Identification Of Weak Signals In Non-cooperative Environment

Recently,there have been various innovations in communications and sig-nal processing,the key challenge of which is how to guarantee the performance of Automatic Modulation Classification(AMC).However,the accuracy of tra-ditional AMC methods cannot satisfy the requirements of wireless communi-cation management.To solve the drawbacks of previous AMC approaches,we investigate the weak signal classification methods in non-cooperative scenar-ios.To improve the accuracy of classifiers,the weak signals should be prepro-cessed first and then identified.On the condition that the target is identifying a small number of signals,we should utilize the small-data for classification.To classify a large number of received signals,we should adopt the big-data.To classify mixed weak signals,we should separate mixed signals and then identify them.The main contents of this paper are as follows.1.Preprocessing for weak signals in non-cooperative scenarios.The frequency offset can severely degrade the performance of AMC meth-ods.However,previous approaches can only coarsely estimate the frequency offset and the approximation error dramatically degrades the accuracy.For channel parameter estimation,we proposed a clustering-based method to di-minish the frequency offset.Compared with traditional frequency offset esti-mation approaches,our proposed method achieves higher accuracy.To solve the drawbacks of the traditional spectral clustering algorithm,we proposed a novel spectral clustering based approach which achieves a better performance.2.Classification for small-data weak signals.The classification approaches for weak signals in non-cooperative scenar-ios can be divided into two main categories,that is,Likelihood-Based(LB)and Feature-Based(FB)methods.To solve the shortcomings of traditional AMC approaches,a dictionary learning based AMC framework is proposed.Further-more,we propose a novel dictionary learning method called FBCDDL which outperforms the previous algorithms:(1).FBCDDL is more robust to the in-terference than previous methods;(2).FBCDDL achieves a lower complexity than previous algorithms;(3).FBCDDL can guarantee its convergence while previous approaches cannot.Experimental results show that FBCDDL per-forms better over interference than other AMC methods with shorter execution time.3.Classification for big-data weak signals.With the development of communication systems,large-scale applications become more and more important.In this paper,we propose an AMC method using the Convolutional Neural Network(CNN)with a novel penalty in the loss function,which performs better than previous approaches:(1).our proposed approach extracts features automatically while previous classifiers cannot;(2).our proposed method achieves higher accuracy than previous algorithms.Ex-perimental results verify that our proposed classifier outperforms than previous approaches.4.Classification of mixed week signals in non-cooperative scenarios.Mixed weak signals dramatically degrade the performance of AMC algo-rithms.To solve the drawbacks of previous methods,we proposed a novel approaches which enjoys the advantages as follows:(1).our algorithm can be utilized in various scenarios while previous methods cannot;(2).the accuracy of our classifier is higher than that of previous approaches.