Scribd Logo
Upload

Welcome to Scribd!

  • Theory and Application of Radiation Boundary Operators - Moore T.G., Blaschak G.B, Taflove A, Kriegsmann G. A - 1988

    Uploaded by

    samouille6666
    5 views16 pages
    Theory and Application of Radiation Boundary Operators - Moore T.G., Blaschak G.B, Taflove a, Kriegsmann G. a - 1988

    Original Title

    Theory and Application of Radiation Boundary Operators - Moore T.G., Blaschak G.B, Taflove a, Kriegsmann G. a - 1988

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document
    Theory and Application of Radiation Boundary Operators - Moore T.G., Blaschak G.B, Taflove a, Kriegsmann G. a - 1988

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    5 views16 pages

    Theory and Application of Radiation Boundary Operators - Moore T.G., Blaschak G.B, Taflove A, Kriegsmann G. A - 1988

    Uploaded by

    samouille6666
    Theory and Application of Radiation Boundary Operators - Moore T.G., Blaschak G.B, Taflove a, Kriegsmann G. a - 1988

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    of 16
    Tutorial Theory and Application of Radiation Boundary Operators ‘THOMAS G. MOORE, srupexr sewn, tse, JEFFREY G. BLASCHAK, stubeNt MEMBER, IEEE, ‘ALLEN TAFLOVE, senior mesaer, te, ano GREGORY A. KRIEGSMANN Invited Review Paper Absract—A succinct unified review fx provided of the theory of ‘diation boundary operators which hat appeared principally In the Recent OSRC rests inclade analysis of the seatering behavior ofboth ‘ectricaly small and lctcally Inge cylinders, reatvely loaded ‘consti sphere, and a simple reentrant duct. New radiation boundary operntor rsa incle the demonstration of the effectiveness of higher fre operators in truncating iledifference time-domain gid. I. Ivmmopueron ITH THE RECENT introduction of the on-surface radiation condition (OSRC) method [1] and the continued growth of finite-difference time-domain (FD-TD) {2} and finite-element [3] techniques for modeling electromag- ic wave scattering problems, the understanding and use of radiation boundary operators has become increasingly impor- tant to the engineering community. Radiation boundary ‘operators have fundamentally different uses in the OSRC and finite-diference/finite-clement methods. Finite techniques use radiation boundary operators in either the time domain or frequency domain to create a radiation boundary condition (RBC) which truncates @ volumetric computational domain electrically close to a modeled target, and yet effectively Manuscript recived October 18, 1987; revised January 18, 1988. This ‘work was sported in pat by NASA Lewis Research Ceater Gast NAG 5 (635 and in pr bythe National Science Foundation Gram MCS-450057, 1.G. Moore and A. Taflove are with the Deparment of Elec ngicering and Comper Science, Technological Taste, Northwest Univesity, Evanson, IL 60201 4 G. Blaschak is withthe Lincoln Laboruoy, Mastachasets Insti of| Teta expo, MAG. pumann i withthe Department of Engincering Scinces and Angled Matemaiss, Techologieal Ine, Nortwesern Univers, rans, IL 201 TEEE Log Number 8823682. simulates the extension of the computational domain to infinity. In contrast, the OSRC method uses the radiation boundary operator directly on the surface of the target to reduce the usual frequency-domain integral equation for the scattered field to either an integration of known quantities or a second-order ordinary differential equation. Each is simply ‘implemented on the target surface. Although the OSRC and finite methods use radiation boundary operators in different ‘manners, both techniques can be greatly enhanced by more effective radiation boundary operators. ‘The purpose of this paper is to provide a succinct unified review of key research that has been performed in the area of radiation boundary operators. Because much of this research has appeared in the applied mathematics and computational physics literature over the past ten years, its results and implications are generally not well known by the engineering ‘electromagnetics community. This paper will also present some recent results from the application of these operators to ‘engineering problems. In particular, we will examine two basic types of radiation boundary operators and give examples showing their use in both the FD-TD and OSRC methods. Specifically, in Section Il we will discuss the theory behind radiation boundary operators. In Section IN the radiation boundary operators will be used to construct new radiation boundary conditions for a two-dimensional FD-TD grid of higher order than those currently used; and the effectiveness of the new radiation boundary conditions will be tested. In Section IV, the radiation boundary operators will be used in the OSRC method to approximately solve the problem of scattering from a perfectly conducting cylinder. Section V ‘concludes with @ discussion ofthe research activities that are ‘ongoing inthe areas of radiation boundary operators and their applications. IL. Twony ‘There are two basic types of radiation boundary operators: ‘mode annihilating and one-way wave equation approxima- tions. Each of these radiation boundary operators possesses different characteristics and forms. In this section, the two different types of radiation boundary operators are examined in detail. (0018-926/88/1200-1797901.00 © 1988 IEEE. 1798 ‘A, Mode-Annihilating Operators ‘The first type of radiation boundary operator to be discussed is the mode-annihilating differential operator. This type of ‘operator is based on the idea of killing the terms (herein referred to as ‘‘modes”) ofthe far-field expansion of outward ‘propagating solutions tothe wave equation. One can view the idea of killing modes of the scattered fields as first being proposed by Sommerfeld in the form of the Sommerfeld ‘radiation condition 4] which annihilates the first mode in the expansion. Later, researchers [5] extended the Sommerfeld theory and created an operator that annihilates the next mode in the expansion. Independently, other researchers created a ‘general operator that kills an arbitrary number of modes in the ‘expansion as derived and presented in [6]. It isthe theory that appeared in (6] that will be reviewed in this section. For this section we will proceed as follows. In Section I1- Al) the scattered fields are written in terms of a far-field expansion, and the effect that the Sommerfeld radiation condition has on the expansion is presented. The operators derived in (6], are presented in Section 11-A2) for the full three-dimensional case and are specialized to two dimensions in Section H-A3). 1) Far-Field Expansions and the Sommerfeld Radiation Condition: We consider here solutions U(R, 8, $, £) to the scalar wave equation vU-U, w ‘and the associated Helmholtz equation for time-harmonic VU+KU=0 @ where the wave speed ¢ has been scaled to unity and the ‘harmonic wave is assumed to have time dependency e~". The radiating solutions of the scalar wave equation (i.e, solutions ‘propagating in directions which are outward from the origin of ‘a spherical coordinate system) can be expanded in a conver- gent series of the form [7], UR, 0,6, 92 SAEED @ ‘This result was extended to the time-harmonic case for both ‘vector and scalar fields (8). For the scalar Helmholtz equation, itis proved in (8) that UR, 6, 8) ® isa convergent expansion fr scalar wavefunctions that satisfy the Sommerfeld radiation condition. ‘The Sommerfeld radiation condition (4), given by Him RU ikU)=0 oa where Up denotes a derivative respect to R, is satisfied by the each term of (4). By using the correspondence —jk = a/8t, the Sommerfeld condition is extended to fim RUn+U)=0 (3b) ‘which is satisfied by each term of the expansion in (3). The [BEE TRANSACTIONS ON ANTENNAS AKD PROPAGATION, VOL 36, NO. 12, DECEMBER 1985 ‘Sommerfeld radiation condition can be viewed as an operator on the far-field expansion of U giving the asymptotic result, Ld 2) (3-8) vnc faerie eee maces naomasonoucbemenans a tone ge Sbrcoe Wa oe on a Ree eee ees Spon mente ee hoe arene Sealey i hor nas se setts ee ee Tec ori ete ea eae an alate See Storer sno oe tenon eat heey raons ne Pace Se somal ot, gay mig « ty wrt i odn Peaea shy cee = o ROU(R, 0, 6)= 3 RFF (0, 6) +3 RM EME, 8). [Now define the intermediate operator, a La ai and observe that applying L* to both sides of (7) annihilates the first sum and makes the leading order term ofthe second ssum be O(R“*=!). We have LRU) = O(R-™") (8) which accomplishes the goal of annihilating the frst m terms of the far-field expansion. A more useful way to express this result is asa single operator acting on U only. This is achieved by inductive arguments [9]. For m = 1, (a) Oy R (= *) u+u=01R- o) an? - which canbe writen as a = (3-03) =O(R). Cs) “The frst opertor inthe soquence is then 1 aatsh (000) ‘which, when applied to both sides of (4), annihilates the first term of the expansion. Similarly, mo (eae) (100) MOORE «a: THEORY ARD APPLICATION OF RADIATION BOUNDARY OPERATORS, annihilates the first two terms. In general, the recursion = » ao(ve22") a. ee eee ee tna a emas aen Se B,U=O(R*-!) for any function U satisfying the expansion in (4). In the literature, B, has been utilized as a boundary condition ay a2) B,U=0 a3) for the wave function U. This condition becomes more sccuraz, in powers of R-i, asthe order of the operator increases. The original application of (13) was fo truncate & ‘computational domain while accurately modeling the outward propagation of waves to infinity. Further application of By, particularly B,, is found inthe OSRC method for computing scattering from two-dimensional, convex, conducting. and ‘homogenous dielectric bodies. 3) Operators for Two-Dimensional Wave Propagation: Extension of By for use with wave functions U(r, 8 1) in two space dimensions proceeds inthe time harmonic case from an ‘expansion presented in (10), 107, = Hk SOs 0 G, a 4) Which has the far-field result [9], 2 UG, =| em Mr O= that is analogous to (4). A sequence of boundary operators is ‘defined {6} by the recursion relationship as), a(t) a8 Pa nated oo a! ay oo wn ‘The operator B, annihilates the first terms of the expansion (16) and yields By Or", as) ‘The utility of these operators will be demonstrated in Section IV in the OSRC calculation of electromagnetic 179 ‘scattering from a conducting circular cylinder illuminated by @ ‘TE polarized plane wave. The first two operators used there aa ‘or Br (19) and wbtm (Bello) om In the derivation of (196) the recursion relation, (16) produces 1 second-order r derivative. It is conveniently eliminated by the substitution from the Helmholtz equation, ecenee need oF Pr ar We 20) B. One-Way Wave Equations A partial differential equation which permits wave propaga~ tion only in certain directions is called a “‘one-way wave ‘equation."" Fig. 1 shows a finite two-dimensional Cartesian 2 1800 Fig. 1. Twordimension! Canesian computational domain, where L” is defined as Lo = D,-DNI-# 40) with nt (4c) the operator L* is similarly defined except for a “+” sign before the radical. In (11) it is shown that at a boundary, say at x = 0, the application of L~ to the wave function U will exactly absorb a plane wave incident at any angle and traveling in the —x direction. Thus L-U=0 @) applied at x = 0 functions as an exact analytical RBC which absorbs wave motion from the interior of the spatial domain {= @& y):0 RDC’. By sacrificing performance near 0 0°, the L7 RBC extends the point at which is less than one percent to about @ = 60 B. Numerical Experiments [Numerical experiments are now reported which clearly ‘measure the amount of nonphysical reflection a given RBC produces as a pulse propagates through a grid boundary. Fig. 4(a) shows two domains on which the two-dimensional FD-TD algorithm is computed simultaneously for the transverse ‘magnetic (TM) case. On the boundary ofthe test domain Oy a test RBC is applied. Each point in @ has a corresponding ‘member in the substantially larger domain M,. A tine source is located at grid position (50, 25) in both domains. The source produces outward propagating, cylindrical waves which are spatially coincident in both domains up until time steps when the waves interact with the boundary of fy. Any reflection from the boundary of r makes the solution at points within Q differ from the solution at corresponding points within Qp. The ‘wave solution at points within Op represents the desired numerical modeling of free-space propagation up until time steps when reflections from its own boundary enter the region ‘of {corresponding to @. By calculating the difference inthe solutions in 1p and Q at each point at each time step, a measure of the spurious reflection caused by the boundary of is obtained, We define atthe nth time step DG D=EM N- EMS) (40a) for all (i, ) within the test domain, where £7 i the solution [MOORE tl: THEORY AND APPLICATION OF RADIATION BOUNDARY OPERATORS 1803 wo? wot 05 = man 108 = tee, 107 Reflection Cootficlent 010 20 a0 40 60 60 70 80 Angle of incldanco « Reflection Coefficient eb 0 = wa = mee) 104 ot 2 90 40 80 60 70 80 90 ‘Angle ot incidence Fig. 2. Reflection coefciet verus wave angle of lchence, (a) Palé RBC. (}) Cchysbey o sbinterval RBC, 102 108 wot wos = Pate | a Chat on sie 0? 10% vo 20 a0 «0 80 60 70 80 90 ‘Angle of incidence Fig. 3. Comparion of third order RBCs [IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL, 36, NO. 12, DECEMBER 1988 a, — source at (eoas) a, 1 ® ox. Pa 2 owe. & 2 ocw. & > compact guise yom Time Step » Fig. 4. Pulse sues. 2) Compustionsl domains.) Wave source, within 0; and E? isthe solution at points in Q,. D(, J) is the local error in the test domain caused by its grid boundary reflections. We also define a global reflected error measure, E=SY DMA) (406) for all (i, /) within , which measures the total reflected error within the test grid atthe nth time step. ‘The source used in the numerical experiments is the pulse ‘obtained from (20) and is defined as follows: E50, 25, n) Ee [Zot et ok tou mom sr 0, bor (1a) where (ai) and 51 is the time step used in the simulation. In all experiments we maintained 8¢ = 2.5 x 10"! sec and A = 2c 4, where cis the speed of light in free-space and A is the space increment of the finite difference grid. The time profile of the pulse defined in (41a) is shown in Fig. 4(6). This pulse was selected because it has an extremely smooth transition to zero. AAs discussed in [20], the pulse has its first five derivatives vanish at £ = 0, 7 and is a good approximation to a smooth ‘compact pulse This pulse has very litle high-frequency content which is important because ofthe deleterious effects of grid dispersion (dependence of numerical wave phase velocity upon spatial wavenumber). Grid dispersion and its relation to RBC's is discussed in [21}, and [22]. This problem is compounded in higher dimensions by anisotropies of the numerical wave phase velocity with wave vector angle in the grid [3], [23] and is a subject of current research aimed at further reduction of arid boundary reflection coefficients. ‘The source point in Fig. (a) is 25 cells from the boundary ‘of Bat y = 0. With the specification A = 2c t, disturbances atthe source point require 50 time steps to propagate to the ‘boundary at y = 0. At time step 70, the peak of the pulse just starts to passthrough the boundary. We choose to observe the reflection at the first row of grid points away from the y = 0 ‘boundary (along J = 1) at time step 2 = 100. This permits the ‘bulk of the outgoing pulse to pass through the boundary and excite the largest observable reflection, MOORE eo: THEORY AND APPLICATION OF RADIATION BOUNDARY OPERATORS oe = yen = ype) Normalized Reflection Along J=1 ny oom Error so 7 100 = yee e2) = mee) Time Step Figs. 5 and 6 show the local and global reflection errors ‘observed for the Padé and L> RBCs. Fig. 7 compares the third-order conditions. In Figs. 5@), 6(@), and 7(@), DG, 1) ‘has been normalized with the peak value of the incident pulse which strikes the y = 0 boundary at timestep n = 70 at grid Peston (50, 0). The pulse experiment result rein agreement withthe reflection coefficient analysis by showing that higher order RBC's do perform better than lower order RBC’s in actual simulations. However, comparison of the third-order ‘LE RBC tothe third-order Padé RBC does not indicate any particular performance advantage. The improved wide-angle performance suggested in {15] isnot evident in these exper- ments. IV. Appucarion oF Moe-ANNiniLaTina OreRaToRs: OSRC ‘The on-surface radiation condition method [1] is a new ‘analytical technique by which it is possible to construct ‘accurate approximation of two- and three-dimensional scatter- ing problems involving convex and simple reentrant targets. In this section, two areas of application willbe examined. In the first application, the OSRC method will be applied to compute rrr measres, Pal RBC. a) Locl err at ® 100, @) Global ero. the scattering cross section of two canonical convex targets: 1) a circular cylinder illuminated by both a transverse electric (TE) and transverse magnetic polarized plane wave; and 2) fan acoustic sphere with a constant surface impedance. In the second application, the OSRC method will be applied to the scattering of a plane wave by a canonical reentrant geometry: the open end of a semi-infinite flanged paralle-plate wave- ‘guide. Before either of the cases is examined, some back- ‘ground discussion on the OSRC method is necessary. A. Background ‘The OSRC method is based upon the application of a radiation boundary operator, such as those discussed in Section Il, directly on the surface of the target. The effect of this is to relate the surface currents to known field quantities through a simple expression; thus the problem reduces to solving an equation along the contour of the target. Only second-order operators will be considered here because they fare the most widely used. The method used in this paper will bbe the same one used in [I]. For completeness, recently ‘proposed variations will be reviewed as wel. 1806 on. = men = ee, Normalize Retction Along J=1 88 200 008 (Grd Poston, t @ cn: 09: so 78 +1 i SS 135 150 Time Step o Error measures, Chebyshev on sobnteval RBC. (2) Loca err atm = 100.) Global ero Fi. 6. ‘The original method developed for two-dimensional electro- magnetic targets is to apply a mode-annibilating radiation ‘boundary operator locally at each point on the surface of the target {1}. This is conveniently done by noting that an osculating circle can approximate the target's surface locally at each point. The operator then works on the fields as if they ‘were emanating from within the local osculating circle. In two dimensions, this is accomplished by making the following a8 (2m aan cea 1 Lee 2) 1a (42c) Pee ast where m isthe outward normal, 5 is an arc length parameter, and «(s) is the curvature of the target at s. This produces the ‘second-order two-dimensional surface boundary operator {1} soto donor [ow seat] & ea A three-dimensional surface boundary operator developed for acoustic targets is presented in (24). There, the following substitutions are made in the three-dimensional modeannii- lating radiation boundary operator: (30) a RHO (438) vs Vy ey 7 (630) where V-V is the surface Laplacian and His the mean curvature. This produces the second-order three-dimensional ‘MOORE: THEORY AND APPLICATION OF RADIATION BOUNDARY OPERATORS ooo ome oe: oon order ROE chaby. sus Normalized Reflection Along J= 1 00008 ‘doer RACs = Chay. si Time Step Fig. 7 Comparison of Padé to Chebyshe on subiateval. (a) Local eror surface boundary operator, a ~k 2 U=V-VU+%—K—jkEDU. (34) ‘A variation of this surface boundary condition is presented in {25} in which a slighly diferent radiation boundary operator is used. It is atm 2 ya us20r—-kHNe. 480 ‘The fundamental difference between (434) and (43e) is that the latter annihilates terms of order (KR)-* as compared to (KR)? for the former. (Other methods have recently been presented for deriving a general surface boundary condition for OSRC. A sequence of surface boundary operators is derived in [26] by directly factoring the wave equation, as in Section II-B but in a general coordinate system based on the local properties ofthe target's surface. These results differ from (424), (434), and (43e) by a ° = 100. (Global err. term proportional tothe derivative of the curvature multiplied bby k-2, No published evidence exists at this time to indicate that this term has any beneficial effects. The most recent derivation of surface boundary conditions for OSRC was presented in [27], which demonstrates that the surface ‘boundary condition can be derived directly from geometrical acoustics/optics by making the assumption that the surface ofthe target isa phase front. The resulting boundary condition + VU+Qk+ H?~Ko—jkH)U (4a) Where xo is the Gaussian curvature. The corresponding surface boundary condition in two dimensions is aye 3 «je 2 va 23 243k 26-10) 5, 50+ [ae ; sung] u j ee 0! Bistatc RCS. IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VO. 6, NO. 12, DECEMBER 188 = eet OGRC ung 8 7 anc ung a2 120140 160 180 Phi (degrees) = bat SRC using *OSRC ining 2 oa 40 60 80 We note that (44b) also differs from (424) in the terms proportional tothe derivative of the target's curvature (which are formally O(K-) corrections, but, in turn, differs slightly from the operator derived in [26]. More importantly, [27] shows how this method is deriving surface boundary operators can be applied to the vector scattering case of OSRC. However, no validations have been published to date. Now that the surface boundary conditions have been derived, the application of the OSRC method to scalar problems is straightforward. We set BU =0 45) ‘on the surface of the target, where Bis one of the surface boundary conditions described above. What results is an expression that relates the scattered field to its normal

    You might also like