Research On Blind Separation For Frequency-hopping Signals Based On Time-frequency Analysis

With the performance improvement of Frequency-Hopping(FH) communication system, FH communication reconnaissance face severer test. Traditional signal processing methods cannot adapt to the signal frequency changing with time. According to the characteristics of FH signals, it is imperative to design the targeted algorithm of FH communication reconnaissance. Based on time-frequency analysis, the FH signals underdetermined blind source separation(UBSS) model is established and investigated to solve the problem of FH signals blind separation.The implementation principle and signal characteristic of FH communication system are discussed first in this paper. In order to describe the FH signals more accurately, the characteristics of common time-frequency analysis algorithms are analyzed. This paper has focused on the research of FH signals blind separation under the linear time-delayed underdetermined model.As for the estimation of the mixing matrix, the limitation of current algorithm for single source time-frequency(TF) points extraction are analyzed first. And the standard for single source TF points detecting is introduced. The column vectors of mixing matrix are estimated by the time-frequency ratio matrix constructed by the selected single source TF points. Simulation results show that the improved method has better mixing matrix estimation performance especially in low signal-to-noise ratio(SNR) conditions.Aiming at the problem of mixing matrix changing with FH signals frequency, the time-varying mixing matrix detecting algorithm is presented based on the method of locating frequency hopping points. And the TF observation data is processed piecewise according to the frequency hopping points. And the column vector of estimated mixing matrix estimated is rearranged according to the direction of arrival(DOA), avoiding the order ambiguity between estimated signals. In addition, to separate the FH signals in the TF domain, the subspace projection algorithm is applied for signal separation. And the method of TF domain de-noising is proposed to improve the signal separation performance. The simulation results demonstrate the effectiveness of the proposed FH signals BSS algorithms.