Data Mining Technology Of Spectrum Sensing Data

With the rapid development of wireless communication technologies, the ever increasing demand for limited spectrum resources will eventually cause spectrum scarcity problem. The contradictions between the low utilization of the fixed spectrum allocation policy and the shortage of the radio spectrum resources are becoming more and more prominent. Cognitive Radio offers a solution to this problem, which intelligently detect unused spectrum and share it without harmful interference to primary users. It has been a research hotspot on how to mine and find out spectrum hole and use it intelligently to meet the increasing need for bands. In This dissertation, data mining technology is used to find the spectrum usage pattern for cognitive users. The main work and innovations of this dissertation are mainly as follows.Firstly, a cognitive radio communication system based on spectrum data mining has been comprehensively presented in this dissertation. Compared with the traditional CR system, the new system has been improved with the module of control center which were equipped with data mining function. With the data mining module, intelligently and efficiently channel access can be reached by CR users.Second, this dissertation designed a sequential pattern mining scheme of0-1time series which were the result of the original spectrum data compared with the decision threshold, higher sentenced to0, lower sentenced to1. A time series segmentation algorithm based on sliding window of approximate entropy is proposed to find the change moment of pattern. And then channels’ usage patterns of every segment were provided for CR users’dynamic access to enhance the spectrum efficiency. The TV broadcasting cognitive wireless communication system is simulated and implemented to verify the algorithm. Thirdly, this dissertation improved SVM algorithm with two aspects:the clustering preprocessing of the training set and the correlation selection of testing set. Clustering method was used to divide the training set into subset of different modes, which can avoid the entire time series using only one functional model and can be more flexible and precise. The correlation analysis was presented to find out the optimal testing set and to build the better model to predict the future trend. The prediction results of the load data and the number of the authorized user’s time series illustrate the effectiveness of the multi-class model and the relevant validation.This thesis studied the data mining technology of spectrum sensing data, recommended the data mining information to the CR users for dynamically spectrum access and helped them to make the spectrum access strategies, which is meaningful in both theoretic and application.