Research On An Improved COID

Data Mining, which is Knowledge Discovery in Database, is to find those information with special laws, special significance and potential value from a number of data. It is a new cross subject which combines machine learning, artificial intelligence, databases and other disciplines. And it has a wide range of applications in the financial, economic, demographic, life cycle and other fields.In recent years, with the development of communication technology, wireless communication environment becomes increasingly complex and communication signals in a wide frequency band are usually modulated in different modulation methods. So how to effectively recognize these signal modulation methods has become a focus of signal recognition field. Besides, with the continuous development of human science and technology, data processing is required to have even greater storage capacity, faster speed, and be more accurate when processing outliers data.Because clustering algorithm is an important part of data mining, its development has also been widely concerned. However, if there are outliers in the data set, using most clustering algorithms often cannot obtain accurate results. So a COID algorithm (Cluster-outlier Iterative detection) is put forward.COID algorithm combines clusters and outliers, and we detect outliers by the relationship between them to reach an reasonable clustering. To improve the practicability of this algorithm, COID algorithm is further improved. Finally, experimental results show that the improved COID algorithm has better feasibility, effectiveness and accuracy.In this paper, for several typical modulation methods, firstly the received signal after pretreatment is clustered through using improved COID algorithm; then dynamically excluding outliers and getting the value of the effectiveness evaluation function; finally the function value as the characteristic parameter is input to support vector machine(SVM) in order to recognise the modulation method. This method has great flexibility and intelligence in dealing with data points, so when dealing with the data in low signal noise ratio (SNR) environment, it also can achieve better modulation mode recognition.Concretely, research work can be divided into the following several parts:1.Introduces some common algorithms in data mining, and explains the concepts of the data characteristics distinguishing and correlation. Focuses on the classification of clustering algorithms and their advantages and disadvantages.2.Because the traditional clustering algorithm can not deal with outliers, a new clustering algorithm COID is presented and from clustering cluster center selection to the accuracy of the iterative process are made thorough research. Then the initialization-independent Prim cluster center algorithm is proposed to make the computing process more logical. Finally, a large number of simulation results show the improved algorithm COID has better feasibility, effectiveness and accuracy.3.the improved COID algorithm combining with support vector machine theory is used for modulation identification which using different constellation parameters to identify the communication mode of the signal modulation.In the case of low SNR, it also can accurately output the modulation method, and its recognition rate is higher than K-means clustering and the original COID algorithms.