Study On Single Channel Separation Algorithm For Digital Modulation Signal Based On ICA And Matrix Factorization

Communication reconnaissance is very important and it is the foundation of thecommunication countermeasure. Reconnaissance system faces a complex and changingelectromagnetic environment. Signals are more and more intensive and band is moreand more crowded. In the high-density signal environment, communicationreconnaissance equipment usually intercepts multiple time-frequency aliasing signalswith single antenna. In order to accurately obtain the communication parameters of anoperational object, single-channel time-frequency aliasing signal separation technologyhas become the core technical issue which needs to be solved urgently for thecommunication reconnaissance equipment development. For this applicationenvironment, this thesis conducted research on Single-channel time-frequency aliasingsignal blind separation algorithms, aiming to provide a theoretical basis forcommunication reconnaissance equipment.The main works of this thesis are as follows:1. The mathematical model of single channel source separation problem isestablished. A modeling method for transition between single-channel andmulti-channel is studied, and this model can apply to the conventional multi-channelsource separation algorithm.2. A single channel separation algorithm based on independent component analysisis studied. The separation criteria and optimization algorithms based on independentcomponent analysis are summarized. Through constructing a multi-channel separationmodel and choosing reasonable parameters for the model which are combined with thecharacteristics of the digital modulation signal, a plurality of independent componentsare decomposed by independent component analysis. Then the appropriate componentsare selected to reconstruct the source signal, and signals are separated finally. Theproblems of how to set the model parameters and select the component are analyzed indetail. This algorithm is not very complicated, and it has certain engineering applicationprospects.3. A single channel separation algorithm based on non-negative matrix factorization is studied. The characteristics of non-negative matrix factorizationalgorithm are analyzed. Through combining the characteristics of the non-negativematrix factorization algorithm and the digital modulation signal, the determinantstandard and minimum related constraint are added to the original objective function,and the weight of each constraint condition is balanced. Finally, signals are separated byusing improved nonnegative matrix decomposition on the multi-channel observationdata. As the non-negative matrix factorization algorithm is not based on statisticalproperties, there is no limit to the independence of the signal. It expanded the traditionalblind separation assumptions, and provided a new approach for the blind separation.4. A single channel separation algorithm based on basis-expansion is studied.Mixed signals are decomposed to several intrinsic mode functions which are seen asbasis vectors by using improved empirical mode decomposition. Combining with thecharacteristics of digital modulation signal, signals which are aliasing in time-frequencydomain are separated by using improved empirical mode decomposition for severaltimes. Since the basis-functions are generated adaptively by empirical modedecomposition which is based on signal feature, this algorithm doesn’t need trainingdata.