MmWve Mssive MIMO Bsed Wireless Bchul for 5G Ultr-Dense Networ Zhen Go, Linglong Di, De Mi, Zhocheng Wng, Muhmmd Ali Imrn, nd Muhmmd Zeeshn Shir rxiv:508.03940v3 [cs.it] 27 Nov 205 Abstrct Ultr-dense networ (UDN) hs been considered s promising cndidte for future 5G networ to meet the explosive dt demnd. To relize UDN, relible, Gighertz bndwidth, nd cost-effective bchul connecting ultr-dense smll-cell bse sttions (BSs) nd mcro-cell BS is prerequisite. Millimeter-wve (mmwve) cn provide the potentil Gbps trffic for wireless bchul. Moreover, mmwve cn be esily integrted with mssive MIMO for the improved lin relibility. In this rticle, we discuss the fesibility of mmwve mssive MIMO bsed wireless bchul for 5G UDN, nd the benefits nd chllenges re lso ddressed. Especilly, we propose digitlly-controlled phse-shifter networ (DPSN) bsed hybrid precoding/combining scheme for mmwve mssive MIMO, whereby the low-rn property of mmwve mssive MIMO chnnel mtrix is leverged to reduce the required cost nd complexity of trnsceiver with negligible performnce loss. One ey feture of the proposed scheme is tht the mcro-cell BS cn simultneously support multiple smll-cell BSs with multiple strems for ech smllcell BS, which is essentilly different from conventionl hybrid precoding/combining schemes typiclly limited to single-user MIMO with multiple strems or multi-user MIMO with single strem for ech user. Bsed on the proposed scheme, we further explore the fundmentl issues of developing mmwve mssive MIMO for wireless bchul, nd the ssocited chllenges, insight, nd prospect to enble the mmwve mssive MIMO bsed wireless bchul for 5G UDN re discussed. Index Terms Ultr-dense networ (UDN), mmwve bchul, mssive MIMO, precoding/combining. I. INTRODUCTION THE explosive trffic demnd is chllenging current cellulr networs, including the most dvnced 4G networ. It hs been the consensus tht future 5G networ should relize the gols of thousnd-fold system cpcity, hundredfold energy efficiency, nd tens of lower ltency. To relize such ggressive 5G version, ultr-dense networ (UDN) hs been considered s promising system rchitecture to enble Gbps user experience, semless coverge, nd green communictions []. In UDN, s shown in Fig., the mcro-cell bse sttions (BSs) with lrge coverge usully control the user scheduling, resource lloction, nd support high-mobility users, while Z. Go, L. Di, nd Z. Wng re with Tsinghu Ntionl Lbortory for Informtion Science nd Technology (TNList), Deprtment of Electronic Engineering, Tsinghu University, Beijing 00084, P. R. Chin (E-mils: goz@mils.tsinghu.edu.cn; {dill, zcwng}@tsinghu.edu.cn). D. Mi nd M. A. Imrn re with Institute for Communiction Systems (ICS), Home of 5G Innovtion Center (5GIC), University of Surrey, Guildford, UK (E-mils: {d.mi, m.imrn}@surrey.c.u). M. Z. Shir is with the Deprtment of Electricl nd Computer Engineering, Texs A&M University t Qtr (TAMUQ), Eduction City, P.O. Box 23874, Doh, Qtr (E-mil: muhmmd.shir@qtr.tmu.edu). This wor ws supported in prt by the Interntionl Science & Technology Coopertion Progrm of Chin (Grnt No. 205DFG2760), the Ntionl Nturl Science Foundtion of Chin (Grnt No. 62085 nd 627266), the Beijing Nturl Science Foundtion (Grnt No. 442027), nd the Foundtion of Shenzhen government. Fig.. MmWve mssive MIMO bsed wireless bchul for 5G UDN. mny ultr-dense smll-cell BSs with much smller coverge provide the high dt rte for low-mobility users. Due to ultrdense smll-cell BSs, better frequency reuse cn be chieved, nd energy efficiency cn be lso substntilly improved due to the reduced pth loss in smll cells []. To enble UDN, relible, cost-effective, nd Gighertz bndwidth bchul connecting mcro-cell BS nd the ssocited smll-cell BSs is prerequisite. It hs been demonstrted tht bchul with 0 GHz bndwidth is required to effectively support UDN [2]. Conventionl opticl fiber enjoys lrge bndwidth nd relibility, but its ppliction to UDN s bchul my not be n economicl choice for opertors due to the restriction of deployment nd instlltion. Hence, wireless bchul, especilly millimeter-wve (mmwve) bchul, is more ttrctive to overcome the geogrphicl constrints. The dvntges of mmwve bchul re: A lrge mount of underutilized bnd in mmwve cn be leverged to provide the potentil Gighertz trnsmission bndwidth, which is different from scrce microwve bnd in conventionl cellulr networs [3]. A lrge number of ntenns cn be esily employed for mmwve communictions due to the smll wvelength of mmwve, which cn improve the signl directivity (reduce the co-chnnel interference) nd lin relibility (mitigte the lrge pth loss) for mmwve bchul [4]. This rticle combines mmwve with lrge number of ntenns, which is lso referred to s mmwve mssive MIMO, to provide wireless bchul for future 5G UDN. The contributions of this rticle re listed s follows: We discuss the fesibility nd chllenges of the mmwve mssive MIMO bsed bchul for UDN, where its dvntges, differences compred with conventionl mssive MIMO woring t sub 3 6 GHz for rdio ccess networs (RAN) re lso ddressed. Moreover, the sprsity of mmwve mssive MIMO chnnels is stressed. We explore ey issues nd potentil reserch directions
2 of the cost-effective mmwve mssive MIMO for UDN bchul. Especilly, digitlly-controlled phse-shifter networ (DPSN) bsed hybrid precoding/combining nd the ssocited compressive sensing (CS) bsed chnnel estimtion is proposed. We ddress the benefits of the wireless bchul for 5G UDN with the technique of mmwve mssive MIMO, which my provide vible pproch to relize the novel bchul networ topology, scheduling strtegy, efficient in-bnd bchul in mmwve. II. FEASIBILITY AND CHALLENGES OF MMWAE MASSIE MIMO FOR WIRELESS BACKHAUL IN 5G UDN In UDN, smll cells re densely deployed in hotspots (e.g., office buildings, shopping mlls, resident prtments) with high dt rte to provide trffic offlod from mcro cells, since the lrge mjority of trffic demnd comes from these hotspots. Hence, the bchul between the mcro-cell BS nd the ssocited smll-cell BSs should provide lrge bndwidth with relible lin trnsmission. Besides, power efficiency nd deployment cost re lso ey considertions for opertors. A. MmWve is Suitble for Wireless Bchul in 5G UDN Trditionlly, mmwve is not used for RAN in existing cellulr networs due to its high pth loss nd expensive electron components. However, mmwve is especilly suitble for bchul in UDN due to the following resons. High Cpcity nd Inexpensive: The lrge mount of underutilized mmwve including unlicensed -bnd (57-67GHz) nd lightly licensed E-bnd (7-76GHz nd 8-86GHz) (the specific regultion my vry from country to country) cn provide the potentil Gighertz trnsmission bndwidth [3]. For exmple, more thn one Gbps bchul cpcity cn be supported over 250 MHz chnnel in E-bnd [2]. Immunity to Interference: Trnsmission distnce comfort zone for E-bnd is up to severl ilometers due to the rin ttenution, while tht for -bnd is bout 500-700m due to both the rin nd oxygen ttenution. Owing to the high pth loss, mmwve is suitble for UDN, where the improved frequency reuse nd reduced intercell interference re expected. It should be pointed out tht rin ttenution is not big issue for mmwve used in UDN. If we consider the very hevy rinfll of 25mm/hr, the rin ttenution is only round 2 db in E- bnd if we consider the distnce of bchul lin is 200m in typicl urbn UDN [3]. Smll Form Fctor: The smll wvelength of mmwve implies tht mssive ntenns cn be esily equipped t both mcro nd smll-cell BSs, which cn improve the signl directivity nd compenste severe pth loss of mmwve to chieve lrger coverge in turn [4]. Hence the compct mmwve bchul equipment cn be esily deployed with low cost sites (such s light poles, building wlls, bus sttions) nd short instlltion time. B. MmWve Mssive MIMO is Different from Microwve Mssie MIMO Inheriting the dvntges from conventionl microwve mssive MIMO, mmwve mssive MIMO hs the flexible bemforming, sptil multiplexing, nd diversity. Hence mmwve mssive MIMO brings not only the improved relibility of bchul lin, but lso new rchitecture of bchul networ including the flexible networ topology, scheduling scheme, which will be further detiled in Section I. However, compred with conventionl microwve (sub 3 6 GHz) mssive MIMO used for RAN, the implementtion of mmwve mssive MIMO lso brings chllenges s follows. First, the cost nd complexity of trnsceiver including high-speed nlog-digitl converters (ADCs) nd digitlnlog converters (DACs), synthesizers, mixers, etc., in mmwve communictions re much lrger thn tht in conventionl microwve communictions. Hence, mssive low-cost ntenns but limited number of expensive bsebnd (BB) chins cn be n ppeling trnsceiver structure for mmwve mssive MIMO, which, however, chllenges conventionl precoding/combining schemes. Second, the number of ntenns in mmwve t both mcro nd smll-cell BSs cn be much lrger thn tht in conventionl microwve mssive MIMO due to the much smller wvelength of mmwve. This implies the chllenge tht chnnel estimtion in mmwve mssive MIMO cn be more difficult even when time division duplex (TDD) leverging the chnnel reciprocity is considered. Even for TDD-bsed mmwve communictions, the synchroniztion nd clibrtion error of rdio frequency (RF) chins to gurntee the chnnel reciprocity re not trivil [5]. Third, since single-ntenn users re typiclly considered in microwve mssive MIMO due to the limited form fctor, only chnnel stte informtion t trnsmitter (CSIT) is required for precoding. However, for mmwve mssive MIMO where ech smll-cell BS cn be equipped with mssive ntenns, precoding in the uplin nd combining in the downlin t smll-cell BSs re lso necessry, since precoding/combining cn effectively support multiple strems nd directionl trnsmission for the improved lin relibility. Therefore, chnnel stte informtion t receiver (CSIR) is lso required for mmwve mssive MIMO, which indictes nother chllenge tht chnnel estimtion cquired in the uplin by leverging the chnnel reciprocity should lso be feedbc to smll-cell BSs. III. MMWAE CHANNEL CHARACTERISTICS As discussed bove, the mmwve mssive MIMO bsed bchul is pt to the trnsceiver with the limited number of BB chins. Compred with microwve mssive MIMO using full digitl precoding, precoding/combining with the smller number of BB chins thn tht of ntenn elements cn me mmwve mssive MIMO suffer from certin performnce loss, which is lrgely dependent on the propgtion condition of mmwve mssive MIMO chnnels.
3 " # $ Fig. 2. Energy probbility distribution of singulr vlues with descending order of mmwve mssive MIMO chnnel mtrix versus different L s, where N T = 52 nd N R = 32. A. MmWve Chnnels with Sptil/Angulr Sprsity Extensive experiments hve shown tht mmwve mssive MIMO chnnels exhibit the obviously sptil/ngulr sprsity due to its high pth loss for non-line-of-sight (NLOS) signls, where only smll number of dominted multipths (typiclly, 3 5 multipths in relistic environments [6]) consist of mmwve MIMO multipth chnnels. If we consider the widely used uniform liner rry (ULA), the point-to-point mmwve mssive MIMO chnnel cn be modeled s [6] H= N T N R ρ L α l T (θ l )b R (ϕ l)= l= $ N T N R A T DB R ρ, () where N T nd N R re the numbers of trnsmit nd receive ntenns, respectively, ρ is the verge pth loss, L is the number of multipths, α l is the complex gin of the lth pth, θ l [0,2π] nd ϕ l [0,2π] re zimuth ngles of deprture or rrivl (AoD/AoA). In ddition, T (θ l ) = [ NT,e j2πdsin(θ l )/λ,,e ] j2π(nt )dsin(θ T l)/λ nd [ b R (ϕ l ) = NR,e j2πdsin(ϕ l )/λ,,e ] j2π(nr )dsin(ϕ T l)/λ re steering vectors t the trnsmitter nd receiver, respectively, A T = [ T (θ ) T (θ 2 ) T (θ L )], B R =[b R (ϕ ) b R (ϕ 2 ) b R (ϕ L )], nd the digonl mtrix D = dig{α,α 2,,α L }, where λ nd d re wvelength nd ntenn spcing, respectively. B. Low-Rn Property of MmWve Mssive MIMO Chnnels The sptil/ngulr sprsity of mmwve chnnels with smll L (e.g., 3 5) nd mssive MIMO chnnel mtrix with lrge N T, N R (dozens even hundreds) implies tht mmwve mssive MIMO chnnel mtrix hs the low-rn property [7]. For exmple, Fig. 2 provides the energy probbility distribution of singulr vlues of H with descending order ginst different L s, where N T = 52, N R = 32, nd pth gins follow the independent nd identiclly distributed (i.i.d.) complex Gussin distribution. It cn be observed tht the mmwve mssive MIMO chnnel mtrix hs the obvious low-rn property. If we consider single user (SU)-MIMO with CSIT for precoding nd CSIR for combining, the low-rn chnnel mtrix indictes tht the number of effective independent strems which cn be exploited is smll. Theoreticl nlysis hs shown tht the cpcity of MIMO systems over sprse mmwve chnnels ppers ceiling effect with the incresed number of BB chins [7]. Hence, we cn leverge the finite number of BB chins to mximize the bchul cpcity over sprse mmwve chnnels, where the number of BB chins cn be s smll s the effective rn of mmwve mssive MIMO chnnel mtrix. I. KEY ISSUES OF DESIGNING MMWAE MASSIE MIMO FOR 5G UDN BACKHAUL A. Hybrid Precoding/Combining Design In order to relize the relible point-to-multiple-points (P2MP) bchul lin, mmwve mssive MIMO for UDN PA LNA Fig. 3. Comprison of precoding/combining schemes, where PA denotes power mplifier nd LNA denotes low-noise mplifier: () Anlog precoding/combining scheme in mmwve multi-ntenn systems; (b) Digitl precoding in microwve mssive MIMO for RAN; (c) Conventionl hybrid precoding/combining in mmwve mssive MIMO for RAN; (d) Proposed DPSN bsed hybrid precoding/combining for mmwve mssive MIMO in UDN bchul.
4 bchul should exploit the flexible bemforming nd sptil multiplexing to simultneously support multiple smll-cell BSs nd provide multiple strems for ech smll-cell BS, which chllenges conventionl precoding/combining schemes. ) Overview of Existing Precoding/Combining Schemes: Conventionl mmwve multi-ntenn systems utilize single RF chin nd nlog (e.g., ferrite bsed) phse-shifters for precoding/combining s shown in Fig. 3 (), but it is limited to SU-MIMO with single strem. Full digitl precoding in microwve mssive MIMO, s shown in Fig. 3 (b), cn simultneously support multiple single-ntenn users, i.e., MU-MIMO, but it requires one specific RF chin to be connected to ech ntenn, which cn be unffordble in mmwve communictions [7]. Recently, the hybrid precoding/combining scheme consisting of nlog nd digitl precoding/combining s shown in Fig. 3 (c), hs been proposed for mmwve mssive MIMO with the reduced cost nd complexity of trnsceiver. However, stte-off-the-rt hybrid precoding/combining schemes re usully limited to SU-MIMO with multiple strems or multi-user (MU)-MIMO with single strem for ech user [4], [6] [8]. 2) Proposed DPSN Bsed Hybrid Precoding/Combining: Multi-User nd Multi-Strem: To support multi-user nd multi-strem, we propose the DPSN bsed hybrid precoding/combining scheme s shown in Fig. 3 (d), which cn effectively reduce the cost nd complexity of trnsceiver. Specificlly, consider the mcro-cell BS hs N M BB chins, where N M N M BB ntenns but NBB M, while ech smll-cell BS hs N Sm ntenns but NBB Sm BB chins, where NSm NBB Sm. The number of simultneously supported smll-cell BSs is K. denotes the mmwve mssive MIMO chnnel mtrix ssocited with the mcro-cell BS nd the th smll-cell BS, nd it cn be expressed s follows ccording to singulr vlue decomposition (SD): H C NM N Sm H = [ U U 2 ] with N M Σ 0 0 Σ 2 0 0 > N Sm [ ( ) ( 2 ) ] U Σ ( ), (2) where both [ ] U U2 C NM N M nd [ ] 2 C NSm N Sm re unitry mtrices, Σ CR R nd Σ 2 C (NSm R ) (N Sm R ) re digonl mtrices whose digonl elements re singulr vlues of H, nd R is the effective rn of H. The pproximtion in (2) is due to the low-rn property of H with Σ 2 0, so tht U CNM R ( ) nd C R N Sm. Eq. (2) indictes tht NBB M nd NSm BB cn be reduced to R in SU-MIMO due to only N s = R effective independent strems. ( ) Moreover, we cn use the precoding mtrix P = U nd the combining mtrixc = to effectively relize the independent multi-strem trnsmission [9]. To chieve this gol, we cn use the emerging low-cost silicon-bsed SiGe nd CMOS bsed progrmmble DPSN [0] to relize prtil precoding/combining in the nlog RF. With the cscde of the digitl precoding mtrix P d, C R R (or combining mtrix C d, C R R ) nd nlog precoding mtrix P, C R N M (or combining mtrix C, C NSm R ), we cn use P d, P, (or C, C d, ) to pproximte P (or C ). Consider the precoding for instnce, we cn use the following itertive pproch to cquire P d, nd P, tht cn minimize P P d, P, F with the constrint tht elements in P, re constnt modulus. We initilize tht P P. Then, we perform the following opertions itertively until P, nd P d, converge: ) every element of P, hs the sme phse with the corresponding element in P ; 2) P d, P (P, ), 3) P (P d, ) P. Note tht P, lwys meets the constrint of constnt modulus nd () is the Moore-Penrose pseudoinverse. Similrly, we cn cquire C d, nd C, ccording to C with the sme pproch. Besides, some power lloction strtegies such s wterfilling cn be integrted in the digitl bsebnd precoding/combining to further improve the chievble cpcity. Furthermore, consider the downlin MU-MIMO, where the chnnel mtrix between mcro-cell BS nd K smll-cell BSs cn be denoted s H C NM KN Sm, nd it cn be ) represented s H=[H H 2 H K ] with H U ( Σ for K ccording to (2). Hence we further obtin H [ { U U 2 K] U dig Σ,Σ 2, } { (,Σ K ), ( ) ( ) } dig 2,, K, where H for K re ssumed to shre the sme effective rn R = R. For precoding/combining in the proposed MU-MIMO system, the nlog precoding ] mtrix t T C KR N M mcro-cell BS isp = [ P T, PT,2 PT,K nd the nlog nd digitl combining mtrices for the th smll-cell BS cn be C, nd C d,, respectively. To further (3), eliminte ) the multi-user interference, digitl precoding P d = ( Pd P Ũ is proposed t the mcro-cell BS, where Pd = dig{p d,,p d,2,,p d,k } nd Ũ = [ U U 2 U K]. The precoding/combining in the uplin of mmwve mssive MIMO bsed bchul is similr to the downlin, which will not be detiled in this rticle owing to the spce limittion. The proposed precoding/combining scheme cn digonlize the equivlent chnnel P d P Hdig{C,C 2,,C K } with C = C, C d, to relize multi-user nd multi-strem trnsmission, which is essentilly different from existing schemes. Moreover, thns to the obvious low-rn property of mmwve mssive MIMO chnnel mtrix s shown in Fig. 2, the proposed precoding/combining with the reduced number of BB chins only suffers from negligible performnce loss, which will be shown in Section. B. CSI Acquisition for MmWve Mssive MIMO To effectively relize the proposed DPSN bsed hybrid precoding/combining scheme, relible CSI cquisition scheme with low overhed is nother chllenge. ) Chllenging Chnnel Estimtion for MmWve Mssive MIMO: As we hve discussed in Section II-B, mmwve mssive MIMO my suffers from the prohibitively high overhed for chnnel estimtion, nd clibrtion error of RF chins s well s synchroniztion re lso not trivil in TDD. Additionlly, due to the much smller number of BB chins thn tht of ntenns, the effective dimensions tht cn be exploited for chnnel estimtion will be substntilly reduced lthough mssive ntenns re employed. Furthermore, chnnel estimtion in the digitl bsebnd should consider the
5 Fig. 4. CS-bsed chnnel estimtion for mmwve mssive MIMO bsed UDN bchul: () Corse chnnel estimtion; (b) AoA nd pth gins estimtion t smll-cell BS; (c) AoD nd pth gins estimtion t mcro-cell BS. chrcteristics of phse-shifter networs t both mcro-cell BS nd smll-cell BSs, which cn me it more complex. Finlly, due to the strong signl directivity of mmwve, relible chnnel estimtion requires the sufficient received signl power, which mens t lest prtil CSIT is necessry to ensure bemforming t the trnsmitter to mtch mmwve MIMO chnnels. 2) Overview of Existing Chnnel Estimtion Schemes: Wireless locl re networs (WLANs) (IEEE 802. d) relies on the bemforming trining to compenste the lrge pth loss in 60 GHz [7]. The specific trining consists of three phses: ) sector level sweep is to select the best trnsmit nd optionlly receive ntenn sector; 2) bem refinement is used for fine djustment of bemforming; 3) bem trcing cn djust bemforming during dt trnsmission. In wireless personl re networs (WPANs) (IEEE 802.5.3c), codeboo is designed in scenrios of indoor communictions with the smll number of ntenns [7], where the bemforming protocol is similr to tht in IEEE 802. d. However, both of them only consider the nlog bemforming (precoding). [6] proposed n hierrchicl multi-resolution codeboo bsed chnnel estimtion for hybrid precoding/combining scheme. However, the proposed scheme my suffer from the destructive interference between the pth gins when multiple pths re summed up in the erlier stges of the proposed lgorithm [6]. 3) Proposed CS-Bsed Chnnel Estimtion for MmWve Mssive MIMO: Some unique fetures of mmwve mssive MIMO chnnels cn be leverged to llevite the chllenging problem of chnnel estimtion. Due to the fixed BSs loction, mmwve mssive MIMO chnnels used for bchul eep lmost unchnged for long time. Such long coherence time of chnnels indictes tht chnnels re not necessry to be estimted very frequently compred with tht used for RAN. The low-rn property of mmwve mssive MIMO chnnel mtrix indictes tht lthough the dimension of mmwve mssive MIMO chnnel mtrix cn be huge, its effective degrees of freedom (DoF) cn be smll. This inspires us to reconstruct chnnel mtrix with significntly reduced mesurements (sub-nyquist smpling) under the frmewor of CS []. By leverging these fetures, we propose CS-bsed chnnel estimtion scheme s illustrted in Fig. 4, which consists of the following three phses: Phse : corse chnnel estimtion, s illustrted in Fig. 4 (), ims to cquire prtil CSIT to generte the pproprite bemforming ptterns for the following fine chnnel estimtion with the improved received signl power. Specificlly, the mcro-cell BS sequentilly brodcsts L M predefined bemforming ptterns in L M successive time slots, while in every time slot, ech smllcell BS sequentilly receives signl with L Sm combining ptterns in L Sm successive sub-time slots. Then ech smll-cell BS feedbcs the indices of severl optiml bemforming/combining ptterns to the mcro-cell BS. Phse 2: chnnel estimtion t smll-cell BS, s shown in Fig. 4 (b), ims to estimte AoA nd pth gins t ech smll-cell BS. The mcro-cell BS performs bemforming ccording to the feedbc, while the th smllcell BS estimtes AoA nd pth gins by exploiting the finite rte of innovtion (FRI) theory (nlog CS) []. With the id of the predefined trining signls S C TSm gin NM BB, the received signls t the smll-cell BS is SP d P A T DB R C C d ccording to (), where the index is omitted, Tgin Sm is the time overhed. Since DPSN cn disble some phse-shifters to set some elements of C to be zeros, the AoA nd pth gins estimtion cn be solved by the specific lgorithms of FRI theory, e.g., estimting signl prmeters virottionl invrince techniques (ESPRI) lgorithm []. Phse 3: chnnel estimtion t mcro-cell BS, s shown in Fig. 4 (c), ims to estimte AoD nd pth gins t the mcro-cell BS. The specific procedure is similr to Phse 2, where the th smll-cell BS trnsmits trining signl while the mcro-cell BS estimtes chnnels. The FRI theory [] cn be used to ccurtely cquire the super-resolution estimtion of AoA/AoD to effectively distinguish multiple pths with smll ngulr difference, which cn lso relx the required resolution of the bemforming ptterns in corse chnnel estimtion. According to the estimted prmeters nd Eq. (), both mcro-cell BS nd smllcell BSs cn cquire the complete CSI for the following precoding/combining. 4) Other Issues of Chnnel Estimtion: There still remins some problems to be investigted further, such s the optiml bemforming/combining ptterns in corse chnnel estimtion [6], [7], trining signls for AoA/AoD nd pth gins estimtion [8], low-complexity high-ccurcy CS-bsed chnnel
6 " #$ % " & ) & * " #$ % " & ) & * + & + & ',- ',- Fig. 5. Cpcity comprison between the proposed hybrid precoding/combining scheme nd the optiml full digitl one: () K fctor = 0 db; (b) K fctor = 0 db. estimtion lgorithms, effective chnnel feedbc scheme, dynmic chnnel trcing to combt sudden blocge or slow chnnel chnges. For instnce, the microwve control lin with only limited resource cn be used to feedbc the estimted prmetric AoA/AoD, since the number of AoA/AoD is typiclly 3 5 [6]. Regrding the CS-bsed chnnel estimtion lgorithm, in ddition to FRI theory, other CS pproches such s low-rn mtrix reconstruction re expected to be tilored for mmwve mssive MIMO with low complexity []. Besides, the proposed chnnel estimtion scheme (including the corse chnnel estimtion nd the following estimtion of AoA/AoD nd pth gins) is used to initilly build the UDN bchul lin, where the ltency cn be negligible. Once the bchul lin is built, only the estimtion of AoA/AoD nd pth gins is required to trc the chnnels nd then djust the corresponding precoding/combining, where the trining sequences nd dt cn be multiplexing in the time domin, so the ltency cn be lso negligible. C. Low-Complexity Hybrid Precoding/Combining for MmWve mssive 3D MIMO For UDN in urbn re, the precoding/combining scheme for mmwve mssive 3D MIMO cn exploit both zimuth nd elevtion to chieve the improved performnce for bchul lin. Hence, ULA bsed hybrid precoding/combining scheme nd the ssocited chnnel estimtion proposed in this rticle should be extended to mmwve mssive 3D MIMO in the future. Additionlly, SD nd wterfilling my impose higher computtionl complexity on the hybrid precoding/combining in the 3D MIMO. Therefore, low-complexity hybrid precoding/combining schemes re desired for prcticl system design. For instnce, the sptil/ngulr sprsity of mmwve mssive MIMO chnnels nd the geometric structure of mmwve mssive 3D MIMO my be exploited to reduce the complexity of SD, while other sub-optiml power lloctions cn be considered to pproch the performnce of wterfilling with much low complexity. D. Smpling with Low-Resolution ADC Smll cells in future UDN cn provide the Gbps user experience, which requires the lrge trnsmission bndwidth. To relize mmwve mssive MIMO bsed bchul, the cost of conventionl high-speed ADC with high resolution cn be unffordble, while low-resolution ADC with low hrdwre cost is ppeling. By fr, -bit ADC bsed signl detector nd precoding/combining hve been investigted for mmwve mssive MIMO [7], [2]. However, further efforts re still needed to generlize the ssocited results of -bit to more generl cses, nd constelltion mpping, chnnel estimtion, trining signls, etc., my need to be reconsidered if lowresolution ADC is dopted.. PERFORMANCE COMPARISON Fig. 5 compres the cpcity (bit per chnnel use, bpcu) of the proposed DPSN bsed hybrid nd the optiml full digitl precoding/combining schemes in the downlin, where both the wterfilling power lloction nd equl-power lloction re investigted. In simultions, ULA is considered t both mcro nd smll-cell BSs, the woring frequency is 60 GHz, = 32. For the optiml full digitl scheme, NBB Sm = NSm nd NBB M = NM, where the idel CSIT nd CISR re ssumed s the upper bound of cpcity. In the proposed scheme, NBB Sm = 4 nd NM BB = 6, K = 4, N M = 52, nd N Sm where cses of idel CSI nown by trnsceiver nd unidel CSI cquired by the proposed CS-bsed chnnel estimtion scheme re considered. For mmwve mssive MIMO chnnels, L in simultions follows the discrete uniform distribution U d [2,6], nd AoA/AoD follow the continuous uniform distribution U c [0,2π). For pth gins, we consider Ricn fding consisting of one LOS pth nd L equl-power NLOS pths, where pth gins follow the mutully independent complex Gussin distribution with zero mens, nd K fctor denotes the rtio between the power of LOS pth nd the power of NLOS pth. Fig. 5 shows tht the proposed hybrid scheme with idel CSIT nd CSIR suffers from negligible cpcity loss compred with the optiml full digitl scheme, lthough the proposed scheme only uses much smller number of BB chins. This is becuse the proposed scheme exploits the lowrn property of mmwve mssive MIMO chnnel mtrix, where cpcity exhibits ceiling effect when the number of BB chins re sufficiently lrge. Moreover, with the incresed
7 Tgin Sm, cpcity of the proposed scheme with CS-bsed chnnel estimtion pproches tht with the idel CSI. This is becuse the incresed number of mesurements cn improve the chnnel estimtion performnce. Besides, schemes with wterfilling power lloction outperform these with equlpower lloction, which indictes tht wterfilling or other power lloctions should be considered in prcticl system design for the improved bchul cpcity. I. BENEFITS AND OPPORTUNITIES OF MMWAE MASSIE MIMO BASED BACKHAUL NETWORK A. Point-to-Multiple-Points Bchul In conventionl wireless bchul networ, point-to-point (P2P) nd P2MP re two typicl networ topologies. A generl consensus is P2MP hs the lower totl cost of ownership thn tht in P2P [2]. By fr, P2P hs been widely used in both microwve nd mmwve bchul systems, while P2MP hs been implemented in sub 6 GHz licensed bnd. However, there re no stisfctory P2MP bsed bchul solutions in mmwve, which is urgently desired by industry. In this rticle, the proposed mmwve mssive MIMO bsed wireless bchul enbles mcro-cell BS to simultneously support multiple smll-cell BSs, which cn provide the vible pproch to relize the P2MP bchul in mmwve. B. Bem Division Multiplex (BDM) Bsed Scheduling Time division multiplex (TDM) bsed scheduling hs been proposed for mmwve bsed bchul [2]. However, this scheme my suffer from the ltency, since bchul lins between different smll-cell BSs nd mcro-cell BS re multiplexing in different time slots. In this rticle, the proposed mmwve mssive MIMO bsed bchul networ cn relize the BDM bsed scheduling due to the flexible sptil multiplexing nd hybrid bemforming. In the prcticl bchul networ, ccording to the bchul lod, the mcro-cell BS cn flexibly combine the TDM nd BDM to support more smll-cell BSs. For instnce, lins with hevy lod or without LOS pth cn be ssigned more bem resources, while multiple lins with light lod cn be multiplexing in the time domin. Additionlly, TDM bsed in-bnd mmwve bchul is recently proposed for the reduced cost, where bchul nd RAN shre the sme frequency bnd [2]. Obviously, with the proposed mmwve mssive MIMO scheme, the BDM bsed scheduling my be nother competitive solution for the in-bnd mmwve bchul with lower ltency. C. TDD is Suitble for MmWve Bchul Networ For frequency division duplex (FDD)-bsed mmwve bchul, the uplin nd downlin hve to use different bnds. However, the regultion in mmwve my be different in different countries. This indictes tht one single device my be not suitble in vrious countries. By contrst, the uplin nd downlin in TDD shre the sme bnd. Hence one single device cn be employed in vrious countries. Moreover, since different opertors will employ UDN in the sme res, the mutul interference of bchul networs must be considered. Compred to FDD with the different uplin/downlin chnnels, TDD is more esier to find clen spectrum nd void interference. Moreover, since the symmetric trffic is dominnt in bchul networ, TDD cn flexibly djust the rtio of time slots in the uplin nd downlin ccording to the trffic requirement [5]. For the prcticl TDD mmwve mssive MIMO bsed bchul, the dptive interference mngement is desired to void mutul interference of different opertors UDN, nd utomted configurtion solutions re expected for the plug-nd-ply bchul networ, especilly for unlicensed -bnd. While for licensed E-bnd, spectrum regultion needs to be further improved. II. CONCLUSIONS This rticle discusses promising wireless bchul bsed on mmwve mssive MIMO for future 5G UDN. We hve explored the fundmentl issues of the implementtion of the mmwve mssive MIMO for wireless bchul. Especilly, by leverging the low-rn property of mmwve mssive MIMO chnnel mtrix, we propose the DPSN bsed hybrid precoding/combining scheme nd the ssocited CS-bsed chnnel estimtion scheme. The proposed scheme cn gurntee the mcro-cell BS to simultneously support multiple smll-cell BSs with multiple strems for ech smll-cell BS. This is essentilly different from conventionl hybrid precoding/combining used for RAN. The proposed scheme cn provide the vible pproch to relize the desired P2MP bchul topology nd novel BDM bsed scheduling, nd it my lso fcilitte the in-bnd bchul in mmwve. Additionlly, some potentil reserch directions to enble mmwve mssive MIMO bsed wireless bchul re highlighted, which my become ctive reserch topics in ner future. REFERENCES [] C.-L. I, C. Rowell, S. Hn, Z. Xu, G. Li, nd Z. Pn, Towrd green nd soft: A 5G perspective, IEEE Commun. Mg., vol. 52, no. 2, pp. 66-73, Feb. 204. [2] R. Tori nd A. Sridhrn, Point-to-multipoint in-bnd mmwve bchul for 5G networs, IEEE Wireless Commun., vol. 53, no., pp. 95-20, Jn. 205. [3] L. Wei, R. Q. Hu, Y. Qin, nd G. Wu, Key elements to enble millimeter wve communictions for 5G wireless systems, IEEE Wireless Commun., vol. 2, no. 6, pp. 36-43, Dec. 204. [4] S. Hn, C.-L. I, Z. Xu, nd C. Rowell, Lrge-scle ntenn systems with hybrid precoding nlog nd digitl bemforming for millimeter wve 5G, IEEE Commun. Mg., vol. 53, no., pp. 86-94, Jn. 205. [5] P. Chn, E. Lo, R. Wng, E. Au,. Lu, R. Cheng, W. Mow, R. Murch, nd K. Letief, The evolution pth of 4G networs: FDD or TDD? IEEE Commun. Mg., vol. 44, no. 2, pp. 42-50, Dec. 2006. [6] A. Alhteeb, O. E. Aych, G. Leus, nd R. W. Heth, Chnnel estimtion nd hybrid precoding for millimeter wve cellulr systems, IEEE J. Sel. Topics Signl Process., vol. 8, no. 5, pp. 83-846, Oct. 204. [7] A. Alhteeb, J. Mo, N. Gonzlez-Prelcic, nd R. W. Heth, MIMO precoding nd combining solutions for millimter-wve systems, IEEE Commun. Mg., vol. 52, no. 2, pp. 22-3, Dec. 204. [8] S. Hn, C.-L. I, Z. Xu, nd S. Wng, Reference signls design for hybrid nlog nd digitl bemforming, IEEE Commun. Lett., vol. 8, no. 7, pp. 9-93, Jul. 204. [9] D. Tse, Fundmentls of wireless communiction, Cmbridge, U.K.: Cmbridge University Press, 2005. [0] Y. Yu, P. Bltus, nd A. Roermund, Integrted 60 GHz RF bemforming in CMOS, Springer, 20. [] Y. C. Eldr nd G. Kutynio, Compressed Sensing: Theory nd Applictions, Cmbridge University Press, My 202. [2] S. Wng, Y. Li, nd J. Wng, Multiuser detection in mssive sptil modultion MIMO with low-resolution ADCs, IEEE Trns. Wireless. Commun., vol. 4, no. 4, pp. 256-268, Apr. 205.