Improved Sinusoidal Parameter Estimation through Iterative Linear Programming Schemes

Publication Type:

Conference Paper


SMC Conference 2011 (2011)


Sinusoidal modeling systems are commonly employed in sound and music processing systems for their ability to decompose a signal to its fundamental spectral information. Sinusoidal modeling is a two-phase process: sinusoidal parameters are estimated in each analysis frame in the first phase, and these parameters are chained into sinusoidal trajectories in the second phase. This paper focuses on the first phase. Current methods for estimating parameters rely heavily on the resolution of the Fourier transform and are thus hindered by the Heisenberg uncertainty principle. A novel approach is proposed that can super-resolve frequencies and attain more accurate estimates of sinusoidal parameters than current methods. The proposed algorithm formulates parameter estimation as a linear programming problem, in which the L1 norm of the residual component of the sinusoidal decomposition is minimized. Shared information from iteration to iteration and from frame to frame allows for efficient parameter estimation at high sampling rates.

smc2011_submission_81.pdf328.24 KB
SMC paper: