Abstract: The digital modulation with the characteristics of constant envelope and compact spectrum is one of the research emphases of digital communication. However, few research was carried on the single channel blind separation of this kind of signal. It is of great significance to solve the problem of single channel blind separation of continuous phase modulated signals. The paper is based on optimal Bayesian estimation criteria and the real posterior probability distribution of the system state is approached by means of important functions. The improved particle filtering algorithm is applied to solve the problem of single channel blind separation of continuous phase modulated signals and realize the sequential estimation of the symbols and unknown parameters. The proposed algorithm makes use of more information of received signal waveform through over-sampling and recursion of the data. So the noise can be effectively suppressed and ISI brought by continuous phase can be overcome. Simulation takes GMSK modulated signal as example which is widely used in communication and the results show that the proposed algorithm is superior to standard particle filtering algorithm and has favorable performance in symbol estimation and parameter convergence.