Research On MQAM Signals Modulation Recognition Algorithm Based On Clustering Theory

Modulation recognition of communication signals plays an important role in signal detection and signal demodulation and also is an important part of signal analysis, which is widely used in military and civil. In recent years, with the rapid development of modulation technology, new modulations appear constantly, because the MQAM (Mary Quadrature Amplitude Modulation) modulation mode is flexible, high bandwidth efficiency and more and more popular in satellite communication and network communication. Therefore, researching MQAM signals modulation recognition has the important practical significance. This paper presents adaptive subtraction clustering and semi supervised fuzzy clustering algorithm reconstructing constellation map of MQAM signals, introducing effective function and the constellation diagram circle radius, completing automatic recognition of MQAM signals. The main work and research results are as follows:The first, study on parameter selection:The initial clustering centers are chosen by adaptive subtraction clustering based on signal noise ratio. Field radius has a great influence on the choice of centers. The choice of these parameters is studied in this article, through simulation, finding the appropriate parameter values, obtaining the better initial clustering centers.The second, positioning clustering centers by semi supervised clustering:By the adaptive subtraction clustering to obtain initial clustering centers, the location of clustering centers may not be accurate in the low signal-to-noise ratio. Therefore, this paper introduces semi supervised fuzzy clustering algorithm to clustering two times, around each initial cluster centers tagging portion of sample points and according to the relations of each labeled sample points with centers giving membership values; Then, the clustering centers and membership values are updated by the guiding of labeled sample points. Theoretical analysis and experimental results show that high order modulation signals’ complexity and computation is reduced by positioning algorithm of semi supervised. The clustering centers position more accurate by guidance of the labeled sample points.The third, introducing effective function judging the rationality of clustering: Semi supervised clustering updating the centers position and then completing the constellation reconfiguration. Because the influence of the noise reconstructing constellation may exist deviation, thus affecting the classification and recognition of the receiving end. For this problem, this paper proposes clustering effective function to determining the clustering rationality and then judging the initial division of modulation. Experiments show that when the clustering results is same as the constellation map, the value of effective function is maximum, whereas a smaller.The fourth, the design of classifier:In this paper, the circle radius of constellation diagram and the effective function are used as characteristic parameters respectively to classification and recognition the modulation mode of MQAM signals.Because the value of effective function can determine the clustering rationality, so it can be used as the input of support vector machine classifiers and then classification recognition MQAM signals. Through the simulation experiments, the recognition rate is greater than or equal to 95% in the error rate 15%, so it can be competent the classification and recognition of MQAM signals. But this method is complex to calculate effective function value and cost long time to train classifiers’.Through the comparison of MQAM square constellation diagram with reconstruct constellation map in paper, each point on the constellation diagram can be partition circles of different radius.According to the standard constellation diagram determining the scope of each order signal radius value, so the circle radius of constellation diagram can be used for identification classification modulation mode of MQAM signals. This method is easy to extract eigenvalue of the circle radius of constellation diagram, less computation, modulation recognition judgment is convenient; In low signal to noise, there is still a high recognition rate when the number of clustering centers is not accurate and the centers’position exists deviation.