Applying Pattern Recognition To Interference Analysis In GSM Networks

With the rapid development of mobile communication network and GSM network environment increasingly complex, there emerged many problems that affect call quality, such as drop rate increasing、reducing base station coverage、network indicators sharp decline, so network optimization and maintenance work which mainly integrate existing resources becomes a key work. And interference analysis becomes a difficult work of network optimization because of various interference types and complicated positioning process. Now uplink interference location work mainly use live single point scan, which is strongly dependent on the long-term work experience summary of staff and can’t adapt to changing network conditions quickly and mixed interference identification needs several confirmation test of scene. So this method is difficult to spread widely.To solve these problems this paper uses FAS data which on behalf of the interference characteristics to locate cell from two points:on the one hand cells are located which have similar interference graphics using shape matching method, on the other hand cells are learned which have similar characteristics using pattern recognition method. The spectrum goes through shape representation, shape matching, type identification to identify whole network-wide cells which influenced by single interference. And then feature vector of the training samples is extracted to input to random forest learning model to establish stable model after perfect template library. Then belonged type classes of cells are predicted based on learning model. Through this method workers can take necessary measures to monitor network station effectively to locate cells received single interference and cells mixed interferences.This system can locate cells which can’t recognized by other methods using FAS, NCS and MEE data, to illustrate the validity and completeness of this method. The result of mixed interference shows that these cells meet characteristics of belonged types. The combination of both methods enhances the accuracy and completeness of positioning cells. This method applies shape matching and pattern recognition technology to interference cells location and provides a new solution, which not only can quickly identify single batch interference, but also can identify mixed interference, and can identify other possible type interferences in future.On the basis of this method, this system use C#language, SQL Server Database and OpenCV in.NET platform to develop single interference module and mixed interference module. After actual data verification it achieved a good result.