Linear Prediction Approach to the Robust Parameter Estimation for the Damped Sinusoids

In this paper, we focus on the problem of parameter estimation for the damped sinusoids, which are corrupted by impulsive noise. To provide a robust initial guess for the current parameter estimators, the robust weighted linear prediction (RWLP) estimator is developed, where the parameter estimates are obtained by minimizing the weighted ℓ
_p
-norm of the linear prediction (LP) error vector. The Markov optimum weighting matrix is derived, and an iteratively reweighted least-squares (IRLS) procedure is devised to calculate the LP coefficient estimates. The simulation results demonstrate the robustness of the RWLP estimator by comparing with the ℓ
₂
-norm based counterpart, and the computational efficiency by comparing with the ℓ
_p
-MUSIC algorithm.