Two Far-field Approximation Based Fast IPO Methods for Scattering Analysis of Large Objects
Article information
Abstract
To reduce the numerical complexity of the conventional iterative physical optics algorithm to compute the scattering by a large scatterer, we propose two fast algorithms to accelerate the far-field interaction between the facets located within the far-field region for each facet. The interaction is approximated using the far-field approximation based on the octree grouping of the facets. The first algorithm considers the exact lit and shadow check between two facets, which corresponds to the physical optics approximation. The second algorithm mitigates the burden of the exact shadow check by checking at the leaf box, which can reduce the numerical complexity further. The accuracy and computation time of the two algorithms are examined for several objects and compared to those of the conventional iterative physical optics algorithm scheme and the multilevel fast multipole method.
I. Introduction
The radar cross section (RCS) analysis for large-scale objects such as aircraft, ships, etc. is essential in electromagnetic scattering problems. Accurate and efficient computation of the RCS of electrically large targets poses significant challenges due to the substantial computational burden required to fully calculate the electromagnetic interactions over a widely distributed object.
Full-wave methods, such as the method of moments (MoM) and its accelerated variants like multilevel fast multipole method (MLFMM), can offer highly accurate solutions for scattering problems, but demand significant computational resources. Consequently, this can limit their application to electrically large scatterers [1].
Iterative physical optics (IPO) is a high-frequency approximation (HFA) technique that predicts the surface current based on the physical optics (PO) approximation [2]. Initially developed for analyzing open-ended cavities made of perfect electric conductor (PEC) materials, the IPO method was later extended to impedance media satisfying the impedance boundary conditions (IBC) [2, 3]. Beyond cavities, the IPO scheme has been applied to RCS prediction for realistic large objects like tanks, aircraft, and rough ocean surfaces [4–6]. This method offers significantly faster computation than full-wave solvers while maintaining sufficient accuracy.
To efficiently analyze electromagnetic scattering from large structures, the fast far-field approximation (FaFFA) technique has been introduced [7]. FaFFA, similar in principle to the fast multipole method (FMM), reduces computational complexity down to O(N1.33) by approximating the interactions between spatially distinct sub-scatterers. In this approach, well-separated surface facets are grouped, and the interactions are computed between these groups rather than individual elements. The method was later extended to the IPO as demonstrated in [8], where it significantly accelerates the iterative process by approximating group interactions. Furthermore, in [9], a hybrid acceleration technique IPO-MLFMM was proposed, employing multilevel grouping and the addition theorem, leading to even greater computational efficiency for large-scale problems. Subsequently, various acceleration techniques based on domain decomposition were proposed in [10, 11], enabling more efficient scattering analysis of electrically large structures.
While the FaFFA scheme significantly accelerates IPO computation, it cannot exactly account for the lit/shadow check among facets, which is crucial for the PO approximation. This limitation can reduce the accuracy of the fast IPO scheme. To partially mitigate this negative effect, a simple shadow check scheme, widely used in various applications, is applied to the fast IPO scheme in [12]. Furthermore, acceleration through hardware advancements has been actively researched. Some approaches parallelize the iterative current update process on CPU-GPU heterogeneous platforms using frameworks like compute unified device architecture (CUDA) [13]. Others have focused on accelerating the primary bottleneck of determining shadow and lit regions by using dedicated ray-tracing engines like NVIDIA OptiX, which reduces the preprocessing time required for visibility checks [14].
First, we propose a fast IPO scheme that incorporates a full shadow check based on the far-field approximation and the MLFMM acceleration algorithm. To further enhance its speed, we introduce a second method that allows a small, controlled error in the shadow check, which further accelerates the proposed fast IPO scheme. Section II details the proposed scheme, while Section III presents several numerical examples. The convention of ejwt is assumed and suppressed throughout the paper.
II. Fast Iterative Physical Optics
1. Conventional IPO Update Equation
The surface of the scatterer is discretized into many small flat triangular facets. The surface is assumed to consist of an inhomogeneous impedance material, whose normalized surface impedance is
The MFIE for the impedance object is written as
where
where
To accelerate the computation of Eq. (1), the facets are grouped and formed into the octree, and the interactions among the facets in cubes are approximated by the far-field approximation. The near-field interaction between the facets is directly calculated using Eq. (2) like the conventional IPO method.
2. Far-field Approximation between Cubes
Fig. 1 shows the coordinates of two cubes located in the far-field region relative to each other, whose length is Lc and Lc’, respectively. First, the far-field criteria for the cubes are derived.
Using the Taylor expansion,
where
where λ0 is the free-space wavelength. Interactions that do not meet this far-field criterion are treated as near-field and are calculated rigorously using Eq. (2).
When two cubes are in the far-field region, the scalar Green’s function, e−jk0R/4πR can be approximated using the following approximations: for the phase term,
Eq. (6) can be expressed in terms of the multiplication of three terms: the translation term, e−jk0Rcc’/4πRcc’, the disaggregation term,
For the PO method, only the radiated field from facets located within the lit region of a facet is considered. Thus, two acceleration algorithms can be formulated, denoted as fast far-field IPO algorithm 1 (FF-IPO1) and FF-IPO2. The FF-IPO1 can exactly consider the lit and shadow condition between the facets like the conventional IPO scheme. On the other hand, the FF-IPO2 consider the lit and shadow condition between the leaf boxes, which is the smallest box in the octree. It can allow small errors but can improve numerical efficiency.
3. FF-IPO1
When the far-field condition is satisfied, Eq. (1) can be simplified by applying Eqs. (6) and (7) to Eq. (1). First, the required spatial derivatives of the scalar Green’s function in Eq. (1) can be asymptotically approximated as
Fig. 2 shows the overall procedure of the accelerated computation for the FF-IPO1, which is very similar to MLFMM, except for the inexact computation of the Green’s function. The leaf box containing the observation facets is denoted as cq, where the subscript q is the box index. First, determine the largest box, denoted as
where sfp is the surface of the p-th facet. Nl is the total number of leaf boxes. Nf(q) is the number of facets contained within the q-th leaf box (cq). For the q1-th facet in cq,
4. FF-IPO2
In the computation of FF-IPO1, the filtering process for facets located within the shadow region can be somewhat cumbersome. Since the leaf box typically contains a very small number of facets (e.g. fewer than ten), performing the shadow check only at the leaf-box-to-leaf-box level introduces an approximation error. However, this error is often sufficiently small to be acceptable. The box-wise shadow check offers a significant advantage: the aggregation for each leaf box can be calculated once and then reused for all far-field interactions, which drastically reduces the numerical complexity. This constitutes the main difference between FF-IPO1 and FF-IPO2. Apart from this box-wise approach, the overall procedure of FF-IPO2 is otherwise identical to that of FF-IPO1. The updating equation, Eq. (9) can be expressed as
where
III. Numerical Results
To validate the far-field approximation of the dyadic Green’s function, Eq. (8), the source cube, c’ with a side length of 4λ0, is centered at the origin. A point source is located at one of the cube’s vertices, specifically at (2λ0, 2λ0, −2λ0). When considering a point observation (
It is observed that the far-field approximation is very accurate, and its accuracy may increase for greater distances between the box center and the observation point. Consequently, the proposed far-field interaction calculation scheme is very accurate. The phase comparison for the double curl equation shows behavior very similar to that for the single curl case shown in Fig. 4(b), which is omitted.
To evaluate the performance of the two proposed FF-IPO algorithms, three large and complicated objects are considered, which have different dimensions and operating frequency. To assess computational efficiency and accuracy, the scattering from the objects is computed using the conventional IPO, the proposed FF-IPO schemes, and the commercial FEKO MLFMM. The iterative process for three IPO schemes is terminated by the same stopping criterion, detailed in [4]. The iterations stop when the residual error falls below a tolerance of 10−3 or when the error begins to diverge.
The accuracy is then quantified using the normalized root mean squared error (NRMSE), defined as
where Nσ is the number of observation points. σMLFMM and σIPO are the RCS values, computed by the MLFMM and IPO, respectively. The subscripts, “max” and “min’ denote the maximum and minimum values of σMLFMM.
The objects considered are a fighter aircraft, an unmanned aerial vehicle (UAV) [15], and a missile. Only the missile is positioned above two rough ocean surfaces. The fighter comprises three distinct materials: a dielectric radome, dielectric canopy, and an impedance body. The dielectric radome and canopy are modeled as an impedance surface with an appropriate surface impedance as ηr = 0.53 [16] and ηr = 0.71 + j0.1 [17] for the radome and canopy, respectively. The body’s normalized surface impedance (η) is 0.38 – j0.06 [17]. The UAV consists of a homogeneous material with a PEC substrate coated with a laminate radar-absorbing structure (RAS). The RAS is a nonmagnetic dielectric layer, whose thickness and relative permittivity are given by t = 0.02λ0 and ɛr = 6.6 – j2.8, respectively. Consequently, the normalized impedance of the UAV can be approximated as
Fig. 5 compares the RCS results for three different targets, which are computed by the conventional IPO, two proposed FF-IPO schemes and the MLFMM as a reference solution. The NRMSE for the conventional IPO, FF-IPO1, and FF-IPO2 are given by 17.12 m2, 17.12 m2, and 17.19 m2, respectively. As expected, the error for the FF-IPO2 shows a slight increase. In Fig. 5(a) and 5(b), the error between FF-IPO1 and FF-IPO2 can be observed in the low RCS intervals, but the difference is negligible.
RCS comparison for three targets: (a) fighter, (b) UAV, and (c) missile above rough ocean surface (160λ0×160λ0×76λ0).
Fig. 6 illustrates the source of this error. It occurs because the group-wise visibility check in the FF-IPO2 scheme erroneously includes some shadowed facets (light gray) in the lit region interaction. Since the leaf box contains a small number of facets, the number of these misclassified facets is very small compared to the total number of facets. Fig. 7 compares the computational cost of the two proposed FF-IPO schemes. The cost is computed as the ratio of the FF-IPO scheme’s cost to that of the conventional IPO scheme. As expected, both computation time and the number of multiplications decrease with increasing the far-ratio in Table 1. The FF-IPO2 scheme demonstrates superior efficiency compared to the FF-IPO1 scheme. For the larger ocean case, for example, the simulation time of FF-IPO1 and FF-IPO2 is reduced to approximately 33% and 25%, respectively.
Example of visibility misclassification in the FF-IPO2, showing the exact lit facets (dark gray) and erroneously classified facets (light gray).
Finally, the asymptotic numerical complexity (number of multiplication) is estimated as the number of facets (N) increases. Fig. 8 shows this complexity: the conventional IPO exhibits a complexity of O(N2), while those of FF-IPO1 and FF-IPO2 are reduced to O(N1.8) and O(N1.33), respectively. Here, O(·) is the big O symbol. The numerical complexity of the (dis)aggregation and translation is summarized in Table 2 for the simulation geometry. The translation is the main bottleneck of the FF-IPO2. To compute the complexity in Table 2, only the far-field case is considered, so that the number off multiplications for the aggregation is less than O(N).
IV. Conclusion
This paper introduces two FF-IPO algorithms to accelerate the RCS analysis of large-scale targets. Both algorithms apply the far-field approximation to reduce the computational burden of the conventional IPO scheme. Specifically, the FF-IPO1 scheme precisely performs the lit/shadow check on a facet-wise, while the FF-IPO2 scheme conducts this check solely between leaf boxes. Although this approximation in FF-IPO2 may introduce a slight error, it significantly enhances numerical efficiency. For three complicated objects, the proposed FF-IPO schemes are compared against the MLFMM. As expected, the accuracy of the FF-IPO1 scheme is maintained accuracy without degradation. The FF-IPO2 scheme’s accuracy was only slightly reduced, compared to that of the conventional IPO scheme. However, the numerical burden is drastically lowered, exhibiting a numerical complexity of O(N1.33). For instance, the maximum simulation time reduction for the larger ocean simulation is 33% for FF-IPO1 and 25% for FF-IPO2. Further reductions are attainable for larger scatterers, as the far-field interactions increase rapidly with size. Consequently, these fast IPO schemes prove highly effective for analyzing scattering from large objects.
Notes
This work was supported by the Agency For Defense Development Grant Funded by the Korean Government (UD230016JD).
References
Biography
![]()
Jeong-Un Yoo, https://orcid.org/0000-0003-2363-296X received the B.S. degree in electronic engineering from Gangneung-Wonju National University, Gangneung, Korea, in 2017. He is currently working toward a Ph.D. degree in electronic engineering at Inha University, Incheon, Korea. His research interests include electromagnetic numerical analysis.
![]()
Il-Suek Koh, https://orcid.org/0000-0003-0014-2466 received B.S. and M.S. degrees in electronics engineering from Yonsei University, Seoul, Korea, in 1992 and 1994, respectively, and a Ph.D. from the University of Michigan, Ann Arbor, MI, USA, in 2002. In 1994, he joined LG Electronics Ltd., Seoul, as a research engineer. He is currently a professor at Inha University, Incheon, Korea. His research interests include wireless communication channel modeling and numerical and analytical methods for electromagnetic fields.
