31 August 2011

App Note 29 part 3

This app note has several interesting appendices (of course, Jim's appendices are always interesting).

Appendix A discusses a common topology for a 5V-to-plus-and-minus-15V converter. The topology discussed is the Royer topology, and here, Jim thoroughly badmouths it. His main complaints are the noise (the switching residue) at the output and the current consumption (due to the magnetic saturation behavior). He says, "The converter's inherent operation ensures noisy outputs" and he suggests several schemes to minimize or avoid the noise. The "bracket pulse" and strobed-output operation are worthy of Rube Goldberg. He claims, "While useful, none of these arrangements offer the flexibility of the inherently noise free converters shown in the text." That sentence may be a bit of an overstatement. "Noise free" is an exaggeration.

The funny thing about this appendix is that he returns to the Royer topology for LCD backlight lamps, and expends quote a bit of ink extolling their virtues in that application. Of course, his complaints are still valid in that application, but I wonder if he ever wished that he had tempered his language here a little bit.

Appendices B and C we have seen before in previous app notes.

Appendix D discusses inductor selection. He says "Electromagnetic theory, although applicable to these issues, can be confusing, particularly to the non-specialist." He suggests the purchase of an inductor kit (such as shown in Figure D2) and a cut-and-try method for inductor selection. Jim sure loved his kits and decade boxes! (See previous app notes for praise for his General Radio 1432 resistor decade boxes.) This approach is a practical one, particularly if you can avoid the horrors shown in Figure D6. Yikes.

Appendix E covers converter efficiency and sources on inefficiency. There is good discussion and good advice all around here (with a brief mention of using germanium).

I already covered Appendix F last week.

Appendix G discusses magnetics and inductor selection (again!), and contains another great quote about his own inspiration (page AN29-44): "As a purveyor of switching power ICs we incur responsibility towards addressing the magnetics issue (our publically spirited attitude is, admittedly, capitalistically influenced)."


29 August 2011

App Note 29 part 2

The main section of this application note contains six major sections of converter circuits, from micropower supplies to high-voltage sources. There's a lot of good stuff here.

The first major section discusses several 5V to +/-15V converters (Appendix A also covers a 5V-to-15V converter, but a special case). Figure 1 is a "low-noise" converter that limits the slew rate of the transformer drive to limit the high-frequency harmonics at the output. This trick is a good one, although the circuit here is complicated by the overdrive (and underdrive) that the MOSFET gates require to fully turn on and off. Figure 4 goes for even lower noise by using sine wave drive (amplifier A1 is a 16-kHz Wien bridge oscillator!), but efficiency is only 30%. Figure 6 uses a single inductor (similar to the battery-powered design in Figure 22 of App Note 8). Figure 8 has low quiescent current.

The second major section discusses micropower converters. Figure 12 is a tricky micropower boost converter, using the VC pin on the LT1070 to duty-cycle the part into a low-power state when the output is lightly loaded. (Am being I paranoid? I think Jim is mocking me in footnotes 2 and 3 on page AN29-11). Figure 19A is a micropower buck converter using the same trick, and Figure 19B has multiple outputs. Figures 20 and 23 are single-cell 5V converters, using the LT1017 and LT1018 low-voltage comparators (these parts have been previously discussed: see App Note 15). Figure 23 uses an LT1070 at 1.5V, but it requires a start-up circuit to get at least 3V on the supply pin.

The third major section discusses high-efficiency conversion (other trick are discussed in Appendix E). Figure 32 uses synchronous rectification to reach 90%. Figures 35 and 37 show isolated and nonisolated designs that are 75% efficient.

The fourth major section discusses wide-input-range converters, starting with a 40V-to-60V-input telco converter with 5V output in Figure 38 (almost the same as Figure 4 in App Note 25). Figure 40 is a flyback converter that produces a -5V output from an input between 3.5V and 35V. Figure 42 is a buck converter and Figure 44 is a buck-boost converter, similar to the flyback in Figure 40. Figure 46 is another example of a linear regulator with a switching loop that controls the voltage across the linear regular (we've seen this trick before, too; see App Note 2 Figure 5).

The fifth major section discusses high-voltage and high-potential supplies. Figure 49 produces 1000V out (make note of the damper network), while Figure 50 produces a floating 1000V output. Figure 51 uses a piezoceramic transformer (foreshadowing of future topics, here) to achieve an isolation voltage up to 20 kV (the output voltage is only 10V, but the common-mode voltage can be huge).

The final major section discusses switched-capacitor converters, using the LT1054 and the LT1026, (similar to circuits we saw in App Note 11). He also includes some new tricks, like the diode-capacitor voltage multipliers in Figure 60, and the LT1020 duty-cycle trick in Figures 58 and 59. The main section of the text concludes with Figure 61, the high-power switched capacitor converter that we previously seen a couple times before: Figure 23 in App Note 3 and Figure 12 in App Note 8.

Best quote (so far) is on page 22: "These 'boingies' can be seen in trace B on Figure 43B." That deserves to be in the technical lexicon. Then again, I adore jargon.


28 August 2011

Scope Sunday 7

When I started these "Scope Sunday" posts two months ago, I thought that I'd have plenty of new material from the upcoming MIT Flea Markets. The recipe was simple: I'd by a scope, or a plugin, or some test equipment, and post about it. I also had planned a trip out to California which included the De Anza Flea Market in Cupertino, so I was sure that I'd be buying too much gear this summer and blogging about it.

The reality of this summer's shopping has been less satisfying. I was at the MIT Flea Market last week, and after a quick walk around, the only piece of Tektronix gear that I saw was a 520A Vectorscope for $30 (guaranteed not to work). No, thank you.

When I was at the De Anza Flea Market two weeks ago, I also found nothing. Nothing! It was very disappointing! Maybe it's just the August doldrums (happens every year), but I really was expecting to find more oscilloscopes and test equipment to choose from.

After a deeper search of the MIT Flea Market last week, I did find some additional gear at some unreasonable prices: a 465B for $275 and a broken function generator for $150. You know, these "Scope Sundays" posts are going to become impossible if I don't get some better luck. Maybe I'll go back to trolling Craigslist (or should it be described as "trawling"?). Maybe it's just a sign that I need to fix some of the gear that I already own, rather than acquiring more gear. (Nah, can't be.)

I did, however, purchase an HP 3400A true RMS voltmeter at the MIT Flea, but I'm going to save that until after App Note 61.

Looking forward to September!

26 August 2011

App Note 29 part 1

"Some thoughts on DC-DC converters." 44 pages.

At 44 pages, this app note is the longest one so far (but not the longest one ever, by far). It's also the first one with a coauthor (Brian Huffman). Given the length, I'll cover this one over the next few days. Today, I can't help but to talk about the oscilloscopes!

First, there are eleven traces in Figure 7! How did he do that? I don't know of any Tektronix mainframe that allows for 11 traces on a single display. You can get 8 traces using a 556 with two 1A4 plugins (which we've seen before, for example, see App Note 3 Figure 16), but I don't know how to get eleven. I suspect a double exposure with the camera.

Second, there are some great measurements here. Figure 5 has a trace at 20 microvolts per division. Figure 33 has three traces with current probes at 2 amps per division. There are some high-voltage measurements, too (20 volts per division in Figure 2), but we've seen higher (for example, 200 volts per division in Figure D7 of App Note 25).

Appendix F contains some more sage advice on instrumentation. Again he starts by talking about probes, but after that discussion, we finally, finally have some explicit oscilloscope recommendations. After dismissing the more modern Tektronix scopes (the 2445 and 2446 were modern at the time), he recommends his favorite, the 547 with a four-trace type 1A4 plugin. I'm surprised that he suggested the three-bay 7603 (with two 75-MHz 7A18 plugins) as an equivalent mainframe. There are much nicer (and four-bay!) 7000-series mainframes available. You certainly don't need the bandwidth of the specialty 7104 here, but I definitely prefer the 7704A and 7904A mainframes to the sluggish 100-MHz 7603. (However: in most applications, Jim preferred low-bandwidth scopes. It's actually good advice.)

He also heaps significant praise on the 556 dual-beam scope, and Figure F2 contains the first actual photograph of a scope (a type 556 dual-beam oscilloscope with 1A7 plugin). He also discusses some specialty low-level and differential plugins, including the 1A7 and 7A22 plugins (with 10 microvolt sensitivity) and the differential comparator plugins W, 1A5, and 7A13.


24 August 2011

App Note 28

"Thermocouple measurement." 20 pages.

Another classic Jim Williams topic: thermocouples. Thermocouples are another entry in that great category of "important to instrumentation, but often ignored by circuit designers" topics. Clearly Jim had a lot of experience with thermocouple instrumentation, and it shows in this app note.

This app note is the first one starting with a history lesson (the first words are "In 1822"), along with the oldest reference (also 1822) on page AN28-13 (well, the oldest so far). He goes on to describe a variety of temperature sensors in general and thermocouple behaviors in particular. The chart on page AN28-2 is a useful summary of the characteristics of thermocouples, thermistors, platinum RTDs, sensor diodes, and IC temperature sensors. The chart on page AN28-3 lists half-a-dozen kinds of thermocouples, and the accompanying text discusses cold-junction compensation and ice baths.

The rest of the app note describes the LT1025, which provides cold-junction compensation without an actual ice bath. Several op amps are suggested for the precision job of thermocouple-output amplification, including the precision LT1001, the chopper amplifiers LTC1050 and LTC1052, and (the now discontinued?) bipolar LTKA00 and LTKA01 amplifiers.

Three circuits (Figures 13, 15, and 17) are given for isolated applications. The best circuit is Figure 15, with 0.01% accuracy. He spent a lot of time working with isolated applications. Again, I think isolated signal chains are in the category of "important to instrumentation, but often ignored by circuit designers" topics. You don't spend a lot of time discussing them in school, but in the real world, isolation is often needed. Practicing engineers (especially integration engineers) probably spend more time worrying about grounding and isolation than any other concern.

The final circuits implement linearization, with a variety of techniques: offset (Figure 20), diode breakpoint (Figure 21), analog multiplication (Figure 22), and digital (Figure 23). Figure 23 is accompanied by nearly three pages of assembly code. The software comes with an appropriate disclaimer from Jim: "Including of a software-based circuit was not without attendant conscience searching and pain on the author's part. Hopefully, the Analog Faithful will tolerate this transgression." Still, I teased him about it.

The best quotes come from page AN28-19 in Appendix A's discussion of error sources: "Minimizing sensing error is the manufacturer's responsibility (we do our best!), but tracking requires user care... As a general rule, skepticism is warranted, even in the most "obviously simple" situations. Experiment with several sensor positions and mounting options. If measured results agree, you're probably on the right track. If not, rethink and try again."

The app note concludes with another cartoon.

22 August 2011

App Note 25

"Switching regulators for poets: A gentle guide for the trepidatious." 24 pages.

This application note is a classic. I think it's the first one where Jim's personality and sense of humor really shine through his writing. There are a lot of gems here, even just on the front page: the title is a classic, along with the discussion of Everyman and the poets, and "my poetry ain't very good." I also like his acknowledgment of the "encouragement" from the Captains of his corporation. Page AN25-24 sports his first cartoon, which will become something of a trademark.

The app note itself is pretty short: only twelve pages of text (and twelve pages of appendices). Discounting Figure 1, there are really only five application circuits here: Figures 4, 6, 9, 17, and 18. The best circuit is the monster in Figure 9, the 100W off-line switching regulator, an impressive achievement.

The appendices contain some great advice. Appendix B is a very practical treatment of compensation for switching converters. It's not as analytical as I would like (what's new?), but it is practical and exhaustive for this application (and it isn't the mess that App Note 18 contains).

Appendix C is worthy of particular note: Figure C1 shows the first oscilloscope labeled as a Type 547 with a four-channel Type 1A4 plug in (which we've known all along). However, I believe that the pictures in Figures C2, C3, and C4 are from Jim's Tek 556 (note the damaged gradicule). If you look back at Figure 16 in App Note 18, you'll see the same damaged gradicule with six traces on the screen, which requires a 556 with two 1A4 plug ins. Figures C5 and C6 show some nonideal turn-off and turn-on effects in diodes, which is the topic of a future app note.

The evolutionary design approach in Appendix D is very good method. I have seen more than one engineer attempt to power up a design such as Figure 9 all at once, and the fireworks are often worthy of the 1812 Overture.

The best (funniest) quote appears on page AN25-13: "The author acknowledges Carl Nelson's abundance of commentary, some of which was useful, during preparation of this work", although the parenthetical statement on page AN25-4, "(ground as I say, not as I do)", is a close second.

21 August 2011

Scope Sunday 6

Speaking of test equipment (it's not all Tektronix around here), I am dumbfounded by the announcement this week that Hewlett-Packard was going to shed its computer business. Wasn't getting into the computer business the whole reason HP spun off Agilent and bought Compaq ten years ago? I predict that in ten years, HP will be a management consulting company, and ten years after that, it will be a zucchini farm. Despite the incomprehensibility of the logic behind this announcement, it does make a recent acquisition of mine even funnier.

On the left is the 1999 HP Test and Measurement Catalog, and on the right is the 2000 Agilent Test and Measurement Catalog (also known as the transition of the coming apocalypse). One good sign was a picture of the HP 200A audio oscillator on the inside front cover of the Agilent catalog, but I'm afraid the other omens were bad (at least bad for HP).

I am reminded of a picture and caption of David Packard, Bill Hewlett, and Bill Gates that Jim had on his bulletin board at work. I've tried to recreate it (and paraphrase it) below.

Packard: "What's that young man talking about?" Hewlett: "Hell if I know, but I know I don't like it."

UPDATE: My friend Eugene sent me a link to this letter, which is an appropriate addition to this conversation:

Dear Fred:

I have no personal knowledge of computers nor does anyone in our organization have any appreciable knowledge.

Sorry we can't help you out in this regard.

Sincerely yours,

William R. Hewlett

Letter from Steve Blank, Elephants Can Dance – Reinventing HP.

19 August 2011

App Note 23

"Micropower circuits for signal conditioning." 24 pages.

This app note discusses techniques for designing extremely low-power circuits.  As Jim says, "Although micropower ICs are available, the interconnection of these devices to form a functioning micropower circuit requires care."

The first two circuits are low-power temperature sensors (the first with a thermistor and the second with a thermocouple).  The next two circuits are strain-gauge amplifiers, which use sampling techniques to achieve micropower performance, exploiting low duty cycle to give low power consumption.  (Many more bridge circuits are coming up in App Note 43.)  The next three circuits are temperature-monitoring applications for the LTC1040, LTC1041, and LTC1042 family of micropower comparator circuits.  Figure 6 is a clever 4mA-to-20mA current-loop thermistor amplifier that uses the current loop signal as the power source.

Figure 9 is a micropower SAR analog-to-digital converter.  Successive approximation is a good technique for low-power A-to-D, but unfortunately, now that SA registers (like the 74C905 used here) have been discontinued, it's harder to implement in discrete form.  Figure 11 shows a micropower single-slope A-to-D converter, which only consumes 100 microamps (the recently discontinued 74C906 could easily be replaced by another low-offset CMOS switch).  Figure 14 is a micropower sample and hold (SAR A-to-D converters require a S&H front end). This circuit cleverly uses the programming pin on the LT1006 to turn down the power consumption in the hold mode.

The best circuit is Figure 16, a micropower 10-kHz voltage-to-frequency converter.  This circuit is also known as "The Zoo Circuit" and Jim wrote a chapter in his first book dedicated to it.  Note the quote on page AN23-13: "A nice day at the San Francisco Zoo…, instrumental in arriving at the final configuration, is happily acknowledged."  Rather than discuss the circuit here, I'll wait until I review his first book.  (However, I will note here that there is an error in the schematic: the base of Q4 should be connected to its collector.)  Figure 20 is a higher-speed V-to-F converter, reaching 1 MHz.

The final circuits are power regulators.  Figure 22 is a switching regulator (using a discontinued 74C907 as the switch, we've got the whole family now).  Figure 25 is a switching regulator that maintains a constant voltage drop across the LT1020 regulator, using the integral comparator.  I like the start-up circuitry here.  Figures 28 through 31 show off other tricks using the LT1020.

The box sections at the end (why not appendices?) cover a number of topics.  Box Section A discusses low-power techniques and the design evolution of the zoo circuit.  Box Section B discusses the LTC1040, LTC1041, and LTC1042 family of micropower comparator circuits.  And finally, Box Section C discusses the effects of test equipment on micropower circuits (or, stated another way, suggestions that might power your circuits from the input source: just turn up the amplitude on that pulse generator).

Best quote (from page AN23-19): For example, everyone "knows" that "MOS devices draw no current." Unfortunately, Mother Nature dictates that as frequency and signal swings go up, the capacitances associated with MOS devices begin to require more power. It is often a mistake to automatically associate low power operation with a process technology. While it's likely that CMOS will provide lower power operation for a given function than 12AX7s, a bipolar approach may be even better.

15 August 2011

App Note 22

"A monolithic IC for 100MHz RMS-DC conversion." 16 pages.

App Note 22 is a first chapter in the sad story of an underappreciated design. This app note discusses the LT1088, an IC building block for RMS-to-DC conversion. Jim advocated, designed, and was the primary evangelist of this product. In addition to this app note, Jim also co-authored the 1986 ISSCC paper that described its function and fabrication. For someone with a instrumentation background, an RMS-to-DC converter is an important functional block. If you look in future app notes, you see that he often used this part in his circuits, and he referenced this app note (and the ISSCC paper) quite a few times. It should have been a popular part (and I think that Jim was proud of it), but unfortunately the LT1088 was a poor seller and has been discontinued.

The first part of the app note discusses challenges for RMS-to-DC conversion, and explains how the feedback approach (Figures 2 and 3) is the preferred solution. The text on page AN22-4 (and Figures 5 and 6) discusses the problems and proprietary solution to the thermal conductivity of the die attach (the complexity of which may be a contributing factor to the part's discontinuance). (The box section on page AN22-15 details a method for measuring the thermal resistance of the die attachment.) Figure 7 is a die photo, showing the extreme symmetry necessary for good performance. (Also, there's an inside joke here about "Counts' Theorem" (that 9 does not equal 10) that I don't fully appreciate. Someone is going to have to explain it to me. Anyone?)

Figure 8 shows a complete application circuit for the part, and its performance is shown in Figures 11 through 16. Figure 16 shows the stereotypical response for a thermal-computation system such as this one, where the positive-going and negative-going transient responses are different (because of the differences in heating and cooling time constants). Figures 20 and 21 show protection circuitry to prevent damage to the LT1088.

As Jim says, "some applications may require buffering the LT1088's relatively low input impedance." Several buffers are shown in Figures 23, 24, and 25. Figures 24 and 25 are from App Note 21's collection of composite amplifiers.

I think the best (most interesting) circuit is Figure 27, the RF leveling loop, using a AD539 wideband multiplier, a wideband discrete amplifier (from App Note 21 Figure 6), and a LT1088 RMS-to-DC converter.

Best quote (from page AN22-1): "Applications such as wideband RMS voltmeters, RF leveling loops, wideband AGC, high crest factor measurements, SCR power monitoring, and high frequency noise measurements require the advantages of thermally based conversion." There are a large number of applications; it's too bad it wasn't a more popular part.

14 August 2011

Scope Sunday 5

In honor of Tektronix's 65th anniversary, which they celebrated this past week, here's another great old scope.  The Tektronix 511 was the first Tektronix product, introduced in 1947.  This scope is a 511A, which was introduced a few years later.

I actually bought this scope (serial number 2520) in California in January 2011 with the hope of showing it off to Jim (and perhaps working with him to restore it).  Unfortunately, he passed away before I even got a chance to tell him about it.

The purchase was almost a nightmare.  Almost.  I bought this scope from a guy on Craigslist.  He didn't know anything about it; he had found it (and quite a bit of old ham gear) in the basement of his new house when he moved in, and he was trying to sell it all for some free space and a few bucks.  (Apparently, the previous owner had passed away, and his kids just sold the house with everything in it.)

When I arrived to purchase it, he had it in the kitchen, set up and plugged in.  He said, "I've tested it and it shows a trace on the screen," and he flipped on the power switch.  My heart stopped!  Conventional wisdom is that these old scopes require a careful power-on ritual after being stored for a long time: the electrolytic capacitors go bad, the high-voltage section goes bad, and dirt and grime get into bad places.  Many aficionados recommend cleaning the scope, checking all the big capacitors (especially the "black beauties of death"), and bringing up the power slowly with a variac and an ammeter.  This guy just flipped the switch (as much as I wanted, I figured screaming "Oh, God, no!" would have been rude).  Thankfully, there was no smoke, and after the trace appeared, I quickly turned the scope off, and said "I'll take it."

I wonder how many old scopes have been damaged by over-enthusiastic sellers trying to demonstrate that "it works" to a buyer?

12 August 2011

App Note 21

"Composite amplifiers." 12 pages.

This application note discusses several topologies for composite amplifiers.  In a sense, it continues the theme from App Note 18 of improving the performance of op amps using additional (often discrete) gain stages.  I love these discrete designs.

Figure 1 shows the basic topology, with a precision amplifier driving the DC point of a high-speed amplifier.  This configuration is sometimes called a Goldberg amplifier, where the precision LT1012 monitors the voltage of the summing junction and drives the noninverting input of the high-speed LT1022 to correct its offset voltage.

Figure 2 uses the same approach.  A pair of discrete FETs replaces the input stage of the LT318A (using the offset pins, like the circuit in Figure 16 of App Note 13 and the low-noise preamp in National AN-299, which Jim also wrote).  Again, the LT1012 monitors the summing junction and drives the noninverting input of the FET pair to correct the offset voltage of the FETs and LT318A.  This circuit uses a lag network (10-pF and 1-k) at the summing junction for stability (and if you look closely at the step response in Figure 3, you can see the long-tail transient of the lag, impacting the settling time).

The FET probe circuit in Figure 4 is a copy of Figure 14 in App Note 9.  Figure 6 improves on the gain of Figure 4, allowing the gain to be precisely unity (or larger).  The circuit in Figure 8 is another DC-stabilized fast amplifier, using a high-speed NPN pair and a LT1010 buffer.  Figure 9 uses a PNP level-shifting stage to trade some speed for output voltage swing.  A FET buffer improves the bias current at the summing junction of the input NPN pair.

Figure 11 is a current-feedback amplifier, using LT1010 buffers as the input and output drivers, and a discrete current mirror for the gain stage.  The LT1001 provides a low-frequency offset-voltage correction loop.  The best circuit, the "Son of Godzilla Amplifier" in Figure 12, replaces the LT1010 buffers with discrete buffers.  This circuit achieves an impressive bandwidth of 110 MHz and slew rate of 3000 V/us (using 2N3904 and 2N3906 transistors!), even at a gain of 20. The two op amps provide feedback offset correction and feedback biasing ("Without closed-loop control, the circuit will quickly go into thermal runaway and destroy itself.")

The final three circuits show a few other composite designs.  Figures 14 and 16 show approaches for low-noise amplifiers, and Figure 17 shows a paralleling trick for LT1010 buffers (this trick was described by Bob Widlar in his App Note 16; see his Figure 43).

Best quote (from page AN21-3): "Note that rise time is limited by the pulse generator and not the circuit."  (He also mentions this deficiency in the caption of Figure 13.)  Jim will fix this embarrassing problem (a slow pulse generator) in a future app note!

10 August 2011

App Note 18 part 2

The box section on page AN18-12, entitled "The oscillation problem (frequency compensation without tears)" is a bit of a muddle. He seems resolutely determined to avoid the phrase "phase margin". Too bad. Some of the explanation seems tortured in order to avoid saying it. (I feel bad about the following write up, because I really don't like this part of the app note, and Jim can't defend himself. I should have just skipped it.)

Rather than his distinction between "local oscillations" and "loop oscillations", I prefer to distinguish between the "intentional feedback loop" and the "unintentional feedback loops". The first step in the construction of any feedback system is to make sure that the major loop, the intentional feedback loop, is stable. If the major loop doesn't have enough phase margin, then the system will be unstable, regardless of how well it's constructed. (If there are minor loops, they need to be stable, too, of course.) The second step is to make sure than the construction doesn't introduce additional negative phase shift to the major loop, or introduce any unintentional feedback loops through parasitics.

In his parlance, "local oscillations" are caused by a subcircuit, usually the boost stage. Most of these are caused by unintentional feedback loops, as he says, "transistor parasitics, layout and circuit configuration," although a poorly designed intentional feedback loop (a minor loop) within the boost stage could certainly be unstable. I think the important troubleshooting step is to compare what you built with what you meant to build (hence my preference for the distinction between intentional and unintentional loops). Nonetheless, the advice is still good. For example, "avoid high f_t transistors unless they are needed" (that is, minimize the high-frequency gain for all your unintentional loops).

(A side note here: He mentions the use of damper networks, but doesn't mention the drawbacks. In Figure 5a the additional gain of the inverters is removed at high frequency by the 100-ohm and 200-picofarad lag. I'd like to see a careful measurement of how badly that compensation network degrades the settling time.)

His next paragraph is a bit tortured: "Loop oscillations are caused when the added gain stage supplies enough delay to force substantial phase shift. This causes the control amplifier to run too far out of phase with the gain stage." This "too far out of phase" statement is confused. The real issue is the phase margin of the loop. If the phase margin of the combined loop (the control amplifier and the boost stage) is too low, then you've got trouble. It's not an issue of being "in phase".

The advice in the next paragraph is slightly off-base. "If the booster stage has higher gain bandwidth than the control amplifier, its phase delay is easily accommodated in the loop." This isn't exactly right, and counterexamples could be constructed (such as when the gain of the booster stage is larger than unity). Again, the real issue is the phase margin of the combined loop. It is usually true that if the gain-bandwidth of the booster is large, then its phase shift at the unity-gain frequency of the loop will be small, and the phase margin will be sufficient. But there's no guarantee. If you want a guarantee, then you need to determine the combined loop transmission and determine the phase margin.

Repeat after me: "Find the phase margin."


08 August 2011

App Note 18 part 1

"Power gain stages for monolithic amplifiers." 16 pages.

This app note contains several different discrete output stages for op amps. The three major themes here are high output current, high output (rail-to-rail) swing, and high output voltage.

The first major theme is current boosting. The first circuits (in Figure 1) exploit Widlar's LT1010 power buffer (up to 150 mA), in the application for which it was designed.  I really like the circuit in Figure 2, which uses the power-supply terminals for unintended purposes. By sensing the current in the supply pins, you can tell whether the output buffer is sourcing or sinking current, and then drive huge currents (up to 3 amps in the MJE2955 and MJE3055) based on that measurement. The power-supply pins don't always have to connected to just power! This circuit is a nice reminder than buffers are really four-terminal devices (and op amps are five-terminal devices), and all of the terminals can be used in a clever design.

Figure 3 is a fast output stage, using a feed-forward path, similar to Figure 8 in App Note 6. The op amp is being used as a low-frequency error servo, while the feed-forward through the JFET provides the high-frequency path, with a slew rate of one thousand volts per microsecond.

The second major theme is "voltage-gain" stages for nearly rail-to-rail output swing. Using CMOS inverters as "linear" gain elements (as in Figure 5a) weirds me out. I just can't get over my distrust of digital circuits to use them this way. Is the gate behavior in the linear region reliable enough? I guess so. Figure 5b uses bipolar transistors to drive closer to the rails at higher currents. The circuits in Figure 5 are run off a five-volt rail; Figure 7 is another (nearly) rail-to-rail output stage, this time for plus-and-minus 15V rails.

The third major theme is high-voltage output stages, with four example circuits. Figure 9 is roughly similar to some of the other output buffers, but using high-voltage transistors and driving the output node to plus-and-minus 125 volts. (I appreciate the comment that the input common-mode voltage limits require a minimum gain of 11 in the non-inverted connection. In other words, "Remember to do the math!")

Figure 11 is a high-voltage stage, similar to Figure 9, but that uses vacuum tubes. (There aren't many (modern) op-amp circuits that require a 12.6VAC filament supply.) Unfortunately, he calls them "Mr. De Forest's Descendants". I know that he is trying to funny, but De Forest deserves no credit for the invention of the vacuum tube. Don't get me started (instead, I'll just refer you to to Chapter 1 of "The Design of CMOS Radio-Frequency Integrated Circuits" by Thomas Lee).

Figure 13 is an extremely high-voltage output stage, driving up to +1000 volts, but powered just from +28V. The basic trick here is the integral boost switching regulator and the transformer. The current limiting is done by the comparator C1 and the diode network, which brute-forces off the oscillator, the darlington drive, and the drives to the MOSFETs. Finally, Figure 15 is implements a bipolar high-voltage step-up stage, by restoring the "polarity" of the output voltage after the transformer and rectifier with a SCR-based synchronous demodulator.

Best quote (from page AN18-8): "The transistor inverter [in Figure 11] is necessitated because our thermionic friends have no equivalent to PNP transistors."
I'll discuss the box section on frequency compensation next time.


07 August 2011

Scope Sunday 4

The Tektronix 7104 is the fastest analog oscilloscope ever produced. Originally designed in the 1970s for the US Atomic Energy Commission, it has a 1-GHz bandwidth and a special CRT that can display single-shot high-speed events in normal room light. I've been looking to buy a Tek 7104 for several years, but I haven't seen any for sale in New England (other than some ridiculous prices and/or shipping on eBay). Then, finally, last fall, something happened and my luck changed.

It started at the MIT SwapFest in September. I hadn't seen a 7104 at the MIT SwapFest in at least three years. Most months, you'll see a 7603 or a 7704 with the occasional 7904. But in September, I finally found a 7104, and I bought it for $175 from a mechanical engineer who didn't appreciate what he had. He knew it was a 1-GHz scope (much more than he needed), but he didn't appreciate it as the precious gem it truly is. The plug-ins were an odd assortment, including one 7a19, one 7a29p, and two 7b70.

Then, not four days later, I was at the surplus store up in New Hampshire, and in front of the store, sitting outside, they had a 7104 (with one plug-in, a 7a29 and no timebases) on sale for $100. Inside the store I picked up a 7b10 and a couple of 7b15 for $20 each.

Then, lightning struck a third time. I won an eBay auction for a (local) 7104 for $153 (but without any plug-ins). I drove an hour and picked it up that Friday.

I hadn't seen a (reasonably priced) 7104 for sale in the northeast in years, and then I find three excellent deals in less than a month.

However, I was petrified to turn them on. I feared that the jugs would have a bad filament or excessive burn-in. I really didn't expect a $100 7104 to be in complete working order.  But I finally got up the nerve to test them. My good friend Eugene came over to my house one Sunday afternoon and brought his Tek 109 pulser. Buzz! We tested the 7a29 and 7a19 plug-ins in a 7904 and then turned our attention to the 7104s.

Throwing caution to the wind, we plugged in the scopes, and (after turning down all the intensity knobs and while holding our breaths) we powered them up. After a fair bit of fiddling, the traces appeared, and the working bandwidths of the systems were confirmed.

So, even better, they all work. The three pictures you see here were taken on three different mainframes (but with the same 7a29 and 7b15). Life is good.

05 August 2011

App Note 17

"Considerations for successive approximation A-to-D converters." 8 pages.

Successive approximation is an important analog-to-digital conversion technique, and there are several clever circuits here, but all of the designs use the AM2504 successive approximation register chip, which is no longer available. Certainly, there are plenty of available monolithic ADC chips that internally use the SAR technique, but there are very few stand-alone register chips anymore, so these discrete-design techniques are dated. (The AM2504 family is obsolete, but ON Semiconductor still makes the MC14549 family; however, I think that's it.) For some reason, this app note feels even more obsolete than App Note 1 and the discontinued LT1005.

Figure 1 shows a basic ADC topology using the AM2504 SAR and an AM6012 DAC (good luck finding one of those, too). The conversion time is quoted at 12 microseconds. Note that the only Linear Technology parts here are the LT1021 reference and the LT1011 comparator. (Also note that the AM2504 in Figure 1 is labeled "SAR Register". Is that like "PIN Number" and "ATM Machine"? Sorry, Jim.)

Figure 3 shows a two-speed scheme (similar to App Note 13 Figure 31) that speeds up the conversion clock on the lower bits (when you don't need as much time for settling because the output of the DAC isn't moving as much), achieving 7.5 microseconds. Page AN17-3 shows how to calculate the required comparator gain, and Figure 5 shows how to use a Schottky-diode-bounded LT318A pre-amplifier to meet the gain spec with the much faster, but lower gain, LT1016 comparator (down to 3.5 microseconds). Figure 7 shows a faster discrete pre-amp, using a cascaded differential pair from a CA3127 transistor array (the little base-current compensation circuit is nice).

The best circuit is the fastest: Figure 9, the highest-speed SAR ADC in the app note. Several techniques for speed are used, including the fast preamp and LT1016. (Note that the voltage reference and comparator are still the only LTC parts, but at least there are three LT1016s.) A feedback loop determines the clock speed (fast for large voltage errors, and slow for small voltage errors, allowing extra time for fine settling) and an active clamp (the 74121 and FET) assists the DAC in quick settling. "The circuit achieves a full 12-bit conversion in 1.8us, about the practical limit with off-the-shelf components" (even if the major components (DAC and SAR) aren't carried on anyone's shelf anymore).

The box section on page AN17-8 gives a descriptive analogy for the SAR technique, and talks about some of additional DAC considerations, such as output capacitance, (meaningful) settling time, and monotonicity. The best quote is (page AN17-8) "The successive approximation technique is probably as old as the first crude weighing scale ever constructed."

04 August 2011

App Note 16 and Sheep

There is no App Note 16. (Well, there is, but Jim Williams didn't write it. Bob Widlar wrote it about his LT1010 fast power buffer.) So we'll skip App Note 16 and go to App Note 17 tomorrow.

In the meantime, speaking of Bob Widlar, here's something completely different. This weekend, when I was looking for the letter that Jim sent to me about my 453 aviation adventure (Scope Sunday 3), I found these pictures that Jim sent to me of Bob Widlar and his sheep.

Bob Pease tells the sheep story here [dead link].

Unfortunately, the above link to National's web site is now dead (thank you, Texas Instruments, that's great!), but here's the text that Pease wrote:
First of all, Widlar did not bring in a goat to chew down the unmowed lawns at National (when the money for gardeners was cut back). That would be absurd. Widlar would not do that. What he brought in was a sheep. I can prove it, because Fran Hoffart showed me a picture of the sheep.

Widlar brought the sheep in the back seat of his Mercedes-Benz convertible. That would be nice to document with a photo, but Fran didn't get a photo of the sheep's arrival. However, Bob Dobkin told me that he drove up with Widlar and the sheep, after Widlar bought the sheep in Morgan Hill for $60. Dobkin said that after the sheep was tied up to a tree in front of National's headquarters, the news photographers only took 20 minutes to show up. At the end of the day, Widlar went over to a bar (Marchetti's) and took the sheep with him. He left the sheep with the bartender.
Originally published in Electronic Design, July 25, 1991.

03 August 2011

App Note 15

This app note suffers from a curious self-contradiction, to which Jim almost admits in the last section: "No commercially available logic, processor or memory family will operate from 1.5V. Many of the circuits described [herein] normally work in logic-driven systems." In other words, there are some great circuit designs here that work within 1.5-volt power supply, but what circuits interface to the ADC, VFC, and S&H? The A-to-D converter and the V-to-F converter have "logic" level outputs, and the sample-and-holds have "logic" level inputs (where "logic" means high (1.5V) and low (0V) voltages), but there is no logic family that works at 1.5V (or at least there wasn't in 1985; ironically, these circuits are probably more useful today than they were then).

But, if you go ahead and use Figure 13's flyback regulator (based on a Bob Widlar design) to produce a five-volt supply to power your logic, why not run the whole system from five volts? In fact, App Note 11 has already covered clever tricks for five-volt circuits, and App Note 8 has already covered power-supply design techniques for battery-powered systems (especially App Note 8 Figure 18). So what are we doing here?

Why do people climb Mt. Everest? Because it's there. Why does Jim design circuits for 1.5V operation? Because he can. Getting active circuits to work at 1.5V is deep magic, and Jim has it. Part of the magic is the LT1017/1018 comparator, part of the magic is Widlar's LM10 op amp, and part of magic is clever circuit design. Figures 1 and 3 show how to use capacitor charging and the low-voltage comparators to implement a V-to-F converter and a single-slope A-to-D converter, respectively. Figure 6 is a ramp-and-hold circuit using the LM10 op amp.

The best circuit is Figure 7, a more traditional sample-and-hold with a JFET switch and a charge pump to get enough negative voltage to turn the switch off. (Is a charge pump cheating? Maybe.) This S&H requires careful hand selection of the JFETs to get a below-500-millivolt pinch-off transistor for Q1, and 500-microvolt matching for Q2 and Q3. Do you want low voltage operation? You pay for it!

Figure 9 is another temperature-compensated crystal oscillator (see App Note 12 Figure 6), using a voltage-boost stage that is shown in Figure 11 here (and also App Note 8 Figure 18 and similar to App Note 6 Figure 7; he must have had a box of Triad transformers).

The App Note concludes with a box section on "Components for 1.5V operation" which basically just says "LT1017, LT1018, and LM10; good luck, sparky!" (He also mentions the LT1004 and LT1034 voltage references.) He discusses the pros and cons of silicon and germanium diodes and transistors (if you have a stash of germanium diodes and transistors, that is). There is also a graph of battery voltage versus life for carbon-zinc and mercury, which is quite interesting.

Best quote (for the specifications, from page AN15-8): "The LM10 op amp-reference runs as low as 1.1V; the LT1017/LT1018 comparator goes down to 1.2V. The LM10 provides good DC input characteristics, although speed is limited to 0.1V/us. The LT1017/LT1018 comparator series features microsecond range response time, high gain and good DC characteristics." Amazing.

02 August 2011

App Note 14

I'm surprised at the shear number of voltage-to-frequency converters that have appeared in these early application notes. Jim really had a V-to-F obsession going on here. So far, we've seen V-to-F converters in App Note 3 (1 circuit), App Note 7 (7 circuits), App Note 9 (3 circuits), App Note 11 (1 circuit), and App Note 13 (3 circuits), and now we've got a whole 20-page app note devoted to them (and there's more to come in future app notes). Did Linear Technology have that many requests for help with voltage-to-frequency applications? Or did Jim just love building them?

Of course, if the progression from App Note 7 Figure 8 (100 dB of dynamic range), to App Note 9 Figure 21 (150 dB), to Figure 1 of this App Note (160 dB) is any indication, he's getting good at designing V-to-F converters. Really good.

This app note is dedicated to ten more circuits that implement voltage-to-frequency converters. The box section (starting on page AN14-19) contains a brief tutorial, showing five different basic topologies (it's good introductory reading before the rest of the app note). Clearly, the best circuit is the star (and start) of the show, the "King Kong V-to-F" in Figure 1. This circuit is an amazing topology (compare it to Figure 21 in App Note 9). The text says "This circuit, similar to those employed in oscilloscope triggering applications..." and if you peek ahead at the references, you'll see a reference to the trigger circuit in the Tektronix 2235. Here's the relevant portion of the schematic from the 2235 service manual:

This circuit is an impressive example of Jim's talent for synthesis. Marrying together the brawn of speed (the trigger circuit with the ECL 10K and 10H gates) and the brain of linearity in the feedback path (the LTC1043 switched capacitor and the LTC1052 chopper amp), this Frankenstein monster really does achieve the best of both worlds. (Staying with the monster-movie theme: it was either that, or a gorilla joke.) A lesser engineer would have just ended up with a muddled mess. This circuit is an example of great design (the feedback path), outright theft (the trigger circuit), and a few gruesome hacks (the linearity correction and the one-gigaohm resistor in the collector of the 2N3904). Respect.

Figure 5 is another Pease-type V-to-F converter, but a "highly modified, high speed variant" of a Pease-type. Hmm. It would be interesting to check Bob Pease's publications in this time period and see if there was some friendly one-upmanship going on between these two former National Semiconductor colleagues.

The rest of the circuits cover a wide variety of V-to-F schemes and functions. Figure 8 is quartz stabilized for improved temperature coefficient. The "ultra-linear" design in Figure 10 includes additional inputs and hooks for a processor-driven gain-and-offset calibration scheme. Figure 12 is a 1.5-volt single-cell circuit exploiting the LT1017 comparator (pressed into service as an op amp, too, here). (More single-cell circuits are coming up in App Note 15 tomorrow). Figure 14 is another sine-wave output VCO (again abusing the under-appreciated AD639; see App Note 13 part 2). Figures 17 and 19 have 1/x transfer functions, and Figure 21 has an exponential transfer function (which, instead of volts-per-hertz, implements volts-per-octave for electronic-music applications). Finally, Figure 23 implements a ratiometric converter, measuring the ratio of two resistors.

The best quote is the (possibly) subtle tweak at Bob Pease (on page AN14-4): "The charge feedback scheme used is a highly modified, high speed variant of the approach originally described by R. A. Pease." In other words, Bob had a good idea, but it was too slow!

01 August 2011

A Bibliography of Jim Williams

After digging up Jim's application notes from National Semiconductor last week, I started thinking that it would be useful to have a complete compiled bibliography of all of Jim's publications. So I spent a little bit of time (OK, a lot of time) searching around, and I've put together a draft of "A Bibliography of Jim Williams". In 2009, Jim told me that he had over 350 publications. I've found 288 so far. Please notify me of any omissions.

The bibliography can be found at http://web.mit.edu/klund/www/jw/jwbib.pdf.

Back to our regularly scheduled Linear Tech app notes tomorrow.

National App Notes

Last Thursday, I mentioned the National Semiconductor Application Note 294 that Jim wrote. On Friday, a commenter referred to a circuit in NSC App Note 299 (which Jim also wrote).  Jim worked for National in the Linear Integrated Circuits Group from 1979 to 1982. During this period, he wrote many app notes, but getting a complete list of his notes is a bit of a mystery hunt. There are several unfortunate reasons for this difficulty:
  1. National doesn't always print bylines with author's names on their app notes.
  2. National regularly deletes old app notes from their archives.
  3. National sometimes updates the publication date of their app notes upon revision.
However, I spent some time digging, and I think I now have a complete list. To find which application notes he wrote, I had to infer Jim's authorship based on the right time period and other clues. One reliable clue was the inclusion of photographs of Jim's Tektronix 556 oscilloscope with the damaged graticule. In other cases, I made educated guesses based on his use of references, footnotes, or subject matter. A reference to one of Jim's past publications is a good hint, a footnote discussing the Hewlett Packard HP200 oscillator is a dead giveaway!

Based on this research, there are (at least?) twenty-one application notes that he wrote.  They are App Notes 256, 260, 262, 263, 264, 265, 266, 269, 272, 285, 286, 288, 289, 292, 293, 294, 295, 298, 299, 301, and 311. Not bad for three years' work!

You can find most of the app notes on National's master list.  For more details on the frustrations of this mystery hunt, see http://web.mit.edu/klund/www/jw/jw-nsc.html.  (After I finish reading all of the Linear Tech app notes, it will be interesting to go back and reread all of his National app notes.  I should have done that first.)