Wireless Spectrum Prediction And Application Based On Data Mining

Radio spectrum is a valuable natural resource.As the transmission carrier for wireless communication,it plays an important role in radio communications.In recent years,with the rapid development of various communication technologies and the expansion of radio business,spectrum resources become increasingly tense.However,a series of spectral measurements have shown that some licensed spectrums are underutilized under the fixed spectrum management scheme.To resolve the above contradiction,the concept of cognitive radio(CR)came into being.In the CR network,secondary users(SUs)can dynamically access the free authorized bands,which contributes to improving spectrum utilization and alleviating the shortage of spectrum resources.Spectrum prediction helps to reduce spectrum usage conflicts during access and ensure the communication quality of primary users(PUs)and SUs.Meanwhile,it can guide SUs to spectrum detection,which helps to reduce the time consumption and energy consumption of the sensing process.At present,spectrum prediction has become an important part of cognitive radio research.In this paper,the continous 48-hour spectrum measurements were conducted in the uplink bands of GSM900 service(ranging from 890 MHz to 915 MHz),broadcasting service(ranging from 470 MHz to 806 MHz)and the uplink bands of GSM1800 service(ranging from 1710 MHz to 1785 MHz)in Chengdu,Sichuan Province.And then,the collected data were analyzed and researches were carried out on spectrum prediction in time and frequency dimensional based on data mining method.Finally,a prototype system for radio spectrum monitoring data mining was developed.The major accomplishments are listed as follows:(1)The prediction of future spectrum occupancy pattern was realized based on partial periodic pattern mining algorithm.With it,frequent usage patterns of PUs were extracted.And then,strong association rules were generated from those frequent patterns,which helps to obtain the regularity of spectrum usage.Thus,the spectrum prediction in the time dimension can be realized.The results show that this method has achieved satisfactory results in the prediction of the three service bands with the prediction accuracy was about 90%.Meanwhile,the influences of two main parameters of the algorithm,the minimum confidence of patterns and the minimum confidence of rules,on the mining results was analyzed in detail.Moreover,the prediction results of traditional Frequent Pattern Mining(FPM)method and partial periodic pattern mining method were compared,which shows the advantages of the latter in spectrum prediction.(2)In the case where the SUs need multiple channels to transmit data,the spectrum prediction in the frequency dimension is helpful for improving the throughput and transmission rate of the spectrum detection.To this end,the density-based clustering algorithm was introduced to deal with the channel clustering based on their correlations,thus realizing the spectrum prediction in frequency dimension.Firstly,the effectiveness of the proposed method was validated between the comparisons with the existing algorithm based on the minimum entropy increment(MEI).And then,it was used to analyze the above three services and obtained a high accuracy.In TV service and GSM900 uplink service,the prediction accuracy of the experimental data of was mostly as high as 90% and above with ? = 0.3 and MinPts = 2.Experimental results of GSM1800 uplink service show that the prediction accuracy was more than 87% with ? = 0.55 and MinPts = 2.(3)A prototype system for radio spectrum monitoring data mining was developed,which integrated the models and algorithms that included in the process of spectrum monitoring data analysis and mining.The prototype system is capable of unified management,statistical analysis and mining of radio spectrum monitoring data.