Sie sind auf Seite 1von 6

Chasing Chebyshev

Steve C. Cripps

e all have our pet peeves. As a somewhat frequent long-distance flyer, I have to admit that one of mine is the talkative airline passenger. There are actually two scenarios. Your next-door stranger with whom you will be rubbing arms, shoulders, and who knows what else for the next ten hours can usually be silenced fairly quickly with a few monosyllabic replies and supporting body language. More irritating, I find, is the extended introductory earful from a bigmouthed specimen in the row behind, who is sitting next to a more receptive neighbor than myself. Topics usually includealthough are by no means restricted tothe rags-to-riches life story, the firebrand political views, the impeccable siblings, and the never-ending litany of incompetent management in high corporate places. Fortunately, the monolog does not usually survive the noise of takeoff, and at least sitting in front you avoid being asked the dreaded question, So whats your line of business?

Steve C. Cripps is with Hywave Associates, The Mangold House, Hare Lane, Buckland St. Mary, Chard TA20 3JS Somerset, U.K., stevehywave@aol.com.
Digital Object Identifier 10.1109/MMM.2007.906917

take the well-worn option of using the E word. At least it seems to help in the shutting-up process, as does pulling out the latest issue of IEEE Transactions on Microwave Theory and Techniques and pretending that you are able to read it. I was recently talking with a somewhat distinguished gentleman from academia who has done much work in the information/communi PHOTODISC cation theory area, and I was amused by his solution to this probI could at this point head down a lem. When I asked about his main well-trodden path and complain about research activity, he smiled and said I the standard reaction when you answer guess Ive spent most of my life chasing with the E word (engineer). Yes, pro- Shannon. Taking this lead, and thinkfessional engineers should hold the same ing about my own technical career, I esteem in the public eye as doctors and thoughtperhaps with a pinch or two lawyers. Yes, its tedious that when you of saltI could say that I have been qualify your reply with another E chasing Chebyshev a lot of the time. I (electronic engineer) they assume you realize this doesnt have quite the same can help them fix their TV set and, fur- buzz about it as chasing Shannon, but I thermore, that fixing TV sets is more or suppose that takes us down the old E less what you do for a living. My trouble word path again. The fact is, one way is that Im not sure many of my activities or another, I have spent an awful lot of these days qualify as engineering, time designing matching networks. As early as the first week in my first even under my own definition of the term. How do you explain to a complete, job, I was staring at the inevitable Smith nontechnical stranger that you are cur- chart, trying to figure out how to match rently measuring memory effects in RF the input impedance of a primordial power transistors? The S word (scien- GaAs field effect transistor (FET) device tist) seems to be out of fashion at the over a 500-MHz bandwidth around 3 moment, so I suppose in the end I just GHz. It was immediately apparent that

34

December 2007

computer-aided design (CAD) optithe simple line-and-stub matching ditions and practical considerations from mization problems I had learned to solve in my those of three-dimensional filter design, (c) a double-section, four-element undergraduate courses were inade- and I know of only a small number of design based on the exact synthequate. Seeking with nave enthusiasm a worthwhile efforts to present the subject sis and bandpass transformation solution in the literature, I was assured from the viewpoint of the active circuit of an n = 2 Chebyshev prototype, that the classical reference on the sub- designer [2][4]. Rightly or wrongly (and heres yet with additional impedance transject was in the company library. This, of formation. course, refers to the famous volume by another Microwave Byte), the convenFor the recordand Im not proud of Matthai, Young, and Jones (MYJ) [1]. I tional way of designing broader bandquickly discovered, however, that a width active circuits is to take the device thisI think it took me about ten years to senior colleague kept the only copy and match the input and output to 50 . work my way through this little list, and locked away in his desk. I rather think Analog IC designers may cringe, but in there are no prizes for guessing which he enjoyed being the local expert and a world where only discrete transistors solution wins, although the margin of vicwas reluctant to lend it to anyone are available, and where device and tory is still to me quite surprising. This is else. I actually still remember the package parasitics all but eliminate the shown in Figure 2, where each network is evening I finally managed to persuade possibilities of combining transistors swept for transmission and input return him to part with it, albeit just for an into higher gain op-amp-like blocks, it is loss. The simple single low-pass network evening. Tea, dinner, and even the then a well tried and tested approach that has is a tremendous workhorse in RF and statutory trip to the local pub were post- served the higher GHz frequency indus- microwave design, but it is clearly not up poned indefinitely as I eagerly scanned try well enough. Figure 1 shows a to this task; the same would apply to the the esteemed volume for all the answers matching problem that I would quantify complementary but much less used single to my matching problems. I was disap- as being of intermediate difficulty; high-pass section. On the other hand, the pointed. Although without question a matching a 10- resistance into a broad- four-element Chebyshev network of (c) landmark text, the problems it tackled band 50- termination over a single shows impressive performance, with a seemed somewhat distantly related to octave frequency bandwidth. I will sum- minimum return loss of almost 18 dB mine, and I became quickly frustrated marize my own lifes experiences with across an octave bandwidth. It does, howwith the generality of much of the this kind of a problem by showing three ever, have a less convenient topology. Series capacitors have never been popular analysis. It was really a book about fil- basic approaches: ters, not matching networks for active (a) The Occams Razor, a single sec- as matching elements due to their fundamicrowave devices, which for the most tion, low-pass two-element network mentally lumped nature. They must be part had probably not been invented (b) a double low-pass, four-element implemented in the form of discrete comwhen the book was written. network, using an intuitive ponents which have to be emplaced, solIt actually took me the better part of design approach with subsequent dered, or die attached. Specific values demanded by the design may the next 20 years to get the not be available with sufficient whole matching thing straighttolerance, and substantial paraened out, by which time it was 0.49 nH sitics must be absorbed by the a bit late since the new world S11 design. They also block the dc of RF wireless seemed to have 10 1.04 pF supply, which is highly inconless need for it. Thereby hangs venient in active component another story (and, rest (a) applications. More on the assured, another Microwave ZM downs and ups of high-pass Bytes!), but for the present I 0.37 nH 0.85 nH networks later. thought I would try to share The network in Figure 1(b) my current knowledge of the S11 is the logical extension to the subject, even if only for poster10 1.69 pF 0.74 pF simple low-pass structure in ity. As usual, I will probably Figure 1(a). It delivers a greatoffend a number of readers (b) ly improved matching perforwho think that matching is ZM mance, has a convenient topomerely a byproduct of filter 2.51 pF 0.62 nH logy for implementation at design, and as such has been higher frequencies, and appears defined in precise terms for S11 10 0.65 nH somewhat intuitive in its opermany decades (and, indeed, in 0.63 pF ation. Basically, an intermeditomes such as MYJ). So be it. (c) ate impedance (Zm ) is selected The fact is that the kind of matching problems presented such that the matching is done by active circuit design have a Figure 1. Three matching networks: (a) single low-pass, in two smaller steps. There different set of boundary con- (b) two-section low-pass, and (c) high-pass/low-pass network. have been various rules of

36

December 2007

thumb proposed for selecting this intermediate impedance. A popular reasoning, based on Smith chart Q-curves, suggests using the geometric mean of the source and load terminations as the intermediate impedance. This seems sensible enough, given that each section needs to have the same bandwidth, and we know that bandwidth is a function of the normalized transformation ratio. The only problem with this is which frequency to use in order to compute the specific element values. Again, rules of thumb abound, and both the geometric and arithmetic mean of the specified band limits are cited. Both, it turns out, do not give an optimum set of component values, but in the modern era of CAD, almost anything will do since we will just hit the optimization button to obtain the optimum element values for a given return loss target. This is actually the design procedure which gives the circuit elements in Figure 1(b) and the response in Figure 2(b). It turns out that the optimizer has found a network which corresponds very closely to the optimum

solution using this topology, an n = 2 Chebyshev response. In the past, it was necessary to use an exact design approach [5], which gives essentially the same element values. This procedure can itself be performed using numerous CAD tools and design wizards, some of which are available as shareware. Given the substantial matching improvement over a single low-pass network, many are sufficiently happy with the result that they feel this is all they need to know about matching. Indeed, what I have described thus far will solve a wide range of matching problems, wide enough to take many through their technical careers. But it turns out there is quite a bit more depth to this problem, even if we confine ourselves to four matching elements. The cascaded low-pass topology has some practical benefits in terms of supplying bias to active devices and in implementing the elements. In many cases, the series inductances can be replaced with lengths of transmission line, and shunt capacitors can be realized as open

0 Ins Loss (dB) -0.1 -0.2 -0.3 -0.4 -0.5 4 5 6 7 8 (c) (b) (a)

0 Ret. Loss (dB,dotted) -10

circuit stubs or lumped capacitors. In a sense, one could say that although officially having four elements, in practice there may effectively be just two. The third plot in Figure 2(c) can still be classified as a four-element network and shows yet further performance improvement over the double low-pass network. Although the series-C, shuntL topology may not be so convenient in many practical cases, I rather feel that the performance benefit is not so widely appreciated; it has certainly not been widely used in my observations. The official design procedure for this third network is quite complicated and derives directly from filter synthesis theory [6]. From my viewpoint, I have always felt that the filter part is a means to an end, and the real magic in designing such networks lies in a series of network transformations originally described by Kuroda. These are known as the Kuroda identities, and some of these were adapted by Norton [2] to address matching problems. [Note that I have cited secondary references here, because I have personally had difficulty finding the original papers.] For my purposes, I have actually further abbreviated one of Nortons results, giving the single transformation shown in Figure 3. If we happen to have a circuit which contains a shunt/series inductor combination, we can flip the inductors around to give a series/shunt configuration with transformed inductor values. This transforms the input impedance by a factor of n2 , where n=1+ Ls Lp

-20

Figure 2. Matching performance of networks in Figure 1: (a) single-section low-pass, (b) two-section low-pass, and (c) high-pass/low-pass.

LS ZL n 2Z

LST ZL

LP

LPT

n =(1+LS /LP ) LPT =nLP LST =n(n 1)LP

Figure 3. Impedance transformation using series shunt inductors.

at all frequencies. A remarkable result, and not at all intuitive to me. One can see, perhaps, some vague parallels with a transformer, but rest assured the relationships apply with no coupling between the inductors (there is a complementary transformation that uses capacitors). Anyway, the point is that in order to leverage this result we need to jimmy up a circuit that contains a series/shunt inductive pair, so we can use the transformation to get our broadband match. The most obvious way of doing this is to

38

December 2007

Ls LS S11
10

Lp

Lp

Figure 4. Filter structure based on series shunt resonators. provide each inductance with an attendant capacitance in order to reduce the frequency characteristics of the inductors, as shown in Figure 4. So, you guessed it, we have ended up with a filter. With some reluctance, therefore, we now have to figure out the best way of assigning values to the four elements to obtain a decent bandpass characteristic. But to get the best bandwidth and simultaneously the highest impedance transformation, we also wish to maximize the ratio of the series and shunt inductors. It is instructive to pursue this using traditional Smith chart graphics rather than filter synthesis algebra, which can give the answers but for me lacks the intuitive understanding of the process. Lets start with a 10- source impedance and assume we wish to transform, or match, this to a higher resistive load. So, first of all, we add a series resonator. At the output, this has the effect of spreading out the impedance/frequency trajectory along the 10- constant resistance circle on the Smith chart (see Figure 4). If we stipulate that our midband resonant frequency is the geometric mean of the upper and lower band edge frequencies, the impedance arc will straddle the horizontal axis in a symmetrical fashion. We now add the shunt resonator and stipulate that its resonant frequency will be maintained at the defined midband frequency. So as we reduce the value of the shunt inductor (and, thus, simultaneously increase the shunt capacitor to maintain the same resonant frequency), the impedance arc will bend inwards and ultimately cross over itself, the endpoints sliding along a constant conductance circle, as shown in Figure 4. Heres the interesting part. If we draw a circle centered on the horizontal axis, whose left-hand intersection with the axis coincides with our 10- starting point, this circle represents a target area for maintaining the return loss below a stipulated level if we shift the reference characteristic impedance to the impedance represented by the position of center of the circle (see magnified plot, Figure 5).

||= .135

10

13.1

Figure 5. Multiple choices of series and shunt elements, keeping a constant stipulated maximum reflection.

n2

3 2 1

0.1

0.2

0.3 Ls

0.4

Figure 6. Transformation ratio as a function of inductor values.

So we can adjust the two inductor values such that the impedance loop stays within the circle, just touching it at the band edges. I find this an entertaining and instructive exercise to do using a simulator with a nice fast tuning feature (such as Microwave Office). So we can stipulate the minimum final return loss of the network by appropriate selection of the circle center location. For a given circle size, we then have to determine which pair of inductance values gives the greatest ratio of Ls/Lp. Figure 5 shows three cases in a continuum of possible pairs of values, where the circle radius corresponds to a reflection coefficient of 0.135, or 0.08 dB transmission loss, for a reference impedance of 13.1 . One can plot a few more values (Figure 6) and quickly home in on an inductance pairing that gives the maximum ratio, which is shown as the middle (red) trace in Figure 5. Interestingly, this optimum case is the one where the impedance trajectory crosses the horizontal axis at the center of the circle. This indicates that we will see a perfect match at two points in the band, while maintaining a mimimum return loss represented by the radius of the circle at all other frequencies. It so happens that the return loss plot of this optimum case corresponds precisely to an n = 2 Chebyshev response, and the values we determined here could have been found by synthesising an n = 2 Chebyshev filter. So not only does the Chebyshev filter give the maximum bandwidth for a stipulated maximum inband return loss, it also allows for the greatest impedance transformation factor by having the largest possible Ls/Lp ratio. Although well known, this result has always bothered me. After all, Chebyshev just came up with an abstract mathematical function (he died, by the way, in 1894). The fact that it seems to be the optimum response for both bandwidth and impedance transformation can certainly be demonstrated well enough, but I have never been quite sure about where and how these mystical properties arise. Looking again at the red trace in Figure 5, it seems to jump out of the paper, as if trying to tell us that it represents something special. There is almost a languid elegance

40

December 2007

about the way it gyrates inside the prescribed boundary, leaving one in no doubt that there is nothing better. Its not so much a case of chasing Chebyshevhis shadow keeps falling across our path at every turn. Anyway, to finish our matching problem, we now deploy the inductor transformation. The inductor ratio Ls/Lp at the point of maximum transformation is .315/.33, and the value of n is 1.955. So we scale the impedance to the right of the inductors by a factor of 3.82, transforming the 13.1 to 50.1 , giving the topology and element values originally shown in Figure 1(c). Now heres the thing. After all that work, the final network looks almost like something we could have tried a lot earlier. Rather than matching to some intermediate impedance using a double lowpass section, we seem to have ended up doing basically the same thing, but with a high-pass, low-pass cascade. It turns out that this is a better topology for implementing the two-step approach. Traditionalists may cringe, but once again a CAD optimizer will home in on the correct optimum values, given this starting topology and only very approximate element values. But the key point is that the frequency characteristics of the high-pass and low-pass sections complement each other to give a more optimal broadband match. This is illustrated in Figure 7, where the impedance traces of the input and output sections are plotted at the midpoint of the network. It can be seen that the midpoint impedances are not so far off being conjugately matched over the entire band.

8 8

LP

4 Combined 4

HP

Figure 7. Low-pass section (red, looking towards 50- termination) and high-pass section (red, towards 10- load) frequency trajectories (48 GHz, 1-GHz points).

42

December 2007

I have restricted this discussion to twosection networks and matching broadband resistive impedances. Unfortunately, matching active devices isnt as simple as this, because there are usually parasitic reactances which need to be absorbed. It is well known that as the number of sections is increased, the match and the impedance transformation ratio will increase, albeit with diminishing returns as n gets higher. This can also improve the chances of finding a parasitic absorption scheme. But Im not sure that I have seen so many examples of the high-pass/low-pass cascade being extended to many multiple sections. As I noted earlier, such structures have not in the past been especially convenient to implement, and cascaded low-pass structures have been more common for the seriously challenging matching applications. Times change. The highpass network appears to be back in favor, despite its underused past. I could design a multisection, composite lowpass/high pass structure and demon-

strate some quite impressive matching ratio over a wide, multi-octave band. I could claim that a given physical length shows a better bandwidth and matching ratio than a more conventional cascaded low-pass structure. I could then write a paper titled Novel Broadband Matching Structure Using Cascaded Right-Handed and LeftHanded Transmission Line Elements. High-pass is definitely back in again these days, and I think this is a key issue in the plethora of such papers that are currently appearing in MTT and APS journals and conference proceedings. Personally, I do not accept that a cascade of a few lumped elements constitutes a transmission line, even if the individual elements are physically small in comparison to a wavelength. I do accept that recent improvements in circuit board technology, particularly plated vias and multilayers, along with the availability of smaller chip components and radio frequency integrated circuit (RFIC) processes, have given the

high-pass network something of a renaissance. If high-pass is in, it would indeed appear possible to design circuits that demonstrate useful feats of matching, phaseshifting, and bandwidth and which are arguably more compact than the traditionally more favored low-pass networks. But highpass filters have always been an integral part of the classical filter synthesis dynasty, and the recent wide-ranging rediscovery of their useful properties can hardly be claimed as revolutionary, nor can the placement of a few highpass sections along a length of conventional transmission line be claimed to reverse the polarity of at least two fundamental constants of the universe. On the other hand (Im right-handed, by the way), by way of offering an olive branch to my many correspondents in the meta world, I will concede that taking the concept into two and three dimensions may well be a different matter. I also recognize that this extension is, for the most part, the main focus of the metamaterials community and its sponsors. I myself in the June 2007 installment of Microwave Bytes promoted the concept that conventional two-dimensional structures could benefit from some more extensive study, and there may yet be undiscovered structures that have useful practical applications. So the same may indeed apply to meta-structures. But it remains to be seen whether either of these research paths will end up being remembered as solutions without problems in a world where active components continue to usurp older passive techniques.

References
[1] G. Matthaei, L. Young, and E. Jones, Microwave Filters, Impedance Matching Networks, and Coupling Structures. New York: McGraw-Hill, 1964 (reprinted Artech House, 1995). [2] L. Besser and R. Gilmore, Practical RF Design for Modern Wireless Systems, vol. 1. Norwood, MA: Artech House, 2003. [3] B.J. Minnis, Designing Microwave Circuits by Exact Synthesis. Norwood, MA: Artech House, 1996. [4] P. Abrie, Design of Microwave Amplifiers and Oscillators. Norwood, MA: Artech House, 1999. [5] G. Matthaei, Tables of Chebyshev impedancetransforming networks of low-pass filter form, IEEE Trans. Microwave Theory Techn., vol. 19, no. 8, pp. 939964, 1964. [6] S. Cripps, Advanced Techniques in RF Power Amplifier Design. Norwood, MA: Artech House, 2002, Ch. 7.

44

December 2007

Das könnte Ihnen auch gefallen