An Advanced 3D Ray Launching Method for Wireless Propagation Prediction Greg Durgin, Neal Patwari, Theodore 8. Rappaport Mobile and Portable Radio Research Group at Virginia Polytechnic Institute and State University Bradley Department of Electrical and Computer Engineering Blacksburg, VA 24061-0350 (540)231-2967 ‘Abstract - Foe radio propagation prediction, recent simmula- tions invotving ray tracing offer unprecedented aceuracy (1), [11], (13}, {14}, These techniques surpass statistical chan- net models and provide a bounty of additional information including RMS delay spread, angle of strival, and overall ‘wideband channel impulse response. In partieular, three dimensional ray tracing produces an accurate, determinis- tic channel model for wireloss system design. This paper presents a new 3D ray tracing technique of unprecedented pete and accurac lation results are earpated to tmierocell measurements at 1900 MEL. 1. Iwrropucrio ‘Three main areas of error in ray tracing simulation are propagation modeling errors, database errors, and kine= ‘matic errars. Propagation modeling errors come from. the geometrical optics models used to describe radio wave be= wiot. Database errors stem from the limitations of a ite, numerical deseription of the world, The ability of y tracing algorithm to find and interpret radiation paths determines the kinematic errors of a simulation. This pa- per facuses on method to launch and interpret rays that model the waveftonts of radio propagation; we present. a new methor: that eliminates many of the kinematic errors associated with a ray launching scheme. ‘This paper concentrates on three major ray launching concepis. It first establishes methods for geodesic ray launching, originally proposed by Seidel [13 The following section reviews the principles and inconsistencies ‘of interpreting ray information. using the reception sphere model found! in the literature [11]. Finally, the paper intro- duces 8 new method for Interpreting ray information using a special weighting function to construct ray-trared waver fronts. The distributed wavefront model has several im- ‘portant advantages over the reception sphere models used in previous literature (Le. [13], [15)). This paper provides fundamental principles to construct an extremely acewrat three dimensional ray launcher for propagation prediction, ‘unlike any previously proposed, II, PROPAGATION AND DaTAWASE MopgtiNa ‘Am environment database for 3D rey Isunching consists of iat polygons that represent the surfaces of bulldings and ‘Thie work is sponsored by the National Science Foundation Fras- fcental Foculty Fulomship ander Grant Xo, NCR.O2S3808 an the MPIC: Tneustlal Aint Progean 0-7803-3650-2/97 $10.00 ©1997 IEEE Fax: (540)241-2968 —http://www.mprg.ce.vt.edu terrain. Flat palygons are easy to describe numerically and allow simple reflection calculations. The well-defined edges also lend themselves to diffraction calculations [6]. A paly- gon database may be constructed cosily using computer aided design (CAD) software. Figure 1 shows an exazmple of 8 ray-troeed environment in AutoCAD" (5). Fig. 1. Am exarple of rays traced through an urban enviroament (Htowlyn, VA), A key issue for implementing widesprend, sitespecific propagation tools is the formal specification of ensiron- meat databases, These databases must. contain all of the information relevant to © propagation simulation such as geometry, material properties, and surfore characteristics, For this reason, current research is working to develop the Real Wortd Database Format (RWDF), a standard method for constructing a site-specific propagation database, LIL, Ray Tracixe Tecrsiques An alternative to ray launching that avoids the geomet cal complexities of a ray launching algorithm is tho method of smoges, ‘The method of images uses image theary ta place artificial sources in the envizangnent that model re fleetlons from the fla: planos of database [3], [7]. Because Image theory determines mcact radiation paths, the method of images introduces no errors inta the radiation paths i finds. However, this method anly works for reflected modes of propagation, since diffraction introduces infinite degxees of freedom in the ditection af a vay path. The primary drasrback to image theory algorithms is their heavy dependence on the number of elements i 788‘the environment database. For single reflection, each additional surface in the database can double the num- her of images available to a receiver. Therefore, image theory should not be used to render complicated, three- dimensional cityscapes, which require multiple reflections from hundreds of surfaces, Image theory is best suited for simplified, two-dimensional city blocks. [A versatile ray tracing technique, and the one explored. in the following sections, launches rays from a. transmit ter and reflects them through the environment. Flr this method, the launched rays that pass arbitrarily close to a, receiver establish the actual rediation paths, Although this paper concentrates on reflection, ray launching. techniques are extendible to diffraction, The principle advantage of this method is its ability ¢0 quickly render a complicated BD scene since the algorithm has anly a linear dependence (on the number of datalase elements. The overall process ing time can be reduced farther by incorporating: spatial discrimination, such as a bounding volume Mierarchy (2). IV. Gropesic Ray Lauxcaive Jn a ray launching scheme, rays emannte from a unit sphere ceutered on the transmitter loration. Launch palate around this sphere follow a ragular, computer generated geometry, There are two desirable characteristics of ray Tannett geometcy 1» Large-Scale Uniformity « The launch points must dis- trite evenly around the sphere so that all regions of space are illuminated equally by rays. Large-scale uniformity delivers unbiased ray coverage in three di- 1 Small-Seade Uniformity - The lacel pestern of rays ine pinging on a wavelroat should be a predictable, uni- form pattern. This corresponds ta equal angles be- tween a ray and its neighbors, which assists the inter- pretation of wavefront information. Ray shooting from the vertices of regular polyhodrons isthe only way to exactly satisfy the two uniformity criteria (15), Sinoe no regular polyhedron lias more than twenty vertices, high precision ray tracing; must use other geometties [12], A. Geodesic Geometry ‘The geodesic sphere arises by tessllating the faces of 2 regular polyhedron and extrapolating tse intersection points to the surface af « sphere [4]. Figure 2 depicts the geodesic facets and vertices that result when the sides of an. icasabedron are subdivided into smaller equllateral trian. ales, The geodesic vertices provide ray launch points with, ‘aquivalent angular separation around the entire sphere {13} ‘Maredives, each ray will have exactly six neighbor rays that surround the original ib a predictable hexagonal ‘The receiver schemes presented in this paper expl regularity when calculating power levels, B, Geodesic: Aberrations ‘Unifortnity allows for simple interpretation of a ray Jaunch, ‘The reception sphere model takes advantage of Fig. 2, An icosahedron (a etenelfate to produce a geodesic sptere (8). Rags lunch from the vertices of thin struct. the uniformity of a geodesic launcher to collect: rays us ing a simple circular arca. Because the geodesic sphere approsimates uniformity, there are some discrepancies in angular separation among the launched rays. For instance, each ray emanating from ane of the twelve vertiers of the icosahedron has only five neighboring rays rather than the usual six (see Figure 3}. This type of aberration becomes, insignificant for heavily tessellated spheres (22). f Res ay RRROooaS Sane . Coe Fig. 9. Examples of fee-ncighbor aberzatians on a geodmic wphece with semellation frequency N= 8. ‘Tlasallation freuen the umber of trlangies alo te ecge of Scowabede Another aberration occurs in the angular separation. be- tween rays, There is an average discrepancy of twenty per- eit between the smallest and largest angular separations on a sphere (15). Angular separation plays an important part in aay tracing calculation. It determines the distance to the closest ray on the same wavelroat « vital informa- tion for interpreting simulation results, Regardless of the teatellation frequency; there will be a twenty percent varin- tion between the minimum and maximum separation on a geodesic sphere. High tessellotion frequency, however, docs mitigate the variations in localized angular separation [12] Character the angulae separation for each ray chould ‘use the maximum separation angle with respect to its neighbors, rather than a single separation value for all geodesic rays. Despite aberrotions, an average value for angular sepa ration 18 & useful parameter in a sithblation, ‘The angular resolution of a geodesic sphere is a function of the tessel- lation frequency NN, the number of triangles on the edge of an icosahedron face. Assuming that the 204? total sodesic facets are congruent equilateral triangles, Equa- 7868tion (1) finds the average radial separation by equating, their total area with the 47 steradians of a unit sphere. or ~ N 0) Tin (1) Se eo sic to tage eeide tag store Mates a nears aap Serres Steaks oasis uaa V, THE RECEPTION SPHERE (Onee raye are traced chrough a scene, a programm must. interpret the results by measuring voltage or power levels at arbitrary points in space. A common method to in terpeet the traced ray Information fs the reception sphere ruodel. This method assures uaiform ray launching from fn geodesic sphere in an environment composed of flat sur- faces, such as buildings or planar segments of terrain. The simple essumptions and easy implementation of the recep: tian sphere model make it a useful ray tracing algorithen. A, Reception Sphere Mechanics “The reception sphere model surrounds @ receiver point with a sphore of varying size. Rays that intersect this sphere contribute to the total received power. One of the traits of the reception sphere model i its simple fmple- mentation sinew the sphere-ray Interseetion tests ate easy ta desesibe suabhensaticalls. “The sige of the reception sphere depends on the charac- teristics of the incoming ray. Because rays spread out as they leave the souree, the reception sphere must increase size accordingly, Fugu 4 shows a two dimensional exatn- ple of Biting reception sphere sires with ieoming rays to suarautce the eollection of exactly one ray from each wavo- font. Ite sphere is too large, more than one ray from the sume wavelront adds co the total pamer. An undersized sphere can mis waveleont altogether @ spe ne L05 we ‘Diptera ig. 4. Reception epice sine Sr li-oFight (LOS) tnd enticed ‘wovefrate. ‘The meres ate aled ta fuaranies tke secepbon cf (ee and iy ene ray from ch wavethnt B. Ray Double Counting 1m two dimensions, reception epteres work perfect; 8 thint dimension adds complexity and ambiguity. Fig- ture & shows the ideal impingement of grodesiclaunched rays onto 8 spherieal wavelront. ‘The minimum radius for ‘reception sphere to guarantee the collection of wt Feast ‘one ray from a maveftont is ly the distance between rays [11], (13). This radlus wweeps outa circular area acrons the ‘wavefront where sometimes two rays fall within the sphere, registering additional voltage and power. (Minoan Sine + fm . 4 — Boule Count ‘Fig. Receptlon sphere doable cousking an a uniform wavetrost. ‘The aterm representa she ies] hmpingement of raps ute ota erlang Fig 6, Double count ero white regions) any a asa wave ‘The double counting errors occur frequently across a measured wavefront. Pigure 6 highlights the double count ing on a map of an ideal wavefront. The peobebility of a randomly placéd rectiver experiencing & double count fcornee from the geometry of Figure 6. For the ideal ease, double countings accur with a probability of 2 — 1 oe 20.9% ofthe time. ‘Tis probability fs indepondont of tessel- lation frespency and worsens with increased geodesic and reflecting surfsee aberrations Incoherent accumulation af this additional power will sult in 6 +3 dB errer, However, the arrival angles and the path lengths of tao rays from tbe sarae wavefront are iden ileal. Since their araplitudes and phases are-equal a coher eat addition will result in twice as much predicted voltage and, subsequery.0.++6 dB pawer exror. Asturing rancor location of n rocefver long & geodesic wavefront, a coherent voltage measurement with reception spheres wil increase the mean power by at least 1.25 dB and Inteoduce an ad- ditional 24 dB of standard deviation errot. Slamulations that do not use facalied anular separation to calculate reception sphere sizes exhibit even worse statistics [16] VL Distramuren Wavernonts ‘The method of distributed wavefronts reanedics the probe lems inherent with the reception sphere model while mai 787