Machine Learning Classification Algorithm And Database Based Spectrum Sensing

Spectrum sensing,as one of the key technologies and research contents of cognitive radio,is of great significance to solve the contradiction between the increasingly tense spectrum demand and efficient spectrum utilization.In order to overcome the shortcomings of traditional spectrum sensing in real environment and the difficulties of realizing conditions,this thesis explores the application of database and machine learning methods to spectrum sensing technology.In the actual cognitive radio network environment,due to multipath loss,complex wireless propagation environment and other practical factors,the signal propagation attenuation is serious,which leads to the inaccuracy of secondary user(SU)'s identification for licensed spectrum status,and it is difficult to effectively support SU's accurate judgment of primary user(PU)status,coupled with hidden nodes.Problems,etc.,are more likely to cause SU interference on PU and other serious problems.To solve these problems,a hybrid AdaBoost algorithm based on machine learning is proposed in cooperative spectrum sensing.The algorithm uses SVM as the first sub-classifier and single-level decision tree as the remaining sub-classifiers.The weight of each subclassifier is determined by its classification performance.The better the classification performance,the higher the weight of the classifier.Finally,according to the weight of each classifier,the final classifier is formed,which effectively enhances the performance of the classifier as a whole,improves the probability of detection,and better protects PU from SU interference.In view of the limited performance of traditional spectrum detection algorithms and the high cost of SU implementation in order to ensure the detection effect,a spectrum sensing scheme based on geo-location fingerprint database is proposed.The scheme establishes a geo-location fingerprint database by estimating the time when the base station reference signal arrives at SU.Before detecting the PU signal,SU periodically reports the fingerprint information of the current location.The network end uses fingerprint matching mechanism to lock the geographical location of SU,and reads the available status of the corresponding PU licensed band at the location,so as to control the further operation of SU,that is,access PU band,detect PU band or not access PU band.On the basis of guaranteeing the expected perception performance,this sensing mechanism can greatly simplify the perception process of SU terminals,reduce a large number of unnecessary perception operations,and significantly reduce the energy consumption of SU in the perception process.