盧光躍,葉迎暉,孫 宇,彌 寅
克服噪聲不確定度的擬合優(yōu)度檢驗(yàn)頻譜感知算法*
盧光躍**,葉迎暉,孫 宇,彌 寅
(西安郵電大學(xué)無線網(wǎng)絡(luò)安全技術(shù)國家工程實(shí)驗(yàn)室,西安710121)
針對(duì)已有的基于擬合優(yōu)度(GoF)檢驗(yàn)的頻譜感知算法易受到噪聲不確定度影響的問題,利用矩估計(jì)法或特征分解估計(jì)法對(duì)噪聲方差進(jìn)行實(shí)時(shí)估計(jì),將采樣數(shù)據(jù)處理為標(biāo)準(zhǔn)正態(tài)分布的信號(hào),最后通過GoF檢驗(yàn)來感知主用戶的存在性。在減小GoF算法復(fù)雜度的同時(shí),克服了噪聲不確定度對(duì)算法性能的影響,仿真結(jié)果也表明了所提算法的有效性。
認(rèn)知無線電;頻譜感知;擬合優(yōu)度檢驗(yàn);噪聲方差估計(jì)
認(rèn)知無線電(Cognitive Radio,CR)是一種動(dòng)態(tài)頻譜管理技術(shù),旨在解決當(dāng)前日益嚴(yán)重的頻譜資源匱乏、頻譜利用率不高的問題,其核心思想是允許次用戶(Second User,SU)在主用戶(PrimarY User,PU)不使用授權(quán)頻段時(shí)動(dòng)態(tài)接入該頻段,而當(dāng)PU重新使用授權(quán)頻段時(shí)能夠及時(shí)撤出,以免干擾PU通信??梢?,CR的前提條件和首要任務(wù)是頻譜感知。
經(jīng)典的頻譜感知方法主要有能量檢測算法(EnergY Detection,ED)、循環(huán)平穩(wěn)特征檢測算法(CYclostationarY Feature Detection,CFD)、匹配濾波檢測算法(Matched-Filtering,MF)、基于特征結(jié)構(gòu)的感知算法和基于擬合優(yōu)度檢測(Goodness of Fit,GoF)的感知算法等。ED算法[1-2]實(shí)現(xiàn)簡單且不需要任何先驗(yàn)信息,但它對(duì)噪聲不確定度敏感,噪聲不確定度往往會(huì)造成算法性能的急劇下降。CFD算法[3]復(fù)雜度高,MF算法[4]必須預(yù)知PU的先驗(yàn)知識(shí)(如信號(hào)波形、調(diào)制方式等),并且對(duì)于同步的要求也比較高?;谔卣鹘Y(jié)構(gòu)的感知算法[5-6]主要利用接收信號(hào)協(xié)方差矩陣特征值和特征矢量的性質(zhì)進(jìn)行感知,其中基于特征值的感知算法主要有最大最小特征值之比(Maximum-Minimum Eigenvalue,MME)算法和最大最小特征值之差(Difference betWeen the Maximum eigenvalue and the Minimum Eigenvalue,DMM)等算法[7-8],基于特征矢量的頻譜感知算法主要有特征模板匹配(Feature TemPlated Matching,F(xiàn)TM)算法和子空間投影(SubsPace Projection,SP)算法等[9]。該類算法檢測性能優(yōu)于ED算法,不需要預(yù)知PU先驗(yàn)知識(shí),缺點(diǎn)是復(fù)雜度較高。GoF算法[10-15]將頻譜感知轉(zhuǎn)化為一種擬合優(yōu)度檢測問題,即假設(shè)檢驗(yàn)統(tǒng)計(jì)量服從某一特定分布,若感知頻段不存在PU信號(hào),則采樣數(shù)據(jù)應(yīng)服從該分布,否則采樣數(shù)據(jù)將偏離特定的分布。一般地,假設(shè)噪聲服從均值為0、方差為σ2的高斯白噪聲,頻譜感知問題便轉(zhuǎn)變?yōu)闄z驗(yàn)采樣數(shù)據(jù)是否服從均值為0、方差為σ2的正態(tài)分布問題。文獻(xiàn)[15]中給出常用的GoF檢測準(zhǔn)則包括Kolmogorov-Smirnov(KS)準(zhǔn)則、Ander_ son-Darling(AD)準(zhǔn)則和Cramer-von Mises(CM)準(zhǔn)則等。文獻(xiàn)[10]表明GoF算法性能優(yōu)于能量檢測算法,但缺點(diǎn)是需要噪聲方差先驗(yàn)知識(shí),噪聲不確定度對(duì)其有很大影響,且只適用于實(shí)信號(hào)。文獻(xiàn)[11]給出了一種適用于復(fù)信號(hào)的改進(jìn)算法,其原理是當(dāng)PU信號(hào)不存在時(shí)采樣數(shù)據(jù)的能量服從卡方分布,用卡方分布代替正態(tài)分布進(jìn)行擬合優(yōu)度檢測,其缺點(diǎn)同樣是算法性能受噪聲不確定度影響。
本文在感知過程中動(dòng)態(tài)估計(jì)噪聲方差,并對(duì)接收信號(hào)進(jìn)行相應(yīng)處理,從而解決了噪聲方差波動(dòng)對(duì)算法性能影響的問題。
頻譜感知問題通??梢员硎緸橐粋€(gè)二元假設(shè)檢驗(yàn)問題,其模型如下:
GoF算法利用H0與H1條件下接收數(shù)據(jù)概率分布函數(shù)之間的差異進(jìn)行感知。假設(shè)采樣點(diǎn)數(shù)為N,將SU接收數(shù)據(jù)k)按照升序排列,得到的新序列記為γ(k),于是,有γ(1)≤γ(2)≤…≤γ(N),則其經(jīng)驗(yàn)譜分布FN(γ)表示[11]為
顯然,若H0成立,采樣數(shù)據(jù)服從均值為0、方差為σ2的正態(tài)分布,其分布函數(shù)F0(γ)為
當(dāng)N趨于無窮大時(shí),F(xiàn)N(γ)依概率1收斂于F0(γ);若H1成立,由于有PU信號(hào)存在,F(xiàn)N(γ)將偏離F0(γ)。因此,頻譜感知轉(zhuǎn)化為如下的擬合優(yōu)度檢驗(yàn)問題:
采用AD準(zhǔn)則進(jìn)行擬合優(yōu)度檢驗(yàn),文獻(xiàn)[15]給出FN(γ)與F0(γ)之間的距離:
式中:Zi=F0(γ(i))。取為檢驗(yàn)統(tǒng)計(jì)量,則當(dāng)大于等于門限T時(shí),判斷H1成立,反之判斷H0成立。因此,門限T可由下式確定:
H0條件下的分布與噪聲分布無關(guān),其極限分布[16]如下:
式中:αj=(-1)jΓ(j+0.5)/(Γ(0.5)j?。镚am_ma函數(shù)。事實(shí)上,當(dāng)N≥5時(shí)上式即收斂,因此若給定虛警概率Pf,即可求得或查表[15]得到相應(yīng)的門限T,如Pf=0.01,T=3.875;Pf=0.1,T=1.933。
由式(3)和(6)可以看出,當(dāng)噪聲方差σ2未知時(shí),將無法確定F0(γ)和Zi;而實(shí)際中,即使噪聲方差σ2已知,其值也會(huì)隨環(huán)境變化而隨時(shí)變化,即會(huì)存在噪聲不確定度問題,此時(shí)F0(γ)和Zi也將發(fā)生變化,也需要實(shí)時(shí)更新。因此,GoF算法需要知道σ2;且當(dāng)存在噪聲不確定度時(shí),算法性能必將受到影響。
如果能夠在感知過程中實(shí)時(shí)估計(jì)噪聲方差,進(jìn)而動(dòng)態(tài)更新F0(γ(i)),則可消除噪聲方差對(duì)算法性能的影響。然而,根據(jù)式(6),此方法在每次感知中均需利用式(3)動(dòng)態(tài)計(jì)算F0(γ(i)),進(jìn)而計(jì)算Zi;不同的噪聲方差需要重復(fù)計(jì)算式(3),這必將增加算法復(fù)雜度。為此,可根據(jù)估計(jì)出的噪聲方差?σ2,對(duì)采樣數(shù)據(jù)作如下處理:
為了估計(jì)σ2,當(dāng)SU進(jìn)行頻譜感知時(shí)PU信號(hào)保持不變[10,13],可采用矩估計(jì)法估計(jì)σ2;另一方面,當(dāng)SU進(jìn)行頻譜感知時(shí)PU信號(hào)時(shí)刻在變,采用特征分解估計(jì)法估計(jì)σ2。
3.1 矩估計(jì)法
由辛欽大數(shù)定理可知[17],當(dāng)N→∞時(shí),樣本m階矩依分布收斂于總體m階矩,樣本矩的連續(xù)函數(shù)收斂于相應(yīng)的總體矩的連續(xù)函數(shù),因此,可用樣本m階矩作為總體m階矩的估計(jì),用樣本方差的無偏估計(jì)代替σ2:
矩估計(jì)法運(yùn)算簡單,復(fù)雜度為O(N)。
3.2 特征分解估計(jì)法
由于接收數(shù)據(jù)協(xié)方差矩陣攜帶噪聲的統(tǒng)計(jì)信息,對(duì)其作特征值分解最小特征值即對(duì)應(yīng)噪聲方差,因此估計(jì)接收數(shù)據(jù)協(xié)方差矩陣成為關(guān)鍵。首先將采樣數(shù)據(jù)k)等分為k段,,…,,其中每一段數(shù)據(jù)有N/k個(gè)采樣點(diǎn)。構(gòu)造數(shù)據(jù)矩陣Y:
則接收數(shù)據(jù)協(xié)方差矩陣Rγ可估計(jì)為
式中:S=E[ssH]為PU信號(hào)協(xié)方差矩陣。因?yàn)閚是均值為0、方差為σ2的高斯白噪聲,所以有
式中:I為k×k維單位陣。因此,Rγ的特征值結(jié)構(gòu)如下:
式中:λ是矩陣S的最大特征值。由此可知,除Rγ的最大特征之外,其余特征值可以看作噪聲方差的估計(jì),因此可通過求除最大特征值外的特征值進(jìn)行平均得到噪聲方差σ2的估計(jì)值?σ2。特征分解估計(jì)法不需任何先驗(yàn)信息且適用范圍大,其復(fù)雜度為O(k3)。
綜上,所提算法的步驟可描述如下:
(1)給定Pf,結(jié)合式(7)和式(8)或查表求得對(duì)應(yīng)的判決門限T;
(3)根據(jù)式(9)對(duì)接收數(shù)據(jù)作相應(yīng)處理,得到新數(shù)據(jù)?γ并將其作升序排列:
為了驗(yàn)證上述分析,本節(jié)對(duì)傳統(tǒng)GoF算法與Modified-GoF算法進(jìn)行仿真比較。仿真中,假設(shè)信道是理想信道,PU發(fā)送信號(hào)s=1,Pf=0.01,N=3 200,根據(jù)文獻(xiàn)[15]可知,T=3.875;當(dāng)存在噪聲不確定度α?xí)r[7],真實(shí)的噪聲方差在區(qū)間[B-1σ2,Bσ2]取值,其中B=100.1α。因?yàn)镋D算法和GoF算法需要知道噪聲方差,因此假設(shè)σ2=1。
圖1和圖2分別給出信噪比為-25 dB時(shí),GoF和Modified-GoF算法在α=0 dB和α=1 dB時(shí)檢驗(yàn)統(tǒng)計(jì)量的概率密度函數(shù)(Test Statistic PDF,TSPDF)??梢?,當(dāng)不存在噪聲不確定度(α=0 dB)時(shí),兩種算法在H0情況下和H1情況下其TS-PDF有一定的分離度,此時(shí)通過設(shè)置門限T可使算法在滿足適當(dāng)Pf的情況下獲得較好的檢測性能。當(dāng)α=1時(shí),Modified-GoF算法TS-PDF基本保持不變,但GoF算法的TS-PDF變化較大且與H1時(shí)的TS-PDF重疊部分增加,此時(shí)若仍采用原門限判決,將導(dǎo)致虛警概率變大,檢測性能下降。由此可知,GoF算法受噪聲不確定度的影響,而Modified-GoF算法將不受其影響。
圖1 GoF算法的TS-PDFFig.1 Test satistic PDF of GoF
圖2 Modified-GoF算法的TS-PDFFig.2 Test statistic PDF of Modified-GoF
圖3 描述了兩種算法在α=0 dB和α=1 dB兩種情況下的檢測性能。由圖3可以看出,在α=0 dB時(shí)兩種算法Pd與Pf曲線基本重合,兩者檢測性能相當(dāng)。但當(dāng)α=1 dB時(shí)GoF算法Pf躍升為0.5,此時(shí)算法將失效;而Modified-GoF算法Pf仍為0.01,性能仍保持穩(wěn)定。為進(jìn)一步說明所提算法的優(yōu)越性,圖4給出了α=0時(shí)DMM算法[7]、ED算法和Modified-GoF算法檢測性能曲線。由圖4可知,相同條件下DMM算法性能優(yōu)于ED算法3~4 dB,而Modified-GoF算法性能在低信噪比情況下遠(yuǎn)優(yōu)于其他兩種算法。
圖3 GoF算法和Modified-GoF算法檢測性能比較Fig.3 Performance comParison betWeen GoF and Modified-GoF
圖4 DMM、ED和Modified-GoF算法檢測性能Fig.4 Performance comParison among DMM,GoF and Modified-GoF
圖5 為α=0 dB時(shí)Modified-GoF算法在N取不同值時(shí)的檢測性能曲線。可以看出算法Pf始終能夠滿足給定的要求,而Pd的性能隨N的增加而增加。如N=3 200時(shí)的性能比N=2 000時(shí)的性能好2 dB,比N=1 000時(shí)的性能好5 dB。圖6給出了N=100和N=300的小采樣點(diǎn)情況下兩種算法的性能比較??梢姡诓蓸狱c(diǎn)數(shù)較小時(shí),Modified-GoF算法的檢測性能略好于GoF算法,盡管其虛警概率滿足預(yù)設(shè)的要求,但也略高于GoF算法,這主要是因?yàn)椴蓸狱c(diǎn)數(shù)小時(shí)噪聲方差估計(jì)精度不高造成的。由圖5和圖6可知,Modified-GoF算法在小采樣點(diǎn)數(shù)時(shí)具有較好的檢測性能。
圖5 采樣點(diǎn)不同時(shí)Modified-GoF算法檢測性能Fig.5 The Performances of Modified-GoF against different samPles
圖6 小采樣點(diǎn)情況下Modified-GoF算法檢測性能Fig.6 The Performances of Modified-GoF at small samPles
快速有效的頻譜感知算法是認(rèn)知無線電的研究重點(diǎn),已有的基于擬合優(yōu)度頻譜感知算法能用較小的樣本點(diǎn)數(shù)實(shí)現(xiàn)較理想的感知效果,但其需要噪聲方差這一先驗(yàn)信息以及受噪聲不確定度的影響。本文提出一種改進(jìn)的擬合優(yōu)度檢測算法,該算法無需知道噪聲方差,且不受噪聲不確定度的影響。同時(shí),相比于已有的GoF算法[10],用查表法替代Zi=F0(γ(i))的計(jì)算,從而減少了運(yùn)算量。仿真結(jié)果表明:Modified-GoF算法不受噪聲不確定度的影響,且相同條件下低信噪比時(shí)算法性能優(yōu)于DMM算法和ED算法,與GoF算法性能相當(dāng);同時(shí)在小采樣情況下,所提算法能夠以較小的感知時(shí)延而保持較理想性能。但是,GoF類算法在快衰落信道下性能會(huì)降低,未來我們將會(huì)重點(diǎn)研究快衰落信道下如何提高GoF算法的檢測性能。
[1] DIGHAM F F,ALOUINI M S,SIMONM K.On the energY detection of unknoWn signals over fading channels[J].IEEE Transactions on Communications,1967,5(1):3575-3579.
[2] CABRIC D,TKACHENKO A,BRODERSENR W.ExPer_ imental studY of sPectrum sensing based on energY detec_ tion and netWork cooPeration[C]//Proceedings of the ACM 1st International WorkshoP on TechnologY and Poli_ cY for Accessing SPectrum.NeW York:IEEE,2006:1-5.
[3] CABRIC D,MISHRA S M,BRODERSEN R W.ImPle_ mentation issues in sPectrum sensing for cognitive radios [C]//Proceedings of the 38th Asilomar Conference on Signals SYstems&ComPuters.Pacific Grove,California:IEEE,2004:772-776.
[4] DANDAWTEH A V.Statistical tests for Presence of cY_ clostationaritY[J].IEEE Transactions on Signal Process_ ing,1994,42(9):2355-2369.
[5] 盧光躍,彌寅,包志強(qiáng),等.基于特征結(jié)構(gòu)的頻譜感知算法[J].西安郵電大學(xué)學(xué)報(bào),2014,19(2):1-12. LU GuangYue,MI Yin,BAO Zhiqiang,et al.The cooPera_ tive sPectrum sensing algorithms based on eigenvalue structure of the received signal[J].Journal of Xi′an Uni_ versitY of Posts and Telecommunications,2014,19(2):1-12.(in Chinese)
[6] 彌寅,盧光躍,關(guān)璐.特征值類頻譜感知算法的仿真分析[J].西安郵電大學(xué)學(xué)報(bào),2014,19(5):27-33. MI Yin,LU GuangYue,GUAN Lu.Simulation and analY_ sis of eigenvalue-based cooPerative sPectrum sensing al_ gorithms[J].Journal of Xi′an UniversitY of Posts and Tel_ ecommunications,2014,19(5):27-33.(in Chinese)
[7] ZENG Y,LIANG Y C.Maximum-minimum eigenvalue detection for cognitive radio[C]//Proceedings of 2007 18th International SYmPosium on Personal,Indoor and Mobile Radio Communications.Athens,Grence:IEEE,2007:1-5.
[8] 王穎喜,盧光躍.基于最大最小特征值之差的頻譜感知技術(shù)研究[J].電子與信息學(xué)報(bào),2010,32(11):2572-2574. WANG Yingxi,LU GuangYue.DMM based sPectrum sensing method for cognitive radio sYstems[J].Journal of Electronics&Information TechnologY,2010,32(11):2572-2574.(in Chinese)
[9] 孫宇,盧光躍,彌寅.子空間投影的頻譜感知算法研究[J].信號(hào)處理,2015,31(4):83-89. SUN Yu,LU GuangYue,MI Yin.The research of sPectrum sensing method based on subsPace Projection[J].Journal of Signal Processing,2014,31(4):83-89.(in Chinese)
[10] WANG H,YANG E H,ZHAO Z,et al.SPectrum sensing in cognitive radio using goodness of fit testing[J].IEEE Transactions on Wireless Communications,2009,8(11):5427-5430.
[11] JINM,GUO Q,XI J,et al.SPectrum sensing based ongoodness of fit test With unilateral alternative hYPothesis [J].Electronics Letters,2014,50(22):1645-1646.
[12] ARSHAD K,MOESSNER K.Robust sPectrum sensing based on statistical tests[J].IET Communications,2013,7(9):808-817.
[13] 沈雷,王海泉,趙知?jiǎng)牛?認(rèn)知無線電中基于擬合優(yōu)度的頻譜盲檢測算法研究[J].通信學(xué)報(bào),2011,32(11):27-34. SHEN Lei,WANG Haiquan,ZHAO Zhijin,et al.Blind sPectrum sensing based on goodness of fit test for cogni_ tive radio in noise of uncertain PoWer[J].Journal on Communications,2012,32(11):27-34.(in Chinese)
[14] LEI S,WANG H,SHEN L.SPectrum sensing based on goodness of fit tests[C]//Proceedings of 2011 Interna_ tional Conference on Electronics,Communications and Control.Zhejiang:IEEE,2011:485-489.
[15] STEPHENS M A.EDF Statistics for goodness of fit and some comParisons[J].Journal of the American Statisti_ cal Association,1974,69(347):730-737.
[16] ANDERSON T W,DARLING D A.AsYmPtotic theorY of certain"goodness of fit"criteria based on stochastic Processes[J].Annals of Mathematical Statistics,1952,23(1):143-143.
[17] 盛驟,謝式千,潘承毅,等.概率論與數(shù)理統(tǒng)計(jì)[M].4版.北京:高等教育出版社,2008:141-151. SHENG Zhou,XIE Shiqian,PAN ChengYi,et al.ProbabilitY theorY and mathematical statistics[M].4th ed.Beijing:Higher Education Press,2008:141-151.(in Chinese)
盧光躍(1971—),男,河南南陽人,1999年于西安電子科技大學(xué)獲博士學(xué)位,現(xiàn)為西安郵電大學(xué)教授,主要從事通信信號(hào)處理、無線傳感網(wǎng)絡(luò)以及認(rèn)知無線電等方面的研究;
LU GuangYue Was bron in NanYang,Henan Province,in 1971.He received the Ph.D.de_ gree from Xidian UniversitY in 1999.He is noW a Professor.His research concerns signal Processing,Wireless sen_ sor netWork and cognitive radio.
Email:tonYlugY@163.com
葉迎暉(1991—),男,浙江麗水人,西安郵電大學(xué)碩士研究生,主要研究方向?yàn)檎J(rèn)知無線電頻譜感知技術(shù);
YE Yinghui Was born in Lishui,Zhejiang Province,in 1991.He is noW a graduate student.His research concerns sPec_ trum sening in cognitive radio.
Email:connectYYh@126.com
孫 宇(1990—),男,內(nèi)蒙古烏蘭察布人,西安郵電大學(xué)碩士研究生,主要研究方向?yàn)檎J(rèn)知無線電頻譜感知技術(shù);
SUN Yu Was born Wulanchabu,Neimenggu Autonomous Region,in 1990.He is noW a graduate student.His research concerns sPectrum sensing in cognitive radio.
Email:Yu_sun90@163.com
彌 寅(1986—),男,甘肅慶陽人,碩士,西安郵電大學(xué)助教,主要研究方向?yàn)檎J(rèn)知無線電頻譜感知技術(shù)。
MI Yin Was born in QingYang,Gansu Province,in 1986.He is noW a teaching assistant With the M.S.degree.His research concerns sPectrum sening in cognitive radio.
Email:miYin0404@163.com
A Novel Anti-noise-uncertainty Spectrum Sensing Method Using Goodness of Fit Test
LU GuangYue,YE Yinghui,SUN Yu,MI Yin
(National Engineering LaboratorY for Wireless SecuritY,Xi′an UniversitY of Posts and Telecommunications,Xi′an 710121,China)
For the Problem that the Performance of the existing sPectrum sensing algorithm based on Good_ ness of Fit(GoF)test is sensitive to the noise uncertaintY,the moment estimation method or the eigen de_ comPosition estimation method is emPloYed to estimate the noise variance timelY and the samPling data is Processed,using the estimated variance,to be normal Gaussian signal.FinallY,the GoF test is used to de_ tect the existence of the Primer user.The ProPosed method is free of noise uncertaintY With loWer comPlexi_ tY.Simulation results shoW the effectiveness of the ProPosed method.
cognitive radio;sPectrum sensing;goodness of fit test;estimation of noise variance
The National Natural Science Foundation of China(No.61271276,61301091);The Natural Science Foundation of Shaanxi Province(2014JM8299)
TN92
A
1001-893X(2016)01-0026-06
10.3969/j.issn.1001-893x.2016.01.005
盧光躍,葉迎暉,孫宇,等.克服噪聲不確定度的擬合優(yōu)度檢驗(yàn)頻譜感知算法[J].電訊技術(shù),2016,56(1):26-31.[LU GuangYue,YE Yinghui,SUN Yu,et al.A novel anti-noise-uncertaintY sPectrum sensing method using goodness of fit test[J].Telecommunication Engineer_ ing,2016,56(1):26-31.]
2015-06-23;
2015-08-27 Received date:2015-06-23;Revised date:2015-08-27
國家自然科學(xué)基金資助項(xiàng)目(61271276,61301091);陜西省自然科學(xué)基金資助項(xiàng)目(2014JM8299)
**通信作者:tonYlugY@163.com Corresponding author:tonYlugY@163.com