Noisy Communication Signal Detection Technology Research

In modern wireless communication system, lots of modulation signal has been used.However, because of the complicated communication environment, signals will carrynon-negligible noise with a little prior knowledge, which makes the processing of the signalvery hard and it will be much harder to get the useful signal for the non cooperation side.Therefore, whether for military or civilian, how to get and identify the signal effectively isvery important.The detection problem of seven usually used digital communication signal (2ASK,4ASK,2FSK,4FSK,2PSK,4PSK,16QAM) under Gaussian noise is researched in the paper.Firstly, sparse decomposition matching pursuit algorithm is used to decrease the noise ofthe signal and to reconstruct it. Relative to the noise of Gauss distribution, any signal willhave some structure, whose characteristics are matching to the properties of atoms. When thesignal are corresponding to the useful atoms, they will be extracted, then the remaining part isjust the noise. Then use the similarity coefficient to identify the reconstructed signalmodulation method.Simulation results show that: the algorithm is suitable to the noisycommunication signal.When the Gauss noise are superimposed on more than one communication signals, theusually used identify algorithms are useless, as a result of which, blind source separationtechnology is used in the paper, and the basic theories and corresponding mathematic modelsare given. What is more, for the Gauss noise condition, the paper adds an albino offsettechnology to the classic Fast ICA(Fast Independent Component Analysis) algorithm toreduce the deviation caused by the Gauss noise, which forms the offset Fast ICA algorithm.Simulation results show that: under different signal-noise ratio, offset Fast ICA is moresuitable to the separation of the noisy signal than classic Fast ICA, and which can achieve abetter separation performance index.Finally, based on the support vector machine theories, the paper analyzes and comparesthe multiple classification models that extended from the two classification models. For theidentification of seven modulation signals, one to many multi-classification and binary treemulti-classification methods are proposed based on the five instantaneous information.Simulation results show that: Compare to the one to many multi-classification, the binary tree multi-classification methods has a higher identification accuracy and efficiency, and thebinary tree multi-classification methods can achieve a93%identification rate when the signalto noise ratio is not lower than10dB, which illustrates the feasibility of the method.