Traditional time-frequency analysis is restricted by the Nyquist sampling theorem. As the amount of information increases, higher requirements are needed in sampling rate, transmission velocity, and storage space. Moreover, bilinear Wigner-Ville distribution is suffered from cross terms when processing multi-component signals. Using the kernel function based methods to suppress cross terms can decrease time-frequency concentration. In this paper, compressive sensing is combined with time-frequency analysis to solve the above problems. Under the framework of compressive sensing based time-frequency analysis, the restriction of Nyquist sampling theorem can be lessened, and the Wigner-Ville distribution can achieve suppressed cross terms with high time-frequency concentration. Simulations are provided for mono-component signal, multi-component signal, and bat sound signal, based on different window functions such as the rectangular window or the Gaussian window, to verify that the compressive sensing based time-frequency representation reconstruction is superior to the traditional reconstruction method. Moreover, we analyze the relationship between different sample regions and the performance of the reconstructed time-frequency representations, in terms of the mean-square-error (MSE) and time-frequency concentration measurement (CM).