Interval methods for sinusoidal parameter estimation: A comparative analysis (Q1582468)

The paper compares three algorithms to estimate sinusoidal parameters in various applications such as signal modeling and estimation of directions of arrival. The estimation problem arises from a signal that consists of sinuoids plus an additive noise. Three algorithms are presented: EM interval method (EMIM), grouped coordinate descent based on interval methods with and without least squares. The estimation problems are decomposed via the EM algorithm and then the interval methods are applied for global optimization. EMIM has the best performance when drawing the main attention on estimation accuracy, although it has the drawback of computational complexity and slow convergence.