Research And Design On VoIP Calls Sorting

As an important product of fusion of Internet and telecommunications networks, VoIPservices have gained rapid and sustained development in the whole world in recent years.Although our country has not yet officially open the VoIP business operations, there still exist alarge number of illegal and implicit VoIP gateway operations in the network because of the lowthreshold of VoIP technologies and the high-profit of the corresponding business. These Gatewayoperations not only cause a great diversion of the traditional telecom business, but also provide aconvenient channel for the spread of the bad information. How to implement effective control ofVoIP services and purify the environment of telecommunication networks has become an urgenttopic to be studied.This dissertation is supported by the theme project in the field of information technologybelonging to "Twelfth Five-Year Plan" National High-Tech Research and Development Programof China (863). The main task is to conduct research on the key technologies of real-time sortingof the VoIP calls, which can provide technical support for supervising the VoIP business.Thispaper proposed the technical idea which first uses VoIP traffic data to find the operating VoIPservices gateway address and then sorts VoIP calls in real time through the obtained gatewayaddress database. According to the idea, two types of the key technologies are focused in thispaper. The first one is the VoIP traffic identification technology which has nothing to do with thespecific protocols. Based on this, the rapid screening of various types of VoIP service data in theInternet can be achieved. The second one is VoIP gateway node identification technology whichis used for finding VoIP gateway addresses from the VoIP business data quickly. On this basis,the paper has designed the real-time sorting subsystem of the VoIP calls in combination withother mature technologies.The main innovations and achievements of this dissertation are as follows:AVoIP traffic identification method based onC4.5decision tree is proposed. Traditionalmethods can only identify the single object, are unable to identify unknown applications and lackof adaptability to the network. Aiming at such shortcomings, according to traffic classificationbased on machine learning theory, several statistical characteristics to reflect the VoIP voicepacket transmission law are selected, and a method of VoIP traffic identification based on C4.5decision tree is designed. The test results of the classifier built on Weka platform show that theC4.5decision tree classifier is much better than the performance of other machine learningclassification algorithms. Moreover, the two new features can reflect the nature of the continuousVoIP voice packet law, which lead to the significantly superior recognition accuracy compared with the traditional characteristics.A SVM-based VoIP gateway node identification method is proposed. Learning from theidentification of other nodes in the Internet, in the light of the rules and characteristics of theVoIP call, this algorithm summarized a number of features that can be used for nodeidentification from the difference between the VoIP gateway node and the node of similar PC,and extracted the statistical characteristics combined with traffic distribution in the time domain.The features are merged to describe the training samples. Considering the advantages of theSVM classification in the case of small sample size and high dimensionality, the SVM classifieris built on the Weka platform. The experimental results show that the SVM recognitionperformance of the gateway node are significantly superior to other classification algorithms. Atthe same time, the experiment also assesses the contribution of the selected features forclassification, which further confirms the validity of the selected feature.A VoIP call real-time sorting subsystem based on the above-mentioned technologycombined with the existing mature technology is designed. The subsystem takes front-end toback-end co-processing architecture. Each VoIP call belonging to known gateway achievesreal-time accurate sorting in the front-end through the mature5-tuple match and signalingmonitoring technologies. In the back-end, it used traffic identification and gateway nodesidentification techniques to analysis Internet traffic data rapidly, discover new gateway addressin a timely manner, and load the front-end. As a result, the two-ends can feedback mutually,which formed the VoIP calls sorting mechanism with self-learning ability.