Chapter 9: Electromagnetic Waves 9.1 Waves at planar boundaries at normal incidence 9.1.1 Introduction Chapter 9 treats the propagation of plane waves in vacuum and simple media, at planar boundaries, and in combinations confined between sets of planar boundaries, as in waveguides or cavity resonators. Chapter 10 then discusses how such waves can be generated and received by antennas and antenna arrays. More specifically, Section 9.1 explains how plane waves are reflected from planar boundaries at normal incidence, and Section 9.2 treats reflection and refraction when the waves are incident at arbitrary angles. Section 9.3 then explains how linear combinations of such waves can satisfy all boundary conditions when they are confined within parallel plates or rectangular cylinders acting as waveguides. By adding planar boundaries at the ends of such waveguides, waves can be trapped at the resonant frequencies of the resulting cavity, as explained in Section 9.4. Sections 9.5 then treat waves in anisotropic, dispersive, and ionized media, respectively. 9.1.2 Introduction to boundary value problems Section 2.2 showed how uniform planes waves could propagate in any direction with any polarization, and could be superimposed in any combination to yield a total electromagnetic field. The general electromagnetic boundary value problem treated in Sections 9.1–4 involves determining exactly which, if any, combination of waves matches any given set of boundary conditions, which are the relations between the electric and magnetic fields adjacent to both sides of each boundary. These boundaries can generally be both active and passive, the active boundaries usually being sources. Boundary conditions generally constrain E and/or H for all time on the boundary of the two- or three-dimensional region of interest. The uniqueness theorem presented in Section 2.8 states that only one solution satisfies all Maxwell’s equations if the boundary conditions are sufficient. Therefore we may solve boundary value problems simply by hypothesizing the correct combination of waves and testing it against Maxwell’s equations. That is, we leave undetermined the numerical constants that characterize the chosen combination of waves, and then determine which values of those constraints satisfy Maxwell’s equations. This strategy eases the challenge of hypothesizing the final answer directly. Moreover, symmetry and other considerations often suggest the nature of the wave combination required by the problem, thus reducing the numbers of unknown constants that must be determined. The four basic steps for solving boundary value problems are: 1) Determine the natural behavior of each homogeneous section of the system in isolation (absent its boundaries). - 255 - 2) Express this natural behavior as the superposition of waves characterized by unknown constants; symmetry and other considerations can minimize the number of waves required. Here our basic building blocks are usually uniform plane waves, but other more compact expansions are typically used if the symmetry of the problem permits, as illustrated in Section 4.5.2 for cylindrical and spherical geometries, Section 7.2.2 for TEM transmission lines, and Section 9.3.1 for waveguide modes. 3) Write equations for the boundary conditions that must be satisfied by these sets of superimposed waves, and then solve for the unknown constants. 4) Test the resulting solution against any of Maxwell’s equations that have not already been imposed. Variations of this four-step procedure can be used to solve almost any problem by replacing Maxwell’s equations with their approximate equivalent for the given problem domain. 9.1.3 Reflection from perfect conductors One of the simplest examples of a boundary value problem is that of a uniform plane wave in vacuum normally incident upon a planar perfect conductor at z ≥ 0, as illustrated in Figure 9.1.1(a). Step 1 of the general boundary-problem solution method of Section 9.1.2 is simply to note that electromagnetic fields in the medium can be represented by superimposed uniform plane waves. (a) x Ex(z,to) (b) Ex(z,t) input reflected x σ=∞ input y 0 σ=∞ y 0 z z standing wave λ Figure 9.1.1 Plane wave at normal incidence reflecting from a perfect conductor. For this incompletely defined example, the initial part of Step 2 of the method involves refinement of the problem definition by describing more explicitly the incident wave, for example: E ( z, t ) = x̂E o cos ( ωt − kz ) ⎡⎣ Vm-1 ⎤⎦ (9.1.1) where the wave number k = 2π/λ = ω/c = ω(μoεo)0.5, (2.3.24). The associated magnetic field (2.3.25) is: - 256 - H ( z, t ) = ŷ ( E o ηo ) cos ( ωt − kz ) ⎡⎣Am-1 ⎤⎦ (9.1.2) This unambiguously defines the source, and the boundary is similarly unambigous: σ = ∞ and therefore E = 0 for z ≥ 0. This more complete problem definition is sufficient to yield a unique solution. Often the first step in solving a problem is to ensure its definition is complete. Since there can be no waves inside the perfect conductor, and since the source field alone does not satisfy the boundary condition E // = 0 at z = 0, one or more additional plane waves must be superimposed to yield a valid solution. In particular, we need to match the boundary condition E // = 0 at z = 0. This can be done by adding a single uniform plane wave propagating in the -z direction with an electric field that cancels the incident electric field at z = 0 for all time t. Thus we hypothesize that the total electric field is: E ( z, t ) = x̂ ⎡⎣ E o cos ( ωt − kz ) + E1 cos ( ωt + kz + φ ) ⎤⎦ (9.1.3) where we have introduced the constants E1 and φ. In Step 3 of the method we must solve the equation (9.1.3) that characterizes the boundary value constraints: E ( 0, t ) = x̂ ⎡E ⎣ o cos ( ωt − 0 ) + E1 cos ( ωt + 0 + φ ) ⎤⎦ = 0 (9.1.4) ∴E1 = −E o , φ = 0 (9.1.5) The result (9.1.5) yields the final trial solution: E ( z, t ) = x̂Eo [ cos ( ωt − kz ) − cos ( ωt + kz )] = x̂ 2Eo ( sin ωt ) sin kz (9.1.6) H ( z, t ) = ŷE o [ cos ( ωt − kz ) + cos ( ωt + kz )] ηo = ŷ ( 2E o ηo ) ( cos ωt ) cos kz (9.1.7) Note that the sign of the reflected H wave is reversed from that of the reflected E , consistent with the reversal of the Poynting vector for the reflected wave alone. We have used the identities: cos α + cos β = 2 cos ⎡⎣( α + β ) 2⎤⎦ cos ⎡⎣( α − β ) 2⎤⎦ (9.1.8) cos α − cos β = −2sin ⎡⎣( α + β ) 2⎤⎦ sin ⎡⎣( α − β ) 2⎤⎦ (9.1.9) Also note that H ( z, t ) is 90o out of phase with E ( z, t ) with respect to both time and space. - 257 - We also need a trial solution for z > 0. Inside the conductor E = H = 0 , and boundary conditions (2.6.17) require a surface current: A ⎡⎣ m-1 ⎦⎤ Js = n̂ × H (9.1.10) The fourth and final step of this problem-solving method is to test the full trial solution against all of Maxwell’s equations. We know that our trial solution satisfies the wave equation in our source-free region because our solution is the superposition of waves that do; it therefore also satisfies Faraday’s and Ampere’s laws in a source-free region, as well as Gauss’s laws. At the perfectly conducting boundary we require E // = 0 and H ⊥ = 0 ; these constraints are also satisfied by our trial solution, and therefore the problem is solved for the vacuum. Zero-value fields inside the conductor satisfy all Maxwell’s equations, and the surface current Js (9.1.10) satisfies the final boundary condition. The nature of this solution is interesting. Note that the total electric field is zero not only at the surface of the conductor, but also at a series of null planes parallel to the conductor and spaced at intervals Δ along the z axis such that kznulls = -nπ, where n = 0, 1, 2, ... That is, the null spacing Δ = π/k = λ/2, where λ is the wavelength. On the other hand, the magnetic field is maximum at those planes where E is zero (the null planes of E), and has nulls where E is maximum. Since the time average power flow and the Poynting vector are clearly zero at each of these planes, there is no net power flow to the right. Except at the field nulls, however, there is reactive power, as discussed in Section 2.7.3. Because no average power is flowing via these waves and the energy and waves are approximately stationary in space, the solution is called a standing wave, as illustrated in Figures 7.2.3 for VSWR and 7.4.1 for resonance on perfectly reflecting TEM transmission lines. 9.1.4 Reflection from transmissive boundaries Often more than one wave must be added to the given incident wave to satisfy all boundary conditions. For example, assume the same uniform plane wave (9.1.1–2) in vacuum is incident upon the same planar interface, where a medium having μ,ε ≠ μo,εo for z ≥ 0 has replaced the conductor. We have no reason to suspect that the fields beyond the interface are zero, so we might try a trial solution with both a reflected wave Er(z,t) and a transmitted wave Et(z,t) having unknown amplitudes (Er and Et) and phases (φ and θ) for which we can solve: E ( z, t ) = x̂ ⎡E ⎣ o cos ( ωt − kz ) + E r cos ( ωt + kz + φ ) ⎤⎦ (z < 0) (9.1.11) E t ( z, t ) = x̂E t cos ( ωt − k t z + θ ) (z ≥ 0) (9.1.12) H ( z, t ) = ŷ ⎡⎣ E o cos ( ωt − kz ) − E r cos ( ωt + kz + φ ) ⎤⎦ ηo (z < 0) (9.1.13) H ( z, t ) = ŷE t cos ( ωt − k t z + θ ) ηt (z ≥ 0) (9.1.14) - 258 - where k = ω μ o ε o , kt = ω με , ηo = μ o /ε o , and ηt = μ/ε . Using these four equations to match boundary conditions at z = 0 for E // and H // , both of which are continuous across an insulating boundary, and dividing by Eo, yields: x̂ ⎡⎣cos ( ωt ) + ( E r E o ) cos ( ωt + φ ) ⎤⎦ = x̂ ( E t E o ) cos ( ωt + θ ) (9.1.15) ŷ ⎡⎣cos ( ωt ) − ( E r E o ) cos ( ωt + φ ) ⎦⎤ ηo = ŷ ⎣⎡( E t E o ) cos ( ωt + θ ) ⎤⎦ ηt (9.1.16) First we note that for these equations to be satisfied for all time t we must have φ = θ = 0, unless we reverse the signs of Er or Et and let φ or θ = π, respectively, which is equivalent. Dividing these two equations by cos ωt yields: 1 + ( E r Eo ) = E t Eo (9.1.17) ⎣⎡1 − ( E r Eo ) ⎦⎤ ηo = ( E t E o ) ηt (9.1.18) These last two equations can easily be solved to yield the wave reflection coefficient and the wave transmission coefficient: E r ( ηt ηo ) −1 = E o ( ηt ηo ) +1 Et 2ηt = Eo ηt + ηo (reflection coefficient) (9.1.19) (transmission coefficient) (9.1.20) The wave transmission coefficient Et/Eo follows from (9.1.17) and (9.1.19). When the characteristic impedance ηt of the dielectric equals that of the incident medium, ηo, there are no reflections and the transmitted wave equals the incident wave. We then have an impedance match. These values for Er/Eo and Et/Eo can be substituted into (9.1.11–14) to yield the final solution for E ( z, t ) and H ( z, t ) . The last step of the four-step method for solving boundary value problems involves checking this solution against all Maxwell’s equations—they are satisfied. Example 9.1A A 1-Wm-2 uniform plane wave in vacuum, x̂E + cos ( ωt − kz ) , is normally incident upon a planar dielectric with ε = 4εo. What fraction of the incident power P+ is reflected? What is H ( t ) at the dielectric surface (z = 0)? - 259 - Solution: P− P+ = E − E + 2 2 = ( ηt -ηo ) ( ηt +ηo ) , using(5.1.19). Since ηt = μo 4εo = ηo 2 , therefore: P− P+ = ⎡⎣( ηo 2 ) − ηo ⎤⎦ ⎡⎣( ηo 2 ) + ηo ⎤⎦ 2 2 = ( −1 3) = 1 9 . For the forward wave: E = x̂E + cos ( ωt − kz ) and H = ŷ ( E + ηo ) cos ( ωt − kz ) , where E + so E + = ( 2ηo ) 0.5 = ( 2× 377 ) 0.5 2 2ηo = 1 , ≅ 27 [V/m]. The sum of the incident and reflected magnetic fields at z = 0 is H = ŷ ( E + ηo ) [ cos ( ωt ) − (1 3) cos ( ωt ) ⎤⎦ ≅ ŷ ( 27 377 )( 2 3) cos ( ωt ) = 0.48 ŷ cos ( ωt ) ⎡⎣A m-1 ⎤⎦ 9.2 Waves incident on planar boundaries at angles 9.2.1 Introduction to waves propagating at angles To determine electromagnetic fields we can generally solve a boundary value problem using the method of Section 9.1.1, the first step of which involves characterization of the basic quasistatic or dynamic fields and waves that could potentially exist within each separate region of the problem. The final solution is a linear combination of these basic fields and waves that matches all boundary conditions at the interfaces between the various regions. So far we have considered only waves propagating along boundaries or normal to them. The general case involves waves incident upon boundaries at arbitrary angles, so we seek a compact notation characterizing such waves that simplifies the boundary value equations and their solutions. Because wave behavior at boundaries often becomes frequency dependent, it is convenient to use complex notation as introduced in Section 2.3.2 and reviewed in Appendix B, which can explicitly represent the frequency dependence of wave phenomena. For example, we might represent the electric field associated with a uniform plane wave propagating in the +z direction as E o e− jkz , where: E ( z ) = E o e− jkz = E o e− j2πz λ (9.2.1) Eo = x̂ Eox + ŷ Eoy (9.2.2) This notation is simpler than the time domain representation. For example, if this wave were x-polarized, then the compact complex notation x̂E x would be replaced in the time domain by: { } E ( t ) = R e x̂E x (z)e jωt = x̂R e {( R e [ E x (z)] + jIm [ E x (z)]) ( cos ωt + jsin ωt )} = x̂ { R e [ E x (z)] cos ωt − Im [ E x (z)] sin ωt} - 260 - (9.2.3) The more general time-domain expression including both x and y components would be twice as long. Thus complex notation adequately characterizes frequency-dependent wave propagation and is more compact. The physical significance of (9.2.1) is divided into two parts: E o tells us the polarization, amplitude, and absolute phase of the wave at the origin, and e-j2πz/λ ≡ ejφ(z) tells us how the phase φ of this wave varies with position. In this case the phase decreases 2π radians as z increases by one wavelength λ. The physical significance of a phase shift φ of -2π radians for z = λ is that observers located at z = λ experience a delay of 2π radians; for pure sinusoids a phase shift of 2π is of course not observable. Waves propagating in arbitrary directions are therefore easily represented by expressions similar to (9.2.1), but with a phase φ that is a function of x, y, and z. For example, a general plane wave would be: E ( z ) = Eoe − jk x x − jk y y − jk z z = E o e− jk i r (9.2.4) where r = x̂x + ŷy + ẑz and: k = xk ˆ x + yk ˆ y + zk ˆ z (9.2.5) We call k the propagation vector or wave number k . The wave numbers kx, ky, and kz have the dimensions of radians per meter and determine how rapidly the wave phase φ varies with position along the x, y, and z axes, respectively. Positions having the same value for k • r have the same phase and are located on the same phase front. A wave with a planar phase fronts is a plane wave, and if its amplitude is constant across any phase front, it is a uniform plane wave. x planes of constant phase ⎯k •⎯r = constant λx |⎯k | ≡ ko = 2π/λo λo ⎯r2 kx ⎯r1 ⎯k θ kz = 2π/λz λz z Figure 9.2.1 Uniform plane wave propagating at angle θ in the x-z plane. - 261 - The vector k points in the direction of propagation for uniform plane waves. The geometry is represented in Figure 9.2.1 for a uniform plane wave propagating in the x-z plane at an angle θ and wavelength λo. The planes of constant phase are perpendicular to the wave vector k because k • r must be constant everywhere in such a plane. The solution (9.2.4) can be substituted into the wave equation (2.3.21): ( ∇2 + ω2με ) E = 0 (9.2.6) This substitution yields47: ( ) ⎡ − k 2 + k 2 + k 2 + ω2με ⎤ E = 0 x y z ⎣ ⎦ 2 k x 2 + k y 2 + k z 2 = k o 2 = ω2με = k = k • k (9.2.7) (9.2.8) Therefore the figure and (9.2.7) suggest that: k x = k • x̂ = k o sin θ, k z = k • ẑ = k o cos θ (9.2.9) The figure also includes the wave propagation vector components x̂ kx and ẑ kz. Three projected wavelengths, λx, λy, and λz, are perceived by observers moving along those three axes. The distance between successive wavefronts at 2π phase intervals is λo in the direction of propagation, and the distances separating these same wavefronts as measured along the x and z axes are equal or greater, as illustrated in Figure 9.2.1. For example: λ z = λ o cos θ = 2π k z ≥ λ o (9.2.10) Combining (9.2.8) and (9.2.10) yields: λ x −2 + λ y −2 + λ z −2 = λ o −2 (9.2.11) The electric field E ( r ) for the wave of Figure 9.2.1 propagating in the x-z plane is orthogonal to the wave propagation vector k . For simplicity we assume this wave is ypolarized: E ( r ) = ŷ E o e − jk i r 47 2 ( ∇ E= ∂ 2 2 ∂x + ∂ 2 2 ∂y + ∂ 2 ∂z 2 (9.2.12) ) E = − ( k x 2 + k y2 + k z2 ) E . - 262 - The corresponding magnetic field is: H ( r ) = − ( ∇ × E ) jωμo = ( x̂∂E y ∂z − ẑ∂E y ∂x ) jωμo (9.2.13) = ( ẑ sin θ − x̂ cos θ ) ( E o ηo ) e− jkir One difference between this uniform y-polarized plane wave propagating at an angle and one propagating along a cartesian axis is that H no longer lies along a single axis, although it remains perpendicular to both E and k . The next section treats such waves further. Example 9.2A If λx = 2λz in Figure 9.2.1, what are θ, λo, and k ? −2 Solution: By geometry, θ = tan −1 ( λ z λ x ) = tan −1 0.5 ≅ 27° . By (9.2.11) λ o−2 = λ −2 x + λz −0.5 = ( 0.25 +1) λ −2 λ z = 0.89λ z . k = x̂k x + ẑk z = x̂k o sin θ + ẑk o cos θ , z , so λ o = 1.25 where k o = 2π λo . Alternatively, k = x̂ 2π λ x + ẑ 2π λ z . 9.2.2 Waves at planar dielectric boundaries Waves at planar dielectric boundaries are solved using the boundary-value-problem method of Section 9.1.4 applied to waves propagating at angles, as introduced in Section 9.2.1. Because the behavior of waves at an interface depends upon their polarization we need a coordinate system for characterizing it. For this purpose the plane of incidence is defined as the plane of projection of the incident wave propagation vector k upon the interface, as illustrated in Figure 9.2.2(a). One cartesian axis is traditionally defined as being normal to this plane of incidence; in the figure it is the y axis. (a) (b) plane of incidence k θi Hz x z interface H E • x k kz kr θi θr kz z y Figure 9.2.2 Uniform plane wave incident upon a planar interface. We know from Section 2.3.4 that any pair of orthogonally polarized uniform plane waves can be superimposed to achieve any arbitrary wave polarization. For example, x- and ypolarized waves can be superimposed. It is customary to recognize two simple types of incident - 263 - electromagnetic waves that can be superimposed to yield any incident wave polarization: transverse electric waves (TE waves) are linearly polarized transverse to the plane of incidence (y-polarized in the figure), and transverse magnetic waves (TM waves) have the orthogonal linear polarization so that the magnetic field is purely transverse (again if y-polarized). TE and TM waves are typically transmitted and reflected with different amplitudes. Consider first a TE wave incident upon the planar interface of Figure 9.2.2(b) at the incidence angle θi. The corresponding H lies in the x-z plane and is orthogonal to E . H points downward in the figure, corresponding to power S = E × H propagating toward the interface, where S is the Poynting vector for the incident wave. The wavelength of the wave above the interface is λo = 1/(f με ) in the medium characterized by permittivity ε and permeability μ. The medium into which the wave is partially transmitted is characterized by εt and μt, and there the wave has wavelength λt = 1/(f μ t ε t ) and the same frequency f. This incident TE wave can be characterized by: Ei = ŷ E o e jk x x− jk zVz ⎡⎣ m-1 ⎦⎤ (9.2.14) z ⎡ Hi = − ( Eo η) ( x̂ sin θi + ẑ cos θi ) e jk x x − jk zA ⎣ m-1 ⎤⎦ where the characteristic impedance of the incident medium is η = (9.2.15) μ/ε , and H is orthogonal to E. The transmitted wave would generally be similar, but with a different ηt, θt, Et, and k t . We might expect a reflected wave as well. The boundary-value-problem method of Section 9.1.2 requires expressions for all waves that might be present in both regions of this problem. In addition to the incident wave we therefore might add general expressions for reflected and transmitted waves having the same TE polarization. If still other waves were needed then no solution satisfying all Maxwell’s equations would emerge until they were added too; we shall see no others are needed here. These general reflected and transmitted waves are: E r = ŷ E r e− jk rx x− jk rzVz ⎡⎣ m-1 ⎤⎦ (9.2.16) z ⎡ H r = ( E r η) ( − x̂ sin θr + ẑ cos θr ) e− jk rx x − jk rzA ⎣ m-1 ⎦⎤ (9.2.17) E t = ŷ E t e jk tx x − jk tzVz ⎡⎣ m-1 ⎤⎦ (9.2.18) z ⎡ H t = − ( E t ηt )( x̂ sin θt + zˆ cos θt ) e jk tx x− jk tzA ⎣ m-1 ⎤⎦ (9.2.19) Boundary conditions that must be met everywhere on the non-conducting surface at x = 0 include: - 264 - Ei // + E r // = E t // (9.2.20) Hi // + H r // = H t // (9.2.21) Substituting into (9.2.20) the values of E // at the boundaries yields: E o e− jk z z + E r e− jk rz z = E t e− jk tz z (9.2.22) This equation can be satisfied for all values of z only if all exponents are equal. Therefore e − jk z z can be factored out, simplifying the boundary-condition equations for both E // and H // : Eo + E r = E t (boundary condition for E//) (9.2.23) Eo E E cos θi − r cos θr = t cos θt η η ηt (boundary condition for H//) (9.2.24) Because the exponential terms in (9.2.22) are all equal, it follows that the phases of all three waves must match along the full boundary, and: k iz = k rz = k tz = k i sin θi = k i sin θr = k t sin θt = 2π λ z (9.2.25) This phase-matching condition implies that the wavelengths of all three waves in the z direction must equal the same λz. It also implies that the angle of reflection θr equals the angle of incidence θi, and that the angle of transmission θt is related to θi by Snell’s law: sin θt k i c t με = = = μt εt sin θi k t ci (Snell’s law) (9.2.26) If μ = μt, then the angle of transmission becomes: ⎞ ⎛ θt = sin −1 ⎜ sin θi ε ⎟ εt ⎠ ⎝ (9.2.27) These phase-matching constraints, including Snell’s law, apply equally to TM waves. The magnitudes of the reflected and transmitted TE waves can be found by solving the simultaneous equations (9.2.23) and (9.2.24): η cosθi − ηcosθ t ηn ' − 1 = E r E o = Γ ( θi ) = t ηt cos θi + ηcosθ t ηn ' +1 - 265 - (9.2.28) E t /E o = T(θi ) = 2ηt cos θi 2ηn ' = ηt cos θi + ηcos θt ηn ' + 1 (9.2.29) where we have defined the normalized angular impedance for TE waves as ηn' ≡ ηtcosθi/(ηcosθt). The complex angular reflection and transmission coefficients Γ and T for TE waves approach those given by (9.1.19) and (9.1.20) for normal incidence in the limit θi → 0. The limit of grazing incidence is not so simple, and even the form of the transmitted wave can change markedly if it becomes evanescent, as discussed in the next section. The results for 2 incident TM waves are postponed to Section 9.2.6. Figure 9.2.6(a) plots Γ(θ) for a typical dielectric interface. It is sometimes useful to note that (9.2.28) and (9.2.29) also apply to equivalent TEM lines for which the characteristic impedances of the input and output lines are ηi/cosθi and ηt/cosθt, respectively. When TM waves are incident, the corresponding equivalent impedances are ηicosθi and ηtcosθt, respectively. Example 9.2B What fraction of the normally incident power (θi = 0) is reflected by a single glass camera lense having ε = 2.25εo? If θi = 30o, what is θt in the glass? Solution: At each interface between air and glass, (9.2.28) yields for θi = 0: ΓL = (ηn' - 1)/(ηn' + 1), where ηn' = (ηglasscosθi)/(ηaircosθt) = (εi/εg)0.5 = 1/1.5. Thus ΓL = (1 - 1.5)/(1 + 1.5) = -0.2, and |Γ|2 = 0.04, so ~4 percent of the power is reflected from each of the two curved surfaces for each independent lense, or ~8 percent total; these reflections are incoherent so their reflected powers add. Modern lenses have many elements with different permittivities, but coatings on them reduce these reflections, as discussed in Section 7.3.2 for quarter-wave transformers. Snell’s law (9.2.26) yields θt = sin-1[(εi/εt)0.5 sinθi] = sin-1[(1/1.5)(0.5)] = 19.5°. 9.2.3 Evanescent waves Figure 9.2.3 suggests why a special form of electromagnetic wave is sometimes required in order to satisfy boundary conditions. Figure 9.2.3(a) illustrates how the required equality of the z components of the incident, reflected, and transmitted wave propagation vectors k controls the angles of reflection and transmission, θr and θt. The radii of the two semi-circles correspond to the magnitudes of ki and k t . Figure 9.2.3(b) shows that a wave incident at a certain critical angle θc will produce a transmitted wave propagating parallel to the interface, provided k t < k i . Snell’s law (9.2.26) can be evaluated for sin θt = 1 to yield: θc = sin −1( ci c t ) for ci < c t (critical angle) - 266 - (9.2.30) Figures 9.2.3(b) illustrates why phase matching is impossible with uniform plane waves when θ > θc; kz > kt . Therefore the λz determined by λ and θi is less than λt, the natural wavelength of the transmission medium at frequency ω. A non-uniform plane wave is then required for phase matching, as discussed below. (a) loci x k i = ω ( με ) ki kz θi where θr (b) 0.5 z kz kz θc θc 0.5 kr kz z kt θt loci kt k t = ω (μt εt ) where k i = ω ( με ) ki kr θt loci where loci x k t = ω (μt εt ) 0.5 where 0.5 < ki Figure 9.2.3 Angles of reflection and transmission, and the critical angle θc. The wave propagation vector k t must satisfy the wave equation ( ∇2 + k t 2 ) E = 0 . Therefore the transmitted wave must be proportional to e− jk t i r , where k t = k̂k t and k z = k i sin θi , satisfy the expression: k t 2 = ω2μ t ε t = k tx 2 + k z 2 (9.2.31) When kt2 < kz2 it follows that: ( k tx = ± j k z 2 − k t 2 ) 0.5 = ± jα (9.2.32) We choose the positive sign for α so that the wave amplitude decays with distance from the power source rather than growing exponentially. The transmitted wave then becomes: E t ( x, z ) = ŷ E t e jk tx x − jk z z = yˆ E t eαx − jk z z (9.2.33) Note that x is negative in the decay region. The rate of decay α = (ki2sin2θi - kt2)0.5 is zero when θi = θc and increases as θi increases past θc; Waves that decay with distance from an interface and propagate power parallel to it are called surface waves. - 267 - The associated magnetic field H t can be found by substituting (9.2.33) into Faraday’s law: H t = ∇ × E ( − jωμ t ) = − ( E t ηt )( x̂ sin θt − ẑ cos θt ) eαx − jk z z (9.2.34) This is the same expression as (9.2.19), which was obtained for normal incidence, except that the magnetic field and wave now decay with distance x from the interface rather than propagating in that direction. Also, since sinθt > 1 for θi > θc, cosθt is now imaginary and positive, and H is not in phase with E . As a result, Poynting’s vector for these surface waves has a real part corresponding to real power propagating parallel to the surface, and an imaginary part corresponding to reactive power flowing perpendicular to the surface in the direction of wave decay: ( * S = E × H = ( − jαx̂ + k z ẑ ) E t 2 ) ωμ t e2αx ⎡⎣Wm-2 ⎦⎤ (9.2.35) 2 The reactive part flowing in the - xˆ direction is + jα E t ωμ t e2ax and is therefore inductive (+j), corresponding to an excess of magnetic stored energy relative to electric energy within this surface wave. A wave such as this one that decays in a direction for which the power flow is purely reactive is designated an evanescent wave. An instantaneous view of the electric and magnetic fields of a non-uniform TE plane wave formed at such a dielectric boundary is shown in Figure 9.2.4; these correspond to the fields of (9.2.33) and (9.2.34). E • H E(t) × × × × • • • H(t) x standing wave in x direction • • • • • × • × × × y × × × × × × • • • • • • • • • × × × × λio/2 direction of propagation ε, μ θi ki εt, μt × λz/2 • z R e {S} I m {S} × evanescent wave Figure 9.2.4 Evanescent wave traveling in the +z direction at a dielectric interface. - 268 - The conventional notation used here indicates field strength by the density of symbols or field lines, and the arrows indicate field direction. Small circles correspond to field lines pointing perpendicular to the page; center dots indicate field lines pointing out of the page in the +y direction and center crosses indicate the opposite, i.e., field lines pointing into the page. The time average wave intensity in the + ẑ direction for x negative and outside the dielectric is: { Pz = 0.5R e E × H * } = ( k z E t 2 2ωμ t ) eαx ⎡⎣ W m-2 ⎦⎤ (9.2.36) Since the real and imaginary parts of S are orthogonal, there is no decay in the direction of propagation, and therefore no power absorption or heating of the media. Beyond the critical angle θc the power is perfectly reflected. In the next section we shall see that the real and imaginary parts of S are often neither orthogonal nor parallel. 9.2.4 Waves in lossy media Sometimes one or both of the two media are conductive. This section explores the nature of waves propagating in such lossy media having conductivity σ > 0. Section 9.2.5 then discusses reflections from such media. Losses can also arise if ε or μ are complex. The quasistatic relaxation of charge, current, and field distributions in lossy media is discussed separately in Section 4.3. We can determine the nature of waves in lossy media using the approach of Section 2.3.3 and including the conduction currents J in Ampere’s law: ∇ × H = J + jωεE = σE + jωεE = jωεeff E (9.2.37) where the effective complex permittivity εeff is: εeff = ε ⎡⎣1 − ( jσ ωε ) ⎤⎦ (9.2.38) The quantity σ/ωε is called the loss tangent of the medium and indicates how fast waves decay. As we shall see, waves propagate well if σ << ωε, and decay rapidly if σ > ωε, sometimes within a fraction of a wavelength. Substituting εeff for ε in k2 = ω2με yields the dispersion relation: k 2 = ω2με ⎡⎣1− ( jσ ωε ) ⎤⎦ = ( k '− jk '') 2 where we define the complex wavenumber k in terms of its real and imaginary parts as: - 269 - (9.2.39) k = k '− jk '' (9.2.40) The form of the wave solution, following (2.3.26), is therefore: E ( r ) = ŷE o e − jk ' z −k '' z ⎡⎣ v m-1 ⎤⎦ (9.2.41) This wave has wavelength λ′, frequency ω, and phase velocity vp inside the conductor related by: k ' = 2π λ ' = ω v p (9.2.42) and the wave decays exponentially with z as e-k"z = e-z/Δ. Note that the wave decays in the same direction as it propagates, corresponding to power dissipation, and that the 1/e penetration depth Δ is 1/k" meters. Inside conductors, λ′ and vp are much less than their free-space values. We now need to determine k' and k". In general, matching the real and imaginary parts of (9.2.39) yields two equations that can be solved for k' and k": ( k ' )2 − ( k '' )2 = ω2με (9.2.43) 2k ' k '' = ωμσ (9.2.44) However, in the limits of very high or very low values of the loss tangent σ/ωε, it is much easier to evaluate (9.2.39) directly. In the low loss limit where σ << ωε, (9.2.39) yields: k = ω με ⎡⎣1− ( jσ ωε ) ⎤⎦ ≅ ω με − jση 2 ( σ << ωε ) where the approximate wave impedance of the medium is η = (9.2.45) μ/ε , and we have used the Taylor series approximation 1+ δ ≅ 1 + δ/2 for δ << 1. In this limit we see from (9.2.45) that λ′ and vp ≅ c are approximately the same as they are for the lossless case, and that the 1/e penetration depth Δ ≅ 2/ση, which becomes extremely large as σ → 0. In the high loss limit where σ >> ωε, (9.2.39) yields: k = ω με ⎣⎡1 − ( jσ ωε ) ⎦⎤ ≅ − jμωσ ≅ ωμσ − j = ± ωμσ (1− j) 2 - 270 - ( σ >> ωε ) (9.2.46) (9.2.47) The real and imaginary parts of k have the same magnitudes, and the choice of sign determines the direction of propagation. The wave generally decays exponentially as it propagates, although exponential growth occurs in media with negative conductivity. The penetration depth is commonly called the skin depth δ in this limit (σ >> ωε), where: δ = 1 k '' ≅ ( 2 ωμσ ) 0.5 [ m] (skin depth) (9.2.48) Because the real and imaginary parts of k are equal here, both the skin depth and the wavelength λ' inside the conductor are extremely small compared to the free-space wavelength λ; thus: λ ' = 2π k ' = 2πδ [ m ] (wavelength in conductor) (9.2.49) These distances δ and λ' are extremely short in common metals such as copper (σ ≅ 5.8×107, μ = μo) at frequencies such as 1 GHz, where δ ≅ 2×10-6 m and λ' ≅ 13×10-6 m, which are roughly five orders of magnitude smaller than the 30-cm free space wavelength. The phase velocity vp of the wave is reduced by the same large factor. In the high conductivity limit, the wave impedance of the medium also becomes complex: η= μ εeff = μ ≅ ε (1 − jσ ωε ) jωμ ωμ = (1 + j) σ 2σ (9.2.50) where +j is consistent with a decaying wave in a lossy medium. The imaginary part of η corresponds to power dissipation, and is non-zero whenever σ ≠ 0. Often we wish to shield electronics from unwanted external radiation that could introduce noise, or to ensure that no radiation escapes to produce radio frequency interference (RFI) that affects other systems. Although the skin depth effect shields electromagnetic radiation, high conductivity will reflect most incident radiation in any event. Conductors generally provide good shielding at higher frequencies for which the time intervals are short compared to the magnetic relaxation time (4.3.15) while remaining long compared to the charge relaxation time (4.3.3); Section 4.3.2 and Example 4.3B present examples of magnetic field diffusion into conductors. Example 9.2C A uniform plane wave propagates at frequency f = c/λ = 1 MHz in a medium characterized by εo, μo, and conductivity σ. If σ ≅ 10-3ωεo, over what distance D would the wave amplitude decay by a factor of 1/e? What would be the 1/e wave penetration depth δ in a good conductor having σ ≅ 1011ωε at this frequency? Solution: In the low-loss limit where σ << ωε, k ≅ ω/c - jση/2 (9.2.45), so E ∝ e-σηz/2 = e-z/D where D = 2/ση = 2(εo/μo)0.5/10-3ωεo = 2000c/ω ≅ 318λ [m]. In the high-loss limit k - 271 - ≅ (1 ± j)(ωμσ/2)0.5 so E ∝ e-z/δ, where δ = (2/ωμσ)0.5 = (2×10-11/ω2με)0.5 = (2×10-11)0.5 λ/2π ≅ 7.1×10-7λ = 0.21 mm. This conductivity corresponds to typical metal and the resulting penetration depth is a tiny fraction of a free-space wavelength. 9.2.5 Waves incident upon good conductors This section focuses primarily on waves propagating inside good conductors. The field distributions produced outside good conductors by the superposition of waves incident upon and reflected from them are discussed in Section 9.2.3. Section 9.2.4 showed that uniform plane waves in lossy conductors decay as they propagate. The wave propagation constant k is then complex in order to characterize exponential decay with distance: k = k '− jk" (9.2.51) The form of a uniform plane wave in lossy media is therefore: E ( r ) = ŷ E o e − jk ' z−k "z ⎡⎣v m-1 ⎤⎦ (9.2.52) When a plane wave impacts a conducting surface at an angle, a complex wave propagation vector k t is required to represent the resulting transmitted wave. The real and imaginary parts of k t are generally at some angle to each other. The result is a non-uniform plane wave because its intensity is non-uniform across each phase front. To illustrate how such transmitted non-uniform plane waves can be found, consider a lossy transmission medium characterized by ε, σ, and μ, where we can combine ε and σ into a single effective complex permittivity, as done in (9.2.38)48: εeff ≡ ε (1 − jσ ωε ) (9.2.53) If we represent the electric field as E o e− jkir and substitute it into the wave equation ( ∇2 + ω2με ) E = 0 , we obtain for non-zero E the general dispersion relation for plane waves in isotropic lossy media: ⎡( − jk ) • ( − jk ) + ω2μεeff ⎤ E = 0 ⎣ ⎦ k • k = ω2μεeff 48 (9.2.54) (dispersion relation) ∇ × H = J + jωE = σE + jωE = jωε eff E - 272 - (9.2.55) Once a plane of incidence such as the x-z plane is defined, this relation has four scalar unknowns—the real and imaginary parts for each of the x and z (in-plane) components of k . At a planar boundary there are four such unknowns for each of the reflected and transmitted waves, or a total of eight unknowns. Each of these four components of k (real and imaginary, parallel and perpendicular) must satisfy a boundary condition, yielding four equations. The dispersion relation (9.2.55) has real and imaginary parts for each side of the boundary, thus providing four more equations. The resulting set of eight equations can be solved for the eight unknowns, and generally lead to real and imaginary parts for k t that are neither parallel nor perpendicular to each other or to the boundary. That is, the real and imaginary parts of k and S can point in four different directions. It is useful to consider the special case of reflections from planar conductors for which σ >> ωε. In this limit the solution is simple because the transmitted wave inside the conductor propagates almost perpendicular to the interface, which can be shown as follows. Equation (9.2.47) gave the propagation constant k for a uniform plane wave in a medium with σ >> ωε: k≅± ωμσ (1− j) 2 (9.2.56) The real part of such a k is so large that even for grazing angles of incidence, θi ≅ 90o, the transmission angle θt must be nearly zero in order to match phases, as suggested by Figure 9.2.3(a) in the limit where kt is orders of magnitude greater than ki. As a result, the power dissipated in the conductor is essentially the same as for θi = 0o, and therefore depends in a simple way on the induced surface current and parallel surface magnetic field H // . H // is simply twice that associated with the incident wave alone ( H ⊥ ≅ 0 ) ; essentially all the incident power is reflected so the incident and reflected waves have the same amplitudes and their magnetic fields add. The power density Pd [W m-2] dissipated by waves traveling in the +z direction in conductors with an interface at z = 0 can be found using the Poynting vector: Pd = 1 R e 2 {( ⎧⎪ TE 2 ⎪⎫ 2 i ⎬ = 1 R e ⎨⎧ 1 ⎬⎫ ηi 2 Hi T E×H •zˆ z=0+ = R e ⎨ ⎪⎩ 2ηt ⎪⎭ 2 ⎩ ηt ⎭ * ) } (9.2.57) The wave impedance ηt of the conductor (σ >> ωε) was derived in (9.2.50), and (9.2.29) showed η cos θi 1: that T = 2η'n η'n +1 ≅ 2η'n for TE waves and η'n = t ηi cos θt ( ) η t ≅ ( ωμ t 2σ ) T TE ( θi ) ≅ 2η'n 0.5 (1+ j) (9.2.58) ( η'n +1) ≅ 2η'n ≅ 2ηt cos θt ηi = ( 2ωμt εi μiσ )0.5 (1+ j) cos θi - 273 - (9.2.59) Therefore (9.2.57), (9.2.58), and (9.2.59) yield: Pd ≅ σ μi H 2 4ωμ t εi = H ( z = 0 ) 2 ωμ ⎡ W m 2 ⎤ ⎦ 8σ ⎣ 2ωμ t εi i 2μi σ (9.2.60) A simple way to remember (9.2.60) is to note that it yields the same dissipated power density that would result if the same surface current Js flowed uniformly through a conducting slab having conductivity σ and a thickness equal to the skin depth δ = { 2/ωμσ : 2 } J 2 ωμ ⎡ 2 * Pd = δ R e E • J = J δ = s = H ( z = 0 ) W m 2 ⎤⎦ 2 2σ 2σδ 8σ ⎣ (9.2.61) The significance of this result is that it simplifies calculation of power dissipated when waves impact conductors—we need only evaluate the surface magnetic field under the assumption the conductor is perfect, and then use (9.2.61) to compute the power dissipated per square meter. Example 9.2D What fraction of the 10-GHz power reflected by a satellite dish antenna is resistively dissipated in the metal if σ = 5×107 Siemens per meter? Assume normal incidence. A wire of diameter D and made of the same metal carries a current I. What is the approximate power dissipated per meter if the skin depth δ at the chosen frequency is much greater than D? What is this dissipation if δ << D? Solution: The plane wave intensity is I = ηo H + 2 2 [W/m2], and the power absorbed by a good 2 ωμ 4σ , where the magnetic field near a conductor is given by (9.2.61): Pd ≅ 2H + good conductor is twice the incident magnetic field due to the reflected wave. The fractional power absorbed is: ( Pd I = 4 ωμ σ ηo = 4 ωεo σ ≅ 4 2π1010 ×8.8×10−12 5×107 2 ) 0.5 = 4.2×10-4. If 2 δ >> D, then a wire dissipates I R 2 watts = 2 I σπD 2 [W/m2]. The magnetic field around a wire is: H = I πD , and if δ << D, then the power dissipated per meter 2 2 ωμ 4σ = I ωμ 4σ πD [W/m2], where the surface area for is: πD H dissipation is πD [m2]. Note that the latter dissipation is now increases with the square-root of frequency and is proportional to 1/ σ , not 1/σ. 9.2.6 Duality and TM waves at dielectric boundaries Transverse magnetic (TM) waves reflect from planar surfaces just as do TE waves, except with different amplitudes as a function of angle. The angles of reflection and transmission are the - 274 same as for TE waves, however, because both TE and TM waves must satisfy the same phase matching boundary condition (9.2.25). The behavior of TE waves at planar boundaries is characterized by equations (9.2.14) and (9.2.15) for the incident electric and magnetic fields, (9.2.16) and (9.2.17) for the reflected wave, and (9.2.18) and (9.2.19) for the transmitted wave, supplemented by expressions for the complex reflection and transmission coefficients Γ = Er/Eo, (9.2.28), and T = Et/Eo, (9.2.29). Although the analogous behavior of TM waves could be derived using the same boundary-value problem solving method used in Section 9.2.2 for TE waves, the principle of duality can provide the same solutions with much less effort. Duality works because Maxwell’s equations without charges or currents are duals of themselves. That is, by transforming E ⇒ H , H ⇒ −E , and ε ⇔ μ, the set of Maxwell’s equations is unchanged: ∇ × E = −μ∂ H ∂t → ∇ × H = ε∂E ∂t (9.2.62) ∇ × H = ε∂E ∂t → −∇ × E = μ∂H ∂t (9.2.63) ∇ • εE = 0 → ∇ • μH = 0 (9.2.64) ∇ • μH = 0 → ∇ • εE = 0 (9.2.65) The transformed set of equations on the right-hand side of (9.2.62) to (9.2.65) is the same as the original, although sequenced differently. As a result, any solution to Maxwell’s equations is also a solution to the dual problem where the variables and boundary conditions are all transformed as indicated above. The boundary conditions derived in Section 2.6 for a planar interface between two insulating uncharged media are that E // , H // , μH ⊥ , and εE ⊥ be continuous across the boundary. Since the duality transformation leaves these boundary conditions unchanged, they are dual too. However, duality cannot be used, for example, in the presence of perfect conductors that force E // to zero, but not H // . (a) TE H E E ' = −H (b) TM • • θi ε,μ,η εt, μt, ηt H' = E ε', μ', η' θi εt', μt', ηt' ε' = μ μ' = ε η' = η-1 = (ε/μ)0.5 εt' = μt μt' = εt ηt' = ηt-1 Figure 9.2.5 Dual TE and TM waves incident upon a dual planar boundary. - 275 - Figure 9.2.5(b) illustrates a TM plane wave incident upon a planar boundary where both the wave and the boundary conditions are dual to the TE wave illustrated in (a). The behavior of TM waves at planar boundaries between non-conducting media is therefore characterized by duality transformations of Equations (9.2.62–65) for TE waves, supplemented by similar transformations of the expressions for the complex reflection and transmission coefficients Γ = Er/Eo, (9.2.28), and T = Et/Eo, (9.2.29). After the transformations E ⇒ H , H ⇒ −E , and ε ⇔ μ, Equations (9.2.14–19) become: H i = ŷ H o e jk x x − jk z z ⎡⎣Am -1 ⎤⎦ (9.2.66) Ei = ( Hoη) ( x̂ sin θi + ẑ cos θi ) e jk x x − jk z z ⎡⎣Vm-1 ⎤⎦ (9.2.67) H r = ŷ H r e − jk rx x − jk z z ⎡⎣Am -1 ⎤⎦ (9.2.68) E r = ( H r η) ( x̂ sin θr - zˆ cos θr ) e− jk rx x − jk z z ⎡⎣Vm-1 ⎤⎦ (9.2.69) H t = ŷ H t e jk tx x− jk z z ⎡⎣Am-1 ⎤⎦ (9.2.70) E t = ( H t ηt )( xˆ sin θt + zˆ cos θt ) e jk tx x − jk z z ⎡⎣Vm-1 ⎤⎦ (9.2.71) The complex reflection and transmission coefficients for TM waves are transformed versions of (9.2.28) and (9.2.29), where we define a new angle-dependent ηn by interchanging μ ↔ ε in ηn′ in (9.2.28): ( ) ( ηn −1 +1) H r Ho = ηn −1 −1 ( (9.2.72) ) H t Ho = 2ηn −1 ηn −1 +1 (9.2.73) ηn −1 ≡ ηcos θi ( ηt cos θt ) (9.2.74) These equations, (9.2.66) to (9.2.74), completely describe the TM case, once phase matching provides θr and θt. It is interesting to compare the power reflected for TE and TM waves as a function of the 2 2 angle of incidence θi. Power in uniform plane waves is proportional to both E and H . Figure 9.2.6 sketches how the fractional power reflected or surface reflectivity varies with angle of incidence θi for both TE and TM waves for various impedance mismatches, assuming μ = μt - 276 - 2 and σ = 0 everywhere. If the wave is incident upon a medium with εt > ε, then Γ → 1 as 2 θi → 90° , whereas Γ → 1 at the critical angle θc if εt < ε, and remains unity for θc < θ < 90o (this θc case is not illustrated). Γ (a) 2 1• ε ≅ 100εo (b) TE TM H E TM wave • ε, μ θB θB θt ε ≅ 10εo ε ≅ 2εo 0• θB θB θi • o θB 90 or θc θt Figure 9.2.6 Power reflected from planar dielectric interfaces for μ = μt. Figure 9.2.6 reveals an important phenomenon—there is perfect transmission at Brewster’s angle θB for one of the two polarizations. In this case Brewster’s angle occurs for the TM polarization because μ is the same everywhere and ε is not, and it would occur for TE polarization if μ varied across the boundary while ε did not. This phenomenon is widely used in glass Brewster-angle windows when even the slightest reflection must be avoided or when pure linear polarization is required (the reflected wave is pure). We can compute θB by noting Hr/Ho and, using (9.2.72), ηn = 1. If μ = μt, then (9.2.74) yields εt0.5 cos θi = ε0.5 cos θt. Snell’s law for μ = μt yields ε0.5 sin θi = εt0.5 sin θt. These two equations are satisfied if sin θi = cos θt and cos θi = sin θt. Dividing this form of Snell’s law by [cos θi = sin θt] yields: tan θi = (εt/εi)0.5, or: θB = tan −1 ε t εi (9.2.75) Moreover, dividing [sin θi = cos θt] by [cos θi = sin θt] yields tan θB = cos θt, which implies θB + θt = 90°. Using this equation it is easy to show that θB > 45o for interfaces where θt < θi, and when θt > θi, it follows that θB < 45o. One way to physically interpret Brewster’s angle for TM waves is to note that at θB the polar axes of the electric dipoles induced in the second dielectric εt are pointed exactly at the angle of reflection mandated by phase matching, but dipoles radiate nothing along their polar axis; Figure 9.2.6(b) illustrates the geometry. That is, θB + θt = 90o. For magnetic media - 277 - magnetic dipoles are induced, and for TE waves their axes point in the direction of reflection at Brewster’s angle. Yet another way to physically interpret Brewster’s angle is to note that perfect transmission can be achieved if the boundary conditions can be matched without invoking a reflected wave. This requires existence of a pair of incidence and transmission angles θi and θt such that the parallel components of both E and H for these two waves match across the boundary. Such a pair consistent with Snell’s law always exists for TM waves at planar dielectric boundaries, but not for TE waves. Thus there is perfect impedance matching at Brewster’s angle. Example 9.2E What is Brewster’s angle θB if μ2 = 4μ1, and ε2 = ε1, and for which polarization would the phenomenon be observed? Solution: If the permeabilities differ, but not the permittivities, then Brewster’s angle is observed only for TE waves. At Brewster’s angle θB + θt = 90° , and Snell’s law sin θ t μ . But sin θ t = sin ( 90° − θB ) = cos θB , so Snell’s law becomes: = sin θB μt tan θB = μ t μ = 2 , and θB ≅ 63° . says 9.3 Waves guided within Cartesian boundaries 9.3.1 Parallel-plate waveguides We have seen in Section 9.2 that waves can be reflected at planar interfaces. For example, (9.2.14) and (9.2.16) describe the electric fields for a TE wave reflected from a planar interface at x = 0, and are repeated here: Ei = ŷ E o e jk x x − jk z z ⎣⎡ V m-1 ⎦⎤ (incident TE wave) (9.3.1) E r = ŷ E r − jk x x− jk z z (reflected TE wave) (9.3.2) Note that the subscripts i and r denote “incident” and “reflected”, not “imaginary” and “real”. These equations satisfy the phase-matching boundary condition that kz = kzi = kzr. Therefore |kx| is the same for the incident and reflected waves because for both waves kx2 + kz2 = ω2με. If the planar interface at x = 0 is a perfect conductor, then the total electric field there parallel to the conductor must be zero, implying E r = −Eo . The superposition of these two incident and reflected waves is: E ( x, z ) = ŷ E o ( e jk x x − e − jk x x ) e − jk z z = ŷ 2 jE o sin k x xe − jk z z - 278 - (9.3.3) Thus E = 0 in a series of parallel planes located at x = d, where: k x d = nπ for n = 0, 1, 2,... (guidance condition) (9.3.4) Because kx = 2π/λx, these planes of electric-field nulls are located at xnulls = nλx/2. A second perfect conductor could be inserted at any one of these x planes so that the waves would reflect back and forth and propagate together in the +z direction, trapped between the two conducting planes. We can easily confirm that the boundary conditions are satisfied for the corresponding magnetic field H ( x, z ) by using Faraday’s law: H ( x, z ) = − ( ∇ × E ) jωμ = − ( 2Eo ωμ ) ( x̂ jk z sin k x x + ẑ k x cos k x x ) e− jk z z (9.3.5) At x = nλ x 2 we find H ⊥ = H x = 0 , so this solution is valid. Equation (9.3.4), kxd = nπ (n = 1, 2,...), is the guidance condition for parallel-plate waveguides that relates mode number to waveguide dimensions; d is the separation of the parallel plates. We can use this guidance condition to make the expressions (9.3.3) and (9.3.5) for E and H more explicit by replacing kxx with nπx/d, and kz with 2π/λz: ( ) − j2πz E ( x, z ) = ŷ2 jE o sin nπx e λ z d 2E H ( x, z ) = − o ωμ ( E for TEn mode) ( ) ( ) − (9.3.6) j2 πz nπx nπ nπx ⎤ λz ⎡ 2π ⎢⎣ x̂ j λ z sin d + zˆ d cos d ⎥⎦ e (9.3.7) These electric and magnetic fields correspond to a waveguide mode propagating in the +z direction, as illustrated in Figure 9.3.1(a) for a waveguide with plates separated by distance d. * The direction of propagation can be inferred from the Poynting vector S = E × H . Because this is a TE wave and there is only one half wavelength between the two conducting plates (n = 1), this is designated the TE1 mode of a parallel-plate waveguide. Because Maxwell’s equations are linear, several propagating modes with different values of n can be active simultaneously and be superimposed. These fields are periodic in both the x and z directions. The wavelength λz along the z axis is called the waveguide wavelength and is easily found using kz = 2π/λz where k2 = kx2 + kz2 = ω2με: ( λ z = 2π k 2 − k x 2 ) −0.5 2 2 = 2π ⎡⎣( ω c ) − ( nπ d ) ⎦⎤ - 279 - −0.5 (waveguide wavelength) (9.3.8) (a) TE1 mode (b) TM1 mode λz E(t) d × × × × x • • • • • × × × × × • • direction of propagation ×E(t) • • • λ×z • ××× × • • • • • × × × H(t) • • H(t) z x× × × × • • • • • × × × × × • • • z Figure 9.3.1 TE1 and TM1 modes of parallel-plate waveguides. Only those TEn modes having n < ωd cπ = 2d λ have non-imaginary waveguide wavelengths λz and propagate. Propagation ceases when the mode number n increases to the point where nπ d ≥ k ≡ 2π λ ≡ ω c . This propagation requirement can also be expressed in terms of a minimum frequency ω of propagation, or cut-off frequency, for any TE mode: ωTEn = nπc d (cut-off frequency for TEn mode) (9.3.9) Thus each TEn mode has a minimum frequency ωTEn for which it can propagate, as illustrated in Figure 9.3.2. mode number n evanescence 3 TE3 propagation TE2 2 1 0 TEM ωTEM ωTE1 ωTE2 ωTE3 cut-off frequency TE1 ω Figure 9.3.2 Propagation frequencies of TEn modes in parallel-plate waveguides. The TE0 mode has zero fields everywhere and therefore does not exist. The TM0 mode can propagate even DC signals, but is identical to the TEM mode. This same relationship can also be expressed in terms of the maximum free-space wavelength λTEn that can propagate: λTEn = 2d n (cut-off wavelength λTEn for TEn mode) (9.3.10) For a non-TEM wave the longest free-space wavelength λ that can propagate in a parallel-plate waveguide of width d is 2d/n. - 280 - If ω < ωTEn, then we have an evanescent wave; kz, (9.3.6), and (9.3.7) become: k z 2 = ω2με − k x 2 < 0 , k z = ± jα (9.3.11) ( E for TEn mode, ω c kz Figure 9.3.4 Dispersion relation and vp and vg for a parallel-plate waveguide. The phase and group velocities of waves in parallel-plate waveguides do not equal c, as can be deduced the dispersion relation plotted in Figure 9.3.4: 2 k z 2 = ω2με − k x 2 = ω2 c2 − ( nπ d ) (waveguide dispersion relation) (9.3.14) The phase velocity vp of a wave, which is the velocity at which the field distribution pictured in Figure 9.3.1 moves to the right, equals ω/k, which approaches infinity as ω approaches the cut-off frequency ωTEn from above. The group velocity vg, which is the velocity of energy or information propagation, equals dω/dk, which is the local slope of the dispersion relation ω(k) and can never exceed c. Both vp and vg approach c = (με)-0.5 as ω → ∞. Example 9.3A A TM2 mode is propagating in the +z direction in a parallel-plate waveguide with plate separation d and free-space wavelength λo. What are E and H ? What is λz? What is the decay length αz-1 when the mode is evanescent and λo = 2λcut off? Solution: We can superimpose incident and reflected TM waves or use duality to yield H (x,y) = ŷ Hocoskxx e − jk z z , analogous to (9.3.3), where kxd = 2π for the TM2 mode. Therefore kx = 2π/d and kz = 2π/λz = (ko2 – kx2)0.5 = [(2π/λo)2 – (2π/d)2]0.5. Ampere’s law yields: E = ∇× H jωε = ( x̂∂H y ∂z + ẑH y ∂x ) jωε = ( x̂k z cosk x x + ẑjk x sink x x )( H o ωε)e − jk z z The waveguide wavelength λz = 2π/kz = (λo-2 – d-2)0.5. Cutoff occurs when k z = 0 , or ko = kx = 2π/d = 2π/λcut off. Therefore λo = 2λcut off ⇒ λo = 2d, and αo-1 = (-jkz) = (kx2 – ko2)0.5 = [d-2 – (2d)-2]-0.5/2π = d/(30.5π) [m]. - 282 - 9.3.2 Rectangular waveguides Waves can be trapped within conducting cylinders and propagate along their axis, rectangular and cylindrical waveguides being the most common examples. Consider the rectangular waveguide illustrated in Figure 9.3.5. The fields inside it must satisfy the wave equation: ( ∇2 + ω2με ) E = 0 ( ∇2 + ω2με ) H = 0 (9.3.15) where ∇ 2 ≡ ∂ 2 ∂x 2 + ∂ 2 ∂y 2 + ∂ 2 ∂z 2 . x b H(t) E(t) 0 z a λz/2 y Figure 9.3.5 Dominant mode in rectangular waveguide (TE10). Since the wave equation requires that the second spatial derivative of E or H equal E or H times a constant, these fields must be products of sinusoids or exponentials along each of the three cartesian coordinates, or sums of such products. For example, the wave equation and boundary conditions ( E // = 0 at x = 0 and y = 0 ) are satisfied by: E x = sin k y y ( A sin k x x + Bcos k x x ) e− jk z z (9.3.16) E y = sin k x x ( C sin k y y + D cos k y y ) e− jk z z (9.3.17) provided that the usual dispersion relation for uniform media is satisfied: k x 2 + k y 2 + k z 2 = k o 2 = ω2με (dispersion relation) (9.3.18) These field components must also satisfy the boundary condition E // = 0 for x = b and y = a, which leads to the guidance conditions: k x b = nπ (9.3.19) k y a = mπ (guidance conditions) - 283 - (9.3.20) We can already compute the cut-off frequency of propagation ωmn for each mode, assuming our conjectured solutions are valid, as shown below. Cut-off occurs when a mode becomes evanescent, i.e., when ko2 equals zero or becomes negative. At cut-off for the m,n mode: 2 ωmn με = k y 2 + k x 2 = ( mπ a ) + ( nπ b ) 2 2 ωmn = ⎣⎡( mπc a ) + ( nπc b ) ⎤⎦ 0.5 2 (9.3.21) (cut-off frequencies) (9.3.22) Since a ≥ b by definition, and the 0,0 mode cannot have non-zero fields, as shown later, the lowest frequency that can propagate is: ω10 = πc a (9.3.23) which implies that the longest wavelength that can propagate in rectangular waveguide is λmax = 2πc/ω10 = 2a. If a > 2b, the second lowest cut-off frequency is ω20 = 2ω10, so such waveguides can propagate only a single mode over no more than an octave (factor of 2 in frequency) before another propagating mode is added. The parallel-plate waveguides of Section 9.3.1 exhibited similar properties. Returning to the field solutions, E must also satisfy Gauss’s law, ∇ • εE = 0 , within the waveguide, where ε is constant and Ez ≡ 0 for TE waves. This implies: ∇ • E = 0 = ∂E x ∂x + ∂E y ∂y + ∂E z ∂z = ⎡k ⎣ x sin k y y ( A cos k x x −Bsin k x x ) + k y sin k x x ( C cos k y y − D sin k y y ) ⎤ e− jk z z ⎦ (9.3.24) The only way (9.3.24) can be zero for all combinations of x and y is for: A=C=0 (9.3.25) k y D = −k x B (9.3.26) The electric field for TE modes follows from (9.3.17), (9.3.18), (9.3.25), and (9.3.26): E = ( E o k o ) ( x̂k y sin k y y cos k x x − ŷk x sin k x x cos k y y ) e − jk z z (9.3.27) where the factor of ko-1 was introduced so that E o would have its usual units of volts/meter. Note that since kx = ky = 0 for the TE00 mode, E = 0 everywhere and this mode does not exist. The corresponding magnetic field follows from H = − ( ∇ × E ) jωμ : - 284 - H= ( Eo ηk o 2 ) ( x̂k x k z sin k x x cos k y y + ŷk y k z cos k x x sin k y y ) − jẑ k x 2 cos k x x cos k y y e− jk z z (9.3.28) A similar procedure yields the fields for the TM modes; their form is similar to the TE modes, but with E and H interchanged and then shifted spatially to match boundary conditions. The validity of field solutions where E and H are interchanged also follows from the principle of duality, discussed in Section 9.2.6. The most widely used rectangular waveguide mode is TE10, often called the dominant mode, where the first digit corresponds to the number of half-wavelengths along the wider side of the guide and the second digit applies to the narrower side. For this mode the guidance conditions yield kx = 0 and ky = π/a, where a ≥ b by convention. Thus the fields (9.3.27) and (9.3.28) become: E TE10 = E o x̂ ( sin k y y ) e− jk z z (dominant mode) HTE10 = ( Eo ωμ ) ⎡⎣ ŷk z sin ( πy a ) − jẑ ( π a ) cos ( πy a ) ⎤⎦ e− jk z z (9.3.29) (9.3.30) The fields for this mode are roughly sketched in Figure 9.3.5 for a wave propagating in the plusz direction. The electric field varies as the sine across the width a, and is uniform across the height b of the guide; at any instant it also varies sinusoidally along z. Hy varies as the sine of y across the width b, while Hx varies as the cosine; both are uniform in x and vary sinusoidally along z. The forms of evanescent modes are easily found. For example, the electric and magnetic fields given by (9.3.29) and (9.3.30) still apply even if kz = (ko2 - kx2 - ky2)0.5 ≡ -jα so that e-jkz becomes e-αz. For frequencies below cutoff the fields for this mode become: E TE10 = x̂ E o (sin k y y)e−αz ( (9.3.31) ) H TE10 = − j πE o ηak o 2 ⎡⎣ ŷ αsin (πy a ) + ẑ(π a ) cos(πy a )⎤⎦e−αz (9.3.32) The main differences are that for the evanescent wave: 1) the field distribution at the origin simply decays exponentially with distance z and the fields lose their wave character since they wax and wane in synchrony at all positions, 2) the electric and magnetic fields vary 90 degrees out of phase so that the total energy storage alternates twice per cycle between being purely electric and purely magnetic, and 3) the energy flux becomes purely reactive since the real (time average) power flow is zero. The same differences apply to any evanscent TEmn or TMmn mode. - 285 - Example 9.3B What modes have the four lowest cutoff frequencies for a rectangular waveguide having the dimensions a = 1.2b? For the TE10 mode, where can we cut thin slots in the waveguide walls such that they transect no currents and thus have no deleterious effect? Solution: The cut-off frequencies (9.3.22) are: ωmn = [(mπc/a)2 + (nπc/b)2]0.5, so the lowest cut off is for TEmn = TE10, since TE00, TM00, TM01, and TM10 do not exist. Next comes TE11 and TM11, followed by TE20. The wall currents are perpendicular to⎯H, which has no x component for the dominant mode (9.3.30); see Figure 9.3.5. Therefore thin slots cut in the x direction in the sidewalls (y = 0,a) will never transect current or perturb the TE10 mode. In addition, the figure and (9.3.30) show that the z-directed currents at the center of the top and bottom walls are also always zero, so thin zdirected slots at those midlines do not perturb the TE10 mode either. Small antennas placed through thin slots or holes in such waveguides are sometimes used to introduce or extract signals. 9.3.3 Excitation of waveguide modes Energy can be radiated inside waveguides and resonators by antennas. We can compute the energy radiated into each waveguide or resonator mode using modal expansions for the fields and matching the boundary conditions imposed by the given source current distribution J . Consider a waveguide of cross-section a×b and uniform in the z direction, where a ≥ b. If we assume the source current J s is confined at z = 0 to a wire or current sheet in the x,y plane, then the associated magnetic fields H + and H − at z = 0 ± δ , respectively (δ→0), must satisfy the boundary condition (2.1.11): H + − H − = J s × ẑ (9.3.33) Symmetry dictates H − ( x,y) = −H + ( x,y) for the x-y components of the fields on the two sides of the boundary at z = 0, assuming there are no other sources present, so: ẑ × (H + − H − ) = ẑ × (Js × ẑ ) = Js = 2 ẑ × H + ( x,y) (9.3.34) To illustrate the method we restrict ourselves to the simple case of TEm,0 modes, for which E and Js are in the x direction. The total magnetic field (9.3.28) summed over all TEm,0 modes and orthogonal to ẑ for forward propagating waves is: H + total = ŷ ∞ ∑ ⎣⎡E m,0mπ (ηak o2 )⎤⎦k zm sin(mπy a ) = (Js 2) × ẑ m =0 - 286 - (9.3.35) where Em is the complex amplitude of the electric field for the TEm,0 mode, and ky has been replaced by mπ/a. We can multiply both right-hand sides of (9.3.35) by sin(nπy/a) and integrate over the x-y plane to find: ∞ ∑ ⎡⎣ E m,0mπ ( ηak o2 )⎤⎦ k zm ∫∫ m =0 sin ( mπy a ) sin ( nπy a ) dx dy A = 0.5 ∫∫ Js ( x, y ) sin ( nπy a ) dx dy (9.3.36) A Because sine waves of different frequencies are orthogonal when integrated over an integral number of half-wavelengths at each frequency, the integral on the left-hand side is zero unless m = n. Therefore we have a simple way to evaluate the phase and magnitude of each excited mode: E n,0 = ⎡⎣ηk o 2 nbπk zn ⎦⎤ ∫∫ Js ( x,y)sin(nπy a ) dx dy (9.3.37) A Not all excited modes propagate real power, however. Modes n with cutoff frequencies above ω are evanescent, so kzn = [ko2 - (nπ/a)2]0.5 is imaginary. The associated magnetic field remains in phase with Js and real, and therefore the power in each evanescent wave is imaginary. Since all modes are orthogonal in space, their powers add; for evanescent modes the imaginary power corresponds to net stored magnetic or electric energy. The reactance at the input to the wires driving the current Js is therefore either capacitive or inductive, depending on whether the total energy stored in the reactive modes is predominantly electric or magnetic, respectively. A more intuitive way to understand modal excitation is to recognize that the power P delivered to the waveguide by a current distribution J s is: P= ∫∫∫ * E • Js dv [V] (9.3.38) V and therefore any mode for which the field distribution E is orthogonal to Js will not be excited, and vice versa. For example, a straight wire in the x direction across a waveguide carrying current at some frequency ω will excite all TEn0 modes that have non-zero E at the position of the wire; modes with cutoff frequencies above ω will contribute only reactance to the current source, while the propagating modes will contribute a real part. Proper design of the current distribution J s can permit any combination of modes to be excited, while not exciting the rest. - 287 - 9.4 Cavity resonators 9.4.1 Rectangular cavity resonators Rectangular cavity resonators are hollow rectangular conducting boxes of width a, height b, and length d, where d ≥ a ≥ b by convention. Since they are simply rectangular waveguides terminated at both ends by conducting walls, and the electric fields must still obey the wave equation, ( ∇2 + ω2με) E = 0 , therefore E for TE modes must have the form of the TE waveguide fields (9.3.27), but with a sinusoidal z dependence that matches the boundary conditions at z = 0 and z = d; for example, equal forward- and backward-propagating waves would form the standing wave: E = ( E o k o ) ( x̂k ysink y ycosk x x - ŷk x sink x xcosk y y ) ( Asink z z + Bcosk z z ) (9.4.1) where B = 0 ensures E // = 0 at z = 0, and kz = pπ/d ensures it for z = d, where p = 1, 2, ... Unlike rectangular waveguides that propagate any frequency above cut-off for the spatial field distribution (mode) of interest, cavity resonators operate only at specific resonant frequencies or combinations of them in order to match all boundary conditions. The resonant frequencies ωmnp for a rectangular cavity resonator follow from the dispersion relation: 2 2 ωmnp 2με = k y 2 + k x 2 + k z 2 = ( mπ a ) + ( nπ b ) + ( pπ d ) 2 2 2 ωmnp = ⎣⎡( mπc a ) + ( nπc b ) + ( pπc d ) ⎤⎦ 0.5 ⎡ s-1 ⎤⎦ ⎣r 2 (9.4.2) (cavity resonances) (9.4.3) The fundamental mode for a cavity resonator is the lowest frequency mode. Since boundary conditions can not be met unless at least two of the quantum numbers m, n, and p are non-zero, the lowest resonant frequency is associated with the two longest dimensions, d and a. Therefore the lowest resonant frequency is: 2 2 ω101 = ⎡⎣( πc a ) + ( πc d ) ⎤⎦ 0.5 [ radians/sec] (lowest resonance) (9.4.4) Cavity resonators are therefore sometimes filled with dielectrics or magnetic materials to reduce their resonant frequencies by reducing c. The fields for the fundamental mode of a rectangular cavity resonator, TE101, follow from (9.4.1) and Faraday’s law: E = x̂ E o sin ( πy a ) sin ( πz d ) (fundamental waveguide mode) - 288 - (9.4.5) ( ) H = jE o πωc2 η ⎡⎣ ŷ sin ( πy a ) cos ( πz d ) d − ẑ cos ( πy a ) sin ( πz d ) a⎤⎦ (9.4.6) The total energy w [J] = we(t) + wm(t) in each mode m,n,p of a cavity resonator can be calculated using (2.7.28) and (2.7.29), and will decay exponentially at a rate that depends on total power dissipation Pd [W] due to losses in the walls and in any insulator filling the cavity interior: w ( t ) ≅ w o e− Pd t w = w o e−ωt Q (9.4.7) Wall losses and any dissipation in insulators can be estimated by integrating (9.2.60) and (2.7.30), respectively, over the volume of the cavity resonator. The energy stored, power dissipation, and Q can be quite different for different modes, and are characterized by wmnp, Pd,mnp, and Qmnp, respectively, as defined by either (3.5.23) or (7.4.43): Qmnp = ωwmnp Pdmnp (9.4.8) Example 9.4A What are the lowest resonant frequency and its Q for a perfectly conducting metallic cavity of dimensions a, b, d if it is filled with a medium characterized by ε, μ, and σ. Assume Q >> 1. Solution: The lowest resonant frequency ω101 is given by (9.4.4), where c = (με)-0.5: ω101 = π(με)-0.5(a-2 + d-2)0.5. Q101 = ω101wT101/Pd101 where the total energy stored wT101 is twice the average electric energy stored since the total electric and magnetic energy storages are equal. At each point in the resonator the time-average electric energy 2 density stored is We = ε E 4 [J m-3] and the time-average power dissipated is 2 σ E 2 , [W m-3] so the electric-energy/dissipation density ratio everywhere is ε/2σ, and thus w T101 Pd101 = ε σ , so Q101 = πε ( με ) 9.4.2 −0.5 ( a −2 + d −2 )0.5 σ . Perturbation of resonator frequencies Often we would like to tune a resonance to some nearby frequency. This can generally be accomplished by changing the shape of the resonator slightly. Although the relationship between shape and resonant frequency can be evaluated using Maxwell’s equations, a simpler and more physical approach is taken here. The energy stored in a resonator can be regarded as a population of N trapped photons at frequency f bouncing about inside. Since the energy E per photon is hf (1.1.10), the total energy in the resonator is: w T = NhJf [ ] (9.4.9) - 289 - If we force the walls of a resonator to move slowly toward its new shape, they will move either opposite to the forces imposed by the electromagnetic fields inside, or in the same direction, and thereby do positive or negative work, respectively, on those fields. If we do positive work, then the total electromagnetic energy wT must increase. Since the number of photons remains constant if the shape change is slow compared to the frequency, positive work on the fields results in increased electromagnetic energy and frequency f. If the resonator walls move in the direction of the applied electromagnetic forces, the externally applied work on the fields is negative and the energy and resonant frequency decrease. The paradigm above leads to a simple expression for the change in resonant frequency of any resonator due to small physical changes. Consider the case of an air-filled metallic cavity of any shape that is perturbed by pushing in or out the walls slightly in one or more places. The electromagnetic force on a conductor has components associated with both the attractive electric and repulsive magnetic pressures on conductors given by (4.1.15) and (4.1.23), respectively. For sinusoidal waves these pressures are: 2 (electric pressure) (9.4.10) 2 (magnetic pressure) (9.4.11) Pe = −εo E o N 4 ⎣⎡ m-2 ⎤⎦ Pm = μo Ho N 4 ⎣⎡ m-2 ⎦⎤ But these pressures, except for the negative sign of Pe (corresponding to attraction), are the electric and magnetic energy densities [J m-3]. The work Δw done in moving the cavity boundary slightly is the pressure Pe/m applied, times the area over which it is applied, times the distance moved perpendicular to the boundary. For example, Δw equals the inward electromagnetic pressure (± energy density) times the increase in volume added by the moving boundary. But this increase in total stored electromagnetic energy is simply: Δw T = Nh Δf = −(Pe + Pm )Δvolume = Δw e − Δw m (9.4.12) The signs for the increases in electric and magnetic energy storage Δwe and Δwm and pressures Pe and Pm are different because the pressures Pe and Pm are in opposite directions, where Δwe = WeΔvol, and Δwm = -PmΔvol = -WmΔvol. Δwe is defined as the electric energy stored in the increased volume of the cavity, Δvol, assuming the electric field strength remains constant as the wall moves slightly; Δwm is defined similarly. The main restriction here is that the walls cannot be moved so far that the force density on the walls changes, nor can their shape change abruptly for the same reason. For example, a sharp point concentrates electric fields and would violate this constraint. Dividing (9.4.12) by wT = Nhf yields the frequency perturbation equation: - 290 - Δw T w T = Δf f = ( Δw e − Δw m ) w T = Δvol ( We − Wm ) w T (frequency perturbation) (9.4.13) A simple example illustrates its use. Consider a rectangular cavity resonator operating in the TE101 mode with the fields given by (5.4.37) and (5.4.38). If we push in the center of the top or bottom of the cavity where H ≅ 0 and E ≠ 0 we are reducing the volume allocated to electric energy storage, so Δwe is negative and the resonant frequency will drop in accord with (9.4.13). If we push in the sides, however, the resonant frequency will increase because we are reducing the volume where magnetic energy is stored and Δwm is negative; the electric energy density at the sidewalls is zero. In physical terms, pushing in the top center where the electric fields pull inward on the wall means that those fields are doing work on the moving wall and therefore lose energy and frequency. Pushing in where the magnetic fields are pushing outward does work on the fields, increasing their energy and frequency. This technique can be used to determine experimentally the unknown resonant mode of a cavity as well as tuning it. 9.5 Waves in complex media 9.5.1 Waves in anisotropic media There are many types of media that can be analyzed simply using Maxwell’s equations, which characterize media by their permittivity ε, permeability μ, and conductivity σ. In general ε, η, and σ can be complex, frequency dependent, and functions of field direction. They can also be functions of density, temperature, field strength, and other quantities. Moreover they can also couple⎯E to⎯B, and⎯H to⎯D. In this section we only treat the special cases of anisotropic media (Section 9.5.1), dispersive media (Section 9.5.2), and plasmas (Section 9.5.3). Lossy media were treated in Sections 9.2.4 and 9.2.5. Anisotropic media, by definition, have permittivities, permeabilities, and/or conductivities that are functions of field direction. We can generally represent these dependences by 3×3 matrices (tensors), i.e.: D = εE (9.5.1) B = μH (9.5.2) J = σE (9.5.3) For example, (9.5.1) says: D x = ε xx E x + ε xy E y + ε xz E z D y = ε yx E x + ε yy E y + ε yz E z (9.5.4) D z = ε zx E x + ε zy E y + ε zz E z - 291 - Most media are symmetric so that εij = εji; in this case the matrix ε can always be diagonalized by rotating the coordinate system to define new directions x, y, and z that yield zeros off-axis: ⎛ εx ⎜ ε=⎜ 0 ⎜0 ⎝ 0 εy 0 0⎞ ⎟ 0⎟ ε z ⎟⎠ (9.5.5) These new axes are called the principal axes of the medium. The medium is isotropic if the permittivities of these three axes are equal, uniaxial if only two of the three axes are equal, and biaxial if all three differ. For example, tetragonal, hexagonal, and rhombohedral crystals are uniaxial, and orthorhombic, monoclinic, and triclinic crystals are biaxial. Most constitutive tensors are symmetric (they equal their own transpose), the most notable exception being permeability tensors for magnetized media like plasmas and ferrites, which are hermetian49 and not discussed in this text. One immediate consequence of anisotropic permittivity and (9.5.4) is that D is generally no longer parallel to E , as suggested in Figure 9.5.1 for a uniaxial medium. When εxx ≠ εzz, E and D are parallel only if they lie along one of the principal axes. As explained shortly, this property of uniaxial or biaxial media can be used to convert any wave polarization into any other. x D Dx = εxxEx εxx ≠ εzz E Ex Ez z Dz = εzzEz Figure 9.5.1 D and E in an anisotropic medium. The origins of anisotropy in media are easy to understand in terms of simple models for crystals. For example, an isotropic cubic lattice becomes uniaxial if it is compressed or stretched along one of those axes, as illustrated in Figure 9.5.2(a) for z-axis compression. That such compressed columns act to increase the effective permittivity in their axial direction can be understood by noting that each of these atomic columns functions like columns of dielectric between capacitor plates, as suggested in Figure 9.5.2(b). Parallel-plate capacitors were 49 Hermetian matrices equal the complex conjugate of their transpose. - 292 - discussed in Section 3.1.3. Alternatively the same volume of dielectric could be layered over one of the capacitor plates, as illustrated in Figure 9.5.2(c). (a) (b) V + z d y x - A A εo ε εz > εx = εy (c) d/2 εo ε ⇒ εeff(a) ⇒ εeff(b) Figure 9.5.2 Uniaxial crystal and anisotropically filled capacitors. Even though half the volume between the capacitor plates is occupied by dielectric in both these cases, the capacitance for the columns [Figure 9.5.2(b)] is greater, corresponding to a greater effective permittivity εeff. This can be shown using Equation (3.1.10), which says that a parallel plate capacitor has C = εeffA/d, where A is the plate area and d is the distance between the plates. The capacitances Ca and Cb for Figures 9.5.2(b) and (c) correspond to two capacitors in parallel and series, respectively, where: Ca = ( εA 2d ) + ( εo A 2d ) = ( ε + εo ) A 2d = εeff (a) A d −1 −1 Cb = ⎣⎡( εA2 d ) + ( εo A2 d ) ⎤⎦ −1 = ⎡εε ⎣ o ( ε + εo ) ⎤⎦ 2A d = εeff (b) A d (9.5.6) (9.5.7) In the limit where ε >> εo the permittivity ratio εeff(a)/ εeff(b) → ε/4εo > 1. In all compressive cases εeff(a) ≥ εeff(b). If the crystal were stretched rather than compressed, this inequality would be reversed. Exotic complex materials can exhibit inverted behavior, however. Since the permittivity here interacts directly only with E , not H , the velocity of propagation c = 1/ με depends only on the permittivity in the direction of E . We therefore expect slower propagation of waves linearly polarized so that E is parallel to an axis with higher values of ε. We can derive this behavior from the source-free Maxwell’s equations and the matrix constitutive relation (9.5.4). ∇ × E = − jωμH (9.5.8) ∇ × H = jωD (9.5.9) ∇•D = 0 (9.5.10) ∇•B = 0 (9.5.11) - 293 - Combining the curl of Faraday’s law (9.5.8) with Ampere’s law (9.5.9), as we did in Section 2.3.3, yields: ∇ × ( ∇ × E ) = ∇ ( ∇ • E ) − ∇ 2 E = ω2μD (9.5.12) We now assume, and later prove, that ∇ • E = 0 , so (9.5.12) becomes: ∇ 2 E + ω2μD = 0 (9.5.13) This expression can be separated into independent equations for each axis. Waves propagating in the z direction are governed by the x and y components of (9.5.13): ( ) (9.5.14) ( ) (9.5.15) ⎡ ∂ 2 ∂z 2 + ω2με ⎤ E = 0 x⎦ x ⎣ ⎡ ∂ 2 ∂z 2 + ω2με ⎤ E = 0 y⎦ y ⎣ The wave equation (9.5.14) characterizes the propagation of x-polarized waves and (9.5.15) characterizes y-polarized waves; their wave velocities are (μεx)-0.5 and (μεy)-0.5, respectively. If εx ≠ εy then the axis with the lower velocity is called the “slow” axis, and the other is the “fast” axis. This dual-velocity phenomenon is called birefringence. That our assumption ∇ • E = 0 is correct is easily seen by noting that the standard wave solution for both x- and y- polarized waves satisfies these constraints. Since ∇ is distributive, the equation is also satisfied for arbitrary linear combinations of x- and y- polarized waves, which is the most general case here. If a wave has both x- and y-polarized components, the polarization of their superposition will evolve as they propagate along the z axis at different velocities. For example, a linearly wave polarized at 45 degrees to the x and y axes will evolve into elliptical and then circular polarization before evolving back into linear polarization orthogonal to the input. This ability of a birefringent medium to transform polarization is illustrated in Figure 9.5.3. In this case we can represent the linearly polarized wave at z = 0 as: E ( z = 0 ) = E o ( x̂ + yˆ ) (9.5.16) If the wave numbers for the x and y axes are kx and ky, respectively, then the wave at position z will be: ( E(z) = Eoe− jk x z xˆ + ye ˆ j( k x − k y ) z ) (9.5.17) The phase difference between the x- and y-polarized components of the electric field is therefore Δφ = (kx - ky)z. As suggested in the figure, circular polarization results when the two - 294 - components are 90 degrees out of phase (Δφ = ±90o), and the orthogonal linear polarization results when Δφ = 180o. 90o Δφ (x axis is slow) y z x o 180 0o 45 linear polarization o -45o linear polarization left-elliptical polarization left-circular polarization 270o Figure 9.5.3 Polarization conversion in a birefringent medium. Polarization conversion is commonly used in optical systems to convert linear polarization to circular, or vice-versa, via a quarter-wave plate for which Δφ is 90o, equivalent to a quarter wavelength. A half-wave plate (Δφ = 180o) reverses the sense of any polarization. Example 9.5A A certain birefringent medium is characterized by μo, εx = 2εo, εy = 2.002εo. How thick D must a quarter-wave plate be if λ = 5×10-7[m] in free space (visible light)? At what thickness D' might this same plate rotate appropriate linear polarization 90 degrees? − jk D Solution: The phase lags along the x and y axes arise from e− jk x D and e y , respectively, and the difference is π/2 = (ky - kx)D for a quarter-wave plate. But ki = ω(μoεi)0.5, so (ky - kx)D = ω(μoεx)0.5[(1 + 1.001)0.5 - 1] D ≅ (ω/cx)0.5 0.0005 D = π/2. Since ω/cx = 2π/λx, therefore D = 2000λx/4 where λx = 5×10-7(εo/εx)0.5. Thus D = 500λx = 0.18 mm, which is approximately the thickness of a Vu-Graph transparency that acts as a quarter-wave plate. A differential phase lag of π yields 90° polarization rotation for waves linearly polarized at an angle 45° from the principal axes x and y, so the thickness would be doubled to ~0.36 mm. 9.5.2 Waves in dispersive media Dispersive media have wave velocities that are frequency dependent due to the frequency dependence of μ, ε, or σ. These frequency dependencies arise in all materials because of the non-instantaneous physical responses of electrons to fields. Often these time lags are so brief that only at optical frequencies do they become a significant fraction of a period, although propagation over sufficiently long paths can introduce significant cumulative differences in effects across any frequency band or gap. Only vacuum is essentially non-dispersive. - 295 - The principal consequence of dispersion is that narrowband pulse signals exhibit two velocities, the phase velocity vp of the sinusoids within the pulse envelope, and the group velocity vg at which the pulse envelope, energy, and information propagate. Because energy and information travel at the group velocity, it never exceeds the velocity of light although phase velocity frequently does. A simple way to reveal this phenomenon is to superimpose two otherwise identical sinusoidal waves propagating at slightly different frequencies, ω ± Δω; superposition is valid because Maxwell’s equations are linear. The corresponding wave numbers are k ± Δk, where Δk << k and Δω << ω. Such a superposition for two sinusoids propagating in the +z direction is: E(t, z) = E o cos [( ω + Δω) t − ( k + Δk ) z ] + Eo cos [( ω − Δωt ) − ( k − Δk ) z ] = E o 2 cos ( ωt − kz ) cos ( Δωt − Δkz ) (9.5.18) where we used the identity cos α + cos β = 2 cos[(α+β)/2] cos[(α−β)/2]. The first factor on the right-hand side of (9.5.18) is a sine wave propagating at the center frequency ω at the phase velocity: vp = ω k (phase velocity) (9.5.19) The second factor is the low-frequency, long-wavelength modulation envelope that propagates at the group velocity vg = Δω/Δk, which is the slope of the ω(k) dispersion relation: vg = ∂ω ∂k = ( ∂k ∂ω) −1 (group velocity) (9.5.20) Figure 9.5.4(a) illustrates the original sinusoids plus their superposition at two points in time, and Figure 9.5.4(b) illustrates the corresponding dispersion relation. ω + Δω (a) E(z, to) (b) ω z ω - Δω ω E(z, t1) z vp peak null 0 vg slope = Δω/Δk = vg slope = ω/k = vp k k null Figure 9.5.4 Phase and group velocity for two superimposed sinusoids. - 296 - Note that this dispersion relation has a phase velocity that approaches infinity at the lowest frequencies, which is what happens in plasmas near the plasma frequency, as discussed in the next section. Communications systems employ finite-duration pulses with Fourier components at all frequencies, so if such pulses travel sufficiently far even the envelope with its finite bandwidth will become distorted. As a result dispersive media are either avoided or compensated in most communications system unless the bandwidths are sufficiently narrow. Compensation is possible because dispersion is a linear process so inverse filters are readily designed. Section 12.2.2 discusses dispersion further in the context of optical fibers. Example 9.5B When ω = co0.5, what are the phase and group velocities vp and vg in a medium having the dispersion relation k = ω2/co? Solution: vp = ω/k = co/ω = co0.5 [m s-1]. vg = (∂k/∂ω)-1 = co/2ω = co0.5/2 [m s-1]. 9.5.3 Waves in plasmas A plasma is a charge-neutral gaseous assembly of atoms or molecules with enough free electrons to significantly influence wave propagation. Examples include the ionosphere50, the sun, interiors of fluorescent bulbs or nuclear fusion reactors, and even electrons in metals or electron pairs in superconductors. We can characterize fields in plasmas once we know their permittivity ε at the frequency of interest. To compute the permittivity of a non-magnetized plasma we recall (2.5.8) and (2.5.13): D = εE = εo E + P = εo E + nqd (9.5.21) where q = -e is the electron charge, d is the mean field-induced displacement of the electrons from their equilibrium positions, and n3 is the number of electrons per cubic meter. Although positive ions are also displaced, these displacements are generally negligible in comparison to those of the electrons because the electron masses me are so much less. We can take the mass mi of the ions into account simply by replacing me in the equations by mr, the reduced mass of the electrons, where it can be shown that mr = memi/(me + mi) ≅ me. To determine ε in (9.5.21) for a collisionless plasma, we merely need to solve Newton’s law for d(t) , where the force f follows from (1.2.1): 2 f = qE = ma = m ( jω) d (9.5.22) 50 The terrestrial ionosphere is a partially ionized layer at altitudes ~50-5000 km, depending primarily upon solar ionization. Its peak electron density is ~1012 electrons m-3 at 100-300 km during daylight. - 297 - Solving (9.5.22) for d and substituting it into the expression for P yields: P = nqd = −ned = Ene2 m−1 ( jω) −2 (9.5.23) Combining (9.5.21) and (9.5.23) yields: 2 D = εo E + P = εo ⎡⎣1+ ne2 m ( jω) εo ⎦⎤ E = εo ⎡⎣1− ω2p ω2 ⎦⎤ E = εE (9.5.24) where ωp is defined as the plasma frequency: ( ωp ≡ ne2 mεo ) 0.5 ⎡⎣radians s-1 ⎤⎦ (9.5.25) The plasma frequency is the natural frequency of oscillation of a displaced electron or cluster of electrons about their equilibrium location in a neutral plasma, and we shall see that the propagation of waves above and below this frequency is markedly different. The dispersion relation for a collisionless non-magnetic plasma is thus: ( k 2 = ω2με = ω2μo εo 1− ωp 2 ω2 ) (9.5.26) which is plotted as ω(k) in Figure 9.5.5 together with the slopes representing the phase and group velocities of waves in plasmas. ω normal wave propagation k = ω/c; slope = c ωp slope = vg = ∂ω/∂k evanescence slope = vp = ω/k 0 k = (ω/c)[1 - (ωp/ω)2]0.5 k Figure 9.5.5 Dispersion relation and velocities for a simple plasma. Using the expressions (9.5.19) and (9.5.20) for phase and group velocity we find for plasmas: 2 v p = ω k = c ⎡1 − ( ωp ω) ⎤ ⎣ ⎦ −0.5 - 298 - (9.5.27) vg = ( ∂k ∂ω) −1 2 = c ⎡1 − ( ωp ω) ⎤ ⎣ ⎦ 0.5 (9.5.28) Since vpvg = c2, and since vg ≤ c, it follows that vp is alway equal to or greater than c. However, for ω < ωp we find vp and vg become imaginary because normal wave propagation is replaced by another behavior. From (9.5.26) we see that when ω < ωp: k = ω 1 − (ωp /ω) 2 = ± jα c η= (9.5.29) μ ω μ 2 = ηo / 1 − ( ωp ω) = ∓ j o α ε (9.5.30) Therefore an x-polarized wave propagating in the +z direction would be: E = x̂E o e −αz (9.5.31) H = ŷ η−1 Eo e−αz = jŷ ( α ωμo ) Eoe−αz (9.5.32) where the sign of ±jα was chosen (-) to correspond to exponential decay of the wave rather than growth. We find from (9.5.32) that H(t) is delayed 90o behind E(t), that both decay exponentially with z, and that the Poynting vector⎯S is purely imaginary: ∗ ( S = E × H = − jzˆ α E o 2 ) ωμo e−2αz (9.5.33) Such an evanescent wave decays exponentially and carry only reactive power and no timeaverage power because of the time orthogonality of E and H. Reactive power implies that below ωp the average energy stored is predominantly electric, but in this case the stored energy is actually dominated by the kinetic energy of the electrons. It is this extra energy that allows the permittivity ε to become negative below ωp although μo remains constant. The frequency ωp below which conversion from propagation to evanescence occurs is called the cut-off frequency, which is the plasma frequency here. Example 9.5C What is the plasma frequency fp [Hz]of the ionosphere when ne = 1012 m-3? Solution: fp = ωp/2π = (1012e2/mεo)0.5/2π ≅ 9.0 MHz. - 299 -

6.013 Electromagnetics and Applications, Chapter 9: Electromagnetic Waves

Related Documents

Recommended Documents

Description

Related Documents

Free up your schedule!

Take 5 seconds to unlock