+ All Categories
Home > Documents > GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work...

GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work...

Date post: 20-Jul-2020
Category:
Upload: others
View: 3 times
Download: 0 times
Share this document with a friend
42
GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY TRACING Masters Thesis Defense Blake Troksa Advisor: Dr. Branislav Notaros Committee: Dr. Sudeep Pasricha Dr. Hamid Chitsaz
Transcript
Page 1: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

GPU ACCELERATED CONE BASED SHOOTING BOUNCING

RAY TRACING

Masters Thesis Defense

Blake Troksa

Advisor: Dr. Branislav Notaros

Committee: Dr. Sudeep Pasricha

Dr. Hamid Chitsaz

Page 2: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

OVERVIEW

Ray Tracing Overview

• Shooting-Bouncing Ray Tracing

Geometric Calculations

• Ray Generation

• Mesh Creation

Post Processing

• Sphere Intersections

• Double Counting

Parallelization of Ray Tracing

• GPU Acceleration

• Coalesced Memory Accesses

• Speedup from Parallelization

Results

• Comparison with another SBR algorithm

• Comparison with Image Theory

• Comparison with FMM-FFT

• Comparison with Commercial Software

Future Work

• Hybridization

Page 3: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

RAY TRACING

Applications of Ray

Tracing in CEM

Two types in

CEM

Image Theory

Shooting

Bouncing Rays

Time Complexity

Searching for Facet

Intersections

Page 4: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

RAY TRACING CONT.

• Maxwell’s Equations are linear in linear, homogenous, and anisotropic propagation domains

• Assume infinite planar facet interfaces

Page 5: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

IMAGE THEORY

• Exact Path Calculation

• O(Nk)

• N – Number of

observation points

• K – Number of facets

Page 6: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

SHOOTING-BOUNCING

RAYS(SBR)

• Intuitive implementation of ray tracing

• Rays launched from transmitting antenna source point

• Path of ray is traced until intersection with a facet

• Previous intersection next intersection

• Electric field is calculated once the ray has reached a reception point

• Advantages of SBR

• Computationally quick form of ray tracing

Page 7: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

THE SBR ALGORITHM

• Two Subsections

• Geometric Path Calculations

• Mesh Creation

• Ray Generation

• Post Processing

• Electric Field Calculations

• Double Count Removal

Page 8: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

SBR ALGORITHM INPUTS

• Number of Rays

• Limit on number of Reflections

• Lossy material vs. metallic

• Geometry definition (Mesh)

• OBJ file

• Location of observation points

Page 9: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

MESHES

• LiDAR Data

• Building Blueprint

Page 10: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

RAY GENERATION

• The Icosahedron

• Provides a consistent angle between the points distributed along each face.

• Enables easy computation of separation angle for each ray

• Batching

𝑛 ∗ (𝑛 + 1)

2

Page 11: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

RAY DENSITY

• Sampling Density of Rays

• Ideally the sampling density would rise to infinity

• Spatial Angle for Rays

• Rays approximate the field information of the volume that surrounds them

• Allows for conservation of power that all rays within that volume would contain provided infinite ray density

Page 12: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

Tubes Cones

Tracing of single ray

Double Count

Removal

Trouble with Curvature

Cover entire space without

overlap

Page 13: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

SPHERE INTERSECTIONS

• Intersection of the Rays

• Rays are received at spheres as opposed to planes or points

• Spheres model an omni-directional cross-section of the cone at an observation point.

• Reception sphere continuously grow

𝑅𝑎𝑑𝑖𝑢𝑠𝑆𝑝ℎ𝑒𝑟𝑒 = Τ𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒 ∗ 𝛼 3

Page 14: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

SPHERE INTERSECTIONS

Page 15: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

SEPARATION ANGLE

• Angle across face of Icosahedron

• Same across all faces

Page 16: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

LOSSY DIELECTRIC WAVEGUIDE

• Image Theory Solution

• Long

• 1km

• Lossy Dielectric

• Dry Concrete (εr=5)

• Frequency = 1Ghz

Page 17: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

STATIC ALPHA PER-RAY ALPHA

EFFECT OF SEPARATION ANGLE

Reference results obtained here [1]

Page 18: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

MIN PER-RAY ALPHA MAX PER-RAY ALPHA

EFFECT OF SEPARATION ANGLE CONT.

Reference results obtained here [1]

Page 19: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

GEOMETRIC CALCULATIONS

• NVIDIA Ray-Tracing Application Programming Interface

• NVIDIA has invested much research in creating computationally quick ray tracing programs for use in rendering applications.

• We take advantage of the tools and software developed by NVIDIA

• Binary Space Partition Tree

• K-D treehttps://www.google.com/search?q=nvidia+optix&source=lnms&tbm=isch

&sa=X&ved=0ahUKEwjB6tmLl57jAhWPXM0KHXOTCTQQ_AUIEygE&

cshid=1562343531524818&biw=1280&bih=622#imgrc=sMPWGl7tpSmc

HM:

Page 20: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

ELECTRIC FIELD CALCULATIONS

• Attenuation of Electric Field

• Plane waves now interact by Fresnel

coefficients

• Loss due to distance traveled

Page 21: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

REFLECTED RAYS

• Decomposition into

normal and parallel

polarizations

• Angle in = Angle out

Page 22: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

DOUBLE COUNTING

• Ray’s that intersect the same triangles represent an approximation of rays with the same image theory path

• Reception spheres of the same image theory ray overlap at the observation point

• The electric field contribution is counted twice

Page 23: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

DOUBLE COUNT REMOVAL

• Removal based on

Adjacent Rays

• Rays stored in map

lookup

• Other techniques

• Removal based on check

of sphere size

Page 24: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

WITH DOUBLE COUNT

REMOVAL

WITHOUT DOUBLE

COUNT REMOVAL

EFFECTS OF DOUBLE COUNT REMOVAL

Reference results obtained here [1]

Page 25: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

ACCELERATION

• Yields efficient parallelization specifically on GPU’s.

Rays Propagate Independently

• Due to the independence of ray paths, field calculations for each ray are also independent

• GPU’s capable of handling the computing of the Fresnel coefficients

Electromagnetic Field Calculations

• Computationally intensive to re-compute the size of a sphere for each ray after each reflection.

• GPU’s efficiently handle this mathematically simple but extensive process.

Ray Sphere Intersections

Page 26: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

COALESCED MEMORY ACCESSES

• Access to global memory should be in a coalesced fashion for the threads located in each thread block.

• NVIDIA GeForce 1060 GPU’s have a warp size of 32

• Limits number of reads to global memory and increases speedup

Page 27: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

CONFIGURATION OPTIMIZATION

• Threads per block

• Rays per Thread

• Block size

Threads Per Block

Page 28: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

PARALLEL ICOSAHEDRON

Page 29: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

SPEEDUP

• Comparison between earlier iteration of

the algorithm

• No double count removal

Page 30: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

SBR AND SBR

Reference results obtained here [2]

Page 31: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

SBR AND IT

Reference results obtained here [1]

Page 32: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

REMCOM WIRELESS

INSITE

Option of shooting

bouncing ray

tracing

Exact Path

Corrections

GPU Acceleration

Page 33: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

OUR SBR AND REMCOM CONT.

Reference results obtained here [2]

Page 34: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

SBR AND FMM-FFT

• FMM-FFT

• Full-wave solver

• Supercomputer

• 4 hours and 54

minutes

Reference results obtained here [3]

Page 35: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

CONVERGENCE

• PEC Waveguide Tested

• Known Analytical Solution

• TE10 Mode

Page 36: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

RSS MAPS WITH SBR

• Received Signal Strength

• Laborious and time consuming process

• Simulated on Colorado School of Mines Edgar

Mine

Page 37: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

MEASUREMENTS

• Ground Plane Reflection

• Friis Formula

Page 38: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

MINE MEASUREMENTS

Page 39: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

FUTURE WORK

• SBR Hybridization

• Image theory - exact path adjustments

• MoM/FEM

• Diffraction

• More Real-world testing

• Adaptive sampling

• Adaptive observation point checks

Page 40: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

ACKNOWLEDGEMENTS/REFERENCES

• This work was supported by the National Science Foundation under grant ECCS-1646562.

1. D. Didascalou, Ray-optical wave propagation modeling in arbitrarily shaped tunnels, 2000.

2. Shin-Hon Chen and Shyh-Kang Jeng, "SBR image approach for radio wave propagation in tunnels with and without traffic," in IEEE Transactions on Vehicular Technology, vol. 45, no. 3, pp. 570-578, Aug. 1996.

3. A. C. Yucel, W. Sheng, C. Zhou, Y. Liu, H. Bagci and E. Michielssen, "An FMM-FFT Accelerated SIE Simulator for Analyzing EM Wave Propagation in Mine Environments Loaded With Conductors," in IEEE Journal on Multiscale and Multiphysics Computational Techniques, vol. 3, pp. 3-15, 2018.

4. M.F.Cátedra and J.Perez, Cell Planning for Wireless Communications. Norwood, MA, USA: Artech House, 1999.

5. B. M. Notaros, Electromagnetics, New Jersey : PEARSON Prentice Hall; 2010.

6. V. Mohtashami and A. A. Shishegar, "A new double—counting cancellation

technique for ray tracing using separation angle distribution," 2008 IEEE

International RF and Microwave Conference, Kuala Lumpur, 2008, pp. 306-310.

Page 41: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

THANK YOU

Page 42: GPU ACCELERATED CONE BASED SHOOTING BOUNCING RAY … · ACKNOWLEDGEMENTS/ REFERENCES • This work was supported by the National Science Foundation under grant ECCS- 1646562. 1. D.

SBR, FMM-FFT, REMCOM


Recommended