In the analysis of power grid waveforms, the presence of amplitude or phase steps can disrupt the estimation of frequency and rate-of-change-of-frequency (ROCOF). Standard methods based on phasor-models fail in the extraction of signal parameters during these signal dynamics, often yielding large frequency and ROCOF deviations. To address this challenge, we propose a technique that approximates components of the signal (e.g., amplitude and frequency variations) using dictionaries based on parameterized models of common signal dynamics. Distinct from a previous iteration of this method developed by the authors, the proposed technique allows for the identification of multiple steps in a window, as well as the presence of interfering tones. The method is shown to improve signal reconstruction when applied to real-world waveforms, as compared to standard static and dynamic phasor-based algorithms.