A Research Of FLOCS-based Automatic Modulation Classification On Graph

Automatic Modulation Classification(AMC)is a communication signal processing technology widely used in military and civilian fields.It classifies received signals dam-aged by noise and interference with pre-set criteria.Automatic Modulation Classifica-tion on Graph(AMC_G)is an emerging AMC technology.At present,a comprehensive theoretical system has not yet been formed,and there is big research gap needs address-ing and great potential of future applications.In AMC_G,the Digtal Signal Processing on Graphs(DSP_G)theory and AMC thechnology is combined,so that the classification task of received signals are generalized from traditional time or frequency domain to graph domain.A graph-domain representation of modulated signal is established based on it,from which stable features are extracted,achieving accurate classification of modulation type.The assumption of Gaussian distribution noise in the conventional AMC_Gapproaches can lead to the analytical optimal solutions.However,the actual noise/interference in prac-tice does not comply with Gaussian distribution,which cause a decline of classification performance of conventional AMC_Galgorithm.Consequently,new robust and efficient AMC_Gtechnique for Non-Gaussion noise is practically significant.Funded by the Na-tional Natural Science Foundation of China,this paper focuses on the Fractional Lower Order Cyclic Spectrum(FLOCS)based automatic modulation classification algorithm on graph,and the primary content is as follows:1.A graph-domain mapping and representation method based on FLOCS was pro-posed to address the problem that the second-order statistics is destroyed by Alpha stable distribution noise.A fractional lower order cyclic analysis was performed on the commu-nication signals,FLOCS was then calculated,and the received signal was mapped to the graph-domain based on it,and therefore corresponding graph-domain representation was established.2.A feature extraction and classification algorithm based on variable-scale modula-tion candidate sets was proposed to address the lack of scalability of the feature extraction algorithms in existing AMC_Galgorithm.Based-on the graph-domain representation of the communication signal,the optimal features were extracted using the Kullback-Leibler di-vergence to build a graph-domain feature library for the standard modulation candidates set,and meanwhile,an efficient update of the graph-domain feature library was achieved when signals were modified in modulation candidates set.Combining this with the Ham-ming distance based graph-domain classifier,the modulation type was automatically rec-ognized.3.A Probability of Correct Classification(Pcc)theoretical estimation method was proposed to address the problem that the performance of existing AMC_Galgorithm was evaluated only through simulation.In addition to the simulation experiment,a Pcc theo-retical evaluation link was added for a high-precision estimation of Pcc and fast update of it when the modulation candidates set was resized.Simulation results showed the graph domain mapping algorithm based on FLOCS was effective and verified the scalability of the feature extraction algorithm based on the variable-scale modulation candidates set.It confirmed the theory of Pcc estimation method can realize high-precision Pcc measurement and fast update of it when the mod-ulation candidates set was resized.