A General 3D Non-Stationary Massive MIMO GBSM for 6G Communication Systems
AA General 3D Non-Stationary Massive MIMOGBSM for 6G Communication Systems
Yi Zheng , Long Yu , Runruo Yang , and Cheng-Xiang Wang
National Mobile Communications Research Laboratory, School of Information of Science and Engineering,Southeast University, Nanjing 210096, China. Purple Mountain Laboratories, Nanjing 211111, China. * Corresponding Author: Cheng-Xiang WangEmail: { zheng yi, yulong, yangrr, chxwang } @seu.edu.cn Abstract —A general three-dimensional (3D) non-stationarymassive multiple-input multiple-output (MIMO) geometry-basedstochastic model (GBSM) for the sixth generation (6G) communi-cation systems is proposed in the paper. The novelty of the modelis that the model is designed to cover a variety of channel charac-teristics, including space-time-frequency (STF) non-stationarity,spherical wavefront, spatial consistency, channel hardening, etc.Firstly, the introduction of the twin-cluster channel model isgiven in detail. Secondly, the key statistical properties such asspace-time-frequency correlation function (STFCF), space cross-correlation function (CCF), temporal autocorrelation function(ACF), frequency correlation function (FCF), and performanceindicators, e.g., singular value spread (SVS), and channel ca-pacity are derived. Finally, the simulation results are given andconsistent with some measurements in relevant literatures, whichvalidate that the proposed channel model has a certain value asa reference to model massive MIMO channel characteristics.
Index Terms —Massive MIMO, STF non-stationarity, GBSM,channel hardening, channel capacity
I. I
NTRODUCTION
Compared with the fifth generation (5G) communicationsystems, the 6G communication systems have attracted moreand more attention because of almost a thousand times trans-mission rate and capacity [1], [2]. Massive MIMO technologyis an efficient way to increase capacity and spectral efficiencyfor 6G communication systems, which refers to that the basestation (BS) is equipped with a large number of antennas upto one hundred or even thousands of antennas. Besides, withmore and more antennas exploited at BS side, the channelamong different users become approximatively orthogonal,which is called channel hardening phenomenon (or favorablepropagation conditions) [3], [4]. Therefore, the interferenceamong the users can be removed, which makes the 6Gcommunication systems inherently robust.There are mainly three new characteristics for massiveMIMO channel: spherical wavefront, spatial non-stationarity,and channel hardening phenomenon. Spherical wavefrontrefers to that the distance between the transmitter (Tx) andreceiver (Rx) or cluster is less than the Rayleigh distance L /λ , where L represents the antenna array size, λ denotesthe wavelength. Channel measurements showed that the angleof departure (AoD) along the antenna array gradually shifts[5] and line-of-sight (LOS) path azimuth angle shifts along the array [6], which demonstrated the spherical wavefront. Birth-death process along the array brings spatial non-stationarity.The clusters appear and disappear along the array randomly,which was verified by the fact that received power of LOS pathvaries along the array [6]. In [7], theoretical analysis showedthat the correlation matrix at user side becomes a diagonalmatrix under favorable propagation conditions and channelhardening phenomenon is obvious. The above characteristicsbring new requirements for massive MIMO channel modeling.A general massive MIMO channel model for 6G communi-cation systems should be at least suitable for millimeter wavecommunications, vehicle-to-vehicle (V2V) communications,3D communication environments, and high-speed train (HST)communications. In [8], a two-dimensional (2D) parabolicwavefront model was proposed, and a 3D parabolic wavefrontmodel was further developed [9]. However, the model onlyconsidered the movements of the Rx and clusters, so it was notapplicable to V2V scenario. Similarly, a twin-cluster channelmodel in [10] did not consider the movement of the Tx. Ageneral 3D non-stationary 5G channel model and a general 3Dnon-stationary channel model for 5G and beyond were givenin [11] and [12], respectively. The two models only consideredthe cluster evolution in space domain and time domain,which was not suitable for millimeter wave communicationsystems needing to consider cluster evolution in frequencydomain. References [13] and [14] demonstrated a multi-ringchannel model and a multi-confocal ellipse channel model,respectively. Both of them were 2D channel models withoutconsidering elevation characteristic. Reference [15] proposeda 3D ellipsoid model, which did not consider the movement ofthe Tx and cluster. Reference [16] proposed a millimeter wavemassive MIMO channel for HST communications withoutconsidering spatial consistency and V2V communications. Theabove mentioned channel models do not consider all therequirements for 6G massive MIMO channel modeling.To the best of our knowledge, the general 3D massiveMIMO GBSM considering STF cluster evolution, sphericalwavefront, spatial consistency, and channel hardening for 6Gcommunication systems is still missing in the literature. Thispaper presents a general 3D massive MIMO GBSM based onthe model in [12] so as to fill the above research gaps. The a r X i v : . [ ee ss . SP ] J a n ontributions of the paper are summarized as follows. Firstly,the proposed channel model is suitable for millimeter wavecommunications, V2V communications, 3D communicationenvironments, and HST communications by adjusting channelparameters. Secondly, the cluster evolution is further extendedfrom space and time domain in [12] to STF domain. Thirdly,the large scale parameters (LSPs) and small scale parameters(SSPs) are generated according to the positions of Tx andRx using the sum-of-sinusoids (SoS) method, which makesthe model inherently spatially consistent. Finally, the STFnon-stationarity, spatial consistency, and channel hardeningcharacteristics are verified by simulation results.The remaining paper is structured as follows. Section II de-scribes the proposed general massive MIMO GBSM in detail.In Section III, statistical properties and performance indicatorsof the presented model are derived. Simulation results andanalysis are given in Section IV. Finally, conclusions are drawnin Section V.II. A GENERAL M ASSIVE
MIMO GBSMAs illustrated in Fig. 1, large uniform rectangular arrays(URAs) are adopted at the BS and mobile station (MS)sides in this model. Suppose that the BS is Tx and the MSis Rx. The URA at BS (MS) side is formed by uniformlinear arrays (ULAs) in two dimensions. There are M T ( M R )antenna elements symboled as A Tp ( p = 1 , , · · · , M T ) ( A Rq ( q = 1 , , · · · , M R )) and spaced at a distance δ T ( δ R ) inone dimension. In another dimension, there are N T ( N R )antenna elements symboled as A Tu ( u = 1 , , · · · , N T ) ( A Rv ( v = 1 , , · · · , N R )) spaced at a distance δ T ( δ R ). In the M T ( M R ) antenna elements dimension, the angle of elevation is β TE ( β RE ) and the angle of azimuth is β TA ( β RA ). In order to calculateconveniently, we consider the M T ( M R ) antenna elementsas a ULA. All the ULAs in the N T ( N R ) antenna elementsdimension can be added to form the whole URA. Multi-bouncepropagation is simplified as twin-cluster propagation. The pathbetween the first bounce cluster C An and the last bounce cluster C Zn is abstracted by a virtual link. The total number of pathsfrom A Tp to A Rq at time t is N qp ( t ) . The number of scatterersin the n th path is M n ( t ) . The Tx, Rx, and clusters canmove with arbitrary velocities and trajectories. Furthermore,all the parameters are time-variant. For clarity, the remainingdefinitions of the parameters are shown in Table I. A. Channel Impulse Response (CIR)
The complete channel matrix is comprised of large scalefading (LSF) part and small scale fading (SSF) part. The LSFconsists of path loss (
P L ), shadowing ( SH ), blockage loss( BL ), and gas absorption loss ( AL ). The theoretical channelmatrix is presented as H = [ P L · SH · BL · AL ] / H s (1)where H s is the SSF matrix and can be further represented as H s = [ h qp ( t, τ )] M R N R × M T N T (2) where h qp ( t, τ ) can be acquired by the summation of LOScomponent and non-line-of-sight (NLOS) components. h qp ( t, τ ) = (cid:115) K RF ( t ) K RF ( t ) + 1 h Lqp ( t, τ )+ (cid:115) K RF ( t ) + 1 h Nqp ( t, τ ) (3)where K RF ( t ) is Rician factor, h Lqp ( t, τ ) is LOS componentand h Nqp ( t, τ ) NLOS components. LOS component can berepresented as h Lqp ( t, τ ) = (cid:34) F q,V p ( φ RE,LOS ( t ) , φ RA,LOS ( t )) F q,H p ( φ RE,LOS ( t ) , φ RA,LOS ( t )) (cid:35) T · (cid:34) e jθ VpVpLOS − e jθ HpHpLOS (cid:35) (cid:34) F p,V p ( φ TE,LOS ( t ) , φ TA,LOS ( t )) F p,H p ( φ TE,LOS ( t ) , φ TA,LOS ( t )) (cid:35) · e j πf c τ LOSqp ( t ) δ ( τ − τ LOSqp ( t )) . (4)NLOS components can be represented as h Nqp ( t, τ ) = N qp ( t ) (cid:88) n =1 M n ( t ) (cid:88) m =1 (cid:34) F q,V p ( φ RE,m n ( t ) , φ RA,m n ( t )) F q,H p ( φ RE,m n ( t ) , φ RA,m n ( t )) (cid:35) T · e jθ VpVpmn (cid:113) κ − m n ( t ) e jθ VpHpmn (cid:113) κ − m n ( t ) e jθ HpVpmn e jθ HpHpmn · (cid:34) F p,V p ( φ TE,m n ( t ) , φ TA,m n ( t )) F p,H p ( φ TE,m n ( t ) , φ TA,m n ( t )) (cid:35) (cid:113) P qp,m n ( t ) e j πf c τ qp,mn ( t ) · δ ( τ − τ qp,m n ( t )) (5)where ( · ) T denotes the transpose operation. F T ( R ) p ( q ) ,V p ( · ) and F T ( R ) p ( q ) ,H p ( · ) represent the vertical polarization and horizontalpolarization at Tx (Rx) side, respectively. κ m n denotes thecross polarization ratio. An C Zn C z z x D T A R A Tp A T TM A R RM A Rq A ,m n TE A,m n T TE RE RA n RA,m E,m n R m n T d m n R d p,m (t) n T d ,m (t) n Rq d (t) A v (t) Z v (t) T v (t) R v (t) qp D u T A T TN A AT Rv A R RN A Fig. 1. A general 3D massive MIMO GBSM for 6G communication systems.ABLE ID
EFINITION OF K EY C HANNEL M ODEL P ARAMETERS . Parameters Definition f c Carrier frequency D Distance from A T to A R at initial time d T ( R ) m n Distance from A T ( R )1 to the m th scatterer in C A ( Z ) n at initial time d T ( R ) p ( q ) ,m n ( t ) Distance from A T ( R ) p ( q ) to the m th sactterer in C A ( Z ) n at time td qp,m n ( t ) Distance from A Tp through the m th scatterer in C An and the m th scatterer in C Zn to A Rq at time tτ qp,m n ( t ) Delay from A Tp through the m th scatterer in C An and the m th scatterer in C Zn to A Rq at time tv T ( t ) , v R ( t ) , v A n ( t ) , v Z n ( t ) Speeds of the Tx, Rx, cluster C An , and cluster C Zn at time tα TA ( t ) , α RA ( t ) , α A n A ( t ) , α Z n A ( t ) Azimuth angles of movements of the Tx, Rx, cluster C An , and cluster C Zn at time tα TE ( t ) , α RE ( t ) , α A n E ( t ) , α Z n E ( t ) Elevation angles of movements of the Tx, Rx, cluster C An , and cluster C Zn at time tφ TA,LOS , φ TE,LOS
Azimuth angle of departure (AAoD) and elevation angle of departure (EAoD) from A T to A R at initial time φ RA,LOS , φ RE,LOS
Azimuth angle of arrival (AAoA) and elevation angle of arrival (EAoA) from A R to A T at initial time φ T ( R ) A,m n , φ T ( R ) E,m n AAoD (AAoA) and EAoD (EAoA) from A T ( R )1 to the m th scatterer in C A ( Z ) n at initial time P qp,m n ( t ) Power of the ray from A Tp through the m th scatterer in C An and the m th scatterer in C Zn to A Rq at time t B. Channel Transfer Function (CTF)
Take the Fourier transform of the CIR, we will get the CTFas H qp ( t, f ) = (cid:115) K RF ( t ) K RF ( t ) + 1 H Lqp ( t, f )+ (cid:115) K RF ( t ) + 1 H Nqp ( t, f ) . (6)where H Lqp ( t, f ) is LOS component and H Nqp ( t, f ) NLOScomponents.
C. STF Cluster Evolution
The proposed channel has the characteristic of STF non-stationarity. Clusters may appear and disappear in STF do-main. The space-time evolution is modeled jointly. For initialmoment t i and antenna element A Tp ( A Rq ), the cluster isrepresented as C Tp ( t i ) ( C Rq ( t i ) ). At the next moment t i + (cid:52) t ,the cluster evolves into C Tp +1 ( t i + (cid:52) t ) ( C Rq +1 ( t i + (cid:52) t ) ). Thespace-time evolution process can be modeled as C Tp ( t i ) E −→ C Tp +1 ( t i + (cid:52) t ) ( p = 1 , , · · · , M T − (7) C Rq ( t i ) E −→ C Rq +1 ( t i + (cid:52) t ) ( q = 1 , , · · · , M R − . (8)By defining λ G and λ R as the generation rate and recom-bination rate of the cluster, the survival probabilities of theclusters at Tx and Rx sides can be represented as P T survival ( (cid:52) t, δ p ) = e − λ R [ ( (cid:15) T ) +( (cid:15) T ) +2 (cid:15) T (cid:15) T cos ( α TA − β TA ) ] / (9) P R survival ( (cid:52) t, δ q ) = e − λ R [ ( (cid:15) R ) +( (cid:15) R ) +2 (cid:15) R (cid:15) R cos ( α RA − β RA ) ] / (10)where (cid:15) T = δ p cosβ TE D Ac , δ p = ( p − δ T , ( (cid:15) R = δ q cosβ RE D Ac , δ q =( q − δ R ), and (cid:15) T = v T (cid:52) tD Sc ( (cid:15) R = v R (cid:52) tD Sc ) represent the distancedifferences caused by array evolution and time evolution, respectively. D Ac and D Sc are scenario-dependent coefficientsin space domain and time domain, respectively. Combinedwith frequency evolution, the total survival probability isrepresented as P survival ( (cid:52) t, δ p , δ q , (cid:52) f ) = P T survival ( (cid:52) t, δ p ) · P R survival ( (cid:52) t, δ q ) · P survival ( (cid:52) f ) (11)where P survival ( (cid:52) f ) can be further represented as [16] P survival ( (cid:52) f ) = e − λ R F ( (cid:52) f ) Dfc (12)where F ( (cid:52) f ) and D fc are determined by channel measure-ments. Furthermore, the number of the clusters which arenewly generated by STF evolution can be represented as E [ N new ] = λ G λ R (1 − P survival ( (cid:52) t, δ p , δ q , (cid:52) f )) . (13)III. S TATISTICAL P ROPERTIES AND P ERFORMANCE I NDICATOR
In this section, statistical properties and performance indi-cators of the proposed 3D massive MIMO GBSM are derived.
A. The STFCF
According to (6), we can get H qp ( t, f ) and H ∗ q (cid:48) p (cid:48) ( t + (cid:52) t, f + (cid:52) f ) , then the STFCF can bedefined as (16) at the bottom of the page, where E [ · ] defines expectation operation, and ( · ) ∗ defines theconjugation operation. R Lqp,q (cid:48) p (cid:48) ( t, f ; (cid:52) t, (cid:52) f, δ T , δ R ) and R Nqp,q (cid:48) p (cid:48) ( t, f ; (cid:52) t, (cid:52) f, δ T , δ R ) represent the STFCF of LOScomponent and NLOS components, respectively. B. The Space CCF
In terms of (3), we can get h qp ( t ) and h ∗ q (cid:48) p (cid:48) ( t ) easily. Thespace CCF can be denoted as ρ qp,q (cid:48) p (cid:48) ( t ; δ T , δ R ) = E (cid:2) h qp ( t ) h ∗ q (cid:48) p (cid:48) ( t ) (cid:3) = K RF ( t ) K RF ( t ) + 1 · ρ Lqp,q (cid:48) p (cid:48) ( t ; δ T , δ R ) + 1 K RF ( t ) + 1 · ρ Nqp,q (cid:48) p (cid:48) ( t ; δ T , δ R ) . (15) . The Temporal ACF According to (3), we can get h qp ( t ) and h ∗ qp ( t + (cid:52) t ) . Thetemporal ACF can be denoted as r qp,qp ( t ; (cid:52) t ) = E (cid:2) h qp ( t ) h ∗ qp ( t + (cid:52) t ) (cid:3) = (cid:115) K RF ( t ) K RF ( t ) + 1 · K RF ( t + (cid:52) t ) K RF ( t + (cid:52) t ) + 1 r Lqp,qp ( t ; (cid:52) t )+ (cid:115) K RF ( t ) + 1 · K RF ( t + (cid:52) t ) + 1 r Nqp,qp ( t ; (cid:52) t ) . (16) D. The FCF
In terms of (6), we can get H qp ( t, f ) and H ∗ qp ( t, f + (cid:52) f ) .The FCF can be denoted as κ qp,qp ( t, f ; (cid:52) f ) = E (cid:2) H qp ( t, f ) H ∗ qp ( t, f + (cid:52) f ) (cid:3) = K RF ( t ) K RF ( t ) + 1 κ Lqp,qp ( t, f ; (cid:52) f ) + 1 K RF ( t ) + 1 κ Nqp,qp ( t, f ; (cid:52) f ) . (17) E. The SVS
The channel matrix can be represented as singular valuedecomposition H = U Σ V (18)where U and V are used to represent unitary matrixes, Σ isused to represent K × M diagonal matrix. K and M denotethe number of users and Tx antenna elements, respectively.Furthermore, the SVS can be calculated as κ svs = max k σ k min k σ k (19)where σ k ( k =1, 2, · · · , K) are the singular values, and κ svs isSVS. F. Channel Capacity
Channel capacity is the maximum rate in channel wherethe bit error rate tends to zero. There are M T and M R antennas at Tx and Rx sides, respectively, and the Tx doesnot know the channel state information. If we choose signalcovariance matrix as identity matrix I M T , which means thesignals are independent and equi-powered at the transmitantennas, channel capacity can be represented as [17] C = log (cid:20) det( I M R + ρM T HH H ) (cid:21) (20)where det [ · ] defines the determinant, ( · ) H defines the conju-gate transpose operation, I M R defines the identity matrix ofsize M R , and ρ defines the signal-to-noise ratio (SNR). IV. R ESULTS AND A NALYSIS
The statistical properties and performance indicators of themodel are simulated and analyzed in this section. LSPs withspatial consistency are generated through the SoS method. Asshown in Fig. 2, it is the LSP of delay spread in an area of 300m ×
300 m and its parameters are set to 300 sine waves withthe ACF modeled as a compound function of Gaussian andexponential decay. It can be seen obviously that the continuousspatial variation of delay spread factor is realized.
A. The Temporal ACF
Fig. 3 illustrates the temporal ACF. Fig. 3 (a) represents theACF changing with different velocities at Rx side. When thevelocity at Rx side becomes larger, the coherence time willbecome shorter. The coherence time refers to the time differ-ence when the ACF equals to a given threshold, which canbe determined by system requirements. Fig. 3 (b) representsthe ACF changing with different carrier frequencies. Whenthe carrier frequency becomes larger, the coherence time willbecome shorter. The reasons for above phenomenon is that thelarger velocity and carrier frequency lead to larger Dopplershift. Larger Doppler shift makes the channel more fluctuantand uncorrelated.
B. The Space CCF
Fig. 4 illustrates the space CCF. The measurement wasconducted in a campus environment at 2.6 GHz carrier fre-quency with 128 antenna elements ULA at BS side [18]. Thesimulation result is consistent with the measurement, whichvalidated the presented model.
C. The FCF
The FCF is shown in Fig. 5. The channel with different clus-ter azimuth spread values 3, 5, and 7 has different coherencebandwidths 680 MHz, 475 MHz, and 260 MHz, respectively.What should be noted that is the coherence bandwidth refersto the frequency separation when the FCF equals to 0.5. Theabove phenomenon indicates that the larger cluster azimuthspreads will reduce the correlation of the channel.
D. STF Cluster Evolution
Cluster evolutions in STF domain are shown in Fig. 6 (a),Fig. 6 (b), and Fig. 6 (c). Different antennas at the sametime instant and frequency, or the same antenna at differenttime instants and frequencies will see different clusters, whichindicates the channel is non-stationary in STF domain. R qp,q (cid:48) p (cid:48) ( t, f ; (cid:52) t, (cid:52) f, δ T , δ R ) = E (cid:2) H qp ( t, f ) H ∗ q (cid:48) p (cid:48) ( t + (cid:52) t, f + (cid:52) f ) (cid:3) = (cid:115) K RF ( t ) K RF ( t ) + 1 · K RF ( t + (cid:52) t ) K RF ( t + (cid:52) t ) + 1 · R Lqp,q (cid:48) p (cid:48) ( t, f ; (cid:52) t, (cid:52) f, δ T , δ R ) + (cid:115) K RF ( t ) + 1 · K RF ( t + (cid:52) t ) + 1 R Nqp,q (cid:48) p (cid:48) ( t, f ; (cid:52) t, (cid:52) f, δ T , δ R ) (14) ig. 2. Delay spread with spatial consistency in a 2D area. E. The SVS
Fig. 7 illustrates the cumulative distribution functions(CDFs) of SVSs of simulation results and measurement in[19]. The channel measurement was performed in indoorscenario at 1.4725 GHz with a virtual 128-element ULA. Thesimulation results agree with the measurement data. When M T gradually increases to 128, the SVS gradually decreases tobelow 1 dB. The above phenomenon manifests that the channelbecomes more and more stable, and the channel vectors amongusers become approximately orthogonal. F. Channel Capacity
The uplink sum-rates in the measured channels and thepresented model are compared in Fig. 8 [20]. The simulationresults are consistent with the measurement data. With thenumber of antennas increasing at BS side, the uplink sum-rates also increase within a certain SNR range.V. C
ONCLUSIONS
The paper has proposed a general 3D massive MIMOGBSM for 6G communication systems. The presented model (a) Normalized time difference, t (s) T e m po r a l A C F Analytical, v R =5 m/sSimulation, v R =5 m/sAnalytical, v R =15 m/sSimulation, v R =15 m/s (b) Normalized time difference, t (s) T e m po r a l A C F Analytical, f c =4.3 GHzSimulation, f c =4.3 GHzAnalytical, f c =5.3 GHzSimulation, f c =5.3 GHzAnalytical, f c =6.3 GHzSimulation, f c =6.3 GHz Fig. 3. (a) Temporal ACF with different velocities at Rx side (b) TemporalACF with different carrier frequencies ( β TA = π /10, β RA = π /12, α TA = π /10, α RA = π /12, λ G =20/m, λ R =1/m, D Ac =40 m). Antenna index A b s o l u t e v a l ue o f s pa c e CC F AnalyticalSimulationMeasurement
Fig. 4. Absolute value of space CCF at different antenna spacings at theBS side ( f c =2.6 GHz, M T =128, M R =1, β TA = π /10, β RA = π /12, α TA = π /10, α RA = π /12, δ T = λ/ , λ G =20/m, λ R =1/m, D Sc =40 m, NLOS). A b s o l u t e v a l ue o f l o c a l F C F AS =3 AS =5 AS =7 Fig. 5. Absolute value of FCF with different cluster azimuth spreads( f c =28 GHz, β TA = π /10, β RA = π /12, α TA = π /10, α RA = π /12, δ T = λ/ , λ G =20/m, λ R =1/m). can support arbitrary velocities and trajectories at both Tx sideand Rx side, which are equipped with URAs. Meanwhile, ithas studied cluster evolution in STF domain to support STFnon-stationary communication scenario. In addition, the spatialconsistency of LSPs generation has been proved by using themethod of parameter generation with spatial consistency. Thesimulations about temporal ACF, space CCF, FCF, SVS, andchannel capacity are conducted. Analytical, simulation results,and measurements have also been compared to verify thevalidity of the channel model. The novel GBSM proposed inthis paper will play a significant part in the development of6G communication systems.A CKNOWLEDGMENT
This work was supported by the National Key R&D Programof China under Grant 2018YFB1801101, the National Natural Sci-ence Foundation of China (NSFC) under Grant 61960206006 andGrant 61901109, the Frontiers Science Center for Mobile Infor-mation Communication and Security, the High Level Innovationand Entrepreneurial Research Team Program in Jiangsu, the HighLevel Innovation and Entrepreneurial Talent Introduction Programin Jiangsu, the Research Fund of National Mobile Communications ig. 6. (a) Cluster evolution in space domain ( f c =5.3 GHz, M T =128, δ T = λ/ , D Ac =40 m) (b) Cluster evolution in time domain ( f c =5.3 GHz, v R =15 m/s, v T =0 m/s, D Sc =40 m) (c) Cluster evolution in frequency domain( f c =38 GHz, v R =15 m/s, v T =0 m/s). The SVS[dB] CD F M T =4, M R =2, SimulationM T =4, M R =2, MeasurementM T =32, M R =2, SimulationM T =32, M R =2, MeasurementM T =128, M R =2, SimulationM T =128, M R =2, Measurement Fig. 7. CDFs of SVSs with different numbers of antennas at Tx side( f c =1.4725 GHz, β TA = π /3, β RA = π /4, α TA = π /3, α RA = π /3, δ T = λ/ , λ G =20/m, λ R =1/m, LOS). Research Laboratory, Southeast University, under Grant 2020B01,the Fundamental Research Funds for the Central Universities underGrant 2242019R30001, the Huawei Cooperation Project, and theEU H2020 RISE TESTBED2 project under Grant 872172, theNational Postdoctoral Program for Innovative Talents under GrantBX20180062. R EFERENCES[1] X.-H. You, C.-X. Wang, J. Huang, et al., “Towards 6G wirelesscommunication networks: Vision, enabling technologies, and newparadigm shifts,”
Sci. China Inf. Sci. , vol. 64, no. 1, Jan. 2021, doi:10.1007/s11432-020-2955-6.[2] C.-X. Wang, J. Huang, H. Wang, et al., ”6G Wireless Channel Mea-surements and Models: Trends and Challenges,” in
IEEE Veh. Technol.Mag. , vol. 15, no. 4, pp. 22-32, Dec. 2020.[3] H. Q. Ngo, E. G. Larsson, and T. L. Marzetta, “Aspects of favorablepropagation in massive MIMO,” in
Proc. EUSIPCO , Lisbon, Sept. 2014,pp. 76–80.[4] J. Chen, “When does asymptotic orthogonality exist for very largearrays?” in
Proc. IEEE GLOBECOM , Atlanta, GA, Dec. 2013, pp. 4146–4150.[5] Y. Lu, C. Tao, L. Liu, K. Liu, “Spatial characteristics of the massiveMIMO channel based on indoor measurement at 1.4725 GHz,”
IETCommunications , vol. 12, no. 2, pp. 192–197, Nov. 2017. -10 -5 0 5 10 (a) SNR[dB] S u m r a t e [ bp s / H z ] N T =5, N R =5Measurement -10 -5 0 5 10 (b) SNR[dB] S u m r a t e [ bp s / H z ] N T =5, N R =32Measurement -10 -5 0 5 10 (c) SNR[dB] S u m r a t e [ bp s / H z ] N T =5, N R =64Measurement -10 -5 0 5 10 (d) SNR[dB] S u m r a t e [ bp s / H z ] N T =5, N R =128Measurement Fig. 8. The uplink sum-rates with different numbers of antennas at Rx side( f c =2.6 GHz, β TA = π /10, β RA = π /12, α TA = π /10, α RA = π /12, δ T = λ/ , λ G =20/m, λ R =1/m).[6] J. Huang, C.-X. Wang, R. Feng, et al., “Multi-frequency mmWavemassive MIMO channel measurements and characterization for 5Gwireless communication systems,” IEEE J. Sel. Areas Commun. , vol.35, no. 7, pp. 1591–1605, July 2017.[7] J. Li and Y. Zhao, “Channel characterization and modeling for large-scale antenna systems,” in
Proc. ISCIT , Incheon, Sept. 2014, pp. 559–563.[8] C. F. Lopez, C.-X. Wang, and R. Feng, “A novel 2D non-stationarywideband massive MIMO channel model,” in
Proc. IEEE CAMAD ,Toronto, ON, Oct. 2016, pp. 207–212.[9] C. F. L´opez and C.-X. Wang, “Novel 3-D non-stationary widebandmodels for massive MIMO channels,”
IEEE Trans. Wireless Commun. ,vol. 17, no. 5, pp. 2893–2905, May 2018.[10] S. Wu, C.-X. Wang, e. M. Aggoune, M. M. Alwakeel and Y. He, “Anon-stationary 3-D wideband twin-cluster model for 5G massive MIMOchannels,”
IEEE J. Sel. Areas Commun. , vol. 32, no. 6, pp. 1207–1218,June 2014.[11] S. Wu, C.-X. Wang, e. M. Aggoune, et al., “A general 3-D non-stationary5G wireless channel model,”
IEEE Trans. Commun. , vol. 66, no. 7, pp.3065–3078, July 2018.[12] J. Bian, C.-X. Wang, X. Gao, X.-H. You, and M. Zhang, “A general3D non-stationary wireless channel model for 5G and beyond,”
IEEETrans. Wireless Commun. , accepted for publication.[13] H. Wu, S. Jin, and X. Gao, “Non-stationary multi-ring channel modelfor massive MIMO systems,” in
Proc. WCSP , Nanjing, Oct. 2015, pp.1–6.[14] S. Wu, C.-X. Wang, H. Haas, et al., “A non-stationary widebandchannel model for massive MIMO communication systems,”
IEEETrans. Wireless Commun. , vol. 14, no. 3, pp. 1434–1446, Mar. 2015.[15] L. Wang, J. Chen, X. Yang, N. Ma and P. Zhang, “A 3D ellipsoid modelfor isotropic and non-isotropic scatterers co-existing massive MIMOchannels,” in
Proc. IEEE WCNCW , Marrakech, Morocco, Apr. 2019,pp. 1–6.[16] Y. Liu, C.-X. Wang, J. Huang, J. Sun and W. Zhang, “Novel 3-Dnonstationary mmWave massive MIMO channel models for 5G high-speed train wireless communications,”
IEEE Trans. Veh. Technol. , vol.68, no. 3, pp. 2077–2086, Mar. 2019.[17] A. Paulraj, R. Nabar, and D. Gore,
Introduction to Space–Time WirelessCommunications.
Cambridge, U.K.: Cambridge Univ. Press, 2003.[18] S. Payami and F. Tufvesson, “Channel measurements and analysis forvery large array systems at 2.6 GHz,” in
Proc. EUCAP , Prague, Mar.2012, pp. 433–437.[19] Q. Li, L. Liu, C. Tao, et al., “Performance analysis of a massive MIMOsystem in indoor scenario,” in
Proc. ISAPE , Hangzhou, China, Dec.2018, pp. 1–4.[20] X. Gao, M. Zhu, F. Rusek, F. Tufvesson, and O. Edfors, “Large antennaarray and propagation environment interaction,” in