Finite Element Analysis of Reverberation Chambers: A Two-Dimensional Study at Cutoff Charles F. Bunting, Ph.D. Old Dominion University Dept. of Eng. Technology Norfolk, VA 23529 e-mail: cfblOOf@et.odu.edu Karl J. Moeller C. J. Reddy, Ph.D. Stephen A. Scearce Electromagnetics Res. Br. Mail Stop 490 NASA Langley Research Center Hampton, VA 23681 Hampton University Dept. of Electrical Engineering Hampton, VA 23668 Old Dominion University Dept. of Electrical Engineering Norfolk, VA 23529 Abstract: A two-dimensional analysis of reverberation chambers is performed at cutoff. The structure considered is lossless and corresponds to an infinite Q chamber. The concept of frequency stirring is used to generate field data for a discrete set of modes and the resulting statistics are analyzed. The total field and power are examined for TE modes. This analysis yields statistics similar to the expected reverberation chamber statistics for the fields. Preliminary data for mechanical stirring is also presented. The twodimensional model may yield considerable insight into the nature of the loss experienced on real airframes, and also provide an error measure on data in the non-ideal mode-stirred chamber. INTRODUCTION + Wu concludes that the amount of frequency shift for a given stirrer essentially determines the random nature of the fields. Wu also makes an important observation regarding the Figure 1. Basic 2D Reverberation Chamber. consequences of the shifting eigenfrequencies in noting that the effect of a large stirrer is essentially a form of random A reverberation chamber is an enclosure consisting of metal walls with a metallic paddle wheel (denoted a “stirrer” modulation. This modulation is considered to have both or “tuner”) essentially forming a high quality factor (Q) cavity amplitude and frequency modulation. Wu concludes that an with continuously variable boundary conditions. The fields ineffective (electrically small) stirrer is unable to provide inside the chamber for a given tuner position are completely adequate frequency modulation effects. Hill [3] examined the concept of electronic mode stirring deterministic. However, given the nature of the variable initially suggested by Loughry [4]. Hill’s analysis provides boundary condition, the ability of a given source to couple considerable insight into the connection between a twoenergy into certain modes, and the passband characteristic due the chamber Q, the fields are typically characterized by dimensional versus and three-dimensional analysis. Electronic statistical means. Specifically the probability density mode stirring essentially implies that instead of simply stirring functions for the in-phase and quadrature component of the via mechanical means to obtain a statistical field distnbution, electric and magnetic fields are normally distributed [ 11. The the source can be swept over a narrow bandwidth to effect the field magnitudes are Rayleigh distributed and the power is shift in eigenfiequencies as suggested by Wu [2]. Loughry exponentially distributed [l]. The degree to which the actual uses bandwidth-limited white Gaussian noise as the fields conform to the specific statistical description is modulation source. Hill indicates that there is freedom in the dependent upon the number of modes that can simultaneously signal used, but constrains the signal to be spectrally flat over exist at a given frequency. The number of modes that can the bandwidth. Hill’s analysis employs a two-dimensional exist at a given frequency is a function of the cavity Q and the rectangular cavity with an electric line source. By utilizing a numerical analysis scheme such as finite operating frequency. elements the effects of introducing loss may be accurately The transmission line matrix (TLM) method was simulated, however, for this paper a two-dimensional lossless employed by Wu [ 2 ] to examine the characteristics of the siniulation will be performed. A two-dimensional model may fields in the presence of a large stirrer in the reverberation yield insight into the nature of the loss experienced on real chamber. Wu used a 2-D cavity with a 1-D stirrer model to airframes and also provide a n error measure on data in the esainine the shifting of the eigenfrequencies and the effects on ideal mode-stirred chamber. the mode amplitudes for a variety of different stirrer sizes. This paper will consider the basic 2D geometry as depicted in Figure 1. An anti-symmetric tuner may be rotated to provide mechanical mode stirring, however the focus of this early work is to examine the use of a frequency stirring approach. A FORTRAN code applying finite element techniques has been written that provides the electric field at any point within the chamber. FINITE ELEMENTS FOR THE REVERBERATION CHAMBER [5] Consider Maxwell's curl equations in a source free, homogeneous, isotropic, time harmonic form given by and (7) The transverse components will be discretized using edge elements and the longitudinal component used node-based Lagrange fimctions. After integrating and summing over all elements in the domain the following eigenvalue problem of the form Ax = ABx for the cutoff frequency is obtained v x E = -j w f i , V x H = jwEE, Although analytical results are available for a large class of problems for static, quasi-static, and dynamic conditions, numerical methods typically must be applied whenever the geometry does not coincide with a separable coordinate system. A full-field approach is required when the medium is characterized by higher order modes, hybrid modes such as for a general inhomogeneous media, or a waveguide or cavity with imperfectly conducting walls. Other problems can be expressed in a scalar form when there is a distinct separation between TE and TM modes. In general this approximation cannot be applied and a fill vector formulation is required. The solution for the fields inside the reverberation chamber will be required to support hybrid modes in the presence of any metallic discontinuity or an inhomogeneity due lossy objects. For the finite element approach the electric field in terms of all three vector components is the desired field quantity. A two dimensional representation for the fields may be expanded in terms of the transverse field and the longitudinal field where RESULTS A two-dimensional model of NASA Langley Research Center's B chamber with dimension 3.96m x 7.10m was generated and is depicted in Figure 2. The resulting system of equations was on the order of 1O,OOOx1O,OOO and was solved using a CONVEX 220 using a sparse eigenvalue solver in FORTRAN. The data was then processed in MATLAB. The solution of eq. (8) has on the order of 10,000 eigenvalues and eigenvectors as possible values of cutoff frequency and corresponding field configurations. Although there are a huge number of solutions available, only those solutions corresponding to a discretization of 10 samples per wavelength were retained. This level of discretization allowed accurate solutions up to 300 MHz, and resulted in a possible 171 valid cutoff frequencies. A typical result for the field behavior is shown in Fig. 3 for the x-component of the electric field, Ex. Note that the hghest field levels are in the immediate vicinity of the stirrer as would be expected since s and setting Expressing the del operator as V = VI + 2 y = 0 for cutoff, it is possible to write two separate equations one for the transverse part and another for the z component in terms of the cutoff frequency, Figure 2. Discretization for the two-dimensional geometry. leading to the following weighted residual form: the fields become infinite near the tuner. Additionally, consider a contour plot of the s-component of the electric field as given in Fig. 4. In this view it is possible to see the large the modal fields represented by the eigenvectors exist at amplitude variations of the field near the tuner. discrete frequencies and the chamber Q does not directly enter Typical experimental work using a reverberation chamber into the formulation. The implication is that the computed involves the rotation of the tuner and a measurement of the modal fields are for an infinite Q chamber, and that multireceived power to determine such characteristics as the stirring moding does not occur. ratio, or a measurement of the VSWR based on reflected power at the transmit port. The rotation of the tuner provides a variable boundary condition that results in a deterministic Frequency Stirring variation of the field by applying finite elements. One issue of Motivated by the notion that the action of stirring excites concern for the eigenvalue problem of eq. (8) involving the cutoff frequency is that although all modes are found the modes weighted by a factor governed by the quality factor corresponding to the particular geometrical configuration, the of the chamber, and noting that frequency stirring may be weighting of the modes is not possible. In an actual used in place of mechanical stirring [3] an interesting reverberation chamber the source does not couple energy experiment was performed. For a fixed tuner position equally between the possible modes. Another concern is that computing all the eigensolutions, within the accuracy constraint based on proper discretization, corresponds to Interpolated Ex, Frequency (MHz) 4 0 8 1143, Date=K-Feb1998 chmbq computing all the possible modes that could be excited in the 04 chamber. The solutions for cutoff result in a field that is either 03 TE or TM as the frequency is varied. Consequently, when the solution is TE only Ex and Ey exist and when the solution is 02 TM only E, exists. Figure 5 depicts the amplitude of Ex at 01 one point in the chamber as the frequency is varied. The field G o variations increase with frequency. This is related to the 01 increased mode density at higher frequencies. A histogram -0 2 was constructed using MATLAB in Fig. 6 based on the data -0 3 of Fig. 5. Any single component should have a Gaussian 04 distribution [ 11. The results are approximately Gaussian A considering the limited number of data points used (101 total -2 .J points for TE). y In m. Mode Num =21 x in m. Actual Grid Size -100 Figure 3. The x-component of the electric field for a frequency of 108.1143 MHz. Ex m).Element Number =5021. Dale=C+Mar-1998 0.5 InlerpOlatedEx. Frequency (MHz) =108.1143. Dale=25Fet-1998 chmbr2 0.4. 03 15 02 1 01 0.3. 0.2 c N ' fi0.10: 0 05 -, U- -01 -0.1 ; o 1 - . -0.2 -02 E -05 -0.3. -03 > . Jy -0.4. 1 -04 -0.5 .o 15 3 2 1 0 1 2 0 5 3 x in m. Actual Gnd Size =lo0 Figure 4. Contour plot of the x-component of the electric field for a frequency of 108.1143 MHz. 210 50 100 150 200 250 3CU 350 Figure 5. The x-component of the electric field for a fixed point inside the chamber as a function of frequency. Hslogramb r E% Element Number =502!. Dale=Z5Febl998 Histogramfor Elo&. Element Number =5GZ1. Dale=C&Mar-1998 14 5 1 12 10 -E $ 8 .R c 6 r 4 2 0 005 01 015 02 025 03 035 04 EloGE amplilude.20 bim. xbar =-O 83694. ybar =-I 617 0 Figure 6. The histogram of the data of Figure 5. Figure 8. The histogram for Figure 7. '(E). Element Number =5021. E,, EO l l (E). Element Number =5021. Dale=C&Mar-IPB 0,45 I -! E 025 2 02. 0.05 - 0.1 u-008 0.06 . 0.04 0.02 0 01 0 1 0.12 ~ 0.1 I 0.14L - 0.15 I 0.16 0 35 - I I O.I8i I 0.41 0.3 Dale=06Mar-1998 0.2 50 100 150 200 250 f i n MW. xbar=-O.@3694. ybar=-l 617 300 0 1 50 350 100 150 200 250 f i n MM.xbar =4.83694. ybar=-1.617 350 3M Figure 9. The power for a single point in the reverberation chamber. Figure 7. The total field for a fixed point in the reverberation chamber. It is also of interest to examine the total field. Considering a field that is TE we can write Etot as, Hislopramb r Elar,2. 25 , , , Element Number =5021. D a l e W a r - 1 9 8 8 , , , , , , Examining the total field for the same point described earlier in Figs. 4-5 results in the results of as shown in Fig. 7 and Fig. 8. The histogram of the data in Fig. 7 closely resembles the expected Rayleigh distribution [ 11. It is also possible to examine a term that is proportional to the power where 2 EIotE2 amplilude.20 bins, xbar=-0.83694. P CC Elf,l= Ef + E t The data proportional to the power is depicted in Fig. 9 with the corresponding histogram in Fig. 10. The data resembles the expected exponential distribution [I]. 21 1 ybar=-1.617 Figure 10. The histogram for Figure 9. Mechanical Stirring The action of mechanical stirring is an important aspect of the reverberation chamber analysis. Applying finite elements to a rotating stirrer requires a significant amount of data processing since the field will need to be found for each tuner position. For the purposes of this investigation the tuner was rotated in 1.6” increments for a total of 225 steps. The increment was chosen to match the step size used by NIST when the NASA Langley chambers were characterized in early 1997. The finite element discretization was reduced to a system s u e of 3,000 unknowns resulting in an upper frequency limit of 150 MHz. The eigenvalue problem of eq. (8) was solved for each position of the tuner which resulted in a total of 43 eigenvalues between the frequencies of 20 MHz and 150 MHz. Figure 11 depicts the total field at approximately 60 MHz for at a point located far from the tuner and greater than half a wavelength from any walls. The total field is periodic and relatively smooth. Another interesting study not shown in this paper is the change in frequency as a function of tuner position. Additionally, movies of the field have been made in which each frame corresponds to the solution for a particular tuner position. available random access memory (RAM) required for in-core processing. Virtual storage, or out-of-core processing, of the data can also be used, but greatly slows the solution time. Taking advantage of sparsity in the resulting system of equations helps significantly. In spite of the advantages gained by using the sparse eigenvalue solver optimized for the CONVEX the system of equations was limited to the value used in this paper at around 1O,OOOx1O,OOO. The CONVEX 220 at NASA-Langley has 1 GB of random access memory. By taking advantage of sparsity the dimensions of the matrices A and B in eq. (8) are reduced to approximately 10,000~7. However, the computation of the eigenvectors results in a full matrix. Work is continuing in the development of a high speed sparse complex eigensystem solver that will allow the solution of much larger systems (with on the order of 60,000 unknowns) to be analyzed. Other current work includes the coupling of a source to the fields in the 2D geometry by solving the eigenvalue problem for the propagation constant and including a source. The resulting solution would contain the proper mode weighting for the particular configuration depending upon the stirrer position. Future work will examine the three dimensional problem at low frequencies that will include the source. E tatal. mode number =7. I =-1.0975. y =a66 ACKNOWLEDGEMENTS This work was sponsored by the American Society of Engineering Education under an ASEE Faculty Fellowship at NASA Langley during the Summer of 1997. REFERENCES [ l ] Lehman, T.H., “A statistical theory of electromagnetic fields in complex cavities,” EMP Interaction Notes, No. 494, 1993. Figure 11. The total field for a 225 tuner steps. [2] D.L. Wu and D.C. Chang, “The effect of an electrically large stirrer in a mode-stirred chamber,” IEEE Trans. on Electromag. Compat., vol. 31, no. 2, pp. 164-170, 1989. CONCLUSIONS This paper has presented a two-dimensional analysis of reverberation chambers using finite element techniques. The results of the finite element analysis of the reverberation chamber illustrate the potential utility of a 2D representation for enhancing the basic statistical characteristics of the chamber. One important advantage of employing a numerical scheme such as finite elements to analyze reverberation chambers is that the field values at virtually any point in the structure can be easily obtained without perturbing the field as can occur when using an antenna to monitor the fields. The challenge of agplying finite element techniques to reverberation chambers is that enormous computational resources are required. The primary limitation is in the [3] D.A. Hill, “Electronic mode stirring for reverberation chambers,” IEEE Trans. on Electromag. Compat.? vol. 36, pp. 294-299, 1994. [4] T.A. Loughry, “Frequency-stirring: an alternative to mechanical mode-stirring for the conduct of electromagnetic testing,” Philipps Laboratory, Kirtland AFB, NM PL-TR-91-1036, NOV.1991. [5] Reddy, C.J., M.D. Deshpande, C.R. Cockrell. and F.B. Beck, “Finite element method for eigenvalue problems in electromagnetics,” NASA Technical Paper 3485. NASA Langely Research Center, Dec. 1994.