresolution in microscopy and the Abbe resolution limit. J. Opt. Soc. Am. 57(10), 1190 (1967)
Milanfar, P. (ed.) Super-Resolution Imaging. Digital Imaging and Computer Vision (2010)
Odendaal, J., Barnard, E., Pistorius, C.: Two-dimensional superresolution radar imaging using the MUSIC algorithm. IEEE Trans. Antennas Propag. 42(10), 1386–1391 (1994)
Park, S.C., Park, M.K., Kang, M.G.: Super-resolution image reconstruction: a technical overview. IEEE Signal Process. Mag. 20(3), 21–36 (2003)
Prony, R.: Essai expérimental et analytique: sur les lois de la dilatabilité de fluides élastique et sur celles de la force expansive de la vapeur de l’alkool, à différentes températures. J. Éc. Polytech. 1(2), 24–76 (1795)
Puschmann, K.G., Kneer, F.: On super-resolution in astronomical imaging. Astron. Astrophys. 436, 373–378 (2005)
Roy, R., Kailath, T.: ESPRIT—estimation of signal parameters via rotational invariance techniques. IEEE Trans. Acoust. Speech Signal Process. 37(7), 984–995 (1989)
Rudin, W.: Real and Complex Analysis, 3rd edn. McGraw-Hill, New York (1987)
Rust, M.J., Bates, M., Zhuang, X.: Sub-diffraction-limit imaging by stochastic optical reconstruction microscopy (STORM). Nat. Methods 3(10), 793–796 (2006)
Schmidt, R.: Multiple emitter location and signal parameter estimation. IEEE Trans. Antennas Propag. 34(3), 276–280 (1986)
Shahram, M., Milanfar, P.: Imaging below the diffraction limit: a statistical analysis. IEEE Trans. Image Process. 13(5), 677–689 (2004)
Shahram, M., Milanfar, P.: On the resolvability of sinusoids with nearby frequencies in the presence of noise. IEEE Trans. Signal Process. 53(7), 2579–2588 (2005)
Slepian, D.: Prolate spheroidal wave functions, Fourier analysis, and uncertainty. V. The discrete case. Bell Syst. Tech. J. 57, 1371–1430 (1978)
Stoica, P., Babu, P.: Sparse estimation of spectral lines: Grid selection problems and their solutions. IEEE Trans. Signal Process. 60(2), 962–967 (2012)
Stoica, P., Moses, R., Friedlander, B., Soderstrom, T.: Maximum likelihood estimation of the parameters of multiple sinusoids from noisy measurements. IEEE Trans. Acoust. Speech Signal Process. 37(3), 378–392 (1989)
Stoica, P., Moses, R.L.: Spectral Analysis of Signals. Prentice Hall, New York (2005)
Stoica, P., Nehorai, A.: Statistical analysis of two nonlinear least-squares estimators of sine-wave parameters in the colored-noise case. Circuits Syst. Signal Process. 8(1), 3–15 (1989)
Stoica, P., Soderstrom, T.: Statistical analysis of MUSIC and subspace rotation estimates of sinusoidal frequencies. IEEE Trans. Signal Process. 39(8), 1836–1847 (1991)
Tang, G., Bhaskar, B.N., Shah, P., Recht, B.:. Compressed sensing off the grid. Preprint
Acknowledgements
E.J. Candès is partially supported by AFOSR under grant FA9550-09-1-0643, by ONR under grant N00014-09-1-0258 and by a gift from the Broadcom Foundation. C. Fernandez-Granda is supported by a Fundación Caja Madrid Fellowship. We thank Carlos Sing-Long for useful feedback about an earlier version of the manuscript.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Thomas Strohmer.
Appendices
Appendix A: Proof of Lemma 2.5
We use the construction described in Sect. 2 of [6]. In more detail,
where \(\alpha, \beta\in\mathbb{C}^{\vert T \vert }\) are coefficient vectors,
and G(0)=1; here, G (ℓ) is the ℓth derivative of G. If f lo is even, G(t) is the square of the Fejér kernel. By construction, the coefficients α and β are selected such that for all t j ∈T,
Without loss of generality we consider t j =0 and bound q(t)−v j in the interval [0,0.16λ lo]. To ease notation, we define w(t)=q(t)−v j =w R (t)+iw I (t), where w R is the real part of w and w I the imaginary part. Leveraging different results from Sect. 2 in [6] (in particular the equations in (2.25) and Lemmas 2.2 and 2.7), we have
The same bound holds for w I . Since w R (0), \(w_{R}'(0)\), w I (0) and \(w_{I}'(0)\) are all equal to zero, this implies \(\vert w_{R}(t) \vert \leq C' f_{\mathrm{lo}}^{2} t^{2}\) and \(\vert w_{I}(t) \vert \leq C' f_{\mathrm {lo}}^{2} t^{2}\) in the interval of interest, which allows the conclusion
Appendix B: Proof of Lemma 2.7
The proof is similar to that of Lemma 2.4 (see Sect. 2 of [6]), where a low-frequency kernel and its derivative are used to interpolate an arbitrary sign pattern on a support satisfying the minimum-distance condition. More precisely, we set
where \(\alpha, \beta\in\mathbb{C}^{\vert T \vert }\) are coefficient vectors, G is defined by (A.1). Note that G, G (1) and, consequently, q 1 are trigonometric polynomials of degree at most f 0. By Lemma 2.7 in [6], it holds that for any t 0∈T and \(t \in\mathbb{T}\) obeying |t−t 0|≤0.16λ lo,
where C ℓ is a positive constant for ℓ=0,1,2,3; in particular, C 0≤0.007, C 1≤0.08 and C 2≤1.06. In addition, there exist other positive constants \(C_{0}'\) and \(C_{1}'\), such that for all t 0∈T and \(t\in\mathbb{T}\) with |t−t 0|≤Δ/2,
for ℓ=0,1. We refer to Sect. 2.3 in [6] for a detailed description of how to compute these bounds.
In order to satisfy (2.18) and (2.19), we constrain q 1 as follows: for each t j ∈T,
Intuitively, this forces q 1 to approximate the linear function v j (t−t j ) around t j . These constraints can be expressed in matrix form,
where
and j and k range from 1 to |T|. It is shown in Sect. 2.3.1 of [6] that under the minimum-separation condition this system is invertible, so that α and β are well defined. These coefficient vectors can consequently be expressed as
where S is the Schur complement. Inequality (B.2) implies
where κ=|G (2)(0)|=π 2 f lo(f lo+4)/3.
Let ∥M∥∞ denote the usual infinity norm of a matrix M defined as \(\|M\|_{\infty}= \max_{\|x\|_{\infty}= 1} \|Mx\|_{\infty}= \max_{i} \sum_{j} |a_{ij}|\). Then, if ∥I−M∥∞<1, the series M −1=(I−(I−M))−1=∑ k≥0(I−M)k is convergent and we have
This, together with (B.4), (B.5) and (B.6) implies
for a certain positive constant C κ . Note that due to the numeric upper bounds on the constants in (B.2) C κ is indeed a positive constant as long as f lo≥1. Finally, we obtain a bound on the magnitude of the entries of α
where C α =C κ C 1/(1−C 0), and on the entries of β
for a positive constant C β =C κ . Combining these inequalities with (B.3) and the fact that the absolute values of G(t) and G (1)(t) are bounded by one and 7f lo respectively (see the proof of Lemma C.5 in [6]), we have that for any t
where t i denotes the element in T nearest to t (note that all other elements are at least Δ/2 away). Thus, (2.19) holds.
The proof is completed by the following lemma, which proves (2.18).
Lemma 6.1
For any t j ∈T and \(t \in\mathbb{T}\) obeying |t−t j |≤0.16λ lo, we have
Proof
We assume without loss of generality that t j =0. By symmetry, it suffices to show the claim for t∈(0,0.16λ lo]. To ease notation, we define w(t)=v j t−q 1(t)=w R (t)+iw I (t), where w R is the real part of w and w I the imaginary part. Leveraging (B.7), (B.8) and (B.2) together with the fact that G (2)(t) and G (3)(t) are bounded by \(4 f_{\mathrm{lo}}^{2}\) and \(6 f_{\mathrm{lo}}^{3}\) respectively if |t|≤0.16λ lo (see the proof of Lemma 2.3 in [6]), we obtain
The same bound applies to w I . Since w R (0), \(w_{R}'(0)\), w I (0) and \(w_{I}'(0)\) are all equal to zero, this implies |w R (t)|≤Cf lo t 2—and similarly for |w I (t)|—in the interval of interest. Whence, |w(t)|≤Cf lo t 2. □
Appendix C: Proof of Corollary 1.3
The proof of Theorem 1.2 relies on two identities
which suffice to establish
To prove the corollary, we show that (C.1) and (C.2) hold. Due to the fact that \(\| \epsilon \| _{2}^{2}\) follows a χ 2-distribution with 4f lo+2 degrees of freedom, we have
for any positive γ by a concentration inequality (see [24, Sect. 4]). By Parseval, this implies that with high probability \(\| Q_{\mathrm{lo}}x -y \| _{\mathcal {L}_{2}}=\| \epsilon \| _{2}\leq\delta\). As a result, x est is feasible, which implies (C.1) and furthermore
since by the Cauchy-Schwarz inequality \(\| f \| _{L_{1}} \leq\| f \| _{\mathcal{L}_{2}}\) for any function f with bounded L 2 norm supported on the unit interval. Thus, (C.2) also holds and the proof is complete.
Appendix D: Extension to Multiple Dimensions
The extension of the proof hinges on establishing versions of Lemmas 2.4, 2.5 and 2.7 for multiple dimensions. These lemmas construct bounded low-frequency polynomials which interpolate a sign pattern on a well-separated set of points S and have bounded second derivatives in a neighborhood of S. In the multidimensional case, we need the directional derivative of the polynomials to be bounded in any direction, which can be ensured by bounding the eigenvalues of their Hessian matrix evaluated on the support of the signal. To construct such polynomials one can proceed in a way similar to the proof of Lemmas 2.4 and 2.7, namely, by using a low-frequency kernel constructed by tensorizing several squared Fejér kernels to interpolate the sign pattern, while constraining the first-order derivatives to either vanish or have a fixed value. As in the one-dimensional case, one can set up a system of equations and prove that it is well conditioned using the rapid decay of the interpolation kernel away from the origin. Finally, one can verify that the construction satisfies the required conditions by exploiting the fact that the interpolation kernel and its derivatives are locally quadratic and rapidly decaying. This is spelled out in the proof of Proposition C.1 in [6] to prove a version of Lemma 2.4 in two dimensions. In order to clarify further how to adapt our techniques to a multidimensional setting we provide below a sketch of the proof of the analog of Lemma 2.1 in two dimensions. In particular, this illustrates how the increase in dimension does not change the exponent of the SRF in our recovery guarantees.
4.1 D.1 Proof of an Extension of Lemma 2.1 to Two Dimensions
We now have \(t \in\mathbb{T}^{2}\). As a result, we redefine
where w is a constant.
The proof relies on the existence of a low-frequency polynomial
satisfying
where \(C_{a}'\) and \(C_{b}'\) are constants. Proposition C.1 in [6] constructs such a polynomial. Under a minimum distance condition, which constrains the elements of T to be separated by 2.38λ lo in infinity norm (as explained in [6] this choice of norm is arbitrary and could be changed to the ℓ 2 norm), [6] shows that q satisfies (D.1) and (D.3) and that both eigenvalues of its Hessian matrix evaluated on T are of order \(f_{\mathrm{lo}}^{2}\), which implies (D.2).
As in one dimension, we perform a polar decomposition of P T h,
and work with \(v_{j} = e^{-i \phi(t_{j})}\). The rest of the proof is almost identical to the 1D case. Since q is low frequency,
Next, since q interpolates e −iϕ(t) on T,
Applying (D.3) and Hölder’s inequality, we obtain
Setting t j =(0,0) without loss of generality, the triangle inequality and (D.2) yield
Combining (D.5), (D.6) and (D.7) gives
and similarly
By the same argument as in the 1D case, the fact that \(\hat{x}\) has minimal total-variation norm is now sufficient to establish
and
Rights and permissions
About this article
Cite this article
Candès, E.J., Fernandez-Granda, C. Super-Resolution from Noisy Data. J Fourier Anal Appl 19, 1229–1254 (2013). https://doi.org/10.1007/s00041-013-9292-3
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00041-013-9292-3
Keywords
- Deconvolution
- Stable signal recovery
- Sparsity
- Line spectra estimation
- Basis mismatch
- Super-resolution factor