Recognition Research Of Abnormal Radio Signal Based On Support Vector Machine

With the rapid development of modern information technology, the radio spectrumresources is becoming an important resource for the widespread use of human society,However, various interfering signals appear constantly, which has increased the difficulty ofradio monitoring. As we all know the identification of signal type is one of the most importanttask in signal monitoring, its identification reach of abnormal signal has very importanttheoretical and practical significance.Radio signal identification is a typical problem of multi-classification. what’s more,support vector machine (SVM) is good at solving high-dimensional classification problems,and it has been widely used in many other fields. This paper we apply the SVM to radiodetection, then propose the radio identification method of abnormal signal types of FMbroadcast band and C-band based on support vector machine model, and the main contentsare as follows:1. An overview on various SVM kernel function theory、modeling and parameterselection are discussed, their impact on the effectiveness of support vector machine classifierare analyzed and the abnormal radio signal recognition experiment which select for differentparameter values and penalty factor is finished. Experimental results show that the choice ofkernel function, parameter settings can bring a significant effect on support vector machinegeneralization capability, thereby affecting the recognition effect of radio abnormal signalrecognition system.2. Take the radio signal of broadcast band as the research background. We collect thedevice data, preprocess data and extract feature, build support vector machine model, selectthe appropriate parameters, and ultimately achieve broadcast abnormal signal classification.In order to fatherly validate the feasibility that the SVM method is able to apply to identifythe radio signal, under the same conditions, we select BP neural network classification as thecomparative experiment. The comparative study found that our optimization parameters andextracted feature has a good results using the methods of SVM classification, furthermore therecognition accuracy rate is higher than traditional neural network classification methods.3. On the basis of two categories, we achieve the multi-classification of support vectormachines as the research object to C-band signal for the study. This paper mainly constructtwo kinds of multi-class support vector machine based on the " one to one (OAO)" and "oneto many (OAA)", then analyzes the advantages and disadvantages of the two algorithms andconducted experiments comparing with the BPNN. Experimental results show that although the results of these two multi-classification SVM methods are different, but its classificationaccuracy are both significantly higher than the classification accuracy of BP neural network.4. In the experiment of C-band radio signal recognition, in order to further investigate thefactors that affect the recognition performance of SVM, the paper we use differentoptimization algorithms and kernel functions. The comparative studies show that signalrecognition rates of PSO algorithm and RBF kernel function are higher and better than otheralgorithms and kernel function methods, that is to say that SVM is an effective new methodfor C-band abnormal signal identification.