Iterative algorithms to approximate canonical Gabor windows: Computational aspects

Rate of convergence, degree of approximation (41A25) General harmonic expansions, frames (42C15) Linear operator approximation theory (47A58) Signal theory (characterization, reconstruction, filtering, etc.) (94A12) Sampling theory in information and communication theory (94A20)

Abstract: In this paper we investigate the computational aspects of some recently proposed iterative methods for approximating the canonical tight and canonical dual window of a Gabor frame (g,a,b). The iterations start with the window g while the iteration steps comprise the window g, the k^{th} iterand gamma_{k}, the frame operators S and S_{k} corresponding to (g,a,b) and (gamma_{k},a,b), respectively, and a number of scalars. The structure of the iteration step of the method is determined by the envisaged convergence order m of the method. We consider two strategies for scaling the terms in the iteration step: norm scaling, where in each step the windows are normalized, and initial scaling where we only scale in the very beginning. Norm scaling leads to fast, but conditionally convergent methods, while initial scaling leads to unconditionally convergent methods, but with possibly suboptimal convergence constants. The iterations, initially formulated for time-continuous Gabor systems, are considered and tested in a discrete setting in which one passes to the appropriately sampled-and-periodized windows and frame operators. Furthermore, they are compared with respect to accuracy and efficiency with other methods to approximate canonical windows associated with Gabor frames.

Review text

In the paper an analysis of the iterations to obtain two operators allowing the exact reconstruction of \(L_2\) functions is presented. The paper represents a masterpiece of numerical analysis by its mathematical level as well as by its length (30 pages), so it is a true memoir. The reconstruction is based on Gabor Families (GF) which can be described as follows. Let be given the objects: \(a> 0\) (the time-shift), \(b> 0\) (the frequency shift), \(g\in L_2\) (the window) by means of which one constructs a 2-parameter family of functions (a kernel), \(g_{uv}(t)= \exp(2\pi jvt)g(t- u)\), \(t\in\mathbb{R}\). This is exactly the sliding window known in signal analysis. Next, the Gabor System (GS), denoted by \((g,a, b)\), is defined as the family of the translates of the kernel \((g,a,b)= \{g_{na,mb},m,n\in \mathbb{Z}\}\). With respect to this family, one may inquire whether it represents a frame, with the bounds \(0< A\leq B<\infty\). If this is true, one says that \((g,a,b)\) is a Gabor Frame (GF). In this, case, along with the GF, one introduces the frame operator \(S\) by \[ Sf= \sum^\infty_{m,n=-\infty}(f, g_{na,mb}) g_{na,mb}, \] where \((f,g)\) denotes the scalar product. These facts are already known from wavelet theory. A relevant fact related to GF was realized in recent times by introducing two special canonical reconstructions formulae using two canonical windows, namely \(g^t= S^{-1/2}g\), \(g^d= S^{-1}g\) which provide the following representations for any \(g\in L_2:f= \sum^\infty_{m,n=-\infty} (f,g^t_{na,mb}) g^t_{na,mb}= \sum^\infty_{m,n=-\infty} (f,g^d_{na, mb}) g^d_{na, mb}\). Moreover \((g^t, a,b)\) and \((g^d, a,b)\) are GF themselves. The content of the paper lies in providing a complete analysis of some iteration schemes, existing in literature, which lead to the canonical windows. Each iteration starts with the window \(g\) which leads to the GS \((g,a, b)\), a GS with the bounds \(A\) and \(B\). At each step, say \(k\), a window \(\gamma_k\), with \(\gamma_0= g\), is produced together with its GF \((\gamma_k,a, b)\) and its operator frame \(S_k\). The iteration comprises the preceding window, the frame operator and some constants. Two strategies for scaling the steps were considered: norm scaling and initial scaling. It is interesting to point out that the convergence of the iteration is insured if the frame ratio \(A/B\) obeys to a specific inequality. For the numerical experiments, two windows were selected: a Gabor signal and the secant one. Since the theory of iterations and their convergence were developed in the continuous cases, simulations and tests were performed in the discrete case using appropriate sampling and periodization of windows and frame operators. Comparison of the used iteration was favorable when compared with other ones. Finally, we think we are faced with an important contribution to signal restoration by the use of Gabor frames.

Recommendations

Cited in

(10)

Describes a project that uses

Uses Software

ATLAS

This page was built for publication: Iterative algorithms to approximate canonical Gabor windows: Computational aspects

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2373369)