Computational Electromagnetics and Antenna Modelling

Introduction and Context

Computational electromagnetics (CEM) is the numerical study of electromagnetic fields, currents, charges, radiation, scattering, and interaction with materials and structures. In antenna engineering, CEM complements analytical theory and measurement. Analytical solutions are available only for a limited class of canonical geometries, whereas practical antennas contain finite conductors, dielectric substrates, feeds, radomes, packaging, losses, and nearby objects. Numerical modelling makes it possible to predict performance before fabrication and to investigate designs that cannot be treated reliably by closed-form methods.

A professional antenna simulation is not simply the operation of a software package. It is a modelling workflow involving physical idealisation, formulation selection, geometry preparation, meshing, solver settings, convergence assessment, validation, uncertainty analysis, and engineering interpretation. A visually smooth field plot does not establish that a result is correct. The central question is whether the numerical solution is sufficiently accurate for the decision being made.

This chapter develops the principal full-wave and high-frequency techniques used in antenna modelling: the Method of Moments (MoM), finite-element method (FEM), finite-difference time-domain (FDTD), physical optics (PO), and asymptotic techniques. It also examines meshing, convergence, validation, sensitivity, uncertainty, professional workflows, and a representative antenna-design case.

Learning Objectives

After completing this chapter, the reader should be able to:

1. Explain the physical and mathematical basis of major computational electromagnetic methods.

2. Select MoM, FEM, FDTD, PO, or an asymptotic method for a specified antenna problem.

3. Construct suitable computational domains, boundaries, materials, ports, and meshes.

4. Interpret convergence, validation, uncertainty, and sensitivity results.

5. Critically evaluate numerical outputs and identify common modelling errors.

6. Develop a defensible simulation-to-measurement workflow for antenna design.

1. Electromagnetic Modelling as a Boundary-Value Problem

Maxwell’s equations describe the electromagnetic fields. In differential form,

\[

\nabla\times \mathbf{E}=-\frac{\partial \mathbf{B}}{\partial t},\qquad

\nabla\times \mathbf{H}=\mathbf{J}+\frac{\partial \mathbf{D}}{\partial t},

\]

with

\[

\nabla\cdot\mathbf{D}=\rho,\qquad \nabla\cdot\mathbf{B}=0.

\]

For linear, isotropic media, \(\mathbf{D}=\epsilon\mathbf{E}\), \(\mathbf{B}=\mu\mathbf{H}\), and \(\mathbf{J}=\sigma\mathbf{E}+\mathbf{J}_s\), where \(\epsilon\), \(\mu\), and \(\sigma\) are permittivity, permeability, and conductivity. In sinusoidal steady state with the \(e^{j\omega t}\) convention, the wave equation in a homogeneous source-free region may be written as

\[

\nabla^2\mathbf{E}+k^2\mathbf{E}=0,

\qquad k=\omega\sqrt{\mu\epsilon}.

\]

Loss can be represented using complex permittivity or a complex wavenumber. The numerical problem is completed by specifying geometry, material properties, excitation, and boundary conditions. Typical boundaries include perfect electric conductor (PEC), perfect magnetic conductor (PMC), impedance or absorbing boundaries, periodic boundaries, and radiation conditions at infinity.

A computational model therefore contains four distinct descriptions:

• Geometry: surfaces, volumes, gaps, ports, and nearby objects.
• Physics: constitutive properties, conductor models, losses, and excitations.
• Discretisation: basis functions, cells, elements, or rays.
• Numerical solution: linear-system solution, time marching, eigenvalue extraction, or asymptotic evaluation.

The operating wavelength \(\lambda=2\pi/k\) is a primary scale. Geometrical features much smaller than \(\lambda\) may still be important at a feed, edge, slot, or material interface; consequently, a uniform wavelength-based mesh is rarely optimal.

2. Method of Moments

2.1 Integral-equation formulation

The Method of Moments converts an electromagnetic integral equation into a matrix equation. For a perfectly conducting wire or surface, the unknown is usually the induced current. The electric field produced by a current distribution is represented through a Green’s function. Symbolically, an operator equation has the form

\[

\mathcal{L}(\mathbf{J})=\mathbf{E}^{\mathrm{inc}},

\]

where \(\mathbf{J}\) is the unknown current and \(\mathbf{E}^{\mathrm{inc}}\) is the incident or impressed field. The current is expanded using basis functions:

\[

\mathbf{J}(\mathbf{r})\approx \sum_{n=1}^{N} I_n\mathbf{f}_n(\mathbf{r}).

\]

Testing the equation with functions \(\mathbf{w}_m\) gives

\[

\sum_{n=1}^{N} Z_{mn}I_n=V_m,

\]

or, in matrix notation,

\[

\mathbf{Z}\mathbf{I}=\mathbf{V}.

\]

Here \(\mathbf{Z}\) is the impedance matrix, \(\mathbf{I}\) contains expansion coefficients, and \(\mathbf{V}\) represents excitation projections. The entries depend on the Green’s function, basis functions, frequency, and material environment.

For a thin-wire antenna, the unknown may be a longitudinal current. For surface-integral-equation models, Rao–Wilton–Glisson (RWG) basis functions are widely used on triangular meshes. RWG functions preserve appropriate continuity of normal current across neighbouring triangles and are well suited to electric-field integral equations.

2.2 Strengths and limitations

MoM is especially efficient for open-region radiation and scattering because the Green’s function analytically incorporates radiation to infinity; no large air volume or absorbing outer boundary is necessarily required. It is therefore attractive for wire antennas, conducting surfaces, reflectors, and arrays.

The principal limitation is that the matrix is generally dense. If \(N\) unknowns are used, direct factorisation may require approximately \(O(N^3)\) operations and \(O(N^2)\) storage. Iterative solvers, fast-multipole methods, adaptive integral methods, and low-rank compression reduce this burden, but conditioning and convergence remain important.

MoM may become less convenient for strongly inhomogeneous three-dimensional dielectric volumes, complex lossy media, or geometries containing many electrically large volumetric regions. Hybrid formulations can combine surface integral equations with FEM or physical optics.

2.3 Antenna quantities from MoM

Once currents are known, input impedance is obtained from the ratio of terminal voltage to current,

\[

Z_{\mathrm{in}}=\frac{V_{\mathrm{in}}}{I_{\mathrm{in}}}=R_{\mathrm{in}}+jX_{\mathrm{in}}.

\]

For a lossless one-port antenna, radiated power is related to the real part of input impedance. With a specified reference impedance \(Z_0\), the reflection coefficient is

\[

\Gamma=\frac{Z_{\mathrm{in}}-Z_0}{Z_{\mathrm{in}}+Z_0},

\qquad

\mathrm{VSWR}=\frac{1+|\Gamma|}{1-|\Gamma|},

\]

provided \(|\Gamma|<1\). These expressions assume a consistent reference plane and compatible impedance definitions.

3. Finite-Element Method

3.1 Local discretisation

The FEM divides the computational region into finite elements, commonly tetrahedra in three dimensions. The field is approximated locally using shape functions. For a frequency-domain electric-field formulation, a weak form is obtained by multiplying the governing equation by a test function and integrating over the domain. A representative form is

\[

\int_\Omega (\nabla\times\mathbf{W})\cdot\mu^{-1}(\nabla\times\mathbf{E})\,d\Omega

-\omega^2\int_\Omega \mathbf{W}\cdot\epsilon\mathbf{E}\,d\Omega

=\text{source and boundary terms}.

\]

The exact expression depends on the selected formulation, material tensors, losses, and boundary conditions. Edge-based vector elements, especially first-order Nédélec elements, are used for electromagnetic fields because they enforce tangential-field conformity and avoid many spurious modes associated with inappropriate nodal elements.

After discretisation, FEM produces a sparse matrix system. If the unknown vector is \(\mathbf{x}\), the system can be represented as

\[

\mathbf{A}(\omega)\mathbf{x}=\mathbf{b}.

\]

For eigenvalue problems, the system often takes the generalized form

\[

\mathbf{K}\mathbf{x}=\omega^2\mathbf{M}\mathbf{x},

\]

where \(\mathbf{K}\) and \(\mathbf{M}\) are stiffness-like and mass-like matrices.

3.2 Open boundaries and absorbing regions

FEM is naturally suited to bounded domains, whereas antennas radiate into an unbounded exterior. The exterior is truncated using absorbing boundary conditions or, more commonly, a perfectly matched layer (PML). A PML is a deliberately constructed anisotropic or complex-stretched region designed to absorb outgoing waves with minimal reflection. It must be sufficiently thick, correctly graded, and separated from strong near fields when possible.

FEM is powerful for dielectric-loaded antennas, waveguide feeds, radomes, cavities, and strongly inhomogeneous structures. It offers local mesh refinement and sparse algebra. However, the air region and PML can become expensive for electrically large open problems. Poorly designed PMLs, incorrect element order, and insufficient refinement near sharp edges are common sources of error.

4. Finite-Difference Time-Domain Method

4.1 Yee discretisation

FDTD solves Maxwell’s curl equations directly in time. The Yee algorithm places electric and magnetic field components at staggered positions in space and time. This arrangement supports the discrete curl relationships and provides a robust explicit update scheme.

In a simple one-dimensional, source-free, lossless case, the electric-field update has the general structure

\[

E^{n+1}=E^n+\frac{\Delta t}{\epsilon\Delta x}(H^{n+1/2}_{\text{left}}-H^{n+1/2}_{\text{right}}),

\]

with a corresponding magnetic-field update. Exact signs and component indices depend on coordinate orientation. The method is explicit: new field values are calculated from known previous values.

Stability requires a Courant–Friedrichs–Lewy condition. In a three-dimensional uniform Cartesian grid,

\[

\Delta t\leq \frac{S}{c\sqrt{(1/\Delta x)^2+(1/\Delta y)^2+(1/\Delta z)^2}},

\]

where \(c\) is wave speed in the reference medium and \(S\leq 1\) is a safety factor. For cubic cells of side \(\Delta\), this becomes \(\Delta t\leq S\Delta/(c\sqrt{3})\). The condition is a stability requirement, not a guarantee of accuracy.

4.2 Advantages and limitations

FDTD is broadband: one time-domain excitation can yield a frequency response over a band through Fourier transformation. It is attractive for ultra-wideband antennas, transient radiation, electromagnetic compatibility, pulsed systems, and nonlinear or time-varying materials. Its disadvantages include staircasing on curved surfaces, large memory requirements, long simulation times for high-Q structures, and a time step controlled by the smallest cell in the model.

Subcell models, conformal FDTD, nonuniform grids, PMLs, and parallel computing improve performance. Nevertheless, a narrow gap or tiny feed feature can force a very small time step throughout an explicit computation.

5. Physical Optics and Asymptotic Techniques

5.1 Physical optics

Physical optics approximates the induced current on a sufficiently large, smooth, highly conducting illuminated surface by the local incident magnetic field:

\[

\mathbf{J}_{\mathrm{PO}}\approx 2\hat{\mathbf{n}}\times\mathbf{H}^{\mathrm{inc}}

\]

on the illuminated region, with current commonly set to zero on the geometrically shadowed region. Here \(\hat{\mathbf{n}}\) is the outward surface normal. The approximation assumes an electrically large conducting body and neglects or simplifies edge diffraction, multiple scattering, curvature-transition effects, and detailed resonances.

PO is efficient for reflector antennas, large metallic platforms, and radar-cross-section estimates. It is often used with physical theory of diffraction (PTD), which adds approximate edge-diffraction currents. Geometrical optics (GO) models rays, reflection, and refraction, while geometrical theory of diffraction (GTD) and uniform theory of diffraction (UTD) extend ray methods to edges, tips, and shadow boundaries.

5.2 Domain of validity

There is no universal electrical-size threshold at which an asymptotic method becomes exact. Accuracy depends on curvature, observation direction, edge geometry, material contrast, polarization, and whether the result is dominated by a resonance or diffraction phenomenon. PO is generally unreliable for electrically small objects, narrow slots, strongly resonant cavities, and situations in which creeping waves or multiple interactions dominate.

Hybrid methods use a full-wave solver for critical regions and an asymptotic method for electrically large surroundings. This can provide a useful balance between accuracy and computational cost, but the interface between regions must conserve fields and avoid double counting.

6. Selecting a Computational Method

Method selection should be based on physics, not software familiarity. A practical decision framework is:

| Problem characteristic | Preferred method or formulation |

|---|---|

| Thin wires and conducting surfaces in free space | MoM |

| Complex dielectric and metal volumes, waveguides, radomes | FEM |

| Broadband transient or ultra-wideband antenna | FDTD |

| Electrically large reflector or metallic platform | PO/PTD or hybrid asymptotics |

| Large structure with a detailed local feed | Hybrid FEM–MoM, FEM–PO, or domain decomposition |

| Resonant closed or semi-closed cavity | FEM or eigenmode methods |

| Periodic array or unit cell | FEM, MoM, or FDTD with periodic boundaries |

The final choice also depends on available memory, desired outputs, frequency span, material complexity, and whether fields, currents, impedance, transient waveforms, or far-field patterns are required.

7. Meshing and Discretisation Quality

A mesh must resolve both electromagnetic variation and geometric detail. A commonly used starting guideline is to place several elements per wavelength in the medium, where

\[

\lambda_m=\frac{\lambda_0}{\sqrt{\epsilon_r\mu_r}}

\]

for a lossless homogeneous material with relative parameters \(\epsilon_r\) and \(\mu_r\). This is only a starting point. Higher-order elements may achieve a given accuracy with fewer elements per wavelength, while sharp edges, thin substrates, narrow slots, skin depth, and strong material gradients require local refinement.

Important mesh regions include:

• feed gaps and lumped ports;
• conductor edges and corners;
• dielectric interfaces;
• high-curvature surfaces;
• strongly coupled or closely spaced elements;
• regions of rapidly varying near fields;
• PML interfaces and terminating boundaries.

A good mesh is not necessarily uniform or maximally fine. It is adapted to error indicators and engineering sensitivity. Excessive refinement can increase computational cost without improving the quantity of interest.

For lossy conductors, the skin depth is

\[

\delta=\sqrt{\frac{2}{\omega\mu\sigma}},

\]

under the good-conductor approximation \(\sigma\gg\omega\epsilon\). If a volumetric conductor is explicitly resolved, its mesh must represent the relevant current decay; otherwise, an impedance or surface-impedance boundary may be more appropriate.

8. Convergence, Verification, and Validation

8.1 Convergence

A convergence study examines how a computed quantity changes as discretisation is refined. Possible quantities include resonant frequency, input impedance, accepted power, gain, radiation efficiency, or a selected field value. A result is not converged merely because the solver reports completion.

For a mesh sequence with characteristic size \(h\), one may examine

\[

Q(h_1),\;Q(h_2),\;Q(h_3),

\]

where smaller \(h\) denotes refinement. A practical stopping criterion is that the change in the quantity of interest between successive refinements is below a specified engineering tolerance. For example, one may require resonant frequency changes below 0.5% and peak realized-gain changes below 0.2 dB, depending on application.

Frequency convergence and mesh convergence are distinct. A frequency sweep can be finely sampled while the spatial discretisation remains inadequate. Conversely, a highly refined mesh does not compensate for an incorrectly defined port or boundary.

8.2 Verification and validation

Verification asks whether the equations were solved correctly. It includes code checks, residual monitoring, mesh refinement, comparison with canonical solutions, conservation tests, and independent solver or formulation comparisons.

Validation asks whether the mathematical model represents the physical device. It requires comparison with measurements or trusted experimental data, including uncertainty in fabrication, materials, connectors, cables, calibration, and test environment.

For a one-port antenna, a useful power balance is

\[

P_{\mathrm{accepted}}=P_{\mathrm{rad}}+P_{\mathrm{loss}},

\]

where accepted power excludes reflected power. Significant unexplained imbalance indicates a possible port, material, normalization, or post-processing error.

8.3 Residuals and conditioning

For \(\mathbf{A}\mathbf{x}=\mathbf{b}\), the relative residual can be defined as

\[

r=\frac{\|\mathbf{b}-\mathbf{A}\mathbf{x}\|}{\|\mathbf{b}\|}.

\]

A small residual indicates that the discrete equations are being solved accurately, but it does not prove that the mesh, geometry, or physical model is accurate. Ill-conditioned matrices can amplify data and round-off errors. Near resonances, nearly touching conductors, extreme material contrasts, and poor basis choices often create conditioning difficulties.

9. Uncertainty and Sensitivity Analysis

Real antenna performance depends on uncertain variables such as substrate permittivity, thickness, conductor dimensions, solder geometry, connector position, surface roughness, and manufacturing tolerances. Reporting only a nominal simulation can therefore be misleading.

Let the response be \(y=f(x_1,x_2,\ldots,x_n)\). A local sensitivity to variable \(x_i\) may be estimated by

\[

S_i\approx\frac{f(x_i+\Delta x_i)-f(x_i-\Delta x_i)}{2\Delta x_i}.

\]

This central finite difference is appropriate when the response is sufficiently smooth and \(\Delta x_i\) is small enough to represent local behaviour but large enough to avoid numerical noise. A normalized sensitivity is often more interpretable:

\[

S_i^{\mathrm{norm}}=\frac{x_i}{y}\frac{\partial y}{\partial x_i}.

\]

For independent small uncertainties, first-order variance propagation gives

\[

\sigma_y^2\approx\sum_i\left(\frac{\partial y}{\partial x_i}\right)^2\sigma_{x_i}^2.

\]

This approximation assumes small variations, a locally linear response, and negligible covariance. If variables are correlated, covariance terms must be included. Near a sharp resonance, nonlinear or Monte Carlo analysis is safer. Latin hypercube sampling, polynomial chaos, surrogate models, and adjoint sensitivities can reduce the cost of uncertainty quantification.

Sensitivity analysis supports design decisions: a variable with high sensitivity deserves tighter manufacturing control, improved measurement, or robust redesign. It also helps distinguish numerical uncertainty from physical tolerance uncertainty.

10. Professional Simulation Workflow

A defensible workflow proceeds as follows:

1. Define the question. Specify frequency range, output quantities, tolerances, and pass/fail criteria.

2. Idealise deliberately. Decide which screws, seams, cables, coatings, and losses materially affect the result.

3. Build parametric geometry. Use named dimensions, version control, and explicit coordinate systems.

4. Assign materials carefully. Use frequency-dependent complex properties where justified and document data sources.

5. Define excitation and reference planes. Verify port orientation, impedance, mode content, and de-embedding.

6. Select solver and boundaries. Check that the method’s assumptions match the electromagnetic regime.

7. Generate a physics-aware mesh. Refine gaps, edges, interfaces, and high-field regions.

8. Run baseline and convergence cases. Record mesh statistics, solver tolerances, residuals, and computational resources.

9. Check conservation and plausibility. Inspect current continuity, field symmetry, power balance, and limiting cases.

10. Validate independently. Compare with measurement, an analytical result, or a second numerical formulation.

11. Perform sensitivity and uncertainty studies. Propagate realistic tolerances, not arbitrary perturbations only.

12. Report reproducibly. Archive geometry, material files, solver version, mesh, settings, scripts, and post-processing definitions.

Automation through Python, MATLAB, or solver scripting is valuable for parameter sweeps, mesh studies, regression testing, and uncertainty analysis. Automation should include checks for failed solves, nonphysical results, inconsistent units, and accidentally changed boundary conditions.

11. Illustrative Case Study: Printed Patch Antenna with a Radome

Background

Consider an illustrative 5.8 GHz rectangular microstrip patch antenna intended for a compact wireless sensor. The antenna is fabricated on a dielectric substrate and enclosed by a thin polymer radome. The design must meet a target impedance bandwidth, maintain acceptable broadside gain, and tolerate manufacturing variation in radome spacing and substrate permittivity.

Problem

An initial model predicts a resonance at 5.8 GHz, but the measured prototype resonates at 5.65 GHz and has approximately 1 dB lower realized gain. The engineering team must determine whether the discrepancy is caused by numerical error, inaccurate materials, fabrication tolerances, or the radome model.

Analysis and approach

A frequency-domain FEM model is selected because the structure contains dielectric volumes, a feed transition, and a radome. The air region is terminated with a PML. The patch edges, coaxial transition, substrate interface, and radome corners receive local refinement. A coarse, intermediate, and fine mesh are solved, and resonance is defined consistently as the frequency of minimum \(|S_{11}|\).

The mesh study gives the following illustrative results:

| Mesh | Elements | Resonant frequency | Minimum \(|S_{11}|\) |

|---|---:|---:|---:|

| Coarse | 42,000 | 5.83 GHz | −18.1 dB |

| Intermediate | 91,000 | 5.81 GHz | −18.7 dB |

| Fine | 188,000 | 5.80 GHz | −18.8 dB |

The small change from intermediate to fine mesh indicates adequate numerical convergence for resonance, although gain and efficiency require separate checks. A second model replaces the detailed coaxial connector with a calibrated wave port and compares the reference plane. The two models agree within the project tolerance after de-embedding.

Next, the substrate relative permittivity is varied over its measured tolerance range, and radome spacing is varied according to mechanical drawings. A 2.5% increase in effective permittivity shifts resonance downward, while reduced radome spacing increases capacitive loading and also shifts resonance downward. The nominal material database had used a room-temperature value that was not representative of the measured laminate batch.

A measurement model is then constructed including the actual substrate thickness, copper thickness, connector launch, and radome. The revised simulation predicts a resonance near 5.66 GHz and a gain reduction close to the measured value. The remaining difference is within the combined uncertainty of chamber calibration, material characterisation, and fabrication.

Outcome

The investigation shows that the original discrepancy was not primarily a mesh problem. It resulted from incomplete physical modelling: the effective dielectric properties and radome spacing were inaccurate. The final design retunes the patch dimensions and specifies a controlled radome-to-patch distance. A tolerance study confirms that the revised design remains within the required impedance band for most manufactured units.

Learning points

• A converged numerical solution can still be physically invalid if material data are wrong.
• Port definition and reference-plane consistency are essential for meaningful \(S\)-parameters.
• Resonance convergence does not automatically establish gain or efficiency convergence.
• Measurement disagreement should be investigated through a hierarchy: units and geometry, ports, mesh, boundaries, materials, fabrication, and test uncertainty.
• Sensitivity results can guide manufacturing specifications and redesign.

12. Challenges, Limitations, and Governance

All computational methods involve approximation. Errors may arise from geometric simplification, discretisation, iterative solver tolerance, truncated domains, inaccurate constitutive data, uncertain contact resistance, or incorrect interpretation of field quantities. Antenna models can also be numerically sensitive near resonances, where a small dimensional or material change produces a large response change.

Professional governance requires traceability. Simulation files should be version-controlled, and critical results should be reproducible by another engineer. Commercial software does not remove the need for engineering judgement. Licences, proprietary material databases, and cloud-computing services may introduce data-security and reproducibility concerns. When AI-assisted geometry optimisation or surrogate modelling is used, training data, objective functions, constraints, and validation cases must be documented. AI-generated designs must not be accepted without electromagnetic verification and, where appropriate, physical testing.

Safety and regulatory implications also matter. Antenna models used for exposure assessment, radar performance, satellite links, or certification must use validated methods and appropriate worst-case assumptions. A visually attractive simulation must never substitute for required compliance measurements.

13. Contemporary Developments and Best Practices

Current CEM research is advancing fast multipole and low-rank methods, GPU acceleration, domain decomposition, reduced-order modelling, topology optimisation, adjoint-based gradients, uncertainty quantification, and digital-twin workflows. Machine learning is increasingly used for surrogate models and inverse design, but its reliability depends on adequate training coverage and physics-based validation.

Best practice is to combine methods when appropriate rather than seeking one universal solver. For example, a detailed FEM feed model may be coupled to a MoM or PO model of a large platform. The coupling must be tested against overlapping benchmark cases. Adaptive meshing should respond to estimated error or response sensitivity, not only to geometry size. Results should be reported with units, coordinate conventions, mesh data, solver tolerances, boundary definitions, and uncertainty estimates.

Chapter Summary

Computational electromagnetics transforms Maxwell’s equations into numerically solvable models of practical antennas. MoM is efficient for conducting wires and surfaces in open space but produces dense systems. FEM provides flexible local discretisation for complex dielectric and volumetric structures and requires careful treatment of open boundaries. FDTD directly advances fields in time and is valuable for broadband and transient problems, subject to stability and memory constraints. PO, PTD, GO, GTD, and UTD provide efficient approximations for electrically large structures but are limited near resonances, edges, and strong multiple interactions.

Reliable antenna modelling requires more than selecting a solver. Geometry, materials, ports, boundaries, meshing, convergence, residuals, power balance, and validation must be considered together. Verification establishes numerical correctness, whereas validation assesses physical realism. Sensitivity and uncertainty analysis reveal whether manufacturing and material variations threaten performance. Professional workflows are documented, reproducible, automated where useful, and supported by independent checks and measurement.

Key Terms

Computational electromagnetics: Numerical solution of electromagnetic field problems.

Method of Moments: Integral-equation technique that expands unknown currents or fields in basis functions.

Finite-element method: Local weighted-residual discretisation using finite elements.

Finite-difference time-domain: Explicit time-domain solution of discretised Maxwell curl equations.

Physical optics: High-frequency approximation of induced currents on illuminated conducting surfaces.

Perfectly matched layer: Artificial absorbing region designed to minimise reflection at a truncated boundary.

Rao–Wilton–Glisson basis: Edge-based basis function widely used for triangular surface meshes.

Mesh convergence: Reduction of solution change as spatial discretisation is refined.

Verification: Demonstration that the discrete equations are solved correctly.

Validation: Demonstration that the model represents the physical system adequately.

Residual: Norm of the remaining algebraic error in a solved matrix equation.

Conditioning: Sensitivity of a numerical system to perturbations in data or arithmetic.

Skin depth: Characteristic depth over which current decreases in a good conductor.

Sensitivity analysis: Study of response variation caused by changes in input parameters.

Uncertainty quantification: Estimation and propagation of imperfectly known inputs.

Domain decomposition: Division of a problem into coupled computational subdomains.

Near field: Electromagnetic region close to an antenna where reactive and radiating terms coexist.

Far field: Region where angular radiation characteristics are established and fields approximate transverse waves.

Surrogate model: An inexpensive approximation of a computationally expensive simulation.

Review Questions

Short Answer Questions

1. What are the principal unknowns in a typical MoM antenna formulation?

2. Why is a PML used in FEM or FDTD antenna simulations?

3. State the distinction between verification and validation.

4. What is the purpose of mesh convergence testing?

5. Why can a small residual fail to guarantee an accurate antenna result?

Long Answer Questions

1. Compare MoM, FEM, and FDTD in terms of formulation, computational structure, strengths, and limitations.

2. Explain the assumptions and limitations of physical optics and its relationship to diffraction corrections.

3. Describe a complete professional workflow for modelling a dielectric-loaded antenna.

4. Discuss how material uncertainty and manufacturing tolerance affect antenna performance predictions.

5. Explain why ports, reference planes, and power normalisation require special attention in antenna simulation.

Analytical / Application Questions

1. Select an appropriate computational method for a broadband electrically small antenna mounted on a complex dielectric housing. Justify the selection and identify likely numerical risks.

2. Design a mesh-convergence study for an antenna whose acceptance criteria are resonance within 0.5% and realized gain within 0.2 dB. Specify quantities, refinement strategy, and stopping logic.

3. A simulated antenna agrees in resonance with measurement but differs in gain by 1.5 dB. Develop a systematic investigation covering numerical, physical, measurement, and uncertainty causes.

Multiple Choice Questions with Answers

Question 1

Which method most directly uses an integral equation with unknown surface or wire currents?

A. Finite-element method

B. Method of Moments

C. Finite-difference time-domain

D. Geometrical optics

Correct Answer: B. Method of Moments

Question 2

What is the main purpose of a perfectly matched layer?

A. To increase conductor conductivity

B. To impose periodic excitation

C. To absorb outgoing waves at a truncated domain

D. To calculate the antenna feed impedance

Correct Answer: C. To absorb outgoing waves at a truncated domain

Question 3

For a cubic FDTD cell of side Δ, the three-dimensional Courant limit has the form

A. Δt ≤ SΔ/c

B. Δt ≤ SΔ/(c√3)

C. Δt ≤ Sc/Δ

D. Δt ≤ cΔ/S

Correct Answer: B. Δt ≤ SΔ/(c√3)

Question 4

Which method is generally most suitable for an electrically large metallic reflector?

A. Physical optics

B. Nodal static analysis

C. Lumped-circuit analysis only

D. Thermal finite elements

Correct Answer: A. Physical optics

Question 5

What does verification primarily assess?

A. Whether the physical prototype was manufactured correctly

B. Whether the numerical equations were solved correctly

C. Whether the antenna meets a regulatory limit

D. Whether the material supplier is qualified

Correct Answer: B. Whether the numerical equations were solved correctly

Question 6

The skin depth of a good conductor decreases when

A. Frequency decreases

B. Conductivity decreases

C. Frequency or permeability increases

D. The conductor is removed

Correct Answer: C. Frequency or permeability increases

Question 7

Which feature commonly requires local mesh refinement?

A. A region with no fields or material interfaces

B. A feed gap or sharp conductor edge

C. A uniformly distant observation point

D. A purely graphical label

Correct Answer: B. A feed gap or sharp conductor edge

Question 8

A small algebraic residual by itself does not prove physical accuracy because

A. Residuals measure only how well the discrete system was solved

B. Residuals always equal measurement error

C. Residuals are unrelated to linear systems

D. Residuals apply only to thermal problems

Correct Answer: A. Residuals measure only how well the discrete system was solved

Question 9

Which assumption is required for the simple first-order variance propagation formula?

A. Inputs must be exactly zero

B. The response is locally approximately linear and uncertainties are suitably small

C. All variables must be perfectly correlated

D. The solver must use FDTD

Correct Answer: B. The response is locally approximately linear and uncertainties are suitably small

Question 10

For a one-port antenna, accepted power is ideally equal to

A. Radiated power minus dissipated power

B. Reflected power plus incident power

C. Radiated power plus loss power

D. Reactive power only

Correct Answer: C. Radiated power plus loss power