Use, Abuse, and Misuse of Amplifiers

Bob Pease

Hello and welcome everyone to the lecture on the Use, Abuse, and Misuse of Amplifiers co-sponsored by National Semiconductor and Arrow Electronics. My name is Rick Timmons and I’m Senior Vice President of Engineering at Arrow Electronics, I am very excited to partner with National Semiconductor for this seminar. I’d like to introduce Bob Pease, Staff Scientist from National and Electrical Engineer Hall-of-Famer to this seminar, the Use, Abuse, and Misuse of Amplifiers.

Thank you, Rick. I’m glad to be here today. First thing I’ve got to do is introduce my colleague Paul Rako, -- that’s how you spell his name. He’s going to be handling the questions because there might be a lot of Q & A and we want to make sure we don’t get confused. And it’s a challenging job to run this computer and see which question we answer next. So, that’s what Paul is good for.

Now, your computer will let you ask me a question. So, somewhere on the front of your computer, when you’re watching this stuff, it will tell you where to ask a question; you can send them all in. There may be more questions than we can answer today; we’ll try to answer the questions of general interest. And, I’ll answer all the rest of them in a day or two as soon as I can. Some of them, of course, you may ask me a question that sounds simple, but I may have to ask you for more information. So, when we get back to you, we’ll try to get answers to everything. Don’t be afraid to ask me questions now or online or at the end of the talk or later or any old day.

Somebody already asked: “Can I see this Web cast later?” Yes, you can see it later. You can’t see it one hour after the show, but a few hours after the show, or for sure tomorrow, you can see this as the archive rerun of what we’re talking about today. And I am a real engineer and I have a real address, and if you need to mail me anything, that’s not too hard to do. A lot of people ask, “Gee, Bob, I wanted to see your recent column on X.” Well, the answer is it’s really not hard. National’s Website, which you already know how to find because if you didn’t know how to find it, you wouldn’t be there, is national.com/rap. You can go to places like ED Columns, or Recent Columns, and you can easily see the columns that you’re interested in. We’ve got way over 90 columns on the Web at this time. And some of my other Websites at www.national.com/rap are things like VBE stuff, Trekking, Horrible Pictures, Good Technical Stuff on Band Gaps.

Now, some companies have already outlawed PowerPoint presentations. Paul was just saying, “Larry Ellison at Oracle has forbidden PowerPoint presentations at Oracle.” If you can’t say it on a chalkboard with some colored markers, then you’re in the wrong job. So when PowerPoint is outlawed, only outlaws will have PowerPoint.

My next Trek will only be 79,000 feet of ascent and only 69,000 feet of descent because we end up higher than we start, sometime in October. We have a question here – please.

Q How good a mountain-biker must you be in order to trek Nepal?

That’s an interesting question. You don’t have to be a very good biker. I’m not a very good biker, but I had no problems. We only went up to, you know, 18,000 feet by mountain bike, and down the other side. It was exciting at times, but it was mostly easy riding. And it was a lot more pushing the bike than we thought.

Next comment. You’ve probably all seen about my book, which has been translated into German, Dutch, and German and French. I like that one: “One cut and it works. One cut and - - kaboom.” Also, it’s translated into Russian. I’m told it’s a very good translation, troubleshooting stuff. This is a book that’s good for technicians and it’s also good for engineers and it’s good for students and it’s good for teachers. And, unfortunately, they've raised the darn price a lot: 35 bucks, 36 bucks now. It’s a shame. This is the book where you can’t tell where my ratty beard leaves off and where my ratty breadboard continues. This is the book where the best part of it is the back page: “Fools you are when you say you like to learn by your experience. I prefer to profit by the mistakes of others and avoid the price of my own.” And I put that on the back of my next book, which is "How To Drive Into Accidents and How Not To." And pretty soon we’re going to get into the real stuff of op amps, but at first I said, We should really point out that: “the best model of a cat is a cat; preferably the same cat.” And that was Norbert Wiener who invented that one, that saying. I tend to agree with it.

Now, Bob Pease says, “How can I help you get your analog task done with less grief?” Well, two general rules are: Always use an appropriate CF feedback cap around your op amp, and always add appropriate bypass caps on your amplifier or on your circuit. Now, let’s go into A in a little more detail. On a general-purpose op amp, you should always add a small feedback cap, 20 or 200 pico farads, unless you can prove it is not needed. On a current feedback amplifier, you should never add CF unless you can prove it’s okay.

How do you prove it? Section AA-Prime. Don’t just look to see if your circuit is oscillating; bang on it and see if it rings. Hey, that’s pretty deadbeat. It didn’t ring. If it rings, you know it may not be oscillating right now, but it might be soon. So, watch out for that; don’t just see if the oscillation dies out. If it doesn’t die out darn fast, you’re close to trouble.

And B - power supply bypassing. Every group of one of linear circuits needs some ceramic and some electrolytic or tantalum bypass caps. Hot, fast amplifiers really need to get a few capacitors right next to the power supply terminals of that amplifier. And if you have a hot amplifier that’s faster than 2 MHz, 4 MHz, a data sheet will remind you, yeah, you got to add your microfarads. And if it’s a really hot op amp like 100 MHz, 200 MHz, it will remind you, you have to have the capacitor about a tenth of an inch maximum away from the power supply pins. If in doubt, bang on the thing.

Now, a friend of mine asked me awhile ago, “Gee, this Application note published in the magazine says, 'When you have an op amp, the op amp always has a lag.' The lag is sort of proportional inversely to the gain-bandwidth product. If it’s a fast gain-bandwidth product, the lag is less, but there’s always a lag.” Oh, heck. What do you do about that? He proposed, “You add a lag in the feedback, which cancels out the lag in the path so that two traces go up together.” Isn’t that wonderful? Hey, isn’t that nice? No problem with lag, except for, wait a minute, one little thing. Is it a good idea to put a lag in the feedback loop? If you have a 30 MHz op amp and you want to cancel the lag with the 30 MHz op amp, putting a lag in the feedback is not necessarily a good idea. I went out and got some of these op amps. Well, it’s not a disaster; it’s a caution that when you actually build it, you look and it shows you can see the ringing at about 22 MHz, with a Q of about 2. Maybe it’s Q of 1½; I don’t recall. Maybe it’s a Q of 3. But, this guy wrote in that you should use AD 829, or a similar high-speed video amplifier, to get less lag. And of course, when you’re doing fast conversion, data conversion, A to D converter, you don’t want to have the lag of the signal; you want to have the signal with minimum lag. And I explained this just a little bit to show this is the normal triangle wave that he puts in at a volt per microsecond or something or other, 10 volts per microsecond. And this is the normal summing errors, the normal lag, and it’s noticeable, it’s significant. And when you cancel it out with a similar or like amplifier, nearly identical amplifier, it all cancels out, except for one little detail.

Let’s look a little closer here. Look at the signal. I misrepresented it, but I didn’t mean to. It really does look like he says it does, no question. I didn’t mean to put a hook on there or a catch. But when you look at this signal, it really does overshoot and bounce and ring and ring for about 300 nanoseconds. So if you’re trying to do something with a realistic signal like 1 volt peak-to-peak and 5 microseconds rep rate, that’s 200-KHz. Your lag would normally be several millivolts; might be 5, 7 millivolts. And consequently, it overshoots that, it can’t help it, and then it goes back and it finally settles out with a half a microsecond ringing. Now, if you don’t mind that, you could probably use the circuit. You know, it probably won’t oscillate. But I’m nervous because I don’t like things that ring that badly. I think this approach would work better. If you've got to use a reasonably quick op amp, use an R and a C. This has a lag. You want to put a lag in the feedback. Let the capacitance here give you the immediacy so it don’t oscillate.

Questions? Ah, please. Which one is a good one?

Q Well, Sean wonders if we should physically bang on the part. I thought it was just a demonstration of mechanical -- resonance.

Ah, where do we have our basket of paper? Oh, here we go. Let's say we have a circuit like this, and we want to bang on it. So we’ll take the output and just put in a square wave. And, 0.01 and 10K; you bang on the output, and it would ring. I mean you take this output right here and put 10K in a square wave generator and the output would ring a little bit. So, that’s what I mean by banging on it. You don’t have to literally bang the table, although that’s pretty deadbeat. At least it’s not going, ding, ding, ding. And, with a circuit, you want it to die out. Now this circuit, with the R and the C, you could bang on this until it, sort of, it might overshoot just a little bit, but it settles out and does not give much problem. And, when you’re asking it to have low slew error, it would do pretty good at having the ramp error go away. Another question?

Q George wants to know if it’s preferable to use ferrites, but I’m not sure the context of that. Was that with regard to the -- perhaps you can clarify it with another question.

Well, yeah, I can’t answer that question right now. We use ferrites for certain switch mode regulators because ferrites are compact, they’re cheap, they’re wonderful for certain jobs; they’re awful for other jobs. But, we do use ferrites, and people use switching regulators and see that we recommend a particular type of ferrite. But, in general, there's a lots of places where iron is the right thing to use.

Q Someone wants to elaborate on the “lag” term.

Oh, God. Let’s see. If you have this op amp, this first op amp, you put in a signal. And the output lags the input. So you put in a step and the input goes up. The input goes up as a ramp and the output lags. It’s behind; it’s late. So, there’s a certain amount of a lag, and this is a certain number of nanoseconds, or whatever it is, it’s a lag. You put an RC circuit in there and guess what? It’s a lag. So, a lag is a well-known term. Lots of people like to use poles and zeroes and I don’t; I refuse to. I like to use lags and leads. And the lag is simply an RC lag or equivalent.

Q This is like the difference between time domain and frequency domain.

I’m not a big frequency domain man; I’m a time domain man.

Q I think I’ll will need to get back with him on that because --

Runt pulse?

Q Uh-huh.

With what circuit?

Q LM6762. Do you remember?

Absolutely not. We can’t answer a detailed question like that right now, but we’ll do it later.

Q Yeah, we'll look over it.

Sure. Except, there’s not enough information here. You can make any op amp give a runt pulse if you put in a runt pulse.

Q What is a runt pulse?

A runt pulse is a very small pulse, smaller than expected. Like you’re expecting a 10 MHz clock, which would be 50 nanoseconds per cycle. If you then have a 5-nanosecond pulse and that’s a runt pulse, and many systems can’t handle that. So, runt pulses do happen. But this guy who asked, “What is the LM6762 doing giving a runt pulse?” We do not have enough information to answer that today. But if you tell us more about it, we’ll try to answer it later.

Q What's the best way to parallel up to two amps to get more drive current?

The general question of two amps. We have lots of op amps that are going to put out four amps. You don’t need to parallel them. If you need to parallel op amps to get 8 amps, we know how to do that. What kind of signal is it that you want more of? Send us more information on the question; your question is a little un-informing. But I’m sure we can answer it, and you may not have to parallel a thing. One more question and we’ll get back to the --fun? You got any more questions?

Q Martin wants us to say more about ferrites, but in what context?

I love ferrites in switch mode regulators. I don’t use them for a damn thing else. And if you do, you may know a lot more about it. But, I can’t possibly answer every question about ferrites – enough. Okay?

Now, we’re going to get into the heart of this. In this application, misuse, wrong common mode range, overdrive inputs or outputs. We were just discussing this. How come the output does not respond when you have a signal going in at plus 6 volts, like 6 volts and 6 ¼ volts on a comparator? You take an old LM339 and nothing happens because it’s outside its common mode range. LMC660, nothing happens. LM324, nothing happens. We make a lot of amplifiers and comparatives that do work right up to the plus rail and a little bit beyond, so you can make a unity-gain follower that goes all the way up to 6.9 volts really nicely, but we make a lot of parts that don’t. So, if you use a part that don't have an input common mode range or an output swing of 6.9 volts, then don’t get mad. Just get a better part, a more suitable part, because people do ask us, “How come?” And the answer is, “Look at the data sheet, man.”

Now, here’s another topic. It doesn’t come up too often, but the op amps are getting so small and the output capabilities are getting so large you can drive audio signals; just enormous amounts. Now we make a cute little audio amplifier. I can’t even remember the part number. But it’s a MicroSMD part. And the true size of it is bigger than that spec, but it’s smaller than this spec by about a factor of four. And it will just push all over the place, and it gets warm and it doesn’t care; it’s fine. But a lot of op amps can put out a lot of current, they overheat themselves, and they’ll overdo it, lose accuracy; they may stop working. You could probably destroy some of them. So you have to avoid gross abuse and gross overheating of a tiny little amplifier that has no way to get the heat out.

Now, this small audio amplifier I’m referring to can put out enough noise, it won’t deafen you, but you could fill up this room; it could talk as loud as I can with a small efficient speaker. So there's lots of amplifiers that can drive wonderful outputs, but not all of them can do it without overheating themselves. This small audio amplifier that we’re making has built-in thermal limits, so it turns itself off when it gets too hot; it doesn’t mind. But, for example, LPC662, it’s a slow little amplifier; it draws what, a quarter of a milliampere per channel. But it can put out 5 milliamperes guaranteed, but it could put out about 50 milliamperes. And that’s going to get hot. If you’re running on four volts, it might not get flaming hot. If it’s running on eight, or 14 volts, it’s going to get hot, or 15 volts.

Another possibility is some amplifiers, if they’re small, can overheat with any kind of extra heating source, such as, you hit the rails with it, you drive the input, your output hits the rails, and the internal drivers can get too warm. How about a nice little op amp that’s small and it’s got about 100 MHz bandwidth, and you put a capacitive load on it, and the poor little dear breaks into oscillation and hammers as hard as it can to drive that capacitive load, and it gets hot. Please avoid that. Now, we have some excellent new op amps, which are rail-to-rail on the input, rail-to-rail on the output; it is bipolar, it's not CMOS. But it has almost perfect infinite tolerance of CL. You put a capacitive load on it; it slows down. You put more load; it slows down some more. But it doesn’t go berserk.

Now, one guy was asking just recently, “Gee, I put in V1; from a divider I put in V2. As I heat up this thermistor and V2 moves up and down, V1 moves up and down. How come?” Well, some of our amplifiers do have resistors and diodes in them. Now, a FET input op amp usually don't have those resistors. So, you usually don’t have any interaction. If you put a FET op amp in there, you wiggle this up and down, and this doesn’t wiggle at all. But with bipolars, it tends to wiggle because you have to protect the inputs. The PNP inputs may not need much protection from overdrive differential, but the NPNs do, so we have to have the resistors and diodes. And when this moves around a lot, this moves with it. It sort of resistively divides. So, when this voltage moves, yeah, that moves; if that bothers you - whatever.

Another guy said, “Your LM317, the output is oscillating at 90 hertz.” And I said, “The LM317 cannot oscillate at 90 hertz, but are you talking about the tiniest little LM317L in the smallest little package?” He said, “Yes, how did you know that?” And I said, “Well, it’s going into thermal limit.” And thermal limit often goes in and out of operation at about 90 hertz. So, I told him to go back to put a little tiny bit of heat sinking on it to get a little bit of heat out of the poor little thing, and it should be okay.

Next topic: Protecting against inputs. If you have ESD pulses, one, two, four, 800 volts or more, or 1,000 or 2,000 or 4,000, or more large signals, or 160 volts peak sine wave at 60 hertz, or more, that can really beat up your input of your circuit, whether it’s an op amp or any other part. Even if you have an op amp that’s rated for more than 2KV or four, there’s more than 4KV on a cold North Dakota day. When it gets dry, you can raise dozens of kilovolts of ESD, which can blast most circuits unless you put some resistors and stuff in there to protect it, such as usually resistors and diodes can do it. We’ll discuss the kinds of diodes and we’ll discuss the kind of resistors. I’ll be through with this in about two minutes and then we can go look at some more questions; how’s that? For example, what if you had a whole volt of signal? You want one volt here, 10 volts, here, one volt here. If you just put three diodes there, they’re going to start conducting a little bit, unless you put in say a 9k, 1k, divider. So, when this goes to plus one volt, plus one volt, plus one volt, plus 10 volts, plus one volt, this is bootstrapped. When this is one volt, it’s going to start to move, but this is still bootstrapped. So the capacitance is bootstrapped, the leakage is bootstrapped, these diodes will not leak very much, and you could make a circuit like that survive lots and lots of KV of ESD because it will all get clamped before it gets anywhere near the input of the op amp.

Oh, let’s see. Let’s talk about the kinds of diodes. With 1N914’s, they’re fast; they can actually take 75 milliamperes DC and a couple hundred milliamperes of transient pulse. But they are leaky, five, 10 nanoampers at room temp. Much worse if you ever let them get warm. So you normally don’t use 1N914s, unless you can tolerate one of the world’s leakiest diodes, although I’ve got to say this is leakier. Something like the 1N484 and 1N457, ordinary low leakage diode, you can’t even find them anymore. They’re low leakage but they’re not very fast, and they’re not very low leakage either. 1N4002, it’s amazing. This is a one-amp rectifier. What's it got for leakage? A few picoamperes? Nobody’s going to guarantee that, but you can easily go out and select something inside 50 picoamperes and you’ll never find a bad 1N4002. It will leak as bad as 50 picoamperes. So, for some applications, hey, you could put in some 1N914s, you could put in some 1N4002s. The 1N914s will turn on fast; these won’t. These will handle the ampere and these won’t. So, there's a whole bunch of funny things you can do with store-bought diodes. And we’ll also discuss in a little while the 1N5817, which is a Schottky.

This thing can actually carry the three amperes, and it can turn on in less than a nanosecond. Unfortunately, there are some cases where one nanosecond is not fast enough, and we’ll be discussing that later.

Now, we’ve been talking about diodes. The lowest leakage diode I know is an LED GaliumArsenide, or whatever, sealed up in pure darkness. And as the sub-picoampere diodes, it’s one of the better ones you can get. Although actually, the input diodes at the clamp of an LMC662, you can get four diodes for half a buck, and those are low leakage, too. Those are also sub, subpicoampere.

Now, if you take a 2N3904 and tie the base to the collector, and you look at the emitter versus that collector, that is a very good diode. It is extremely fast. It may be one of the fastest turn-on diodes I know. It turns on fast, it turns off fast, and it has very low leakage, subpicoampere also at room temp. Unfortunately, it can only stand off a few volts because the base-emitter junction does not stand off a lot of volts. Now, you can get a pretty good low leakage diode because your base-collector junction of that same transistor is good for about 50 volts. It’s very low leakage, but it is not very fast. It’s not very fast to turn on; it’s not very fast to turn off.

There’s one other thing you can buy if you really need blast proofing is a Transzorb, and I think they still make them, and they’re much better than zeners. They say, “Well, it’s like a zener, but it’s about 20 times more robust at taking sheer damage and current times voltage than any zener. You can get them in all sorts of voltages. I don’t know how much they cost these days, but there are applications where a Transzorb will protect you, and a lot of these circuits don’t quite protect you.

Q Bob, how about a JFET hooked up as a diode for low leakage? I’ve seen that in circuits.

There are JFETs that have low leakage, but they’re not a hell of a lot better than a 2N3904; those are subpicoampere, also.

Now, let me drag out of my book, which has a lot of junk information and a lot of useful information. It has Appendix E somewhere in here, and has a list of various diodes. And we got a Schottky rectifier. I’m going to go down this list. The Schottky rectifier is very high leakage. It’s got several micro-amperes. There are other Schottky rectifiers that will leak milliamperes, but they will soak up and absorb amperes, like that 1N5819, 5817, 5818, it’s all in the same family.

B: Piece of Germanium. Look how soft it is. It's soft as a grape. Now, F is sort of interesting because F is a big old 1N4001 family which doesn’t leak very much. It’s not high conductance, it doesn’t have a good slope. Here’s a good slope, but it’s a healthy old rectifier. Here’s a good slope, L. What is L? L is a 2N3904, base emitter junction. That is one of the steepest curves; it’s just about theoretical, and is almost as good, but that’s a LM3046, which is a much smaller device.

What make is K? Oh, LM194; that’s an even bigger device. So there’s a very nice clamp if you need a clamp; it’s about $2.00 per clamp. You wouldn’t normally waste your money. Whereas, L is very steep, M is pretty shallow, M is the collector base junction of the 2N3904. So the collector-base junction is really inferior at having a poor slope, and the base emitter junction is very steep; it has a very good slope. So this is one of the things I like, every time I could find my book, I can always find this information. It’s never going to go away. But I can always find it. It’s always useful to publish some things that you can find later.

What if you have large signals? If you have plus or minus one volt, you can protect it with clamp diodes to ground, but for plus or minus 10 volts, or five volts, you don’t want to fool around with that. So you clamp it against the supply. What diode do you need? Like if you’re trying to go real fast, the 1N914 might be one picofarad; it might be appropriate. How much blast proofing do you need? You have to engineer it. There’s no simple answer. You could put low leakage diodes with 10K, 10K, 10K; low leakage, not much delay, or you might put a meg, a meg, a meg. It’s not a big deal. There are many things you can do. So, what ”R” values? Whatever you need, you have to engineer it, 1K, 10K, a meg. Why would we put two 10K resistors in series? If you blast this with ESD sparks, you’re walking across the rug, and you reach over and you touch the input to your little amplifier, then you buy a new amplifier because you killed it. Well, to avoid that, you could put 10K in series with 10K. And if you put a couple thousand volts across 10K these days, 10K tends to break down across its spirals because most resistors these days are spiral-cut with thin-film, thick-film resistors. You used to be able to buy, for a penny, 10K Allen Bradley quarter watt, or half watt. You put a couple of them in there and they do not break down. And they can survive large KV without sending through huge shoot-through currents.

Q Because they’re carbon composition.

Carbon composition, Allen Bradley. They’re hard to get these days; I don’t know if you can get them at all. People say, “Well, our one percent film resistors, our one-penny film resistors are better.” Well, they ain't better for everything. They’re inferior for protecting your devices. So, here I draw an example with a couple of resistors, or maybe four you could put in series because if you really have bad ESD, four in series may help you survive it. Otherwise, the current just short circuits through here and blows out your diodes. Diodes do not have an infinite amount of ESD standoff. Put a good 20KV blast on the diodes and they may get unhappy. And I also show a resistor here because if you pull this a couple of volts above the plus supply, the op amp might become unhappy. Even if you only do it for 100 nanoseconds, it might be okay and it might not, but it’s better to put 10K here to protect the input so you don't get many milliamperes into the input. Often, you really need all these things engineered, and sometimes you don’t. Sometimes this is overkill.

Here’s another application where we bootstrap it. This has to swing plus or minus 10 volts, the output swings plus or minus 10 volts, whatever it has to do, and what if this diode leaked? Well, it only leaks to a voltage the same as the input, which is the same as the output. There’s no voltage across here to speak of, maybe one or two millivolts. So, for a normal operation, you get no leakage from the diodes, but when you blast it through 50 volts or 1,000 volts, it goes right through this diode and right through this diode, and you’re more or less protected. You can usually survive with this circuit, things you can’t do without that circuit.

Now, I saw a mailing recently from a bunch of very wild and crazy guys in Michigan who are really very wise about audio. And let’s say we have a high-power audio amplifier, and we do this actually sometimes. We take this audio amplifier like the old LM12 or the LM2876, or all these high-powered audio amplifiers. And we take the screws in the heat sink and we loosen them until the op amp goes right past 200 degrees C. The op amp turns itself off to protect itself. The op amp is well protected, but the load, it goes brupt, brupt, brupt and might not be protected because when you turn the thing off, you might have a woofer running at, oh, 3 amperes. You turn off the op amp and the woofer draws a three-amperes right through the tweeter and wrecks the tweeter. So the point is that a high-powered amplifier may protect the tweeter because it don't turn off, and a medium-power amplifier might damage the tweeter because it turns off and lets the woofer beat the heck out of the tweeter. Now, I think you got a chance in this situation, and I can’t prove it - - by putting a clamp to the plus supply. And we clamp to the minus supply with a little diode, and I don’t want to tell you what kind, but maybe a -- you wouldn’t need a fast Schottky – a 1N4001, 1N4002 ought to do this job, although I’m not going to take my tweeters and try it. My amplifier doesn’t turn off because it’s vacuum tubes, and it works. And living in San Francisco, it does not waste energy; it just keeps my house warm. So I don’t have a problem with this.

Q Great.

That’s the end of this subsection.

Q Would you like to take a few questions, Bob?

Okay, we are well past halfway through it. Sure, let's take some good questions.

Q A couple of by-passing. Bob must work for a cost-effective company because he wants to use one bypass cap between the plus and minus rails instead of two, one from the plus rail to ground and one from ground to the minus rail.

Oh, I never said that, nope. People have asked me, why don’t I just put one capacitor from the plus rail to the minus? There may be cases where that’s a good idea; I can’t name one. If you try it and it’s okay, you have my permission to try it, but I don’t think I’m going to recommend it. I think you’re better off with one bypass to each rail from ground. In most cases, I think it will work much better. There might be cases where it’s not much better, but I don’t recommend that. I think you’ll find it’s not a good solution.

Q Okay. Martin clarified his ferrite question. He wants to know if he can use a ferrite to keep the noise from the switching power supply out of an analog linear circuit for the power.

He can try it. I’m not an expert on that. If he’d like, we can try to get somebody else to call him back, like Jon Cronk, who’s knowledgeable about that. In some cases, I think he’s right; you could do that. In other cases, it depends upon how low a noise you need. If -- if he only needs an improvement of 5dB, and the ferrite gives him 10dB, he should go away happy. And if he ain't happy, we’ll have to help him out some other way because if he could, for example, fax or e-mail us his circuit and he tells us how much he needs and what his frequencies of noise are. I just published an article. This engineer should go to the recent ”What’s all this Ripple Rejection Stuff, Part III"? And I will tell you where to find the darn thing. Go to www.national.com/rap and we show several applications where you can use an amplifier to get that noise down possibly a lot better than a ferrite. It depends on the frequency. Every application like that, every case is different. You go to ED Columns, and you go down just about a half inch down in the List of Columns, and it’s right there. Oh, I should write the name of it: "Ripple Rejection Part III." You can read "Ripple Rejection Part I and II," which are not -- Part II isn’t that bad. And Part III gives more results, and that might be useful because if you need a lot of ripple rejection improvement, I can’t tell from this distance whether you have a chance with a ferrite. It depends on the frequency; it depends on the ferrite. Go ahead and try it because if it don't work, you know it doesn’t work. And if it does work, you’re done.

Q Some of these questions we could spend an hour on alone.

We’re not going to do too much.

Q Joe wants to know about using discrete parts for a photo multiplier tube amplifier to get low noise as opposed to op amps. He’s heard or he’s read an article that you can ­­ get better noise in a photo multiplier tube amplifier using discrete components.

Uh, sometimes yeah. There's two places he should read. First of all, you go the same place. You go to my website, www.national.com/rap, go to ED Columns, and you go back about one year to ”Transimpedance Stuff.” So if you want pretty good noise, you can do awfully well with good monolithic amplifiers with good transimpedance amplifier, which means putting in a feedback resistor. If the best amplifier you could buy ain't good enough, go get the book by Hobbs, Philip Hobbs. You go to Amazon.com and you ask for Photo Optical Systems? Something about optical systems, photo optical systems. Ah, "Electro-Optical Systems." Look up that book; it’s going to cost you 150 bucks and it’s worth it. And he does give advice on how to cut down the noise. Sometimes you can just use one or two transistors ahead of a cheap op amp, a $1.00 op amp, and get almost as good as a very fancy custom-engineered system. It depends on how low noise you need. It’s complicated. There's no simple answers, but going here first, going here second. And this is not mentioned in the book, so you have to look up Philip Hobbs and his "Electro-Optical Systems." And that will help solve a lot of your problems.

Q This ties into Dev’s question -- can you mention some of the best books on this subject?

Which subject?

Q I assume just analog op amps, abuse of op amps. Your book, of course, is first.

Yeah, that’s one of the better ones. There are so many aspects of it. There's no simple answers on that. This is better for keeping you out of trouble than almost every other book put together.

Q Someone wants to know how to have a gain of 600 in a noisy environment, Chalo Marcelo. And I wish him luck.

Well, if you put it in a box, the amplifier doesn’t know what’s in that noisy environment. So, if you have a signal inside the box, and it’s low noise, you amplify it with a low noise amplifier and you bypass with capacitors, the supplies pretty well, you’ve got a good chance of getting a low noise output. And there ain't anything easy about it. First of all, you have to try it. After you try it, you’ve got a good chance. Another question.

Q One fellow wants us to ignore these silly questions and another fellow wants -- Joel wants to know how long it took you to grow your beard? He thinks it’s cool.

Forty years. Next question.

Q This fellow wants to add a 1-volt offset to a 40 megahertz square wave and make a differential output that’s 10-bit accurate.

Differential out -- does that mean push-pull output? Now, let’s think about that.

Q That was James.

40 megahertz square wave. Well, if he wants a push-pull output, he takes a good dual op amp and he puts in the signal, and let’s say 50 ohms, and 300 ohms, and 300 ohms and add in a volt, and that’ll add in plus one volt. And then you take an inverter, and it gives you minus one volt, and if you want to add that -- this is just R and R. So now you've got a push-pull output. There is no such thing in the world as a differential output unless you mean a floating output, which I don’t think he wants. But you can easily add in a little bias negative to push the output up – no problem. Do it. Next question.

Q John wants to make a variable gain amplifier using an analog switch in the feedback path to add and change resistances.

He has my permission to do it.

Q I’ve had bad luck doing that.

Well, if you want it at 17 megahertz, it ain't easy.

Q Right.

That requires a little engineering. Why doesn’t he try it, and then when he’s tried it, he can send me notes on what’s wrong with it, and we’ll tell him what he did wrong because there’s a thousand kinds of variable gain amplifiers that you might want. And he might design one that’s perfect, in which case he wouldn’t even tell me he had a problem; he might design it right. It’s not that hard to do it; just do it. And if it don't work, then you make a list of what you wished would happen and a list of what doesn’t happen that you wish would happen, and we’ll fix it. It’s not that hard.

Q You’re also switching in stray capacitance and switching in a varactor that changes with the applied signal.

What if 1N914s don’t leak? Well, he didn’t say 17 megahertz.

Q No, no.

We don’t know what he wants.

Q We’re not sure what he needs.

That’s too broad a question. But if you try it, it might work. And if it don't work, then you know -- then you write down a list of what you forgot to ask in the first place. Okay, one more question.

Q This ties into one of our own discussions of how do we terminate unused parts of an op amp? You have a quad op amp and you’re not using two or three of the amps, how do you terminate that so that --

Well, I usually say, for example, on an LM339, LM324, you tie one of the inputs to ground and either tie the other one to ground or tie it in feedback. Now you found a case the other day where the thing would blow up if it got too hot because if this is ground and this goes to plus 10 millivolts, and this wants to be plus 5 millivolts, it will try really hard and it will draw too much current and blow up. Most op amps don’t do that. So you have some CMOS op amps, you have some 741s; tie the inputs to the minus rail and they’re okay. In a few cases, maybe tie them to the plus rail. In a few cases, you might tie them to any other voltage. If you had a part that did get hot because it’s drawing too much current, you could tie them to plus one volt, and then they won’t get hot. One more question.

Q Eduardo wants to do like EKG signals, very low, low, low value differential signals, and he was looking for some precautions or tips for amplifying these type of very low level --

Okay, I’m not an expert, and that is enough to write a book about, but I can list about four precautions. First of all, EKG-type signals, if you’re going to put them on my body, you want to make darn sure it doesn’t blow up. So, we got to make sure that they’re resistors and the resistors, and maybe capacitor, say 10K and maybe 1 meg and .01. If you didn’t bypass the one meg, it’ll get awful noisy. And you got to put diodes in there to protect it so it doesn’t blow up. And so no matter what happens in there, it’s not going to put 100 volts on me -- things like that. You need a low noise amplifier. Typically FETs are low noise these days in that bandwidth. The impedances are usually high, so what you want to do is find out how other people do EKG preamps because -- you don’t want to ask me. You want to ask the guy who is selling EKG preamps. Go out and borrow an HP machine that does EKGs, and borrow the data sheet that they use to make their preamp, and look at that one. And learn from that guy; don’t learn from me. But that’s one of the precautions.

Q And read our life support policy for their EKG.

And that’s an important policy that says you take an LM394CN and you pay a dollar for it and if it blows up and kills somebody, we told you, you shouldn’t do that. You should buy metal can parts or guaranteed parts. So we do have parts that are guaranteed for life support and for places like that where you want to be really careful. We’ll be happy to talk to you about it. There’s nothing simple.

Q We have simple questions and complex ones, and we’re taking some of the simple ones because they’re quicker to answer; we’re running out of time. Maurice wants to know for a low noise pre-amplifier, inverting or non-inverting architecture?

Depends on the application. You could try it both ways. It’ll take you five minutes to do one and it’ll take you four minutes to do the other. In nine minutes you’ve done both. You could figure out which one you like better. Sometimes inverting is fine, okay?

Q How much current -- Steven wants to know how much current you can push into an un-powered op amp input.

Many of our op amps have an absolute max current of five or 10 milliamperes. So, if you want to protect it with resistors and diodes, and you keep it down to five or 10 milliamperes, you’re usually very safe. I can’t think of many exceptions. There may be cases where you can put in a little more, but you don’t want to get too pushy.

Q Here’s a great question from Kim that I know you know the history of where the bootstrap comes from, the term when we say we "bootstrapped an amplifier."

All right. Let’s say we have an op amp that has a certain input impedance. We want to drive it, drive the input. But it looks like a big lag because here’s a CR, looks like a big lead. This resistance is a heavy load. But, if we drive this and then take the midpoint of this capacitor and drive it again, this is called a bootstrap connection where as this goes up and down and this goes up and down and this goes up and down, there ain't no voltage across here. This is pulling this up by its bootstraps; the amplifier has to drive this. This goes back to when you had cathode followers and the bootstrap resistor comes in here and the bootstrap comes in here. And that’s only about what, 70, 80 years old?

Q Uh-huh.

So, that’s what it’s called. It’s like picking your boots up by your own bootstraps, which with boots you can’t do it, but with input impedances, you can.

Q I think we have some bad news for David. He wants to use a potentiometer on a circuit with a gain of an op amp can be changed from minus one to plus one continuously. How do we go through zero? That’s my first question.

Oh, let’s see. This is a standard circuit. I’m going to be lecturing about this about 7:00 tonight. You put your pot -- let’s say it’s a 10K pot. You put in 10K, 10K. It doesn’t make any difference of the value. When the pot is up here, you’ve got a unity gain follower and the gain is plus one. You put it down here, it's a unity gain inverter, it’s minus one. You’re putting the pot in the middle; it’s about zero – standard applications. So, it’s a unity gain inverter with a plus input tied to the tap on the pot. It works. Standard circuit. Not everybody remembers it.

Q We have 60 questions stacked up, so I think we’re going to have a long discussion.

Let's do one more and then -- then we’ll go do some more standard stuff. Proceed.

Q There are so many interesting ones here.

Well, yeah, that's good. We’ll get -- we'll get them all answered sooner or later. If they’re really interesting, we’ll make a column out of it, if it’s okay with you. We won’t use your name if you don’t want us to. Somebody said they were going to ask this question. Who first said, “Oscillators don’t; amplifiers do?" Well, what is it that oscillators don’t and amplifiers do?

Q I don’t know.

Oscillate. And my guess is 1937, 1943, that was a very popular phrase because a whole bunch of technicians were learning to use op amps or use tubes and other electronic things. And they built amplifiers that did sort of amplify when they weren’t oscillating. And they would also design oscillators that didn’t quite oscillate. I could do that; that doesn’t bother me. So my guess is it’s an unnamed technician from 60, 70 years ago.

Q Here’s a great question from Anoci (phonetic) because it’s about interpersonal problems in engineering. How do you deal with your EMC engineer that requests you to slow down the switching of a diode so it doesn’t radiate as much EMC?

What do they call them, soft turn-on diodes/soft turn-off diodes? People like Schottky diodes because they turn on and off real fast. So you put a Schottky diode into something, and put in a sine wave, and it makes a nice crisp waveform. Let’s say that this goes up, then it comes down. It turns off pretty well. If you have a snap action diode, it’ll stay on and turn off eventually with a snap. But there are soft diodes that sort of turn off kind of slow. And there are soft rectifiers that can do what you want in these high-speed circuits, and I’ve never used one in my life. I wouldn’t know who would sell them. But if you shop around and ask properly, there’s a chance you can do it. On the other hand, you know that the soft diode is going to hurt efficiency a little bit.

Q I’m taking the first question so we can stay fresh here. Muhinder -- might be related to my friend Sonja (phonetic). He wants to know how you can get a differential output from an op amp and still get full dynamic range.

Well, as I just said a couple of foils ago, it’s really not hard because nobody wants a differential output amplifier. Let me explain what a differential output floating output amplifier is. Here’s an example.

You take an op amp, make a pair of op amps. Now, this one’s called a differential in/differential out amplifier. So you wiggle the signal differential, you got a differential signal output. You wiggle the input common mode and the output goes up and down. So you tie the two inputs together and the output goes up and down. You can’t do that and have full differential range, which is why people don’t do this anymore. Now he wants a differential input?

Q No, output.

Differential output. Well, okay, to get good differential output, you put any op amp you want here and you put a unity gain inverter; it gives you a push-pull output. And, if that isn’t what he wants, then he has misstated what he wants. So, if that doesn’t make him happy, then he should explain what’s wrong with it, and then he should re-explain what he wishes would happen because I just told him something that will do what he wants; it gives full utilization of the supplies. If you've got plus/minus 15-volt supplies, you could get, oh, 56 volts peak-to-peak-to-peak-to-peak on the output. So, it’s not that hard to do this. And you can get full output utilization just by using push-pull. If somebody says, I could sell you a differential amplifier, differential output, and it doesn’t do what he wants, then he shouldn’t use that. He should use this.

Q Bob has a great question that with an increasing load on an op amp, the quiescent current, the supply current, goes up above and beyond the extra current used for the load. And he seems to understand the topology of the drive makes sense, but then the question got cut off because apparently of the software's involved. Yes.

Well, that’s a specific question. Not all op amps do that. My guess is when you have an op amp running on a plus or minus 1 milliampere, and you ask for 10 milliamperes, you ask it to source 10 milliamperes, and you’ll have a little more than 11 on the plus supply. Except you could have less than 11; you could have 10.9 on the plus supply, or you could have 11.2. It’s usually not a big deal. If he has an amplifier that is grossly bad, then we should discuss that case later because we can't answer that kind of question. But most op amps do not and should not and will not do that. Okay?

Q Some questions like John wanted to know if we have a .75 nanovolt per root hertz noise voltage amp. We have got a great selector guide that he can search parametrically.

Yeah. Well, there are ways of getting .75 nanovolts. You take two 1-nanovolt per root hertz amplifiers, put them in parallel, and they will do point-seven. I think we can do about 1.15 nanovolts per hertz, so you would have to put two op amps in parallel. And to do that, you need a very low source impedance. I have an amplifier coming up, so let's wait until I get to that one. Okay?

Q And it might be the difference between a video. Does he need this at video frequencies or down at audio rumble frequency? That’s important for us to know, too.

Yeah, I’ve been building low-noise amplifiers recently and getting very low noise. And it’s coming up! So, I’ll show you what I’ve done, okay? My turn.

Okay. Okay, my friend, other Paul Grohe, says, LMH6624 does .92 nanovolts per hertz. Well, last week I didn’t know about this one, but I do now. And you put two of those in parallel, and let me show you how to do it pretty quickly. You have Vin with low source impedance. Otherwise, the current will kill you. And you take the first amplifier and you take the second amplifier and you add together the outputs, say R and 100R. And this R also has to be very low in R 100R. Then combine and just add the two outputs with a couple of resistors, and then amplify it again as needed. And now this .92 drops to about .70 nanovolts per hertz. So it will be better than .75 nanovolts per hertz. And these are inexpensive enough you can use them. These will work very well above 10 kilohertz. I don’t think it’s very low noise at 10 hertz. The 1/F corner is not wonderful, but thank you very much Paul Grohe, that’s a good tip. Now I know about the LMH6624.

Okay, my turn. A guy called me up; I think it was honestly more than 20 years ago. He said, “Well, we just plugged in our power supply to our circuit board, and we put the minus and plus 12 volts on the plus and minus 12-volt terminals. We want to launch it tomorrow. Should we launch it? You know, the minus and plus 12 volts was only applied for a few seconds. And I said, Gee, if it’s working at all, it's still horrifying. Throw it away. I would not take that LM108A/MIL38510 and put that in my car radio. It does nothing trustworthy about any op amp with a minus or plus 12 volts on the supplies. Throw it away. And then a year later, some other unfortunate guy came back with the same question, a different guy. I told him the same answer. And I said, By the way, what if you put some anti-reversal diodes on there? And he said, Wouldn’t adding the diodes hurt the reliability? And I said, Compared to the reliability you got after you blew up the LM108, no, it would greatly improve reliability.

We’ll discuss several other little things you could do. What about anti-reversal stuff? Radio – an in expensive radio. You put your battery on it and then you absent-mindedly cross it. Isn't it wonderful that 9-volt batteries have these little prongs and you can put one between it and short it out, and if you put it -- put -- and touch them in there, it’ll blow up? Isn’t that a shame? But, if you put a diode in series, it wouldn’t blow up. Except now, you lose a lot of your battery life because if this gets to 6-volts, so this is 6.6, you’ve lost a fraction of the voltage of your battery. It’s less efficient, it’s wasteful, your battery life is less. Then hey, here’s another great idea. You put a diode, a big rectifier (1N4002) across the inputs. Yeah, you put it across the battery, and it croaks the battery and rips out everything in sight – explodes your battery, especially a car battery. Oh, it’s wonderful.

Well, those aren’t such wonderful solutions, although if you don’t care about your voltage, you could set 12.6 volt supplies; you’d get about 12 here. Some people could use this. Some people could use this if you have a current limited supply. If you’re always using a 12-volt supply that you could short and it doesn’t care. This is not a terrible solution. It doesn’t waste any power; it just gets warm when you cross the power supply coming in. But this is not wonderful. And I said, How about this? And I invented this, and I said, “This is such a wonderful invention, I’m going to apply for a patent on it.” And we searched and nobody had invented this thing up until about 1990, but one guy had done it about eight months before I invented it. You think, oh, in the 1980s. Why didn’t anybody think of this? Well, by finally in 1989, some guy did invent it. He invented it a few months ahead of me. But, you put our N-channel FET in here and you put it backwards. So instead of having the current flowing in the drain and out the source, it flows in the source and out the drain, which is okay. This resistor is optional; I just put it in there for a joke. You could put zero ohms, you could put a mega ohm; it doesn’t make any difference. Well, the mega ohm would turn on a little slower. Now, does this have to be a big fancy FET? If you need a lot of load current, you could put in a $2.00 FET and switch dozens of amperes with a tiny IR drop. If you were switching in for a radio, you could just use a 2N7000 that just has a few ohms of impedance; nobody really cares. Now you put it the correct way, the correct power supply, and this thing turns on and the current goes through it. You swap this power and this thing just turns off. You turn this minus and this doesn’t turn on; it just does nothing. This is a reasonably good circuit and that’s on page 164 of this book, or whatever it is. So, this is a fairly good what would you call it? A fairly good application to avoid the damage and the loss of series diodes or the disaster of shunt diodes, and it has very low loss. And it only costs, you know, a dime or two.

Here’s another good solution to a system. You make the PC card edge connector a little bit shorter for your power pins. You plug in your ground, and you start to shove it in this way, at this angle, and this hits before those. You shove it in that way and this hits before those. No matter what you do, ground is always connected before the power supplies. And then you can install your anti-reversal rectifiers here. As I was saying, when you plug it in, this is pretty safe, and you can have an anti-reversal thing on your plus 12 and your plus 5 and your minus 12. And you could even have an anti anti-reversal between your plus 5 and your plus 12 so that this doesn’t go appreciably above the plus 12. That is a very good system solution for many cases. One question.

Q Asa has a very interesting subtle question about banging to see if the op amp rings. He wanted to know, do you do that on the output of the amp?

You can do it at the input, you could do it at the output, you can do it on the power supply. It’s not that critical how you do it. You do it to see if it’s ringing. If it isn’t ringing, you don’t have to do it very often.

Q Jay wants to know about protection diodes and audio circuits that minimize the noise added.

The diodes you might add to protect an audio circuit need not turn on at all, so they should add no noise to the circuit. I can’t think of how they would add noise.

Q Grant brought up, we were talking about conventional through-hole resistors breaking down across the spiral cuts. What about surface mount resistors?

I think it’s generally true that the larger the resister, the more abuse it will withstand, and not just watts, but volts and stuff. And some of those are cut. And you look underneath the protective covering, and the bigger ones can take more abuse and will break down less than the smaller ones. So if you really have to stand off hundreds of volts -- in the old days, people used to say, this resistor is rated for one watt or 600 volts, whichever comes first. You can’t put 1,000 volts on it even if it’s less than a watt. These days, people don’t talk about it, but I’m sure it’s still true. And you might have to ask the resistor manufacturer.

Q David’s touting a book by Jerald Graeme about transimpedance amps for photo diodes.

That’s pretty good but I don’t think it goes at all beyond my invention. I think my column on transimpedance amplifiers, goes further than Jerry Graeme’s book. I asked him some questions but I never quite got an answer. I asked him, “Did you treat this in your book? I read your whole book, I think. Did you discuss this topic? I didn’t see it.” He never quite got around to answering my question. I do, in my column, on transimpedance amplifiers, I do recommend his book. And it’s a pretty good book. And it teaches us some basic stuff. But the Phil Hobbs' book is much more advanced. And, if you’re really trying to push state-of-the-art, Phillip Hobbs’ book is pretty good. I haven’t yet proved I know more than Hobbs, but I could sneak up real close to him. But he’s a good guy to learn from if you really want high performance. And I'll have another mention of that in awhile. Yeah.

Q Bob has a question about your book. He thinks he may have found a typo, or he has a question. Let’s see. A signed book, on the diode curve page that you were talking about, E and F seemed to be connected and run with the opposite slope. Is that a typo?

Not a curve slope, E and F? Wait a second. They have different slopes maybe. No, F is just pointing to one of the curves. As F points to a curve that is otherwise unlabeled. The curve that’s F is not quite as steep as E, but E and F both slant similarly, and the little thing with the arrow is just pointing to that curve because I couldn’t fit everything in there.

Q All right.

One more and we’ll go to section G; we’re doing great.

Q Boy, we have a ton of questions here. I’m sorry we’re not going to be able to get through all of them. I know this is a common problem Fredrik has: Driving an analog to digital converter input. And I’m not quite sure where he’s going, he's with an op amp that sinks a lot of current at the input. Is he talking about a current feedback amp that has low input impedance or --

Unclear. The question is not quite clear; we should probably answer that later. The whole question of driving A to D converter inputs is tough and it’s not usually impossible. I know Nick Gray in our Data Conversion Group is usually an expert on that. You can’t have one solution to make everybody happy. So if you say, “I want to do 72-bit resolution,” then we have to use the most precision op amp. If you want to do it at 72,000 megahertz, then we have to use the fastest amplifier. There’s no simple answer to make anybody or everybody happy, but we can probably get you one solution that will make you happy.

Okay, off to some more stuff. I think I’m all done with this section. My goodness. I had a question from a guy who had a little problem with an amplifier. Problem: Low noise pre-amp is needed to get a signal from one microvolt up to 10,000 microvolts up to the 2 volts to drive a big A to D converter, such as a 16-bit A to D converter. And the signal source is about 500 Picofarads. It’s a capacitive source, and he designed a little op amp circuit. So he got some low noise op amps and he put in a meg ohm here and 100 ohms here and a mega ohms here and a mega ohm here and 10 picofarad. Wow, ten picofarads. And he put in these low-noise amplifiers and he fed it into the 16-bit A to D converter after taking a lot of gain. And he engineered it up, he ran it in SPICE, and it seemed to work real well, until they went out and built 200 PC boards with 400 channels and 1,600 op amps. And SPICE said this would be a good low-noise amplifier. It’s funny. When they got some built, they were noisy, and they looked at the next channel, and it was noisy. And they looked at all the other channels, and they were -- well, how much noise? Well most of us don’t run into this, but I think this sets a new high of 24,000 bits -- 24,000 noise-to-signal ratio in the signal bandwidth, which was about 10 kilohertz to 100 kilohertz. Now also, it was 16,000 LSBs of noise from DC to 100 hertz. In other words the output was wandering around so much it wouldn’t even stay in the middle of the range of the A to D converter. Oh, 24,000 LSB. Now, that’s not a new high, but it’s getting up there.

What did he do wrong? Well, the first thing he did wrong was he trusted SPICE, and he didn’t breadboard it, and he may have used the wrong model. But the model of this AD797 is very low voltage noise. It’s down like we were saying, 0.92 nanovolts per hertz. Very low voltage noise, but guess what? Very high current noise. Now one time I thought I looked at the SPICE models of an AD797; I tried to look at it again. I think it has a SPICE model with one picoampere per root hertz of noise. So part of the problem is this current, if it goes through a source impedance of 500 picofarads, it’s disastrous. If a DC goes through a meg ohm, mass disasters. If it goes through 1 megohm here DC coupled, that’s the source of your noise. So, he doesn’t have to throw all those op amps away, but I did tell him he had some problems, and he wasn’t going to get there unless he threw away SPICE -- wrong op amp choice; bad SPICE model maybe. He didn’t use the SPICE model correctly. He didn’t build a breadboard. He trusted SPICE -- bad choice of R and C levels. What can he do?

Q That answers Myron’s question about, do you believe in analog simulators, especially for AC simulation?

If -- if you applied it right and if you knew it was going to be lousy, and if you got an answer saying it wasn’t lousy, you’d say there’s something wrong with spice. You have to be smarter than SPICE. I’m giving a three-hour session Tuesday night and Thursday night on SPICE down at San Jose State. And I’m warning people, and I’m going to show them this example: don’t trust SPICE. You have to be smart enough that when SPICE lies, you have to say something is wrong.

Q Joshua wants to know how to make amplitude in different phase detector.

I don’t know how to do it. Ask somebody else.

Q Gain it up and clamp it and --

Nah. Anyhow, I told this guy, you can recover; you can get much lower noise. You can’t get to one to one Noise-to-Signal ratio; you can’t get to one LSB. Well, he might get down to 40 LSBs. He could probably use AD797s or similar in the output stage, but you don’t couple them with 10 picofarads and 1 meg. You couple them with 10 nanofarads and 2K. Then the output wander and drift will be about 1,000 times better; it’ll be about 16 LSB peak-to-peak of DC noise. Not too bad. Then at the input, you could maybe use 797s or similar. You don’t have to use a 797 here exactly. But you take a 2N5486, which is a low noise part, 2N5486. And my point is, you got to buy a 1,000 of them because you’re going to use them not just 10 at a time or 20, but maybe 80 or 160 at a time to get low noise. You set up a little AC amplifier with some DC coupling and you make one of these, and then you make another 10 more, and you parallel 10 of these, and then you do it again. For each group of 10, you need match FETs; that’s why you need 1,000 of them so the VGS matching is in the right ballpark so these are all in each group sharing the current. So, if you use 40 of these, you don’t need 40 capacitors. But you might need four capacitors.

Now, your signal source impedance is about 500 picofarads. If you only have a few picoamperes of noise, the noise current won’t kill you. It’s not like one picoampere per root hertz, which really hurts you with that kind of signal source. This has a chance of getting down well below .7 nanovolts per hertz. Now, this is what the guy was asking a little while ago: “What if your source impedance is bothering you?” He wanted to get .75 nanovolts per hertz. This will do it, although it wouldn’t necessarily be a good choice to do it for DC. There are no simple answers. If you’re trying to do it for AC, this would probably do .74 nanovolts per hertz using several dozen 2N5486s and several dozen (times 4) milliamperes. You got to use a lot of milliamperes through these things to get low noise. And then you need a low noise second stage and then a reasonably low noise output stage. This may work pretty well for low noise. Oh, let’s do two more questions and then we got two more -- we only got two more groups. We’re doing great.

Q I'm sorry to say we've got like 60 questions. We're not going to get --

Well, pick a couple…..

Q Here’s an integrator question and a differentiator question. Chris had a question about integrator configurations, specifically a cap in the feedback loop versus a series RC in the feedback group, and how this affects the speed and the stability of the integrator. And he thanks us in advance.

If you have an output, if you want an integrator you do integrate. Now sometimes you put resistor around it because you want to approximate. If you want an integrator, it integrates. Could you put a zener on it so it won’t peg? Yeah, if that’s what you want. Would you put a series on it? If you want an integrator with a resistor and a lag, you could do that. So you put in a step up and the output instead of just integrating down, the output will step down and then integrate. You want that, you can do it. People do things like that. It is often a good idea to do that. I can’t tell for sure, but you can do it. Next question.

Q There are so many good ones.

I have a week’s worth of stuff to finish all off these 159 questions.

Q Somebody caught that we said carbon compensation instead of composition.

Composition. I'm sorry. You try giving a lecture for two hours and you don't make no tongue error, well, you're doing pretty good.

Q A differentiator circuit: How do you improve the noise when the input slope is slow?

Well, first of all, differentiators are often theoretically made with a C and RF, but you can’t just do this. If you just do this, it will just plain oscillate because that’s the way to make an oscillator. You get a lag and a lag and it oscillates; we’ve been discussing that. But, if you put a little bit of resistor, like this is input cap is C, and you put in CF= .01 C. And you put in RF and you put in .002 RF. This has a chance of being stable. But, if it’s noisy, it’s because your signal is noisy. A differentiator like this has gain that goes up and eventually it comes down. Putting the R in here and putting more R, a higher R if you had to, and a larger C if you had to, would bring this down a little bit, but it might hurt your fidelity of differentiation. Try some of these things and see what it does. But, if you just have nothing here and no picofarads here, yeah, it’s going to be noisy as heck. It’s going to ring and scream and be awfully noisy, especially near the crossing frequency where this gain goes up and the op amps gain comes down. And that may not answer his question, but that’s a good start. And if he fools around with this, he should be able to get some improvement. And if can’t, then he should define his exact problem. Next question. One more question and then it’s my turn.

Q Greg wants help with a competitive parts, so we’re going to skip on Greg’s.

Well, we’ll help him define in our part, but we’ll do that later.

Q Right. Marion wanted to know if op amps lose a lot of their gain when they’re operated close to the supply rails.

Lousy ones do; good ones don’t.

Q Ours are good.

Ours are. I think all of ours are good, even an LM324. You get it near the plus rail, it works pretty well until it just doesn’t work. And then it stops.

Q Right. Kim wanted to know how to make a virtual ground, a mid-rail ground, the best way.

Well, let’s for example say, plus 12, and you want to have a 6 volt supply. And one of the things you could do, you could take a six-volt zener, but it’s usually better to take the resistor. You could add a follower if you want to use a three-cent device. But if you want to use a 10-cent device, this would be much more accurate just to put a follower here. One good thing about that is if this is a little noisy, you might want to put a little capacitor there, so you spend two more pennies. And sometimes this is a very good solution. And if that ain't a good solution, you have to say why it isn't a good solution, and then we’d help you find a better one. But these three solutions are usually pretty good. That’s 6.6 volts. That's 6.6 and this is around six volts. And this is around six volts. That’s pretty good.

Q We have enough time for a quick one. Matthew wants to sum two analog signals in a non-inverting solution with one op amp. So a mixer or --

Yeah, yeah, yeah. Now, the general case is, he didn’t say what gain he wanted, did he?

Q No.

Okay. Let’s here's R and R and I would just say 4R. The guy wants to gain a four. You put in R and R and 4R, you put in a signal here, you put in a signal here, your gain is plus four. Or if you put in R and R and R, the gain is plus one. But you have to tie these to ground; you don’t have to do a heck a lot of with them. But you can fool around and change your gain by changing this ratio, but the adding is done over here. And it’s really not hard. There’s one minor drawback. If you wiggle this one, it kicks back into this one. So, if you had a source that has lousy output impedance, it might kick this point a little bit. That’s why people usually like to use inverting amplifiers because you wiggle this signal and it doesn’t kick back into this one. But R and R and R and maybe a little R over here; you don’t need seven Rs to make this thing work. You might do it with 2 or 3. But that’s how you do it. You add them into the plus input and let the minus input do the feedback. Okay? My turn. My turn.

Speaking of plus and minus inputs, if you have an op amp and you put in the signal in the plus input, which is the normal application, it has a gain of plus 10. R -- 9R -- gain is 10. Fine, it works. You put in no volts; you put in .1 volt, the output goes up to plus one volt, for example. But, you run this thing in SPICE, you see. I wonder if this thing is working. Do I have the op amp inputs connected forward or wrong? We at one time had an application where Paul, the other Paul, found some op amps that were crossed up on the inputs. If we had run them in SPICE, were we going to find this? And it turns out in some cases with certain kinds of SPICE you cross the input, you put in the signal, and the output goes up 1 volt, just as if you hadn't crossed the input. Oh, my God. That means if you run a SPICE run on this, you might not even tell if you had the inputs crossed. That is not nice. This is true whether you have a real op amp circuit with lots of transistors, or whether it’s a macro model, or it’s a SPICE model, or a simplified model, the op amp can go up when you put a signal up. It won’t peg like you want it to. How come? We finally found out; it took years and years to find it out. The output appears to work if the time step is too big, if you have it in an automatic mode and it selects a time step that’s automatic, the time step is too big. When you force the time step to be small enough and the output will peg. So, don’t expect SPICE to act like the real world because SPICE is beastly.

Two more questions and then I’ll finish my junk.

Q Joshua has a great question; it comes up all the time. Is there any way to get rid of noise coming in from the ground plane? And it’s a switching power supply noise that’s polluting his ground plane.

There are ways of getting an output that’s not too noisy, but by definition, where is ground referred to? There is a way to do it, and it will take me more than a minute. Maybe we’ll do that at the end. Save that ground question until the end, could you?

Q Okay.

One more.

Q Somebody wanted to know why the LM6365 -- Randy, wanted to know why we end-of-lifed or discontinued that. We’re looking into that. Be careful, Randy. Sometimes we discontinue a certain package, like a ceramic package, and there are other versions of the parts. So make sure that that’s end of life.

Okay, I can give this one a quick answer. I know we just hadn't added and subtracted with two inputs. Let’s have four resistors all nominally equal, maybe 1 percent, maybe .01 percent. So we have a VN here referred to ground A, and here’s ground B. If ground A is jumping up and down and VN is referred to it, then VN referred to A is not noisy. You can take this signal, it is not noisy and you can get this signal moved over here. And then as this moves up and down, this don't. It is possible to do this. Now, if it’s really, really, really noisy, you may have some problems. Got it?

Okay, survey. Who’s going to do surveys?

Q They’re going to launch the survey for us later.

Oh, they’re going to do a survey. There will be a survey; we hope you will be able to see it and to reply to it if convenient. We hope you’ll ask us more questions. We hope you will have some fun, and we hope you will like the answers we bring you later. We hope you come back and look at the archives later.

Q Yes. Allen had a question about he's using a CLC450 and we’ve got an improved VIP10 process that’s far superior. And he wanted to know op amps should drive impulse transformers. And he’s looking for a replacement for a CLC450. We’ve got a lot of work on crossovers that if it’s not available now, we’ll certainly have something very soon.

Yeah, if he looks in basically the LMH family, he may find something that’s as good as or better than or plenty good enough; I don’t know what he needs -- what he needs because pulse transform applications are sometimes obscure, but there’s a good chance we can help him out.

Q Yes, our crack engineer, Paul GROHE, is furiously writing. I think he knows exactly -- he says go to our website www.national.com/see/clc2lmh.

And that gives a whole bunch of selections that way?

National.com. That one you should have remembered by now. National.com/clc2lmh for converting over from old CLC parts to LMH6 parts.

Q He's going to find it for us.

He'll find it for us.

Q We’ll have it back in a minute. Somebody, Asa, wanted to know, are 741s widely used in new designs?

Relatively not so much. They’re still excellent for certain things like driving capacitive loads, but a lot of people are doing low voltage work, and 741s are not really suitable for anything below plus or minus 5 volts. And people don’t use them for that much anymore. They’re not cheap, they’re not excellent for anything, they’re not low noise, they’re not wonderful, but if you have them around, use them.

Q Matthew wrote back on that two input circuit -- mixing two inputs in a unity in a gain of one. It was a unity gain situation.

Oh sure. Where did that piece of paper go?

Q Oh, we have so many here. This is analog PowerPoint.

RRR. Okay, unity gain. Watch this. RRR -- you could actually put in ½ R. R RRR – all four resistors equal gain equals plus 1 for each.

Q I’ll let you write that good and big.

So, this is the circuit that will do it, a gain of 1 for each input here. You would also do a gain of minus 1 for each input here. But if you’re not using these inputs, you have to tie them down to ground, and this will give you a gain of plus 1 for each input. Okay.

Q www.national.com/see/clc2lmh

Good. So, almost everything we wrote down was correct, but it’s national.com/see/clc2lmh, and I’ll leave that up there for a minute. Okay, my turn.

We had a customer who was running a fairly basic high-powered flasher. Blink, blink, blink. Real big, what would you call it, photo discharge tube? And he had an ordinary LM339 type device to rattle back and forth with some Rs and some Cs and positive feedback to rattle at, I’ll pick a number, 10 kilohertz -- not a big deal. Not blindingly fast or anything, driving in to the gate of a middle-sized FET to drive a transformer to drive a diode to generate plus 300 volts. And I think one of the other sections of the LM339 triggered this thing, and one of the other sections said at 300 volts, you turn this all off. Fine. So we built up a bunch of these and they seemed to work, and we built a bunch more, and some of them were dying. And the LM339 was dying near its output. And some of them would die in the first minute and some of them would die in the first hour and some would die in the first day and some would die in the first week and some would die in the first month and some of them kept on playing forever. Fortunately, he had enough of them we were able to de-cap it and we discovered the output was all blowing up right at the output wire, output metal, output pin. What’s going on? Well, the flash lamp fires. This drops from about 300 volts to about 30 volts in way less than a microsecond; less than 10 nanoseconds. These things turn on really fast, they draw huge currents for a while, and they pull this voltage down. And this goes down 280 volts fast and this one does, too. And this goes down negative a lot, too. And this goes down a lot, too. Why wouldn’t it? But isn’t there a clamp out of there, a Schottky, 1M5817? Isn’t that a clamp? Well, you know, it’s almost a clamp and it’s almost a good enough clamp to save the thing, but it ain't quite big enough. The output pin of the LM339 is hammered to minus 32 volts. So, the falling edge is just excessive; the pulse is one nanosecond wide. How can you put 30 volts across the 1N5817? Well, you do. And it turns on in about a nanosecond, but it’s a nanosecond too late. Oh heck. So we finally help the guy out a little bit because he says, “Look, the data sheet on your part says you can put 32, 34 volts, 36 volts on the LM339.” We said, “Wait a minute. If you look at the data sheet, it says minus .3 volts minimum, plus 32 volts max.” Minus 32 volts is greatly excessive. We figured there was an ampere flowing near this. So, we took his 12-ohm driver. You really want 12 ohms to prevent oscillation of this little FET. We said, “Put six ohms here, put 6 ohms here, catch it with a 1N5817, move it over just a tiny little bit to give this a chance. Then, put in another 12 ohms in series of the FET, and put -- wrap five turns of wire around the resistor. And that way, when this moves really, really fast, this is slower -- a little bit slower. It's got a chance to survive; we never had anymore fails. But that was a tough one.

Now, we’re nearly to the end of time. Last page of advice is eschew SPICE; avoid digital computers, and avoid digital circuits. Whenever possible, don’t blow them all up. Oh, thank you very much, and we still have time for more questions.

Q About three minutes. Greg has a great question. This is my favorite so far. I think Greg has the makings of a great engineer because he says will your book help me think in the time domain? If not, can you recommend something?

Yes. My book is lousy to help you in the frequency domain; it's not wonderful. I’m not a good frequency-domain man. Frequency-domain guys have other problems, but they don’t usually do as well as I do in the time domain. My book is generally good in the time domain.

Q Our pal Nick Gray just wrote in. Nick works for us in the Analog Digital Converter Group. And he’s going to have a seminar on December 10th where he says he will discuss ground-plane noise.

Oh, that’s excellent, yeah.

So Nick’s an expert on that, a great resource. Even if you’re not doing A to D converters, there are all kinds of great things to learn from Nick.

Now, do you want to read that and then we can still do one more question?

Announcer: Thank you Bob and thank you everyone for attending this online seminar co-sponsored by Arrow Electronics and National Semiconductor. Please take some time at the end of this presentation to fill out the short survey and join National for their next online seminar, December 10th, entitled "Controlling Noise and Radiation in Mixed Signal and Digital Systems." Thank you.

Very good, yeah. That ought to be interesting. I’ve heard Nick -- some of Nick’s advanced practice lectures on that, and he really knows his stuff on it. One more question.

Q Peyman has a great question: What’s the best way to get two separate analog signals with different magnitudes to end up being the same magnitude? I hope he doesn’t mean against the rail. I don’t know much about automatic gain control.

I don’t either. I think what he’s trying to say is he wants to get one of the signals that's smaller to be bigger, the same size as the other. He might need a rectifier. You want to do that at 1 hertz, you do it one way; you want to do at megahertz, it’s another way. I am not an expert on any of those things. So, I -- I’ll pass on that one. I don’t know how to help him much.

Q That’s it.

We quit! We quit! Thank you very much, friends. Bye-bye.

(End of Presentation)