Bounding, clamping techniques improve circuit performance Ci yeuit configurations that limit amplifiers’ input or output excursions can yield improvements in symmetry and transient response and simultaneously protect circuits from overdrive damage. Robert Pease, National Semiconductor Corp By applying the bounding and clamping techniques described in this article, you can reduee saturation delays arising from overdrive and prevent damage to your amplifier cireuit or its sueceeding stages. ‘The techniques also allow you to obtain funetions that are otherwise difficult to realize—eg, absolute-value (preci- sion-rectifier) circuits and high-symmetry clamps. Sometimes it's necessary to use a feedback bound if you need prompt response from a cireuit. Consider, for example, Fig a’s integrator cireuit without the zener feedback bound (shown by dashed Inne). If the input Fig 1—A simple, unbounded integrator circuit (a) can suffer from long delays when recovering trom ampliior Saturation—trace C in (b). Adding the bounding-diode (dashed-line) circuit eliminates the dolay (race B). EDN NOVEMBER 10, 1983 voltage is near +10V, the output can hit its negative rail at —13V. When the input next goes negative, you might expect the circuit to start immediately integrat- ing upward. However, as Fig Ih shows, the output response can be delayed because the amplifier has been “out of control” so long: The feedback eapacitor charges to 28V (not just 18V); the higher voltage translates to an extra 16-msee delay before the output can begin to move In some servo or control-loop systems, this delay can degrade performance to an intolerable level. However, you can avoid the delay by using the feedback clamp. ‘The zener-diode pair allows the op amp to retain control VERTICAL | HORIZONTAL TovibIV. | 10mSECIDIV sviplv | 10mSECIDIV sviplv | 1omsecrovSimple bounding degrades speed Fig 2—Bounding an inverter's output to +9V, this feedback configuration is Yar rom ideal. The zener diodes high capacitance considerably degrades the stage's ran slent response—trace B in (b). at all times; therefore, as soon as the input goes negative, the output rises. Indeed, the op amps output never reaches —13V, because as soon as it attains approximately ~10.7V, a zener diode conduets and holds the summing junction at OV. Any excess current fed to the input does no harm; it’s merely shunted off by the feedback bound. Feedback bounds can help in other areas also. Imagine a situation, for example, in which an inverter cireuit must drive a load from OV to at least =8V, but damage or overload considerations dictate that the ‘output not exceed +9V. A pair of ideal zener diodes, implemented as the Model IN4738 diodes are in Fig 2a’s cireuit, would do the job, Real zener diodes, however, are far from ideal; even the best of them have high capacitance. Because the N4738's 45-pF junction capacitance acts as a large damper, it degrades the unbounded cireuit’s fast response (Fix 2b. trace A), slowing it down considerably (trace B). You can avoid this undesirable slowdown by using the circuit in Fig 3. Here the feedback bound is not merely a pair of zener diodes, but a network of zeners and signal diodes. When the op ‘umnp’s output Cries Lo slew rapidly, a cupacitive current flows through the zeners; however, it flows through Rs to ground instead of through D; or Dy. This network’s speed performance is vastly improved over that of Fig 2's circuit. In fact, you'd be well advised to add the 2-pF feedback capacitor to avoid overshoot and ringing. Fig 3's circuit does not give perfect: performanee, however, because zener tolerane- ‘es can cause the positive and negative limits to be mismatched by as much as 0.5V ur even by 10%. Fortunately, the next cireuit (Fig 4) yields inherently high symmetry, along with other benefits. 8 VERTICAL | HORIZONTAL “SviDIV | 5 #SEGIDIV When driven to its positive limit, the ontpnt voltage is Vrooxy+Vixox+Vis- At the negative limit, it's Vonoxy+ Vowow+ Vis. Inasmuch as the dominating term (ener voltage Vps) is the same for both expressions, and the signal diodes’ forward voltages tend to match within 10 or 20 mV, you can see that this clreult’s bound symmetry shows improvement by a factor of 10 to 20 compared with the basic 2-zener circuit, Note that when Vour is +8.5V, Ds is definitely ON, but D, is just preparing to conduct. Thus, the low Dy Dy = LOWLEAKAGE 0A FAST-RECOVERY TYPES, DEPENDING ON APPLICATION Fig 3—This configuration eliminates the zener diode's capacitive effects by shunting capaciive currents to ‘ground. EDN NOVEMBER 10, 1983.Circuit tricks nullify zener capacitance leakage of D, and Dz is responsible for the circuit's excellent linearity and accuracy in its linear region. For this reason, a high-leukuge diode such as a INST is inappropriate, but a collector-base junction is a wise choice (see box, “So you say you need a diode?”), Resistors Ry and Ry ensure that Dy is always biased ON for all conditions at the input or output. Improve the integrator Did the addition of zener diodes to Fig 1a’s circuit fully solve the integrator’s problem? Not quite: The design of zener bounds is not so trivial. Consider, for example, Pig 5a's configuration, Assume the output is driven near its limit. When the input signal goes to zero, the integrator’s output should hold at +10V, but instead it droops down a full volt because of zener leakage (b). Close inspection of a typieal 10V zener’s So you say you need a diode? spec sheet reveals, for example, that I-A leakage at 7.6V is acceptable. Considering the 1-pA max at 7.6V, a leakage of 10 to 100 pA could exist at 8.5 or 9.5V (because of a soft knee)—and this high figure would not be a reason for rejection by the zener's manufacturer. Leakages of this order, however, can cause gross errors in a simple bound circuit, Fortunately, you ean also use the high-performance bound cireuitry of Fig 3 or Fig 4 with the integrator. With these substitutions, the integrator performs with excellent accuraey to within 0.5V of its limit without leakage problems. This statement is valid even if the zener diode is leaky or has a soft knee, because the leakage flows through R; (Fig 3) or Ry, Re (Fig 4). Thus it can't flow into the summing point and cause errors. ‘You can, therefore, use standard, low-cost components, When you need an appropriate diode for a critical application, you're often faced with a difficult choice. Diodes have combinations of characteristics, 30 it's neces- sary to consider the particular parameters that are crucial to your circuit. The nearby table gives salient characteristics for fast-re- covery and low-leakage diodes, as well as home-brew diodes made from transistor junctions. Note, for example, that the ubiquitous 1N914 (or 1N4148) features fast recovery, but its relatively leaky. A 2N4250's col- lector-base junction, on the other hand, has extremely low leakage, but slow recovery time. The same device's emitter junction, mean- while, offers very fast recovery, but will withstand only 6V and is moderately capacitive So you can see, your choice inevitably in- volves a tradeoff. The nearby figure (a) and (b) show typical hookups for _transistor-junotion diodes. What about monolithic diode arrays? Forget it—these diodes are likely to have excessive leak ages and capacitances to the substrate: moreover, when one diode conducts, it injects carriers in such a way that every other diode in the IC array develops a high leakage. Its a little known fact that slow, low-leakage diodes often not only tum off slowly, but they also turn on slowly. The photo shows the output voltage of Fig 11b’s circuit in the special case in which V2=0 (both zeners shorted), Riy=1 kA, D,=1N914 and D,=1N457. The 1N914 tums on and off very quickly. The 1N457 stores charges and conducts for 500 nsec after the input voltage falls to zero. It also fails to start conduct- ing until 40 nsec after its terminal voltage has risen above U./V. Fortunately, you can avoid these covershoots by using emitter-base junctions and 1N314s. Be warned that the overshoot shown is excessive mainly at repetition rates below 20 kHz—it actually shrinks and looks accept- able at frequencies above 800 kHz, so I's a particularly insidious and sneaky kind of overshoot. Note also that not all manufactur- ers’ 1N457s exhibit this phenome- non, so take care in qualifying your particular vendor. For extremely low leakage (for example, in Fig 13a’s circuit), consider using collector-base junctions. ff, for instance, you need leakage lower than 100 pA, then it’s often much less costly to test transistors with a 98% yield to this spec than to shop for a diode with low leakage but a high price. Some low-leakage transistors to consider are the 2N930, 2N3707 and 2N4250. However, most other inexpensive, small signal transistors have an excel- lent yield to a leakage spec of 30 pA max at 5V, even though the spec sheet guarantees 10 nA. ‘What about zener diodes? Typi- cally, 6.2V (and higher) zeners have crisp knees, low leakage and adequately low series resistance. ‘On the other hand, devices below 4.7V usually have soft knees, unacceptable leakage, high series resistance, high noise and poor stability. For low-voltage bound- ing, therefore, you'd be well ad- vised to consider "IC zeners”, such as the LM385-1.2 and -2.5 or the LM336-2.5 and -5.0 for 1.2, 2.5 and 5V needs; the adjustable LM385 suits use in 1.2 to 5.3V applications. EDN NOVEMBER 10, 1983in these two high-performance bound cireuits in a variety of applications. Adjustable bound is better You might have noticed that some of the circuits presented could have substantial temperature coeffi- cients (2 or 8 m¥/C), because of the sener-diode drift. Fortunately, you can use a mieropower voltage refer- ence—such a3 the LM385—to improve the TC (the reference’s equivalent circuit is shown in Fig 6a). Applied in Fig 6b’s cireuit, the device senses D;'s ON voltage, and uses this voltage to compensate for the forward voltage of D, and Dy or Dzand Ds, whichever is the ease. The resulting bound voltage has a low 0.2-mV/°C ‘TC and a reasonably low (400 at 1 mA) impedanee at any bound voltage from 3.75 to 6.8V. The LM385 itself is TYPICAL DIODE CHARACTERISTICS limited to 5.8V, but you could add an LM336-5 or similar 5Y referenee-diode’ IC to make, for example, a high-performance -+10V adjustable bound. Insert the device between points A and B in Fig 6b and adjust Rj for the desired output voltage. ly, if your application can’t tolerate leakage- induced’ errore at high temperatures, consider the high-performance cireuit in Fig 7. By adding a Giode/resistor stage to a normal bound cireuit, you ean obtain excellent, low-leakage performance at tempera- tures as high as 125°C. ‘The bound-current leakage is less than 1 nA at +10.4V output at all temperatures. In addition, the output is limited to less than *12V for ‘bound currents as high as I mA, Note that an LM11A is a good choice for this cireuit—its bias eurrent is less than 0.15 nA; moveover, its offset measures less than 0.12 mV over —55 to +125°C. RECOVERY |CAPACITANCE | MAXIMUM | LEAKAGE | SERIES we TIME ow Wocrmae | “Savior | resistance ines ansee 18 av 7a 260 (FAST: GOLD 0OPED) ‘nus OR ina aeons ape | ororw | sopen 20 WOW LeaKace) 1B ‘00NSEC iF aov eps 3a COLLECIOR BASE Soneton 200 NSEC apr | ontooov | apa ae ieuirTER open ave, OB EMITTER StREHOE! NSEC: oer stosv | y0PA as * » “ nog Diode characteristics vary wildly, as is evident in the table. Your choice almost invariably involves @ tradeoff involving speed, Cepecitence, vokage breakdown and leakage. 10 ofton wieo to uee trancicor junctions as dlodesa)’s two possible connections eld extremely low leakage and very fast recovery time, respectively. The photo shows a 1N487's long turn-on time in Fig 11b's Cireut. Trace A isthe circuit's input voltage, B isthe diode volage. TRAGE a 2viDIV o2vioI VERTICAL | HORIZONTAL. 50 NSECIDIV 50 NSECIDIV EDN NOVEMBER 10. 1953 zatSingle-zener network enhances symmetry 12 “FEEDBACK foro = ey mw: ovMak Fig 4—Improving on Fig 3's circult this feedback connec: lion eniiances syruinavy by a factor of 10 to 20 by using the ‘same zener diode fo set both positive and negative limits. ne Yap teens Ret — ay ssa si2a+ Bey SiOBES = LoWseaRace PES WOneT CAGE LEAKY ZENER Fig 6_Improve temperature oocttiolente by uaing an “IC zener’, (a) gives the IC's equivalent circut: (b) shows & TVPICALZENER ‘rout whose output limits are set by a resistor rato. av wv iy The bound etreuits presented to this point all have finite slope when overdriven. For example, when the input current increases from its normal 10 to 100 2A levels to 1 or 5 mA beyond the desired limiting input threshold, the output voltage can increase by as much as 80 to 100 mV. To obtain a very sharp knee in precision analog circuit, you might want to hold this error down to 1 mV. You can attain this low error, but at the eost of some speed anid complenity, Pig 8 shows a general-purpose limiter using two or more op amps. If sections A and B are connected, for example, Vovr will be —Viva or Vixe, whichever is more positive. Thus, if Vise is , the output will normally be —Visa, but it will be limited to ~8V as its most negative value. If, instead, you disconnect section B and connect VERTICAL | HORIZONTAL 2viply | 5mSECIDIV Fig 5—Zener-diode leakage degrades performance in this Circuit It can cause the output to droop a full volt more hegative, for example, than the theoretical Bounded TOV. (Overcome this problem by using Fig 3's or Fig #8 circut. 22 EDN NOVEMBER 10, 1983Notes: ROTTawsisrons = pnazeo ouND CURRENT a = 10044 =10yA = = 10008 ~ 1098 198 Fig 7—Eliminating leakage effects, this circu operates at temperatures as high as 125°C. Using a low-bias-current ample, It limts outputs to = 12V. Note the use of transistor junctions as low-leakage diodes, EDN NOVEMBER 10, 1983 288Complex network negates zener leakage onverringe = SecrIONA section INONINVERTNG) Noes: Fig 8 Featuring a very sharp knee, this circu limits output ‘change to less than 1 mV for inputs as large as several vos beyond the limiting threshold. The eircult also performs the absolute-vale (rectication) function seetion C, then Vour=—Vina or +Vixc, whichever is more positive. In the ease that both Viva and Vayc are comnected to the signal input, the cireuit performs a positive absolute-value (precision-rectifier) function, with exellent accuraey at frequencies below 1 kHz. What are the circuit's shorteomings? Frequency response, for one. Even with high-speed amplifiers and fast diodes, performance is poor above 10 kHz. In nase NOTES OP AMPS = LMasH OR Las Fig 9—Active feedback provides clean, sharp limiting. 1s the basic building block for Fig 10's complete circu set Posmive Feeback SELECTOR ser Y Neoarwve notes: AAUBIGDES = see oR 4a cHoOsE E, = cyt OP AMPS = (uSeAOR Luc4aon Lae? TRACE | VERTICAL | HORIZONTAL A TOVIDIV. | 100 »SECIDIV 8 sviolv | 100uSECIDIV osvioly | 100 uSECIDIV Fig 10—This variable-iimit, active-feedback configuration allows you to use either capacttve or resistive feedback to ‘configure an integrator or an inverter, respectively. noose (0p amps to suit your speed requirements. EDN NOVEMBER 10, 1988Reference ICs simulate ideal zeners addition, the method has the disadvantage of only providing a negative limit (or a positive one by reversing all diodes). On the other hand, it performs a useful function by selecting the highest of a group of voltages. For example, if you use four op amps and eight diodes, then you can obtain an output that’s the most positive of four inputs. What about active feedback? Intuitively, it would seem that you could make a high-precision limiter by using an op amp that senses when the output signal approaches the limit voltage, then reaches back and provides a feedback current to stop the outputs advance. The cireuit in Fig 9 is an example of such a scheme. It admittedly looks as if it'll oscillate and misbehave, but lab tests prove it doesn't. The complete version of the circuit is shown in Fig Wa, You can obtain clean limiting (b) for capacitive or resistive feedback (Zp), but the circuit won't handle both cases unless you use damping eapacitors (Cx) equal to Y4o the feedback capacitor. In order of increasing speed, you can use LM324, LM348 or LF347 op amps to develop a faster response. Outside the feedback loop Tf you want to limit your cireuit's output by clipping the input excursions, you can use a simple clamping scheme such as that shown in Fig Ia. However, the zener diodes’ leakage and capacitance ereate some problems in this cireuit. Fig 11b suggests a way to Fig 11—Clamp your op amp's inputs with these simple ‘ireuits. The configuration in (b) eliminates the effects of the Zener diodes’ capacitance. In choosing D, and D:, you must ‘consider a speedioakage tradeofl. Bounding and clamping: what and why? Bounding, clamping, limiting and there's no attenuation (and no eral seconds.) clipping—what do these terms current drain by the clamping © To avoid thermal errors mean, and what distinguishes network). When the signal ex: caused by an amplifiers them? They all apply to voltage- ceeds the preset threshold, how- output stage overheating in level restrictions, at either a cit- ever, the clamping network con- a limit condition uit’s input or output. As for their ducts heavily, preventing the To avoid degradation or out- differentiation, it's somewhat un- amplifier from being driven any Fight damage to an amplifi- clear—in fact, they're often used further. ers inputs. interchangeably for a given circuit Given this behavior, what rea- To avoid overdriving or dam- configuration. sons are there for limiting? There aging circuits connected to For the purposes of this article, are many: an ampiifier's output. however, “bounding” pertains to © To eliminate undue delays To obtain specific symmetri- limiting circuits inserted in an when an integrator must re~ cal or asymmetrical shap- amplifier’s feedback loop; “‘clamp- cover from its supply- ing. ing" to networks applied to the determined limit. © To obtain a nonlinear curve amplifiers input terminal. The for- © To eliminate undue delays for simulation of nonlinear mer generally produces a sharp when any amplifier circuit circuits (for example, using ‘output-clipping action; the latter must recover from its sup- analog-computer — tech- simply prevents overdrive. ply-determined limit, (Not all niques). Usually, an input signal is cou- amplifiers recover in a like @ Tocontrol anoscillator's am- pled resistively to a unity-gain manner—some chopper- plitude. follower's input. For small signals, stabilized designs take sev- To rectify an ac signal 286 EDN NOVEMBER 10, 1983