A common method for parameter estimation is adopting the inherent cyclostationary properties which vary periodically in the communication signals. Since the communication signals exhibit sparsity at the cycle frequency domain, the random measurements can be utilized to reduce the number of samples and lighten the burden of sampling hardware, and then the parameter estimation can be completed based on the compressive samples. However, in this kind of sparse modeling, it is inevitable that the continuous parameter space is discretized into a finite set of grid points, which will lead to basis mismatch. The signals cannot be expressed sparsely under an assumed finite dictionary, e.g. Fourier basis and DFT basis, therefore the parameter estimation accuracy is seriously affected by basis mismatch. In this paper, a gridless CA reconstruction which can locate the nonzero cyclic frequencies on an infinitely dense grid is proposed by utilizing the atomic norm to describe the continuity and sparsity of the cycle frequency domain. Numerical results demonstrate that the proposed method can reduce the mean square error of estimation effectively.