An Interim Channel Model for Beyond-3G Systems Extending the 3GPP Spatial Channel Model (SCM)

Daniel S. Baum and Jan Hansen Jari Salo ETH Zürich, Zürich, Switzerland Helsinki University of Technology, Espoo, Finland {dsbaum,hansen}@nari.ee.ethz.ch [email protected]

Giovanni Del Galdo and Marko Milojevic Pekka Kyösti Ilmenau University of Technology, Ilmenau, Germany Elektrobit Ltd., Oulu, Finland {giovanni.delgaldo,marko.milojevic}@tu-ilmenau.de [email protected]

Abstract— This paper reports on the interim beyond-3G (B3G) While the WINNER project only started recently, there is channel model developed by and used within the European an immediate demand for models suitable for initial usage. WINNER project. The model is a comprehensive spatial channel This document presents the result of our studies in form of a model for 2 and 5 GHz frequency bands and supports band- model that is used for initial evaluation of B3G technologies in widths up to 100 MHz in three different outdoor environments. It outdoor scenarios within the WINNER project. further features time-evolution of system-level parameters for challenging advanced communication algorithms, as well as a Contributions. Our specific contributions are as follows: reduced-variability tapped delay-line model for improved usabili- • ty in calibration and comparison simulations. We analyze shortcomings of a selected spatial channel model standard with respect to the identified require- Keywords- channel model, beyond-3G, MIMO, SCM, 3GPP ments from other WINNER Work Packages. • We evaluated results found from literature search and I. INTRODUCTION derived from our own measurement data to devise In recent years Multiple-Input Multiple-Output (MIMO) missing parameters. wireless communication techniques have attracted strong • We propose a set of backward compatible extension to attention in research and development due to their potential the 3GPP Spatial Channel Model (SCM). benefits in spectral efficiency, throughput and quality of service. Only recently, however, has this technology been This paper summarizes the results reported in [4]. considered to be included in wireless communication system standards, such as IEEE 802.11n for wireless LANs (WLAN), II. 3GPP SCM IEEE 802.16 for broadband fixed wireless access (FWA), and 3GPP high-speed downlink packet access (HSDPA) for cellu- We have identified two publications ([1], [2]) defining lar mobile communications. spatial / MIMO radio channel models that are commonly accepted and used. Other publications focus mainly on aspects Any wireless communication system needs to specify a and certain effects of the radio channel. As the 802.11n model propagation channel model that can act as a basis for perfor- is targeted towards indoor applications, we have selected the mance evaluation and comparison. With advancing com- 3GPP SCM as a basis for outdoor channel model extensions. munication technologies, these models need to be refined as further characteristics of the channel can be exploited and thus A. Properties need to be modeled. To enable MIMO, the standardization groups 802.11 and 3GPP thus first defined spatial channel The SCM is a so-called geometric or ray-based model models suitable for their applications [1], [2]. based on stochastic modeling of scatterers. It defines three environments (Suburban Macro, Urban Macro, and Urban Upcoming communication systems will be based on a new Micro) where Urban Micro is differentiated in line-of-sight set of system parameters (e.g. extended bandwidth and new (LOS) and non-LOS (NLOS) propagation. There is a fixed frequency bands), a broader range of and additional scenarios number of 6 “paths” in every scenario, each representing a (e.g. mobile to mobile, mobile hotspot), and new communica- Dirac function in delay domain, but made up of 20 spatially tion techniques (e.g., tracking algorithms). This triggers new separated “sub-paths” according to the sum-of-sinusoids requirements on the underlying channel models. method [5]. Path powers, path delays, and angular properties for both sides of the link are modeled as random variables The European WINNER project [3], which is part of the defined through probability density functions (PDFs) and Framework 6 effort, is currently researching the outline of a cross-correlations. All parameters, except for fast-fading, are system design of such a B3G system. In WINNER, it is the drawn independently in time, in what is termed “drops”. goal of Work Package 5 to come up with channel models that suit the needs in the project.

This work has been performed in the framework of the IST project IST- 2003-507581 WINNER, which is partly funded by the European Union. The authors would like to acknowledge the contributions of their colleagues.

0-7803-8887-9/05/$20.00 (c)2005 IEEE TABLE 1. MIDPATH POWER-DELAY PARAMETERS values, and avoids that single sub-paths become delay- resolvable. Furthermore, lumping together a number of sub- Scenario Suburban Macro, Urban Micro Urban Macro paths keeps the fading distribution of that tap close to Rayleigh and thus aids a potential implementation with a classic No. mid-paths per path 3 4 Gaussian-distributed number generator. We found that 4 is the Mid-path power and 1 10/20 0 ns 6/20 0 ns absolute minimum number of sinusoids to yield a reasonable delay relative to 2 6/20 7 ns 6/20 5.8 ns paths Rayleigh distribution. 3 4/20 26.5 ns 4/20 13.5 ns The number of mid-paths, and the power and delay 4 - - 4/20 27.6 ns parameters chosen for each mid-path are tabulated in Table 1. The mid-path powers, i.e. number of sub-paths, were chosen by B. Shortcomings considering the decreasing power with delay while staying The SCM was defined for a 5 MHz bandwidth CDMA above the minimum number of sub-paths. The delays for the system in the 2 GHz band, whereas the currently defined mid-paths were then derived by employing the method from WINNER system parameters are 100 MHz bandwidth in both 2 [9] with the DS set to the predetermined value of 10 ns and the and 5 GHz frequency range [6]. Other issues are the drop based predetermined set of powers given for the mid-paths. concept, i.e., no short-term system-level time-variability in the In SCM, each sub-path has an angle relative to the path model, the lack of Ricean K-factor models (LOS support) for mean angle assigned to it. By perturbing the set of sub-paths macro scenarios, and the lack of a wider range of scenarios. assigned to a mid-path, the AS of that mid-path can be varied. It has been reported, e.g. [10], that the intra-cluster AS III. INTERIM BEYOND-3G CHANNEL MODEL conditioned on the intra-cluster delay is approximately Our main goal for the extension was to keep it simple, independent of the delay. Hence, the mid-path ASs (ASi, where backward-compatible, and within the conceptual approach of i is the mid-path index) were optimized such that the deviation the SCM. This approach provides consistency and compara- from the path AS (ASn, where n is the path index), i.e. the AS bility. In the following we discuss the underlying concepts and of all mid-paths combined, is minimized. The result is the reasoning behind the proposed extensions. tabulated in Table 2.

A. Bandwidth B. Frequency Range To extend the model in a way such that its characteristics 1) Path-Loss Model remain unchanged if compared at the original 5 MHz resolu- The SCM path-loss model is based on the COST-Hata- tion bandwidth, we add intra-path delay-spread (DS), which is Model [11] for Suburban and Urban Macro and the COST- zero in the SCM. A possible power-delay profile (PDP) is a Walfish-Ikegami-Model (COST-WI) [11] for Urban Micro. one-sided exponential function. This approach of so-called Some relevant references on path-loss were found ([12]-[18]), intra-cluster DS was originally proposed by Saleh and however only few of them allow direct comparison between Valenzuela for indoor propagation modeling [7]. The intra- equivalent measurements at 2 and 5 GHz. These few however cluster DS model has also been adopted for outdoor scenarios indicate that the most significant difference can be attributed to in the COST 259 [8] model. Following the SCM philosophy, different gains in free-space path-loss, which is 8 dB higher at which is partly based on COST 259, we use this as our guide- 5 GHz compared to 2 GHz. Thus, for comparability reasons, line. The path DS was chosen under the following considera- we propose a 5 GHz path-loss model that has an offset of 8 dB tions 0 • 3 mid-paths In SCM, all paths within a scenario have the same path -5 4 mid-paths azimuth-spread (AS). Equivalently, we set the path DS exponential PDP to be constant. The path AS and DS then define the -10 minimum observable total (over all paths) AS and DS. -15 • Both from measurements and intuition it follows that this minimum total spread lies somewhere between -20 zero and a fraction of the mean total spread. -25 • The error in power between an exponential PDP and the SCM definition (no DS) is illustrated in Figure 1. -30 For a path DS of 10 ns, this error is slightly below -20 dB and can be considered reasonably small. We set it -35 equivalent to this value for all paths. power belowerror original signal in dB -40 We split the 20 sub-paths into subsets, denoted “mid- -45 paths”, which we then move to different delays relative to the 0 1 original path. Even though a mid-path consists of multiple sub- 10 10 paths, it remains a single tap (delay-resolvable component). delay-spread in ns This approach limits the diversity increase to reasonable Figure 1. Relative power of channel impulse response difference when path- DS is added, compared at 5 MHz bandwidth TABLE 2. SUB-PATHS TO MID-PATHS ASSIGNMENT AND RESULTING shadowing and make no differentiation for frequency range. NORMALIZED MID-PATH ANGLE-SPREADS

Mid- 3 mid-path configuration 4 mid-path configuration C. Other Extensions path Pwr Sub-paths ASi / Pwr Sub- ASi / 1) LOS for All Scenarios ASn paths ASn In the SCM, the LOS model, consisting of path-loss and 1 10/20 1, 2, 3, 4, 5, 0.9865 6/20 1, 2, 3, 1.2471 Ricean K-factor definition, is a switch selectable option for 6, 7, 8, 19, 20 4, 19, 20 Urban Micro only. We extend the K-factor option to cover also 2 6/20 9, 10, 11, 12, 1.0056 6/20 5, 6, 7, 0.9145 Urban and Suburban Macro scenarios as follows. Urban and 17, 18 8, 17, 18 Suburban Macro are assigned the same parameters. The 3 4/20 13, 14, 15, 16 1.0247 4/20 9, 10, 0.8891 probability of having LOS is calculated as [21] 15, 16 4 - - - 4/20 11, 12, 0.7887 PLOS = (1 - hB/hBS)(1 - d/dco), dco < 300, hBS > hB 13, 14 and is zero otherwise. Here, hBS is the BS height, hB the average to the current 2 GHz model. height of the rooftops, and dco is the cut-off distance. Values for these parameters are proposed in [8]. There are some issues here though. The COST-231-Hata- Model was derived for the purpose of GSM coverage We use the empirical K-factor model presented in [22] for prediction and has a distance range of 1-20 km. The 5 GHz typical (American) suburban environments and BS heights of band on the other hand is likely going to be used for short- approximately 20 m. In [23], an excellent agreement with this range high-throughput services. In this case, a path-loss based model was reported based on independent measurements under on the COST-WI model with a distance range of 0.02-5 km is similar conditions. We propose the following parameters: MS much more suitable. Note that this model has also been antenna height 1.5 m, MS beam-width: 360°, and selection of accepted by the ITU-R and was selected as Urban / Alternative season: summer. The resulting model is

Flat Suburban path-loss model in the IEEE 802.16 standard for K = 15.4 - 5.0 log10(d), fixed wireless access [19]. Furthermore, the model distin- guishes between LOS and NLOS situations. In conclusion, we where d is the BS-MS distance in m, and K is in dB. propose to use the COST-WI model as an alternative path-loss 2) Time-Evolution model for all scenarios with the following parameters: base The literature on dynamic / non-stationary channel models, station (BS) antenna height: Macro – 32 m, Micro – 10 m; that is, channel models with time-varying channel parameters, building height: Urban – 12 m, Suburban – 9 m; building to is relatively scarce. Initial references of dynamic channel building distance: 50 m, street width: 25 m, mobile station models appear to be [24]-[26]. Dynamic channel models for (MS) antenna height: 1.5 m, orientation: 30° for all paths, and indoor environments are developed in [27]-[29], of which the selection of: Macro – medium sized city / suburban centres, first reference focuses on dynamic delay-domain characteriza- Micro – metropolitan. The results are summarized in Table 3. tion and the latter two also incorporate spatial dynamics of the 2) Delay-Spread, Angle-Spread and Ricean K-factor indoor channel. The standard [30] defines a simple model for Preliminary measurement analysis and literature findings varying tap delays and tap birth-death. However, this model is [20] indicate that DS and AS statistics do not significantly intended for receiver testing and does not represent a realistic deviate with doubling of the channel frequency and we thus channel. leave both parameter definitions unchanged in a first approxi- The concept of drops in SCM can be seen as relatively mation. Similarly, we propose using the 2 GHz K-factor for 5 short channel observation periods that are significantly separat- GHz range. We apply the same argument in the case of ed from each other in time or space such that the channel TABLE 3. PATH-LOSS MODEL parameters become constant and independent during these periods. Our approach is to virtually extend the lengths of these Scenario Suburban Urban Urban periods by adding short-term time-variability of some channel Macro Macro Micro parameters within the drops. All channel parameters remain 31.5 + 34.5 + 34.53 + SCM path- NLOS independent between drops. The three effects we model are 35.0 log (d) 35.0 log (d) 38.0 log (d) loss (dB), 10 10 10 discussed in the following. 30.18 + d is in m LOS - - 26.0 log10(d) a) Drifting of Path Delays and Angles SCM shad. NLOS 8 8 10 The paths, sub-paths, and spatial sampling instants (within std. dev. (dB) LOS - - 4 a drop) are indexed by n, m, and k, respectively. We assume 7.17 + 11.14 + 31.81 + that the positions of scatterers are fixed during a drop. As a Alternative. NLOS 38.0 log (d) 38.0 log (d) 40.5 log (d) short-range 10 10 10 consequence, the scatter angles as seen from the BS (angles-of- 30.18 + 30.18 + 30.18 + departure, AoDs) do not change, with the exception of the LOS path-loss (dB) LOS 26.0 log10(d) 26.0 log10(d) 26.0 log10(d) AoD in LOS scenarios. This assumption is valid in many cases Alt. shad. std. NLOS 10 10 10 of practical interest. Based on the fixed-geometry assumption, Dev. (dB) LOS 4 4 4 the scatter angles as seen from the MS (angles-of-arrival, θ 5 vs. 2 GHz AoAs, nm,AoA,k) as well as the sub-path delays change during a (N)LOS + 8 dB + 8 dB + 8 dB path-loss drop due to the MS movement. Similarly, the LOS directions from BS and MS (θBS,k and θMS,k, respectively) vary in time. In TABLE 4. DISTRIBUTION PARAMETERS FOR dn,0 = dmin + X AND the following equations λ is the wavelength, c is the speed of CORRELATION DISTANCES FOR SHADOW FADING light, v is the velocity of the MS, θ is the direction of MS v d Parameters of log(X) 50% movement, D is the sample density per half wavelength, d Scenario min correlation S MS-BS (m) Mean var. point (m) is the MS-BS distance (LOS path length), and dnm is the distance between MS and the last-bounce scatter (LBS) of the ττ− 22+ n 1 Suburban Macro 10 ττ− 1 200 mth sub-path of the nth path. All angles are defined with N 1 respect to the normal of the antenna broadside with positive ττ− 22+ n 1 angles in counter-clockwise direction. For illustration of the Urban Macro 10 ττ− 1 50 geometry, see Fig. 5.2 of [2]. The update equations for the AoA N 1 Urban Micro 10 1 5 / AoD of the LOS sub-path (when present) are θMS,k+1 = θv – γk 2 for dMS-BS,k+1 < dMS-BS,k, θMS,k+1 = θv – 180° + γk otherwise, and θ θ θ θ a headache for accurate simulations as the simulation time BS,k+1 = BS,k – ( MS,k+1 – MS,k), with grows exponentially with the number of random parameters. ddldl=+−222cos()ϕ , As a practical add-on, we have thus defined a set of fixed MSBSk−+,1 MSBSk − , MSBSk − , MSBSk − , values for the power, delays, and angular parameters of the ϕ λ paths tabulated in Table 5. This is similar to the SCM link-level dMS−− BS,, ksin( MS BS k ) , , γ = arcsin l = model. However, while the latter targets 3GPP comparability, k 2D dMS−+ BS,1 k S our solution is close to the SCM system-level model and furthermore optimized for small frequency autocorrelation. and ϕMS-BS,k = θv – θMS,k. The update equations for the AoAs of θ θ ξ the scattered (non-LOS) sub-paths are nm,AoA,k+1 = v – k for The parameters were derived as follows. The fixed delays dnm,k+1 < dnm,k, θnm,AoA, k+1 = θv – 180° + ξk otherwise, where of the 6 paths were fitted to the PDP of the SCM system-level model using the method from [9]. These delays were then d sin(ϕ ) ξ = arcsin nm,, k nm k , perturbed until a satisfactory frequency decorrelation was k  dnm,1 k+ achieved. Mean angles were randomized until the total ASs roughly equalled the expected values for AS. =+−22 ϕ , ddldlnm,1 k+ nm , k2cos() nm , k nm , k IV. IMPLEMENTATION and ϕnm, k = θv – θnm,AoA,k. The sub-path delays are updated ac- 1 cording to τnm,k+1 – τnm,k = cos(Φk) l/c, where Φk = ϕMS-BS,k for The original SCM has been implemented in MATLAB and the LOS sub-path, and Φk = ϕnm,k for other sub-paths. is available [38] under a public license. Please check the referenced website for updated information about any Initial values for k=0 are generated according to [2]. For extensions of the implementation. calculating the AoA drift, the initial distance between MS and LBS is required. This distance is unknown since SCM is not a single-bounce geometrical model and hence cannot simply be REFERENCES inferred from the geometry. Instead, we propose a simple [1] V. Erceg, L. Schumacher, P. Kyritsi, A. Molisch, D. S. Baum, et al., “TGn channel models”, IEEE 802.11-03/940r2, Jan. 2004. stochastic model where the initial distance, dn,0, to all LBSs of the nth path is a random variable, independent for all n=1...6. [2] 3GPP, “Spatial channel model for MIMO simulations”, TR 25.996 V6.1.0, Sep. 2003. [Online]. Available: http://www.3gpp.org/ As a plausible PDF for dn,0, we select a log-normal distribution with a constant, small offset and parameters given in Table 4. [3] 6th Framework Programme, Information Society Technologies, Wireless World Initiative New Radio (WINNER), IST-2003-507591, [online] Available: https://www.ist-winner.org/ In general, the angle Φk is a nonlinear function in time. For l much smaller than the MS-LBS (or MS-BS) distance, a [4] D. S. Baum, G. Del Galdo, J. Salo, P. Kyösti, T. Rautiainen, M. linearization of this function yields a good approximation for Milojevic, and J. Hansen, “SCM extensions,” WINNER WP5.5 – Internal, Jan. 2005. observation periods of several meters. One can then use a [5] M. F. Pop, and N. C. Beaulieu, “Limitations of sum-of-sinusoids fading constant change of AoA, AoD and delay over the drop. This channel simulators,” IEEE Trans. Commun., vol. 49, no. 4, Apr. 2001, yields significant savings in computation while resulting in a pp. 699–708. usable model for simulation of dynamic MIMO channels. [6] J. von Häfen, U. Schwark, G. Mange, M. Litzenburger, D. C. Schultz, et al., “System Requirements,“ WINNER D7.1 – Public, Jul. 2004. b) Drifting of Shadow Fading [7] A. Saleh, and R. A. Valenzuela, “A statistical model for indoor multi- The time-evolution of shadow fading is determined by its path propagation,” IEEE J. Select. Areas Commun., vol. SAC-5, no. 2, spatial autocorrelation function. References show that an expo- Feb. 1987, pp. 128–137. nential function fits well and the drifting can thus be modeled [8] L. M. Correia, Wireless Flexible Personalized Communications, COST by a first order autoregressive process. Derived from publica- 259: European Cooperation in Mobile Radio Research, John Wiley & Sons, 2001. Section 3.2 by M. Steinbauer and A. F. Molisch, tions on measurement data around 2 GHz ([31]-[37]), we “Directional channel models”. propose using correlation distances (50% correlation point) [9] J. Kivinen, X. Zhao, and P. Vainikainen, “Empirical characterization of listed in Table 4. wideband indoor radio channel at 5.3 GHz,” IEEE Trans. Ant. Prop., Aug. 2001, pp. 1192–1203. 3) Tapped Delay-Line Model In the SCM, most parameters are defined by their PDFs. While this provides richness in variability, it can turn out to be 1 MATLAB is a registered trademark of The MathWorks, Inc. TABLE 5. TAPPED DELAY-LINE PARAMETERS

Scenario Suburban Macro Urban Macro Urban Micro Power-delay parameters: 1 0 0 0 0 0 0 relative path power (dB) / 2 -2.6682 0.1408 -2.2204 0.3600 -1.2661 0.2840 delay (µs) 3 -6.2147 0.0626 -1.7184 0.2527 -2.7201 0.2047 4 -10.4132 0.4015 -5.1896 1.0387 -4.2973 0.6623 5 -16.4735 1.3820 -9.0516 2.7300 -6.0140 0.8066 6 -22.1898 2.8280 -12.5013 4.5977 -8.4306 0.9227 Resulting total DS (µs) 0.231 0.841 0.294 Path AS at BS, MS (deg) 2, 35 2, 35 5, 35 Angular parameters: 1 156.1507 -101.3376 65.7489 81.9720 76.4750 -127.2788 0.6966 6.6100 AoA (deg) / 2 -137.2020 -100.8629 45.6454 80.5354 -11.8704 -129.9678 -13.2268 14.1360 AoD (deg) 3 39.3383 -110.9587 143.1863 79.6210 -14.5707 -136.8071 146.0669 50.8297 4 115.1626 -112.9888 32.5131 98.6319 17.7089 -96.2155 -30.5485 38.3972 5 91.1897 -115.5088 -91.0551 102.1308 167.6567 -159.5999 -11.4412 6.6690 6 4.6769 -118.0681 -19.1657 107.0643 139.0774 173.1860 -1.0587 40.2849 Resulting total AS at BS,MS (deg) 4.70, 64.78 7.87, 62.35 15.76, 62.19 18.21, 67.80

[10] K. I. Pedersen, P. E. Mogensen, and B. H. Fleury, “A stochastic model GHz for fixed BWA applications,” in Proc. IEEE ICC’02, vol. 1, May of the temporal and azimuthal dispersion seen at the base station in 2002, pp. 272–276. outdoor propagation environments,” IEEE Trans. Veh. Technol., vol. 49, [24] S. J. Papantoniou, “Modelling the mobile-radio channel,” Ph.D. no. 2, Mar. 2000, pp. 437-447. dissertation, no. 9120, ETH Zürich, 1990. [11] COST 231, “Urban transmission loss models for mobile radio in the [25] R. Heddergott, U. Bernhard, and B. Fleury, “Stochastic radio channel 900- and 1800 MHz bands,” TD(90)119 Rev. 2, Sep. 1991. model for advanced indoor mobile communications systems,” in IEEE [12] A. J. Rustako, Jr., V. Erceg, R. S. Roman, T. M. Willis, and J. Ling, Proc. PIMRC ’97, Helsinki, Finland, 1997, pp. 140–144. “Measurements of microcellular propagation loss at 6 GHz and 2 GHz [26] B. Fleury and U. B. R. Heddergott, “Advanced radio channel model for over NLOS paths in the city of Boston,” in Proc. IEEE Magic WAND,” in Proc. ACTS Mobile Summit, Granada, Spain, 1996. GLOBECOM’95, vol. 1, Nov. 1995, pp. 758–763. [27] J. Ø. Nielsen, V. Afanassiev, and J. B. Andersen, “A dynamic model of [13] A. Karlsson, R. E. Schuh, C. Bergljung, P. Karlsson, and N. Lowendahl, the indoor channel,” Wirel. Pers. Commun., vol. 19, 2001, pp. 91–120. “The influence of trees on radio channels at frequencies of 3 and 5 GHz,” in Proc. IEEE VTC’01, vol. 4, Oct. 2001, pp. 2008–2012. [28] T. Zwick, C. Fischer, and W. Wiesbeck, “A stochastic multipath channel model including path directions for indoor environments,” IEEE J. [14] N. Kita, A. Sato, and M. Umehira, “A path loss model with height Select. Areas Commun., vol. 20, no. 6, Aug. 2002, pp. 1178–1192. variation in residential area based on experimental and theoretical studies using a 5G/2G dual band antenna,” in Proc. IEEE VTC’00, vol. [29] C.-C. Chong, D. I. Laurenson, and S. McLaughlin, “The implementation 2, Sep. 2000, pp. 840–844. and evaluation of a novel wideband dynamic directional indoor channel model based on a Markov process,” in Proc. IEEE PIMRC’03, vol. 1, [15] X. Zhao, J. Kivinen, P. Vainikainen, and K. Skog, “Propagation 2003, pp. 670–674. characteristics for wideband outdoor mobile communications at 5.3 GHz,” IEEE J. Select. Areas Commun., vol. 20, Apr. 2002, pp. 507–514. [30] 3GPP, “Base Station (BS) radio transmission and reception (FDD),” TS 25.104 V6.7.0, Sep. 2004. [Online]. Available: http://www.3gpp.org/ [16] K. Yonezawa, T. Maeyama, H. Iwai, and H. Harada, “Path loss measurement in 5 GHz macro cellular systems and consideration of [31] M. Marsan, G. Hess, and S. Gilbert, “Shadowing variability in an urban extending existing path loss prediction methods,” in Proc. IEEE land mobile environment at 900 MHz,” El. Lett., vol. 26, no. 10, May WCNC’04, vol. 1, Mar. 2004, pp. 279–283. 1990, pp. 646–648. [32] E. Perahia, D. Cox, and S. Ho, “Shadow fading cross-correlation [17] C.-C. Chong, C.-M. Tan, D. I. Laurenson, S. McLaughlin, M. A. Beach, between base stations,” in Proc. IEEE Veh. Tech. Conf., vol. 1, May and A. R. Nix, “A new statistical wideband spatio-temporal channel 2001, pp. 313–317. model for 5 GHz band WLAN systems,” IEEE J. Select. Areas Commun., vol. 21, no. 2, Feb. 2003, pp. 139–150. [33] A. Algans, K. Pedersen, and P. Mogensen, “Experimental analysis of joint statistical properties of azimuth spread, delay spread, and shadow [18] X. Zhao, T. Rautiainen, K. Kalliola, and P. Vainikainen, “Path loss fading,” IEEE J. Select. Areas Commun., vol. 20, no. 3, Apr. 2002, pp. models for urban microcells at 5.3 GHz,” COST 273 TD(04)207, 523–531. Duisburg, Germany, Sep. 2004. [19] V. Erceg, K. V. S. Hari, M. S. Smith, D. S. Baum, et al., “Channel [34] M. Gudmundson, “Correlation model for shadow fading in mobile radio systems,” El. Lett., vol. 27, no. 23, Nov. 1991, pp. 2145–2146. models for fixed wireless applications,” IEEE Broadband Wireless Access Working Group, Tech. Rep., IEEE 802.16.3c-01/29r4, Jul. 2001. [35] J. Weitzen and T. J. Lowe, “Measurement of angular and distance [20] J. A. Wepman, J. R. Hoffman, et al., “Impulse response measurements correlation properties of log-normal shadowing at 1900 MHz and its in the 902-928 and 1850-1990 MHz bands in macrocellular environ- application to design of PCS systems,” IEEE Trans. Veh. Technol., vol. ments,” in Proc. IEEE ICUPC’93, vol. 2, Oct. 1993, pp. 590 – 594. 51, no. 2, Mar. 2002, pp. 265–273. [21] H. Asplund and J.-E. Berg, “Parameter distributions for the COST 259 [36] T. Sørensen, “Correlation model for slow fading in a small urban macro directional channel model,” COST 259 TD(99)108, Sep. 1999. cell,” in Proc. IEEE PIMRC, vol. 3, Sep. 1998, pp. 1161–1165. [22] L. J. Greenstein, S. Ghassemzadeh, V. Erceg, and D. G. Michelson, [37] J. Berg, R. Bownds, and F. Lotse, “Path loss and fading models for “Ricean K-factors in narrowband fixed wireless channels,” in Proc. microcells at 900 MHz,” in Proc. IEEE VTC’92, vol. 2, May 1992, pp. WPMC’99, Amsterdam, The Netherlands, Sep. 1999. 666–671. [38] J. Salo, G. Del Galdo, J. Salmi, P. Kyösti, M. Milojevic, D. Laselva, and [23] P. Soma, D. S. Baum, V. Erceg, et al., “Analysis and modeling of C. Schneider. (2005, Jan.) MATLAB implementation of the 3GPP MIMO radio channel based on outdoor measurements conducted at 2.5 Spatial Channel Model (3GPP TR 25.996). [Online]. Available: http://www.tkk.fi/Units/Radio/scm/