UFDC Home  Search all Groups  UF Institutional Repository  UF Institutional Repository  UF Theses & Dissertations  Vendor Digitized Files   Help 
Material Information
Subjects
Notes
Record Information

Full Text 
A SIMPLIFIED METHOD OF SYNTHESIZING LADDER NETWORKS WITH IMAGEPARAMETER HALFSECTIONS By DAVID SILBER A DISSERTATION PRESENTED TO THE GRADUATE COUNCIL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA August, 1960 Copyright by David Silber 1960 ACKNOWLEDGMENTS The author would like to express his gratitude to the members of his supervisory committee for their advice and helpful criticism. He is especially indebted to Dr. T. S. George for his supervision and constant guidance, to Dr. W. H. Chen for his many valuable suggestions and to Dr. H. A. Meyer, Director of the Statistical Laboratory, for his help in obtaining numer ical data by using the University's electronic computer. TABLE OF CONTENTS ACKNOWLEDGMENTS ............... . LIST OF TABLES ............ LIST OF ILLUSTRATIONS . . CHAPTER I. INTRODUCTION . . . 1.1. The Use of Frequency Selective Networks and Terminology . . 1.2. The Image Parameter Method of Network Synthesis . . 1.3. The Insertion Loss Method of Network Synthesis . . . 1.4. Comparison of the Image Parameter and Insertion Loss Methods . . 1.5. Statement of Objectives . . II. THE GENERAL LADDER NETWORK IN TERMS OF ELEMENTARY BUILDING BLOCKS CONNECTED IN TANDEM . 2.1. The Elementary Building Block . 2.2. Methods of Connecting the Elementary Building Blocks . . . 2.3. A Conventional LowPass Ladder in Terms of Elementary Building Blocks . . 2.4. Filters with Incidental Dissipation in Terms of Elementary Building Blocks . III. DERIVATION OF THE IMAGE PARAMETERS FOR LADDER FILTERS HAVING PRESCRIBED INSERTION LOSS CHARACTERISTICS . . . : : : Page ii vi 1 1 3 5 6 8 10 10 12 15 23 26 Page 3.1. Outline of Approach ... 26 3.2. Insertion Loss of a Twosection Ladder Filter in Terms of Image Parameters 28 3.3. Tchebycheff and Butterworth Filters Syn thesized with Image Parameter Halfsections 33 3.4. The Elliptic Function Filter Synthesized with Image Parameter Halfsections 44 IV. ANALYSIS OF RESULTS AND CONCLUSIONS 61 4.1. Analysis of Obtained Data . 61 4.2. Suggested Modification of Zobel Designs with the View of Obtaining More Efficient Filters . . 63 4.3. Limits of Improvement of a Zobel Filter 70 4.4. Summary and Conclusions . 73 REFERENCES . . 76 BIOGRAPHY ......... . 79 LIST OF TABLES TABLE Page 1. CUTOFF FREQUENCY w. FOR THE BUTTERWORTH TWOSECTION FILTER . 40 2. IMAGE PARAMETER VALUES FOR THE TCHEBYCHEFF TWO SECTION FILTER ................. 45 3. IMAGE PARAMETER VALUES FOR THE ELLIPTICFUNCTION TYPE TWOSECTION FILTER . .. .. 58 LIST OF ILLUSTRATIONS Figure Page 2.1. The elementary building block . 11 2.2. Image impedance of the elementary building block at terminals 10 of Fig. 2.1 . 11 2.3. Image impedance of the elementary building block at terminals 20 of Fig. 2.1 . .. 11 2.4. Tandem connection of the elementary building blocks giving a nonsymmetrical ladder filter 13 2.5. Connection of the elementary building blocks by which symmetrical and nonsymmetrical ladder filters may be obtained . . 13 2.6. A special case of the ladder of Fig. 2.5 16 2.7. Possible locations of elements with arbitrary values . . 20 2.8. Effect of adding proportionately equal amounts of resistance on the location of poles and zeros of the insertion voltageration function 24 3.1. A twosection ladder filter . 29 3.2. Elementary building blocks from which the ladder of Fig. 3.1 may be obtained . .. 29 3.3. Equivalent lattice of the ladder of Fig. 3.1 31 3.4. A Butterworth and Tchebycheff type ladder filter and its equivalent lattice . 31 3.5. Insertion loss characteristics for (a) Butterworth and (b) Tchebycheff twosection filters .. .. 36 3.6. Cutoff frequency of elementary building blocks versus maximum attenuation in passband for the design of filters with Butterworth characteristics 41 3.7. A Butterworth filter in terms of elementary building blocks with WUc = 1.272 and Lp 3 db 42 vi  Figure Page 3.8. Image parameters oc rI and r2 versus maximum peak ripple in passband for the design of filters with Tchebycheff characteristic . 47 3.9. A Tchebycheff filter with peak ripple in passband L= 1.5 db in terms of elementary building blocks with wC = 1.0 . . 48 3.10. Insertion loss characteristic of an elliptic function twosection filter . . 51 3.11. Minimum stqpband attenuation L versus selectivity factor k for the twosection efliptic function filter . . 53 3.12. Variation of minimum attenuation in stopband with maximum ripple in passband . 53 3.13. Image parameters rl, r2, r3 and r4 for k 0.7 versus maximum ripple in passband for the design of elliptic function filters . . 60 4.1. Typical insertion loss characteristic in passband for a Zobel filter . . 66 4.2. Typical insertion loss requirement and a possible insertion loss characteristic which satisfies it. 66 4.3. Two types of insertion loss requirements and possible filter characteristics . 71 vii CHAPTER I INTRODUCTION 1.1 The Use of Frequency Selective Networks and Terminology While transmitting information in the form of electrical signals from one point to another, we are faced with the problem of recovering a desired signal from undesired ones and from elec trical noise. When the energies of different signals are in different frequency regions, or bands (a situation which is often found in nature, or can be attained by various techniques in the process of converting the information into an electrical signal at the trans mitting points), the separation of one signal from others makes use of the differences in energyfrequency spectra of different signals. This separation is accomplished by placing some device in the path of the signal (before it reaches the receiving point) which exhibits frequency selective characteristics, i.e., a device which is capable of accentuating (or letting pass through without change) one frequency band, the band containing all or most of the energy of the desired signal, while suppressing (or attenuating) another band. The device usually employed for this purpose consists of an array of electrical components (resistance, capacitance and 1   2 inductance) interconnected in a fashion which produces the desired frequency selectivity. There are some mechanical devices with this characteristic which are sometimes used. A word on terminology: An array of interconnected electri cal components is called an "electrical network," or in short a "network." The manner of interconnection of components, the choice of their type and number required in order to produce a specified electrical characteristic (e.g., frequency selectivity), as well as the mathematical tools employed in network problems, are treated in network theory. If a network consists of electrical components, the values of which are independent of frequency, it is a "linear network"; if the circuit element values are positive constants, it is a "linear passive network." A network inserted in a communication path, which usually consists of two wires, must have two input terminals and two output terminals, hence falls in the category of "fourterminal networks," or, more precisely, "two terminalpair networks." A fourterminal network which has frequency discriminating properties is called an "electric wave filter," or in short, a "filter." Depending on the region of frequencies a filter favors or attenuates, there are "lowpass filters" (i.e., these filters pass low frequencies while attenuating high frequencies), "highpass filters," "bandpass filters," "bandelimination filters" and "multiple passband filters" where a filter has several passbands separated by attenuation bands.  3  The external characteristics of a linear passive four terminal network may be completely described by a set of three independent parameters, which are in general functions of frequency. Of the possible sets of parameters, there are some which can be readily obtained by measurement at the terminals without knowledge of the internal structure of the network; these are the openand shortcircuit impedances. A very convenient set of parameters, the "image parameters," may be obtained as functions of the open and shortcircuit impedances. The image parameters consist of two image impedances and an image transfer coefficient. The external electrical characteristics of a filter in serted in a communication path between transmitting point (gen erator) and receiving point (load) are the operating characteristics called "insertion characteristics"; these consist of "insertion loss" and "insertion phaseshift" characteristics. The first is a measure of the decrease in amplitude of a sinusoidal voltage passing through the filter, the second gives the amount of phase shift (or delay) for a sinusoidal voltage passing from input to output (i.e., from generator to load) terminals of the filter. Both are, in general, functions of frequency. 1.2 The Image Parameter Method of Network Synthesis The problem of designing a filter is to synthesize a net work having a prescribed insertion loss characteristic. There are two distinctly different methods of filter 4 design available. The older one, developed by Zobell* from earlier work of Pupin, Campbell and Foster2, is based on image parameter theory; the resulting image parameter filter is called a Zobel filter. The newer method, originated by Darlington,3 is based on insertion loss theory developed independently by Darlington, Cauer and Piloty at the same time. The outstanding feature of the imageparameter design method is its simplicity. This, coupled with the fairly well performing filters obtainable by the image parameter design tech niques, caused this method to be used almost exclusively in prac tical filter design. The simplicity of this method stems from the "building block" structure of composite filters. Each "building block," or "section," is a fourterminal network which can be fully described by two image impedances Z11 and Z12 and an image transfer coeffi cient 91 "= a + Jip, where oe is the image attenuation coeffi cient and /31 is the image phase shift coefficient. A composite filter is formed by connecting sections in tandem, with image impedances that are equal to each other (matched) at the terminals of the sections which are joined. The total transfer coefficient of the composite filter is simply the sum of the transfer coefficients of individual sections. If these filters could be terminated at both ends by the respective image impedances (i.e., matched at the ends) the image *Raised numbers refer to entries in the References at the end of this dissertation. 5 transfer coefficient would describe their actual (measurable) per formance. However, since the image impedances are generally func tions of frequency, and since the terminating impedances are usually constant resistances, the filters cannot, in general, be matched at the termination. This mismatch causes the quantities of interest, the insertion loss and insertion phase, to be somewhat different from the image attenuation and image phase shift. Moreover, the requirement of matched image impedances at the points of interconnection of individual sections imposes restric tions on the networks obtainable by the image parameter method, as well as on their performance characteristics (i.e., their insertion loss and insertion phase characteristic). These restrictions lead to an inefficient use of network components. 1.3 The Insertion Loss Method of Network Synthesis The insertion loss method of filter design is characterized by the fact that filters obtainable by it have an insertion loss characteristic exactly as prescribed, when inserted between prescribed terminations (hence the name "insertion loss theory"). It is based on the treatment of the entire problem of finding a fourterminal network satisfying given insertion loss requirements. The insertion parameters, rather than the image parameters of Zobel filters, are the primary design parameters from which the circuit element values of the filter are obtained in the last step of a sequence of (often lengthy) operations. The resulting filter, no matter how compli cated, is obtained as an entity, rather than an aggregate of  6 individual building blocks (sections) as in Zobel filters, hence it does not suffer from the matching restrictions imposed on Zobel sections and permits the realization of optimum (in the sense of the number of circuit elements required to meet prescribed charac teristics, and the element values) networks. Another very desirable feature of the insertion loss method is the correction for incidental dissipation it permits. Though the range of compensation is limited by the type of insertion loss function (more specifically, by the distance of the nearest insertionloss zero to the j]axis in complex plane), it permits taking into account losses in components for a great number of cases encountered in practice. 1.4 Comparison of the Image Parameter and Insertion Loss Methods A comparison of the image parameter method and the inser tion loss method of network synthesis reveals that the latter is much more flexible in the results it produces, and much more sophis ticated in the overall approach and in the mathematical tools it employs. As a result of the last, it is taught in most engineering schools only in graduate curricula, and the majority of practical design engineers have no opportunity to acquire a working knowledge of the insertion loss method while in school. An important economical disadvantage of the insertion loss method of network synthesis, as compared with the image parameter method, is the much longer design time required by the insertion  7  loss techniques, resulting in a more costly filter design than in the case of image parameter filters.* On the other hand if the filter designed by the insertion loss method is more economical** than a corresponding Zobel filter (which is often the case), the increased cost of design may be offset when it is to be manufac tured in large volume. The situation in the practical filterdesign field may be summarized as follows: Because of the simplicity of the design procedures and of the acceptable filters the image parameter method yields, and because of the more analytical background and longer design time required of the designers using the insertion loss method, the image parameter method is still used by the majority of design engineers, though using the insertion loss method may often result in a better filter. There have been many contributions in the past to both design methods;*** a great number of these aimed at bridging the gap exist ing between modern theory and practical design methods. The approach from the imageparameter field is, in general, toward the improvement of Zobel filters by better matching, e.g., Ref. 6, by reduction of components in derived terminations, e.g., Ref. 7, or by finding new structures, e.g., Ref. 8. In the insertion loss field the emphasis *The design time may be reduced by employing modern means of computation, e.g., an electronic computer. This, while reducing the required design time, will not, in general, reduce the overall design cost. **An economical filter is one that requires fewer or less expensive components, or its component values may have higher tolerances without upsetting the performance, etc. ***For a summary of recent contributions and a good list of references, see Ref. 5. 8 9 is on a more comprehensive presentation of known material, on simplification of design techniques,10 and on tabulation of designs.11, 12 This dissertation is concerned with a synthesis method for ladder networks which employs "building blocks" similar to the image parameter filter; in fact, the "elementary building block" used is an image parameter halfsection. To form a compo site filter these "building blocks" are connected in tandem in a fashion similar to a Zobel filter, without, however, the funda mental restriction of Zobel filters, that of matching image impedances at the points of interconnection. 1.5. Statement of Objectives The objectives of this dissertation are twofold: 1. The introduction and investigation of a new method of synthesis of ladder networks by which filters can be designed with a simplicity similar to the simplicity of the image parameter method of design, yet yielding more general filters than the Zobel filter of the image parameter method (i.e., yielding filters the choice of element values and the perform ance characteristics of which do not have the limitations of Zobel filters). 2. The derivation of relations and determination of parameter values for use in this synthesis method, by which 9  design engineers familiar only with the image parameter method will be able to design filters with desirable inser tion loss characteristics (Butterworth, Tchebycheff and ellipticfunction type) until now obtainable primarily by the insertion loss method. The contents of this dissertation is divided into three chapters. Chapter II deals with methods of interconnection of the elementary building blocks. It is shown that any lossless ladder can be expressed in terms of image impedance halfsections with one common image parameter. The cutoff frequency is chosen as the common parameter. In Chapter III some methods of finding the image param eters of the elementary building blocks are investigated. The image parameters are then determined for twosection ladder filters having Butterworth, Tchebycheff, and ellipticfunction insertion loss characteristics. Based on results of Chapter III, a modification of Zobel filters is introduced in Chapter IV. This modification permits control of the distortion in passband of Zobel filters; it also makes the economy of these filters (as far as the performance per number of elements is concerned) comparable to the economy of optimum filters. CHAPTER II THE GENERAL LADDER NETWORK IN TERMS OF ELEMENTARY BUILDING BLOCKS CONNECTED IN TANDEM 2.1. The Elementary Building Block Only the lowpass ladder will be considered here, since the highpass, bandpass and bandelimination networks can be ob tained from the lowpass ladder by well known frequency trans formations.13, 14 The elementary building block (Fig. 2.1) is an image parameter mderived lowpass halfsection, the circuit element values of which are determined by the image parameters m, r and c' using the simple relations from image parameter theory:15 cl = r (2.1) c = mr (2.2) 1 m2 c2 ' (2.3) mr c mc = 2 _ m = 1 0 ) O m 1 (2.4) where wc is the cutoff frequency of the halfsection, r is the image impedance at zero frequency, wo is the frequency of infinite attenuation. We also have (Fig. 2.2 and Fig. 2.3)  10   11  h2 S0l T 0 The elementary building block real S\ imaginary (negative) J .. I Image impedance of the elementary building block at terminals 10 of Fig. 2.1. Imaginary real (positive) (r F / i\ i / I \ S/ I \ \ / \ \ negative) Wc l2 c Image impedance of the elementary building block at terminals 20 of Fig. 2.1. Zll Fig. 2.1. Zi r . Fig. 2.2. Fig. 2.3. ZI2  12  r S2] '2(. 1 ( m2) Analysis of equations (2.1), (2.2), (2.3) and (2.4) reveals that the parameters m, r and w> uniquely determine the component c values c1, c2 and h (with the restriction that m is a real and posi tive constant, 0 t m i 1). On the other hand, given cl, c2 and h, the design parameters m, r and wc may be found from the following relations: 2 h r2 = 2 (2.7) Cl m2 = + c2 (2.8) 1W2 1 (2.9) h2(cl + c2) Equations (2.7), (2.8) and (2.9) are obtained from equations (2.1), (2.2) and (2.3) solved for r, m and oo Restricting the design C parameters to be positive, it is seen from the last set of equations that cl, c2 and h2 uniquely determine the design parameters r, m and W C. 2.2. Methods of Connecting the Elementary Building Blocks Fig. 2.4 shows one method of connecting the halfsections  13  4 r4 4 0 r4 V14 V.4 .M0 0 0 co WA V40 14 4) GJ4 0 '30 u00 0 0 u c 0 .0 "4 "0 4 A 41 V14 00 ,4' 00 V0 C00  14  into a ladder network; here we connect the evennumbered terminals with the oddnumbered ones. We note that the evennumbered terminals have an image impedance given by equation (2.5) (i.e., the image impedance between the evennumbered terminals and 0, the common, for one halfsection only, is given by (2.5)), and the odd numbered terminals have an image impedance given by equation (2.6), hence we connect the terminal pairs having different types of image impedances. The image impedance at both terminations of the resulting ladder will not be of the same type; the image impedance at terminals 10 is of the type given by equation (2.5), the constantk midshunt type, whereas, the image impedance at the other termination (at terminal pair 2n0) is of the type given by equation (2.6), the mderived midseries type. Therefore the resulting ladder cannot be symmetrical.* Another method of connection is shown in Fig. 2.5, where we connect evennumbered terminals with evennumbered terminals and oddnumbered terminals with oddnumbered ones, i.e., we connect pairs of terminals having image impedances of the same type. If the ladder contains an even number of halfsections, both image impedances at the terminations will be of the same type, hence the ladder can be symmetrical. If the number of halfsections is odd, the image impedances at the terminations are similar to those of Fig. 2.4. The method of connection shown in Fig. 2.5 results then in *A 4terminal network possesses electrical symmetry if the image impedances at both terminalpairs are equal, i.e., one terminal pair cannot be distinguished from the other pair by external measurements.  15  a more general ladder than the connection of Fig. 2.4 (in fact, Fig. 2.4 can be considered as a special case of Fig. 2.5 in which the components of the second, fourth, etc., section are zero, i.e., h4 = h8 h12 ... h2n 0 and c3 c7 '* = c2n1 = 0) hence this method of connection will be used in subsequent analysis. Since every component of this ladder can assume any value depending on the choice of the image parameters rl, m1, cl' r2, m2' c2' ... rn m Wcn', it is clear that the reverse must hold also, i.e., any ladder network having the configuration of Fig. 2.5 can be syn thesized with image parameter halfsections by proper choice of the image parameters ri, mi and wci 2.3. A Conventional Low Pass Ladder in Terms of Elementary Building Blocks A particular case of the ladder of Fig. 2.5, one in which a reduction in the number of components can be realized, would result if two adjacent parallel resonant circuits in the series branches of the ladder could be combined into one parallel resonant circuit. Obviously this can be accomplished only if the resonant frequencies of the adjacent circuits are equal. The resulting ladders are shown in Fig. 2.6 (a) and (b), where the components of Fig. 2.5 combine as follows: ql = Cl (2.10) q2 h2 + h4 (2.11) a2 = c2c4 (2.12) c2 + c4  16  N 0 ,M 0 4 I c 10 4 o . 0r 0 41I c U 04 eq0 cr V40  17  a2q2 1 h2c2 h4c4 (2.13) 2.., q3 3 c3 + c5 (2.14) q4 h6 + h8 (2.15) a4 c6 c8 (2.16) c6 + c8 a4q 2 = h6c6 = h8c8 (2.17) In Fig. 2.6 (a) the ladder consists of an odd number of half sections (the number of halfsections n odd), hence (see Fig. 2.5) qn+l h2n2 (2.18) an+1 c2n2 (2.19) The ladder of Fig. 2.6 (b) has an even number of half sections (neven), hence here (see Fig. 2.5) n = h2n + h2n2 (2.20) a C2n2 c2n (2.21) n c2n2 + c2n an 2 = h2nc2n = h2n2c2n2 (2.22) n _ qn+1 c2n (2.23) The number of image parameters for the ladders of Fig. 2.6  18  is 3n (three parameters, ri, m, and Wc for each of the n half sections). However, due to the conditions of equations (2.13), (2.17) "' (2.22), some of the parameters will be related by these equations, resulting in the number of independent image parameters being equal to the number of components in the ladder. Consider the problem of expressing a given ladder of the form of Fig. 2.6 in terms of image parameter halfsections, i.e., given a ladder of Fig. 2.6, determine the element values of an equivalent ladder of Fig. 2.5. One question arises immediately: Is there a unique equivalent of the ladder of Fig. 2.6 in the form of one of Fig. 2.5? If not, could there be one specific form of Fig. 2.5 which is advantageous? The first question can be answered by noting that there are 3n components in ladder of Fig. 2.5, the values of which are to be determined from 2n + 1 relations.* Thus there is no unique equiva lent of a ladder of Fig. 2.6 in the form of one in Fig. 2.5 The second question can be answered after investigating the possible equivalent ladders of the form of Fig. 2.5. *For a ladder of Fig. 2.6 (a) we have three equations of the form of equation (2.11), (2.12) and (2.13) for each series branch except the last, i.e., 3(n 1)/2 equations; one equation for each shunt branch, i.e., (n + 1)/2 equations ((n 1)/2 of the form of (2.14) and one equation (2.10)), and two relations (2.18) and (2.19) for the last series branch, giving a total of 3(n 1)/2 + (n + 1)/2 + 2 2n + 1. For Fig. 2.6 (b) we have 3n/2 equations of the form of (2.11), (2.12) and (2.13) for the series of the form of (2.11), (2.12) and (2.13) for the series branches, and (n + 2)/2 for the shunt branches, giving a total of 3n/2 + (n + 2)/2 2n + 1.  19  In order to obtain a particular ladder equivalent, 3n (2n + 1) n I component values of the equivalent ladder of Fig. 2.5 will have to be either chosen or specified by addi tional nl equations* (i.e., we have nl degrees of freedom). It is clear that the nI components, the values of which are arbi trary, will be located in the ladder in accordance with the location of the components in Fig. 2.6 which on division produce the nl degrees of freedom (e.g., one cannot choose arbitrarily the values of the first ni components, starting with cl, c2, h2, c4, h ... up to the (n1) th component in Fig. 2.5). Examination of the ladders in Fig. 2.6 shows that every branch of the ladder which is divided produces one degree of freedom, as shown by the symbol "I" in Fig. 2.7 (a). The arrows indicate where the degrees of freedom originate, hence they also indicate which of the elements may have arbitrary values. Fig. 2.7 (b) and (c) show two possible choices of the location of arbitrary elements: In Fig. 2.7 (b) the first halfsection has no arbitrary elements, the second has one in the series branch and one in the shunt branch, the third halfsection has none, etc. In *The chosen values of the components, though arbitrary in a certain range, should not result in a negative element lest some of the image parameters from which the halfsections can be computed become imaginary. If instead of a component value a corresponding image parameter is chosen, a similar limitation applies in addition to the requirement 0 mi I l(or w < 4). c  20  4 v] I I II 4 SI SII 4 4 H KHtI 4 4 I,.' C) 44 0 0 0 4 0 41 0 #4 i4 (U4 ,4 /"  21  Fig. 2.7 (c) every halfsection, except the last one, has one arbi trary element. There are a number of combinations of the arrangements in Fig. 2.7 (b) and (c) possible.* The arrangement of arbitrary elements in Fig. 2.7 (c) appears advantageous in that it permits choice of the circuit element value of one component (or one parameter) in each halfsection, except the last. Hence the resulting ladder, with the exception of one terminat ing halfsection, can be made to have one arbitrary parameter.** The values of the nl arbitrary elements may conveniently be specified by additional nl arbitrary equations, in which the limitations on the chosen values (see footnote* on previous page) could be included. A simple set of n1 of such equations could be Pl m P =2 P3 = P =n (2.24) where p denotes an image parameter.*** In particular if the same *Similar reasoning applied to the ladder of Fig. 2.6 (b) shows that the conclusions reached in analysis of Fig. 2.6 (a) hold here as well; we may have one arbitrary element in each half section, except in one of the terminating halfsections. **Since the component values of a halfsection can be ex pressed in terms of image parameters r, m and Wc, equation (2.1), (2.2) and (2.3), the fact that one of the component values can be chosen arbitrarily means that one of the image parameters can be chosen arbitrarily. ***We note that from equation (2.24) we have n1 relations of the form p m p p P P3,'* Pp, pn, hence the value of p cannot be chosen but is de ermined from 3n simultaneous equations, 2n + 1 of which are in the form of equation (2.10) to (2.23) with h's and c's substituted by equation (2.1) to (2.4).  22  image parameter is chosen for all p's, the resulting ladder will be very similar to an image parameter Zobel filter1' 15, 16 when p denotes either r or )c&.* A convenient choice for p, equation (2.24), is to let it denote the cutoff frequency a) c, i.e., letting the cutoff frequen cies of all n halfsections be equal. This would make the image parameters m equal for the halfsections with common parallel resonant frequencies aw, (see equation (2.4)). The above leads to the following conclusion: Any conventional ladder filter, Fig. 2.6, can be synthesized by tandem connected (method of Fig. 2.5) image parameter half sections, pairs of which have equal mvalues and all of which have equal cutoff frequencies co . This is equivalent to stating that: Any conventional ladder filter can be obtained from an image impedance Zobel filter by changing the impedance levels rl, r2, rn of its halfsections. In Chapter III some methods will be investigated, by which the required impedance levels ri, the cutoff frequency ojc and the mvalues could be determined so as to obtain a ladder filter having a prescribed insertion loss. *A conventional image parameter filter is called a Zobel filter. It is a filter consisting of connectedintandem image parameter sections or half sections (prototype or mderived) with common cutoff frequencies w and with image impedances (see equations (2.5) and (2.6)) wh ch match each other at the points of interconnection. In terms of our elementary building blocks, a Zobel filter will have jcl= Wc2 ... cn and ri = r2 ... = rn.  23  2.4. Filters with Incidental Dissipation in Terms of Elementary Building Blocks One of the major disadvantages of the image parameter Zobel filter design technique is the fact that dissipation cannot be taken in account in design (though the effect of dissipation may be found by analysis). The modern network theory offers a technique3 which allows one to synthesize a filter with dissipation producing the same form of insertion loss curve as one without dissipation, except for an additional constant (independent of frequency) loss. This tech nique, called predistortion, is based on the effect of dissipation on the location of poles and zeros of the insertion voltage ratio in the complex frequency plane. Given the pole and zero distribution of a desired insertion voltage ratio of a dissipationless ladder filter, e.g., Fig. 2.8 (poles and zeros on solid curve), the addition of positive resistance to each component (in series with an inductance and in parallel with a capacitance) in such amounts that would make the Qfactors of all components equal, would move the zeros and poles in Fig. 2.8 hori zontally a distance d 1/Q to the left, as indicated by the poles and zeros located on the dashed curve. Adding negative resistance to the dissipationless components in the same amounts as above would shift the poles and zeros a horizontal distance d to the right, as indicated by the zeros and poles on the dotted curve* (the dotted *This horizontal shift is equivalent to transforming the complex variable s 6' + jco to s + d, where the (+) and () signs correspond to addition of positive or negative resistance, respec tively.  24  Effect of added positive resistance I'. Lossless L's and C's Effect of added negative resistance X X Effect of adding proportionately equal amounts of resistance on the location of poles and zeros of the insertion voltageratio function Fig. 2.8.  25  curve is called the predistorted curve for an amount of dissipation equal to d). Suppose that the pole and zero distribution on the dotted curve is used to synthesize a dissipationless ladder network.* If now dissipation d is added to each element, the poles and zeros would move to the left a horizontal distance d and the distribution on the solid curve (Fig. 2.8) would be obtained, resulting in the desired insertion loss. The lossless ladder obtained from the predistorted curve (dotted curve) can be represented by elementary building blocks (image parameter halfsections) as outlined in section 2.3. The image parameters r, m and c c of these building blocks will in gen eral not be equal to the image parameters of the building blocks corresponding to a ladder synthesized from the pole and zero loca tions on the solid curve (the pole and zero location of the desired insertion voltage ratio). Thus, given a lossless ladder filter in terms of image parameters r, m and ULc, the effect of predistortion for a certain amount of dissipation can be obtained by a corresponding (to the amount of dissipation) change in the image parameters r, m and W c of the elementary building blocks. *In practice, predistortion is applied to the zeros only, since the application to the poles would result in a network func tion which is not realizable by a lossless passive ladder without mutual (inductive) coupling. Thus the transformation of s to sd is applied only to the numerator of the insertion voltage ratio function, and compensation of the effects of dissipation for the passband only is achieved. Note also that there is an upper limit of d, which must be smaller than the horizontal distance of the nearest zero to the jw axis, for physical realizability with passive networks. CHAPTER III DERIVATION OF THE IMAGE PARAMETERS FOR LADDER FILTERS HAVING PRESCRIBED INSERTION LOSS CHARACTERISTICS In this chapter the image parameters will be determined for the elementary building blocks which, when connected in tandem, result in ladder filters with Butterworth (maximally flat) Tchebycheff (equal ripple in passband only), and elliptic function (equal ripple in passband and in stopband) insertion loss characteristic. The image parameters will be expressed in terms of the critical frequencies (poles and zeros) of the insertion loss function. The investigation will be limited to twosection symmetrical filters. 3.1. Outline of Approach The insertion loss L (in db) as a function of normalized frequency oJ for the above three types of response is given by17 L = 10 log (1 + E2) (3.1)* where E denotes a polynomial or a ratio of two polynomials in W , the form and the coefficients of which determine the type of response. The zeros and poles of E(w) coincide with the zeros and poles of the insertion loss function L(w), as can be seen from equation (3.1). It shall then be attempted to express the *Equation (3.1) is for symmetrical filters only; it will be assumed for convenience that the filter is working between equal resistances of 1 ohm. 26   27  image parameters of the elementary building blocks in terms of these poles and zeros. Method I: One way of accomplishing this could be to find the inser tion loss of a ladder consisting of elementary building blocks in terms of the image parameters of its halfsections. Comparison of this insertion loss with the insertion loss of Butterworth, Tcheby cheff or ellipticfunction type would yield a number of equations which, when solved simultaneously, would give the image parameters m, r and tc in terms of the poles and zeros of the insertion loss function. Method II: Another method of getting the same results for specific numerical cases would be to synthesize the ladder from the given insertion loss function by any suitable method (e.g., Darlington's method) and then express the ladder in terms of image parameter halfsections, obtaining a number of equations of the form of equations (2.10) to (2.23);* substituting for h's and c's equa tions (2.1) to (2.4), a set of equations in mi, ri and Wce are obtained, which could be solved (at least theoretically) simul taneously for the image parameters. The first approach is more general and will be used here. *It is to be noted here that qi and ai are known.  28  Some numerical data will be obtained using the second approach. 3.2. Insertion Loss of a TwoSection Ladder Filter in Terms of Image Parameters The two section symmetrical ladder which will be used in subsequent investigation is shown in Fig. 3.1.* This ladder is ob tained by tandem connection of image parameter halfsections as shown in Fig. 3.2. Comparison of the ladders of Fig. 3.1 and 3.2 yields the following relations: q1 l2 1(3.2) rl wc m12(rl + r2) (3.3) q2 oWc a2 1 m122 (3.4) r12 Wc(r1 + r2) 4 3(r + r4) (3.6) m34 c(r3 + r4) q5 = m34 (3.8) *The Butterworth and Tchebycheff filter can be considered as a special case of the ladder of Fig. 3.1 in which a2 w a4 = 0. 0V "4 0 fa4 N U 43 III 4 4 r  r U .r HF U 413 C4 3 34 "4 6  II ^3 u 3, N U " N 0) NN I3 a   29  0 M V4 "4 .9I II 44 0 *s .44 '3 0 *6 00 *I .94 en .4 0 V4 0 *0 00 '0 NN "4 4 Ak 0 U 0 *s *g .4i 60 0 A NO C Ni~ 00i .9.' fa  30  where rl, r r and r are the impedance levels of the halfsections 1 2 3 4 1, 2, 3 and 4, respectively; m12 is the common myalue of half section 1 and 2; m34 is the common mvalue of halfsection 3 and 4; Wc is the cutoff frequency common to all halfsections; ql, q , q5, a2 and a have dimensions of capacitance; and q2 and q have dimensions of inductance. It is much more convenient to find the insertion loss of a lattice equivalent to the ladder of Fig. 3.1, rather than that of the ladder directly. A lattice equivalent of the ladder of Fig. 3.1 is shown in Fig. 3.3 (a). The series reactance X and the shunt reactance Xb of the lattice of Fig. 3.3 (a) are given by Xa Za WA (3.9) a J Al A 2o2 Zb 1 A342 Xb j" T (A4 W2 A5) (3.10) A (3.11) 1I L1 A2 C1 (3.12) A3 L2C2 (3.13) A 223 (3.14) A4 L2C2C3 (3.14) A5 = C2 + C3 (3.15)  31  (b) Equivalent lattice of the ladder of Fig. 3.1. q2 4 q2 / "1 A Butterworth and Tchebycheff type ladder filter and its equivalent lattice Fig. 3.3. Fig. 3.4.  32  The relation between the lattice and the ladder elements is as follows: A q q5 (3.16) q4a4  q2a2 A2 q + q2a2A1 (3.17) A3 + a2 + q4a4 (3.18) 1 A4 A2A3 q2a2q4a4A1 (3.19) 1 AS i (ql + q3 + q5) (3.20) The insertion loss of the equivalent lattice of Fig. 3.3 is given by equation (3.1), where Bl(w w 2B2 + B3) E W 2 1 (3.21) U}4B U2B5 + 1 B 2 A2A (3.22) AIA5 1 B3 AA (3.24) A2A4 I 1 4 A l (A2A3 A4) (3.25) B5 L (A1A3 + A2 A5) (3.26) The zeros and poles of the insertion loss are given by  33  the zeros and poles of equation (3.21), i.e. )(W 4 2B2 + B3) 4 W 2(z1 2 + z 2) + z 1222] 4 2 W4B4 2 B + 1 + 1 (3.27) 4 5 P12P2 P12 + p22 where z1, z2 and pl, P2 are the frequencies at which the insertion loss is zero and infinite, respectively. It follows from equation (3.27) B2 = z12 + z22 (3.28) B3 z 2z22 (3.29) B" p4lp22 (3.31) Equations (3.27) can be solved for the zeros and poles: zo 0 (3.32) 1 2 B2 B22 21 z1 2 = 2 + [ B (3.33) Po (3.34) B5 ( B2 4B4) (335) pl,2 = 2B4 (3.35 3.3. Tchebycheff and Butterworth Filters Synthesized with Image Parameter HalfSections The Butterworth and Tchebychefftype twosection filter is shown in Fig. 3.4(a) and its insertion loss characteristic is given by equation (3.1) with17  34  E = H6J5 (3.36) for the Butterworth filter, and E HT5(o) (3.37) for the Tchebycheff filter. Here T5 (w) is the Tchebycheff poly nomial of order 5, which can be defined by T5(W) cos (5cos1 ) (3.38)* 0 5 cdJ 1 The insertion loss zeros of the Butterworth filter, as seen from equation (3.36) are at C 0; the Tchebycheff polynomial is zero when 5cos 7T(i + ) i = 0, 1, *** 4 resulting in zero insertion loss at the following frequencies: zo 0 Z 2 cos2 7T (3.39) 2 cos2 3. z2 10 *Tn(x) can be defined in polynomial form as Tn(z) = 2nl Exn  nxn2/1!22 + n(n 3)xn4/2!24 n(n4)(n5)xn'6/3!26 + ** where the summation is stopped when the exponents of x are negative. The form of (3.38) is convenient in that it exhibits the location of the zeros.  35  Equation (3.37) can be written, using equations (3.39), (c22 C T )(W2 C082 37) E 10 10 (3.40) 1 2 Co 1 Cos 10) For both, Butterworth and Tchebycheff, filter types E = H when 1, giving (see Fig. 3.5 (a) and (b) ) H2 100.11P 1 (3.41) where L is the maximum loss in passband in db. P It can be shown that Tchebycheff and Butterworth two section filters possessing electrical symmetry have also geome trical symmetry, i.e. (Fig. 3.4 (a) ) q q5 q2 q4 (3.42) hence the equivalent lattice to the ladder of Fig. 3.4 (a) can be easily found using Bartlett's bisection theorem.18 The equivalent lattice is shown in Fig. 3.4 (b). Comparison of Fig. 3.3 (a) with Fig. 3.4 (b) yields the following relations: L1 = L2 q2 Cl C3 z q, (3.43) C2 (t Inserting equations (3.43) into equations (3.11) to (3.15) gives  36  L(db) t L(db)  60 z1 z2 Insertion loss characteristics for (a) Butterworth and (b) Tchebycheff twosection filters Fig. 3.5.  37  1 A3 2 A qlq2q3 4 2 A5 q + (3.44) Equations (3.44) inserted in equations (3.22) to (3.26) give B1 q12q22q3 2 2q + 2q3 q2q3 B2 7 " qlq2q3 2ql 2q2 + q3 B3 12q22q3 B4 B5 0 (3.45) Equations (3.2)to (3.8), with equations (3.42) and with a2 a = 0 result in 12 = m34 r1 r and r2 r3, giving 1 ql ", c r c r + r2 c2 + (3.46) 2 c 3 r2 c  38  Substitution of equations (3.46) in (3.45) gives (r1 + r2f rl12r2 wc5 B c 2r(1 r12 + r ) Wc4rl(1 r1r2) S= c *1(3.47) 3 r + r Finally, using equations (3.29) and (3.30) in equations (3.47) we get z12 + z 2 W 2(1 r12 + r 2 c rl+ r2 2 2 c4r(1l rlr2) z z2 M + r 2(3.50) The Butterworth Filter The insertion loss zeros of the Butterworth filter are at zero frequency (i.e., at Co 0). Equations (3.50) then become (with zI = z2 = 0) 0 W c2 1 r 2 + rl c 1 ri + r2 0 c 4r(1 rr2)51) r + r Sol.4ng equations (3.51) simultaneously we obtain r 1 rI 0.5 + (1.25)21 1.27202 S 0.78615 (3.52) r2 r  39  Comparison of equation (3.21) with equation (3.36) shows that B1 H, hence from equations (3.41) and (3.47) we get O.Lp (rI + r2) 10 l 1 =( +2 (3.53) 4 2 10 r1 r2 W*c From equations (3.51) we get r + r = r 3, which inserted 1 2 1' in (3.53) with r1r2 1 from (3.52), gives rl 1.27202 cc H0.2 (100.11P 1)0.1 (3.54) The cutoff frequency w0 is tabulated for values of L , the maximum attenuation in passband, from 0.1 db to 5.0 db in Table 1, and is plotted versus L in Fig. 3.6. A two section Butterworth filter in terms of image parameter halfsections, with L 3 db is shown in Fig. 3.7.* The Tchebycheff Filter With the insertion loss zeros of the Tchebycheff filter given by equations (3.39), equations (3.50) become 2 7T 2 3TF 2 1 2 r cos f + cos 10 c 1 + r 2 7T os2 37T c4rl(1 rlr2) 10 10 r + r2 We also have from equations (3.21) and (3.40) *For Lp = 3 db, Jc 1.272 since H = 1 in equation (3.54). 40  TABLE 1 CUTOFF FREQUENCY wc FOR THE BUTTERWORTH TWOSECTION FILTER L in db. Lp wc Lp wc Lp Lc 0.1 1.8526 1.2 1.4263 3.2 1.2612 0.2 1.7265 1.4 1.4011 3.4 1.2503 0.3 1.6560 1.6 1.3792 3.6 1.2399 0.4 1.6071 1.8 1.3597 3.8 1.2300 0.5 1.5698 2.0 1.3421 4.0 1.2205 0.6 1.5396 2.2 1.3261 4.2 1.2113 0.7 1.5143 2.4 1.3113 4.4 1.2025 0.8 1.4925 2.6 1.2976 4.6 1.1939 0.9 1.4732 2.8 1.2847 4.8 1.1857 1.0 1.4560 3.0 1.2726 5.0 1.1776  41  0.2 0.3 0.5 1.0 2.0 3.0 5.0 Maximum attenuation in passband L in db Cutoff frequency of elementary building blocks versus maximum attenuation in passband for the design of filters with Butterworth characteristic 1.6 1.4 Fig. 3.6.  42  M00 * I .4 41 o c 1 V 1CI *'W '4 000 0 " *44 .C.O o ** Sh  43  H Bl' sin2 sin2 :vr 10 10 with which (3.47) becomes H (rl + r2)2 (3.56) sin2 3 1 r 2r2 Instead of attempting to solve directly equations (3.55) and (3.56) for wcj, rl and r it is more convenient to use equations (3.46) with the following equations derived by Belevitch19 7T S 2sin 10 g 3TT 2gsin 10 q2 g2 + cos2 3 2 g2 + cos2 ) q3 = (3.57) g (g2 + cos2 jT where g is related to H by the transformation 1 sinh 5x H g sinh x sinh 5 arc sinh f ) (3.58) and H is given in terms of L the maximum insertion loss in pass band (which is here synonymous with the peak value of ripple in passband), by equation (3.41). Equations (3.46) and (3.57) are readily solved for uC , r1 and r2, giving  44  k) 2 A + B c 4 sin Zt sin 3T 10 10 2 35 2 g sin 3__ r 10 1 (A + B) sin 1 r 0 r2 1A B A 2(g2 + cos2 7L sin j ) 10 10 B = g2 + cos 13 (3.59) A tabulation of W rI and r for values of L from c 1 2 p 0.1 db to 5 db appears in Table 2.* A plot of COc, rl and r2 from Table 2 versus L is shown in Fig. 3.8, and a Tchebycheff filter with L = 1.5 db in terms of image parameter halfsections p is shown in Fig. 3.9. 3.4. The Elliptic Function Filter Synthesized with Image Parameter HalfSections The elliptic function type (often called equal ripple in passband and stopband type) twosection filter which will be used *The data in Table 2 are computed on the University of Florida IBM 650 computer from the equations (3.41), (3.58) and (3.59).  45  TABLE 2 IMAGE PARAMETER VALUES FOR THE TCHEBYCHEFF TWOSECTION FILTER Lp in db; r's in ohms Lp c r r2 r2/rl 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 1.1724 1.1176 1.0891 1.0704 1.0568 1.0464 1.0379 1.0309 1.0250 1.0199 1.0114 1.0047 0.9926 0.9947 0.9908 0.9874 0.9844 0.9819 0.9796 0.7428 0.6680 0.6197 0.5837 0.5547 0.5304 0.5094 0.4909 0.4743 0.4593 0.4329 0.4102 0.3903 0.3725 0.3565 0.3419 0.3285 0.3161 0.3046 0.8638 0.8262 0.7952 0.7685 0.7448 0.7235 0.7040 0.6859 0.6692 0.6535 0.6247 0.5989 0.5753 0.5537 0.5336 0.5150 0.4975 0.4811 0.4656 1.1613 1.2368 1.2831 1.3166 1.3427 1.3640 1.3820 1.3974 1.4109 1.4228 1.4431 1.4599 1.4741 1.4862 1.4968 1.5061 1.5144 1.5218 1.5284  46  TABLE 2  Continued L p rl r2 r2/r 2 0.9775 0.9756 0.9740 0.9724 0.9711 0.9698 0.9686 0.9676 0.9666 0.9656 0.9648 0.2939 0.2838 0.2743 0.2653 0.2569 0.2488 0.2412 0.2339 0.2270 0.2203 0.2140 0.4509 0.4370 0.4238 0.4111 0.3991 0.3876 0.3765 0.3660 0.3558 0.3460 0.3366 1.5344 1.5399 1.5449 1.5495 1.5537 1.5575 1.5611 1.5644 1.5675 1.5704 1.5730 3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4 4.6 4.8 5.0  47  0.9 4 7 Q6 0.7 .. s 3r ", 3 I. I 0 0 S0 .. 03 "^0.3 ..*.... 0.2 01 0.1 0.2 0.5 1.0 2.0 Maximum attenuation in passband L in db P Fig. 3.8. Image parameters Jc r1 and r2 versus maximum ripple in passband for the design of filters with Tchebycheff characteristics . C4 * co * . O F^ ~ if.' o F I .0 48  :1 0 I o 0 V4 ao ** a *o AS . t .0 Ho *4 0 A  a 3 a 4 0) d 0.0 0 S8.0 4 44 A.0 00 94  49  here is shown in Fig. 3.1,* and its insertion loss characteristic, Fig. 3.10, is given by equation (3.1) with3 2 2 2 2 H (I zW )(1 z2 ) " (1 u2z12)(l &2z2) (3.60) The zeros of the insertion loss characteristic are given by 1 2 2K zi k an ( 2 k) 1 z2 k an ( k) (3.61) where an denotes 20, 21 the elliptic sine, defined as sn(u) sin 0 u = F(k, 0) =/ dO (1 k2 sin2 e )2 K denotes the complete elliptic integral of the first kind 7T 7 K K(k) dO o (1 k2 sin2 9)2 and k is the modulus of the elliptic sine. The constant H is given by H 4 (100.1 1)(100.La 1) (3.62) (3.63) (3.64) *This is the special case of elliptic function filters in which all the poles are on the jWaxis, hence it can be realized with a ladder network that does not require mutual coupling.  50  where L is the maximum distortion in passband, and L is the P a minimum attenuation in stopband in decibels (see Fig. 3.10). L , 22 L and k are related by La 0 (i 32 Lp 10 logl (1 + H2 2) 5 A k =7 n2 (. k) sn2 3K, k) (3.65) As can be seen from Fig. 3.10, the modulus k determines the width of the transition band (k is called the selectivity parameter), k = a (3.66) where u)J is the upper frequency limited of the passband (or the effective cutoff frequency), and aj is the lower limit of the a attenuation band. When L is small and L large, which is the case for most p a filter requirements, the various parameters determining the response of an elliptic function type 2section filter are related by the approximate equation La = 10 logl0 (10 IL 1) 50 logl0 q 12.041 (3.67) where q is the elliptic modular function of k,* La and LP are in *The function log q is tabulated in most elliptic function tables, e.g., Ref. 21, pp. 4951.  51  L (db) t z z2 W p I 1a a Fig. 3.10. Insertion loss characteristic of an elliptic function twosection filter  52  decibels. A plot of La versus k for L 0.1 db is given in Fig. 3.11. Figure 3.12 shows the dependence of L on L for a given k. a p In order to find relations between H, z, z2 of equation (3.60) and the image parameters aWc, r to r m12, m34, a procedure similar to one used in the previous section will be followed. The resulting algebraic manipulations are quite lengthy and will be omitted here. The following steps are carried out: 1. Equations (3.16) to (3.20) are inserted in equations (3.22) to (3.26), giving a set of equations, say Set I, relating B's with q's. 2. Equations (3.2) to (3.8) are inserted in Set I, giving a set of equations, say Set II, relating B's with the image parameters. 3. Equate Set II to corresponding equations (3.28) to (3.33), getting a set of equations, say Set III, relating the image parameters to the zeros and poles of the insertion loss function (3.60). 4. Solve equations Set III simultaneously for the image parameters in terms of zeros and poles of the insertion loss function. Step 3 results in five nonlinear equations in seven unknowns (the image parameters); two additional equations are obtained from the symmetry conditions which the ladder must satisfy, giving a total of seven independent equations. These seven equations do not lend themselves readily for use in Step 4,  53  70 60 50  40  30 ... . 20  0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Selectivity factor k Fig. 3.11. 10 0.01 Minimum stopband attenuation La versus selectivity factor k for the twosection elliptic function filter 0.05 0.1 0.5 1.0 5.0 L in db P Fig. 3.12. Variation of minimum attenuation in stopband with maximum ripple in passband J.o 4. O 0 0 4j I 4 . v  54  hence numerical values for the image parameters will be obtained using Method II (Chapter III).* Step 4 is carried out for a set of equations which are close approximations to those obtained in Step 3. Following are the equations obtained in Step 3: 12 = m122 (3.68) wc2 2 1 2 (3.69) WOc (m12 m342)(1 + ml2m34D) H = (3.70) c5 (rlml2 r4(37 z12 + z 1 + rr4(D + m12m34) (3.71) 2 I1 + ma2m34D z2z 2 rr4D (3.72) 1 (3.72) iJ c4 1 + m12m34D D rlml2 r4m34(3.73) r 4m12 rlm34 The equations obtained from symmetry conditions are: r1r3 = r2r4 (3.74) *The numerical values appear in Table 3. They were obtained by analysis of synthesized elliptic function filters in terms of image parameter halfsections. For this purpose the synthesized filters published in Ref. 12 were used.  55  ml2m34  m122 m342 r1r4(rl r2) (r42 rl2)(r1 + r2) Equation (3.75) can be written as r2 (r1m2 r 4m34)(r4m12 + rlm34) (3. rl (r4m12 r1m34)(rlml2 + r4m34) to which a close approximation turns out to be r2 r1ml2 r m34 (3. rl r4ml2 rlm34 with the assumption that* I L r4ml2 + rlm34 (3. r1ml2 + r434 With the approximation of equation (3.78) the following set of approximate relations may be obtained:** H (m12 + m34)(1 + ml2m34) (3. rlc5 2 + z2  uc 2(1 r r2 + l 1 + m12m34 76) 78) 79) 80) (3.81) *This assumption is justified by the range of values of the r's and m's encountered in a great number of filters, as shown by the numerical data in Table 3. For most of the values computed, equation (3.79) turns out to be near 0.99. **In addition to the approximation (3.78), there were also other approximations used which are implied in (3.78) and (3.79), e.g., r1I r4)< (3.75)  56  z12z22 1 r1( C4(1 rlr3) rl + r2ml2m34 From (3.68) and (3.69) we have m12 (1 0 c2zl2) m34 (1 Ljc2z22) Using equations (3.83) in (3.80), we may get (m12 + m34)(1 + m12m34) rl "wtHum 5b c with which r2 may be computed from (3.81) (3.82) (3.83) (3.84) (3.85) (3.86) S2 + z 2 ~L22... Equation (3.82) may be solved for r3, using rl and r2 from (3.85) and (3.86). 1 r3" 1  3 z12z22 1 + r2ml2m34) r1 Wc4 1 c (3.87) Using (3.85) to (3.87) in (3.74), we get r1r3 r r2 (3.88) The cutoff frequency W in equations (3.83) to (3.87) may be computed from the approximation 0.982 + .0 .log1 L 0.982 + 0.01 logo Lp 10 p (3.89)* *This equation was derived from the data in Table 3. r 1 m12m34 2 ri 1 + nl 2m34  57  If more accurate results are desired, the r's may be reevaluated by the following relations, using previously computed values on the righthand side of the equations. (z22 z12) (r + r2m12m34) r = (3.90) H Wc3(r1ml2 r4m34) 1 (z12 2+ 22 1)(1 + r2m12m34 ) r3 2 rl (3.91) r (l + rlml2m34) r2 r r3ml2m34 (3.92) 4 (1 rr) r c 4 2Z 2 z1 z2 With rI, r3 and r4 from above and r2 from (3.74), wcj can be recomputed from we 2 R(z2 z12) (3.93) mlm2 where R is the righthand side of equation (3.75).* A tabulation of rI to r4, m12, m34 and jc appears in Table 3 (see footnote on page 54). A plot of the r's versus L P is shown in Fig. 3.13. *The difference between the recomputed values and the first values of the r's and w c is an indication of the accuracy of the results. For greater accuracy, the image parameters may be computed by an iterative procedure using equations (3.68) to (3.75).  58  comN0%.4'oi^ C t4r. Oa O N4, LM en O% Ln ONN 00000000 co 4 no"^ r 0 ' O4 0 0Ven Ln o 00000000 0 4 co C' 00* P 00000000 . N Cn o T N T in OC* P^ 1 t^ C^ C9 C 00000000 cn 4D fn 0 cc 0 c c0 cooodoodo 00 00 00000 00 CO 0 C r0 00 0 4N0\.I C4 ('00 .to oCh cQ4 o0 ctn 4 tn LM nm%0 r 00000000ooooc ddddddddJ~ *4'~o cii owi0N .0 o < I a% oCNM .o O 0000000 6 6 ft6 6 *4' 0 0000000 G0000000 N # on ooZ% o P oo 0000000  59  00000000 m\ L n C7%4) wo "s w 00000000 M ow 4 oI0 0 rO QO O O NT< 7 > 00000000 *ro 01 C ^ T OOQ'lMNM' 0 0OO40 4C % 14 %0.%0. 0 I 14 (n *00 0 V0 08 00000000 d r0. U il OOMOiOtMNMO M r^ ~ 0 (y 9 ri r 'o 'n'n ' Ir *0000000 2Q 4 iADO<\O'.O*'O 00000000 00000000 r4 '00 r~rl C 1.4 c'4c< 4n 4LriL tr 44*4444* 00000000 ^s reQ000ori in~L'4 00000000. >i~~l  60  0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.05 Fig. 3.13. 0.1 0.2 0.5 1.0 2.0 3.0 Maximum ripple in passband L in db Image parameters rl, r2, r3 and r4 for k = 0.7 versus maximum ripple in passband for the design of elliptic function filters CHAPTER IV ANALYSIS OF RESULTS AND CONCLUSIONS 4.1. Analysis of Obtained Data It can be seen from the data in Table 3 that the r's show very little dependence on the selectivity factor k for quite a large range of values of k (the maximum change in the r's for a constant value of ripple in passband is about 5 per cent for values of k in an interval of about 0.4 to 0.8). If the Tchebycheff filter is considered as a limiting case of an elliptic function filter in which keO,* i.e., all the poles of which were moved to infinity without appreciably changing the location of the zeros, we may use the data in Table 2 for getting the r's for the lower limit of k 0. It is seen that these rvalues differ little from the corresponding r's (i.e., r's for the same ripple in passband) of the ellipticfunction filter. The r's do, however, show strong dependence on the amount of ripple in passband L . P Defining mismatch constants as *This is true only in an approximate sense, when the curve on which the zeros of the elliptic function insertion powerratio are located (in complex frequency phase) approximates an ellipse. The distribution of the zeros on an ellipselike curve is found in most practical cases of elliptic function type ladder filters without mutual coupling.  61   62  u t" "M (4.1)* t rl rl uI r2 (4.2) u2 "r (4.3) r3 it is seen that as the ripple in passband L increases, u in creases, ul decreases, and u2 shows very little dependence on L . The effect of increasing ut and decreasing u1 may be visualized in complex frequency plane as moving the zeros of the insertion voltage ratio toward the j o* axis, i.e., the effect is similar to that of predistortion. This suggests that in networks with incidental dissipation, a compensation for losses similar to predistortion may be achieved by an increase of ut and a decrease of ul, in amounts depending on the amount of losses (i.e., Q or d as defined in 2.4) in the components. A larger change in ut and uI will be required for components with higher losses. As seen from equation (3.89), the cutoff frequency wt varies predominantly with the square root of k, i.e., with the effective cutoff frequency Wj (see Fig. 3.10). The m's are functions of Wc and the location of the poles *r is the terminating resistance, which is equal to 1 ohm in our analysis.  63  of the insertion loss function E(&), equation (3.83) and (3.84).* Since the location of the poles depends on the selectivity parameter k only, as seen from equation (3.61) and (3.63), and since ?j c depends on k predominantly, the m's depend mostly on k and very little on the amount of ripple in passband L . 4.2. Suggested Modification of Zobel Designs with the View of Obtaining More Efficient** Filters The method of designing a Zobel filter to satisfy given insertion loss requirements has been appropriately called the cut andtry method. One reason for this is the fact that the performance of a filter (i.e., its attenuation and phaseshift when inserted between generator and load) is only approximately given by its image attenuation and phase. This approximation is fairly good in the attenuation band, but is very poor in the passband. The labor required to design a Zobel filter that approx imately meets given requirements is usually very small. A commonly employed method of obtaining a filter which satisfies given require ments may be described in the following general steps: 1. Design a Zobel filter which approximately satisfies insertion loss requirements. *In these equations the zeros z, and z2, rather than the poles, appear. However, the poles are located inversely with respect to the zeros in the elliptic function filter, as seen from equation (3.60). **The term "efficient" is used here in the sense of high contribution to filtering action for each component.  64  2. Build a model filter according to design of Step 1. 3. Test the model of Step 2, establishing the difference between its performance and the requirements. 4. Dictated by the results of Step 3, make corrective changes in the design of Step 1, aiming at eliminating the difference between the filter performance and the requirements. 5. Repeat Steps 2, 3 and 4 until the aim of Step 4 is reached. It should be pointed out here that the cost (in terms of effort or dollars) of arriving at a satisfactory design using the above method is usually smaller than the cost of a design carried out by modern networkdesign techniques (i.e., by insertion parameter synthesis, leading to a satisfactory design in a more direct way).* This may partially account for the fact that Zobel designs are still so widely employed, although the modern techniques (in existence now for about twenty years) are known to produce better, often more economical, filters. It is in general easier to correct the performance of a filter (Step 4) in attenuation band than in passband. In fact, unwanted distortion in passband may sometimes be eliminated only by adding components to the filter (e.g., mderived terminal half sections), or by using an additional network (an attenuation equalizer). This is due to the pattern of distortion in passband inherent in *This is particularly so if the designer has reasonable ability to predict the effects of changes in design (i.e., changes of Step 4).  65  Zobel filters, as shown in Fig. 4.1. The insertion loss in passband of Zobel filters with equal image impedances Z, at both terminations, inserted between equal resistances Rt, is given by24 L 10 log10 + Rt) sin2 BJ 10 4 Rt ZI where B is the total phase shift through the filter. The envelope of the insertion loss in passband is L '20 log R) env 10 2 Ri Z ] Since the image impedance ZI is continuous in passband and becomes zero (or infinite) at the cutoff frequency Wc in a manner indicated by Figs. 2.2 and 2.3, termination of the filter by a constant resistance Rt results in an insertion loss envelope in pass band which increases near the cutoff frequency )c as indicated in Fig. 4.1. Most filter requirements allow a certain amount of maximum distortion in passband. Fig. 4.2 shows a typical insertion loss requirement; any insertion loss curve laying in the shaded area would satisfy the requirements. The curve shown may be considered typical for an efficient* onesection filter. *In general, the number of components needed in a filter in creases with increased selectivity k wUp/Wa, with increased min imum loss in attenuation band La and Lh, and with decreased maximum distortion in passband Lp. The curve shown utilizes the limits of the requirements in a fashion which could lead to a filter with fewest components. * 66  Lenvelope  Maximum distortion in passband C1. (2 c Fig. 4.1. Typical insertion loss characteristic in passband for a Zobel filter _____ 4 p p Ca Fig. 4.2, IN W Typical insertion loss requirement (straight lines) and a possible insertion loss characteristic which satisfies it Jh  67  In case of the Tchebycheff and elliptic function filters, utilization of the maximum allowable distortion in passband, i.e., designing a filter that has this maximum distortion in passband, leads to meeting the requirements with a filter having a minimum number of elements. Since the Tchebycheff and elliptic function filters can be designed to produce a predetermined amount of maximum distortion in passband (in contrast to the Zobel filter, in design stage of which there is no provision for obtaining a certain pre determined amount of distortion in passband), these filters can be designed to satisfy given requirements with a minimum number of components.* In case of the Zobel filter, the amount of distortion in passband can be controlled by mismatch at the filter terminals, i.e., by choice of the value for Rt, within a relatively narrow range only. In general, a Zobel design is termed successful if the filter satisfies the requirements in attenuation band, and turns out to have passband distortion which is below (or equal) the speci fied maximum allowable distortion. *Tuttle25 showed in a comparison of the performance of Butterworth, inverted Tchebycheff, ellipticfunction, constantk and mderived Zobel filters that the Zobel filter was inferior to the elliptic function type only. However, a comparison of a constantk Zobel filter with a Tchebycheff filter should prove that the latter has superior performance for the same number of elements.  68  If the requirement calls for a very small distortion in passband (in the 0.01 db. range), the usual way to obtain it with a Zobel design is by the use of mderived terminating halfsections, the image impedance of which (see Fig. 2.3) has less variation over an appreciable portion of the passband than the prototype (constantk) image impedance, Fig. 2.2. This then results in smaller passband distortion due to better matching at the termina tions, the price for which is an increased number of elements (above the number necessary to produce a given selectivity and discrimina tion) required in the mderived terminations.* If the allowable distortion in passband is high (in the I db. range), the Zobel design can take little advantage of it since at best the maximum distortion will be reached at one point in the passband, as indicated in Fig. 4.1.** From this it may be concluded that in order to make a Zobel filter more efficient (comparable to a Tchebycheff or elliptic function filter), some means of 1. Controlling the amount of distortion in passband 2. Achieving equal (or approximately equal) maximum ripple in passband should be included in the design procedure. *In some cases additional attenuation equalizing networks are used instead of, or for severe requirements, with the mderived terminations. **An efficient (in terms of the number of components) design results when the maximum allowable distortion in passband is reached a maximum possible number of times3 as in the case of the Tchebycheff and elliptic function filters.  69  It was shown in Article 4.1 that for the elliptic function filter the values of the r's are determined mainly by the ripple in passband, whereas all the other parameters ( w and the m's) C depend mainly on the selectivity k. This means that the main effect of a change in the r's will be a change in passband distor tion. There is a certain relation between the changes in the individual r's which results in an equal ripple of a certain mag nitude. These changes depend only little on the position of the poles (i.e., Wc and m) in the attenuation band. From the above reasoning the following may be concluded: Given a Zobel filter designed with any image parameters ( r and m's), the amount of distortion in passband may be controlled by changes in the impedance levels (i.e., in the r's) of its halfsections. In particular, for a Zobel filter consisting of mderived sections, an approximately equal ripple in passband of magnitude L will be obtained if the values of the r's are made equal to the corresponding r's of the elliptic function type filter having the same amount of ripple in pass band and approximately the same selectivity k; for a Zobel filter consisting of constantk sections, an approximation to equal ripples of magnitude L will be achieved by letting the p r's of the Zobel filter assume the values of corresponding r's of the Tchebycheff filter with a passband distortion of L . P Since, as it was pointed out in Article 4.1 (p. 61), the  70  rvalues of a Tchebycheff filter having a certain amount of pass band ripple L differ only slightly from corresponding rvalues of an ellipticfunction type filter having the same passband ripple L , it appears that: Making the impedance levels (the r's) of the halfsections of any Zobel filter (i.e., consisting of mderived or constantk sections and having any selectivity) equal to the corresponding r's of a Tchebycheff filter with passband ripples of magnitude L will result in the insertion loss of the thusly modified P Zobel filter to have approximately equal passband ripples of the same magnitude L . P 4.3. Limits of Improvement of a Zobel Filter There are many ways by which one may carry out the details of a Zobel filter design as outlined in Step 1, Article 4.2. These will in general depend to some extent on the type of the insertion loss requirements. Since this subject is well covered in litera ture,15' 16, 26 it will not be repeated here. For the purpose of this article it will be assumed that Step 1 has been carried out for two types of insertion loss require ments shown by the straight lines in Fig. 4.3 (a) and (b),* and that the resulting Zobel filters have an insertion loss as shown by the solid curves in Fig. 4.3. If now the halfsections of these filters are mismatched as *These two types of requirements are among the most commonly encountered in practice.  71  WJp Ja CJOp (b cb h Two types of insertion loss requirements and possible filter characteristics Lb L 1 Fig. 4.3.  W  72  outlined in Article 4.2, an insertion loss indicated by the dotted curves will be obtained. The question arises, how much of an im provement in performance of these filters may be expected from this mismatch? In the case of the filter in Fig. 4.3 (a), the answer comes from the fact that the new insertion loss (dashed curve) can be at best as good as the insertion loss of an elliptic function type filter with passband ripple L and selectivity k j / w The P P a minimum insertion loss in attenuation band Lae of such a filter is given by equation (3.65) or (3.67), or can be found from the curve in Fig. 3.11. Hence, at best, the improved performance of a modi fied Zobel filter will be an increase of the minimum insertion loss in stopband by* 6L = La L (4.4) ae a In the case of the filter in Fig. 4.3 (b), it is convenient 26 to use a transformation which will express the performance of the filter in terms of a performance as in Fig. 4.3 (a). It turns out that the minimum attenuation Lbe in the frequency range wob to W h is approximately the same as the minimum attenuation Lbe for *If the Zobel filter (before modification) had equal minima of insertion loss in passband, as indicated by the solid curves in Fig. 4.3, the minima will not, in general, be equal after the modi fication, i.e., after mismatching the halfsections. For most cases occurring in practice, however, the difference in the minima will be small and can easily be corrected by a small change in the mvalues, the effect of which on the passband ripple is negligible.  73  a filter as Fig. 4.3 (a) with a selectivity of k 6 Wh (4.5)* W 1 h Lbe may be determined in the same manner as Lae for the previous case, giving the maximum possible improvement as ALb = Lbe Lb (4.6) 4.4. Summary and Conclusions A method of synthesis of ladder networks was investigated with which conventional ladder networks may be obtained by proper choice of parameters of elementary building blocks. *The frequency scale in Fig. 4.3 (b) is transformed to an mscale by the relation m (I WJ 2/w 2)1/2, hence m z 0 cor responds to Oj m 1 corresponds to W oo, mb and mh correspond to Lob and WJh, respectively. Multiplication of the mscale by a factor I/mh introduces a new scale, say m' scale, in which m' 0 corresponds to wop, m' I1 corresponds to Wuh, and mb' = mb/mh corresponds to c An inverse transformation of m, i.e., CJ' = c p/(I m, )1/2, results in a frequency scale 6w' which is infinite at Wh, giving an insertion loss requirement with equal minimum from cb' (corresponding to dob) to Lw' oo. The selec tivity of the transformed requirement is k wJ '/Lb ', giving (4.5) upon substitution of w p uj and W b p /(l b)1/2  74  These parameters were determined for three types of two section filters having desirable insertion loss characteristics. As a result of analysis of the obtained parameter values, a method for improving Zobel filters was suggested. The design of a twosection Butterworth filter can be quite simply accomplished with image parameter halfsections having a constant amount of impedance mismatch given by equation (3.52) and a cutoff frequency Wc given by equation (3.54) or Fig. 3.6. The impedance levels of the halfsections, rI and r are constant, i.e., independent of any design parameter; the cutoff frequency cc is a function of the maximum distortion in passband only, as given by equation (3.54).* The design of twosection Tchebycheff filters may be as easily performed using equation (3.58) and (3.59), or the graphs in Fig. 3.8.** *The cttoff frequency W is not a critical design parameter for constant k type lowpass ana highpass filter types, i.e., it has no effect on the shape of the response curve. A change in the cutoff frequency amounts to a linear transformation of the frequency scale of the insertion loss versus frequency curve. **Butterworth and Tchebycheff filters with lossless elements can be designed with not much effort using already developed formulae for element values (i.e., without going through the actual synthesis from the insertion loss function). The engineering value of our approach is in the simplifications resulting from its application to lossynetworks, for which no explicit formulae for circuit element values are available at the present.  75  Computation of the element values by slide rule gives suffi cient accuracy for commonly encountered tolerances in manufacturing of the components.* A twosection filter of the ellipticfunction type can be designed using either the equations (3.83) to (3.89),** or the graphs in Fig. 3.13 with equations (3.83), (3.84), (3.89) and (3.74); use of the graphs is simpler but more restricted in the range of application. *A manufacturing tolerance in component values of + 1.0 per cent to + 2 per cent is quite common for capacitors in the range of approximately 100 A/f to 0.2 /ff and for inductors in the range of approximately 1 mh to 50 hy. **The values of the zeros, equation (3.61), can be found from many elliptic function tables, e.g., Ref. 21; they are tabulated for use in ellipticfunction filter design in Ref. 23. REFERENCES 1. 0. J. Zobel, "Theory and Design of Uniform and Composite Electric Wave Filters," B.S.T.J., 2 (1923) pp. 146, and "Transmission Characteristics of Electric Wave Filters," B.S.T.J., 3 (1924), pp. 567620. 2. M. I. Pupin,"Wave Propagation Over NonUniform Cables and Long Distance Air Lines," Trans. A.I.E.E., 17,(1900), pp. 445507. G. A. Campbell, "On Loaded Lines in Telephonic Transmission," Phil. Mag., 5,(1903), pp. 313330, and "Physical Theory of the Electric Wave Filter," B.S.T.J., 1 (1922), pp. 132. G. A. Campbell and R. M. Foster, "Maximum Output Networks for Telephone Substation and Repeater Circuits," Trans. A.I.E.E. vol. 39 (1920), pp. 231280. 3. S. Darlington, "Synthesis of Reactance 4 Poles," J. Math. Phys., vol. 18 (1939), pp. 257353. 4. W. Cauer, Theorie der linearen Wechselstromschaltungen, Akademische Verlag, Berlin, Germany (1954), 2nd ed. H. Piloty, "Weichenfilter," Zeitschrift fUr Telegraphen und Fernsprechtechnik, vol. 28 (1939), pp. 291298, 333344, and "Wellenfilter, Insbesondere symmetrische und antimetrische mit vorgeschriebenem Betriebsverhalten," Zeitschrift fUr Tele graphen und Fernsprechtechnik, vol. 28, No. 10 (1939) pp. 363375. 5. V. Belevitch, "Recent Developments in Filter Theory," IRE Transactions on Circuit Theory, vol. CT5 (Dec. 1958) pp. 236252. 6. T. Laurent, "Allgemeine physicalische ZusammenhMnge bei Filterketten," Arch. d. El. U. B. 12, H.1 (Jan. 1958) pp. 18, and "Echostatanpassung, eiee neue Methode zur Anpassung von Spiegelparameterfiltern," Arch. d. El. U. B. 13, H. 3 (March 1959) pp. 132140. 7. R. 0. Rowlands, "Composite Ladder Filters," Wireless Engineer, vol. 29 (Feb. 1952) pp. 51. 8. J. E. Colin, "TwoBranch Filter Structures with Three CutOff Frequencies," Cables & Transmission, vol. 11 (July 1957) pp. 179217. 76   77  9. W. Saraga, "Insertion Parameter Filters," TMC Technical J., vol. 2 (March 1951), pp. 2536. 10. A. J. Grossman, "Synthesis of Tshebycheff Parameter Symmetrical Filters," Proc. IRE, vol. 45 (April 1957) pp. 454473. 10. E. Green, AmplitudeFrequency Characteristics of Ladder Networks, Marconi's Wireless Telegraph Co., Essex, England; 1954. 11. J. K. Skwirzynski and J. Zdunek, "Design Data for Symmetrical Darlington Filters," Proc. IRE, vol. 104, pt. c (Sept. 1957) pp. 366380. 12. S. D. Bedrosin, E. L. Luke, and H. N. Putchi, "On the Tabulation of Insertion Loss LowPass Chain Matrix Coefficient and Network Element Values," Proc. Natl. Electronics Conf., vol. 11, (1955) pp. 697717. 13. T. Laurent, Vierpoltherie und Frequenztransformation, Springer Verlag, Berlin, 1956. 14. H. W. Bode, Network Analysis and Feedback Amplifier Design, D. van Nostrand Company, Inc., New York, 1945. 15. T. E. Shea, Transmission Networks and Wave Filters, D. van Nostrand Company, Inc., New York, 1929. 16. F. Scowen, Introduction to Theory and Design of Electric Wave Filters, Chapman & Hall, Ltd., London 1950. 17. W.H. Chen, Elements of Electrical Analysis and Synthesis, McGraw Hill Co. In press. 18. A. C. Bartlett, The Theory of Electrical Artificial Lines and Filters, John Wiley & Sons, New York, 1931. 19. V. Belevitch, "Tchebycheff Filters and Amplifier Networks," Wireless Engineer, vol. 29 (April 1952) p. 106. 20. L. M. Milne Thomson, Die elliptischen Funktionen von Jacobi, Springer Verlag, Berlin, 1931. 21. E. Jahnke and F. Emde, Tables of Functions, Dover Publications, New York, 1945. 22. W. N. Tuttle, "Design of TwoSection Symmetrical Zobel Filters for Tchebycheff Insertion Loss," Proc. IRE, vol. 47 (Jan. 1959), pp. 2936. 78  23. E. Glowatzki, "Sechsstellige Tafel der CauerParameter," Abhandlungen der Bayerischen Akademie fer Wissenschaften, Neue Folge, Heft 67 (1955). 24. W. Saraga, "Insertion Loss and Insertion PhaseShift of Multi section Zobel Filters with Equal Image Impedances," P.O. Elec. Eng. J., vol. 39 (Jan. 1947), pp. 167172. 25. W. N. Tuttle, "Applied Circuit Theory," IRE Trans. on Circuit Theory, vol. CT 4, (June 1957) pp. 2932. 26. J. H. Mole, Filter Design Data, John Wiley and Sons, Inc., New York, 1952.  79  BIOGRAPHY David Silber was born on October 2, 1922, in Lodz, Poland. His undergraduate studies were pursued at the O.V.M. Polytechnikum in Munich, Germany, from which he received the degree of Electrical Engineer in June, 1950. After arrival in the United States in 1951, he pursued graduate studies at the University of Cincinnati, Evening College and Summer School, while employed by the Keleket XRay Corporation in Covington, Kentucky. Since 1954 he has been employed by the Communication Accessories Company in Kansas City, Missouri, taking leave in 1956 to enter the University of Florida for further graduate studies. At the University of Florida he held a graduate fellowship for two years and taught electrical engineering for one year. He received the degree of Master of Science in Engineering in August, 1957. This dissertation was prepared under the direction of the chairman of the candidate's supervisory committee and has been approved by all members of that committee. It was submitted to the Dean of the College of Engineering and to the Graduate Council, and was approved as partial fulfillment of the requirements for the degree of Doctor of Philosophy. August 13, 1960 Dean, ollegeof Enginerin Dean, Graduate School SUPERVISORY COMMITTEE: Chairman ~G'. (C KV E&^^ UNIVERSITY OF FLORIDA 1 II II 11111 Il III III 3 1262 07332 052 4 3 1262 07332 052 4 