Aimed at the problems that in the face of the increasingly complex electromagnetic environment, the feature recognition algorithm for the blind source separation of multifrequency hopping signals is heavy in computation, and the separation result is inaccurate, in combination with the variable speed frequency hopping signal, a strategy of accelerating the hopping speed and varying the hopping speed is adopted. Simultaneously the independent component analysis method is utilized for dealing with the blind separation problem of variable speed frequency hopping signals, and the negative entropy maximization algorithm is used to accelerate the separation speed of traditional independent components. The simulation results and actual frequency hopping data show that compared with other methods this algorithm can effectively separate the multiple variable speed frequency hopping signals without any prior information and low SNR. At the same time, the time domain waveform of the variable frequency hopping signal and the corresponding frequency hopping pattern can be accurately recovered, in the case of 20 dB SNR, the separation similarity coefficient can reach 99%. The analysis mentioned above provides a new solution for the blind separation problem of the variable frequency hopping signal.