This paper introduces the time-frequency decomposition of the sampled data series using the Prony method, and improves it by automatically identifying the number of main mode terms, so that it can be easily convergent and solved. Since Prony decomposition can use rectangular window segmentation to reduce the number of points, the combination of sine functions with oscillation attenuation can be obtained. Time-frequency decomposition is relatively clear and concise, and this paper also compares with other classical methods in illustration.