Download as pdf
Download as pdf
You are on page 1of 16
4.5 Other Biquads 161 integrator_with a positive loss term with one with a negative loss term. This resulted in the Ackerberg-Mossberg biquad of Fig. 4.29. Observe that the improved performance over that of the Tow-Thomas biquad in Fig. 4.10 is obtained at no additional cost. It requires only a wiring change. We will, therefore, in the remainder of this book deal only with the Ackerberg-Mossberg biquad whenever we need to use a two-integrator loop circuit. Nevertheless, the literature on active filters contains a great number of second-order circuits developed for a variety of applications. In the next section we shall study a few of them to present the student with a choice of circuits that may be called on to realize a design requirement. BIQUADS In addition to the three-amplifier biquads of Section 4.4, engineers have invented biquads that use two opamps or even only one opamp. The number of ‘opamps does, of course, affect cost and power consumption, but the main differences between different circuits are their sensitivity to component tolerances, to be discussed in Chapter 12, and their versatility in being able to tealize different transfer functions. 4.5.1 Sallen—Key Circuits Sallen-Key filters were among the first active filters that appeared in the literature (Sallen and Key, 1955). This reference contains a whole catalog of possible structures that permit the realization of various types of transfer functions. One such structure to realize a lowpass function is shown in Fig. 4.31a. To analyze the circuit for its transfer function we obtain first by voltage division (4.96) and write node equations at nodes V, and Vp: (sCz + G2) Va = GaVo (4.97) Vo (Gi + G2 + 5C1) = GiVi + 8C\V2 + GaVa (4.98) For an ideal opamp we have V, — V_ = 0 and we can solve the previous three equations for the transfer function V3/V;. Let us instead assume a real opamp with finite gain A(s) and use Ye V2/A to be able to investigate the effects of A. We next solve Eq. (4.98) for Y and substitute the result into Eq. (4.97): G GM +sCiVa C2 + Gp —- ——- _ ] V, = Gp (: at 2s cinosrsc "G+ tsC Solving for V,, we obtain Gi G2V, + sC,G2V2 S°C\C2 + s[C2 (Gi + G2) + CiG2] + GiGa 162 SECOND-ORDER LOWPASS AND BANDPASS FILTERS Figure 4.31 Sall filter; (a) de gain H gain H =aK. which with Eq. (4.96) and V, — V_ = V2/A results in 1 KGiG.T yA ve 79 = 3 = —___t—__ 7 POG +5 [C4 Gi +62)+616r(1 a ca) +GiGz Y To derive the design equations, let us assume for now an ideal opamp, A = 00. We the lowpass function V2 KG,G2 T(s)= v 1 $°C\Cz + s[C2 (Gi + Ga) + C1G2(1 — K)] + GiG2 For convenience we set Cy = C) = C to obtain 4.5 Other Biquads 163 KGiG2/C? Hod Ti) = eee 8° + s[Gi + G22 — K)/C + GiG2/C"s° + san/O + 0H (4.101) We have again expressed the function in its standard form as in Eq. (4.17a) that lets us identify how the three filter parameters are expressed in terms of components: a GiG2 (4.102) me (4.103) 2 G4Ge-® : H=K>1 (4.104) Inselecting the four elements C, Gi, G2, and K to realize the three filter parameters we have ~ some freedom of choice. Since, in practice, the available capacitor values are usually limited, let us choose a convenient value for C. Also, let us assume that the de gain is not important and can be fixed at H = K. Then we may select Ry = Ro = R and obtain with 1 O=7-% Low Y K-) 3 4105 the element values fi ee es o Race ad K=3-Ga1t Re, ites Re=Q—1/QRs (4106) Rx is arbitrary and can be chosen equal to R to minimize the number of different resistor values. sign a Sallen-Key lowpass filter with fy = 12.5 kHz and no peaking. The de gain is not "specified. Use an LM741 opamp. Brn cording to Fig. 4.13 a value Q = 0.707 is required to avoid peaking. Choosing C = 0.01uF results in 1 2a x 12.5 x 10" x 10% Rp = Ra(2 — 1/0.707) = 746 Q Ry) =R2=Ra = kQ = 1.273 kQ, _ from Eq. (4.106). Figure 4.32 shows the circuit and its performance. The de gain equals 4 dB as designed, and fy = 12.5 kHz. The dip and subsequent rise in gain at high frequencies are _ caused by higher-order o| dynamics. 164 SECOND-ORDER LOWPASS AND BANDPASS FILTERS = 0.01 uF (a) Bonde Plover (b) Figure 4.32, The circuit design for Example 4.7 and test results. (Bode Plotter scales: 3 to 300 K/Hz; —40 to +10 dB.) The cursor is at the —3 dB point (cursor readout: 12.5 kHz, 1.02 4B). Observe that the Sallen—Key circuit has a potential problem: it is clear from Eq that for large Q, when K must be close to 3, Q is critically dependent on the exact of K. To get a measure of how sensitive Q is‘to the value K, let us use Eq. (4.105) at the slope of the function Q(K) versus K, dQ{kK)_ d ( LoD ma aR 3-K})” @-Ky In terms of finite differences, this equation can be recast as AQ © Q?AK or AQ AK oe ee 30 — 1)—— 0 xox =GO-N= It says that rhe percentage error in Q equals K Q = 3Q — | times the percentage e1 The ratio 4.5 Other Biquads 165 og _ 40/2 _ AQ/Q = dk/K AK/K (4.109) is called the sensitivity of Q to K. We shall study sensitivity in more detail in Chapter 12. For now we just note that for large values of Q the Q-sensitivity becomes very large. For instance, if we need to design a peaking lowpass with Q = 10 (see Fig. 4.13) the sensitivity, with Eq. (4.108), is 29, thatis, a 1% errorin amplifier gain K is predicted to result in approximately a30% error in quality factor. Such large errors in filter parameters will generally not be acceptable. They either require expensive processes to get very tight component tolerances, such as a 0.1% error in K to reduce the expected error in Q to 3%, or call for tuning. For a lowpass design, however, where Q is of the order of unity and K * 2, a 1% error in K results in + 2% error in quality factor. We can conclude, therefore, that Sallen—Key fiters are useful only for small values of Q as are normally required in lowpass filters. For high-Q filters we need to develop alternative realizations. They will be discussed in Sections 4.5.2 and 4.5.3. Gain Adjustment in Sallen-Key Filters Another limitation of Sallen—Key lowpass filters is that their de gain is fixed at H = K = 3-1/0. We shall see later that, in particular for cascade designs, the biquad gain must be adjustable if filters with good dynamic range are to be designed. Naturally, we could solve the problem by cascading the Sallen—Key circuit with an opamp-based amplifier of gain less than or larger than 1, but this solution may be too expensive. Let us look at a modification of the circuit that avoid additional opamps. Consider first the problem of realizing a desired value of H_< K. We can reduce the gain by including a voltage divider into the ci h a way that the poles do not get changed. Observe that the filter has a resistor Ri in series with the i input so that the resistance seen from V;, tothe left in Fig. 4.31a equals Ry. We may replace that resistor by a voltage divider consisting of RX and ceo (4.110) . . ad\ Ry as shown in Fig. 4.31b. Then R, and Ry in parallel equal Ry, is : Ri Gi Gi+G, a+(1-a) and the effect of this step multiplies V; in Eq. (4.98) by a < 1. The transfer function will not change except for a multiplying factor a, so that the gain equals H = aK as desired: V2 aKG,G2 T(s) = 2 = ey Vi s°C\C2 + s{C2 (Gi + G2) + CiG2(1 — K)] + GiGr ‘The opposite problem, increasing the dc gain above the value K, is not as easy to solve. Normally, an additional opamp pear to provide the required gain. Ri The Effect of A(s) on the Sallen-Key Circuit Let us investigate what deviations are caused by the operational amplifier. Anticipating this concern, we did include in the circuit analysis leading to Eq. (4.99) the finite amplifier gain A. Starting from that equation we find with C) = Cp = C and G; = G2 = SECOND-ORDER LOWPASS AND BANDPASS FILTERS ce T+ KA 1 2 pe 2 s +00 (3 ea) te To get a manageable expression for the errors, we make the approximation 1/(1 + Kj 1 — K/A for |A| > K. This leads to the equation T(s)= 2 T(s) = Kaj(1 — K/A) : Ss? + swp[3 — K(1 — K/A)] +0 Using the opamp integrator model of Eq. (2.18), A(s) © o/s, we have Kay (1— sK Jon) EE aad at O* Fate) +smG—K) +08 where we introduced the error term 2 nie a IAGeo)i Equation (4.112) can be manipulated into the following form by dividing numerator denominator by 1 + €: : (1—sK/a) 4s (SE) +( lt+e 1+eé 1 As indicated, this expression describes the realized filter. The notation shows that the pole frequency for is decre: T(s)= fox = -§)=f- 4h =K d correctly, H. = K, but T(s) has acquired a zero on the positive axis at § = @/K. Because Op must be finite and positive, Eq. (4.105) requires K_< 3, zero is, therefore, too far from the operating frequencies of the filter to cause noticeable in the transfer behavior. Let us interpret these results: Notice from Eq. (4.113) that € < 9/ |A(jao)| (€ < 0.061 an LM741 at 10 kHz). In Example 4.7 with K = 1.59, this frequency error is A fo/fo © —1 and the error in quality factor is AQ/Q * +14%. We can state, therefore, that the Sallen- filter provides a good lowpass filter with very small «-caused deviations for low values 4.5 Other Biquads 167 Q. We mentioned already that alternative realizations are needed for high-Q circuits, such as Bandpass filters. Grad gard rae. Changing a Lowpass Filter to a Highpass Filter: The RC-CR Transformation ‘We shall discuss in Chapter 9 a general process of frequency transformation that will allow us to convert a given lowpass specification and design into a highpass, bandpass, or other type of filter. The RC-CR transformation is a simple special case of the general method suitable for active RC filters. We shall discuss it here specifically for the Sallen—Key circuit although it is applicable for other RC—opamp configurations. Consider the Sallen—Key lowpass circuit in Fig. 4.31a described by Eq. (4.100), repeated here for convenience as 7j,(s) in Eq. (4.117): V Tis) = i EGG 4.117) i POC: +81G Gi + Gy + CG W+Gr Now let us replace the conductances G, in the lowpass circuit by the capacitors C), and the " capacitors C, by conductances G,, respectively, G7, ie, C=G,=1/R, (4.118) Cr> Ge, ie, Ge=1/Re=C Amplifier gains are not affected, that is, the parameter K and, therefore, the resistors Ra and Rg are not changed. The result of this transformation is the circuit in Fig. 4.33 with the highpass transfer function V> KS°C\C; Tus) = 2 ae = ao (4.119) Vi GiG2 + s[G2 (Cy + Cz) + Gia — K)]} + 8° Ci Cy This equation can be obtained by analyzing the circuit in Fig. 4.33 or directly from Eq. (4.117) by replacing G, by sC), j = 1,2, and sCy by Gy, k = 1,2. Comparing the result with Eq. (4.117), we se8 that the corresponding coefficients have the same numerical values because of the transformation (4.118). More specifically we recognize that the dc gain of the lowpass and, Figure 4.33 Sallen-Key highpass circuit obtained by the RC-CR trans- formation from the lowpass in Fig. 43la. 168 SECOND-ORDER LOWPASS AND BANDPASS FILTERS the high-frequency gain of the highpass are K,, that the pole frequency w in both functi given by van "s cics as in Eq. (4.102), and that Q in both functions is identical, determined by the expt brackets. Let us still observe that the transformation (4.118) as written assumes a no frequency parameter, s/a, and, therefore, normalized elements. The physical element are obtained by denormalizing the frequency by wp, that is the highpass elements Ryp and are obtained from the known lowpass elements Rip and Cp as follows: I and Chup = Riu > . eoCj.ur moRip Any components that do not affect the frequency parameter and enter the transfer only as a ratio, such as Ra and Ry to set K’,, are not changed. Note that we may still a convenient value for C and choose identical resistors R = 1/G as in the lowpass fi obtain from Eq. (4.119) Ks? rj ui $s? + sa — K) +03 where (a as in Eq, (4.105). An example will illustrate the method. Design a Sallen-Key lowpass filter with fy = 1.25 kHz and Q = 5. The de gain is specified. Use the RC-CR transformation to design a highpass with the same specifications pole frequency and quality factor. LM741 opamps are to be used. The process is identical to the one in Example 4.7. Choosing C = 0.1 yF results in 1 a ki = 1.273 2, 2m x 1.25 x 10** x 1077 Rp = RaQ = 1/5) = 2.29 kQ R=R Figure 4.34 shows the lowpass and highpass circuits and their performance. The de gain of COND-ORDER LOWPASS AND BANDPASS FILTERS. Since the RC-CR transformation did not change the filter poles, the earlier wa the sensitivity of Q to the gain K is still valid, as are the discussion and results on finite opamp gain. Observe further that if a highpass with smaller gain is required, analogous to the one used for the lowpass in Fig. 4.31b can be used. The differenceis we have a capacitor in series with the input so that we obtain a capacitive voltage elements aC; and (1 — a)C. For example, if in the previous example a high- of 0 dB is specified ie.,a = 1/K = 0.357, the C = 0.1-P capacitor will have into a series capacitor of 35.7 nF and a shunt capacitor of 64.3 nF. 4 4.5.2 The Single-Amplifier Biquad (SAB) A useful bandpass circuit using only one operational amplifier was developed by T. (1968) and J. J. Friend (1970). It is shown in its bandpass configuration in Fig. 4 capacitors are labeled C because these filters are normally built with identical can determine the circuit's operation by writing a node equation for the inverting input of the opamp, and one for the node labeled V,. The two equations are (2sC + Gi) Ve = aGiVi + SCV2 + sCV_ (sC + Gx) V_ = sCVx + G2V2 with -h/A We have again assumed finite opamp gain A to be able to investigate later which effect have on filter parameters with no need to repeat the analysis. For an ideal opamp, V_ the terms multiplied by V_ will simply be absent. To solve these equations for the function V2/ Vi, we solve Eq. (4.125) for V_ and insert the result into Eq. (4.124). We 1 sC+G2 Vy =-—= (G2. + = ) v; o(a+ 2) and Figure 4.35 Delyiannis-Friend circuit. 4.5 Other Biquads 171 This equation gives us the transfer function as V2 _ Ns) A saGi/C ye ee eee ee 127 TO= T= De a 2 GUC) Guts ™ Ge lee eo The parameter a to set the gain is obtained by a feed-in voltage divider in the same manneras for the Sallen—Key circuit in Fig. 4.31b. We observe that the first effect of finite opamp gain is the multiplying factor A /(1+A) {This is simply the gain ofa voltage butfer, Eq, (2.82), and can be neglected for frequencies less than the opamp’s unity-gain frequency, f < f,)The secondeffect is determined by the term (G;/C)/(1 + A) in the denominator. We shall investigate its conse- quences later and assume for now an ideal opamp, i.e., we consider the ideal transfer function saCG, saGi/C T(s) Ripa eae = 2 8C428CG1 + GG, 8? +52G2/C + G\Ga/C (4.128) Letus rewrite Eq. (4.128) in the standard form with center frequency ay and quality factor Qo. V, H T(s) = 2 = SH o/ Oo) _ (4.129) Yu 5° +5 (wo/Qo) + 0 We can then identify the filter parameters as R > Ro 2aQ3 (4.130) Conversely, we can also express the element values directly in terms of the filter parameters. From Eq. (4.130) we find the design equations as follows: 1 R H RR=—=5, = =40?, and a=—, (4.131) wc?” Ri 25 so that for a chosen value C we have 1 R. H : R2=200—, 2 =—-3,, and a=—; (4.132) pC 45 205 Note that we require H < 2Q3 because a is defined to be less than-1. Design a Delyiannis—Friend bandpass circuit with fp = 12.5 kHz, Qo = 10, and midband gain H = 26 dB. Use an LM741 opamp. BF sotution | On a linear scale, the gain equals H = 19.95; choosing C = nF, we find from Eq. (4.132) 172 SECOND-ORDER LOWPASS AND BANDPASS FILTERS 20 Ry = ———_ = 50.9k2, R= 3.927 x 10~ and Ri/a=1.27kQ, Ri/(1—a) = 1412 The circuit and performance are shown in Fig. 4.36. Compared to the specified measured center frequency and quality factor are fy = 11.6 kHz, reduced by 7.2 design value, and Qo ~ 10.8, an increase of ~ 8%. The gain is H = 26.2 ¢B. O} Figure 4.36 Bandpass circuit of Example 4.9 and test performance. (Bode Plotter scales: 5 to 25 kHz; 0 dB to +30 dB; cursor at 11.6 kHz, 26.1 dB.) Q-Enhancement Notice from Eq. (4.131) that the component ratio Rp/R, in the Delyiannis equals 403. Thus, implementing large quality factors with thi of the required large resistor ratios. To alleviate this difficulty, we should have a 4.5 Other Biquads 173 ‘quality factor. This feat is accomplished by using a small amount of positive feedback as shown in Fig. 4.37. The analysis proceeds exactly as for the previous circuit; the difference is that we now have V,. # 0, so that instead of Eq. (4.126) we obtain V2 V=-—+V, (4.133) atM (4.133) Inserting this expression into Eqs. (4.124) and (4.125) results in (2sC + Gy) V, = aGiVi + sCV2 + sC (Vy — V2/A) (4.134) (sC + G2) (Vs — V2/A) = sCVe + G2V2 (4.135) From the circuit we derive V, = KV2 (4.136) Next, we insert Eq. (4.136) into Eqs. (4.134) and (4.135) and eliminate the internal voltage V,. After a little algebra, we obtain the transfer function ney = [ A ] saGiC Vi 2 6 | Trad k)A ae ~_ SCG (K = 1A) 80? + 2sCGa — =<“ + iG eS ee ““L1T+0=K)AJ 9, 2G2 G, K-1/A GiG2 : 45-3 (|-26-7-ea A 2G2.1—-K+1/A Cc Note that Eq. (4.127) is a special case of Eq. (4.137) obtained for K = 0. Let us use the standard notation and the parameters introduced in Eqs. (4.129) and (4.130), and again assume for now that we are dealing with an ideal opamp, || = oo. This yields ( 1 ) sHwo/Qo $7 + $a (1 — 20.05) /Qo + 0 T(s)= (4.138) 1-K where we introduced the parameter Figure 4.37 Delyiannis-Friend circuit with’ Q enhancement. 174 SECOND-ORDER LOWPASS AND BANDPASS FILTERS toachieve a more convenient notation in the following derivations. Equation (4.138)ist we sought. We see that the center frequency w has not changed from that of the Friend circuit, but that the quality factor is now enhanced to Qo o= me where Qo was the quality factor without positive feedback. The midband gain Q-enhanced bandpass is also increase -d; it has the value a Awe yu Man 8 1 o ee lm —2aQ31—K Qo1 The design of this circuit with prescribed values of wp, Q, and Hp proceeds as First we have to select a value Qo for the quality factor of the Delyiannis—Friend ci Q enhancement. Qo determines the resistor spread by Eq. (4.131).fhere is a freedom in picking Qo, but a study (Schaumann et al., 1990) of the circuit's sensitiv opamp’s gain-bandwidth product and to the passive component values shows that an choice is ) Q 1.5 We shall choose this value for our designs. Next, we find the parameter a from Eq. sai ('-@) With Eq. (4.139), this gives the design value _ a ~ lta which determines the amount of positive feedback in Fig. 4.37. From Eq. (4.141) wel H =H, Ba “ey and the remaining element values, along with a choice of C, are obtained from Eq. (4 R H and =a, 4Q5 205 We still note from Eq. (4.140) that for 2aQ? — 1, Q is a fairly sensitive function of the tap position K. By varying K,, Q can be tuned easily to the desired value. K is to increase Q; if Q needs to be lowered we reduce K. a9 is not affected by this rises and falls with Q. Let us consider an example. 1 R.=200—, R= 2 20506 1 4.5 Other Biquads 175 Repeat the design of Example 4.9 but with a Q-enhanced Delyiannis—Friend bandpass circuit with fo = 12.5 kHz, Q = 10, and midband gain Hp = 26 dB. Use an LM741 opamp. On a linear scale, the gain equals Hx = 19.95; we choose again C = 5 nF. First we pick Qo = 1.5 as suggested in Eq. (4.142), and compute from Eqs. (4.143) and (4.144), sp (1-2) me 20; QO) 45 With these numbers, 1s a 0.189 1-— } =0.189 = — =—— =0. ( wa) 89 and = Fa = Lago = 21? = He 21 — K) = 19.9520 — i Hate K) = 19.9557 (1 ~ 0.159) = 2.517 so that from Eq. (4.146) we find 3 3.927 x 10-* 7.64k2, Ri = ‘ and Ri/a =1.52k2, Ri/(1—a) 1.93 kQ Finally, we choose R. Note that this resistor loads the opamp; it should not be chosen too small. Let us pick R = 10.0 k& and tap the resistor at KR=159kQ, [(1—K)R =8.41 kQ] The performance is shown in Fig. 4.38. Compared to the specified values, the measured center frequency and quality factor are fy = 12.2 kHz, reduced by 2.4% from the design value, and Qo © 10.17, an increase of © 1.7%. The gain is H = 26.2 dB. F] Figure 4.38 The enhanced- @ bandpass of Example 4.10 and its performance. Note the reduced center frequency error compared to that in Example 4.9. (Bode Plotter scales: 5 to 25 kHz; 0 dB to +25 dB; cursor at 12.2 kHz, 26.2 dB.) @) 170 SECOND-ORDER LOWPASS AND BANDPASS FILTERS Figure 4.38 Continued The Effect of A(s) on the Single-Amplifier Biquad We notice that the center frequency and quality factor again have noticeable errors with an implementation based on an ideal opamp. Let us, therefore, reiterate thatitis to consider the effects caused by real opamps before we can hope to arrive at a design. To help in this task, we have in Eqs. (4.127) and (4.137) already presented functions T(s} including their dependence on A(s). Consider Eq. (4.137): If integrator model for A, Eq. (2.18), it should be clear that T(s) is a function of Rather than attempting to find the poles and zeros of that expression, let us be an approximation to give an indication of the errors. We saw that the dominant effe A-dependent term in the denominator D(s). Using the standard notation with found in Eq. (4.138) that D(s) equals =s?+5% (1-292 KoA | D(s)=s +522 (1 2Q0T KS IVA )+ea We recast this equation into the form 1=1/(KA) 2 i+ ia — a oe where we used Eq. (4.139). Although the frequency dependence of this expression difficult to evaluate, we can make it manageable by making reasonable approxii us assume that |(1 — K)A(jw)| >> 1 and make use of 1/(1 +x) © 1—x forx <1. obtain Ds) = +s [1-205 Oe ote | Vl 2 D(s) F458 {1 20} (1 x)! ama} + 5 (e+e)]}+8 In this equation we used the opamp integrator model and we further neglected the term: proportional to 1/A2. If we collect the s?-terms we obtain ~2 4s {1~20%¢[1 - Ds) ¥ 81 +e) +55. (1-203) + w3 0 4.5 Other Biquads 177 _ where we labeled the error term ¢, ms 200 @ £2004 Gel ~ Jao KO —K) AGao)l G— KP ; With Eq. (4.149), the transfer function T(s) of Eq. (4.138) becomes (4.150) @ sH— Qo. ste) +52 (12030) +03 Qo Because K < 1, the first factor is approximately equal to 1/(1 — K) over the frequency range of interest, so that T' can be rewritten as @ es Tose! "Oo +6) =a, + oaree to 20a) + H aoe eu "T= QoVi+ 2, wo/Vi+e 2, { a Pe oaltee ia 20) sHpwon/ On i see eae 4.151) 8? + se@or/Or + oR Mhis result indicates that the center frequency a is reduced, and the quality factor has increased er its enhanced value Q in Eq. (4.140); the realized parameters are Cy Qo x : = _vite=Ovi+e" 4.152) wa oP coi 1t t= Ovi? (4.152) ind the realized gain is approximately unaffected by the opamp, Hg = Ha of Eq. (4.141). For for the design parameters in Example 4.9 we have from Eq. (4.150) _ + 2Q0 Te 1 iA Gaol @— Ky? 1500/12.5 1 — 0.159)? = 0.035 at is, the frequency error equals — 1.7% and the Q error is +1.7%. In the Delyiannis—Friend cuit with no Q enhancement, we had K = 0 and Qo = Q = 10; the error then becomes ee ~ |AG@o)|— 1500/12.5 17 nd the errors would be equal to —8% and +8% for frequency and Q, respectively, as s observed in Example 4.9. We note, therefore, that Q enhancement brings two notable ges for the small price of two resistors. The component ratio is substantially reduced om 400 in Example 4.9 to nine in Example 4.10) and the errors in frequency and quality

You might also like