AMAZON multi-meters discounts AMAZON oscilloscope discounts
Oscillations are the ubiquitous bugaboos of analog-circuit design. Not only can you encounter oscillating op amps, as described in Section 8, but also oscillating transistors, switching regulators, optoisolators, comparators, and buffers. And, if you think about it, latched-up circuits are just the opposite of oscillating ones, so I included them here, too.
Recall the corollary of Murphy's Law that states: "Oscillators won't. Amplifiers will…" oscillate, that is. The knack of spotting and quashing spurious oscillations is, for some fortunate people, a well-developed art. But others have not learned this art well.
I obviously cannot tell you how to solve every kind of oscillation problem. But, I will give you some general principles and then notes on what can go wrong with various components, including comparators and buffers. This information, along with a few suggested procedures and recommended instruments, will get you off to a good start.
Here are some of the types of oscillations that can pop up unexpectedly:
Oscillations at very high frequencies--hundreds of megahertz--because of a single oscillating transistor.
Oscillations at dozens of megahertz arising from stray feedback around a comparator.
Oscillations at hundreds of kilohertz because of an improperly damped op-amp loop, an unhappy linear voltage-regulator IC, or inadequately bypassed power supplies.
Moderate-frequency oscillations of a switching-regulator loop because of improper loop damping.
Oscillations at "60 Hz" or at "120 Hz," or similar line-related frequencies.
Low-frequency oscillations coming from physical delays in electromechanical or thermal servo loops.
As these general descriptions indicate, the frequency of an oscillation is a good clue to its source. An electric-motor loop can't oscillate at 10 MHz, and a single transistor can't (normally) rattle at 100 Hz. So when an engineer complains of an oscillation, the first question I have is, "Oh, at what frequency?" Even though the frequency is often a good clue, engineers often fail to even notice what the frequency was. This omission tends to make troubleshooting by phone a challenge.
At very high frequencies, 20-1000 MHz, the layout of a circuit greatly affects the possibility of oscillation. One troubleshooting technique is to slide your finger around the circuit and see if at any point an oscillation improves or worsens.
Remember, knowing how to make an oscillation stronger is not worthless knowledge that information can provide clues on how to make the oscillation disappear.
I remember being very impressed when a colleague showed me that some of the earliest IC amplifiers had a tendency to self-oscillate at 98 MHz with certain levels of output voltage. Putting a grid-dip oscillator nearby caused increases or decreases in the problem, when its frequency was near 98 MHz. (Heathkit makes one of these instruments whose updated name is "dip meter"--see Section 2.) At that time I didn't have a 100-MHz scope, but I could see the rectified envelope of these high-frequency oscillations on a 25-MHz scope. So, if you see a circuit shift its DC level just because you move your finger near a transistor, you should become suspicious of high-frequency oscillations.
Of course, you will never "slide your finger around' in a circuit with high or lethal voltages.
One of the easiest ways to inadvertently cause a very-high-frequency oscillation is to run an emitter-follower transistor (even a nice, docile type such as a 2N3904) at an emitter current of 5 or 10 mA. In such a case, you can easily get an oscillation at a few hundred megahertz. So, although a good 100-MHz scope cannot spot this kind of oscillation, the resulting radiated noise can cause other circuits to go berserk and can cause an entire system to fail tests for radiated electromagnetic noise.
For example, when the first personal computers were being designed, designers needed a RESET function for their processor. Several designers decided (quite independently) to use the simplest, cheapest possible RESET circuit, as shown in FIG. 1. When they finished their designs and sent the prototype computers to be approved by the FCC, these designs all failed badly. Why? Because the little transistor would run at over 10 mA and, with a bypass capacitor at its base, the transistor would oscillate at a very high frequency. The frequency was so high that none of the designers noticed it, but as the transistor sprayed around a lot of RF energy up at a couple hundred megahertz, the FCC examiners noticed it, causing the computers to fail the tests for radiated RFI. They all had to go back and fix it. How? For such an emitter follower, a 50- or 100 ohm carbon resistor directly in series with the base of the transistor (and not 2 or 3 inches away) can cure this tendency to oscillate.
Sometimes a small ferrite bead is more suitable than a resistor because it will degrade the transistor's frequency response less.
FIG. 1. This was a "popular" circuit for the RESET function,
until the engineers discovered how badly it oscillated.
Oscillations Crop Up
Not all problematic oscillations are high-frequency ones. An unstable switching regulator feedback loop can oscillate at low frequencies. For troubleshooting switching-regulator feedback loops, I first recommend a network analyzer to save you troubleshooting time. A network analyzer facilitates taking data and checking out variations of the circuit in case of trouble. (However, I do tend to put more faith in real-time step response. If that is consistent with the frequency-domain response, fine; if not, I get suspicious... )
Secondly, if an earlier version of your circuit has worked OK, what's the difference between the new one that does not work well and the old one that does? Be sure to keep one or more examples of the old version around so that you can make comparisons when the new circuits have troubles. (Note that I said when, not if.) Thirdly, look for components such as capacitors whose high-frequency characteristics can change if someone switched types or suppliers.
Optoisolators in switching regulators are another possible cause of oscillation trouble due to their wide range of DC gain and AC response. A switching-regulator IC, on the other hand, is not as likely to cause oscillations, because its response would normally be faster than the loop's frequency. But, the IC is never absolved until proven blameless. For this reason, you should have an extra module with sockets installed just for evaluating these funny little problems with differing suppliers, variant device types, and marginal ICs. You might think that the sockets' stray capacitances and inductances would do more harm than good, but in practice, you can learn more than you lose.
When Is an Oscillation Not an Oscillation?
We still get a phone call, every month or so, from somebody complaining about a "120-Hz oscillation" on one of our circuits. (It's a good thing we do, because one of our applications engineers was mentioning such a case recently, and I realized I had forgotten to mention this type of oscillation, so this paragraph got plopped into the text at the last minute. If I hadn't remembered to include this class of "oscillation," I would have been terribly embarrassed.) Now, how can an op amp be "oscillating" at 60 or 120 Hz? Well, it is not impossible for an op amp or regulator to oscillate at this frequency, but it is extremely unlikely. What is surely happening is that there is some noise at power-line frequency getting into the circuit. There are four major ways for this to happen.
1. When there is a diode connected to a delicate input, the ambient light in the room can shine in and generate photocurrents. With fluorescent lights, this is usually at 120 Hz, but the higher harmonics can boom in, too, along with some DC current. As soon as you realize this is happening, it's fairly easy to troubleshoot this by adding a little darkness to the circuit-cover it up with a dark cloth, jacket, or book. You can then localize it and "darken" it permanently. In the case of extremely bright light, it can even come in through the insulators in the base of a TO-99 can-the little ceramic feed-throughs are not really opaque-they let a small fraction of the light in.
Fortunately, plastic DIP packages are very opaque, these days.
2. A power supply can have more 60- or 120-Hz ripple (saw-tooth shape or pulse shape) than you expect. This can be caused by bad connections, a bad capacitor, an open rectifier, or a ground loop. Again, as soon as you recognize that this kind of thing can happen, it's easy to search and cure the problem.
3. Magnetic flux from a transformer gets coupled into your circuit. The two most common sources are a soldering iron close by, or, a power transformer that is saturating a little bit, spraying flux around. This usually has a distinctive shape at 60 Hz with lots of harmonics, and is quite position sensitive. This, too, is fairly easy to recognize. But if you discover that your power transformer is not only running hot, but spraying flux badly, it's not usually easy to relocate it if you are nearly done with your project. I would love to recommend that you assemble your power supply in a little box, 3 feet away from your main instrument, but that is not always feasible. You should at least start out by installing only a power transformer of known quality, with good, known freedom from excessive external flux and saturation. Sometimes you can retrofit your circuit with a toroidal power transformer, but most people don't keep these lying around. They are more expensive but often worth it, as they are more efficient and have less self-heating.
In concept. you might try adding some shielding. Go ahead, put in 1/16-inch of aluminum. It will have no effect at all-for magnetic shielding, you need iron. Go ahead, put in 1/16-inch of iron. That's not much help, because at 60 Hz, it takes about 1¼-inch thickness of steel to do much good. You can try, but it's not an easy way.
Sometimes you can arrange your critical circuits to have smaller loops, so they will pick up less flux. Make neat, compact paths: use twisted pairs; and use layout tricks like that--those can sometimes help. If you haven't tried these before, ask an old-timer.
4. A mechanical vibration can be coupled in through a floppy wire or a high-K ceramic capacitor. If nobody tells you about this one, this is a very tricky tease, not at all easy to guess. Sometimes if you replace the high-K ceramic with an NPO, or a film capacitor, it will solve the problem. Recently we ran a picoammeter, and when the power supply lead ran near the summing point, there was a certain amount of charge, Q = C x V. When the wire was vibrated at line frequency, a 60-Hz current I = V x dC/dT flowed into the input. The current stopped when we guarded the 5-V bus away from the input, and we also added shock-mounting for the whole assembly, to keep out all vibrations.
There are probably a few other ways to get 60-Hz noises into a circuit, so you must be prepared to exercise ingenuity to search for nasty coupling modes. But if the "oscillation" is at exactly line frequency, and it synchronizes with the "line synch" mode of your scope, then it is certainly not a real oscillation. Now, I have seen 59-Hz oscillations, that would fool you into thinking they were at 60 Hz. but that is quite rare. It just goes to show that there are many noises to keep you on your toes. Some are oscillations, and some are "oscillations."
You can best analyze the design of a slow servo mechanism, such as that in FIG. 2, with a strip-chart recorder because the response of the loop is so slow. (A storage scope might be OK, but a strip-chart recorder works better for me.) You might wish to analyze such a servo loop with a computer simulation, such as SPICE, but the thermal response from the heater to the temperature sensor is strictly a function of the mechanical and thermal mounting of those components. This relationship would hardly be amenable to computer modeling or analysis.
Comparators Can Misbehave
Saying that a comparator is just an op amp with all the damping capacitors left out that is an oversimplification. Comparators have a lot of voltage gain and quite a bit of phase shift at high frequencies; hence, oscillation is always a possibility. In fact, most comparator problems involve oscillation.
Slow comparators, such as the familiar LM339, are fairly well behaved. And if you design a PC-board layout so that the comparator's outputs and all other large, fast, noisy signals are kept away from the comparator's inputs. you can often get a good clean output without oscillation. However, even at slow speeds, an LM339 can oscillate if you impress a slowly shifting voltage ramp on its differential inputs. Things can get even messier if the input signals' sources have a high impedance (>> 10 k-ohm ) or if the PC-board layout doesn't provide guarding.
In general, then, for every comparator application, you should provide a little hysteresis, or positive feedback, from the output back to the positive input. How much? Well, I like to provide about twice or three times as much hysteresis as the minimum amount it takes to prevent oscillation near the comparator's zero-crossing threshold. This excess amount of feedback defines a safety margin. (For more information on safety margins, see the box, "An Important Principle") I have never seen this hysteresis safety-factor technique outlined in print for comparators, so you can say you read it here first.
My suggestion for excess hysteresis is only a rule of thumb. Depending on your application, you might want to use even more hysteresis. For example, a comparator in an RC oscillator may operate with 1, 2, or 5 V of hysteresis, which means you can always use more than my "minimum amount" of excess hysteresis. Also, if you have a signal with a few millivolts of noise riding on top of it, the comparator that senses the signal will often want to have a hysteresis range that is two or three times greater than the worst-case noise.
FIG. 2. Stabilizing this heater's slow servo loop and choosing the proper
values for R1, C1, and C2 involved applying a 1V p-p, 0.004-Hz square
wave, Vset, to R2 and observing the LM11C's output with a strip-chart recorder.
Just the Right Touch
Comparators are literally very "touchy" components; that is, you can drastically alter their performance just by touching the circuit with your finger. And because comparators are so touchy, you should be prepared for the probability that your safety margin will change, for better or worse, when you go from a breadboard to a printed circuit layout. There's no way you can predict how much hysteresis you'll need when your layout changes, so you just have to re-evaluate the system after you change it.
For faster comparators, such as the LM3 11, everything gets even touchier, and the layout is more critical. Yet, when several people accused the LM311 of being inherently oscillatory, I showed them that with a good layout, the LM311 is capable of amplifying any small signal, including its own input noise, without oscillating and without any requirement for positive feedback. One special precaution with the LM311 is tying the trim pins (5 and 6, normally) together to prevent AC feedback from the output (pin 7, normally), because the trim pins can act as auxiliary inputs. The LM311 data sheet in the National Semiconductor Linear Databook has carried a proper set of advice and cautions since 1980, and I recommend this advice for all comparators.
With comparators that are faster than an LM311, I find that depending on a perfect layout alone to prevent oscillation just isn't practical. For these comparators, you'll almost certainly need some hysteresis, and, if you are designing a sampled-data system, you should investigate the techniques of strobing or latching the comparator.
Using these techniques can insure that there is no direct path from the output to the inputs that lasts for more than just a few nanoseconds. Therefore, oscillation may be evitable. Granted, heavy supply bypassing and a properly guarded PC-board layout, with walls to shield the output from the input, may help. But you'll probably still need some hysteresis.
For some specialized applications, you can gain advantages by adding AC-coupled hysteresis in addition to or instead of the normal DC-coupled hysteresis. (See FIG. 3.) For example, in a zero-crossing detector, if you select the feedback capacitor properly, you can get zero effective hysteresis at the zero-crossover point while retaining some hysteresis at other points on the waveform. The trick is to let the capacitor's voltage decay to zero during one half-cycle of the waveform. But make sure that your comparator with AC-coupled hysteresis doesn't oscillate in an unacceptable way if the incoming signal stops.
FIG. 3. This zero-crossing detector has no DC hysteresis but 50 mV of AC-coupled
hysteresis.
FIG. 4. When oscillations get nasty, you need a scope with pinpoint triggering
to help you study the problem.
Comparators Do Have Noise
Most data sheets don't talk about the noise of comparators (with the exception of the new NSC LM612 and LM615 data sheets), but comparators do have noise.
Depending on which unit you use, you may find that each comparator has an individual "noise band." When a differential input signal enters this band slowly from either side, the output can get very noisy, sometimes rail-to-rail, because of amplified noise or oscillation. The oscillation can continue even if the input voltage goes back outside the range where the circuit started oscillating. Consequently, you could easily set up your own test in which your data for offset voltage, V,,, doesn't agree withthe manufacturer's measured or guaranteed values. Indeed, it can be tricky to design a test that does agree.
For my tests of comparator V_OS, I usually set up a classic op-amp oscillator into which I build a specific amount of hysteresis and a definite amount of capacitance, so that the unit will oscillate at a moderate, controlled frequency. If you're curious, refer to Section D, which is not trivial.
Another way to avoid V_OS trouble with comparators is to use a monolithic dual transistor as a differential-amplifier preamplifier stage ahead of the comparator. This preamp can add gain and precision while decreasing the stray feedback from the output to the input signal. Refer to the example of a (fairly slow) precision comparator in LB-32 (Ref. 1).
Common-Mode Excursions Unpredictable
After curing oscillation, most complaints about comparators are related to their common-mode range. We at National Semiconductor's applications engineering department get many queries from engineers who want to know if they can violate comparators' common-mode specs. But they're not always happy with our answers. I guess the complaints are partly the fault of the manufacturers for not being clear enough in their data sheets.
By way of contrast, most engineers know well that an op amp's common-mode voltage range, V,,, is defined provided that both inputs are at the same level. This spec makes sense for an op amp because most operate with their inputs at the same level. But in most cases, a comparator's inputs are not at the same level. As long as you keep both inputs within the comparator's specified common-mode range, the comparator's output will be correct.
But if one input is within the common-mode range and the other is outside that range, one of three things could happen, depending on the voltages and the particular comparator involved. For some input ranges you can overdrive the inputs and still get perfectly valid response; for other input ranges, you can get screwy response but cause no harm to the comparator; and for others, you'll instantly destroy the comparator.
For example, for an LM339-type comparator running on a single 5-V supply, if one of its inputs is in the 0-3.5 V range, then the other input can range from 0-36 V without causing any false outputs or causing any harm to the comparator. In fact, at room temperature, the out-of-range input can go to -0.1 V and still produce the correct output.
But, heaven help you if you pull one of the inputs below the -0.1V level, say to -0.5 or 0.7V. In this case, if you limit the comparator's input current to less than 5 or 10 mA, you won't damage or destroy the comparator. But even if no damage occurs, the outputs of any or all of the comparators in the IC package could respond falsely.
Current can flow almost anywhere within the IC's circuitry when the substrate diode (which is inherent in the device's input transistor) is forward biased. It is this current that causes these false outputs.
We'll try to be more clear about V_cm specs in the future. Maybe next time at National Semiconductor, we'll phrase the spec sheet's cautions a little more vigorously.
In fact--Ta-da--here is the correct phrase from the LM612 data sheet : "The guaranteed common-mode input voltage range for this comparator is V- S V, 5 (V+ -2.0 V), over the entire temperature range. This is the voltage range in which the comparisons must be made. If both inputs are within this range, of course the output will be at the correct state. If one input is within this range, and the other input is less than (V- + 32 V), even if this is greater than V+, the output will be at the correct state.
If, however, either or both inputs are driven below V-, and either input current exceeds 10 PA, the output state is not guaranteed to be correct. And, this definition applies nominally to the LM339, LM393, and also to the LM324 and LM358 amplifiers if you are applying them as a comparator. So, you cannot say we are not trying to make our data sheets more clear and precise4ven if it does sometimes take 20 years to get it just right.
Still, if you stay within their rated common-mode range, comparators are not that hard to put to work. Of course, some people disdain reading data sheets, so they get unhappy when we tell them that differential signals larger than +/- 5V will damage the inputs of some fast comparators. But this possibility has existed since the existence of the uA710, so you'll have to clamp, clip, or attenuate the input signal--differential or otherwise--so the fast comparators can survive.
An Unspoken Problem
Something else that does not usually get mentioned in a data sheet is common-mode slew problems. The good old LM311 is one part that is otherwise very well-behaved, but causes some confusion when common-mode slew problems arise. But to some extent all comparators can have these troubles. If one input suddenly slews up to exceed the other's level, you may see an unexpected, extra delay before the comparator's output changes state. This delay arises because the comparator's internal nodes do not slew fast enough for its outputs to respond. For example, a 10-V step can accrue an extra 100-ns delay compared with the delay for a 100-mV step. And if both inputs slew together, the output can spew out indeterminate glitches or false pulses even if the differential inputs don't cross over. Be careful if your circuit has comparator inputs of this sort, yet cannot tolerate such glitches.
Come to think of it, I get occasional complaints from engineers along the lines of, "I've been using this comparator for years without any trouble, but suddenly it doesn't work right. How come?" When we inquire, we find that the comparators have been operating very close to the edge of the "typical" common-mode range, well beyond crowding the limits for years, the latest batch of comparators gives them trouble. Some of our best friends depend on us to have our parts meet those typical specs, and it's always painful for us to tell them that they really ought to depend only on guaranteed specs.
If you need three op amps and one comparator, can you use a single LM324? Well, op amps are not necessarily bad as comparators, but they sure are slow, and the LM324 is among the slowest. Not only is its slew-rate slow, but if you put in an overdrive of just 5 mV more than VOS, the output will respond at only 0.01 V/ p -- not even as fast as its specified slew rate. An LF35 1 or one-fourth of an LF347 will respond somewhat faster. So if you want to use an op amp as a comparator, you'd better need merely a slow comparator. (Note, however, that one LM358 plus one LM392 will give you effectively three-fourths of an LM324 plus one-fourth of an LM339, and the space taken by the two 8-pin mini-DIPs would be only 4% more than that taken by a single 14-pin DIP.) But, even so, some people do use op amps as slow, precision comparators. Even though op amps are generally not characterized as comparators, you can engineer such a circuit successfully. For example, the LM709 minus its compensation capacitors is a surprisingly competent, fairly quick comparator. But, please don't overdrive and damage the inputs.
Conversely, I am occasionally asked, "Can I put some damping capacitors on an LM339 and use it as a unity-gain follower?" The general answer is NO! because the LM339's phase lags are too squirrelly to be controlled by any possible compensation scheme. But I have used the slower LP339 and LP365 successfully this way, as a slow inverter or slow follower.
Even Buffered Circuits Can Oscillate
Any circuit that adds current gain can oscillate--even a buffer. Let's agree that a buffer is some kind of linear amplifier that has a lot of current gain. Some have a voltage gain around 0.90 or 0.95. Others have gains as high as 10 or 20 because their outputs must swing 50 or 100 V p-p-or more. Even emitter followers, which you'd expect to be very docile because their voltage gain is less than 1, have a tendency to "scream" or oscillate at high frequencies. So whether you buy a buffer or "roll your own," watch out for this problem.
Also, a buffer can have a high-frequency roll-off whose slope increases suddenly at 40 or 60 MHz and thus can contribute phase shift to your loop, back down at 6 or 10 MHz. You can beat this problem, but you have to plan. A buffer can also add a little distortion, which the op amp cannot easily cancel out at moderate or high frequencies.
Since buffers don't usually have a spec on this distortion, beware. Also, if you're running the output's quiescent bias current as Class AB, you must be sure that the DC operating current is stable and not excessive. You must set it high enough so that you don't get distortion but not so high that power consumption becomes excessive.
One of my standard procedures for stabilizing a unity-gain follower stage is to put feedback capacitance around just part of the loop ( FIG. 5). This circuit tolerates capacitive loads, because the buffer decouples the load while the feedback capacitor around the op amp provides local stability. Most unity-gain buffers, whether monolithic, hybrid, or discrete, are unstable with inductive sources, so keep the input leads short. A series resistor may help stability, as it does for the LM310, but it will slow down the device's response.
Many high-speed buffers have the chore of driving loads in the range of 50-150 ohm.
Driving these loads can require a lot of current, which leads to overheating. Plan your heat sinks carefully to keep the device from exceeding its rated maximum tempera-ture. Most buffers don't have any thermal shutdown feature, but the new LM6321s and LM6325s show that a good op amp, at least, can have such features designed in.
When using buffers to drive remote loads, be sure that the transmission lines or cables have suitable termination resistances on both ends to prevent reflections and ringing. If you can afford the voltage drop, it's best to put about 50 ohm between the buffer and its cable.
When your buffer provides a lot of extra voltage gain, you must make sure that the gain rolls off in a well-engineered way at high frequencies, or the loop will be unstable.
If the buffer-amplifier has a positive gain, as in Figure 9Sb, you can use capacitive feedback around the main amplifier. But when the buffer-amplifier has a gain of -10 ( FIG. 5c), you may want to apply a feedback capacitor from the input of the buffer-amplifier (the output of the op amp) to the noninverting input of the opamp.
In some cases, you can achieve stability by putting a series RC damper from the noninverting input to ground to increase the noise gain, but this trick doesn't always work.
Damping this loop is tricky, because there is so much gain stacked up in cascade. But the feedback capacitor to the negative input makes this safe and easy.
FIG. 5. Depending on the gain of the buffer, you can use these three schemes
to stabilize a buffered amplifier.
Nasty Latches
A circuit that inadvertently latches up presents a problem exactly opposite that of an oscillating circuit. Or, you could correctly say that a latched-up circuit is an oscillator with zero frequency. Although latched-up circuits demand troubleshooting, the good thing about them is that they sit right there and let you walk up to them and touch them. And you can measure every thing with a voltmeter to find out how they are latched. This state of affairs doesn't mean that troubleshooting them is easy, because sometimes you can't tell how the latched-up circuit got into its present state. And in an integrated circuit, there can be paths of carriers through the substrate that you can't "put your finger on." The worst aspect of latched-up circuits is that some are destructive, so you can't just sit there and let them remain latched up forever. Two approaches for attacking destructive latches are:
Turn off the power quickly, so the latched-up circuit cannot destroy anything. Try turning on power for short pulses and watching the circuit as it approaches the destructive latch condition.
Use an adjustable current-limited supply with zero or small output capacitance, (such as the example in Section 2), so when the circuit starts to latch, the fault condition can easily pull the current-limited power supply's voltage down.
Another way to inadvertently generate a latched-up condition is to turn on the outputs of your multiple-output power supply in the "wrong" sequence. Some amplifiers and circuits get quite unhappy when one supply (usually the positive one) turns on first. Automatic power-supply sequencers can help you avoid this problem. An anti-reversal rectifier across each supply can help, too, and is always a good idea for preventing damage from inadvertently crossed-up power-supply leads or supply short circuits.
I used to get calls every few months from people who asked me if it was okay to ship (or launch) products that contained LM108s that may have had +15 V on their -15 V pins and vice versa. It was always painful for me to tell them, "Don't ship it--junk it. And, next time put antireversal diodes on each supply." Specifically, you should add these antireversal rectifiers across each bus in your system to protect the loads and circuits. Also add an antireversal rectifier across each power supply's output to protect the supplies (FIG. 6). Some people think that leaving parts out is a good way to improve a circuit's reliability, but I've found that putting in the right parts in the right places works a lot better. Refer also to a running commentary and debate on this topic in Section 13.
If you have any doubt that your anti-oscillation fixes are working, try heating or cooling the suspected semiconductor device. In rare cases, passive components may be sufficiently temperature-sensitive to be at blame, so think about them, too. Even if a circuit doesn't get better when heated, it can get worse when cooled, so also take a peek at the circuit while applying some freeze mist.
My point is that merely stopping an oscillation is not enough. You must apply a tough stimulus to the circuit and see whether your circuit is close to oscillation, or safely removed from any tendency to oscillate. This stricture applies not only to regulators but also to all other devices that need oscillation-curing procedures.
For example, if a 47 ohm resistor in the base of a transistor cures an oscillation, but 24 ohm doesn't, and 33 ohm doesn't, and 39 ohm still doesn't, then 47 ohm is a lot more marginal than it seems. Maybe a 75 ohm resistor would be a better idea-just so long as 100- or 120- or 150 ohm resistors are still safe.
In other words, even though wild guesses and dumb luck can sometimes cure an oscillation, you cannot cure oscillations safely and surely without some thoughtful procedures. Furthermore, somebody who has an appreciation for the "old art" will probably have the best results.
FIG. 6. Installing anti-reversal diodes in the system power supplies and
also on each PC board greatly reduces the chances of damaging the supplies
or the circuitry with inadvertent short circuits or polarity reversals.
I certainly do not want to say that technicians can't troubleshoot oscillations simply because they don't know the theory of why circuits oscillate-that's not my point at all. I will only argue that a green or insensitive person, whether a technician or an engineer, can fail to appreciate when a circuit is getting much too close to the edge of its safety margins for comfort. Conversely, everyone knows the tale of the old-time unschooled technician who saves the project by spotting a clue that leads to a solution when all the brightest engineers can't guess what the problem is.
References
1. Linear Brief LB-32, Microvolt Comparator, in NSC Linear Applications Book, 1980-1990