Abstract:A general difficulty of using singular value decomposition (SVD) to split signal and noise subspaces is in the right choice of effective rank. The commonly used method toward this end is to use a fixed threshold. Yet despite its apparent physical significance, the lack of adaptability has strongly limited the popularity of subspace approach in line enhancement. In order to surmount this drawback, a cluster method based algorithm for determining the effective rank is proposed in accordance with the a priori information obtained from the time-frequency
distribution of the noisy sinusoids. Simulation results show that the methodology advocated is effective for solving a class of multiple line enhancement problems.