Sie sind auf Seite 1von 11

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 44, NO.

3, AUGUST 2002

413

Solution of the Transmission-Line Equations Under the Weak-Coupling Assumption


Clayton R. Paul, Fellow, IEEE
AbstractThe transmission-line equations for a three-conductor lossless line in a nonhomogeneous medium are solved in symbolic form (as opposed to a numerical solution) in the time domain under the assumption of weak coupling. The result provides insight into crosstalk between the lines. Numerical solutions of this problem exist but do not illustrate the factors that influence the crosstalk waveforms. The symbolic solution given in this paper makes those factors clear. The validity of the weak-coupling assumption is also investigated. Index TermsCrosstalk, literal solution, transmission lines, weak coupling.

For the receptor circuit we have (2a) (2b) These transmission-line equations are obtained from the perunit-length equivalent circuit shown in Fig. 2 [1], [2]. The quanand are the per-unit-length inductances of the retities and are the perspective circuits, whereas the quantities unit-length capacitances of the respective circuits. The quantiare the per-unit-length mutual inductance and caties and pacitance between the two circuits, respectively. Equations (1) and (2) are coupled via the terms on the right-hand sides of the equations. The terminal conditions must be incorporated into the general solution of (1) and (2) and these are, from Fig. 1 (3a) (3b) and (4a) (4b) to where the line is of total length and extends from . The solution of these equations requires the solution of the transmission-line equations in (1) and (2) along with the incorporation of the terminal loads given in (3) and (4). A number of numerical solution techniques are available for this problem. The finite-difference time-domain (FDTD) method is described in [1]. A SPICE subcircuit model can also be obtained for inclusion in a SPICE or PSPICE solution [1], [2]. In these numerical methods, the various factors that contribute to the crosstalk waveform are not evident; only a numerical solution is obtained. A symbolic solution of these equations along with the incorporation of the terminal loads into that solution would make the effects of the line and load parameters on the waveform clear. Such a solution for a three-conductor lossless line was obtained in [1]. However, in order to facilitate a solution, it was assumed that the surrounding medium was homogeneous. For example, bare wires in air would correspond to this case but wires with dielectric insulations or a PCB would not. The results of this paper will provide a similar solution for the case of an inhomogeneous surrounding medium. Although the solution in [1] for a homogeneous medium was exact, it appears impractical to provide a similar, exact solution for the case of an inhomogeneous medium. In order to facilitate a solution for the case of an inhomogeneous medium, it will be assumed that the two circuits

I. INTRODUCTION ONSIDER the three-conductor line shown in Fig. 1. The ground symbol merely symbolizes the third or reference conductor for the line voltages. This reference conductor may be an infinite ground plane, a wire, a printed circuit board (PCB) land, an overall shield, etc. The other two conductors may also be of various types such as wires, PCB lands, etc. The line is assumed to be a uniform line in that the conductor and surrounding dielectric cross sections do not vary along the line. The line is assumed to be lossless in that the conductors are assumed to be perfect conductors, and the surrounding medium, which may be inhomogeneous as in a PCB, is assumed to be lossless. One conductor with the reference conductor is driven by a source and source resistance and having open-circuit voltage . This circuit is referred to as is terminated in a resistive load the Generator circuit. The other conductor with the reference conductor is referred to as the Receptor circuit and is terminated , and at the far end in a at the near end in a resistive load, . Electric and magnetic fields are produced by resistive load, the voltage and current of the generator circuit and interact with the receptor circuit producing crosstalk voltages at the terminals and . The objective of this paper is to of that circuit, provide a symbolic solution (as opposed to a numerical one) for these crosstalk voltages. The voltages (with respect to the reference conductor), and , and currents of each circuit, and , are functions of position and time . These are related, for the TEM mode of propagation by the transmission-line equations [1]. For the generator circuit we have (1a) (1b)
Manuscript received January 14, 2002; revised March 18, 2002. The author is with the Department of Electrical and Computer Engineering, Mercer University, Macon, GA 31207 USA. Publisher Item Identifier 10.1109/TEMC.2002.801753.

0018-9375/02$17.00 2002 IEEE

414

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 44, NO. 3, AUGUST 2002

Fig. 1. The crosstalk circuit to be investigated.

and on the right-hand side of (1). The weak-coupling assumption assumes that this back interaction is negligble so that (1) approximates to (5a) (5b) The question of how weakly coupled must the lines be in order for this to be valid can be partially answered from the exact solution for lines in a homogeneous medium given in [1]. In that solution, it was found that the characteristic impedance of the generator circuit could be written as where the coupling coefficient is (6) This is very similar to the coupling coefficient between two turns of a transformer. Hence, we would expect that weak coupling would be a reasonable assumption so long as (7) are weakly coupled. The accuracy of this assumption will also be investigated. II. THE WEAK-COUPLING ASSUMPTION The weak-coupling assumption is reasonable for many cases of practical interest and means the following. The voltage and current of the generator circuit produce electric and magnetic fields that interact with the receptor circuit and induce voltages and currents in that circuit. This inducement of voltages and currents in the receptor circuit are through and the terms on the right-hand side of (2). These induced voltages and currents, in turn, produce electric and magnetic fields that provide a back-interaction or second-order effect by inducing voltages and currents in the generator circuit. This back-interaction is symbolized by the terms As an example of a numerical solution which will be shown, consider the PCB shown in Fig. 3(a). This is an example of a coupled microstrip which resembles a PCB. Two 5-mil lands are of thickness 47 on one side of a glass-epoxy board mils. The lands are 5 mils in width and 1.4-mil thickness. They are separated edge-to-edge by 50 mils. A ground plane (the reference conductor) is on the other side of the board. A numerical program, MSTRPGAL.FOR described in [1], was used to determine the per-unit-length parameters giving H m H m pF m pF m

Fig. 2.

The per-unit-length equivalent circuit.

PAUL: TRANSMISSION LINE-EQUATIONS UNDER THE WEAK-COUPLING ASSUMPTION

415

so that the one-way time delay along the line is (11) Assumption #3: The Lines Are Matched at All Ends: (12) This last assumption is very restrictive since it is rarely achieved in practice. Designers attempt to match with various schemes, but a complete match is rarely achieved at all frequency components of the signal on the line. It is this last assumption that causes the actual crosstalk to, in many cases, differ drastically from the predictions of this model. The derivation we will subsequently provide does not assume completely matched lines and/or symmetry. Why are these three assumptions made in the derivation of the Jarvis model? The answer is evidently to simplify the solution of the transmission-line equations. Nevertheless, the so-called Jarvis model detailed in [3][7] gives the predictions of the near-end and far-end crosstalk voltages as (13a) (13b) where the backward and forward coupling coefficients are defined as (14a) (14b) where the total mutual inductance and capacitance are the product of the per-unit-length values and the total line length and . The results and are denoted as for the near- and far-end crosstalk voltages given in (13) are the commonly-used prediction model based on the above three assumptions [3][7]. The backward and forward and , in (14) are defined slightly different coefficients, than in [3][7] but the result is equivalent. Suppose that the open-circuit voltage is a ramp function illustrated in Fig. 3(b). Equation (13a) indicates that the near-end crosstalk is a pulse of length twice the one-way line delay, and (13b) indicates that the far-end crosstalk is also a pulse but has width equal to the rise time of the source since it is proportional to the derivative of the source voltage. These are illustrated in Fig. 4(a). If the source rise time is much greater than , the crosstalk waveforms simthe one-way delay, i.e., plify to those shown in Fig. 4(b). These are identical to the inductive/capacitive coupling model in [1], [2] which are valid for . In order to illustrate the accuracy of the Jarvis model, we will show computed crosstalk results for the PCB shown in Fig. 3(a) having an open circuit voltage in the form of a ramp rising to 1 V with 500 ps rise time as shown in Fig. 3(b). The total line length is 10 inches or 0.254 m. From the above stated per-unit-length , ns, and parameters we compute

Fig. 3. The configuration investigated. (a) Cross-sectional dimensions. (b) Waveform of the source.

Using these parameters, we calculate and . Hence, we would expect that, for this separation, the weak-coupling assumption would be valid. III. THE JARVIS MODEL There is a popular solution for the general case of a three-conductor lossless line in an inhomogeneous medium that is often used for the prediction of time-domain crosstalk in digital circuits [3][6]. This solution was apparently first published by Jarvis [7] and will be referred to as the Jarvis Model. The derivation of the solution was facilitated by making three important simplifying assumptions. Assumption #1: The Lines Are Weakly Coupled: Assumption #2: The Line Is Symmetrical: Symmetry means that the generator and receptor conductors are identical in cross section, and they are equally spaced from the reference conductor in a symmetrical fashion. The PCB in Fig. 3(a) is symmetrical although we will not make that assumption in our general solution. This symmetry provides that the self inductances are equal as are the self capacitances: (8a) (8b) Under the weak-coupling assumption, we may logically define the characteristic impedance of both circuits as (9) and the velocity of propagation on each line as (there are in fact two modes of propagation on the line whose velocities of propagation are different for an inhomogeneous medium [1], [2]) (10)

416

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 44, NO. 3, AUGUST 2002

(a)

(b) Fig. 4. The near-end and far-end crosstalk waveforms of the Jarvis model. (a)  < T . (b) 

 T.

and . The terminal configuration is as shown in Fig. 1 with . The results of the Jarvis model in (13) are compared to the exact results computed with a SPICE subcircuit model generated by SPICEMTL.FOR described in [1]. Fig. 5 shows the near-end and far-end results for this matched line. The Jarvis model provides reasonable prediction accuracy although there is some residual ringing indicating that the lines are not truly matched. In [1] it was shown that it is impossible to

completely match a multiconductor line using only single resistors between the end of each line and the reference conductor. In order to match a multiconductor line, it is required to terminate the near and far ends with a resistive network wherein there are not only resistors between each line and the reference conductor but also between each line. In order to investigate the adequacy of the weak-coupling assumption, we compare the predictions of the Jarvis model to the exact results for an edge-to-edge separation of 10 mils. For this

PAUL: TRANSMISSION LINE-EQUATIONS UNDER THE WEAK-COUPLING ASSUMPTION

417

(a)

(b)

Fig. 5.

=R =Z

The crosstalk waveforms predicted by the Jarvis model compared to the exact results for 50-mil separation between the two circuits and R . (a) Near-end crosstalk. (b) Far-end crosstalk.

=R =

separation, we compute the per-unit-length parameters as H m H m pF m pF m Using these parameters, we calculate and . The total line length is again 10 inches or 0.254 m. From the above stated per-unit-length pa, rameters we compute ns, and and . The terminal configuration is again shown in Fig. 1 with . The

results of the Jarvis model in (13) are compared to the exact results computed with a SPICE subcircuit model generated by SPICEMTL.FOR described in [1]. Fig. 6 shows the near-end and far-end results for this matched line. Although the general waveshapes are similar, there is inaccuracy in amplitude and in time phase. Although , it is evident that for this close 10-mil separation, the weak-coupling assumption is not adequate. IV. SOLUTION OF THE TRANSMISSION-LINE EQUATIONS UNDER THE WEAK-COUPLING ASSUMPTION We now present the solution of the transmission-line equations under the weak-coupling assumption. Essentially we want to obtain the solution to (2) and (5) with the terminations given in (3) and (4) incorporated into that solution. We will not make the final two assumptions of the Jarvis model, i.e., in this solution, the lines need not be symmetrical or matched.

418

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 44, NO. 3, AUGUST 2002

(a)

(b)

Fig. 6.

=R =Z

The crosstalk waveforms predicted by the Jarvis model compared to the exact results for 10-mil separation between the two circuits and R . (a) Near-end crosstalk. (b) Far-end crosstalk.

=R =

A. The Frequency-Domain Solution Observe that (5) are the transmission-line equations of the generator line in the absence of the receptor line (this is the essence of the meaning of weakly coupled). The solution of these equations can be facilitated by taking the Laplace transform of both sides where denotes the Laplace transform variable (15a) (15b) The exact solution for this (essentially two-conductor) line with the terminal conditions incorporated is [1], [2]

(16b) where the characteristic impedance, velocity of propagation, source and load reflection coefficients, and one-way delay of the generator line are defined as (17a) (17b) (17c) (17d)

(16a)

(17e)

PAUL: TRANSMISSION LINE-EQUATIONS UNDER THE WEAK-COUPLING ASSUMPTION

419

Now, we turn to the solution of the receptor circuit equations given in (2). Laplace transforming these give (18a) (18b) The solutions for the generator line voltage and current are given in (16) and hence the right-hand sides of (18) are known and act like sources much in the same fashion as an incident electromagentic wave excites a two-conductor line. The solution can be obtained by writing (18) in the form of state-variable equations as [1]

where the one-way time delay of the receptor circuit is (24a) and the constants and are (24b) (24c) (24d) Incorporating the terminal conditions given in (4) (25a) (25b)

(19) gives the final form of the solution as The solution to these types of equations is well known to be [1] (20) where (21), shown at the bottom of the page, holds and the characteristic impedance and velocity of propagation of the receptor circuit are (22a) (22b) Substituting (16) and (21) into (20) and carrying out the integration yields (27a) where (26a)

(26b)

(27b) (27c) (23a) (27d) and the receptor circuit reflection coefficients are given by (27e) (23b) (27f)

(21)

420

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 44, NO. 3, AUGUST 2002

B. The Time-Domain Solution The solution of (26) in the time domain relies on the important inverse Laplace transform property (28) In other words, multiplying a Laplace transformed function by results in simply delaying that function by . Before we proceed, there are two terms in the denominators of (26) which are troublesome. These can be expanded into infinite series as

Although these are complicated, the scheme to solve them is to sequentially invert the components of (26) using (28). We first invert the terms

(31a)

(31b) using the operator in (28) that simply scales and time shifts . Once this is done, again time shift and scale this to obtain, according to (29), to yield (32a) (32b) so that

(29a)

(29b) For a moderately mismatched line, only the first few terms of these expansions are significant since the reflection coefficients are less than or equal to unity. In order to illustrate this, consider a line having terminal impedances that are twice the characteristic impedance of the appropriate circuit. Hence, the reflection . Their coefficients are and . Consider products are , that is in the form of a periodic trapea source voltage . The zoidal pulse train having equal rise and fall times, primary spectral content of this signal is contained below ap[1], [2]. For this form of the signal, proximately we would like to have the expressions in (29) valid for all frequencies up to this maximum. Substituting into (29) gives, for example

(33a)

(33b) To obtain this waveform, we simply scale and time shift the waveforms in (31) and add the component waveforms in time. Again, for a moderately mismatched generator line only the first few terms of (33) are significant. Finally, we apply (34a) (34b) to yield the resulting crosstalk waveforms

(30a) In the case where the rise/fall time is equal to the one-way delay, , (30a) becomes

(35a)

(35b) (30b) . The exact result is . Using and , using the first three the first two terms gives , and using the first four terms gives terms gives, . Hence, for this moderately mismatched line, we can truncate the series after four terms and obtain an adequate approximation. The solutions in the frequency domain given by (26) can be placed in the time domain using the fundamental result in (28). In the case of a completely matched line, and , all four reflection coefficients are , and the result in (26) zero, reduces to (36a) (36b)

PAUL: TRANSMISSION LINE-EQUATIONS UNDER THE WEAK-COUPLING ASSUMPTION

421

(a)

(b)

Fig. 7. The crosstalk waveforms predicted by the results of this paper compared to the exact results for a 50-mil separation between the two circuits and R Z and R R (open circuit). (a) Near-end crosstalk. (b) Far-end crosstalk.

=1

with the time-domain result (37a) (37b) In the case of a symmetric line, matched, (36) reduces to , which is also

In the time domain, (38) yields (40a) (40b) For the same conditions as the Jarvis model (matched and symmetrical line) the exact result in (40) is identical to the Jarvis model given in (13) with the forward and backward coefficients given in (14). V. AN IMPORTANT SPECIAL CASE (38b) In the case of a symmetric line wherein and , both one-way delays are the same, . This is a frequent situation on PCBs wherein lands on one level are above an innerplane such as is shown in Fig. 3(a). Secondly, de. signers frequently use series matching to make Hence, a useful special case is where (1) the line is symmetrical

(38a)

and we have used the following result in (36b): (39)

422

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 44, NO. 3, AUGUST 2002

(a)

(b)

Fig. 8.

=Z

The crosstalk waveforms predicted by the results of this paper compared to the exact results for 10 mils separation between the two circuits and R and R R (open circuit). (a) Near-end crosstalk. (b) Far-end crosstalk.

=1

and (2) each line is matched at at least one end. When each line or are zero and either is matched at at least one end, either or are zero. In this case, the terms resulting in infinite , series expansions in (29) are zero so that thereby simplifying the general result in (26). We will examine a case which meets these conditions. The line is symmetric. The terminal conditions are , and so that and . The model in (26) simplifies, in the time domain, to

(41a)

(41b)

The predictions are compared to the exact results for a 50-mil separation in Fig. 7 and for a 10-mil separation in Fig. 8. The predictions for the 50-mil separation in Fig. 7 are quite accurate, whereas the predictions for the 10-mil separation in Fig. 8 again show inaccuracy in both amplitude and time phase. This again shows that the weak-coupling assumption is not adequate for the 10-mil separation. It is worth noting that if the one-way time delay for the 10-mil separation is changed from its logically correct value of ns to ns, the results of (41) match the exact results very closely in time phase. Hence, it appears that one consequence of strong coupling between the lines is to reduce the net one-way time delay from its value with the lines isolated from each other. However, reducing the one-way delay used in (41) from its logically correct value is somewhat arbitrary and cannot be relied on for other cases. These final results show that the Jarvis model (which assumes a symmetric line and all terminations completely

PAUL: TRANSMISSION LINE-EQUATIONS UNDER THE WEAK-COUPLING ASSUMPTION

423

matched) cannot be used for a line that does not satisfy the restrictions of that model since the waveforms for the Jarvis model in Figs. 5 and 6 do not resemble those of Figs. 7 and 8. The Jarvis model in (13) is frequently mistakenly used for predictions wherein all terminations are not matched. This results in grossly erroneous predictions with that model. VI. SUMMARY AND CONCLUSIONS The transmission-line equations were solved in symbolic form for a three-conductor line under the weak-coupling assumption. The results are applicable to inhomogeneous media as in a PCB. The symbolic solution illustrates the factors affecting the crosstalk, whereas a numerical solution does not. It was found that so long as the separations between the generator and receptor lines was sufficient to support the weak-coupling assumption, the predictions of the model are very accurate. Although the results of this paper were derived for the case of a three-conductor line, they should also be applicable, under the -conductor lines where weak-coupling assumption, for . This assumes that one circuit (the generator circuit here) is driven by a source (the source of the crosstalk) and the other lines are passive and also weakly coupled to the generator circuit and each other. REFERENCES
[1] C. R. Paul, Analysis of Multiconductor Transmission Lines. New York: Wiley Interscience, 1994. [2] , Introduction to Electromagnetic Compatibility. New York: Wiley Interscience, 1992.

[3] [4] [5] [6] [7]

MECL System Design Handbook, 4th ed., Motorola Semiconductor Products, Phoenix, AZ, 1988. A. Feller, H. R. Kaupp, and J. J. Digiacome, Crosstalk and reflections in high-speed digital systems, in Proc. Fall Joint Comp.Conf., 1965, pp. 511525. J. A. DeFalco, Predicting crosstalk in digital systems, Comp. Des., no. 6, pp. 6975, 1973. S. Rosenstark, Transmission Lines in Computer Engineering, NY: McGraw-Hill, 1994. D. B. Jarvis, The effects of interconnections on high-speed logic circuits, IEEE Trans. Electron. Comput., vol. EC-12, pp. 476487, Oct. 1963.

Clayton R. Paul (S61M70SM79F87) was born in Macon, GA, on September 6, 1941. He received the B.S. degree, from The Citadel, Charleston, SC, in 1963, the M.S. degree, from Georgia Institute of Technology, Atlanta, in 1964, and the Ph.D. degree, from Purdue University, Lafayette, IN, in 1970, all in electrical engineering. He is Emeritus Professor of electrical engineering at the University of Kentucky, Lexington, where he was a member of the faculty in the Department of Electrical Engineering for 27 years. He is currently the Sam Nunn Eminent Professor of Aerospace Systems Engineering and Professor of Electrical and Computer Engineering in the Department of Electrical and Computer Engineering at Mercer University, Macon, GA. He is the author of 12 textbooks on electrical engineering subjects, and has published numerous technical papers, the majority of which are in his primary research area of electromagnetic compatibility (EMC) of electronic systems. From 1970 to 1984, he conducted extensive research for the United States Air Force in modeling crosstalk in multiconductor transmission lines and printed circuit boards. From 1984 to 1990, he served as a consultant to the IBM corporation, in the area of product EMC design. Dr. Paul is a life member of the IEEE EMC Society, and a member of Tau Beta Pi and Eta Kappa Nu.

Das könnte Ihnen auch gefallen