Signal parameter estimation is a crucial issue in SAR/ISAR imaging, especially for multicomponent linear frequency modulated (LFM) signal with single degree of freedom. A new method of parameter estimation based on sparse signal representation is presented in this paper, which expands signal on a set of over-complete basis. The method is analyzed and validated for performance through simulation, with three commonly used signal sparse representation algorithms compared, including BP, FOCUSS and Sparse Bayesian Learning. The result shows that Sparse Bayesian Learning performs better in sparse components than the other two algorithms, which can estimate signal parameters more efficiently.