You are on page 1of 8

NEW PROPAGATION MODEL USING FAST 3D RAYTRACING APPLICATION TO WIFI Jean-Franois Legendre(1), Olivier Gesny(1), Jean-Christophe Bayetto(1), Laurent

Roger(2)
(1)

SILICOM Rgion Ouest, 4 rue de Jouanet 35700 Rennes, FRANCE Email: silicomregionouest@silicom.fr

(2)

DGA, Centre dElectronique de lArmement, BP 7419, 35174 Bruz Cedex, FRANCE Email: Laurent.ROGER@dga.defense.gouv.fr

SHORT ABSTRACT With the growth of high data rate speed cellular communication systems (UMTS, GSM, WiFi, Bluetooth), prediction of the electromagnetic propagation becomes a strategic stake and an active subject of research. To give fast realistic predictions of the field strength and delay spread in wireless mobile communication networks, SILICOM developed the SLC software(note), based on parallel 3D raytracing combined with GO/UTD theory [1]. This paper describes our solution and shows its application to WiFi hotspot study. Keywords : propagation, raytracing, voxels grid, WiFi

INTRODUCTION

SLC is a computation engine containing advanced propagation models that enable to deal with different propagation modes as shown on Fig. 1. :

Transmitter
1

Receiver

Voltage source

LISN (current measurement)

Fig. 1. The four propagation models integrated in software SLC


(note)

This project is supported by DGA (CELAR). ECPS 2005 Conference, 15-18 March 2005, BREST, FRANCE

To compute the main mode (mode 1 : radiating waves traveling between two antennas), we use 3D raytracing and UTD. The main application is radio planning of cellular communication systems. SLC also deals with coupling and conduction propagation modes along cables (modes 2 and 3), as well as radiation of each induced current elements (mode 4 using raytracing model). The main application is EMC in vehicules, shipsas well as conducting and radiating cables. To compute these modes, our raytracing is hybridized with the NEC2 code (the Numerical Electromagnetics Modeling code developed at Lawrence Livermore National Laboratory ) based on the method of moments. 2 OUR APPROACH

2.1 Introduction Rayoptical propagation models, used for field strength prediction, suffer from long computation times when applied to large urban areas or indoor scenarios. University of Stuttgart has implemented a fast and accurate model where scene is represented by rectangular tiles [2]. Taking this work as a starting point, SILICOM has developed its own model where geometry is fully described with 3D triangular facets. This enables us to model any 3D scenes (eg building, vehicules, ships) without the great amount of RAM necessary for precise DTM/DEM raster data. 2.2 Use of hierarchical grid of voxels The other main improvement of our approach is to introduce concepts from digital picture processing, such as use of hierarchical voxels grid, as shown on Fig. 2. This pre-processing allows to improve the computation time of the main raytracing procedure: the intersection of a ray with the 3D scene. The pre-processing aims at selecting the 3D facets intercepted by a ray. Each voxel of this grid contains all the 3D facets that lie on it. Voxels are filled using the incremental line technique and the 2D polygon filling technique. The chosen resolution L of a voxel is very important. If L is too large, a voxel may contain a great part of the 3D scene and there is no improvement. If L is too short, this leads to a great amount of memory in order to store these 3D facets, and to a computation overlay during intersection tests. Assuming a uniform distribution of N facets on a total surface S, the adaptive value L used by SLC is L=0.5(S/N). However, when detailed buildings (with inner walls and furniture) are included in the 3D scene (for outdoor-indoor application), a single grid of voxels is not efficient. SILICOM implemented hierarchical grids of voxels [3] with different resolutions : typically, one grid for the DTM ground, one grid for each building

ECPS 2005 Conference, 15-18 March 2005, BREST, FRANCE

Trees

ground

buildings A

L = resolution Fig. 2. Hierarchical voxels grid and incremental line (in transparent gray) Intersection tests of the (A,B) line with the surroundings involve three steps : 1. SLC checks if the borders of each grid are intersected using Liang Barsky parametric line-clipping algorithm. 2. SLC creates the incremental line of voxels from A to B in order to get all the 3D facets that may be intersected by this line (A,B). 3. Finally, the geometrical test [4] of line (A,B) intersection with each facet is done. This pre-processing enables to handle very complex 3D scenes that contain millions of facets. Fig. 3. (up) shows a realistic example with three levels of details : outdoor, indoor (inner walls produced from architect plan using IntelliCAD) and furniture (PC, table, chair from CAD library). Fig. 3. (down) shows the corresponding voxels grid where red zones mean a lot of triangular facets and blue zones only few.

ECPS 2005 Conference, 15-18 March 2005, BREST, FRANCE

Level 1 : DEM / outdoor

mjgokfdoiuepotiupoeiuposjfspodspispoispofispoipodpreo

Level 2 : indoor

Level 3 : furniture

Fig. 3. Digital geometric model with three levels (above) and voxels grid of level 1 (below)

ECPS 2005 Conference, 15-18 March 2005, BREST, FRANCE

2.3 Ray-tracing principle The first stage of our ray-tracing is based on a pre-processing of the database. Rays are thrown from the transmitter towards the centre of a facet (for reflection) and the centre of the three wedges (for diffraction). Receiver (dynamic step)

No reflection

Transmitter (static preprocessing step) Fig. 4. Raytracing principle For each facet, the reflection image of the transmitter is computed and the 3D beam formed by the image point and facet is stored in a special voxels grid (build by XY projection of each bundle, like its footprint with a zenith sun). For double reflection, we just have to check if the center of the second facet lies in a voxel containing the beam and perform a vectorial intersection test [4] between former reflection image and this new center. Such a dedicated voxels grid avoids to test all the beams, with the well-known sphere test. Fig. 5. shows an example where red zones contain a lot of beams (near transmitter), and blue zones only few. We proceed in the same way for diffraction on wedges ; Kellers cone is stored in the same dedicated voxels grid. For indoor or mix models, transmission through walls is take into account.

Reflection bundle Virtual facet True facet

Receiver position 80 m

Image point

Transmitter position =voxel where reflection is possible

Fig. 5. Reflection bundle and the associated voxels grid

ECPS 2005 Conference, 15-18 March 2005, BREST, FRANCE

For each receiver location, the (dynamic) prediction stage consists in selecting diffraction or reflection rays interception thanks to the voxels grid process. Once a ray is found out, its magnitude, phase and delay is computed with GO/UTD coefficients ; they depend on dielectrical characteristics of material stored in GIS attributes of each facet. 2.4 Validation In order to validate the approach described above, a measurement campaign (the digital terrain model is illustrated on Fig. 3.) using broadband CW signals (from 10 to 1000 MHz) using two ARAs discone antennas was performed. The transmitter antenna is placed into a room where furniture is fully described as shown on Fig. 6. The Receiver antenna (height=1.5 m) is outside 80 meters away from transmitter (Fig. 5.). Our model takes into account the variation of dielectrical characteristics and antenna gain with frequency. Fig. 6. shows the received signal power for one receiver position : It illustrates how SLC results match with broadband measurements.
10 0 -10 -20 -30 -40 -50 -60 -70 10

Power dBm

(Image courtesly of Ondim-CELAR)


100

Frequency MHz

1000

Fig. 6. Broadband propagation model validation and Radiated field inside the room 2.5 Intra and inter system interference analysis Intra system interference means perturbation due to the same system (typically co-channel and adjacent channel). Inter system interference can happen when several systems (radar, communication...) share the same frequency band. For accurate C/I evaluation, the received power (in dBm) is corrected with an additional loss, named FDR [5] Frequency Dependent Rejection. This factor takes into account the transmitter spectrum P(f), the receiver selectivity H(f) and frequency separation f = ft-fr between transmitter and receiver (For WiFi studies, f=25 MHz).

P(f )df
FDR (f ) =10 log10
-

(1)

P(f ) H(f + f ) df
-

ECPS 2005 Conference, 15-18 March 2005, BREST, FRANCE

APPLICATION TO WIFI

New modules have been integrated in software SLC in order to plan WiFi coverage : When the field strength is greater than a fixed level, WiFi connection is possible and the maximum throughput may be obtained from the computed C/I value through hardware characteristics. Table 1 shows the data rate reached via Orinoco Laptop/PDA WiFi card depending of the received power C and the signal to interference ratio C/I : Throughput Recommended Recommended C C/I 11 Mbps -82 dBm 16 dB 5.5 Mbps -87 dBm 11 dB 2 Mbps -91 dBm 7 dB 1 Mbps -94 dBm 5 dB Table. 1. Orinoco PCMCIA Silver/Gold

In january 2004, SILICOM performed a full WiFi hotspot planification for the Palais des Festivals building (Cannes). Our results allowed to reduce the number of Access Points from 52 to only 40.
C (dBm) C/I (dB)

Palais des festivals (Cannes)


Rate (Mbps) Best AP

3D model

Fig. 7. Planification results of the Palais des Festivals hotspot

ECPS 2005 Conference, 15-18 March 2005, BREST, FRANCE

First, Access Points are distributed over the building using geometrical symmetry, wall masks. Once the coverage threshold (-78 dBm for 90 % of the area, see Fig. 6.) is achieved, channel allocation is realized (Fig. 8.) based on the 1x3 indoor channel shape : channels 1, 6 and 11 for indoor, and channels 10 and 12 for outdoor. C/I calculation takes into account frequency dependent rejection with the Wifi channel mask spectrum shown on Fig. 9. Assuming an Orinoco PDA card (Table. 1.), we obtain the throughput distribution over the area.

0 dB

-30 dB -50 dB 22 MHz 44 MHz

C1 25 MHz

C6

C11

Fig. 8. Channel allocation

Fig. 9. Channel mask spectrum

CONCLUSION

In this paper, a deterministic ray tracing model for outdoor/indoor scenario is presented. The model is based on a hydrid accelerated raytracing (with voxels grid concept) that was adapted to WiFi plannification. This propagation model dedicated to WiFi is used by SILICOM engineers ; starting from CAD plans, they build the digital terrain model and can predict WiFi coverage and find the optimal number of Access Points. 5 REFERENCES

[1] J.F. Legendre, Etude de modles de prdiction de la propagation bass sur la thorie gomtrique de la diffraction , Thesis, INSA, Rennes, 18 january 1995 [2] G. Wlfle, R. Hoppe, and F. M. Landstorfer, A Fast and Enhanced Ray Optical Propagation Model for Indoor and Urban Scenarios, Based on an Intelligent Preprocessing of the Database, 10th IEEE Internat. Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC) 1999, Sept. 1999, Osaka, Japan F5-3 [3] P. Slusallek, Computer graphics RayTracing II , Universitt des Saarlandes, 2003 [4] D. Sunday, Intersections of Rays and Segments with Triangles in 3D , may 2001 [5] FDR PROGRAM , National Telecommunications and Information Administration , www.ntiacsd.ntia.doc.gov/msam/FDR/

ECPS 2005 Conference, 15-18 March 2005, BREST, FRANCE

You might also like