An efficient algorithm for slowly-varying frequency estimation

A new method is proposed in order to estimate slowly varying frequencies of complex exponentials. The method is based on computing a suboptimal signal subspace projection matrix, using some columns of the estimated data covariance matrix, and employing this suboptimal matrix to obtain a minimum-norm vector in the noise subspace. Using that suboptimal vector, the frequencies can be estimated. The proposed algorithm is very fast and compares advantageously with the algorithms based on the rank-one updating of the data covariance matrix, which are considered a good choice, given their performance/computational burden ratio. This good performance is shown by simulations