Research On Signal Separation Method Based On Candecomp/Parafac Decomposition And Time Frequency Analysis

In the electronic warfare,the method of using unmanned aerial vehicle(UAV)clusters for passive positioning to obtain the azimuth information of the target radiation source has attracted widespread attention.When the number of target radiation sources is large,the received signal of each UAV is the aliased signal of the original signals of multiple radiation sources through the transmission channel.In order to obtain accurate positioning information,it is a key issue to obtain the original signal of each radiation source from the aliased signals received by multiple drones.Based on the CANDECOMP/PARAFAC Decomposition(CPD)and time-frequency analysis,this paper studies the signal separation methods of the instantaneous aliasing model and the convolutional aliasing model.First,for the modeling problem of multiple UAVs receiving multiple radiation sources,the tensor decomposition algorithm and the generation of aliased signals are studied.The basic concepts and operations of tensors are summarized,and the algorithms of tensor decomposition are analyzed.The modulation information of the original signal sent by the radiation source studied,and the time-frequency characteristics are analyzed.The channel modeling tool is used to generate transmission channels containing various attenuation factors,and the original signals are mixed through the transmission channels to obtain aliased signals,thereby simulating the received signal of the UAVs.Then,researched and used the CPD method to solve the blind signal separation problem under the instantaneous aliasing model.Summarized the statistical characteristics of the aliased signals and stack the received signal into a tensor form which could be processed by the CPD method.Researched the iterative algorithm of CPD based on unconstrained optimization,and used CPD to process and evaluate randomly generated tensor data.Studied the non-negative constraint method and the proximal gradient update method which is based on coherence constraint terms.For the instantaneous mixing model which considers power attenuation,the performance of the separation results of the classical signal separation algorithm and the CPD method is compared.Finally,used the time-frequency analysis method to solve the problem of signal separation under the convolutional aliasing model.Researched and optimized the algorithm for estimating the number of sources based on Gerschgorin’s disk estimation,which improved the performance under low signal-to-noise ratio environment.The tensor rank estimation method is studied and applied to the estimation of the number of radiation sources.Researched and used the block tensor decomposition method to solve the signal separation problem of low-order multipath aliasing model.Researched and used the frequency domain separation algorithm based to deal with the multipath aliasing model,which solves the shortcomings of the block tensor decomposition method which has high computational complexity of and a known number of multipath.The separation results of the CPD and the frequency domain separation algorithm under the instantaneous aliasing model are compared,applied frequency domain algorithms to process aliased signal data sets containing multipath information.Using Crosstalk Index and signal-to-interference ratio evaluation indicators to evaluate the above separation results,it has a good separation effect.