This action might not be possible to undo. Are you sure you want to continue?

# A Generic Space-Time-Frequency Correlation Model and Its Corresponding Simulation Model for Narrowband MIMO Channels

**Xiang Cheng*, Cheng-Xiang Wang*, and David I. Laurensont
**

*Joint Research Institute in Signal and Image Processing, School ofEngineering & Physical Sciences, Heriot- Watt University, Edinburgh, EH14 4AS, UK E-mail: txc48, cheng-xiang.wang}(d,hw.ac.uk tJoint Research Institute in Signal and Image Processing, School ofEngineering and Electronics, University ofEdinburgh, Edinburgh, EH9 3JL, UK E-mail: Dave.Laurenson d9ed. ac. uk

Keyword: Wireless channels, NIIMO, STF correlation, were not well understood. In [19], only 2-D time-frequency (TF) correlation reference and simulation models were deterministic simulation model. studied for frequency correlated single-input single-output (SISO) channels under isotropic scattering assumptions. The Abstract authors in [13] investigated space, time, and frequency For the analysis and design of Multiple-Input Multiple-Output correlation properties separately of MIMO channels based on (MIMO) wireless communication systems with frequency the elliptical geometry of scatterers for microcellular environments. diversity features, e.g., MIMO-Orthogonal Frequency Division However, in [13], no one generic STF correlation function (CF) Multiplexing (MIMO-OFDM) systems, it is often desirable to was given. Moreover, the integral expressions ofthe derived space develop a channel model that can characterise the three- and time CFs can only be numerically evaluated as no closeddimensional (3-D) space-time-frequency (STF) correlation form expressions were found. Rad and Gazor proposed nonproperties over the links of the underlying MIMO channels. geometric 3-D STF correlation models for MIMO outdoor In this paper, we propose a generic 3-D STF correlation channels [8,15,16], where the angle of arrival (AoA) and angle model, which includes many well-known existing models as of departure (AoD) were assumed to be independent. special cases, with closed-form expressions of the STF In this paper, we first derive a generic theoretical reference correlation properties. Based on the developed theoretical model in order to study the 3-D STF correlation properties reference model, a deterministic simulation model is then between the impulse responses of two sub-channels with proposed and its 3-D STF correlation properties are also different carrier frequencies in a narrowband MIMO channel. investigated by providing closed-form expressions. It is Different from [13], the proposed reference model is based on shown that the correlation properties of the simulation model the well-known geometrical one-ring scattering model [1,6], fit those of the reference model very well when the which has widely been used for modelling MIMO channels in parameters of the simulation model are determined by using macrocelluar environments due to its simplicity, and has a the Lp-norm method (LPNM). closed-form expression of the generic STF CF. In contrast to non-geometric models [8,15,16], the proposed model characterises the AoA using the von Mises angular probability 1 Introduction density function (PDF) [2], which is applicable to both isotropic MIMO systems have recently received much attention because and non-isotropic scattering environments, and considers the of their potential for achieving higher data rate and providing interdependence between the AoA and AoD. More importantly, more reliable reception performance compared with traditional we will demonstrate that the derived generic closed-form single-antenna systems for wireless communications. In order expression is valid not only for the 3-D STF CF, but also its to theoretically analyse and design high performance MIMO degenerate 2-D and 1-D CFs, e.g., ST CF and frequency CF. wireless systems under various circumstances, it is of great This means that all the CFs have a uniform expression but importance to have proper theoretical reference models [20] for with different parameters. The derived new 3-D STF the underlying MIMO wireless channels. Furthermore, for the correlation model is a generalization of many existing models practical simulation and performance evaluation of MIMO [1,2,6,7,10,11]. systems, it is advantageous to develop accurate MIMO channel Due to its infinite complexity, the proposed narrowband onesimulation models [14]. Nowadays, 3-D STF correlation models ring STF MIMO reference model cannot be realized directly are required to comprehensively understand the behaviour of in software or hardware. Therefore, the corresponding simulation MNMO wireless channels with frequency diversity features, e.g., in model is very important in practice for the performance MIMO-OFDM systems [4]. evaluation of MIMO wireless communication systems. The Most existing models, e.g., [1,2,6,14], were proposed to second part of this paper uses the reference model as the investigate 2-D space-time (ST) correlation properties of starting point for the derivation of an efficient simulation narrowband MIMO wireless channels, but the frequency model by taking into account all the 3-D STF correlation correlation properties of two sub-channels in a MIMO channel

Authorized licensed use limited to: National University of Singapore. Downloaded on July 16,2010 at 10:00:30 UTC from IEEE Xplore. Restrictions apply.

Rm ) scattered by the nth scatterer. hi. It is usually assumed that R is much smaller than D. .(t) (hitmq(t)) h2.n = ('Cqn + Cnm )/c. 9R}..1p (t) + jh2. the sub-channel complex impulse responses at two different carrier frequencies fc and fcJ can be expressed as: z 4.e. Sn and c is the speed of light. This allows us to assess the performance of the simulation model analytically by comparing its correlation properties with those of the developed generic reference model. 1. . IOR and k ( k 0 ) is a real-valued parameter that controls the angular spread of the AOA O)R For k = 0 (isotropic scattering). we with 1. fD is the maximum Doppler frequency. 1. respectively. Finally. while for k>O (non-isotropic scattering). The one-ring model is appropriate for describing scattering environments where the transmitter base station (BS) is elevated and unobstructed. Restrictions apply. It is shown that the designed MIMO channel simulator matches the underlying reference model very well with respect to temporal. and N is the number of independent scatterers. hmq(t) = htmq(t)+ j11 mq(t) =lim From (1). can restrict our investigations to the following STF CF v . m = 1.) random variables with uniform distributions over [0. In Section 4.2 ).2. 5nR such as the uniform [17]. wrapped Gaussian [18]. The MS moves with a speed v in the direction determined by the angle of motion y. (3) where O5R iS the continuous denotation of ¢5nR when N is infinite.y)(. many different scatterer distributions have been proposed to characterise the AoA. the von Mises PDF [2] is used. D + Rcos(4R ). Sn around the MS. Without a line-of-sight component. Some simulation results are presented and the performance of the resulting simulation model is evaluated in Section 5.properties of MIMO channels.2ff). TIp. it was shown that this PDF fits the real data very well. The proposed procedure is based on the principle of deterministic channel modelling [12].) is the zeroth-order modified Bessel function of the first kind. +2igDt co4. Gaussian [3].RI ( Tq .2.2zT) accounts for the mean value of the AoA.. conclusions are drawn in Section 6. The local scatterers are located on a ring surrounding the MS with radius R. I hlp (t) = 14. it follows that the correlation properties of hi (t) and h tmq (t) are completely determined by the underlying real Gaussian noise processes -=exp4j[LV. we have the following approximate equations wave . Without loss of generality. ( Tmq ) is the travel time of the vmhqjphT (4) Authorized licensed use limited to: National University of Singapore. are independent and identically distributed (i. q = 1. f (^R )_= exp[kcos(eR _'U~)] l2kcIo (k) )I OR E [0. and the cardioid PDFs [5]. ../.. respectively. fpn n - 2 2 [cos(a)+Asin(a)sin(n )1 A (2a) (2b) (2c) fqn Sn + 2 [cos(a)+ sin(a)sin(OnR)1 2 jnl/ Z-R RcoS (on O8). while .2010 at 10:00:30 UTC from IEEE Xplore. The antenna element spacings at the BS and MS are designated by Sr and JR. D>>R>>maAr. (2e) The phases V. In the literature.. The BS and MS have nBS and nMS omni-directional antenna elements in the horizontal plane. whereas the receiver mobile station (MS) is surrounded by a large number of local scatterers...d.S. ..n = (fpn + -nl)/c and Tmq.'p.n nl dn and jnm are the distances as functions of ¢.. .nMs. 2-D. The angle spread seen at the BS is denoted by Al. . Closed-form expressions will be provided for all the 3-D.. (2d) 2 The one-ring narrowband MIMO model A one-ring narrowband MIMO channel model was first proposed in [6] and further developed in [1].lp (t) = lim 3 The new generic STF CF - 1/ >exp[jVn --24fz. and frequency properties. Each scatterer is assumed to be reflected only once. v = 1. respectively. .nR as shown in Fig. where rl. nBS. p.i. E [0.2zT). Inm z R +-RcOs(R 2 -g ). The von Mises PDF is defined as [2] .. Therefore. and 1-D CFs of the simulation model. an efficient deterministic simulation model is proposed and its corresponding 3-D STF CF is derived as a closed-form expression.-2gC rnq. The multi-element antenna tilt angles are denoted by a and .mq(t) ( u. Furthermore. which can approximate all the above mentioned distributions. the von Mises PDF reduces to the uniform distribution. In [2]. Io(.. Downloaded on July 16... The paper is structured as follows. The one-ring narrowband MIMO channel model is introduced in Section 2 and the new generic closed-form 3-D STF CF is derived in Section 3. spatial. we consider uniform linear antenna arrays with nBS = nMS = 2 (a 2 x 2 MIMO channel). According to [1]. i.Sn . denoting the distance between the BS and MS.. the von Mises PDF approximates different distributions depending on different values of k [2].(t) and h. which is related to R and D by A arctan(R/D) R/D.n +2fzDtcos(0 y)](. The AoA of the wave travelling from the nth scatterer towards the MS is denoted by OnR. In this paper. through the link T. (t) ( himq(t)). it is assumed that both R and D are much larger than the antenna element spacings Sr and JR.1(t) ( h mq(t) ) components of the complex impulse response hi. Let us consider a one-ring narrowband MIMO channel model shown in Fig. .mq h1.

g proposed. (t) (SR/2)cos(.2zXM. environments. For a multiple-input single-output (MISO) B = 2k[x cos(y -. (8d) 4 The deterministic simulation model J = 2 + (ST/4)A sin2 a+. X = fLi .2yzAsin a sin. the 3-D STF CFs between h1. -y). B. 1-D space CF phi}ph2..8 .u) -ycos(. x = 2fDZ. we can + 2xy cos(8 . From (7b)./ Authorized licensed use limited to: National University of Singapore. . 1-D space respectively. ST and SR.+2gfDt cosQpf It is worth stressing here that (5) and (6) are the generic expressions which apply to the 3-D STF CF and the = (t)2+ jh2(mq (t) fl+fq subsequently presented 2-D and 1-D CFs differ only in values hm&fq (t) (lOb) of A. .5R24 (8e) In this section. + (ST/2)A sina sinf.2yzAsin asinf8 (9a) symbol definitions are the same as in (1).y cos(. .lq (z) -Ph2 jphi lq (iz) 0 (SIMO case) and 1 -D time CF p([) ph2iphip(i) 0 in isotropic Ph2 iphT (C' X) Phl ihT (C. This means that in isotropic scattering only the derivation of (5a) is given in the Appendix. y=2JfCSR/c. (t) case).u) -zA sin a sinu]. are given by B = 2k[xcos(y -. It should be observed that (4) is a function of the time separation.mq (9b) (9c) = h Ph2lphT (r X)=Oif B = O and C = O . (lOa) = (Sr/2)cosa+D+R.iph2. 0.1Pkmq =0 and 1-D time CF and htq (t) can be directly obtained as Ah.2yXL . B. .Iph2. (8h) and hitq (t) ofthe deterministic simulation model are modelled as S = R cos. and 1-D time CF.u) ./ -. case. For example. When setting the frequency separation (k) {eJo [A+B)1!2] ec ( B)112 (5b) X=O (X=O) and the tilt angle of the BS c=r/2 in (7c).6= ° + 2xXK . RIc. (8f) utilizing only a finite number of scatterers.. the time CF given in [2] is obtained by setting iS = SR = 0 and X = 0 in (6) with k . the ST CF in [6] is obtained by C =zcosa+XT. N. we find that the STF CF Phl 2.zA sin a sin. the impulse responses L = R cosg8 + (ST/2)A sin a sing8 +(SR/2). the ST CF shown in [1] is obtained by (8b) setting X = 0 in (6) with k . and similarly between h11 . Restrictions apply. Hence. 0 . the degenerate 2-D and 1-D CFs can be easily obtained from (7) by setting relevant terms to zero.2ZT). the space CF given in [7] is obtained. (t)+ jh2 1.u). we can get C=O for the 2-D ST CF (single-input Phihf (i'X) = Ph2hJ 7-h2 ) ( multiple-output (SIMO) case) and 1-D time CF. If where by further setting k = 0 (isotropic scattering) in (6).i1 1(t) and h. the Clarke's A = k2 _x2 _ y2 _ z22 sin2 a . 1. n + 2)rDt cos(0nR -7)lJ. + (8i) h/. space separation. M = (ST/2)A2 sin2 a + (SR/2)A sin a sin f.hT2. CF.2 j = = - I h40(t) 1 )]J ¢5. where E[.mq (t + T).] denotes the statistical average with respect to )R and V. which is obtained from the reference model by K = R cos y+ (ST/2)A sin a sin y+ (SR/2)cos(. B. an efficient deterministic simulation model is + RSR cos8 + (STSRI2)A sin ausing. substituting SR = 0 .mq scattering environments. This means that B = 0 in isotropic scattering environments. and C for the 2-D ST CF: will be kept constant during simulation. Therefore. Since the derivations of (5a) and (5b) are similar. and the other A = k2 _x2 _ y2 _z2tA sin2 a . we can =j 1k fie jIo(A+jB) /]-eicIo[(A-Bl2 also obtain C = 0 for the 2-D ST CF (MIMO case). -. the Lee's ST CF in + 2xy cos(.f. (8a) For a MIMO case.1Pk.X) 2. C = zcosa.that B = 0 if k = 0 holds.mq(t+T).lp Oc) p2.l (t) and hU.2)fz tmq.lP Oc) ° Phip hTmq (T X)) l. Consequently. This clearly 4I1 {ek o [(A jB)12]+ e-cIo [(A + jB)12 (5a) indicates that the 2-D ST CF phi. it is clear ph.mq ((6) X) 2PhliophT The proposed generic 3-D STF correlation model with a c (6) closed-form expression (6) includes many existing models as = Ik eJIo [A-jB)112 special cases. -. . By setting the space separation at the BS T = 0 (z-O) and the frequency separation X = 0 (X=O) in (7c). and k = 0 into (6).XS]. setting the where the phases Vln are simply the outcomes of a random frequency separation X = 0 (X=O) in (7) gives the following generator uniformly distributed over [0.8 time CF in [10] is obtained. and frequency separation. (8c) X = 2zxf/c. y = 2)g-. If further with substituting fD = 0 into (6). k = 0 into (6). The corresponding expressions of A. T (8j) s exp{j wn . and C.2fzpf. Substituting (1)-(3) into (4) and after some mathematical manipulations.y)+ 2xzAsin a sin y-_X2J and [11] is obtained by substituting iST =0 =0 .u) . and C for = i> xp{j [!-n . the 2-D ST CF Phuph2 mq(/) ph2]phi mq(/) 0 (MIMO the 3-D STF CF between the complex impulse responses hi. For a SIMO case. For a SISO case. Downloaded on July 16.mq l.2010 at 10:00:30 UTC from IEEE Xplore.mq = P2.y)+ 2xzAsin a sin y. From (5b). X = 0 . and keeping all (8g) the model parameters fixed. (t) = h1.mq (T jBX)/21Phl. the discrete AoAs expressions ofA.

. Figs. sin(C)sin(P cos 0 R + Q sin 0 R)] Phi iph 2.xcosy. Fig. we will apply the LPNM [12] to compute the model parameters {qn=R of the deterministic simulation }o model based on corresponding properties of the reference model. For PhiphJ (X) and hphmq. 1 -D frequency CF pl hT (X) and 2-D space CF phiphmq are identified as key {ton}n of the simulation model. 2 and 3 illustrate the 2-D space-frequency (SF) CFs against the frequency separation and space separation at the BS and MS.p (r) . AI. Fig. (1 7c) 2N. N =30. N = 30. Fig. X) j2~~~~i. A plot of the function phphip (r) of the reference model is shown in Fig. 5-7 clearly demonstrate that the proposed deterministic Authorized licensed use limited to: National University of Singapore. . Similar to (4). R=100 m. As shown in Figs. 7 depicts the 2-D space CF phiphmq of the reference model and phiphmq of the simulation model. respectively. EJCeJ(PCOS0R+QS1n4H) (13) (14a) (14b) (15a) (1 5b) Note that x. Similarly to (5) and (6).8 Z = (d/2)A sin a + (dR/2)sin/. X) (Zi Ph2. at the BS is generally small for the macrocellular case. 5.analyse the properties of the deterministic channel simulator by time averages instead of statistical averages. a=ff/6. Comparing them.mnq ('Z. the 2-D CFs take the maximum values when the frequency separation X=0 and decrease with the increase of the frequency separation. c=3x108 m/s. Note that Tmax z and yTmax and 3RmaX define the upper limits of the ranges over which the approximations pA. with N = 30. y.%) (12b) replace 0. iphTv.hhphip (i) dr} (17a) (1 7b) 1/p 12 hlhf(/- h hf dX} Ph2. The basic parameters are as follows: fC=I GHz.mq X with N n=1 where C is the same as in (6). E(p) :=i r 1 f max max hphmq -Phiphmqd&TdR d . it is obvious that only needs to be determined for this deterministic simulation model.mq X) hi Phiphmq ph. andXmax = 8 MHz. 2-D. 2 2Phi iphT (r. 5Tm = 30k. Figs. 4. we can further get the 3-D STF CF between hkp (t) and (t) as Ph1pht mq (r. Substituting (10) into (11).mq N hk.) denotes the time average operator. The corresponding expressions of P and Q for the degenerate 2-D and 1-D CFs can easily be obtained from (14) by setting some relevant terms to zero. l. k=3 . 2-4 reveal that the shapes of CFs depend on time separation. and Tmax = 0..mq are of interest. and will evaluate the performance of the simulation model. (12) and (13) are the generic expressions which apply to all the 3-D. The discrete AoAs 0) R have hiph properties.2. the three error norms E(P). 6 depicts the 1-D frequency CF PhphT (x) of the reference model and Phiph (%) (-r. g =fT/3. + zA sin a .. and y= 7fT/12 . PpT (&) ~ZiphT (X) and Xmax .~ 2Phi iphT 2. 2-4. we will only focus on the frequency correlation properties based on (6). and 1-D CFs of the deterministic simulation model with different C. Comparing (14) with (7). The 1 -D time CF phiphlq (r) . we can relate A and B to P and Q by A = a2 +b2 _(P2+Q2 B = -2(aP + bQ). and X are the same as defined in (8a)-(8d) above.1phTl. it is shown that (11) can be expressed in the closed-form as Phl (q mh * E(p) ='{ max f Phlphip (J. it is obvious that the normalized antenna space at the MS or BS has a large effect on the 2-D SF CFs. since the angular spread. while P = XY + ycos.x siny. P. while its influence on the 2-D TF CF is negligibly small..p (r) ph. 2-4. respectively. and dR) = 3/ . Y = R+ (5R/2)cos. and Q. Q = XZ + y sin/. E(p) and E(P) can be minimized independently. D=1200 m. This figure also shows the resulting 1-D time CF phiphip (') of the simulation model designed with the LPNM using p = 2. Comparing (a) with (b) in Figs. if we 2--E [sin(C)cos(P cos tnR + Q sin nR) + cos(C) sin(Pcos Gn + Q sin (n By analogy with (6). frequency separation. .mq (r. - [coCs (C) cos (P cos QR + Q sin ¢R) (12a) where p = 1. Downloaded on July 16.2010 at 10:00:30 UTC from IEEE Xplore. when applying the LPNM with p = 2. Then the LPNM requires the numerical been obtained by applying the LPNM on the error minimization of the following three Lp -norms: norm E"(P) in (17c) with p = 2 .u=z. due to the limitation of space.R by ¢>R and ¢> R.. fD=93 Hz.mhq. the 3-D STF CF can be defined as Ph.1 Parameter computation method In this subsection. . and antenna spacing. we find that the influence of the normalized antenna space at the MS is greater than at the BS.X) Khu ip (t)hvtmq (t + )(11) where K. (16a) (16b) From (12) and (13). z. Restrictions apply. Figs. 4 shows the 2D TF CF along with the frequency separation and time separation.q 5 Simulation results In this section.08 s.

M. are obtained by replacing ¢ [10] W.z. and M. [15] H. (1994)." Nco[2T. Valencia. Table of Integrals. Selected Areas in Commun. In this paper. hl.. pR JpOR f:op/. and S. Lee. and space correlation properties. Fulghum and K. Jeffrey. Commun." Proc. frequency. "A space-time model for frequency Pl. Ottawa. . Feeney. After some manipulations and considering (16). 9. (. [11] W. "Mobile Fading Channels". no..Phimq 4." IEEE cos[-2. Byers.937-1. Select. S. "A wideband MIMO x cos [C + P cos + Q sin OR ]OR channel model derived from the geometric elliptical scattering model. pp. Zoltowski. and F.lp (t) and h1t mq (t + r) according to (4): [6] T. and B.simulation model can fit the underlying reference model very References well in terms of time." IEEE Trans. The proposed 3-D STF reference model [3] F. Downloaded on July 16. volume 133. G. The definite integral in the right hand side of the above 138-143. Hogstad. Chen. San Diego. based on the well-known narrowband one-ring [2] A. Chichester: John Wiley & Sons. and J." IEE Proc. (2000). pp. equation can be solved by using [9. M. no. D. Williamson.. by O5R in (2). (2004). Adachi. C. Veh. e. Rad. 3. reference model. 1175-1190.. S. "Space-time-frequency characterization of MIMO wireless channels. H. 727-737. volume 4. Signal Processing. Veh. Technol.. volume 18. Grimm. "Space-time-frequency characterization of 3d non-isotropic MIMO multicarrier propagation channels employing directional antennas. S. Radar. Substituting all the above mentioned terms and New Jersey: IEEE Press. volume 18. pp. pp. pp. and J. (1986). volume 53. Patzold." IEEE Journal on match those of the underlying theoretical model very closely. W. R R Authorized licensed use limited to: National University of Singapore.g. Areas (AR 2-Ty)(t ) + N-x>-Nn E{cos[ Commun. Y.[14] M. Canada." IEEE J. Gradshteyn.cos O + k sing sin [13] M.. . F. Gazor. D. and M. 3. 0.papR + .n + 2. Restrictions apply. Spain." IEEE Trans. A. pp.. "A space-time channel simulator for MIMO channels based on the geometrical form expression for the 3-D STF CF PhilphT (Zr. 5th ed. Jakes Ed. (1994). communication systems with frequency diversity features. L. 0. rk 2. Technol. ~~~~~~~1 Technol. 3. (2004). 506-512. [1] A. (2006). 3. T." IEEE Trans.R/ )/C Tmq = (dq:pR + fRm )/C * The terms CA: Academic. )of lPhmq analysis. Takawira. Paulraj. 550-560. The developed reference model the mobile station. hl. "Spatially and temporally Derivation of (5a): correlated MIMO channels: modeling and capacity In this appendix.) is given by (5a). volume 5." i = lim E ( (18) Proc.. Abdi. and JOR. Abdi.. M. no. Patzold. Appendix [5] G.n + 2.TD cos(¢ -r)]j to space-time modems. Bolcskei. and B. volume 21. eq. where rp = Series. and Products. IEEE VTC. "A parametric model for the distribution of the angle of arrival and the MIMO channel model. 634-643.2010 at 10:00:30 UTC from IEEE Xplore. Fitz. 489-493. Gazor. 417-426. (2003). 425-434. Rad.7Tmq. "Cross-correlation between the envelopes of is general enough to include many well-known existing models 900 MHz signals received at a mobile radio base station as special cases. Barger. volume 20. the closed. + 2zD -R [9] I. we have proposed a novel generic 3-D associated correlation function and power spectrum at STF correlation reference model.T cos( [8] S." Wireless Communications on Mobile Computing. and A. and I.zIo (k) [12] M. M. Numerical results [4] H. (2002). no. Wireless Commun. Hogstad. A. Ryzhik. Parsons. Areas Commun. 3. - cos(R _)r](R )dR . pp. Kaveh.Tf (mq -Tip )-2. Ed. 427439. Microwave Mobile Communications. Kuo.. pp. The corresponding 3-D STF CFs of the simulation model are derived with the closed-form expressions. Select. P. M. (2002). pp. MIMO-OFDM systems. 23692375. "Impact of the propagation environment on the performance of show that the correlation properties of the simulation model space-frequency code MIMO-OFDM. "The Jakes fading model incorporating angular spread for a disk of scatterers. C.ZTmq =Trans. pp. ISWCS'06.. . 6 Conclusions no. A. ON. Molnar. Borgmann. a deterministic simulation model is then proposed. pp. one-ring scattering model.2TfeTnq n+ 2JTfD coso'~ Y)(t + r)l m [7] T. Patzold. we derive the 3-D STF CF phjph1 (T . (2002). Depending on the developed theoretical site. J. 522]. we have predetection diversity combiner. x co s[Vfn . "Level crossing rates of an equal-gain PDF into (18) and considering (14). J. (2006). A. "A space-time correlation model for multielement antenna systems in mobile fading channels. Commun. Kaveh. and H." IEEE J. 2N N-->n=cl (1998). x ] (19) exp [kcos . volume is suitable for the analysis and design of MIMO wireless 51.mqt(X=Eh.l()lm(+) nonselective Rayleigh fading channels with applications = lim -E1cos Vn Ip. pp. (1970).

Phi () (simulation model) hiph with N=30. model. H.5Ol Oj 0 u) 0 2. pp.1 ' 0 2 Frequency separation.[16] [17] [18] [19] [20] IEEE Trans. Wang. (1994). Veh. The 2-D SF CF Phiph. X (s) )o -( I -1 0. S.25 3 0 15 &11X (wave length) Authorized licensed use limited to: National University of Singapore. 5. Signals. Royal Institute of Technology (KTH). no. .0. Technol. 587-592. accepted with minor revision. 1049-1057. Pedersen. IEEE PIMRC. Restrictions apply. pp. X. 3.5 (MIMO). 6. The 2-D SF CF Phh. The 2-D space CF versus the normalized antenna antenna spacing at the BS iT/ = 0 (SIMO) and (b) the normalized spacing at the BS id/. 1. (2007).20 . Mogensen." doctoral thesis.2 0.5 - 40 10 0 40 x ZVN 10 20 X (ms) Fig. 2. N=30) ACF (ref.X 2 o ~Cu 1- *~0. Downloaded on July 16.06 0.. volume 43. IEEE WCNC 2007.5 -05 5 M 2.~0. M.ph." IEEE Trans.. "Stochastic modelling and simulation of frequency correlated wideband fading channels." Proc. (a) (b) the normalized antenna spacing at the MS iR/ (a) the normalized Fig. . c c . model) 1 0S)8 t ACF (sim. . (2005).5 10 0 0 D75 L5232 2. Patzold.. oe 8 z 02 0 0 0. Winters. K. L.02 Time separation. Wireless Commun. (a) mq (Z) versus the frequency separation X and (b) the normalized antenna spacing at the BS dT/IA: (a) the normalized antenna spacing at the MS idR/iA = 0 (MISO) and (b) the normalized antenna spacing atthe MS idR/l = 0. China. H. . "From antenna spacings to theoretical capacities guidelines for simulating MIMO systems.5 u) o en en O 0 5 ' 5 O 7. C. mq (X) versus the frequency separation X and (b) 0 075_ 30 2.5 .6 . Schumacher. Gazor.25 6R/X (wave length) *.5 5 10 0 (MHz) ~15 6T/X (wave length) 22." IEEE Trans. X (MHz) 4 6 8 10 Fig.5u) 0 0. Rad. 4.t . volume 56.I and at the MS dR/I (a) phiphmq (reference model) and (b) phiphmq (simulation model) with antenna spacing at the BS d5/Z = 5 (MIMO).0 1 Fig. (2007).52 1.0.p (r) Fig.04 0. Technol. "Multiple-input multiple-output radio propagation channels: characteristics and models. The 1-D time CF Ph. 0. "Effect of fading correlation on adaptive arrays in digital mobile radio. and Q. N=30) 06 0. and S.15 0. Yao. model) 08 ACF (sim. 18901895.08 0V0. (2002). 7. Hong Kong. The 2-D TF CFp (Pph (a) (MHz) 7. N=30.. 1 ll ACF (ref.75 (a) Fig. pp.5 (MIMO). Salz. E." Proc. Yu. J. pp. K. Sensors and Systems. volume 2. and P.2010 at 10:00:30 UTC from IEEE Xplore. 3. "Space-time-frequency characterization of 3D non-isotropic MIMO multicarrier propagation channels employing directional antennas.5x (MHz) 7. 1050-1063.25 3 0 07 30 8R/X (wave length) 75 15 &11X (wave length) 8R/X (wave length) 2.50 u) % 7.5 - (MHz) 6RIX (wave length) x (MHz) ).5 30 6O x (MHz) 75 10 0 751 1 25 30 611X (wave length) Fig. model.5 10 0 ) versus the time separation r and (b) X (ms) 20 20 frequency separation X : (a) the normalized antenna spacing at the BS ir/IA = 0 and at the MS idR/Il = 0 (SISO) and (b) the normalized antennaspacingattheBS r/TZ = 5 andattheMS d6R/) = 0. Veh. Geometrical configuration of a 2 x 2 narrowband onering channel model with local scatters around the mobile user. (2007). I. and J. The 1-D frequency CF (reference model) and Phiphip (T) phhhhT (y) (reference model) and hiph (simulation model) with N=30.