Introduction Describe the cycle of producing an electrical component. Describe the effect that an electronic component’s mechanical size has on how we model it. 1. 2. 3. 4. 5. a) b) Define specification First design Numerical simulation of components Evaluation Specs fulfilled? Yes -> move to manufacture No -> restart from 1-5 where appropriate It depends on the wavelength. If the mechanical size is smaller than the wavelength, we can use lumped-element models (RLC) and the network concept (voltages and currents). If the mechanical size is comparable with the wavelength, we need to consider distributed element and the field concept. Lumped element models aren’t sufficient. What kind of frequency bands can we expect to play with in electronics? What are the corresponding wavelengths? What is the difference between doing field simulation and circuit simulation? Draw the hierarchy for the numerical methods of solving Maxwell’s equations. Describe what ‘S-parameters’ are. How can we get S-parameters from a time- If the mechanical size is very large compared with the wavelength, we need to use optical principles (PO, GO, GTD). kHz scale to THz scale. Wavelength is calculated using speed of light. E.g. for kHz, c = 3x10^8 m/s, freq = 3000, hence wavelength = 3x10^8/3000 = 100km For THz, the wavelength is of the order of 0.1mm Circuit simulation: Advantages: Very fast, libraries available Disadvantage: Limited to available models, parameter values limited, no multimode propagation, coupling neglected (coupling can be a problem if components are close to each other) Field simulation: Advantages: Very flexible, can model arbitrary things, more accurate Disadvantage: More complex and slower simulation See ‘Overview’ slide, pg 17, intro lecture Dealing with the frequency domain. Over the range of frequency we have some curves about reflection and transmission. Use a Fourier transform to convert it to the domain signal? What is an SPDT? What does multimode propagation mean? What does 2.5D analysis mean? What is a planar circuit? What makes a calculation slow? frequency domain. Single Pole Double Throw. Switches a signal between one of two outputs. Usually we have a fundamental mode that we introduce into a circuit. However, if we have discontinuities in the circuit, then additional higher modes can be generated (some reflected, some propagated, some die out). Circuit simulation ignores these modes, but field simulation doesn’t. Analysis of planar circuits consisting of layered homogeneous media. Interconnects that connect different layers are usually handled as via conductor. A circuit that can be drawn in a plane without any component crossing over another (e.g. a circuit that doesn’t require vias). Number of unknowns, which depends on the model (level of detail) and the size of the problem. Then there’s operations, e.g. heavy math (matrix, eigenvalue ops). Equations and relations Write down the Lorentz force equation, and state it intuitively. Write down the quick notation for gradient (grad), and give its expanded definition as an operator. Write down the quick notation for curl, and give its expanded definition as an operator. What does the symbol E stand for, and what are the various possibilities for units? What does the symbol B stand for, and what are the various possibilities for units? What does the symbol q stand for, and what are its units? What does the symbol D stand for, and what are the possibilities for units? Describe Faraday’s The total force F experienced by a particle of charge q moving with a velocity v in a region of electric and magnetic fields, E and B is given as: F = qE + qVxB Grad f = (df/dx1,df/dx2,df/dx3…df,dxn) Where the d is the partial derivative and f is a scalar function. Upside down triangle without a dot or cross. It’s the upside down triangle with a cross. Need to take the determinant of the matrix. See Wikipedia. Electric field intensity N/C V/m Magnetic flux density (V.s)/(m^2) (Wb)/(m^2) Tesla T Amount of charge (e.g. the charge of a test particle) Coulombs Electric flux density (or electric displacement flux density) related experiment and give the formula. What does the symbol H stand for, and what are the possibilities for units? What does the symbol J (or Jc) stand for, and what are the possibilities for units? List the equation that relates J to E. What does the equation say intuitively? What does σ represent, and what are its units? What is the relationship between the vector fields D and E? What does ϵ represent, and what are its units? What does P represent, what does it mean and what are its units? How does P fit into Maxwell’s equations? What is the relationship between P and E? What is the relationship between the vector fields B and H? What is the meaning of u (mju), and what are its units? What is mju0 and its value? What does the symbol Mm mean, and what does it mean in real life? What are its units? Units: Coulombs per square meter (C/m^2) or (A.s/m^2) Faraday’s experiment involved a small sphere with +Q surrounded by a larger sphere, first grounded then taken away from ground, results in –Q on the larger sphere, and a flux density of D = phi/(4piR^2)a_r Magnetic field intensity, in ampere-turn per meter (A/m) Conduction current density Amperes per square meter (A/m^2) J = sigma E The conduction current density is proportional to the conductivity of the material, and is in the same direction as the electric field, and is proportional to the magnitude of the electric field. Conductivity of a material. Units are siemens per meter (S/m) D = epsilon E Material permittivity. Units of Farads per meter (F/m). Polarization density It is the electric dipole moment induced per unit volume of a dialectric material (which could either be induced by an electric field E, or could be permanent) Units of Coulombs per square meter (C/m^2) D = epsilon0 E + P Where epsilon0 is the permittivity of free space. This basically says that the electric (displacement) flux density would be equal to epsilon0 E in free space, but elsewhere we have to consider the phenomenon of polarization. P = epsilon0 (xsie) E Where xsie is the electric susceptibility, a parameter that signifies the ability of the material to get polarized. B = mju.H Permeability of the material, i.e. a measure of the ability of a material to support the formation of a magnetic field within itself. Units: Henrys per meter (H/m) or (Vs/Am) Mju0 is 4pi x 10^-7 It is the permeability of free space. Magnetization vector (or magnetic dipole moment per unit volume). This vector is zero in free space, otherwise it What is the relationship between Mm and H? List the four constitutive relations. Describe Faraday’s Law and write it in integral and differential form. What does it mean in real life? What does the M stand for in Faraday’s law, and what are its units? Describe Ampere’s Law with the Maxwell correction and write it in integral and differential form. What does it mean in real life? Describe Gauss’ Law for electric fields and write it in integral and differential form. What does it mean in real life? What does the symbol p stand for in Maxwell’s equations, and what are its units? What does the symbol pm stand for in Maxwell’s equations, and what are its unis? represents the effect of magnetization. Units of Amperes per meter (A/m) Mm = (xsim) H Where xsim is the magnetic susceptibility, a parameter that signifies the ability of the material to get magnetized. Jc = sigma E Jm = sigma* M D = epsilon E B = uH Where sigma* is the magnetic resistivity (see Num Techniques in Electromagnetics) A changing magnetic flux density B, or a magnetic current density M through a surface induces a circulating electric field intensity on the boundary of the surface. When a conducting material is present along the boundary of the surface an EMF results. Magnetic current density. Used in the symmetric form of Maxwell’s equations and assumes the existence of magnetic monopoles. Units: Volts per square meter (V/m^2) A magnetic field can be generated by an actual flow of charges (or conduction current density) and also by a changing electric displacement flux density. The electric flux emanating from a closed surface is equal to the charge enclosed by that surface. Volume electric charge density, in coulombs per cubic meter (C/m^3) or (A.s/m^3). Volume magnetic charge density. Used in the symmetric form of Maxwell’s equations, which assumes the existence of magnetic monopoles. Units are (V.s/m^3). State the law of conservation of charge and write The current due to the flow of charges down the corresponding equation. emanating from a closed surface S is equal to the time rate of decrease of charge inside the volume V bounded by that surface. State Lenz’s law. An induced electromotive force (i.e. E) always gives rise to a current whose magnetic field opposes the original change in magnetic flux. Write down Stokes theorem from vector calculus. Write down the divergence theorem from vector calculus. Write down all Maxwell equations in symmetric, See lecture slides. time-harmonic integral form. Write down all Maxwell equations in symmetric, See lecture slides. time-harmonic derivative form. State the continuity equation in the time domain, and state its meaning physically. State the continuity equation in the frequency domain. Write down the identity for (‘curl of the gradient’) Write down the identity for divergence of the curl. Write down the identity for divergence of the gradient. Write down the identity for curl of the curl. Derive the inhomogeneous Helmholtz equation in the time domain. List the formula for the speed of wave propagation (phase velocity). List the formula for the wavelength of an EM wave. List the formula for the propagation (or phase) constant of an EM wave, and describe its meaning. At any point in a given medium, the divergence of the electric current density due to the flow of charges plus the time rate of increase of the electric volume charge density is equal to zero. See Chapter 1 in Numerical Techniques in Electromagnetics. v = 1/sqrt(mju epsilon) Lambda = v/f Represents the change in phase per meter along the path travelled by the wave at any instant. Units of radians per meter. β = 2pi/lambda = omega sqrt(mju epsilon) Think about what the equation is saying…the constant is equal to 2pi/lambda. That is, every one wavelength we have 2pi radians of phase change. So, if we had a wave with a wavelength of 1 meter, then we’d have 2pi radians of phase change per 1 meter. Hence the 2pi/lambda. List the formula for calculating the wave impedence. List the formula for the Poynting vector, and describe what it means. Describe the skin effect and give the formulas for skin depth and resistance. What happens when the thickness of the material becomes smaller? Write down the expanded definition of the div operator. What is the definition of the magnetic vector Assuming a dielectric… Z = sqrt (mju/epsilon) P = ½ Re(E x H*) Describes the time-average power density for electromagnetic waves. The asterisk represents the adjoint (conjugate transpose). Skin effect is the tendency of an AC current to distribute itself within a conductor with the current density being largest near the surface of the conductor, i.e. the electric current flows mainly at the skin of the conductor at an average depth called the skin depth. Sd = sqrt(2/(omega mju sigma)) R=p.L/A If the thickness becomes smaller than 2Sd, then the boundaries overlap and the area becomes smaller, hence the resistance R increases roughly inversely proportionally to t. Otherwise R is roughly constant. Takes a vector and makes a scalar out of it. H = curl A potential A (German style)? Derive the inhomogeneous Helmholtz equation in the frequency domain based on electric sources. What is the definition of the electric vector potential F? What is the meaning of epsilon, and what are its units? What is epsilon0 and its value? Imagine we have a wave coming from air and penetrating a metal. What happens to the speed of the wave and the wavelength? What does this mean with respect to discretization? How can we take account of losses in Maxwell’s equations? What does it mean if a wave is x-polarised? Discuss the boundary equations for the case where a wave is normally incident on a surface that separates 2 media. Let n be the normal vector between the media. What is the purpose of the Helmholtz equation? Use the magnetic vector potential A. Use the first two Maxwell’s equations. Use the Lorentz condition. Use the definition of the wave vector k. -E = curl F Permittivity, a measure of the resistance that is encountered when forming an electric field in the medium. Units are Farads per meter (F/m) or (A.s/V.m). Epsilon0 is the vacuum permittivity, equal to 8.85x10^-12 Mju and epsilon are higher in the metal than in air. The speed is hence smaller, and the wavelength is “much much smaller” for a metal than in free space (and also smaller than for a dielectric). Hence we have to discretise really fine because the wavelength is small. However…for highfrequency the current and field are confined mainly on the surface (skin effect)…it doesn’t penetrate on the inside. So…if we want to discretise, yes, we need a small discretisation, but we just do it on the surface by using the surface impedance. If we consider permittivity and permeability to be complex we can take account of losses. Say the wave is travelling in the z-direction. Polarization describes the path taken by the tip of the electric field intensity vector E in a plane orthogonal (perpendicular) to the direction of propagation. Thus, if we take a slice at z=0 and watch E as time progresses, we’ll see the vector bounce around in +x and –x if it’s x-polarized. Easy. If we have no surface current, then the magnetic and electric field intensities are preserved between the two regions: n x (H1 – H2) = 0 (J if there is a surface current) n x (E1 – E2) = 0 (-M if there is a surface current) n dot (B1 – B2) = 0 (rho_m if there is surface charge) n dot (D1 – D2) = 0 (rho if there is surface charge) Without Helmholtz, we have many equations with six unknowns. With it we have one equation with two unknowns (two components of the vector A). The trade-off is that Helmholtz is a 2nd order DE. How does the Helmholtz equation get even simpler in some cases? Once we have determined the vector potential A by solving the Helmholtz equation, how can we calculate the magnetic and electric field intensity vectors? What are the duality quantities for H, E, J, A, epsilon, mju, rho and omega(frequency omega)? Where are rectangular waveguides used, and why? When do we get TEM waves? What is a Green’s function? Is it possible for a rectangular waveguide to transmit DC waves? What can we say about the fields at the surface of a perfect conductor? Allows us to simplify our differential equations. Time-independence, no charges -> Laplace equation Time-independence, with charges -> Poisson equation Time-dependence, no charges -> Wave equation See 2.3.1 in the slides…also look it up in a book! -E, H, MA, F, mju, epsilon, rho_m, omega(frequency omega…same) High-power transmission…used because there is NO RADIATION, and it can be cooled :). We get it always if we have two separate conductors (e.g. coaxial cable). A rectangular waveguide only has one conductor, so it’s either TE or TM. The Green function G(r) is equivalent to the impulse response h(t) of system theory. Just as h(t) gives the response (in time) to a temporally impulsive source, so G(r) gives the response (in space) to a spatially impulsive current source. The response to a spatially distributed source is obtained by integration, and plays the same role in space that convolution in system theory does in time. No. The electric field components that are parallel/tangential to the conductor must be zero. In addition, the magnetic field components that are normal to the conductor must also be zero. Describe the general process for solving for all the field components in a rectangular waveguide. Describe a Herzian dipole. Thus, the electric field is normal and the magnetic field tangential at the surface of a perfect conductor. It is sufficient to just find the z-component of the vector potential (assuming the propagation direction is z). From that, we can determine the rest of the field components. A Hertzian dipole is a wire of infinitesimally small-length. At the end of the wire, two equal and opposite charges exist. The current in the wire leads to a magnetic field encircling the wire. If our source point is at r’, and our observation point at r, then the magnetic vector potential at point r is related to the conduction current density J and the scalar Green’s function. Consider a radiating Hertzian dipole. What are the non-zero field components in spherical coordinates at a distance r? Consider a radiating Hertzian dipole. At a far distance, which field components are dominant? Consider a radiating Hertzian dipole. At a near distance, which field components are dominant? What does the equivalence principle state? Give the formulas for the momentary and average-in-time quantities of the following: Electric energy density Magnetic energy density Joule’s heat Poynting vector The resulting magnetic field in spherical coordinates only has one non-zero component. The resulting electric field has two non-zero components. Er, Etheta, Hphi Etheta, Hphi Etheta, Er If we have a surface current on a closed volume, we can forget about what’s happening inside the volume. We just need the tangential components of E or H…sufficient for unique description. See slide 36/37 of physical basics. Mathematics Give an example of an inner/closed/bounded problem in electromagnetics. Give an example of an outer/open/unbounded problem in electromagnetics. Give the general format of a second-order partial differential equation in a domain x and y, and describe when it’s linear, and when it’s homogeneous/inhomogeneous. Write out Laplace’s equation in 2 dimensions. Write out Poisson’s equation in 2 dimensions. What kind of electromagnetics problems are described by Laplace’s equation? What kind of electromagnetics problems are Transmission in a waveguide. We know that the parallel/tangential components of the electric field on the PEC is zero, and the magnetic field normal component is zero. There’s also no radiation outside the waveguide, so hence this is an inner problem. Radiation from an antenna. See slide 6/34. D^2/dx^2 (f) + D^2/dy^2 (f) = 0 D^2/dx^2 (f) + D^2/dy^2 (f) = g(x,y) Steady state static configurations with no source term. Steady-state with a source term (e.g. one wall described by Poisson’s equation? Explain the difference between Dirichlet and Neumann boundary conditions. Derive the wave equation for the electric field intensity E. Use the series expansion technique to solve the wave equation at a high level. with a potential on it). In order to have a unique solution to a DE, we have to impose boundary conditions. If we specify the function value on the boundary explicitly, this is Dirichlet boundary conditions. If we don’t specify the function value itself, but rather its normal derivative on the boundary, this is Neumann boundary conditions. Start with the curlE equation and take the curl of it. Use the vector identity to turn it into a 2nd order PDE. Should end up with a result similar to slide 11 CEM_3, aside from a few constants. See slide 11 onwards, CEM_3. CEM_4 How does the method of moments express our unknown solution to make it easier to compute? What are the two general choices that we have for basis function types in the method of moments? What are the disadvantages of using entiredomain basis functions for the method of moments? What does the Galerkin condition in the method of moments state? Given a system Ax = b. Give the formula for calculating each x component using Cramer’s rule. Given a system Ax=b with A square of dimension n. What’s the time complexity to solve the system using Cramer’s rule? Given a system Ax=b with A square of dimension n. What’s the time complexity to solve the system using Gaussian Elimination? What’s the advantage of using Cholesky decomposition over Gaussian Elimination? Describe an electric wall. What are its boundary conditions? It expresses the unknown solution as a sum of known functions (expansion functions or basis functions) with unknown coefficients. Entire domain basis functions (orthogonal, like sin, Chebyshev, Bessel, Lagrange etc.) Subsectional basis functions Loss of generality. In order to use entire-domain basis functions, we have to know something in advance about the field distribution, which means a loss of generality. Test function = basis function Xi = det(Ai)/det(A) Where Ai is the matrix formed by replacing the ith column of A by the column vector b. O(n^4) O(n^3) If we want the determinant of A, we can get it simply by multiplying the main diagonal elements of L, i.e. L11*L22*…*Lnn. The electric wall assumption implies that the tangential component of the E field is zero. The normal component of E is non-zero. The normal component of the magnetic field is zero…it is non-zero in the tangential components. N cross E = 0 N cross H = Jsurface Describe a magnetic wall. What are its boundary conditions? N dot D = psurface N dot B = 0 The magnetic wall assumption implies that the tangential component of the H field is zero. The normal component of H is non-zero. The normal component of the E field is zero…it is non-zero in the tangential components. N cross E = 0 N cross H = Jsurface N dot D = psurface N dot B = 0 CEM_5 When solving matrix equations, we would like that our matrix be diagonally-dominant. Give an example where it’s a problem when we don’t have a diagonally-dominant matrix. Let’s say we have an equation Ax=b. We want to solve the equation by taking the inverse of A, i.e. x = (A^-1)b. The dimension of A is n. What does MATLAB actually do to solve the inverse? What’s the time complexity? When is the only case where we would want to compute the inverse of a matrix? Give the formula for an eigenvalue problem. What are the two possibilities for general analytical solutions to the eigenvalue problem? Consider Ax=b A = [1 2] [2 1] B = [1;1] This doesn’t converge well. Basically we try to solve the equation A*(A^-1) = I We thus have n linear systems to solve to find the inverse! Assuming we can solve each linear system in n^3 for GE, for example, then it will take n^4 for a solution to the inverse problem. If we want to use the inverse of the matrix to solve the problem for a bunch of different b vectors. [A – lambdaI][x] = 0, or Ax = lambda [A – lambdaI][x] = 0 So, this is solved if x is zero (trivial), or Det(A – lambdaI) = 0 (more important!) CEM_5 What’s the basic idea of the Mode Matching technique? To which problems is the Mode Matching The idea is that part of the solution is analytic, and part is solved numerically. This results typically in much smaller matrices. Waveguides and optical structures. technique mainly applied? What are the two parameters which characterise a transmission line, that we’re normally interested in determining if we’re looking at e.g. a microstrip line? To which kind of problems is the MMT applied in 2D? When do we have to be careful in applying a magnetic wall in a simulation? Draw an example of a coplanar line and a microstrip line. Describe a coplanar strip/line in words. What is a dielectric? What’s the difference between an insulator and a dielectric? How do we stop unsymmetric modes propagating in a coplanar line? When we use the MMT for solving the Helmholtz equation for a coplanar line, are there different propagation constants per region, or do we use a propagation constant for the entire structure? List the general steps for applying the MMT for solving for the transmission line characteristic variables (propagation constant, wave impedence) in a coplanar line. Propagation constant, wave impedence. 2D cross-section of a coplanar line, microstrip line. We have to make sure that the problem is truly symmetric. In some cases, like a coplanar line, unsymmetric modes might propagate, and if we use a magnetic wall then we might lose these. See OneNote L5. Consists of two conducting parallel strips or traces mounted on the same side of a dielectric substrate or PCB. A dielectric is an electrical insulator that can be polarized by an applied electric field. When a dielectric is placed in an electric field, electric charges do not flow through the material, as in a conductor, but only slightly shift from their average equilibrium positions causing dielectric polarization. Insulator implies low electrical conduction. Dielectric implies high polarizability. Bring the lines to the same potential. This can be done with very thin bond wires, which connect the two at points of discontinuity. Entire structure. Partition the structure into homogeneous sections, each one with a constant relative permittivity. Write down the Helmholtz equation for the potential in each region. Do a series expansion in each region with orthogonal basis functions that match the boundary conditions. Match the potential expressions at interfaces and multiply with test functions in the MoM style. If Galerkin is used, only one side will see the test functions be orthogonal with the basis functions. Move everything over to the LHS and build up a matrix M with unknown kz entries, such that MA = 0, where A has the amplitudes (coefficients). Describe the ‘relative convergence’ issue in MMT. Give an example of a 3D eigenvalue problem in MMT. Is the MMT used for antennas? Explain. List the pros and cons of the MMT. Solve for det(M) = 0…gives kz. Then we can find the A values, hence the potentials, hence the propagation constant and the wave impedence. Consider a 3D rectangular waveguide with a discontinuity. The idea is that we have to match the relevant field components at the discontinuity using MoM. For each of the two regions in which we are matching, we are free to choose N and M, the respective number of modes we want to include in our approximation for each region. N and M should be chosen considering the physical size of the regions and the problem. If they’re not ‘balanced’, then unphysical solutions may result. This is the ‘relative convergence’ issue. Resonator, i.e. a closed waveguide without input source. No. MMT doesn’t consider radiation well. We can’t break down the field behaviour into sine/cosine functions in this case. Pros: Low numerical effort, as half analytical Very well suited for homogeneous transmission lines, waveguide discontinuities and optical structures Finite conductor thickness/loss can be considered Modelling of field inside the conductor is possible Cons: What are the sources of error in the MMT? No consideration of radiation (bad for antennas) How many modes do we have to consider (heaps for a small pin in a waveguide)? What to do with the tuning screws? Truncation due to the chosen number of modes. Staircase approximation of round objects. Error in evaluation of integrals. “Relative convergence” issue. CEM_6 Describe the flavours of the finite difference method. Describe which equations they each solve, which form of the equations they use, and in which domain they operate. Give the FDFD: Finite difference frequency domain. Based on differential form, Helmholtz equation, frequency domain. abbreviations of each. Describe the trade-off between discretization and round-off error in choosing a mesh width in the finite difference method. List one way for increasing accuracy in the finite difference method. We’re given a 5-point 2D stencil and asked to find the expression for the point on which the stencil sits in terms of the other four points. The point lies on the edge between two materials with epsilon1 and epsilon2. Find the expression for the field at the point. We’re given a 5-point 2D stencil and asked to find the expression for the point on which the stencil sits in terms of the other four points. The point lies on the corner between two materials with epsilon1 and epsilon2. Find the expression for the field at the point. Let’s say we want to find the modes of a hollow waveguide using the wave equation. Is this an eigenvalue problem or a deterministic problem? TE and TM modes use m and n subscripts to indicate the mode. What do the m and n refer to? How do we get solutions for a rectangular waveguide using the finite difference method? FDTD: Finite difference time domain. Based on integral form, Maxwell’s equations, time domain. If the mesh width is very small, then the values of neighbours may be very similar, and taking the difference of two similar values gives a large error (ill-conditioned). If the mesh width is too large, the discretization error comes into play (this is the O(h^2) factor for central differences, for example). A trade-off must be reached which minimises the error. Mesh refinement. General rule is to use a denser mesh at the edges. Also, the wavelength is dependent on the frequency, and we usually need a discretization size around 1/10 of the wavelength. Thus, we need fine discretization for higher frequencies. 2(eps1 + eps2)*x0 = 0.5(eps1+eps2)x1 + eps1*x2 + 0.5(eps1+eps2)x3 + eps2*x4 2(eps1 + eps2)*x0 = 0.5(eps1+eps2)x1 + eps1*x2 + eps1*x3 + 0.5(eps1+eps2)x4 -laplace u = alpha u Where u = the vector potential Alpha = k^2 – B^2 Hence, eigenvalue equation. M: number of half-wave variations of the field in the x-direction. N: number of half-wave variations of the field in the y-direction. ‘Solutions’ in this case are modes, and corresponding propagation constant. Depending on whether or not we want TE or TM waves, we have different boundary conditions. For TM, we have dirichlet boundary conditions…i.e. the vector potential has to be zero on the boundaries. For TE, we have Neumann boundary conditions, i.e. the vector potential has to have a zero gradient on the boundary. Then we just set up the matrix based on the finite difference stencil we’re using. We can find the determinant of the matrix, set it to zero, and solve for the eigenvalues. If we have N points in our domain (mesh), then we’ll get N eigenvalues out of it! What are the pros and cons for the finite difference method? How can we use Dirichlet and Neumann boundary conditions in finite difference method to reduce the computational domain? What’s the trade-off? Each eigenvalue represents a mode, and each eigenvector the corresponding point values for that mode. Pros: Flexible, general tool Low analytical effort Large application domain Good for eigenvalues Cons: Large matrices in 3D Modelling errors Discretisation errors Roundoff error Truncation error of iterative solver. Dirichlet boundary <-> electric wall Neumann boundary <-> magnetic wall Because we reduce the domain to half, this means that only half of the modes are computed. CEM_7 Write down the expressions for dHx/dt, dHy/dt, dHz/dt, dEx/dt, dEy/dt, dEz/dt using Maxwell’s equations in time domain. Draw Yee’s cell. Describe the Leapfrog algorithm used in FDTD. See slide 4/41 of CEM_7. See slide 6/41 of CEM_7. Consider Yee’s cell. We have alternating E and H components in each direction. That is, each E component has only H component neighbours. At timestep n, we update the E field based on the timestep (n-1) E field and the timestep (n0.5) H field. Thus, in timestep n, we only update E values. At timestep n+0.5, we update the H field based on the timestep (n-0.5) H field and the timestep n E field. List the stability criterion for the FDTD algorithm in 1D, 2D and 3D. Thus leapfrog. (c*deltaT)^2 <= deltaX^2 (1D) (c*deltaT)^2 <= [1/(deltaX^2) + 1/(deltaY^2)]^-1 (2D) Describe broadly what the difference is between the FDTD method and the FD method. (c*deltaT)^2 <= [1/(deltaX^2) + 1/(deltaY^2) + 1/(deltaZ^2)]^-1 (3D) FD is in frequency domain. We take Helmholtz’s equation, discretise in space, and we solve a big matrix once. FDTD is in time domain. We take Maxwell’s equations in time domain, and we iterate in time (leapfrog algorithm). At each timestep, we can update each field component without having to solve a matrix. That is, FDTD uses explicit formulation of each field component at each timestep. How do we get scattering parameters from a FDTD output? What are the pros/cons of FDTD over other methods? For FDTD, we have to excite the system, usually done with a Gaussian pulse. FDTD output is a bunch of vectors. Each vector represents a particular field component variation over time. We can get the scattering parameters by doing a Fourier Transform of this data. Pros: No ‘heavy math’, i.e. no matrix equations or eigenvalues to solve Doing the Fourier transform of the timedomain signal gives us the full spectrum, which we wouldn’t get using a frequency domain method (we choose the N in frequency domain methods) Cons: Write the formulas for an FDTD Gaussian timeexcitation pulse in time and frequency domain. What’s the best choice for a time-excitation pulse in FDTD? When do we truncate the Gaussian timeexcitation pulse in FDTD to zero? Many timesteps, which are dependent on the spatial resolution to ensure stability. Finest detail determines timestep E(t) = e^(-(t-tc)/sigma)^2 E(w) = sigma*sqrt(pi)*e^(-(w)/(2/sigma))^2 Gaussian pulse. Step signals (sharp edges) lead to unphysical behaviour and instabilities. When it essentially gets down to machine precision. There’s a formula for figuring this out based on the finest detail in the structure and the dimensionality of the problem. Write the formula for a sine-modulated Gaussian time-excitation pulse used in FDTD. Draw a rough sketch of this pulse. Why might we want to choose this instead of a standard Gaussian pulse? Write the formula for a single-frequency timeexcitation pulse used in FDTD. Draw a rough sketch of this pulse. Why might we want to choose this instead of a standard Gaussian pulse? We have a 3D structure that we want to solve using FDTD. Describe the process of exciting the structure. Bandlimited spectrum is created. See slide 21/41 of FDTD. Sinusoidal excitation will give only one usable frequency point. This isn’t usually what we want. See slide 22/41. We need to define something like a ‘port’, which is a 2D plane on which our input field components (electric, magnetic) are defined. Before we start our simulation, we have to get a spatial distribution on that plane of the field. This is then a 2D eigenvalue problem, which we can solve. Otherwise, we can input an artificial field distribution and let it run for some time until we get the ‘natural’ field distribution. Either of these two methods then gives us the full-amplitude ‘imprint’ excitation for the structure. What happens if we truncate our input signal too early in FDTD? What is the tradeoff when choosing the port size for the input signal in FDTD? How can we avoid numerical effort in modelling conductors in FDTD? Describe the technique. What’s the problem? We then choose a time excitation pulse, like a Gaussian, and we start our simulation. The Gaussian basically acts as a filter for the amplitude of the imprinted spatial excitation…starting at zero, and going up to the nominal amplitude, then going back to zero. We get ripples in the frequency domain, because a rectangular pulse (i.e. the cutoff rectangle) transforms to a sinc function (sinx/x)in the frequency domain, and these ripples will then overlay on our frequency plot…bad. If we choose it too large, we can excite more than one mode, which is bad. If we choose it too small, the electric walls that we use on the boundary of our port will change the field distribution from the ‘real’ distribution, and affect the impedence value. Impedence boundaries. Considers the surface impedence of the conductor, assuming that the skin effect holds and the fields are concentrated on the boundaries. The problem is that the field components don’t exist on the same grid points, so we can’t get a reasonable approximation for the surface impedence. To solve this we can use interpolation. In FDTD, what’s the problem with using the easiest solution of electric/magnetic walls (Dirichlet and Neumann boundary conditions) in our domain? What kind of box boundary conditions are the best to use with FDTD problems? Explain the Liao extrapolation technique for implementing absorbing boundary conditions. What are the problems with the Liao extrapolation technique for doing ABCs in FDTD? What are two ways to implement absorbing boundary conditions in FDTD solvers? Explain the perfectly-matched layers technique for implementing ABCs in FDTD solvers. List the pros and cons of FDTD. Of course, if the conductor is very thin, the skin effect assumption doesn’t hold any more. If the metallic walls are close to the structure under investigation, the field distribution is disturbed, characteristic impedence and propagation constant will be affected. Resonances will shift. If metallic box is large: box modes (resonator modes of the metallic resonator) occur…like a waveguide. In larger boxes, the cut-off (‘cut on’) frequency becomes lower. This gives spikes in the scattering parameters resulting from the box modes, nothing to do with the structure. ABCs…absorbing boundary conditions. Polynomial extrapolation of field data at interior grid points and past time steps to the desired outer-boundary grid point at the latest time step. Marginally stable Errors above 40dB Quality dependent on angle of incident wave. Liao extrapolation technique. PML (perfectly matched layers) The idea is that we add some additional boundary points, which are lossy. The idea is that this acts like a ‘sponge’ to soak up the waves in this region. At the end of the lossy material we still have a PEC which causes reflections, but by the time it gets back the amplitude is hopefully negligible. Pros: Flexible, general 3D tool Lowest analytical effort Large application domain Easy implementation of lumped elements Large matrices, but simple operations Fast for broadband structures Parallelization easy Cons: Pre- and post-processing necessary (pre: input port definition, post: Fourier transform to get the scattering parameters) Open structures need special boundaries Rectangular mesh: staircase List the sources of error in the FDTD method. approximation Iterative process: long calculation times More difficult to model losses in dispersive materials…we have to model the entire conductor if we want it very accurate. If we use the surface approximation then it’s not as good. Discretization error. Truncation of signal (time domain) Box modes (closed environment) Imperfect boundaries Modelling errors Roundoff errors User! CEM_8 What’s the difference between FIT and FD/FDTD? Derive the FIT algebraic equations considering the Yee-style cube. Write them next to the corresponding Maxwell integral equation. Let’s say we do an FIT run and we get a result. If the result is with respect to D, how do we get E and the path-line integral-variable of E? If we use the time-domain algorithm in FIT, what’s the difference between it and FDTD? In FIT we start with Maxwell’s equations in integral form. In FD/FDTD we start with Maxwell’s equations in derivative form. In FIT the quantities that are modelled are integrals, rather than fields. See slide 10/18 in CEM_8. The result is the integral of D over the surface area, so we have to divide by the surface area to get D, then divide by epsilon to get E, then multiply by ds to get e, and this is where we lose exactness. This is how we can relate D and E, but there are approximations in between. In FIT, the variables are the integrals over the line, not the field values themselves. CEM_9 In which domain, frequency or time, does the MoL operate? In principle, MoL is a hybrid of which two other methods that we have studied? List the similarities between MoL and MMT. Frequency. MMT and FD. It basically uses FD to discretise in 2D, and then the 3rd dimension is analytically calculated, like with MMT. Both are semi-analytical (half-numerical) methods. Both are suited for certain classes of structures. True or false. MoL is based on finite differences. Let’s say we want to use MoL to analyse an infinitely long homogeneous transmission line, e.g. a rectangular waveguide, microstrip, CPW etc. What is our goal? What do we want to calculate? Which equation do we have to solve, and how? Which discretization scheme is used in MoL, and along how many dimensions do we discretise? We discretise in only 1 dimension in MoL in a 3D problem…great! What’s the catch? When we do a 1D discretisation in MoL, do we have the same field value on each line? Describe the basic steps for solving a multilayered problem using the MoL. Both require higher effort to derive and understand the methods, as analytical calculation is combined with numerical computation. True. Propagation constant kz Characteristic impedence Field distribution in the cross section We solve the Helmholtz equation and get the kz from an eigenvalue calculation. Uses central finite differences, and discretization is performed in one direction. E.g. we discretise in the x-direction of our waveguide, and calculate analytically in y and z. Analytic calculation is only possible in a homogeneous region, so we have to separate into layers. No…just like in FDTD with the cube, we alternate field components. Discretise into lines in the x-direction. Use alternating solid/dashed lines to represent Ez and Hz components respectively For each layer…that is, each horizontal slide where there is no change in the y-direction, do… 1. Set up the differential equation (Helmholtz) and substitute in the derivative wrt x (finite difference). This gives us a coupled set of ODEs. 2. Do the mathematical transform of the potential matrix to the principle axis to get a diagonal matrix, and hence a system of uncoupled ODEs. 3. Solve the eigenvalue problem by taking the determinant and finding kz and the transformed E and H fields. This is then basically the MODES that we get! Each mode is propagating with its own propagation constant. 4. Propagate the modes (‘transform the impedence in transmission line theory’) analytically across the homogeneous layer section in the y-direction to the next layer interface. 5. At the interface we ‘normalise the impedence’ by tranforming back to get the real E and H fields, considering ALL the modes that we have from the entire layer. What can we ‘leave out’ in our solution procedure during MoL if we want to solve for the fields in an open hollow waveguide? When we do our geile transformation in MoL to decouple the system into a bunch of ODEs, what are we actually doing in a physical sense? What’s the difference between finding modes in the MMT and MoL? That is, is one better or more flexible than the other? Describe the technique used in 2D MoL to transfer from the top and bottom layers to the matching plane in order to find the propagation constant. 6. Match the E and H fields with the next layer using the continuity of the tangential field components and repeat :) In an open hollow waveguide, we either have TE or TM modes. Thus, we don’t need alternating lines for Ez and Hz, we just need one! We’re transforming the equation from spatial domain into modal domain! In MoL, we can deal with one layer at a time, and it can be inhomogeneous in the x-direction. This will then give us the modes of that layer in one swoop…solving the eigenvalue problem! Awesome! In MMT, we had to match the modes at every inhomogeneity. Usually we know the values of the field/impedence at the top and bottom of our layered structure. We start at the top and bottom and follow the normal MoL technique before converging at a matching layer. At the matching layer, we establish a relationship at the matching plane: Z*J = E Now we know that the tangential E must be zero at this metallisation, so we basically have to find the determinant of Z where it’s zero. This gives us the propagation constant. (Er + 1)/2 In the 2D MoL, if we have a microstrip line with substrate and air, what will be an approximate to the effective epsilonRE of the fundamental mode That is, the field is partially in air and partially in of the line? the substrate. We can use this as a starting point in our root-finding algorithm at the matching plane. What does the term ‘dielectric constant’ mean? Synonym for relative permittivity. How do we separate a structure for use with 2 options: MoL in 3D? 1. Separation into longitudinally homogeneous sections, with the cross section allowed to be inhomogeneous. The lines are then in the propagation direction 2. Separation into x-axis-oriented layers (i.e. horizontal slices of bread), with the layers allowed to be inhomogeneous in the x and z direction, where we use discretization. How does MoL work in 3D? What’s basically the only ‘con’ of the MoL? In the MoL tutorial, we discretised with four types of points, filled circles and triangles, and empty circles and triangles. What was the purpose? Actually, we could do it in any dimension we want, just make sure that we look at the problem and see which direction makes the most sense. Basically the same as in 2D, but we use the Kronecker product for the matrices because we’re doing discretisation in 2D this time. Low flexibility…different algorithms for different structures. Best for layered structures or long, thin structures. Because we need the derivatives on neighbouring points. CEM_10 Is an electric structure temperature agnostic? Is the FEM usually used in time or frequency domain? Do we use Maxwell’s equations directly in the variational approach of the FEM? When we use the FEM method with the variational technique, we try to minimise the field energy to find the potentials at all the points. Write down the equation for the energy of the electrical field. How do we get from the electric field intensity and permittivity to the potential? Derive the FEM equations for a single triangular element that leads to the eventual matrix to be solved. We’re doing FEM and we have found our matrix and now we want to insert the boundary conditions for certain boundary points. How do we go about it. When do we have an eigenvalue problem, and when do we have a deterministic problem? What is the main FEM approach to use when we have electrostatical problems with a charge distribution? What about the case of time- No…usually conduction properties, impedence changes with temperature. Frequency domain (i.e. harmonic problems). No, we use an energy-minimisation approach. See slide 26/64. See slides 26-31 of CEM_10. For the row of the matrix corresponding to that potential point (i.e. row n for point n), we remove all entries in that row, put a 1 in the diagonal entry on that row, and put the known potential value for that point on the RHS (i.e. in the b vector). When the vector on the right is a zero vector, we have an eigenvalue problem (we need to find the determinant of the matrix and set it to zero and solve). When the vector on the right is a nonzero vector, we have a source term and it’s a deterministic problem. For the static case we can use the variational approach with field-energy minimisation. For time-dependent fields…don’t know yet! Finish dependent fields? List the FEM pipeline from problem to solution. In which case do we have to do preprocessing of the input in FEM? List some of the rules for good triangular meshes in FEM. this! 1. Preprocessing of input 2. Proper choice of elements 3. Derivation of the governing differential equation 4. Discretization of the 2D domain 5. Derivation of the element matrices and vectors 6. Assembly of the global matrix system 7. Imposition of boundary conditions 8. Solution of the global matrix system 9. Convergence? If yes, do postprocessing, otherwise repeat from 3 with finer discretization. When we have a deterministic problem with an excitation. Basically the same idea as the FD method where we calculated the field distribution on the input port. We have to define the ‘mask’ of the field components to imprint for excitation. For a triangular mesh, the shape of triangles must be close to equilateral Nodes must appear at source points In regions where the solution is expected to have large variations, the elements must be sufficiently small Avoid elements with very large aspect ratios, i.e., the ratio of the largest side to the smallest side Number the nodes in ascending order starting from 1. The numbering of the nodes directly affects the bandwidth of the global matrix There must be no overlap of elements Neighboring elements must share a common edge List the pros and cons of the FEM. An interior node (non-boundary node) must belong to at least three elements Pros: Flexible choice of elements For round and irregular geometries Cons: Elaborate mesh generation (although adaptive mesh refinement possible) Possible unphysical solution (spurious What is a functional, in mathematics? When we search for a state of a system that minimizes the energy functional, what are we doing exactly? modes) It is a map from a space of functions to output values. That is, a functional takes a function for its input argument, so it’s basically a function of a function. We look for a function that makes the functional stationary, i.e. extrema. CEM_11_IE List the general steps of the integral equation method. Do box modes start happening for lower or higher frequencies as the box size is increased? What is the basic idea of the integral equation method? What key properties of a 2D metallisation problem do we use in the integral equation method to derive the equations that we use? 1. Derivation of the appropriate integral equation (different depending on whether we’re in static/dynamic case) 2. Conversion (discretization of the IE into a matrix equation using basis (or expansion) functions and weighting (or testing) functions. This involves the MoM. 3. Evaluation of the matrix elements 4. Solving the matrix equation and obtaining the parameters of interest Lower. Reduction of a 3D problem to a 2D problem, defined on the surface of a conductive scatterer. That is, everything is defined on the surface of the metallisation. The 2D problem is described by an integral equation, where currents and tangential fields on the surface are connected via the dyadic Green’s function. So, we have a surface metallisation which has an unknown current J. We can get the electrical field at any position r (outside the plate) using the integral over the Green’s function multiplied by the current, and adding the incident field (see slide 5). However, we don’t know the current! So what do we do? Well, we know that the tangential electric field on the conducting plate has to be zero. This then allows us to calculate the current distribution on the plate, by satisfying the condition that the tangential component of the total electric field vanishes on the plate . Once we have the current distribution on the plate, we can calculate the electric field anywhere off the plate using the same equation. In the integral equation method, how many dimensions are involved in evaluating the integral of each matrix element? Can the integral equation method model dielectrics? What’s the best method to use for modelling radar reflections on large objects? What determines the complexity of the calculation of the Green’s function in the integral equation method? Describe the spectral domain approach in the integral equation method. 4 No…can only model metallisations. Integral equation method. Whether or not the system is in an open environment or a closed box configuration. If it’s closed box, the Green’s function can be expressed by sine and cosine functions, but then we probably have the issue with box modes etc again. So in the integral equation method we have an integral equation (wow) which relates the electric field to the current via the Green’s function. The Green’s function is usually not really known. So we convert everything into the spectral domain by doing some Fourier transforms. Is the integral equation method a local or a global method? Write the Coulomb integral describing the potential of charge density in free space for the static case. What’s the difference in the matrices that come out of the IE method compared to the matrices that come out of the FEM or FDTD methods? What determines the size of the matrix to be solved in the IE method for a 2D conducting surface? What are possible sources of errors in the IE method? Why do we do this? By transforming into the spectral domain, we eliminate the integral, because it’s a convolution integral or something. Usually a global method. The source point is coupled to each observation point. => dense matrix. See slide 19/34 CEM_11 Matrices are smaller in IE than in FEM or FDTD, but they’re dense. Also, there’s a matrix-fill stage in IE that’s in O(n^2). Determined by the number of elements (i.e. small areas) on the conducting surface. List the pros and cons of the IE method. Pros: Numerical calculation of the Green’s functions. Truncation of series of basis functions (i.e. granularity of the discretised areas in the domain). Box modes in a closed environment Modelling errors (lossy thick conductors, vias etc) Roundoff errors Accurate modelling of radiation (open) Fast-end efficient for layered structures or structures in free space Cons: Adaptive frequency sweep possible. High numerical effort for problems with nonplanar components like dielectric blocks Possible high numerical effort for calculating Green’s functions. CEM_12 Are asymptotic methods for high or low frequencies? What does MLFMM stand for? Describe the MLFMM. What are the pros/cons of MLFMM? High you idiot. Multi Level Fast Multipole Method. Multi Level Fast Multipole Method. Basically the electromagnetic equivalent of the FMM in molecular dynamics. It’s an extension of the Integral Equation method. Instead of coupling every source term to every observer, we use a tree structure to group source terms together, and a tree structure to group observer points together, and only do the interactions at high levels of the tree. Then the interaction results are distributed down through the tree considering distances. Pros: Much faster than MoM/IE Applications to scattering problems, Radar cross sections Parallel computing possible Same meshing as MoM/IE Cons: Give an overview of the high-frequency approximations PO, GO, GTD and UTD with respect to basic properties. Inaccuracy by far field approximations Only applicable when object >> wavelength Physical optics: Current-based, incoming field is treated locally as a planar wave, good for electrically large problems without edges Geometrical optics (ray optics): Field (ray) based, nature of waves not considered, good for electrically large problems without edges Geometric theory of diffraction: Consideration of edges, corners, inaccurate at shadow boundaries Uniform theory of diffraction: Shadow boundaries Describe physical optics in detail. For high frequency only (wavelength >> object size) CURRENT-BASED! Ignores wave effects Incident field is treated as a planar wave. We calculate the current on each point as the current that would be found on a tangent plane of similar material at that point. Current in shadowed regions is taken as zero (no diffraction). Once we have the current, we then use it in the radiation integrals to compute the scattered far field from the target. NOT accurate for edges (how would we define a tangent plane?). What do we mean by a ‘ray-based’ technique in high-frequency methods? Which is a simpler approach in high-frequency methods: GO or PO? Describe GO in detail. PO works well for large, smooth surfaces with low curvature. We don’t worry about calculating the currents in order to find the reflected fields…we just reflect the ray and that’s it. GO…because we need to calculate currents in PO. For high frequency only (wavelength >>> object size) Wave nature of light is not considered…we use rays…RAY-BASED! Each incident ray is reflected directly and superimposed to get reflected electric fields. Current in shadowed regions is taken as zero (no diffraction). What are the limitations of GO and PO? What’s the solution? Describe GTD in detail. We don’t know what to do at edges (same tangential problem as PO). Can’t deal with edges! Solution is GTD and UTD! Geometrical theory of diffraction. Once again, only works if the dimensions of the body >> wavelength. Overcomes the limitations of PO and GO by introducing a diffraction mechanism. GTD assigns a diffraction coefficient to a point depending on the geometry (i.e. whether it’s an edge, corner/tip or curvature). The coefficients are based on known canonical geometry tests. Draw an example of a shadow and a reflected boundary in GTD. What is the limitation with the geometrical theory of diffraction? What’s the solution? Describe the uniform theory of diffraction in detail. See slide 27/34. GTD produces inaccurate results at the shadow boundaries. UTD is a solution! UTD approximates near EM fields as quasioptical and uses ray diffraction coefficients for each diffracting object-source combination. These more complex coefficients are then used to calculate the field strength and phase for each direction away from the diffracting point.