Forum for Electromagnetic Research Methods and Application Technologies (FERMAT) Title (Font: Times New Roman, Size: 24) Name (1) and Name (2) (Font: Times New Roman, Size: 12) (1) Affiliation (Font: Times New Roman, Size: 12) (Email: xxxxx@xxxxx ) (Font: Times New Roman, Size: 9) (2) Affiliation (Email: xxxxx@xxxxx) reflectors for frequency reuse systems, or for frequency selective radomes. The conventional Method of Moments (MoM) [8], [9] is often applied for the solution of moderatesize problems; however, it becomes numerically inefficient for electrically large problems owing to the large sizes of the associated system matrices. Among the direct solvers, the Characteristic Basis Function Method (CBFM) has been extensively used for reducing the size of the impedance matrix system of equations both in the context of printed antennas and scattering problems [10]-[12]. It is based on the generation of physics-based, Characteristic Basis Functions (CBF) to model the current distributions induced in the subdomains. The reduced version of the system matrix arising in the CBFM consumes a relatively small CPU time compared to the Lower Upper (LU) factorization of the conventional MoM matrix. Another approach to ease the solution of large scale problems is the application of hybrid methods, where the MoM is combined with asymptotic techniques Error! Reference source not found.. In this context, a matrix equation is only derived for the MoM region, resulting in a sizable reduction of the number of unknowns. However, the accuracy of such hybrid techniques is often limited, and there is no simple way to improve it. In this paper, a novel method, which combines the CBFM with the Spectral Rotation (SR) approach Error! Reference source not found. is proposed. The technique is based on computing the integrals pertaining to the mutual-interactions in the spectral domain, while the self-interactions are analyzed in the spatial domain Error! Reference source not found.. Furthermore, the computation of the reaction integrals in the spectral domain is speeded up via the SR method, which circumvents the need to deal with threedimensional Fourier transforms for non-planar geometries. The time required to fill the reduced matrix is considerably reduced in comparison to the conventional CBFM when the SR method is applied, but without compromising the accuracy of the final result. The paper is organized as follows. In Section II the CBFM algorithm is briefly reviewed. Section III presents some of the basic concepts and numerical considerations on the spectral rotation approach. Section IV presents some numerical examples which demonstrate the accuracy and the numerical efficiency of the proposed technique. Finally, some concluding remarks are provided in Section V. Abstract—An efficient Characteristic Basis Function (CBF)-based method is proposed to analyze conformal Frequency Selective Surfaces (FSS) that are not amenable to analysis by using conventional numerical methods typically used to model infinite, planar and doubly-periodic FSSs. The technique begins by employing the CBFs to describe the currents induced on the elements. The reaction integrals needed to derive the reduced matrix elements are computed either in the spatial domain, or spectral domain, depending on the separation distance between the blocks, so as to make the process numerically efficient. The spectral domain integrals are evaluated by making use of the Spectral Rotation (SR) on the spectra of the CBFs to alleviate the computational burden to be associated with full Three Dimensional (3-D) Fourier transform which is required in the conventional spectral domain approach applied to non-planar geometries. Numerical results from the new method have good agreement with the fully-spatial Characteristic Basis Function Method (CBFM) and the conventional Method of Moments (MoM). However, the required Central Processing Unit (CPU) time is greatly reduced. Index Terms—Method of Moments (MoM), Characteristic Basis Function Method (CBFM), Spectral domain analysis, Spectral Rotation (SR). I. INTRODUCTION Periodic structures have been widely used and investigated in the past as they find widespread applications as Frequency-Selective Surfaces (FSS) [1]-[3], Electromagnetic or Photonic Bandgap (EBG or PBG) materials [4] and metamaterials. The periodic structures are conventionally modeled as infinite doubly periodic arrays and periodic boundary conditions are applied to a unit cell based on the Floquet’s theorem [5] or the periodic Green’s Function (GF) [6]. Even if the size of the array to be analyzed may be very large, those structures are, in practice, finite in size [7]; hence, an accurate full-wave analysis should appropriately consider the coupling between all the elements, leading to the solution of a large-scale problem. In this work we focus on the analysis of conformal FSS, that are widely employed to realize high performance frequency selective sub1 Forum for Electromagnetic Research Methods and Application Technologies (FERMAT) algorithms, whose calculation can be relatively timeconsuming. In the next section, we show how we can compute the reaction between two currents which reside on two planes whose orientation can be arbitrary, by using a 2-D technique. II. THE CHARACTERISTIC BASIS FUNCTION METHOD In the conventional CBFM, as a first step, the structure under analysis is partitioned into N sub-domains, or blocks (see Fig. 1). The induced currents in each block are solved for separately by illuminating the block with a set of K plane waves with both E- and H-polarizations. The solutions thus obtained are then processed by using the Singular Value Decomposition (SVD) algorithm, to automatically down select and retain the M most linearly independent high-level basis functions for that block, denoted as CBFs. III. SPECTRAL ROTATION APPROACH Let us consider the scenario depicted in Fig. 2, in which a source CBF is defined over support #1 and a test CBF resides over support #2. Fig. 1. Partitioning of a conformal FSS into N blocks (top view). Fig. 2. Support #1 defined over the xy-plane and support #2 defined over the x'y'-plane. Also shown support #1 rotated by an angle r and with respect to the xy-plane. The second step to be performed is the generation of the (M·N) × (M·N) reduced-size matrix by Galerkin testing the CBFs, which are used both as basis and testing functions. Finally, the current distribution induced on the entire structure is expressed as a linear combination of the CBFs, weighted by the solution vector resulting from the inversion of the reduced system of equations Error! Reference source not found.. The generic element of the reduced matrix, whose calculation comprises a reaction integral, can be computed in the spatial domain by using: Z iRED J i (r ) J j (r ') G (r r ') d r ' d r ,j T S where To compute the fields produced by the CBFi which is defined over support #1 at location #2, in spectral domain, we ~ first express the spectrum of J #1,i (k x , k y ) with respect to the spectral coordinates (k x' , k y' ) of x'y'z' system: ~ ~ J #1,i (k x' , k y' ) J #1,i (k x , k y ) is multiplied by an exponential term which incorporates the projection of the vector distance d between the origins of the two supports, over the test system. This step translates the spectrum from the source (#1) to the test position (#2). The final form of the reaction integral in spectral domain reads: spatial CBF while the primed and unprimed notations refer to the source and test coordinate system, respectively. By invoking Parseval's theorem Error! Reference source not found., expression (1) can be rewritten in spectral domain, as: 1 2 2 J ~ ~ ~ i ( k x , k y ) G ( k x , k y ) J j ( k x , k y ) dk x dk y (3) where r and r are the angular differences between the xyz and x'y'z' reference systems. This first step produces a rotation of the spectrum of the source current at #1, such that it ~ becomes parallel to test support #2 (Fig. 2). Next, J #i (k x' , k y' ) (1) G is the spatial domain GF, Ji represents the ith Z iRED ,j k x k x' cos r cos r k y' cos r sinr k z' sin r k y k x' sinr k y' cos r (2) Z 2RED ,1 where G~ ( , ) denote the spectral domain Green's function, and J~ i is the Fourier transform of the ith CBF. 1 2 2 J ~ ~ # 2 ,i ( k x , k y ) G ( k x , k y ) ~ j k d xˆ k y d yˆ k z d zˆ J #1,i (k x , k y ) e x dk x dk y (4) ~ where G ( , ) denotes the spectrum of the dyadic GF and d For planar structures, such as thin plates, two-dimensional (2-D) Fast Fourier Transform (FFT)-based approaches can be successfully employed in order to reduce the memory and CPU time considerably. However, this applies only if the source and observation surfaces are parallel to each other Error! Reference source not found.; otherwise, it becomes necessary to switch to three-dimensional (3-D) FFT is the distance between the source and observation points. The computation of the integral in (4) can be carried out efficiently in the spectral domain for the mutual terms, since the rotated current becomes highly attenuated in the evanescent region of the spectrum. If we recall the 2 Forum for Electromagnetic Research Methods and Application Technologies (FERMAT) definition of the z-component of the wavenumber Error! Reference source not found.: In order to carry out an estimate of the accuracy of the algorithm, we define the Normalized Error (NE), as: k 2 (k 2 k 2 ) if k 2 k 2 k 2 (visible spectrum) 0 x y x y 0 kz = 2 2 2 2 2 2 j (k x k y ) k0 if k x k y k0 (evanescent spectrum) CBFM SR NE (%) 100 (5) note that kx in (3) is complex when k x2 k y2 k 02 , Fields / RCS ( Fields / RCS ( Fields / RCS ) CBFM 2 ) n CBFM 2 n (6) where superscripts 'CBFM-SR' and 'CBFM ' refer to the hybrid and the conventional CBFM, respectively. The normalized error is 1.2% and the Relative Time (RT), defined as the ratio between the CPU time taken by the conventional CBFM and the CBFM-SR is 2.17. because kz is purely imaginary. Thus, when performing the transform of the current, this produces attenuation in the evanescent part of the spectrum when sin(r) ≠ 0 in (3). Additional attenuation is introduced by the translation to be applied to j k d zˆ the spectrum of the field, defined by the term e z in (4). In the evanescent spectrum this exponential term reduces to e ( k x2 k y2 ) k02 d zˆ producing an attenuation related to d z . Let us consider a CBF distributed over a rectangular support with length L=/4 along x and width w = L/10 at 1 GHz. We assume their distribution is sinusoidal along the longitudinal dimension and constant along the transverse one. The behavior of the spectrum before and after a rotation (, ) = (3/8, /4) is reported in Fig. 3(a) and (b), respectively. Fig. 4. Geometry of 7×5 array of x-oriented PEC strips. (a) (b) Fig. 3. (a) Spectrum of bi-dimensional sinusoidal current distribution J and (b) Spectrum of J after a (, ) = (3/8, /4) rotation. It is apparent that the integration interval in the spectral domain is already significantly reduced with the use of the rotation operation only, and this fact can be exploited to obtain an accurate result for the integral with a relatively small number of spectral samples. The relative advantage also grows as the number of blocks N is increased, because the number of self-interactions to be computed when filling the reduced matrix is N while the number of mutualinteractions is N(N-1) Error! Reference source not found.. Fig. 5. Magnitude of scattered E-fields in the far-field region on the elevation plane, = 0, for TE and TM polarizations. IV. NUMERICAL RESULTS For the second example we analyze an array of 109 patch elements arranged as in Fig. 6. A plane wave is incident from the top with TMx - polarization at the operating frequency of 1 GHz. A total of 19620 sub-sectional basis functions are used to model the structure. The bistatic RCS on the elevation plane computed by using MoM, CBFM-SR and conventional CBFM is reported in Fig. 7. The relative time is 4.1 for this example while the normalized error is 2.1% for normal incidence, when we use 4 CBFs. In Fig. 8 we show the NE on the RCS when the incident angle inc is varied from 0 to 90o and we use 4 CBFs. The NE is below 6% over all incident angles. Table 1 shows the NE and the RT as a For the first example we analyze a 7×5 array of xdirected PEC strips wrapped around a cylindrical surface (Fig. 4). The dimensions of the strip are 120mm×30mm. A total of 1995 sub-sectional basis functions are employed to discretize the geometry at 1 GHz. Both TEy and TMx polarizations have been considered, with (, ) = (-, 0), and the magnitude of the scattered x- and z- components of the E-fields are plotted in Fig. 5 (two orders of magnitude lower Ey component is not shown). The agreement between CBFM-SR and CBFM is good, even for TE incidence, for which the scattered fields are smaller than in the TM case. 3 Forum for Electromagnetic Research Methods and Application Technologies (FERMAT) function of the number of employed CBFs. As it can be noted, the method is independent of the number of employed CBFs (the relative error remains constant) and, as expected, the time advantage grows as the number of CBFs is decreased. After SVD, we find 4 CBFs to be sufficient to model the current distribution (see Fig. 7). The CPU time necessary to fill the reduced matrix mutualblocks becomes negligible in comparison to that required to evaluate the self-block interactions when the problem size is very large. This, in turn, leads to a considerable time-saving in comparison to the conventional approach for large scale geometries. while retaining the accuracy of the results. The scattered fields are found to agree well with those obtained when the reduced matrix is generated conventionally. Furthermore, the time-savings of the proposed method grow as a function of the number of unknowns in the blocks. REFERENCES [1] B.A. Munk, Frequency selective surfaces: Theory and design, Wiley, New York, 2000. [2] R. Mittra, C. H. Chan and T. Cwik, “Techniques for Analyzing Frequency Selective Surfaces - A Review”, IEEE Proc., 76, 1593-1615 (1988). [3] C. H. Chan and R. Mittra, “On the Analysis of Frequency-Selective Surfaces Using Subdomain Basis Functions,” IEEE Trans. Antennas Propag., Vol. 38, No. 1, 40–50, Jan. 1990. [4] Y. Rahmat-Samii and H. Mosallaei, “Electromagnetic band-gap structures: Classification, characterization, and applications,” in Proc. Inst. Elect. Eng. Antennas Propagation, Manchester, U.K., Apr. 17–20, 2001. [5] R. E. Collin, Field Theory of Guided Waves, IEEE Press, New York, 1991. [6] K. Yasumoto and K. Yoshitomi, “Efficient calculation of lattice sums for free-space periodic Green’s function,” IEEE Trans. Antennas Propagat., vol. 47, pp. 1050–1055, June 1999. [7] T. A. Cwik and R. Mittra, "The effects of the truncation and curvature of periodic surfaces: A strip grating," IEEE Trans. Antennas Propag., vol. AP-36, no. 5, pp. 612-622, May 1988. [8] R. F. Harrington: Field Computation by Moment Method. The Macmillan Company, 1968. [9] A. F. Peterson, S. L. Ray and R. Mittra, Computational Methods for Electromagnetics. New York: IEEE Press, 1998. [10] V. Prakash and R. Mittra, “Characteristic basis function method: A new technique for efficient solution of method of moments matrix equations”, Microwave Opt. Technol. Lett., vol. 36, no. 2, pp. 95-100, Jan. 2003. [11] R. Mittra, G. Bianconi, C. Pelletti, K. Du, S. Genovesi and A. Monorchio, “A Computationally Efficient Technique for Prototyping Planar Antennas and Printed Circuits for Wireless Applications”, Proceedings of the IEEE, vol. 100, no. 7, pp. 2122-1231, July 2012. [12] J. Wei and Z. Nie, “A wide band Scattering Analysis of Conformal FSS by CBFM and SVD method”, AsiaPacific Microwave Conference (APMC) 2008. Fig. 6. Geometry of 109 PEC patch elements array. Fig. 7. Bistatic RCS on the elevation plane for TM x incidence. The results of MoM, CBFM and hybrid CBFM (CBFM-SR) are compared. Fig. 8. Normalized % error as a function of the incident theta angle inc. Photo V. CONCLUSIONS In this paper, a technique to efficiently analyze the problem of the curved FSS has been presented. By computing the reduced matrix mutual-interactions in the spectral domain, and by making use of the fully-2D spectral rotation method, a considerable time-saving is achieved 4 Bio