資源描述：
《認(rèn)知無(wú)線電中基于機(jī)器學(xué)習(xí)算法的頻譜感知方法研究》由會(huì)員上傳分享，免費(fèi)在線閱讀，更多相關(guān)內(nèi)容在工程資料-天天文庫(kù)。
1、認(rèn)知無(wú)線電中基于機(jī)器學(xué)習(xí)算法的頻譜感知方法研究摘要傳統(tǒng)無(wú)線頻譜資源是按授權(quán)方式靜態(tài)分配的，這會(huì)限制無(wú)線通信的靈活性。而認(rèn)知無(wú)線電(CR)是機(jī)會(huì)性動(dòng)態(tài)分配頻譜資源，提高了頻譜利用率。其中，頻譜感知作為CR首要前提，其主要目標(biāo)是認(rèn)知設(shè)備迅速和智能地識(shí)別頻段中未被占用的可用頻譜，使更多的用戶機(jī)會(huì)性地使用該資源。本論文基于該背景，將傳統(tǒng)頻譜感知技術(shù)與機(jī)器學(xué)習(xí)算法進(jìn)行比較，其中能量檢測(cè)作為傳統(tǒng)算法的代表，通過(guò)對(duì)比得出傳統(tǒng)技術(shù)的缺點(diǎn)與限制范圍。同時(shí)，運(yùn)用支持向量機(jī)(SVM)來(lái)提高頻譜感知性能，并對(duì)比三種SVM核函數(shù)的錯(cuò)誤率來(lái)選擇最優(yōu)的核函數(shù)
2、。首先，本論文闡述基于認(rèn)知無(wú)線電背景運(yùn)用支持向量機(jī)的可行性。隨后，設(shè)計(jì)出無(wú)線環(huán)境的參數(shù)與仿真出能量檢測(cè)算法。然后在仿真環(huán)境中，用歷史數(shù)據(jù)來(lái)訓(xùn)練SVM模型。同時(shí)，對(duì)比線性、多項(xiàng)式和徑向基核函數(shù)，選擇錯(cuò)誤率最低的SVM模型，最后與能量檢測(cè)比較性能。仿真結(jié)果說(shuō)明，SVM的感知性能更接近理論值，比能量檢測(cè)更為可靠與準(zhǔn)確，錯(cuò)誤率為1.6%，在低SNR下檢測(cè)概率比能量檢測(cè)高出18%。關(guān)鍵詞：認(rèn)知無(wú)線電頻譜感知能量檢測(cè)支持向量機(jī)核函數(shù)IResearchonSpectrumSensingMethodofCognitiveRadioBasedon
3、MachineLearningAlgorithmsABSTRACTTraditionalradiospectrumresourcesarestaticallyallocatedaccordingtotheauthorizedways,anditwouldlimittheflexibilityofwirelesscommunications.Thecognitiveradio(CR)allocatesspectrumresourcesopportunistically,anditimprovesthespectrumefficie
4、ncy.AsthemostimportantprerequisiteforCR,thespectrumsensingaimsathelpingthecognitiveequipmentidentifytheunoccupiedandavailablespectrumquicklyandintelligently.Thus,moreuserscanhaveopportunitiestousethespectrumresource.Thispapercomparesthetraditionalspectrumsensingtechn
5、ologywithmachinelearningalgorithmsandgetsthedisadvantagesandlimitsoftraditionaltechniques.Ittakesenergydetectionasarepresentativeofthetraditionalone.Thispaperusesthesupportvectormachine(SVM)toimprovespectrumsensing,andcomparetheerrorratesofthesethreekindsofSVMkernelf
6、unctionstochoosetheoptimalkernelfunction.Firstly,thispaperdescribesthefeasibilityofusingsupportvectormachinebasedoncognitiveradiobackground.Secondly,thethesisdesignsthewirelessenvironmentparametersandsimulatestheenergydetectionalgorithm.Thirdly,theSVMmodelistrainedby
7、historicaldatainsimulationenvironment.Fourthly,thepapermakesacomparisonbetweenlinear,polynomialandradialbasisfunction,toselectthebestonewiththelowesterrorrate.Finally,thepapercomparesthisbestSVMmodelwiththeenergydetection.Asthesimulationresultsshows,theperceptionperf
8、ormanceofSVM,errorrate1.6%,isclosertotheoreticalvalueandismorereliableandaccuratethanenergydetection.Itis18%higherthanenergydetecti
