Acceleration of the Iterative Physical Optics Using Graphic Processing Unit. Abstract

Transcription

1 THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE. 21 Nov.; 2(11), ISSN (Print) ISSN X (Online) Acceleration of the Iterative Physical Optics Using Graphic Processing Unit 이용희 진희철 김경태 Yong-Hee Lee Huicheol Chin Kyung-Tae Kim 요약 RCS(Radar Cross Section) (Iterative Physical Optics: ). (Geometric Optics: GO) SBR(Shooting and Bouncing Rays). PO(Physical Optics),. GPU(Graphic Processing Unit), A-CR(Adaptive Iterative Physical Optics-Change Rate). Abstract This paper shows the acceleration of iterative physical optics() for radar cross section(rcs) by using two techniques effectively. For the analysis of the multiple reflection in the cavity, uses the near field method, unlike shooting and bouncing rays method which uses the geometric optics(go). However, it is still far slower than physical optics(po) and it is needed to accelerate the speed of for practical purpose. In order to address this problem, graphic processing unit(gpu) can be applied to reduce calculation time and adaptive iterative physical optics-change rate(a-cr) method is also applicable effectively to optimize iteration for acceleration of calculation. Key words:, CUDA, A-CR, RCS. 서론 RCS(Radar Cross Section). RCS RCS. PO(Physical Optics),.,,, PO. (Maxwell's equation) MM(Moment Method) FDTD(Finite Different Time Domain). (Department of Electrical Engineering, Pohang University of Science and Technology) Manuscript received August, 21 ; Revised November 11, 21 ; Accepted November 1, 21. (ID No ) Corresponding Author: Kyung-Tae Kim ( kkt@postech.ac.kr) 112 c Copyright The Korean Institute of Electromagnetic Engineering and Science. All Rights Reserved.

2 , λ,. SBR(Shooting and Bouncing Ray), PO [1]. (geometric optics),. SBR,. MM FDTD,, [2] [4]. PO, HRRP(High Resolution Range Profile) ISAR(Inverse Synthetic Aperture Radar), RCS. RCS GPU(Graphic Processing Unit) GPU A- CR(Adaptive Iterative Physical Optics-Change Rate),.. 본론 2-1 반복적물리광학법 () RCS 1 (open) ( ) (equivalent principle).. s E, H s i E, H i open ˆk Sa J ( r ') M ( r ') ˆn S c E, H ˆn J ( ) n r PEC 그림 1. Fig. 1. progress model of the cavity geometry. close (2) (1),. (2) (free space Green's function) SBR. (), n, n 1 (). (1) (). 2-2 GPU 를활용한가속화 11

3 THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE. vol. 2, no. 11, Nov. 21. GPU CPU(Central Processing Unit),,., N. GPU GPU (gain). CPU GPU.,,,,.,,,,. 2- []. GPU kernel, GPU kernel. 2 A-CR, 4 A-CR. kernel GPU kernel, 2 kernel.,, Incident Field A-CR New Aperture Currents GPU kernel EM Aperture Currents Wall Currents RCS Incident Field A-CR New Aperture Currents EM Aperture Currents Wall Currents RCS 그림 2. GPU - / Fig. 2. Diagram of using GPU - before/after. 표 1. Main kernel Table 1. Main kernel algorithm. Main_Kernel Algorithm <Pre Kernel> <Main Kernel> <Post Kernel> Input: I, current data for patches; G, geometry data for patches; N, the number of patches; Ouput: none, an (intermediate) I data stored in I; thread blockidx.x * blockdim.x + threadidx.x; //Start global memory read early from current of patches. initial_current I[thread]; for patch_number to N do induced_current I[patch_number]; if patch_number thread then induced_current induced_current + near_field_green_function(induced_current, G); end if end for I[thread] initial_current + induced_current; return ; GPU [2]. 1 2 main kernel pseudo code. 2 kernel, kernel main kernel. kernel source observer. GPU nvidia 27 CUDA. 114

4 2- A-CR을이용한반복횟수의자동계산.,,. CPU (ray tracing), [2],[],[7]..., [2]. (difference of current) (relative error),. ( )., (),, /. CPU, GPU [],[]. A-CR CPU, [7]. n (4), () n 1. () 1 % (),. 2,., 1., / 1 initial_current I[thread]; []. A-CR ( ), %., [] [8]. A-CR GPU. (4) () 11

5 THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE. vol. 2, no. 11, Nov 시뮬레이션조건. 시뮬레이션결과 CPU Intel(R) Core(TM) i7-2k GHz, RAM DDR 4 1. GB, GPU GeForce GTX 77. GPU, GPU A-CR. θ. 1. A-CR,, 그림. Fig.. target model., 1 [8]. -2, A-CR. -2 기법에따른시뮬레이션결과비교 2 RMSE(Root Mean Square Error). RMSE.,. RMSE 2.77 db, RMSE 1 %. A-CR RMSE. db,, 4. RMSE, MOM A-CR. db. GPU, 2 4. GPU, 2, 8, GPU 표 2. RMSE( : db) Table 2. Computing RMSE of the various techniques applied to (unit: db). Short cylindrical Short rectangular Long cylindrical Long rectangular θ-pol. -pol. θ-pol. -pol. θ-pol. -pol. θ-pol. -pol GPU GPU + A-CR

6 GPU + GPU + A-CR A-CR iteration (a) (θ-pol.) (a) Open-ended short cylindrical cavity(θ-pol.) GPU + GPU + A-CR A-CR iteration (b) ( -pol.) (b) Open-ended short cylindrical cavity( -pol.) GPU + GPU + A-CR A-CR iteration (c) (θ-pol.) (c) Short rectangular cavity(θ-pol.) GPU + GPU + A-CR A-CR iteration (d) ( -pol.) (d) Short rectangular cavity( -pol.) GPU + GPU + A-CR A-CR iteration (e) (θ-pol.) (e) Open-ended long cylindrical cavity(θ-pol.) GPU + GPU + A-CR A-CR iteration (f) ( -pol.) (f) Open-ended long cylindrical cavity( -pol.) GPU + GPU + A-CR A-CR iteration (g) (θ-pol.) (g) Open-ended long rectangular cavity(θ-pol.) GPU + GPU + A-CR A-CR iteration 그림 4. Fig. 4. Result graph of the various techniques applied to (h) ( -pol.) (h) Open-ended long rectangular cavity( -pol.) 1 12., GPU A-CR, GPU, 4., A-CR. short rectan- 117

7 THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE. vol. 2, no. 11, Nov. 21. 표. ( : ) Table. Computing performance of the various techniques applied to (unit : second) Short cylindrical Short rectangular Long cylindrical Long rectangular θ-pol. -pol. θ-pol. -pol. θ-pol. -pol. θ-pol. -pol , ,48.1 4,.81 4,.82 + GPU GPU + A-CR 표 4. ( : kbyte) Table 4. The usage of global memory(unit: kbyte). Difference of current A-CR Short cylindrical Short rectangular Long cylindrical Long rectangular gular-pol.,, 4 (d) A-CR. 4 A-CR, %.. 결론 GPU A-CR. CPU, A-CR 2 8., GPU A-CR. References [1] Hao Ling, Ri-Chee Chou, and Shung-Wu Lee, "Shooting and bouncing rays: calculating the RCS of an arbitrarily shaped cavity", IEEE Transactions on Antennas and Propagation, vol. 7. no. 2, Feb. 18. [2] Fernando Obelleiro-Basteiro, Jose Luis Rodriguez, and Robert J. Burkholder, "An iterative physical optics approach for analyzing the electromagnetic scattering by large open-ended cavities", IEEE Transactions on Antennas and Propagation, vol. 4. no. 4, apr. 1. [],,, " -MoM ",, 2(8), pp. 71-7, 2 8. [4],,, "(Iterative PO) Ka Von Karman Radome ",, 22(12), pp , [] David B. Kirk, Wen-mei W. Hwu, Programming Massively Parallel Processors: A Hands-on Approach, Elsevier, 21. [] Robert J. Burkholder, Tomas Lundin, "Forward-backward iterative physical optics algorithm for computing the RCS of open-ended cavities", IEEE Transactions on Antennas and Propagation, vol.. no. 2, 7-7, 2. [7] S. H. Choi, D. W. Soe, and N. H. Myung, "Scattering analysis of open-ended cavity with inner object", Jounal of Electromagnetic Waves and Applications, vol. 21. no. 12, , 27. [8] H. Chin, J. -H. Yeom, H. -T. Kim, and K. -T. Kim, "Im- 118

8 provement of iterative physical optics using previous information to guide initial guess", Jounal of Electromagnetic Waves and Applications, vol. 124, 47-48, : ( ) 2 : [ 주관심분야 ], RCS 14 2 : ( ) 1 2 : ( ) 1 2 : ( ) : 211 : 212 : /IR [ 주관심분야 ],,,, RCS 28 2 : ( ) : ( ) : ( ) 21 1 : [ 주관심분야 ], 11