1: \begin{abstract}

2: We propose a fast sequential algorithm for the fundamental problem of estimating frequencies and amplitudes of a noisy mixture of sinusoids. The algorithm is a natural generalization of Orthogonal Matching Pursuit (OMP) to the continuum using Newton refinements, and hence

3:  is termed Newtonized OMP (NOMP). Each iteration consists of two phases: detection of a new sinusoid, and sequential Newton refinements of the parameters of already detected sinusoids. The refinements play a critical role in two ways: (1) sidestepping the potential \emph{basis mismatch} from discretizing a continuous parameter space, (2) providing \emph{feedback} for locally refining parameters estimated in previous iterations. 

4: We characterize convergence, and provide a Constant False Alarm Rate (CFAR) based termination criterion.  

5:  By benchmarking against the \CRB, we show that NOMP achieves near-optimal performance under a variety of conditions. 

6: We  compare the performance of NOMP with classical algorithms such as MUSIC 

7: and more recent Atomic norm Soft Thresholding (AST) and Lasso algorithms, both in terms of frequency estimation accuracy and run time.

8: %  Even though our sequential algorithm can be applied to a general sparse approximation problem, we specialize the analysis and simulation results to the canonical problem of line spectral estimation.

9: \end{abstract}

10: