Coppiiqht c 1951 6s the Institute of E lecLrical and Electronics Errginee rs, Inc.

A Practical Method of Designing RC Active Filters

X TH E FRE(jUE NCV range below about 30 cps, the dissipation factors of available inductors are generally too large to perm it the practical design of inductance-capacitance (LC) or resistance-inductancecapacitance (RLC) filter networ k:s. The circuits de-

\vhere the ind ividual t ransfer ratios, [/V,(s) ] / [D(s) ] ,

have the same properties ascribed to (1). I I active elements are added to the in ultiterminal network i n the man ner shown in Fig. 2, the o ver-all transfer ratio, [eo(s) / ei (s) ) , is given by

scribed in the following pages v ere developed and collected to provide an alternative method of realizing sharp cut-off filters at very low frequencies. I n mans' cases the active elements can be si mpJe cathode-follov er circuits that have stable gain, lov output impedance and a large dynamic range.
TANSFER HAHACTKRlSTlCS

.fi passive two-terminal pair network: consist my of resistive and capaci tive elemen ts has an open circ u it tra nsfer ratio of the form

where s is the complex frequency variable (o 1)u) , the

o, and , are real positive const an ts, and the fi, are nonzero. A II the poles of G(x) I ie on the negative real axis
of the s-plane, a property that severel)' I inn its the i ppl t-

ration of passive RC circu its to sharp c u t-off fi lters.

The unbalanced (u l 1)-term inal (a ir RC net v ork

shown in Fig. 1 can be characterizeJ 1\- the relation

Fig. 2II ulti-terminal activ e tie tworl:.

f?enera HJ speaking, it is possible to select an appropriate net v on: and a series of constants K i K e i so that

the poles and zeros of G(s) can be placed anywhere in

the com plex p1i ne. (Complex critical frequencies will

occur, of course, in conjup=ate pairs.) L'nder certain condit ions, ;i1l I ri nsfer functions of a given degree can be achieved v'ith one fi. xed ne tv-on: bx select ion of appropriate s. .4ctua J1-, a miximum of four active elements are re-

qu ired for the circu it of Fig. 2 ; two am pJ ifiers for each

Fig. I The (n -|- I )- terminal pair uetvv on:. Fhis paper is an abridged version of Tech. Rep. Xo. 50, pu blished by Lincoln Lab. M. I.T., Lexington, Mass., in May 6, 1954, u rider same title. may be made to that report for useful design charts of B u tterworth and Tchebycheff filters. The research in this docu ment was supported jointly by the Army, Navy, arid the .4ir Force under contract with M.I.T. j Lt ncoln Lab., Mass. I net. Tech., Cam bridge, I\'lass.

of the tv'o sets of s, one with positive gain, one with negative ga in. Jhe rem a in ing val ues of K i, and K ei can

be obta tried bv means of passive atten uators. in y transfer voltage rat to ord inority realizable by means of passive RLC network:s can be achieved with the circu it of Fig. 2. In addition, a variety of oscillators can also be characterized by the transfer function U(s)
in (3).

J9SS

fallen and bey . A Practical Method o[ Designing RC Actiue Filters

are the zeros of the denominator polynom ia1 which , for convenience, can be normalized in the foI lowing manner. \\'e have

I n most cases it is desirable to limit the appl icat ion

of the general circuit of Fig. 2 to transfer rit ios with

onl) two conjugate poles. . in) given t rin sfer ratio can be achieved by a cascade of simpler c irctiits of this kind and one or more passive RC ne tv\ orks. The second -order transfer I unc I in .

can be realized by means of the circu its of Fig. 3, which

are special cases of that of Fig. 2. Fhc arr:i ngement of

-(-

d -1- 1

Fig. 3(b) includes tv-o act we elements t hit t max be separate ampl ifiers or one impl ifier v ith tv o input
points. The RC passiv e net\s orks genera 11y have tw o capacitors and two resistors each, and i c ircu it-design

proced ure is ava ilable that affords one considerable control over the orders of mign it ride of the com ponen ts.

Fig. 4H igh-pass filter circuit.

In the x-plane, the zeros of De(s) lie on a circle of radius ue and hive a real part equal to du /2.' The shape of the Irequencx- characteristics of D (s) are dependen t onlv on the value of the parameter d ; the con
stant u determ ines their positions in the frequency

The design of circu its of the form ill ustrated in. Fig. 3 can be facilitated by' means of the catalog of possible circuit arrangements that has been compiled and is given in a later section. There are probab I v' other useful circuits of this form that could be added to the catalog, but for most application s the present list of eigh teen networks will be found adequate.
GENERA .%RRRO. 4CH TO SET&-O RK DESIGN

donna in. and be determines the relative ampl itude. The parameter u can be given the physical interpretations resonan t Irequencv, cut-off frequency, etc. , depend ing upon the nature of the numerator of d(s).
It is conven rent, in desig ning a circuit for a given

D (s) , for one to set us = 1 radian per second temporarily, and to establish the required value of d . The network response can then be shifted in frequency to ma by d ivid ing the resistive elements or the capacittve elements of

The basic objective of a design procedure based on the network catalog is the control over the locations of the poles of the transfer voltage ratio (4). These poles

the circuit by the desired value of ua. In most of the netvorks in the catalog, there are five basic design variables : two resistances, two capacitors,
' This i z tr we out y ter d 2 iv-hen d > 2, the zeros I ie on the negativ-e real axis, a case that is not of present in terest.

76

IRE TRAN!3ACTION!3CIRCUIT

TI-IEOR Y
the value of can

the gain K . The relationships between the variables that are independent of d are given with each network. Several additional parameters that have been found useful for designing a network for a given d include two products of a resistance and capacitance (designated F; and Fs), the ratio of the resistances (p), and the ratio of the capacitors (y). The establishment of a specified value of d is accomplished by means of two of these parameters and the gain J. With each network in the catalog is a short table that specifies, for a given choice of parameters, the appropriate group of design relations for d given at the end of the catalog. The form of the n umerator of 6(x) is determined by the particular network chosen for the function. In some cases the n umerator constants can easily be established at the desired val ues; in others an attempt to do this

may severely limit d in the denominator and lead to an curt design. In this case, the n umerator

be realized by means of additional passive or active works. A method of network design is discussed later.
CATALOG OF SECOND-O RD ER AT1\/E ET\YORRS

Defin itions of Parameters:

Para meters

d Fenn tit as G rou p

Parariieters

' d Formulas Group

Al terna ti ve with 2 active i npb *

Parameters Group

Formulas

Alter nat i x'e xxith 2 active inputs:

6Forniulas
Parameters Cnoup

320

Faramerer

Iarameters

AC - l
li - K

J955
Dzsicx or .ficrive NET\\'ORKS BY MEANfi OF THE CATALOG While anyone may start with a given transfer function or with the networks in the catalog and work out his own method for selecting parameters, one approach has
been foUlld Useful and is dSCri bed here for those wishing

(b) or (c). For most purposes, a recommended procedure is the assumption of x:, K, and the sol ution for T. As an example, suppose that a required

to design such cifCUitS most directly. In any event, it is strongly recommended that one work through at least one group of relations in the foregoing section to gain insight into their meaning. The basis of the general procedure suggested below is the necessity that the practical design of an active filter must be carried out within the limitations imposed by the available components. Restrictions on the size of capacitors, n umber and complexity of amplifier stages, and requirements for variability are typical factors that impose practical circuit limitations and must be con- trolled. Realizmg a Speerfied Value of d When a particular network has been selected, one chooses a set of tv-o parameters(p, Ti) , (p, Tz) , (y, 'i) or (y , Tz)and locates the appropriate group of design formulas in the foregoing section. If the problem involves restrictions on the size of capacitors, then is a useful parameter. On the other hand, if control of the resistance values is more important, one may use p. In general, Ti and Ti are equally convenient parameters except where one of them determines a factor in the nu merator of 6(s). In the same section, each formula group includes: (a) The expression for d in terms of Iand the two parameters (x:, T) , selected above (:c stands ior p or y ,

is to be realized by means of network No. 3 in the catalog, using one cathode follower as the active element. (The circuit to be used is shown in Fig. 4.) Suppose further that #z (parallel combination of the two bracing resistors) shall be 1 megohm and that both capacitors shall have the same value. First of all, we shall choose (y, Tz) as our design parameters, so that we may easily control the ratio of the capacitors and the val ue of d . According to the catalog, Formu la Group III for d is indicated. Setting -1, we have

If we set K 0. 9, a reasonable value for the amplifier of Fig. 4, then

T .

4(1 -]- :r) (1 K) d

T for Ti or Tz) ;

(b) The solution of the equation in (a) for T -, (c) The solution of the equation in (a) for I; (d) The minimum value of Isatisfying the equation in (a) with arbitrary :r, positive T [Jq, is obtained by solving the equation 6K(d, x, T)/6 T 0 for T and substit uting the solution, T+c i. into the expression for K(d, :r, T) |; (e) The minimum value of :r satisfying the equation in (a) with arbitrary I, positive T (x has a minimum in the same sense as I, above) ; (f) The value of expression (b) when I-J (:r - ; is the same condition and both values of T are the same in this case) ; (g) The value of expression (c) when d 0 (the significance of J, is discussed below). In establishing the values of :c, T and I, one selects any two of them arbitrarily (in an algebraic sense=with more purpose in the practical sense) , subject to the algebraic limitations I J :r ,.. The value of the remaining parameter is then determined from formula
One might also employ (u, v) as design parameters, but this patr

appears to be tess useful than the others since it provides less control
over the actual magnitudes of the components.

The expression for d in Formula Group III, d (1 -I-y) /( T) -I- Ti(1 K), suggests that a choice of the smaller value of Tz above would result in a more stable circuit, in that variations in the active element will have less effect on the value of d. Up to this point, we have # -1O, Tz 1.59, and ii -C,. Then, making use of the relation Ti Te 1, we find Ci = Cz = 1.59 ;ujf and di = 3.93 X 10. It can be stated as a general rule of thumb that values of d greater than 0.5 can be realized most easily and with the simplest circuits; as d approaches 0. 2, more care becomes necessary in the circuit design. Finally, values of d or be order of 0. 1 or less demand active elements that are more complicated and highly stabilized, and passive elements that have been carefully adjusted within close tolerances. The latter values of d are not generally encountered in low-frequency filters. As stated it has been found most convenient to design second-order networks on the basis of 1 radian per second, and to make a subsequent shift of their characteristics to the appropriate frequency by altering the passive elements. The basic invariants under a frequency transformation of this kind are the parameters

82

IRE TRAN! ACTION!CIRCUIT THEORY

in any event the active gain should generally be at least 1 /d times as stable as the expected val ue of d. There is another 1:ind of instability often characteristic of active RC netivor ks of the kind d isc ussed here, namelx-, their tendenc\ to become oscillators. This tendenc_\ is most previ len I \\ hen the val ue of d is small. Even in some circ uits \vhere the active gain is ostensibly free from d rif I, osc illat ions max be sustaineci by an amplifier that drives itself in to a region of its characteristics \\ here the gain is far greater than expected. .-\ basic cure for :i sit uation of this 1:ind is the use of a feed bacl: ampl ifier Ior the act ive elemen t, such that the g.tin K is given by (6) \\ here A is the giin of the am(l ifier withou t feed back, Q is the feedback rat io derixed from passive elements.
I t is easilx- seen that th e val ue of is absol u tel \- I i mi ted

p and y ; as long as the ratio of resistors and the ratio of capacitors remain constant, thP frQ Ullc y characteristics of the networks will have the same shape. The necessary invaria rice of p and y ind icates the techn ique for making filters with variable cut-off frequencies. Innperfactions in the Actiue Elements

It has been assu med, heretofore, that the active elements of the networks in Figs. 2 a nd 3 and in the

catalog possessed the ideal attrib utes: in fin ite inpu I impedance, zero output impedance and stable gain. I t is therefore i mportan t, in the design of an active network of this kind, for one to insure that the i mperfections in the am plifiers used do not appreciably' deteriorate the desired performance of the circ u it. With regard to fin ite output impedance, it can be seen that in manv cases the active elements d ri ve a portion of the passive netvork through a resist ive element. I n this case, the design ca n be made to incorporate the on tput impedance in the resist i ve elemen t a nd effect ivelv
neu t ralize its effects.

On the other hand, where an a mpl ifier drix-es a ca picitive branch of a netv'ork, it is imperative that the output impedance be considers bl x smaller than an \ of t he resistive elements of the netv-ork. This cond i t ion is most

to 1 /Q , regardless of the val ue of A. The simple cathodefo1lo\\-er circu it ill iist rated in Fig. 4 is an example of this kind of act ix e elemen t . The crit ical va I ue of K for a given net\v-ork:, ,, , is given at the end of the network catalog. .-\ pract ical circ u it design in ust i net ude means for insuring tha t the active gain does not approach this vit ue.

serious when the vaI ue of r/ is very small and the q tin of the active element is close to | 1. The limit a tion of the A djust ment oJ Ph y.s i ca I .Vetzeor#s a mpl ifier ou tpu I i mped ince to i reason i bl \- small v-al ue \\"hen :i ii act i\-c net \\ ork has been constr ucted \vu th will genera 11) preven I an ) sign ifica nt alteration of I he ph \sica1 cont ponen I s, minor ad j ustments in the lat ter network characteristics in the vic in it 5 of th cti t -off I re- a re f reQuen II \ rec] u ired to .ich ieve the performance in quenc_\'(ice) . On the other ha net , t he ;i tten u.it ion I icated 1i\ the desiyn . I I the depart ure from the exachieved in certain network:s in regions \\'eJ I be)-ond c u t- pected chiri- teristics is not large, the t rim in ing of a off will fall short of the expected val ue bee a use the ou t- single ci pacit or or resistor mix su ffice to proper1\ posip ut impedance, though smi 11, is st ill fi n ite. Th is sit ua - t ion the net v on: c h.i ra ter ist ice in the Irequenc x I ot ion his been observed in to\v--pass net\v orks .i t h igh inn in. Tue sh a Jes f I he ch i r;i cterist ics are most eisi ly frequencies and in notch circu its at the n u II freq uenc\ . :i I tered b\- id j ti stment of the ga in fi. Behavior of th is k: ind can best lie in vest iy:iteJ I :i I n the cx'eu t th;it the depart ure from expected chardirect anal\'sis of the part ie u lar circu it involveJ . liii i e :i terist ics is li rgc .ind is not .iccou ntable to the usual onl y ver;' high freq uenc res or n u II freq uen- res are of tolera rices in c-nipoiien ts, one ma tool: to the fo1lov'ing interest in this case, the anal\sis can be simpl ified 1\ the as possible son r-es of error: in isca ten lation of design assu mpt ion of these extreme frequent-5 end it one. Jarimeters, excessive :i nrpl ifier out pu I impedance, poor Under these circu inst an ces, the ou tpu t i tiipe1.i n e n pit-i t or Q. I t is u n rei son rule to ignore the Q of shou Id be negl iglibly small in comparison \\ it h ( 1 ) |irye pi per ci pi c i tors in net v on:s \\'here the resist ive

times the value of resist ive elemen ts of the net worLs. elements are of the or1er of 1 megoh in or more. Fortu natel) , strict requ iremen ts of th is sort to riot \\hen the netv ork J esiy n inc1uJ es an active element
c ur often in low-freq uencv filters. Another interesting departure of the act ive elemen ts from the ideal is the drift in their gain . \\"ith most it the networks in the catalog (those with Ti T 1 ) , the posi tion of the transfer characteristics in the Ireqiiori c I omain is independent of the active element ; \\-ith a 1 e\\ \i h ose ya in is slight I less t h:in + t (e.g. , the circu it of Fig. 4) , i t is usu al lv I ifficu I I to measure or adjust the J u:iiitit \- ( t K) I tree II v i th necessary acc urac\-. I I a poten t iometer is a vi ila ble for trim ming the gain (as in Fig. 4) , one nii\- effe-I the id j ustmen t in a simple min ner 1a\ observing the over-a II network ampl it ttde-

others this is not so. I n both cases, how ever, .i d rift in

gain will result in a change in the act ua I val ue of d :ind in the shape of the I fiQ UfiDc)' th aracterist ics. I t is of ten

frequenc) response. dhe expected frequenc--response

h:iracterist i- for th ree t ran sfer 1 u nc I ions is ill ustrated in in. 5. ote thit , iltho ugh the fritter are \v ritten on

possible, as in the previous example, to reJ rice th is tependence by an appropriate choice of para meters ; b It I

t he bis is of 1 rad ian Jier second, the f requenc y response

is ind icateJ :i t the true resonant " 1 requenc) uo.

326

fallen and bey . A Proctco/ Method o[ Designing RC Actiae Filters

Fig. 5-\mpl itude-freq tie ncy response of several active net works.

It is sometimes desirable to avoid the necessit y tor adjusting the elements of a net w'orl:, part ten larly the passive ones, by their prior selection within specified tolerances. .A lthough the su bject of tolerances has not been studied in detail, the expected variabil ity of the
network characteristics will generall y be of the same order of magnitude as the tolerances in the passive elements, when the active elements are properl )' aJ j usted.

made to absorb some of the factors of /\/(s). The remaining second -order denom ina tor pot ynom rats are each
iden t ified with in i n1ivid ua1 active RC network of a

form shown in the catalog. Practical considerations determine the pairing of the appropriate n umerator factors with the second -order denom in a tors; the relative properties of the various networks are the key to the pairing

A third- or higher-order transfer ratio can be written in the form

(7)

process. to general rules can be given for selecting a group of second-order networks, for the choice is dependent on the part icular requirements of the over-all circuit. H owever, an example can be used to indicate the kini of reason ing one may employ in a given situation. Consider the transfer ratio

in a variety of ways such that

the transfer

ratios,
an d assume we are i nterested in real izing 6(s) with the simplest and most easil y designed circuits possible. First of all, one passive network: is ind icated : either 1 / (s 1- 2), s/(s --| 2) or (s +1)/ (s l 2). The possib Ie secondorder active netv'ork:s will then have functions of the

G;(s) - [/\/;(s) ] / [D;(s) ] , contain first- or second -order pol ynom rats with real coefficients in the nu merator and

denominator. All of the first-order denom inators can be achieved with one passive RC network or by means of a cascade
of isolated RC networks which , incidenta II y, can be

form

Fig. 6----Circuit with transfer function.

Fig. 7----One aIternat ive to circuit of Fig. 6.

327

IRE TRANSACTIONSCIRCUIT THEORY

1955

8allen and Key. A Practical Method o[ Designing RC Actiae filters

Several additional points should be mentioned in regard to the cascading of second-order networks. First, a series of networks involving cathode-follower amplifiers as active elemen ts, or a series of cathode-coupled ampliher networks, has a useful property whereby the bias given to the first amplifier can usually be carried over from stage to stage. This is importan t where resistive elements couple adjacent networks, since it avoids the need for coupling capacitors which , at low IreQuencies, might become fairly large. In designing a circuit, it is always desirable for one to employ only capacitors whose impedance-frQU cy characteristics help to determ me the transfer ratio ; i.e., capacitors that are part of the basic networks in the catalog. Other capacitors, which might be used solely for bypassing, coupling, etc. , will usually have impedances negligible at the frequencies of interest and will be appreciabl y larger than those in the former category. In this regard, it is often advisable, when one is designing an active RC filter to be followed by ampliher circuits, to incorporate single zeros and poles of the 6(s) into the ampl ifier interstage coupling networks, or even to add additional ones that can be cancelled by the litter
network.

Because the last two functions contain zeros at s - 1, the establish ment of which will tend to inhibit the freedom usually desired in the design for d in the denominator, we may j ustifiably simplify the problem by incorporating the zero at 1 in the passive network. Reference to the catalog further indicates that networks for nu merators 1 and s are simple and can be achieved with positive gains less than 1. Finally, since the smaller val ues of d reQuire amplifiers that more closely approximate the ideal active element, we shall associate the s' numerator (network No. 3) with the denominator whose d = 0.5, so that R can absorb the output impedance of the amplifier. A reasonable set of factors of 6(s) is then

A possible circuit is shown in Fig. 6 (page 83) , to xvhich an alternative is given in Fig. 7 (page 83) where the network is based on a different set of factors of 6Is). In general, there are a great number of possible circuits realizing a given 6(s) that can be justified on the basis of particular reQUiremen ts for simplicity, outpu t impedance, stability and so on.

Finally, where possible, amplifiers in a chain of active networks should be placed so that those with the smallest dynamic range appear last. Two examples of low-frQuency hlters that have been constructed and tested in the laboratory are shown in Figs. 8 and 9, on the previous page.