In order to improve the performance of network sorting of frequcencyhopping (FH) signals under the underdetermined condition, an algorithm based on the sparse component analysis (SCA) is proposed. When estimating the mixing matrix, time frequency (TF) ratio matrices are obtained through S transformation, and the single source point is obtained by using the same direction of the real part and the imaginary part of observation signals. Then the mixing matrix is estimated by using the variance algorithm. Then separating signals are obtained by the improved subspacebased algorithm in the TF domain, and separating signals are obtained by S inverse transformation in the time domain. The proposed algorithm improves the mixing matrix estimation precision through the detection of timefrequency monophyletic points, with better separation performance and antinoise performance, and is effective to realize the asynchronous network or synchronous network sorting of FH signals in the underdetermined blind separation of FH signals.