Abstract: Abstract: Time-frequency overlapped signals existing in the wireless spectrum detection and military reconnaissance receiver, which are difficult to be separated and extracted information by using the time-frequency domain filter and conventional blind source separation method. According to the independence and sparsity of single-channel time-frequency overlapped modulation signal in cyclic spectral domain, a new method is proposed to achieve two source signals separation through the cyclic spectrum domain filter in this paper. This method can be stated as follows: Firstly, the cyclic frequency θ is found through the cyclic frequency estimation by the periodogram, where the modulated signal has the maximum spectral correlation, using θ as frequency shift of the LCL-FRESH filter, the modulated signals with the cyclic frequency are extracted from single-channel time-frequency overlapped signal by a filter. Then, time-frequency overlapped QPSK/ BPSK signal adding white noise is simulated in the MATLAB environment. The result indicates that the two source signals can be separated effectively with this method.