Abstract: The traditional denoising method, such as wavelet threshold, is only valid for Gaussian noise, but is powerless for pulse jamming. Singular spectrum analysis developed in recent years can be a good filter at high SNR conditions of these two types of noises, but the process of noise reduction involves a certain subjective factors, and is subject to restrictions of matrix perturbation theory. Moreover, the ability of denoising will decrease with the lower SNR. For the above situation, an improved algorithm was proposed, which applied rank minimization theory to singular spectrum analysis. Simulation results show that denoising effect of the improved algorithm is obvious, which can maximize the reduction of the mean square error of the signals and improve signal to noise ratio; enhance versatility of singular spectrum analysis.