張景 周小平 王培培 李莉
摘 ?要: 利用毫米波信道的稀疏散射特性和張量的空間結(jié)構(gòu),提出了一種隨機(jī)網(wǎng)格張量分解的信道估計(jì)方法,接收信號(hào)被表示為一個(gè)四階張量,采用隨機(jī)張量壓縮對(duì)單個(gè)用戶(hù)信道進(jìn)行解耦;采用網(wǎng)格張量分解方式,將大尺度的用戶(hù)信道張量分解為若干個(gè)小尺度張量,并行且獨(dú)立地分解所有子張量,由相關(guān)因子矩陣估計(jì)信道參數(shù). 仿真結(jié)果表明,該算法能獲得較為準(zhǔn)確的信道參數(shù)估計(jì),有效地降低了信道估計(jì)算法的復(fù)雜度.
關(guān)鍵詞: 毫米波; 多用戶(hù)大規(guī)模多輸入多輸出(MIMO); 信道估計(jì); 隨機(jī)網(wǎng)格張量分解
中圖分類(lèi)號(hào): TN 929.5 ???文獻(xiàn)標(biāo)志碼: A ???文章編號(hào): 1000-5137(2021)01-0108-07
Abstract: By using the sparse scattering characteristics of the millimeter wave channel and the spatial structure of the tensor,a channel estimation method based on random grid tensor decomposition was proposed. The received signal was represented as a fourth-order tensor and the user channel was decoupled by random tensor compression. After that,using grid tensor decomposition method,the large-scale user channel tensor was decomposed into several small-scale tensors,by which all sub-tensors are decomposed in parallel and independently,and the channel parameters were estimated according to the correlation factor matrix. The simulation results showed that the algorithm was able to obtain more accurate channel parameter estimation,which reduced the complexity of the channel estimation algorithm effectively.
Key words: millimeter wave; multi-user massive multiple input and multiple output(MIMO); channel estimation; random grid tensor decomposition
0 ?引言
毫米波大規(guī)模多輸入多輸出(MIMO)是未來(lái)蜂窩網(wǎng)絡(luò)中重要的技術(shù).毫米波波段的大帶寬可以提供每秒千兆位的通信數(shù)據(jù)速率,能夠更好地滿(mǎn)足第五代移動(dòng)通信的需求[1-4].然而,高頻通信也會(huì)造成重大的路徑損失.為了解決這一問(wèn)題,一般在基站和移動(dòng)端之間部署大規(guī)模的天線陣列,以提供波束形成增益.因此,獲取完整的信道狀態(tài)信息,完成預(yù)編碼至關(guān)重要.
在多用戶(hù)上行信道估計(jì)中,訓(xùn)練序列所花費(fèi)的代價(jià)尤為高昂.隨著用戶(hù)終端數(shù)量的增加,所需的導(dǎo)頻序列長(zhǎng)度也大幅增長(zhǎng).此外,毫米波信道的相干時(shí)間比低頻信道短,因此降低多用戶(hù)多天線毫米波系統(tǒng)的導(dǎo)頻成本尤為重要.對(duì)于多用戶(hù)毫米波大規(guī)模MIMO系統(tǒng),GAO等[5]對(duì)不同用戶(hù)的信道上行鏈路進(jìn)行估計(jì),但由于信道的容量系數(shù)較大,僅估計(jì)了每個(gè)用戶(hù)的最強(qiáng)路徑;GONZALEZ-COMA等[6]根據(jù)下行鏈路上的多個(gè)用戶(hù)(MSs)估計(jì)信道,將信道狀態(tài)信息(CSI)反饋給MSs,但該算法缺乏互易性,并且需要假設(shè)信道方向位于特定的網(wǎng)格;AYACH等[7]提出了一個(gè)有硬件約束的稀疏重構(gòu)問(wèn)題來(lái)設(shè)計(jì)預(yù)編碼器,但該問(wèn)題的復(fù)雜度較高,且需要假設(shè)接收端存在完備的CSI;ALKHATEEB等[8]提出了一種基于壓縮感知的多用戶(hù)毫米波系統(tǒng)信道估計(jì)方法,利用基站和用戶(hù)之間的隨機(jī)測(cè)量矩陣估計(jì)下行信道參數(shù);在文獻(xiàn)[9]中,多用戶(hù)空域接入方式按照平行因子(parafac)分解條件進(jìn)行盲源分離,針對(duì)多用戶(hù)上行接入的情況,給出了一種不需要傳輸信道模型和傳感器陣列流形的估計(jì)算法;ZHOU等[10]提出了分層導(dǎo)頻傳輸方案和基于candecomp/parafac(CP)分解的方法,利用多種模式收集多路數(shù)據(jù)的固有低秩結(jié)構(gòu),將接收端信號(hào)張量用于多用戶(hù)信道到基站信道的聯(lián)合估計(jì),由于毫米波信道的稀疏性,該張量具有固有的低CP級(jí),保證了CP分解的唯一性.然而,用傳統(tǒng)的張量分解方法計(jì)算高維大尺度張量問(wèn)題的成本較高,需要消耗大量的時(shí)間和內(nèi)存空間.
傳統(tǒng)的張量分解算法對(duì)數(shù)據(jù)精度要求較高,不適合處理大尺度問(wèn)題.本文作者利用隨機(jī)張量分解方法[11],從大尺度張量中學(xué)習(xí)相干結(jié)構(gòu),將用戶(hù)信道投影到干擾用戶(hù)的正交空間中,抑制不同信道之間的干擾;利用網(wǎng)格張量算法[12]分解壓縮后的用戶(hù)信道張量,將大尺度的張量轉(zhuǎn)化為若干小網(wǎng)格,獲得較為準(zhǔn)確的估計(jì)值.
1 ?系統(tǒng)模型
考慮一個(gè)由N個(gè)用戶(hù)組成的毫米波系統(tǒng),每個(gè)用戶(hù)配備N(xiāo)T根發(fā)射天線和NR根接收天線,假設(shè)相鄰天線元件之間的距離為信號(hào)波長(zhǎng)的一半.每個(gè)發(fā)射天線為每個(gè)用戶(hù)發(fā)射T個(gè)符號(hào),每個(gè)符號(hào)包含K個(gè)子載波,同時(shí)保證同步,則每個(gè)用戶(hù)在第i個(gè)發(fā)射天線第t個(gè)符號(hào)處的發(fā)射信號(hào)可以表示為:
2 ?隨機(jī)網(wǎng)格張量分解和信道估計(jì)
本研究中,接收信號(hào)有4個(gè)參數(shù):用戶(hù)數(shù)量、接收天線、子載波和符號(hào),如圖1所示.對(duì)于第一階段的張量分解,從大張量隨機(jī)地導(dǎo)出小張量,將高維通道投影到干擾用戶(hù)通道的零空間中;對(duì)于第二階段的三階張量分解,傳統(tǒng)的parafac張量因式分解方法在處理大尺度問(wèn)題時(shí),需要大量的時(shí)間和內(nèi)存消耗,因此將采用網(wǎng)格張量分解方法處理大尺度張量.
2.1 隨機(jī)分解張量
從高斯分布中提取隨機(jī)向量作為n階張量模的近似基.這些隨機(jī)向量構(gòu)成測(cè)量矩陣,用于繪制用戶(hù)信道張量切片的列空間如下:
2.2 第二階段張量:網(wǎng)格張量分解
3 ?仿真結(jié)果與分析
本節(jié)將給出傳統(tǒng)導(dǎo)頻、平行因子分解及所提方法的仿真結(jié)果,以此驗(yàn)證所提方法性能的優(yōu)越性.基于寬帶幾何信道模型生成毫米波信道,其中,發(fā)送端天線數(shù)NT為64,接收端天線數(shù)NR為64,假設(shè)有5條可分辨路徑,取5條路徑的時(shí)延為0.1,1.0,1.5,2.0,3.0 Ts,其中,Ts為符號(hào)周期,信道的AOA和AOD在[0,2]內(nèi)均勻隨機(jī)生成,實(shí)驗(yàn)進(jìn)行了1 000次蒙特卡羅仿真.
圖2為各算法的估計(jì)均方誤差(MSE)對(duì)比.從圖2可以看出,本方法優(yōu)于傳統(tǒng)導(dǎo)頻方法和平行因子分解法,特別是在高信噪比(SNR)情況下,本方法的估計(jì)性能優(yōu)勢(shì)更為明顯.此外,本方法中,子張量數(shù)M為4時(shí)的性能略?xún)?yōu)于M為2時(shí),因此,通過(guò)適當(dāng)設(shè)置參數(shù)可以提高隨機(jī)網(wǎng)格算法的性能.
圖3比較了不同信噪比下,各算法的誤碼率(BER).從圖3可以看出,該算法的誤碼率比其他兩種算法都要低.這是因?yàn)殡S機(jī)網(wǎng)格算法中存在張量的空間結(jié)構(gòu),子張量的分類(lèi)減少了層層迭代誤差積累的情況,降低了層次子誤碼率,提高了系統(tǒng)整體的誤碼率性能.
圖4對(duì)比了隨著天線數(shù)量的增加,不同算法估計(jì)所需的時(shí)間,驗(yàn)證了不同算法的時(shí)間復(fù)雜度.由圖4可知,信道估計(jì)的維數(shù)也隨著天線數(shù)量的增加而增加,不同算法消耗的估計(jì)時(shí)間也迅速增加.其中,本方法時(shí)間執(zhí)行最短,且隨天線尺寸增大,消耗時(shí)間增長(zhǎng)緩慢.這是因?yàn)楸痉椒▽⒏呔S張量的網(wǎng)格轉(zhuǎn)化為低維張量,通過(guò)并行工具箱對(duì)數(shù)據(jù)進(jìn)行并行處理,并進(jìn)行迭代優(yōu)化求解.
4 ?結(jié)論
本文作者提出了一種用于多用戶(hù)毫米波大規(guī)模MIMO系統(tǒng)信道估計(jì)的張量因子分解方法.針對(duì)多用戶(hù)毫米波系統(tǒng)信道估計(jì)計(jì)算量較大的問(wèn)題,利用毫米波信道的稀疏散射特性和張量的空間結(jié)構(gòu),采用隨機(jī)網(wǎng)格張量分解算法估計(jì)信道參數(shù).該方法將接收信號(hào)表示為四階張量,將信道參數(shù)估計(jì)問(wèn)題轉(zhuǎn)化為大規(guī)模張量分解問(wèn)題,采用隨機(jī)張量分解方法進(jìn)行張量壓縮,然后采用網(wǎng)格張量分解算法進(jìn)行并行張量計(jì)算,減少了高維矩陣的逆和乘運(yùn)算次數(shù).仿真結(jié)果表明,該算法能獲得準(zhǔn)確的信道參數(shù)估計(jì),有效地降低了信道估計(jì)算法的復(fù)雜度.
參考文獻(xiàn):
[1] RAPPAPORT T S,MURDOCK J N,GUTIERREZ F.State of the art in 60-GHz integrated circuits and systems for wireless communications [J].Proceedings of the IEEE,2011,99(8):1390-1436.
[2] RANGAN S,RAPPAPORT T S,ERKIP E.Millimeter-wave cellular wireless networks:potentials and challenges [J].Proceedings of the IEEE,2014,102(3):366-385.
[3] GHOSH A,THOMAS T A,CUDAK M C,et al.Millimeter-wave enhanced local area systems:a highdata-rate approach for future wireless networks [J].IEEE Journal on Selected Areas in Communications,2014,32(6):1152-1163.
[4] YANG G,XIAO M,GROSS J,et al.Delay and backlog analysis for 60 GHz wireless networks [C]//Global Communications Conference.Washington,DC:IEEE,2016:1-7.
[5] GAO Z,HU C,DAI L L,et al.Channel estimation for millimeter-wave massive MIMO with hybrid precoding over frequency-selective fading channels [J].IEEE Communication Letters,2016,20(6):1259-1262.
[6] GONZALEZ-COMAJ P,RODRIGUEZ-FERNANDEZ J,GONZALEZ-PRELCIC N,et al.Channel estimation and hybrid precoding/combining for frequency selective multiuser mmWave systems [C]//IEEE Global Communications Conference.Singapore:IEEE,2017:1-6.
[7] AYACH O E,RAJAGOPAL S,ABU-SURRA S,et al.Spatially sparse precoding in millimeter-wave MIMO systems [J].IEEE Transactions on Wireless Communication,2014,13(3):1499-1513.
[8] ALKHATEEB A,LEUS G,HEATH W R.Compressed-sensing based multi-user millimeter wave systems:how many measurements are needed? [C]//IEEE International Conference on Acoustics,Speech and Signal Processing.Brisbane:IEEE,2015:2909-2913.
[9] RONG Y,VOROBYOV S A,GERSHMAN A B,et al.Blind spatial signature estimation via time-varying user power loading and parallel factor analysis [J].IEEE Transactions on Signal Process,2005,53(5):1697-1710.
[10] ZHOU Z,F(xiàn)ANG J,YANG L,et al.Channel estimation for millimeter-wave multiuser MIMO sys-tems via parafac decomposition[J].IEEE Trans Wireless Commun,2016,15(11):7501-7516
[11] ERICHSON N B,MANOHAR K,BRUNTON S L,et al.Randomized CP tensor decomposition [J].Machine Learning Science and Technology,2016,1(2):1-29.
[12] PHAN A H,CICHOCKI A.PARAFAC algorithms for large-scale problems [J].Neurocomputing,2011,74(11):1970-1984.
[13] SHU S,RAPPAPORT T S.Millimeter wave MIMO channel estimation based on adaptive compressed sensing [C]//IEEE International Conference on Communications Workshops.Paris:IEEE,2017:47-53.
[14] KOLDA T G,BADER B W.Tensor decompositions and applications [J].SIAM Review,2009,51(3):455-500.
(責(zé)任編輯:包震宇)