Wideband modern low noise analog front ends

[nimish]

Settling time must have a component included in SINAD, ...

Only if non-linear effects are involved. Any linear filter has a particular impulse/step response and coresponding settling time, too, w/o introducing any additional frequencies / spurs in the spectrum.

Sure, and settling time is inherently nonlinear since it includes, among other things, slew rate limits. If I stick an ideal square wave in I get a nonlinearly distorted square wave out, no?

Also how is the bridged t-coil thing any different from a constant resistance Zobel network? And where would I use it? Anywhere to cancel out unwanted shunt capacitance?

OK apparently the technique is usually called bandwidth extension: https://www.researchgate.net/publication/2983288_Bandwidth_Extension_Techniques_for_CMOS_Amplifiers   What secret sauce was Tek using on top of this? Or did they just realize they could use it?

[mawyatt]

Settling time must have a component included in SINAD, ...

Only if non-linear effects are involved. Any linear filter has a particular impulse/step response and coresponding settling time, too, w/o introducing any additional frequencies / spurs in the spectrum.

Sure, and settling time is inherently nonlinear since it includes, among other things, slew rate limits. If I stick an ideal square wave in I get a nonlinearly distorted square wave out, no?

Also how is the bridged t-coil thing any different from a constant resistance Zobel network? And where would I use it? Anywhere to cancel out unwanted shunt capacitance?

OK apparently the technique is usually called bandwidth extension: https://www.researchgate.net/publication/2983288_Bandwidth_Extension_Techniques_for_CMOS_Amplifiers   What secret sauce was Tek using on top of this? Or did they just realize they could use it?

Here's a couple references on Tek secret sauce magic from the 1960s & 70s.

https://vintagetek.org/electronic-design-t-coils/

And Bob Pease's ref.

https://www.electronicdesign.com/technologies/analog/article/21807815/whats-all-this-tcoil-stuff-anyhow

There is a section in Pease's notes that is an absolute must read if one is interested, actually one should the read the entire article and the book Wideband Amplifiers. This is about a discussion with the brilliant Stanford Prof Tom Lee whom I meet while on a IEEE IMS panel session long ago. Here's a snip from "Talking with Lee at the eFlea";

"“They already had T-coil magic that worked for vacuum tubes, but found that transistors were different enough to require new incantations. They had no proper equations for how to design T-coils for bipolar transistors, because the derivation is really, really hard. The input resistance complicates things enormously. Bob says, ‘Ya know, I think I know how to do that.’ He spent the better part of a year trying to figure it out. He says this is the hardest derivation he’s ever done in his entire career. But he figured it out.

“Tek felt, rightly, that having the T-coil equations for the lossy, asymmetrical case was equivalent to having nukes when everyone else only had rocks and pointy sticks, so they kept this knowledge as a trade secret. The T-coil was one important bit of magic that allowed Tektronix to maintain such a lead on everybody else, because they could get double or triple the gain-bandwidth product out of an amplifier just by adding a capacitor and a pair of coupled inductors in the right place.”"


Interestingly the math behind the Ross solution dates back to K T Wang in 1930, which Bob Ross was exposed to at UC Berkley in 1960s.

Best,

[2N3055]

Settling time must have a component included in SINAD, ...

Only if non-linear effects are involved. Any linear filter has a particular impulse/step response and coresponding settling time, too, w/o introducing any additional frequencies / spurs in the spectrum.

Sure, and settling time is inherently nonlinear since it includes, among other things, slew rate limits. If I stick an ideal square wave in I get a nonlinearly distorted square wave out, no?

Also how is the bridged t-coil thing any different from a constant resistance Zobel network? And where would I use it? Anywhere to cancel out unwanted shunt capacitance?

OK apparently the technique is usually called bandwidth extension: https://www.researchgate.net/publication/2983288_Bandwidth_Extension_Techniques_for_CMOS_Amplifiers   What secret sauce was Tek using on top of this? Or did they just realize they could use it?

Here's a couple references on Tek secret sauce magic from the 1960s & 70s.

https://vintagetek.org/electronic-design-t-coils/

And Bob Pease's ref.

https://www.electronicdesign.com/technologies/analog/article/21807815/whats-all-this-tcoil-stuff-anyhow

There is a section in Pease's notes that is an absolute must read if one is interested, actually one should the read the entire article and the book Wideband Amplifiers. This is about a discussion with the brilliant Stanford Prof Tom Lee whom I meet while on a IEEE IMS panel session long ago. Here's a snip from "Talking with Lee at the eFlea";

"“They already had T-coil magic that worked for vacuum tubes, but found that transistors were different enough to require new incantations. They had no proper equations for how to design T-coils for bipolar transistors, because the derivation is really, really hard. The input resistance complicates things enormously. Bob says, ‘Ya know, I think I know how to do that.’ He spent the better part of a year trying to figure it out. He says this is the hardest derivation he’s ever done in his entire career. But he figured it out.

“Tek felt, rightly, that having the T-coil equations for the lossy, asymmetrical case was equivalent to having nukes when everyone else only had rocks and pointy sticks, so they kept this knowledge as a trade secret. The T-coil was one important bit of magic that allowed Tektronix to maintain such a lead on everybody else, because they could get double or triple the gain-bandwidth product out of an amplifier just by adding a capacitor and a pair of coupled inductors in the right place.”"


Interestingly the math behind the Ross solution dates back to K T Wang in 1930, which Bob Ross was exposed to at UC Berkley in 1960s.

Best,

"Wideband Amplifiers" by Peter Starič and Erik Margan is good explanation behind the principles and math of T-coils.
Starič worked in Tek:
https://vintagetek.org/peter-staric/

[nimish]

"Wideband Amplifiers" by Peter Starič and Erik Margan is good explanation behind the principles and math of T-coils.
Starič worked in Tek:
https://vintagetek.org/peter-staric/

I did, and it basically appears to go over inductive peaking and other BW extension techniques inc T-coils but what I was confused by is:

a) How is this different from an input impedance transformation using a constant-R Zobel network?
b) What was Tektronix's specific advantage beyond applying the well-known constant-R Zobel network -- this had been used in transmission lines and audio for 20 years before Tek existed: https://en.wikipedia.org/wiki/Boucherot_cell invented in a 1919 paper that isn't online (!)

And the short answer is "it's not" and Tek's secret sauce was Ross's use of Wang algebra to quickly compute the symbolic solutions to the Kirchhoff equations.

For whatever reason computing Cramer's Rule on matrices was hard, so they lucked out in that the bridged-T coil results in a symmetric system of equations so Ross could use the Wang algebra to quickly do the tedious at the time computation: Ross explains it here

Duffin and Bott did more work: http://pi.math.cornell.edu/~hubbard/botttalk.pdf and that's basically the SOTA today for network synthesis. Duffin's got some other papers on applications of Wang algebra:

https://homepages.math.uic.edu/~kauffman/DuffinLong.pdf

https://www.ams.org/journals/tran/1959-093-01/S0002-9947-1959-0109161-6/S0002-9947-1959-0109161-6.pdf

https://www.jstor.org/stable/1990850

And you can hire Ross ( via samtec ) to consult on SERDES, where using the T-coil techniques and more is ubiquitous to get the ~50GHz BW

[mawyatt]

@ nimish,

You bring up an interesting topic, suggest you create a IEEE Circuits & Systems, or IEEE Proceedings paper on the subject of the Tektronix T-Coil and Ross's contributions or lack thereof.

I'm sure Professors Tom Lee, Hajimiri, Rodwell and others would welcome, critique and maybe even contribute, and we all would gather a much better understanding of what transpired back then and as you seem to infer maybe it's not that big a deal, altho HP couldn't solve this and you can bet they certainly tried!!

Even in hindsight Lee seemed to have some difficulty, per his comment here from Pease's article:

"A Tektronix engineer, whose name Lee can’t recall, wrote him and asked who at Tek told Lee about the T-coil derivation. Lee told him, “I did this independently, since you guys never published this stuff.”  The Tek engineer said T-coils were a Tek trade secret, to which Lee responded, “I’ve got blood-soaked pages to prove that I did this on my own, because I did it wrong many more times than I did it right.” The Tek engineer then engaged in some good-natured ribbing that Lee only solved the symmetrical case, where both coils have the same value.

That engineer referred Lee to John Addis, of the Tektronix Museum near Portland. The engineer said that Addis was “Mr. T-coil.” Addis was Tek's most prolific and able designer of T-coil-based amplifiers, but he pointed Lee to Bob Ross as the engineer who had derived the equations. Addis then introduced Lee to Ross. Professor Lee recounts the story:

“That started a friendship with Bob, who told this amazing story that starts in China in the 1930s. There was this gifted engineer, K.T. Wang, who was way too smart to work on just his power plant management job. He made up electrical engineering brain teasers for himself and then solved them. Over years of doing this to keep his mind from turning into oatmeal, he noticed there was a pattern to some of these problems. They would be easy to set up, then horribly horrible to solve, and in the end the beast would collapse to something much simpler and very elegant. He wondered, ‘Is there a way to get to the end without going through the middle.’ So, he came up with these weird ad-hoc rules that enabled him to shortcut a lot of the ugly math.

“Not knowing anything about Wang, I had come up with my own method and published what is, as far as I know, the first T-coil derivation in the open literature [Planar Microwave Engineering, Ch. 12 Appendix, Cambridge Press, 2004]. It was not a general method. It was only good for the lossless symmetrical case; it cannot be gracefully extended to anything else. Tektronix had the Full Monty, the lossy asymmetrical case, and I wanted to figure out, who the heck did that, and how? I wanted to meet that guy. Well now I’ve met that guy, the brilliant Bob Ross."

Best,

[mawyatt]

If we revisit the past back in the 1960s the high frequency transistors available were 2N918, 2N2857, and a few others we can't remember. New high frequency transistors weren't introduced often and the designers had to squeeze as much bandwidth as they could with what was available and not bank on a much faster device coming along anytime soon.

This is why the Tektronix Ft Doubler Circuit and the T-Coil were so important IMO, they could get much more bandwidth for a given set of transistors than HP or anyone else.

Best,

[Terry Bites]

This is interesting:

[nimish]

@ nimish,

You bring up an interesting topic, suggest you create a IEEE Circuits & Systems, or IEEE Proceedings paper on the subject of the Tektronix T-Coil and Ross's contributions or lack thereof.

I'm sure Professors Tom Lee, Hajimiri, Rodwell and others would welcome, critique and maybe even contribute, and we all would gather a much better understanding of what transpired back then and as you seem to infer maybe it's not that big a deal, altho HP couldn't solve this and you can bet they certainly tried!!

Even in hindsight Lee seemed to have some difficulty, per his comment here from Pease's article:

"A Tektronix engineer, whose name Lee can’t recall, wrote him and asked who at Tek told Lee about the T-coil derivation. Lee told him, “I did this independently, since you guys never published this stuff.”  The Tek engineer said T-coils were a Tek trade secret, to which Lee responded, “I’ve got blood-soaked pages to prove that I did this on my own, because I did it wrong many more times than I did it right.” The Tek engineer then engaged in some good-natured ribbing that Lee only solved the symmetrical case, where both coils have the same value.

That engineer referred Lee to John Addis, of the Tektronix Museum near Portland. The engineer said that Addis was “Mr. T-coil.” Addis was Tek's most prolific and able designer of T-coil-based amplifiers, but he pointed Lee to Bob Ross as the engineer who had derived the equations. Addis then introduced Lee to Ross. Professor Lee recounts the story:

“That started a friendship with Bob, who told this amazing story that starts in China in the 1930s. There was this gifted engineer, K.T. Wang, who was way too smart to work on just his power plant management job. He made up electrical engineering brain teasers for himself and then solved them. Over years of doing this to keep his mind from turning into oatmeal, he noticed there was a pattern to some of these problems. They would be easy to set up, then horribly horrible to solve, and in the end the beast would collapse to something much simpler and very elegant. He wondered, ‘Is there a way to get to the end without going through the middle.’ So, he came up with these weird ad-hoc rules that enabled him to shortcut a lot of the ugly math.

“Not knowing anything about Wang, I had come up with my own method and published what is, as far as I know, the first T-coil derivation in the open literature [Planar Microwave Engineering, Ch. 12 Appendix, Cambridge Press, 2004]. It was not a general method. It was only good for the lossless symmetrical case; it cannot be gracefully extended to anything else. Tektronix had the Full Monty, the lossy asymmetrical case, and I wanted to figure out, who the heck did that, and how? I wanted to meet that guy. Well now I’ve met that guy, the brilliant Bob Ross."

Best,

My distillation is that Ross didn't invent the bridged T-coil as it's a special case of a constant-R bridged-T Zobel network, but he was able to get the design equations for it as well as asymmetric extensions, which is non-trivial. Zobel had invented them before 1928: https://ieeexplore.ieee.org/abstract/document/6769393, and Boucherot had used a special case even earlier in 1919 and perhaps as early as 1895: https://patents.google.com/patent/US548511A/en for what we'd now call passive power factor correction.

So I think it's fair to say Ross was critical to the development of the T-coil as a practical tool to construct constant-R input filters that allowed Tektronix to maintain an edge against the competition for decades, which is a ridiculous achievement if you think about it. From the Staric and Margan book you can see it getting >2x BW improvement with some tradeoffs.

Now, of course, they are everywhere. The BUF802 data sheet even uses a shunt inductive peaking filter to compensate for the 2.4pF input capacitance so as to maintain the 50 ohm input impedance over a longer bandwidth, though their choice of 6.8nH vs, say, 3.3 nH, is a little odd.

Anyway, https://epubs.siam.org/doi/book/10.1137/1.9781611975826 has a good review and https://link.springer.com/chapter/10.1007/978-3-319-67068-3_21 is good too, co-authored by the guy who invented the inerter / J-damper passive mechanical circuit element as used in F1.

[mawyatt]

Would be interesting to see where researchers are today with ultra-wideband techniques, devices and amps. I know DC to 100GHz was passed a couple decades ago, and DC to THz a decade ago, suspect today some folks are approaching literally  "DC to Daylight" 

Best,

[Martinn]

Now, of course, they are everywhere. The BUF802 data sheet even uses a shunt inductive peaking filter to compensate for the 2.4pF input capacitance so as to maintain the 50 ohm input impedance over a longer bandwidth, though their choice of 6.8nH vs, say, 3.3 nH, is a little odd.

When I read that passage in the datasheet, I wondered whether one could apply this technique to an oscilloscope input through terminator? I just purchased a R&S RTB2k which lacks the input termination (so needs a 50 Ohms BNC through terminator for 50 Ohm sources). My guess would be that for 300 MHz BW it's not really necessary - and had I needed 1 GHz, I would have bought a different scope (which would then have had the proper switchable input termination).

[mawyatt]

Now, of course, they are everywhere. The BUF802 data sheet even uses a shunt inductive peaking filter to compensate for the 2.4pF input capacitance so as to maintain the 50 ohm input impedance over a longer bandwidth, though their choice of 6.8nH vs, say, 3.3 nH, is a little odd.

When I read that passage in the datasheet, I wondered whether one could apply this technique to an oscilloscope input through terminator? I just purchased a R&S RTB2k which lacks the input termination (so needs a 50 Ohms BNC through terminator for 50 Ohm sources). My guess would be that for 300 MHz BW it's not really necessary - and had I needed 1 GHz, I would have bought a different scope (which would then have had the proper switchable input termination).

Evidently Tek utilized these T-Coils in Scope probes and Current probes back in the day, and took advantage everywhere of this unique bandwidth extension knowledge. You can imagine what HP was going thru trying to figure out this T-Coil stuff that Tek was utilizing, and exactly why Tek didn't want to file any patents and kept the Ross solutions a trade secret!!

I mean HP had some really smart folks, and you can bet they certainly were trying!!

http://www.davmar.org/TE/TekConcepts/TE/TekConcepts/TekProbeCircuits.pdf

Of course everything seems simple in hindsight!!

Best,

[David Hess]

When I encountered that design problem, it was nearly impossible to find an amplifier specified for settling to better than 0.01%, which is less than 14 bits.
I agree that my original problem was probably thermal:  we always evaluated multiplexer post-amp settling going from high to low (zero), comparing the switching from low/low and high/low.
In my case, the swing from high to low was less than the slew rate, but the long tail that prevented hitting 16-bit accuracy was very slow and therefore probably thermal.

This is the kind of modern "16-bit" operational amplifier was thinking of:

https://www.analog.com/en/technical-articles/lt1468-an-operational-amplifier-for-fast-16-bit-systems.html

[nimish]

Now, of course, they are everywhere. The BUF802 data sheet even uses a shunt inductive peaking filter to compensate for the 2.4pF input capacitance so as to maintain the 50 ohm input impedance over a longer bandwidth, though their choice of 6.8nH vs, say, 3.3 nH, is a little odd.

When I read that passage in the datasheet, I wondered whether one could apply this technique to an oscilloscope input through terminator? I just purchased a R&S RTB2k which lacks the input termination (so needs a 50 Ohms BNC through terminator for 50 Ohm sources). My guess would be that for 300 MHz BW it's not really necessary - and had I needed 1 GHz, I would have bought a different scope (which would then have had the proper switchable input termination).

Evidently Tek utilized these T-Coils in Scope probes and Current probes back in the day, and took advantage everywhere of this unique bandwidth extension knowledge. You can imagine what HP was going thru trying to figure out this T-Coil stuff that Tek was utilizing, and exactly why Tek didn't want to file any patents and kept the Ross solutions a trade secret!!

I mean HP had some really smart folks, and you can bet they certainly were trying!!

http://www.davmar.org/TE/TekConcepts/TE/TekConcepts/TekProbeCircuits.pdf

Of course everything seems simple in hindsight!!

Best,

A certain W. R. Hewlett invented the distributed amplifier concept (look at the Negative Mutual-Inductance Circuit in that paper), so I doubt HP was unaware.

Was he still around by then? Would have been incredible if the knowledge was forgotten.

[mawyatt]

A certain W. R. Hewlett invented the distributed amplifier concept (look at the Negative Mutual-Inductance Circuit in that paper), so I doubt HP was unaware.

Was he still around by then? Would have been incredible if the knowledge was forgotten.

As we suggested earlier, since you have so much conflicting information regarding Tektronix, Ross and the T-Coil, strongly suggest you create a paper and submit to the IEEE  for a peer review. I'm sure this will catch the attention of many knowledgable folks, maybe some still around from that era that participated in the Tek/HP analog scope wars!!

Back then the HP rep tried his best to get us to use HP analog scopes, even almost giving us them. We always ended up using Tektronix, they were just the masters of these Time Domain Instruments. HP also lacked in the triggering circuitry, Tek's clever use of ECL gates and circuitry just worked better in real world lab use, same as the rest of the Tek scope features.

Of course this is just my opinion, and we had no dislike for HP, in fact almost everything else in our labs was HP!!

BTW Bill Hewlett was certainly around back then as he passed away at 87 in Jan 2001, and was President until 1977, so obviously not suffering from any mental lapses during this time.

Best,

[David Hess]

Back then the HP rep tried his best to get us to use HP analog scopes, even almost giving us them. We always ended up using Tektronix, they were just the masters of these Time Domain Instruments. HP also lacked in the triggering circuitry, Tek's clever use of ECL gates and circuitry just worked better in real world lab use, same as the rest of the Tek scope features.

Of course this is just my opinion, and we had no dislike for HP, in fact almost everything else in our labs was HP!!

The story my father tells is that when he was working at a national lab in the 60s and 70s, the quality of the HP oscilloscopes was lacking compared to the Tektronix oscilloscopes, and that was the determining factor.