Study On Multi-Dimensional Overlapped Digital Modulated Communication Signal Modeling Algorithms In Time Domain

In modern warfare, electronic reconnaissance system is faced with complex andchangeable electromagnetic environment. Signals are more and more intensive，frequency bands are more and more crowded and reconnaissance equipment oftenworks with single antenna, which often lead to receiver receiving multi-dimensionaloverlapped signals in time, frequency and spatial domain. In the above cases, theextraction of information and the analysis of signal become very difficult. At the sametime, the conventional signal separation technology such as frequency domain filteringand spatial filtering is failed under the condition of signals overlapping inmulti-dimensional. In the1990s, a signal separation technology based on independentcomponent analysis theory was raised in signal separation aspect and it is called blindsource separation. However, recovering multiple source signals from only one channeldate is a serious morbid problem in mathematics. If time-frequency overlapped signalscan be modeled with high fitting degree combining the signal characteristics, it will begreatly benefit to the analysis and separation of the signals. Time domain modelingalgorithms are studied for the single sensor time-frequency overlapped signals in thisdissertation. Its main content is summarized as follows:（1）Cosine base, Walsh base and Haar wavelet base are analyzed. Based onDiscrete cosine transform, Walsh transform and Haar transform, the model coefficientsof time-frequency overlapped communication signals using basis function expandingare obtained. Then the application of basis function model in blind source separation isanalyzed.（2）Based on data fitting mixed communication signal modeling algorithm isstudied. Based on autocorrelation of time-frequency overlapped communication signals,the normal equations of autoregressive model are established. The coefficients ofautoregressive model are obtained using Levinson-Durbin iterative algorithm. Then theoptimal model order is determined with order determination criteria, and the orderdetermination criteria is also extended to moving average model and autoregressivemoving average model. （3）Due to the nonlinear of normal equations, moving average model isapproximated by higher order autoregressive model. The autoregressive moving averagemodel can be converted to a cascade of autoregressive model and moving averagemodel. So the two model coefficients are easily obtained after autoregressive modelestablished.（4）Multi-channel modeling algorithm is studied based on signal decompositionand different carrier frequency. Single mixedsignal is decomposed into multi-channeldata based on wavelet packet decomposition and independent sub-basis functions areextracted based on Independent component analysis.The dimension of received signalis extended using independent sub-basis functions. Combining signal decompositionthought, single mixed signal is decomposed into multi-channel data using empiricalmode decomposition and multi-channel model is established. Then single mixed signalis converted into multi-channel observation data based on different carrier frequency ofsource signalsand multi-channel model is established. At last signals are separatedthrough over-determined or determined blind source separation algorithm.