So I've got these 990s....

[PRR]

> it happened to end up very close to your version, would ours be fair game for the public?

Clean-room reverse engineering is legal (ask IBM vs Phoenix and Compaq).

Clean-room new-flavor engineering is perfectly legal- Coke can't sue Pepsi for making brown vanilla soda.

I'm not sure how "clean" this room is, with so many of the key players tossing in details. But the forum is all wide-open to the public, and these guys should know what to share and what to keep close to the vest.

If you broke into John's office and stole the secret plans, he has you on criminal trespass. Assuming the size of his lock was in proportion to the value of the trade-secret, he can't legally lose trade-secret status that way.

(If you work for John and can access the plans, he should make you sign an NDA.)

If John leaves the secret plans behind at StarBucks, and you find them, then John has not taken reasonable care to preserve the trade-secret, and loses their trade-secret status.

I Am Not A Lawyer. If you have a trade-secret, either don't tell anybody or tell your lawyer first.

Anyhow..................

It is hardly likely we would come up with the EXACT same thing as some other design. Or that we would come up with something along the same lines that was both good and very-different-details from the several public and proprietary designs. There is a commonly accepted range of gain, more or less, and 10Ω versus 9.1Ω is not a significant design concept. There are four ways to stuff servo up two stages, all quite obvious variations to any quasi-experienced designer. The four choices may give different performance, and the Gurus must have thought hard about that and may not be ready to share their hard-thought wisdom. But JH says in one version the servo config changed late in the design, so it probably works any which way, maybe a little better audio one way (or maybe just a pragmatic choice).

Also if we did hit on the exact same thing, I believe the best thing for John and Jensen to do is keep quiet. If they admit that "our" design is "the same" as "their" design, their design is no longer a trade secret. If they keep mum, their trade-secret is not exposed.

Anyway: I could split some wood, take some measurements, and "duplicate" a Stradivarius Violin. Anybody want to put down a $250,000 deposit to get me started? Do we really think that 5-digit numbers from another fiddle are all I need to build a "same as" fiddle? There's a lot about a fiddle, or a mike-amp, that can't be written down. Even in this case, where the main lumps and small parts are all commercially available, the way you put them together is part of the magic. I don't just mean this-connects-to-that, but layout and orientation and maybe even state of mind of the builder.

 

[dpaton]

IAANAL

I'm well acquainted with cleanroom engineering, as well as the limits to what you can legally borrow from a teardown, having been the guy doing the teardown reports, and I'm in 100% agreement with you. As for the rooms relative 'clean' quotient, it's a public forum. Anything here is fair game I think, independent of the contributor.

I was looking for some kind of blessing from JH, since he seems to be very near the center of it. I have a hunch that since the forum was the point of dissemination that we're in the clear, but I like to be all the way sure. I'd hate to make someone like JH mad...especially since he's been so helpful.

In the mean time, I'll be (slowly) finishing the updates, and trying to build a managable parts/source list. :green:

-dave

 

[PRR]

> I'd hate to make someone like JH mad...

Oh, I sure hope not!

But this is not new ground we are digging. Nor traditionally secret ground.

Deane published lots of details about the 990. A cynic might say "self-promotion", but he could have wrapped more buzz-words around less technical meat and sold just as many 990s. A cynic would note that he didn't spill every bean, but you can only fit so much meat in an AES paper and a few technical notes. It is actually one of the most lucid and comprehensive audio-amp tutorials ever written.

With a good amp, the rest is 99% plain resistor-picking. That's the virtue of op-amp topology: the application design can focus on the application, not the DC biasing and other non-audio problems. You pick an impedance magnitude and some ratios.

Lots of little gotchas. As Steve reminds us, capacitance on the "-" node leads to MHz oscillation. You didn't put a cap there? What about the foot-long leads to the front panel? Ooops. And what happens when a non-perfect pot/switch gets dirty? If you break a gain-set connection, the gain goes to either unity or infinity. Infinite gain is hard on speakers and ears. As one of John's product sheets explains in detail, contact resistance in a pot can give some very strange effects. As Deane's paper says, an amp that is stable on the bench may be very unhappy driving the long cable to the other studio. Doug Self's paper on power amps points out several "small" details that can lead to astonishing increased THD: taking feedback from the wrong point around the output stage throws class AB spikes back to the input even if it looks the same electrically (never forget parasitics).

JH and the other Gurus have been thinking and studying amps like this for 30 years, every day. We weekend dabblers, even with hints, are not going to catch up with them nor rediscover all the little tips. We thank them for what light they want to share! But I don't feel I am pious enough to warrant any "blessing", and commercial operations aren't really in the blessing business unless there is some mutual benefit.

 

[PRR]

> This extra open loop gain gives the 990 superb audio capabilities. A careful examination of the open loop gain and phase response plot in the AES paper reveals that the 990 does not have a typical 6 dB per octave slope to its open loop response. The open loop response has kind of a "hump" to it so that it ends up having much higher gain bandwith product in the audio range than its actual 10 MHz unity gain frequency would suggest. The trick with negative feedback is that it must be carefully applied. The internal compensation of the 990 is extraordinarily well thought out to maximize phase margin and stability.

Hey, no kidding! At 20KHz, the effective GBW is like 70MHz! I don't have the AES paper handy and anyway I wanted to see for myself. From a rough SPICE model:

Red plot is gain in dB. The cursor shows 77.3dB gain at 10KHz. 77.3dB is 7,300:1, times 10KHz is 73MHz.

The 70MHz GBW at 20KHz is due to the 150pFd cap in the pole-split network against the 15Ω emitter dynamic resistance. Beyond the audio band, the emitter inductors throw-away some extra gain so by 1MHz the effective GBW is 22MHz. Another 2:1 of gain is lost somewhere to give a 10MHz unity-cross frequency.

It is hard to see the slope at this scale. The blue plot is calculated effective GBW: gain at a specified frequency, time the frequency. From 200 to 20KHz it runs about 70MHz GBW, but falling off above the audio band. At any frequency, the effective feedback is (roughly) the GBW at that frequency divided by (closed-loop gain times frequency). So at gain of 100 and 20KHz, we have 70,000,000/(100*20,000)= 35 or around 30dB of NFB.

The green plot is the phase response. (Ignore the left side: my DC-bias network and rough modeling is screwing that up.) In a perfect amp it would stay at 180 to infinity. That's impossible: an amp has to fall-off and fall-off implies phase shift. The simple elegant compensation gives 90 degree phase shift through the fall-off range until the amp gain is less than unity (or the specified noise-gain). Look at any 741 datasheet.

90 degree shift is easy to get and very stable. But we can really live with 120-130 degree phase shift. There is no natural 120 shifter, but we can combine a wide-range 1-pole 90 degree shift with added narrow-band 90 degree shift to get 120 degree shift over an octave or two. This dumps excess gain: in this case, we need to get to unity-gain well before the big-die output devices go limp (around 20-50MHz). Also high gain in the MHz area is just a pain for system stability. Here the emitter inductors add phase over about 3:1 of frequency, approx 250KHz-700KHz. The 130Ω in the pole-split compensation adds another kink around 1MHz. The R-C networks under the Vas emitter then bump-up the phase as we approach 10MHz. I omitted the output devices: if they were modeled we probably would not have that phase-bump at 30MHz.

So the phase response is 90 degrees 500Hz to 10KHz, 100-130 degrees 20KHz to 10MHz or more.

This is classic 1950s-textbook Bode-plot feedback stability criteria, except in textbooks they often shaved the stability margin to the bone (to get "best" results from slow amps), and this is conservative in line with the idea that audio should not be slapped-around, just eased into cooperation.

> The internal compensation of the 990 is extraordinarily well thought out to maximize phase margin and stability. -- Q.E.D.

 

[PRR]

> The 30 Ohm emitter resistors are necessary for high frequency stability

To quibble/expand: they are not "necessary", you can just increase the compensation cap and be stable.

But that reduces slew rate and maximum full-voltage output frequency.

Using simple pole-split compensation, there is a strong relationship between stability and slew rate. National's "Tutorial Opamps" explains it well. Stability depends on input device transconductance, while slew rate at that compensation depends on input device current. With tubes and FETs, the Gm/mA is so low it isn't a big deal. BJT transistors have "too much" Gm for the mA they flow. With typical parts, a naked BJT input compensated for unity gain will be slew-limited at a few volts of output level.

Putting in the emitter resistors breaks the relationship. We can run high current for good slewing, yet set the gain relationship lower and stable. That's old news; in fact we didn't have slew trouble when we "had" to use emitter resistors just to control offset voltage (or when even good transistors had high ohmic emitter resistance), the problem came with good consistent BJTs that could be used without emitter resistance.

But by the nature of the beast, the required resistor is sure to increase noise voltage. Naked noise is related to 1/Gm, and to get any benefit the added resistor has to be larger and thus more noise voltage than the BJT.

Deane saw the noise improvement possible with "small" inductors. He picked values that are low noise up well beyond the audio band. Supersonic noise rises a bit but that is not a problem.

It also jiggles both the compensation/slewrate factor, and the phase plot. You can pick a unity-gain frequency and not be locked into a 6dB/oct slant from there. You can decrease the size of the compensation cap and thus improve slew without loss of stability margin.

> The 30 Ohm emitter resistors are necessary for high frequency stability

Oh wait. If you mean coils without resistors, you get a 2-pole response with the compensation cap and certain instability. So in that sense they are necessary for stability.

> example of a similar but not identical concept, check page 545 of Tremaine, second edition. A 5mH inductor is used in series with the cathode of the input tube to cause the stage gain gain to fall at above-band frequencies, with little detriment to the AF gain.

First: note that Sams' draftsmen can't copy a schematic; R1 C1 surely go where I have drawn them, not in series with the 300K.

Hmmmm. Feedback amp with coil stuffed up the underside of the input device. Topologically it is identical to the 990 coils, except that HF gain falls to zero, not a fixed value. Slew is not a problem because input stage Gm is low, the main rolloff has gain after it, and anyway there isn't a lot of NFB going on to aid instability. Maybe just 8dB. You have to get two poles just right to be unstable there. The 5mH inductor actually tends to decrease stability here, but not enough to be a problem. It may also phase-shift enough to give a small ring followed by a steeper rolloff: such semi-stable gimmicks were popular in tubes.

Noise is not an issue. The noise resistance of the tube (say 3K) is higher than the circuit resistances (1.8K), and the transformed mike resistance is even higher (around 20K). Output noise is just mike thermal noise.

It is interesting prior art. I doubt its designer foresaw how useful it would be with far hotter devices. Deane claimed on those improvements. How this published plan affects Deane's patent position, I have to leave to the lawyers. It could be that his claims are valid, yet unenforceable if the infringer claims some other reason for using a choke. Electronic patents have always been a screwy field, as bad as plows. Worse, because you can sketch a schematic in seconds but back when steel plows were high-tech you needed a model to get a patent. If you can draw something not already on file, and find some non-trivial claim for it, you can get a patent. (Today you don't even need a good claim.)

 

[John Hardy]

A 2-stage mic preamp with input bias current compensation and DC servo circuitry is not the easiest thing in the world to build PROPERLY. There are dozens, perhaps hundreds (even thousands) of little details that are critical, and take many years to learn.

Steve Hogan and I have been (in my humble opinion) extremely helpful and generous with information and some of the pieces to the large puzzle of the Jensen Twin Servo 990 Mic Preamp. I do the relatively crazy thing of including the complete schematic of my MPC-1 card in my M-1 data package. Likewise with the schematics for the MPC-600 and MPC-3000 cards in those data packages. I do this so that people will hopefully know what they are getting, and they will understand and appreciate the attention to detail, the rich Corinthian leather (oops, that's the Chrysler Cordova from the 1970s, commercial by Ricardo Montalban).

However, I would not be particularly thrilled if someone were to take all of this generously given information and turn it into a competing product. If someone wants to build a preamp or two for his or her personal use, have fun. A competitive product, I will not offer any blessings (to whatever extent, if any, my blessings are significant).

Yes, there is clean-room reverse-engineering, virgin work, etc. And my blessings don't amount to much legally (or, perhaps, in any way). Then there is respect. Or is there none of that left these days? Then there is capitalism. Competition of the free marketplace. And the proper balance of all of those things.

Well, that should be specific enough, and vague enough.

John Hardy
The John Hardy Co.
www.johnhardyco.com

 

[Flatpicker]

[quote author="John Hardy"]However, I would not be particularly thrilled if someone were to take all of this generously given information and turn it into a competing product...[/quote]John, since you mention this I've always wanted to ask what the majority of the people who buy your 990s do with them? I don't know of any other commercial preamps using these. Beside DIYers like us buying them, do they go into custom consoles or something? I can't imagine people using someone?s op amp in a product without their blessing. Talk about biting the feeding hand... :shock:

 

[soundguy]

the people on this forum who contribute regularly never have been a problem and so far as I can tell, probably never will be. Its the spineless scum manufacturers that just lurk here, never contribute and then many months later release a design that has been discussed in detail here that are the real problem. We should all consider this when posting anything that is close to our hearts. Big big big props to the designers who are bold enough to contribute here knowing very well that this is the case.

dave

 

[Samuel Groner]

Looks slick, but there must be an off-by-one error in it some place.

PRR, I cannot find any error or bug in my spreadsheet; if you put in suitable values in column E, it should work perfectly...

Samuel

 

[dpaton]

After a nice chat on the phone with the company IP guy, I don't think I'll be posting the schematics and board layouts directly. If folks want the compiled genius of the forum, they can email and ask for it, and I'll likely respond, but I have absolutely no desire to enable the aforementioned spineless manufacturers to do what it is they do, and hopefully a little due-dilligance will keep things copasetic, which is I think all anyone wants around here :green: If they want it, they can figure it out themselves.

As always, more important things are taking over my life, so things are delayed, but not forgotten.

-dave

 

[PRR]

> I cannot find any error or bug in my spreadsheet

Your sheet is correct. I was misunderstanding the purpose of column E.