Research On Blind Source Separation Technology In Communication Reconnaissance

With the development of communication technology and the increasing of complex electromagnetic environment, the physical environment of the electromagnetic signal becomes more dense. Multiple signals are often aliasing in both time domain and frequency domain. In order to meet the needs of communication reconnaissance,we need to separate the enemy communication signals from the mixed signals firstly. At present, both the time-frequency domain separation method or airspace separation methods all can not separate the same frequency aliasing signal. This brought great difficulties to communications reconnaissance systems in analysis and identification of non-cooperative signals Therefore,there is an urgent need to study the feasibility processing methods to achieve separating of communication signals effectively. Blind Source Separation techniques just to solve this problem. It can separate independent source signals from the mixed signals which received by the antenna using a weak condition that the source signal are independent,without knowing any prior information of signal. This paper mainly studies the blind signal separation technology applied in communication reconnaissance systerm.Based on in-depth study of the theory of blind source separation,this paper introduced time average fourth cumulant(TAFC) as a measure of independence, theoretically proved blind source separation algorithm can successfully separate determinate cosine signals, or even same frequency cosine signals. Which inspires people to apply blind source separation techniques on complex cosine signals (communication signals, radar signals). We applied the blind source separation technique to existing communications reconnaissance systems, and proposed software and hardware design solutions based on NI PXI communication reconnaissance of the receiving system. For communications reconnaissance signals,simulation tests proved the aliasing extent of each source spectrum does not affect the separation efficiency of the algorithm. If the sampling frequency meets certain requirements,SNR and sampling time are the main factors which affecting the separation result.This paper proposes an improved algorithm based on negative entropy FastICA. The improved algorithm combined the steepest descent method and Newton iteration algorithm,and further using Part of the Newton iteration in Newton iterative process,so that the improved algorithm has higher and uniform iteration speed. Through simulation experiment,we prove it has a good effectiveness on separation of digital modulation mixed signal.According to frequency hopping signals have typical non-stationary characteristics, This paper presents an algorithms of Nonorthogonal Joint Diagonalization and Zero Diagonalization for source separation based on time-frequency distributions(JDZD). This algorithms not only uses the diagonalization part of auto-source term of time-frequency matrix, but also use the zero diagonalization part of cross-source term of time-frequency matrix. It established a unified cost function, to reduce the possibility of eigenvalues degradation by increasing the time-frequency points. Theoretical analysis and simulation results show that this method can effectively separate multiple frequency-hopping network stations in low SNR conditions.