Core Concepts
Iterative algorithms using discrete Fourier transforms are effective for signal denoising.
Abstract
The content discusses iterative algorithms based on discrete and inverse discrete Fourier transforms for signal denoising. It explores the convergence properties, generalizations, and practical utility of these algorithms. The IterativeFT method is highlighted as a powerful technique for recovering periodic spike signals in the presence of noise. Simulation studies demonstrate the superior denoising performance of IterativeFT compared to existing methods like real domain thresholding, frequency domain thresholding, Butterworth bandpass filtering, and wavelet filtering. The paper concludes by suggesting future research directions and acknowledging funding sources.
Problem Statement:
Family of iterative algorithms involving discrete Fourier transforms.
Motivated by uncertainty principle & sparsification operation.
General Convergence Properties:
Key question: Combinations of h(), g(), c() functions enabling convergence.
Interest in scenarios relevant to data analysis applications.
Algorithm 1:
Inputs: x ∈Rn, im (maximum iterations).
Outputs: y ∈Rn, ic (iterations completed).
Detailed definition provided.
Trivial Cases:
Convergence after single iteration if h() and g() are identity functions.
Simplification with only one function being identity function.
Generalizations:
Matrix-valued input possible.
Complex-valued input allowed.
Alternative invertible discrete transform can replace DFT.
Related Techniques:
Relationship to standard DFT analysis discussed.
Comparison with other iterative algorithms like ADMM, Dykstra's algorithm, EM made.
Iterative Convergence under Sparsification:
Subclass involves sparsification functions for h() and g().
Motivated by DFT uncertainty principle.
Stable compromise between real and frequency domain sparsity achieved upon convergence.
Detection of Spike Signals Using Iterative Convergence:
Assessment of practical utility using simulation design.
Performance evaluation relative to varying parameters like n value and sparse proportion.
Comparative Evaluation of Denoising Performance:
Simulation study design outlined.
Comparison with existing denoising techniques - real/frequency domain thresholding, Butterworth bandpass filtering, wavelet filtering.
Superior performance of IterativeFT method demonstrated through simulation results across different spike signal patterns and SNR values.
Stats
arXiv:2211.09284v3 [eess.SP] 24 Mar 2024