faster than fdtd pushing the boundaries of time-domain modeling … · 2006-05-29 · time-domain...

95
Dynamic Dynamic AMR AMR - - FDTD: FDTD: Method Method Dynamic Dynamic AMR AMR - - FDTD: FDTD: Examples Examples Wireless Wireless Channel Channel Modeling Modeling GPU GPU - - based based Time Time - - Domain Domain Modeling Modeling Radiation Radiation from NRI from NRI - - TL TL structures structures Conclusion Conclusion Intro Intro 1 Faster than FDTD Faster than FDTD : : Pushing the boundaries of Time Pushing the boundaries of Time - - Domain Domain Modeling for Wireless and Optical Modeling for Wireless and Optical Propagation Problems Propagation Problems Costas D. Sarris Costas D. Sarris Assistant Professor Assistant Professor The Edward S. Rogers Sr. Department of The Edward S. Rogers Sr. Department of Electrical and Computer Engineering Electrical and Computer Engineering University of Toronto University of Toronto Research supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) and the Ontario Centers of Excellence.

Upload: others

Post on 11-Mar-2020

6 views

Category:

Documents


0 download

TRANSCRIPT

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

1

Faster than FDTDFaster than FDTD: : Pushing the boundaries of TimePushing the boundaries of Time--Domain Domain

Modeling for Wireless and Optical Modeling for Wireless and Optical Propagation Problems Propagation Problems

Costas D. SarrisCostas D. SarrisAssistant ProfessorAssistant Professor

The Edward S. Rogers Sr. Department ofThe Edward S. Rogers Sr. Department ofElectrical and Computer EngineeringElectrical and Computer Engineering

University of TorontoUniversity of Toronto

Research supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) and the Ontario Centers of Excellence.

2

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

AcknowledgementsAcknowledgements

•• Graduate studentsGraduate students– Abbas Alighanbari, Gerard Baron, Titos Kokkinos

•• PostPost--doctoral fellowdoctoral fellow– Dr. Yaxun Liu

•• ColleaguesColleagues–– Prof. George Eleftheriades, Prof. Eugene FiumeProf. George Eleftheriades, Prof. Eugene Fiume

3

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

OutlineOutline

•• IntroductionIntroduction– Finite-Difference Time-Domain (FDTD) – Current challenges; how our work addresses them

•• Dynamically Adaptive Mesh Refinement FDTDDynamically Adaptive Mesh Refinement FDTD– Overview of the algorithm– Applications: Microwave and optical structures– Performance evaluation

•• FullFull--Wave Indoor Channel ModelingWave Indoor Channel Modeling–– The SThe S--MRTD technique applied to channel modelingMRTD technique applied to channel modeling–– GPU Accelerated SGPU Accelerated S--MRTDMRTD

•• Fast FDTD Analysis of Periodic LeakyFast FDTD Analysis of Periodic Leaky--Wave StructuresWave Structures–– FloquetFloquet--based methodology and validationbased methodology and validation–– Application to NRIApplication to NRI--TL based leakyTL based leaky--wave antennaswave antennas

•• ConclusionConclusion

4

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Introduction: TimeIntroduction: Time--Domain ModelingDomain Modeling

IntroIntro

Example :

FDTD discretization of

⎟⎟⎠

⎞⎜⎜⎝

⎛∂

∂−

μ=

∂∂

yE

zE1

tH zyx

⎥⎥⎦

⎢⎢⎣

Δ

−−

Δ

μΔ

+= ++++++++−+++ y

EEz

EEtHHz

2/1k,j,inz

2/1k,1j,iny

k,2/1j,iny

1k,2/1j,inx2/1k,2/1j,i2/1n

x2/1k,2/1j,i2/1n

Marching in time scheme

In FDTD, the computational domain is divided in “Yee’s cells” and Maxwell’s equations are solved by marching in time.

Yee’s cell

5

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

FDTD: Successful, yet..FDTD: Successful, yet..

IntroIntro Simple to implement, versatile, direct update equations (no

matrix assembly, storage).

Pronounced “numerical dispersion” necessitates use

of dense grid.

• “High-order” methods can do much better

Small time step enforced by stability criterion

• CFL criterion:

s<=1

• Maintaining large cell size through “adaptive meshing”

6

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Mesh Refinement in FDTDMesh Refinement in FDTD• Local mesh refinement schemes: Embedding a locally

dense mesh into a coarse mesh.

• Mesh refinement guided by physical intuition; statically defined at the start of the simulation.

• Side-effect: Late-time instability.• Sample references:

Example: Non-uniform mesh for microstrip

I. S. Kim and W. J. R. Hoefer, MTT-T, June 1990.S. S. Zivanovic, K. S.Yee, and K. K. Mei, MTT-T, Mar. 1991.M. W. Chevalier, R. J. Luebbers, and V. P. Cable, AP-T, Mar. 1997.M. Okoniewski, E. Okoniewska, and M. A. Stuchly, AP-T, Mar. 1997.

IntroIntro

7

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Dynamic Mesh Refinement: Motivation Dynamic Mesh Refinement: Motivation

• Time-domain methods register the evolution of a source pulse and its retro-reflections in a given domain.

• Edges, high-dielectric regions etc. are notcontinuously illuminated during an FDTD simulation; local mesh refinement around them is NOT always necessary.

Absorbing boundary

Simulated Structure

Widebandsource

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

8

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Dynamic Mesh Refinement: Previous WorkDynamic Mesh Refinement: Previous Work

Adaptive Mesh Refinement [Berger, Oliger, J. Comput. Physics,1984]:

– Computational fluid dynamic tool for hyperbolic PDEs. – Performs selective mesh refinement by factors of 2.– Allows for the dynamic re-generation of coarse/dense

mesh regions.

Moving-Window FDTD (MW-FDTD, Luebbers et al., Proc. IEEE AP-S, June 2003):

– Single moving window of fixed width, velocity tracking a forward wave in a wireless link.

– Window is terminated into absorbing boundaries (no possibility for reflected wave modeling).

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

9

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Dynamically AMRDynamically AMR--FDTD: OverviewFDTD: Overview

Key features of this work on Dynamically Adaptive Mesh Refinement (AMR)-FDTD

– Combination of the FDTD technique with the AMR algorithm.

– Implementation of a three-dimensional adaptive, moving mesh.

– Evaluation of efficiency and accuracy for electromagnetic problems, considering realisticrealisticapplications.

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

10

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

AMRAMR--FDTD: Root/Child MeshesFDTD: Root/Child Meshes

The AMR-FDTD starts by covering the computational domain in a coarse mesh (called root mesh), of Yee cell dimensions

Every NAMR time steps, checks whether mesh refinement is needed at any part of the domain.

Clustering together cells that have been “flagged” for refinement, it generates a child mesh that covers them, with cell sizes

Recursively, child meshes can be refined by a factor of two if flagged at a later check.

.zyx Δ×Δ×Δ

.2z

2y

2x Δ

×Δ

×Δ

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

11

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

AMRAMR--FDTD: Root / Child Meshes (contFDTD: Root / Child Meshes (cont--d)d)

Mesh generation corresponds to a tree structure.

A

B1 B2 B3 B4 B5

C1 C2

A

B1 B2

B3B4

B5

C1 C2

zyx Δ×Δ×Δ

2z

2y

2x Δ

×Δ

×Δ

4z

4y

4x Δ

×Δ

×Δ

MESH TREEMESH TREE Yee cellsYee cells

Level 1

Level 2

Level 3

A: Level 1 (Root) MeshB1, B2,…, B5: Level 2 MeshesC1, C3: Level 3 Meshes

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

12

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

AMRAMR--FDTD: Time SteppingFDTD: Time SteppingStability condition for the root mesh:

Courant number s < 1.

Keeping the same Courant number in all meshes, the time step of level M mesh is:

NoteNote: Minimum cell size affects the time step of the corresponding mesh level only (asynchronous updates).

Field Field Updates in Updates in AMRAMR--FDTDFDTD

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

13

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Field Update Flowchart: GeneralField Update Flowchart: General

Check the number of time steps; if it is an integer multiple of NAMR, re-generate the

mesh tree.

Update field grid points of the root mesh

Copy fields from the root mesh to the child meshes. Update level M meshes 2M-1 times.

If maximum time step has been reached, terminate. Otherwise go back to (1).

11

22

33

44

55

Copy fields of the child meshes back to the root mesh for the time steps of the root mesh

(interpolating as needed) .

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

14

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Mesh boundary updates: Mesh boundary updates: CPBsCPBsChild/Parent grid points: Never collocated in space Child/Parent grid points: Never collocated in space or time (always or time (always interleavedinterleaved). ).

Transfer of field values from the one mesh to the Transfer of field values from the one mesh to the other involves other involves spatialspatial and and temporaltemporal interpolationsinterpolations. .

: Child mesh: Parent mesh

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

15

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Each NAMR time steps, the mesh tree is regenerated.

Method: Calculate energy in each Yee cell and then gradient throughout the domain.

Adaptive Mesh RefinementAdaptive Mesh Refinement

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

16

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

If bothboth of the following conditions are met :

Adaptive Mesh Refinement (contAdaptive Mesh Refinement (cont--d)d)

: : predefinedpredefined thresholdsthresholds

cell (i, j, k) of the root mesh is flagged for refinementcell (i, j, k) of the root mesh is flagged for refinement

First criterion: Captures energy gradient peaks.

Second criterion: Prevents numerical noise (at later stages) from

triggering spurious refinements.

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

17

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Mesh Refinement is extended at a distance D around a flagged cell:

This accounts for wave propagation within the mesh refinement time window of NAMR time steps.

Flagged cells are clustered in rectangular regions following thealgorithm of [Berger and Rigoutsos, IEEE Trans. Systems, Man, Cybernetics, Sept. 1991].

Adaptive Mesh Refinement: ClusteringAdaptive Mesh Refinement: Clustering

Flagged cells

: “spreading” factorDynamic Dynamic

AMRAMR--FDTD:FDTD:

MethodMethod

18

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Application: Microstrip LowApplication: Microstrip Low--Pass Filter*Pass Filter*

A=40mm, B1=2mm, B2=21mm, W=3mm,

0.8mm substrate of εr=2.2

Vertical electric field magnitude

Time = 0 Number of Child Meshes = 1 Refined volume/total volume = 0.043

*From Sheen et al, IEEEMTT-T, July 1990.

Dynamic AMR-FDTD:Examples

19

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Application: Microstrip LowApplication: Microstrip Low--Pass FilterPass Filter

Time = 100Δt Number of Child Meshes = 1 Refined volume/total volume = 0.134

A=40mm, B1=2mm, B2=21mm, W=3mm,

0.8mm substrate of εr=2.2

Dynamic AMR-FDTD:Examples

Vertical electric field magnitude

20

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Application: Microstrip LowApplication: Microstrip Low--Pass FilterPass Filter

Time = 200Δt Number of Child Meshes = 1 Refined volume/total volume = 0.525

Vertical electric field magnitude

A=40mm, B1=2mm, B2=21mm, W=3mm,

0.8mm substrate of εr=2.2

Dynamic AMR-FDTD:Examples

21

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Application: Microstrip LowApplication: Microstrip Low--Pass FilterPass Filter

Time = 500Δt Number of Child Meshes = 3 Refined volume/total volume = 0.442

A=40mm, B1=2mm, B2=21mm, W=3mm,

0.8mm substrate of εr=2.2

Dynamic AMR-FDTD:Examples

Vertical electric field magnitude

22

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Application: Microstrip LowApplication: Microstrip Low--Pass FilterPass Filter

Time = 800Δt Number of Child Meshes = 3 Refined volume/total volume = 0.28

A=40mm, B1=2mm, B2=21mm, W=3mm,

0.8mm substrate of εr=2.2

Dynamic AMR-FDTD:Examples

Vertical electric field magnitude

23

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Microstrip LowMicrostrip Low--Pass Filter: Evolution of Child MeshesPass Filter: Evolution of Child Meshes

Coverage=Volume of child meshes / total volume of the domain

Dynamic AMR-FDTD:Examples

24

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Microstrip LowMicrostrip Low--Pass Filter: Evolution of Child MeshesPass Filter: Evolution of Child Meshes

In the long-time regime, AMR-FDTD is equivalent to a root-mesh based uniform mesh FDTD (reason for no late-time instability).

Dynamic AMR-FDTD:Examples

25

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

LateLate--Time RegimeTime Regime

Dynamic AMR-FDTD:Examples

No late-time instability observed !

26

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Microstrip LowMicrostrip Low--Pass Filter: SPass Filter: S--parametersparameters

94.6% reduction in execution time

Dynamic AMR-FDTD:Examples

27

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Application: Microstrip Branch Coupler*Application: Microstrip Branch Coupler*A=40 mm, B1=7 mm,

B2=11 mm ,B3=9 mm, W1=2 mm,W2=3 mm,

0.8mm substrate of εr=2.2

94.8% reduction in execution time

*From Sheen et al, IEEE MTT-T, July 1990.

Dynamic AMR-FDTD:Examples

28

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Application: Microstrip Spiral InductorApplication: Microstrip Spiral Inductor

82% reduction in execution time

A1=60 mm, A2=40 mm, B1=24 mm, B2=20 mm ,B3=18 mm, B4=4 mm ,

W1=W2=2 mm,0.8mm substrate of εr=2.2

Dynamic AMR-FDTD:Examples

29

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Optical applications: Power SplitterOptical applications: Power Splitter

• Dimensions are given in microns. • Dielectric constants: 4lens

r =ε 9guider =ε

Dynamic AMR-FDTD:Examples

30

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Power Splitter: TimePower Splitter: Time--Domain ResultsDomain Results

Port 2

• AMR-FDTD with four levels

Port 3

Dynamic AMR-FDTD:Examples

31

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Power Splitter: Numerical Results (contPower Splitter: Numerical Results (cont--d)d)

• Error Metric:

Dynamic AMR-FDTD:Examples

32

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Power Splitter: Wave front TrackingPower Splitter: Wave front Tracking

Dynamic AMR-FDTD:Examples

33

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Power Splitter: Wave front TrackingPower Splitter: Wave front Tracking

Dynamic AMR-FDTD:Examples

34

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Power Splitter: Wave front TrackingPower Splitter: Wave front Tracking

Dynamic AMR-FDTD:Examples

35

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Power Splitter: Wave front TrackingPower Splitter: Wave front Tracking

Dynamic AMR-FDTD:Examples

36

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Dielectric Ring Resonator* (4Dielectric Ring Resonator* (4--level AMR)level AMR)

*Hagness et al., IEEE J. Lightwave Tech., vol. 15, pp. 2154-2165, Nov. 1997.

Dynamic AMR-FDTD:Examples

Port 2

37

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Dielectric Ring Resonator (contDielectric Ring Resonator (cont--d)d)

Dynamic AMR-FDTD:Examples

38

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Dielectric Ring Resonator: LateDielectric Ring Resonator: Late--time regimetime regime

Dynamic AMR-FDTD:Examples

No late-time instability observed !

39

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

ReferencesReferences

1. Y. Liu and C.D. Sarris, “Efficient Modeling of Microwave Integrated Circuit Geometries via a Dynamically Adaptive Mesh Refinement (AMR) - FDTD Technique”, IEEE Trans. Microwave Theory and Tech., vol. 54, no. 2, pp. 689-703, Feb. 2006.

2. Y. Liu and C.D. Sarris, “A Multilevel Dynamically Adaptive Mesh Refinement (AMR)-FDTD Technique Applied to Dielectric Waveguide Structures”, IEEE/OSA Journal of Lightwave Tech., to appear 2006.

3. Y. Liu and C.D. Sarris, “Numerical Error Analysis and Control in a Dynamically Adaptive Mesh Refinement (AMR) - FDTD Technique”, Proc. 2006 IEEE MTT-S International Microwave Symposium, to appear.

Dynamic AMR-FDTD:Examples

40

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Indoor Wireless Channel ModelingIndoor Wireless Channel Modeling

Wireless Channel Modeling

– Advances in computing enable the application of differential methods to large-scale problems.

–– Recent papersRecent papers: • Hybridize FDTD with UTD [Bernardi et al., IEEE MTT-

T, Dec. 2003, Ray-Tracing [Wang et al., IEEE AP-T, May 2000].

• Demonstrate physical effects (e.g. wall attenuation) that information-theoretic models do not accurately account for [Yun et al., IEEE AP-T, April 2004].

41

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Indoor Wireless Channel ModelingIndoor Wireless Channel Modeling

Wireless Channel Modeling

What does What does ““highhigh--orderorder”” mean? mean? – Spatial and/or temporal partial derivatives

approximated within an error ~ Δp.

SecondSecond--order finite order finite difference difference

(FDTD)(FDTD)

HighHigh--order finite order finite differencedifference

(MRTD(MRTD--exceptexcept Haar, Haar, PSTD, spectral methods PSTD, spectral methods

etc) etc) “connection coefficients”

42

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Indoor Wireless Channel ModelingIndoor Wireless Channel Modeling

Wireless Channel Modeling

What can highWhat can high--order methods do for largeorder methods do for large--scale problems?scale problems?– Coarser mesh: Memory savings.– Potential for execution time economy.

Disadvantages/Potential bottlenecksDisadvantages/Potential bottlenecks– Boundary conditions.– Material properties.– Increased inter-processor communication in

domain-decomposition.

43

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

HighHigh--Order SOrder S--MRTD Technique*MRTD Technique*

Wireless Channel Modeling

S-MRTD discretization

:“connection coefficients”

*M. Krumpholz, L. Katehi, IEEE MTT-T, April 1996.

44

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

HighHigh--Order SOrder S--MRTD Technique (contMRTD Technique (cont--d)d)

Wireless Channel Modeling

Dispersion analysis for Coifman SDispersion analysis for Coifman S--MRTD*MRTD*

*A. Alighanbari, C.D. Sarris, IEEE T*A. Alighanbari, C.D. Sarris, IEEE T--AP, Aug. 2006, to appear.AP, Aug. 2006, to appear.

45

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Indoor Wireless Channel ModelingIndoor Wireless Channel Modeling

Wireless Channel Modeling

Case study: 27m x 27m indoor floor plan.Bi-orthogonal Deslariers-Dubuc functions used.

source

46

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

TimeTime--domain field waveforms (contdomain field waveforms (cont--d)d)

• S-MRTD accuracy at 5 cells per wavelength comparable to FDTD accuracy at 20 cells per wavelength.

47

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

TimeTime--domain field waveforms (contdomain field waveforms (cont--d)d)

• S-MRTD accuracy at 5 cells per wavelength comparable to FDTD accuracy at 20 cells per wavelength.

48

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Vertical electric field magnitude at 900 MHzVertical electric field magnitude at 900 MHz

Wireless Channel Modeling

• Wall conductivity 0.002 S/m

FDTD: 52 hrs 36 mins

MRTD: 12 hrs 44 mins

49

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Vertical electric field magnitude at 900 MHzVertical electric field magnitude at 900 MHz

Wireless Channel Modeling

• Wall conductivity 0.05 S/m

FDTD: 52 hrs 36 mins

MRTD: 12 hrs 44 mins

50

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

SystemSystem--level fading modelslevel fading models

Wireless Channel Modeling

•Wall conductivity 0.002 S/m, LOS points

51

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

SystemSystem--level fading models (contlevel fading models (cont--d)d)

Wireless Channel Modeling

•Wall conductivity 0.05 S/m, LOS points

52

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Wall attenuation model Wall attenuation model

Wireless Channel Modeling

• Wall conductivity 0.002 S/m

53

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Wall attenuation model (contWall attenuation model (cont--d)d)

Wireless Channel Modeling

• Wall conductivity 0.05 S/m

54

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

General Purpose Computing on Graphics HardwareGeneral Purpose Computing on Graphics Hardware

• Commodity Graphics Hardware– Inexpensive– Fast

• Intel Pentium 4 3GHz– 6 GFLOPS (Theoretical)– 5.96 GB/sec peak (Theoretical)

• NVIDIA GeForce 6800 Ultra– 53 GLOPS (Observed)– 35.2 GB/sec peak (Observed)

– Faster• CPU

– 1.5x Annual Growth– 60x Per Decade Growth

• GPU– Greater than 2x Annual Growth– Greater than 1000x Per Decade

Growth

Values from “GPGPU: General-Purpose Computing on Graphics Hardware” Short Coarse Slides, SIGGRAPH ‘04

GPU-basedTime-

DomainModeling

55

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

General Purpose Computing on Graphics HardwareGeneral Purpose Computing on Graphics Hardware• Real-Time 3D Graphics

– Interactivity over physical accuracy

– Assumptions for parallelization

• Geometric primitives processed independently

• Interdependencies (i.e. shadows or reflections) only approximated

– Hardware acceleration by

• Fast local memory• Multiple pipelines

– Latest Innovation• Programmable Pipelines

(Shaders)• High-Precision Floating

Point

GPU-basedTime-

DomainModeling

56

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Graphics Accelerated FDTDGraphics Accelerated FDTD

– S. E. Krakiwsky, L. E. Turner, and M. M. Okoniewski, “Graphics Processor Unit (GPU) Acceleration of Finite-Difference Time-Domain (FDTD) Algorithm,” in IEEE International Symposium on Circuits and Systems, May 2004.

– G. S. Baron, C. D. Sarris, and E. Fiume, “Fast and Accurate Time-Domain Simulation with Commodity Graphics Hardware,” in Proceedings of IEEE Antennas andPropagation Society International Symposium, July 2005.

GPU-basedTime-

DomainModeling

57

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Graphics Accelerated SGraphics Accelerated S--MRTDMRTD

– Reformulate update equations; consider them as discrete spatial convolutions:

– Time-stepping is an iterative sequence of filters.

GPU-basedTime-

DomainModeling Weights of discrete convolution

kernels

58

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Implementation: DetailsImplementation: Details

• Step one: Field and Material Property Meshes as Images– Pixel Shader constructs for

sampling 2D and 3D single-precision float pointing arrays or textures

– Textures: input and output streams

• 4 color channels • Up to 16 texture inputs• Up to 4 texture outputs

GPU-basedTime-

DomainModeling

59

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Implementation: DetailsImplementation: Details

• Step two: Realize update equations as sums of Finite Impulse Response filters– Update equations

combination of a filter and blend

– Enforcing source is a so-called screen or overlay

GPU-basedTime-

DomainModeling

60

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Evaluation: PerformanceEvaluation: Performance

– Per Scheme Speedup• GPU consistently faster than CPU• FDTD scheme approaches 10x speedup• S-MRTD schemes approach 30x speedup !!

GPU-basedTime-

DomainModeling

61

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

EvaluationEvaluation• Overall Speedup

– How does the performance equivalent (in terms of accuracy) FDTD and S-MRTD compare?

– Test: All schemes• coarse to very-fine

discretizations • λ/3 to λ/24 • 64x64 to 512x512 cell meshes

– Reference: FDTD• 1024x1024 cell mesh at λ/48

– Simulation Parameters:• 1.0 GHz Gaussian source• 6.4x6.4m2 cavity• 7.55 ns absolute time

GPU-basedTime-

DomainModeling

62

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Evaluation: Overall SpeedupEvaluation: Overall Speedup

– Error

– No practical difference between CPU and GPU (e.g. non-standard arithmetic not an issue)

• Observation: Very-Fine (λ/24) FDTD comparable to Medium (λ/6) DD4 simulation

– Medium discretization 4x coarser (e.g. 16x fewer cells)– Demonstration of S-MRTD’s superior dispersion

characteristics

GPU-basedTime-

DomainModeling

max

63

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Evaluation: Overall SpeedupEvaluation: Overall Speedup– Execution Times (ms): GPU and CPU

– Medium S-MRTD/Very-Fine FDTD Speedups• GPU: 6.67x• CPU: 5.95x• Across Architectures: 78.43x

GPU-basedTime-

DomainModeling

64

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Application: Wireless Channel ModelingApplication: Wireless Channel Modeling

• Indoor Wireless Channel Simulation– 1.0 and 2.4 GHz sources– 15.36x15.36 m2 floor plan– 6-8cm walls

(σ = 0.002 Ω-1 and εr=2.89)– 386.44 ns absolute time– Test: DD4 GPU

• Δx,Δz = 2 cm(λ/15 and λ/6.25)

• 728x728 cell mesh• 16 layer UPML ABC• 17.25 minutes

– Reference: FDTD CPU• Δx,Δz = 0.5 cm

(λ/60 and λ/25)• 3072x3072 cell mesh• 64 layer UPML ABC• 38.53 hours

GPU-basedTime-

DomainModeling

65

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

GPUGPU--Based SBased S--MRTD: NotesMRTD: Notes

– The use of S-MRTD enables the reduction of the memory requirements of a given problem; can make an otherwise GPU-unsolvable problem (due to memory) “fit” in the GPU !

– If so, sustainable 30x speed-ups are possible (much larger than any reported FDTD speed-ups).

– What was long perceived as a disadvantage of S-MRTD, i.e. increased operations per cell, is actually advantageous for GPU implementations.

– Possibility of “real-time” full-wave channel modeling provided by GPU-based S-MRTD.

GPU-basedTime-

DomainModeling

66

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro Floquet theorem : TwoFloquet theorem : Two--dimensional casedimensional case

( ) ( ) pkj perE~prE~rrrrrrr

−=+

ykxkk yxp ˆˆ +=r

Lattice vectorLattice vectorydxdp yx ˆˆ +=r

Bloch wavevectorBloch wavevector

(x)

(y)

dy

dx

Modeling of Infinite Periodic StructuresModeling of Infinite Periodic Structures

Radiation from NRI-TLstructures

67

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Modeling of unit cell through Periodic Boundary Conditions Modeling of unit cell through Periodic Boundary Conditions (PBCs)(PBCs)

dy

dx

• Application of Floquet’ s conditions between the boundaries in the directions of periodicity, enables the dispersion analysis of the structure.

PBC

PBC

Modeling of Infinite Periodic Structures (contModeling of Infinite Periodic Structures (cont--d)d)

(x)

(y)

dy

dx

Radiation from NRI-TLstructures

68

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

FDTD Modeling of Periodic Structures*FDTD Modeling of Periodic Structures*

• Dispersion analysis is carried out in two-steps:

– Enforce the Floquet boundary condition between boundaries for a fixed real real Bloch wavevector kp.

– Sample the electric field inside the periodic structure; determine resonant frequencies kp(ω), from the Fourier transform.

*A. Taflove, “Computational Electrodynamics: The FDTD method”, ArtechHouse, 1995

Radiation from NRI-TLstructures

69

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Motivation: LeakyMotivation: Leaky--Wave StructuresWave Structures

• Leaky-Wave Antennas: – Periodic structures supporting leaky-wave

radiation.– Renewed interest stemming from recent

advances in metamaterials.

z

0z k<β

x

LWAβx

k0

Air

θ

0sin k

xβθ =

Fast (leaky) waves

Forward Endfire: βx=k0

Broadside: βx=0

emerging at angle θ,

z

|βx|/k0 <1:

Radiation from NRI-TLstructures

70

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Problem StatementProblem Statement

• FDTD analysis of LWAs relies on the simulation of truncated periodic structures, long enough to achieve convergence of α, β (computationally expensive).

• Is it possible to use periodic FDTD analysis to extract the attenuation constant of leaky-wave/lossy structures (complex Bloch wavevector) ?

Radiation from NRI-TLstructures

71

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Example: 2DExample: 2D--NRINRI--TL Medium*TL Medium*

*G.V. Eleftheriades et al, IEEE Trans. MTT, vol. 50, Dec. 2002.

Radiation from NRI-TLstructures

72

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Zo 141 Ohm

Lo 5.6 H

Co 1 pF

Stop Band

Light cone (β=ko)

FastFast

WaveWave

RegionRegion

FDTD Results from: T. Kokkinos, C.D. Sarris, G.V. Eleftheriades, IEEE Trans. MTT, vol. 53, Apr. 2005

Example: 2DExample: 2D--NRINRI--TL Medium (contTL Medium (cont--d)d)

73

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Periodic FDTD Analysis of 2DPeriodic FDTD Analysis of 2D--TL NRI MediumTL NRI Medium

[4] Harms et al, IEEE Trans. AP, vol. 42, Sept. 1994.

• Phase progression between periodic boundaries enforced via the “sine-cosine” method*.

Propagation along x-axis (kxd=λ/2)

x-axisy-axisRadiation

from NRI-TLstructures

74

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

FDTD Field Resonances for Slow/Fast WavesFDTD Field Resonances for Slow/Fast Waves

Slow wave: Perfect resonance

Fast wave: Decaying resonance

• In both cases, the real part of the Bloch wavenumber was enforced.

Light cone

0k=β

Fast Waveregion

SlowWaveregion

SlowWaveregion

Radiation from NRI-TLstructures

75

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Periodic FDTD Modeling of LeakyPeriodic FDTD Modeling of Leaky--Wave StructuresWave Structures

• Question:

– Does the enforcement of the phase progression of a wave within a periodic FDTD mesh still allow for the extraction of the attenuation constant ?

• Answer: Inspection of the implementation of PBCs in FDTD.Radiation

from NRI-TLstructures

76

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Implementation of PBCs in FDTDImplementation of PBCs in FDTD

unit cell

2D-Yee cell

Radiation from NRI-TLstructures

77

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Implementation of PBCs in FDTDImplementation of PBCs in FDTD

2D-Yee cell

unit cell

Radiation from NRI-TLstructures

78

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Implementation of PBCs in FDTDImplementation of PBCs in FDTD

2D-Yee cell

unit cell

Radiation from NRI-TLstructures

79

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Implementation of PBCs in FDTDImplementation of PBCs in FDTD

2D-Yee cell

PBC updates are

bi-directional !

unit cell

Radiation from NRI-TLstructures

80

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Periodic Lossy/Radiating Structure

Periodic FDTD Modeling of LWPeriodic FDTD Modeling of LW--Structures, revisitedStructures, revisited

Unit Cell

Radiation from NRI-TLstructures

81

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Periodic FDTD Modeling: Conclusion Periodic FDTD Modeling: Conclusion

• The enforcement of a PBC involving a real Bloch wavenumber β in the FDTD through the sine-cosine method does notnot in generalin general result in:

• Numerical investigation and analytical inspection of the method shows that the condition enforced is:

•• Spatial attenuation can be calculated in a PBCSpatial attenuation can be calculated in a PBC--terminated terminated unit cellunit cell.

dje)r(E)zdr(E β−=+

d)r(E)zdr(E β−=∠−+∠

Radiation from NRI-TLstructures

82

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Attenuation Constant: CalculationAttenuation Constant: Calculation• Complex propagation constant can be calculated

using two samples of the fields E(zi,t) and E(zj,t), assuming:

))t,z(E())t,z(E(ln

zzj)(

j

i

ij FF

−=ωγ

Distance between the sampling points should be close to one period of the structure.

)(Re)( ωγωβ =)(Im)( ωγω =a

))zz()j(exp())t,z(E())t,z(E( ijij −

γ

β+α−=43421

FF

Radiation from NRI-TLstructures

83

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

MetalMetal--StripStrip--Loaded Dielectric LWA*Loaded Dielectric LWA*

Operating Freq. 80 GHz

Periodicity d 2.5 mm

Strip length 1.25 mm

Strip width w 3 mm

* K. L. Klohn et al, IEEE Trans. MTT, vol. 26, Oct. 1978

Unit cell

• Metal strips are used as perturbations on a dielectric waveguide

• Operated at microwaves.• The n=-1 spatial harmonic is

radiated

Radiation from NRI-TLstructures

84

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

MetalMetal--StripStrip--Loaded Dielectric LWALoaded Dielectric LWAPeriodic FDTD Analysis vs. Ansoft HFSS (truncated structure)

Radiation from NRI-TLstructures

85

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Leaky CPWLeaky CPW--based slot antenna arrays*based slot antenna arrays*

Operating Freq. 30 GHz

Periodicity d 2.093 mm

• Capacitively loaded CPW lines.• Operated at millimeter-wave

region.• Radiates its fundamental spatial

harmonic (n=0).

Computational Domain

*A.Grbic, G.V. Eleftheriades, IEEE Trans. AP, vol. 50, Nov. 2002

Radiation from NRI-TLstructures

86

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Leaky CPWLeaky CPW--based slot antenna arraysbased slot antenna arrays

• Dispersion Analysis: β/ko = 0.72 at 30 GHz

Periodic FDTD vs. Agilent Momentum (truncated structure)

Radiation from NRI-TLstructures

87

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Complex Propagation Constant of 2DComplex Propagation Constant of 2D--NRINRI--TL based TL based LWAsLWAs

Open Stop Band

Stop Band

Stop Band

Zo 141 Ohm

Lo 5.6 nH

Co 1 pF

Dispersion Diagram ( β < ko)

Attenuation Constant ( α/ko)

When β→ko fast waves are coupled to non-radiating

surface waves.

Radiation from NRI-TLstructures

88

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Backward Radiation Patterns

Directivity

Complex Propagation Constant of 2DComplex Propagation Constant of 2D--NRINRI--TL based TL based LWAsLWAs (cont(cont--d)d)

Open Stop Band

Forward Radiation Patterns

Radiation from NRI-TLstructures

89

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Zo 141 Ohm

Lo 10 nH

Co 1 pF

Dispersion Diagram ( β < ko)

Attenuation Constant ( α/ko)

o

oo

CLZ

=2

Complex Propagation Constant of 2DComplex Propagation Constant of 2D--NRINRI--TL based TL based LWAsLWAs (cont(cont--d)d)

Closed Stop Band*

*Condition for closed stop-band discussed, derived in: G.V. Eleftheriades et al, IEEE Trans. MTT, vol. 50, Dec. 2002, eq. 29.

Radiation from NRI-TLstructures

90

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Complex Propagation Constant of 2DComplex Propagation Constant of 2D--NRINRI--TL based TL based LWAsLWAs (cont(cont--d)d)

Closed Stop Band

Backward Radiation Patterns Forward Radiation Patterns

Directivity

Radiation from NRI-TLstructures

91

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

SynopsisSynopsisOpen Stop Band

Closed Stop Band

DirectivityRadiation Patterns

Radiation Patterns DirectivityRadiation

from NRI-TLstructures

92

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Equivalent Circuit for 2DEquivalent Circuit for 2D--NRINRI--TL TL LWAsLWAs

Equivalent Circuit for NRI-TL (operating outside the light coneoperating outside the light cone)

• Propagation constants can be calculated via the periodic analysis of each circuit.

• Values of Lo,Co, Lx, Cx are known (loading elements, hosting TL).• Rrad = unknown.

Equivalent Circuit for NRI-TL (operating within the light coneoperating within the light cone)

Radiation from NRI-TLstructures

93

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Equivalent Circuit for 2DEquivalent Circuit for 2D--NRINRI--TL TL LWAsLWAs

Attenuation Constant ( α/ko)

Closed Stop Band

Dispersion Diagram ( β < ko)

•Rrad=2950 Ohms•Calculated by fitting the FDTD results with the equivalent circuit dispersion curve.

Radiation from NRI-TLstructures

94

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

ConclusionsConclusionsAMR-FDTD is a technique that implementsmultiple, dynamically/adaptively generated subgrids in the three-dimensional FDTD.

The mesh generation in AMR-FDTD is a self-adaptive process, based on pre-defined accuracy parameters (CAD-oriented feature).

Microwave/optical applications have shown the excellent computational performance of AMR-FDTD.

Conclusion

95

Dynamic Dynamic AMRAMR--FDTD:FDTD:

MethodMethod

Dynamic Dynamic AMRAMR--FDTD:FDTD:

ExamplesExamples

Wireless Wireless Channel Channel ModelingModeling

GPUGPU--basedbased

TimeTime--DomainDomain

ModelingModeling

Radiation Radiation from NRIfrom NRI--TLTL

structuresstructures

ConclusionConclusion

IntroIntro

Conclusions (contConclusions (cont--d)d)For indoor wireless problem, the feasibility of the S-MRTD technique was demonstrated.

S-MRTD combined with GPU programming delivers a fast modeling tool, further enhanced by the rapid advances in GPU technology.

Fast periodic simulations of radiating structures were achieved with a periodic FDTD technique.

Conclusion