The disclosed iterative decomposing method based on fixed training sequence filling comprises: taking current frame training sequence as the cycle prefix of a virtual frame included current frame data and next training sequence to make the linear convolution of virtual frame and channel impulse response as the cycle circular convolution. This invention introduces much convenience to obtain data and training sequence estimation by frequency-domain DFT method and separate the data and sequence for more precise channel estimation. This computer simulation results show that this invention has little loss on system error code performance, and simplifies balance process.