Hyperbolic secants yield Gabor frames

A. J. E. M. Janssen, Thomas Strohmer

Abstract

We demonstrate that the hyperbolic secant function $g_2(t)=({1/2}πγ)^{1/2} (\cosh πγt)^{-1}$ generates a Gabor frame $(g_2,a,b)$ for any sampling density $ab < 1$, where $a, b, \gamma > 0$. This result is established by relating the Zak transform of the hyperbolic secant to that of the Gaussian $g_1(t)=(2γ)^{1/4} \exp (-πγt^2)$. By applying the Ron-Shen criterion, we confirm the stability of the resulting frame. Furthermore, we show that at the critical density $ab=1$, the canonical tight windows associated with both the Gaussian and the hyperbolic secant are identical, and we characterize the "singular" dual function for the hyperbolic secant.

Introduction

In the field of Gabor analysis, signals are decomposed into a series of time-frequency shifts of a single window function $g$. For such a system to be practically useful in engineering and signal processing, the set $\{g_{m,n}(t) = e^{2\pi i n b t} g(t - ma)\}_{m,n \in \mathbb{Z}}$ must constitute a frame. A frame ensures that the signal can be reconstructed from its coefficients in a numerically stable manner, even when the representation is redundant ($ab < 1$).

The Gaussian function is the most prominent window in Gabor theory due to its optimal localization in both time and frequency. However, other functions with similar localization properties are of significant interest. The hyperbolic secant ($\text{sech}$) is a natural alternative; it possesses a bell-shaped curve similar to the Gaussian but exhibits "heavier tails" (exponential decay rather than quadratic). This paper provides the rigorous proof that the hyperbolic secant indeed generates a Gabor frame for all valid sampling rates.

Methodology and the Zak Transform

The primary tool used to analyze these Gabor systems is the Zak transform, defined for a function $f$ as: $$Zf(t, \nu) = \sum_{k \in \mathbb{Z}} f(t + k) e^{-2\pi i k \nu}$$ The Zak transform is essential because it maps the problem of determining frame bounds into the analysis of a function on the unit square $[0, 1) \times [0, 1)$. To prove that the hyperbolic secant generates a frame, the authors utilize the Ron-Shen criterion, which provides a condition for the lower frame bound to be strictly positive, ensuring no information is lost during the transformation.

The mathematical derivation relies heavily on Jacobi theta functions, specifically $\vartheta_4$. By employing Cauchy’s residue theorem, the authors compute the Fourier coefficients of ratios of these theta functions to bridge the gap between the hyperbolic secant and the Gaussian.

The Factorization Insight

The central discovery of this research is a "factorization trick" that links the Zak transform of the hyperbolic secant ($g_2$) directly to that of the Gaussian ($g_1$). The relationship is expressed as: $$\frac{Zg_2(t, \nu)}{Zg_1(t, \nu)} = \frac{E}{\vartheta_4(\pi\nu; e^{-\pi\gamma})\vartheta_4(\pi t; e^{-\pi/\gamma})}$$ where $E$ is a constant. This formula reveals that the ratio of the two transforms is "separable"—it can be decomposed into a product of a function of time ($t$) and a function of frequency ($\nu$). Because the properties of the Gaussian's Zak transform are well-established, this factorization allows the frame properties of the hyperbolic secant to be inherited and verified with relative ease.

Critical Density and Universal Tight Windows

A particularly striking result occurs at the critical density ($ab=1$). While the Balian-Low theorem states that a Gabor system cannot be a stable frame at this density if the window is well-localized, one can still study the "canonical tight window." This is the window $g_{tight}$ that is "closest" to the original window $g$ while satisfying the condition that the resulting Gabor system is a tight frame.

The authors demonstrate that at $ab=1$, the canonical tight windows for both the Gaussian and the hyperbolic secant are identical. This suggests a deep structural connection between these two functional forms. Furthermore, as the scaling parameter $\gamma$ is varied, these tight windows form a family that interpolates between a square pulse (the Haar window) and a $\text{sinc}$ function.

Conclusion

This work solidifies the role of the hyperbolic secant in Gabor analysis, proving it to be a robust alternative to the Gaussian for signal representation. By providing explicit formulas for the Zak transform and the dual functions at critical density, the paper offers a complete theoretical framework for $\text{sech}$-based Gabor systems. While the current study is purely theoretical, it opens the door for future investigations into the computational efficiency and practical advantages of "heavy-tailed" windows in digital signal processing.

Summary updated

Abstract

We show that $(g_2,a,b)$ is a Gabor frame when $a&gt;0, b&gt;0, ab&lt;1$ and $g_2(t)=({1/2}πγ)^{1/2} (\cosh πγt)^{-1}$ is a hyperbolic secant with scaling parameter $γ&gt;0$. This is accomplished by expressing the Zak transform of $g_2$ in terms of the Zak transform of the Gaussian $g_1(t)=(2γ)^{1/4} \exp (-πγt^2)$, together with an appropriate use of the Ron-Shen criterion for being a Gabor frame. As a side result it follows that the windows, generating tight Gabor frames, that are canonically associated to $g_2$ and $g_1$ are the same at critical density $a=b=1$. Also, we display the ``singular'' dual function corresponding to the hyperbolic secant at critical density.
BibTeX
@inproceedings{janssen2002hyperbolic,
  author = {A. J. E. M. Janssen and Thomas Strohmer},
  title = {{Hyperbolic secants yield Gabor frames}},
  booktitle = {Applied and Computational Harmonic Analysis},
  year = {2002},
  doi = {10.1006/acha.2001.0376},
}
Figure