Core Concepts
The author presents a method for blind deconvolution and estimation of second-order cyclostationary signals, addressing the impact of transfer functions without prior knowledge requirements.
Abstract
The study focuses on solving the dual problem of blind deconvolution and estimation of time waveform for noisy second-order cyclo-stationary signals. It introduces a blind method that eliminates transfer function effects from signals with varying statistics over time. The research aims to improve machine learning model training by aggregating signals from identical systems with different transfer functions. Various applications in telecommunications, radar, mechanics, and more are discussed. The paper outlines related works in the field, defines the problem statement, methodology for deconvolution filter estimation, and CS2 envelope estimation algorithm. Simulations demonstrate robustness under different parameters like TF poles, cyclic components in signals, and noise levels. Error analysis highlights parameter choices and method limitations while concluding with future research directions.
Stats
Signals ๐ฅ(๐ก) each lasting one second and consisting of ๐ = 24,000 samples were generated during each run using (5).
The number of poles in the TF varied between 5 and 20.
The number of cyclic components in ๐(๐ก) ranged between 5 to 20.
SNR varied between -20dB and 20dB.
Quotes
"The impact of the TF becomes even more critical in Machine Learning models."
"Signals examined are constrained by multiplication of a deterministic periodic function and white noise."
"The study augments research by addressing deconvolution & CS2 waveform estimation."