Estimating the amplitudes, frequencies, and phases of sinusoidal components is a core problem in signal processing, but accuracy suffers when frequencies are mismatched. Quasi-harmonic modeling (QHM) and its extensions refine frequency estimates while extracting amplitudes, yet their performance is bounded by a linear model structure that cannot represent higher-order frequency changes.
The authors revisit QHM from the perspective of frequency mismatch and show that the model order is the key constraint. To overcome this, they introduce a higher-order modeling framework that approximates signal components more precisely. Parameter estimation is framed as an optimization problem and solved with the Levenberg–Marquardt algorithm, allowing simultaneous refinement of amplitudes, frequencies, and phases.
Experiments on both synthetic signals and speech demonstrate that the high-order model outperforms existing QHM methods in frequency refinement and parameter extraction, and remains robust under additive noise. The paper is under review for IEEE Transactions on Signal Processing.