Quantitative Assessment of Dual window Efficiency in Gabor Frame Expansions
Sruthi Raghoothaman, Noufal Asharaf
TL;DR
This work investigates reconstruction quality in Gabor frame expansions using compactly supported windows $B_2$, $B_3$, and $\varepsilon_3$ by comparing canonical and multiple alternative dual windows. It applies two dual-construction strategies—a perturbation-based method and a recursive, explicit scheme—across three window families and evaluates reconstruction accuracy with the Average Mean Squared Error on five standard signals. The results show that exponential B-splines $\varepsilon_3$ consistently yield the lowest AMSE, and among duals, symmetric constructions like $k$ (and their perturbations) deliver the most accurate reconstructions, offering practical guidance for time-frequency analysis tasks. Together, these findings provide concrete guidance on selecting window types and duals to achieve stable, high-fidelity Gabor reconstructions in signal processing applications.
Abstract
We investigate the performance of Gabor frame reconstructions using three compactly supported window functions: the second-order B-spline ($B_2$), the third-order B-spline ($B_3$), and the exponential B-spline of order 3 ($\varepsilon_3$). For each generator, various dual windows are considered, including the canonical dual, symmetric and asymmetric duals, and perturbation-based duals constructed via a recent duality result. The reconstruction quality is assessed using the Average Mean Squared Error (AMSE) across five standard test signals: Blocks, Bumps, Heavisine, Doppler, and Quadchirp. Numerical experiments demonstrate that exponential B-splines yield the lowest AMSE among the three, confirming their effectiveness in Gabor frame applications. Among the dual windows, the symmetric dual and its perturbation-based variant consistently achieve the best reconstruction accuracy, making them strong candidates for practical use in signal processing tasks.
