Download - Academy Publisher
Download - Academy Publisher
Download - Academy Publisher
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
ISBN 978-952-5726-09-1 (Print)<br />
Proceedings of the Second International Symposium on Networking and Network Security (ISNNS ’10)<br />
Jinggangshan, P. R. China, 2-4, April. 2010, pp. 129-132<br />
Modeling Seismic Wave Propagation Using<br />
Graphics Processor Units (GPU)<br />
Zhangang Wang 1 , Suping Peng 1 , and Tao Liu 2<br />
1 State Key Laboratory of Coal Resources and Mine Safety, China University of Mining and Technology, Beijing, China<br />
Email: millwzg@163.com<br />
2 School of Earth and Space Sciences, Peking University, Beijing, China<br />
Email: liuluot@126.com<br />
Abstract—The main drawback of the seismic modeling in<br />
2D viscoelastic media on a single PC is that simulations<br />
with large gridsizes require a tremendous amount of<br />
floating point calculations. To improve computation<br />
speedup, a graphic processing units (GPUs) accelerated<br />
method was proposed using the staggered-grid finite<br />
difference (FD) method. The geophysical model is<br />
decomposed into subdomains for PML absorbing<br />
conditions. The vertex and fragment processing are fully<br />
used to solve FD schemes in parallel and the latest updated<br />
frames are swapped in Framebuffer Object (FBO)<br />
attachments as inputs for the next simulation step. The<br />
simulation program running on modern PCs provides<br />
significant speedup over a CPU implementation, which<br />
makes it possible to simulate realtime complex seismic<br />
propagation in high resolution of 2048*2048 gridsizes on<br />
low-cost PCs.<br />
Index Terms—seismic, wave propagation, finite difference,<br />
viscoelastic media, model, GPU<br />
I. INTRODUCTION<br />
The main drawback of the FD method is that<br />
simulations with large model spaces or long<br />
nonsinusoidal waveforms can require a tremendous<br />
amount of floating point calculations and run times. In<br />
recent years the spatial parallelism on high-performance<br />
PC clusters makes it a promising method for seismic<br />
numerical computing in realistic (complex) media [1, 2].<br />
Modern GPUs have now become ubiquitous in<br />
desktop computers and offer an excellent cost-toperformance-ratio,<br />
exceeding the performance of general<br />
purpose CPU by many times, due to their powerful and<br />
fully programmable parallel processing architectures. In<br />
the past few years, programmability of GPU and<br />
increased floating point precision has allowed GPU to<br />
perform general purpose computations [3, 4].<br />
In this article, we have proposed a new<br />
implementation under the current GPU architecture to<br />
improve the real-time FD simulation of wave<br />
propagation in viscoelastic media, including domain<br />
decomposition and processing techniques.<br />
II. THEORY<br />
Early researches on FDM for elastic wave modeling in<br />
complex media gradually formulated finite difference<br />
schemes based on a system of high-order coupled elastic<br />
equations [5-8]. In such studies, however, the earth’s<br />
viscosity has been ignored and synthetic seismograms<br />
fail to model attenuation and dispersion of seismic waves.<br />
Day and Minster [9] made the first attempt to incorporate<br />
anelasticity into a 2-D time-domain modeling methods<br />
by applying a Pade approximant method. An efficient<br />
approach [10] was proposed based on the rheological<br />
model called “generalized standard linear solid” (GSLS),<br />
which was shown to explain experimental observations<br />
of wave propagation through earth materials [11]. Then a<br />
staggered grid finite difference technique was proposed<br />
based on the GSLS to model the propagation of seismic<br />
waves in 2D/3D viscoelastic media [12], for staggeredgrid<br />
operators are more accurate than standard grid to<br />
perform first derivatives for high frequencies close to the<br />
Nyquist limit [8].<br />
A. Staggered-grid Finite difference formulation<br />
The first-order velocity-stress equations of viscoelastic<br />
wave propagation are given by<br />
s<br />
⎧∂σ<br />
xy τ v<br />
ε<br />
∂v<br />
∂<br />
x y<br />
⎪ = μ ( + ) + γ<br />
xy<br />
⎪ ∂t τσ<br />
∂y ∂x<br />
⎪<br />
p<br />
s<br />
∂σxx<br />
τ v<br />
x y<br />
v<br />
ε<br />
∂v<br />
∂ τ ∂<br />
ε y<br />
⎪ = π ( + ) −2μ ⋅ + γxx<br />
⎪ ∂t τσ<br />
∂x ∂y τσ<br />
∂y<br />
⎪<br />
p<br />
s<br />
⎪<br />
∂σ yy τ v v<br />
ε<br />
∂ ∂<br />
x y τε<br />
∂vx<br />
= π ( + ) −2μ ⋅ + γ<br />
yy<br />
⎪ ∂t τσ<br />
∂x ∂y τσ<br />
∂x<br />
⎪<br />
p<br />
s<br />
⎪∂γ xx<br />
1 τ v vy<br />
vy<br />
[ (<br />
ε<br />
∂ ∂ 1)(<br />
x<br />
τ ∂<br />
xx<br />
) 2 (<br />
ε<br />
=− γ + π − + − μ − 1) ⋅ ]<br />
⎪ ∂t τσ τσ ∂x ∂y τσ<br />
∂y<br />
⎨<br />
p<br />
s<br />
⎪∂γ<br />
yy 1 τ [<br />
yy (<br />
ε<br />
∂v<br />
∂v<br />
x y τε<br />
∂vx<br />
⎪<br />
=− γ + π −1)( + ) −2 μ( −1) ⋅ ]<br />
∂t<br />
τσ<br />
τσ<br />
∂x ∂y τ<br />
σ<br />
∂x<br />
⎪<br />
⎪∂<br />
s<br />
γ<br />
xy 1 τ v vy<br />
[ (<br />
ε ∂ 1)(<br />
x ∂<br />
⎪ =− γ<br />
xy<br />
+ μ − + )]<br />
⎪ ∂ t τσ<br />
τσ<br />
∂ y ∂ x<br />
⎪∂vx<br />
1 ∂σ<br />
∂σ<br />
xy<br />
[<br />
xx<br />
⎪ = + + fx<br />
]<br />
⎪ ∂t ρ ∂x ∂y<br />
⎪ ∂ vy 1 ∂ σxy ∂ σ<br />
⎪<br />
yy<br />
= [ + + f<br />
y ]<br />
⎪⎩<br />
∂t ρ ∂x ∂y<br />
The moduli π, μ can be expressed as<br />
π = v<br />
ρR<br />
2 2<br />
p0<br />
1 ,<br />
iw0τ<br />
σ p<br />
+ τε<br />
+ iw0τ<br />
σ<br />
1 1<br />
μ = v<br />
ρR<br />
2 2<br />
s0<br />
1<br />
(1)<br />
(2)<br />
iw0τ<br />
σ s<br />
+ τ<br />
ε<br />
+ iw0τ<br />
σ<br />
1 1<br />
Where v P0 , v S0 denote the phase velocity at the centre<br />
frequency of the source(w 0 ) for P- and S-waves,<br />
respectively. The symbol R denotes the real part of the<br />
complex variable. The constants of stress relaxation<br />
times for both P-and S-waves, can be calculated by<br />
quality factor Q and angular frequency w.<br />
© 2010 ACADEMY PUBLISHER<br />
AP-PROC-CS-10CN006<br />
129