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

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.
Deane Jensen's 990 schematic from his original 990 paper from mid 1979 included the output isolator. It also is shown in the slicked-up version of that paper that was published in the AES Journal in 1980. My 990 data package includes the output isolator in the 990 schematic. The text includes the following:

=========
R15 and L3 (?output isolator?) are not part of the basic op-amp ?triangle? and are not included in the 990. They are available separately and are recommended in many applications for best results. See the Jensen engineering report for details.
=========

There simply wasn't room for the output isolator in the 1.125" x 1.125" x .600" package that I use (API-2520 compatible), and isolation components are not normally included in an op-amp anyway. My schematic shows 1% resistor values instead of the 5% values that Deane specified, and a PN4250A transistor (Q10) is used in place of the original 1N914B diode (CR3 in the Jensen schematic).

My 990 data package is available at:

http://www.johnhardyco.com/pdf/990.pdf

The Deane Jensen AES paper "JE-990 Discrete Operational Amplifier" can be downloaded for a fee from the AES, or you can get a printed copy from Jensen Transformers or me just for the asking.

John Hardy
The John Hardy Co.
www.johnhardyco.com
 
Steve, John, thanks a bunch!

That's what I love about the Lab...direct access to the guys who make it all tick :grin:

PRR- Thanks for the photoshop tweak...I posted those as I was headed out for the evening, and didn't have a chance to play with them. I may tweak them all after I get a little sleep. Maybe. :roll:

Steve- The tightness of the layout is really spectacular. I spent about an hour at the office at the end of the day trying to match it all up with the 990 circuit on John's site, but quit after a while. I've spent a good portion of my career as a small signal analog EE, and this is really gorgeous. I assume that I can stick these into the 990 Twin-Servo schematic and life will be good, despite the higher Vios? I noticed that the original version takes feedback from the FB pin, ad the load on the apropriately labelled load pin...any reason not to go with Deane's design?

Also, would I be safe in assuming that the difference between the B and C models is entirely in the parts installed, not the layout or any further tricks? Lower Vios is always a good thing in my opinion.

Yes, a stepped attenuator is in the cards for the mic preamp. I'll probably have a trim pot as well, but not more than a step size worth of adjustment.

John- I really like the idea of taking the feedback from before the load isolator personally. Having the feedback loop as pure as possible is a good thing, as I think one should only add FB compensaton as required, not because it's convenient.
Not having the extra ohms of reactance in there (because it's a resistor with an inductor around it...hardly purely resistive, but you know all that) lets the op-amp handle the transients, while letting the load isolator do it's thing, and not be actively compensated for. Pretty cool IMO. It's how I built all of my LM3886 amps.

Some closing points:
-Yes, I'll be doing the twin-servo, in stereo.
-The pair will get their own case with their own power supply (+/-24V). The octal preamps are getting bumped.
-Stepped attenuators for gain. There should be a META just on those. Anyone have a handy calculator for the ladder network? I can't do rev-log in my head, and the search function isn't working for me at 1am.
-Bargraph meter. Kinda like John's, but not at all. Only 6 LEDs plus clip. Simplicity rules. I'll share the design when I'm done. I plan on tapping the +in on the first stage..buffered with w FET opamp of course. I dont want to disturb the nice signal headed to the 990s.
-Transformers...recommended alternatives to the uber-nice but rather expensive JT-16ATB and JT-11BM? I know I'll get the good stuff eventually, but dropping 4 bills on iron is quite beyond me right now :cry:

Gentlemen, thank you again for the wonderful information. I love the Lab!

-dave
 
dave, I would definitly add the bias compensation network John Hardy shows in the schemo of his one-990-pre. Especially with the higher input bias current of the amps you got.

You may need to change the servo to a noninverting config, as shown in the schemo I mentioned.

Samuel
 
> I really like the idea of taking the feedback from before the load isolator personally.

You MUST take NFB before the isolator! That's the whole point.

The opamp (any amp) is a low-pass filter (no amp goes to infinite Hz). For stable NBF, it has to be a 1-pole low-pass (or good approximation). But the darn real-world is full of capacitors, like cable capacitance. The output stage, no matter how good, has an output impedance, resistive or inductive (rarely capacitive). That plus capacitance gives a second low-pass, another phase-shift, and NFB instability.

The general no-brain quick-fix is to slug the amp so slow that a reasonable amount of capacitance doesn't get you into trouble. TL071s can stand a lot of wire on the output, though generally anything with a jack on it has 47Ω between the chip and NBF, and the jack.

That 47Ω or similar really helps. No matter how bad the capacitance gets, the opamp sees a load no lower than 47Ω at any frequency, can make some gain with low phase-shift, and stay stable.

But in some cases you can't stand 47Ω on the output. Obviously it would clobber an 8Ω speaker-load, even if reduced to 4Ω. Deane was probably looking at bus-drivers that might have from 10K to 75Ω load and wanted less-than 1dB level shift and -80dB cross-talk if somehow an output got patched to the bus. There are times when you need under 1Ω source impedance, and yet be able to drive a mile of cable without NFB instability.

It is a tough problem.

> original 990 paper from mid 1979 included the output isolator.

Yes. It finishes the design and makes it generally useful in any audio situation.

> isolation components are not normally included in an op-amp anyway.

Yes, they aren't. Each case is different. If the 990 drives 2" of wire to a fader, the iso is not needed. If the application needs a true 600Ω output, you gonna have a resistor, and the iso is pointless, likewise if driving an L-C passive low-pass filter. In some cases you might want an iso with more or less L or R. But Deane figured values that would work well in most cases where a naked output risks trouble, thus freeing the system designer from dreary calcs and tests; more time to work on the audio system detailing.

> There simply wasn't room for the output isolator

Aside from purely pragmatic details like "will it fit?", it is a matter of taste and goal. In the module, it needs space and an extra pin. It also costs more than a penny: hand winding isn't cheap, or the machine to wind a billion cheaply isn't cheap. If most users don't need it, why put it in the module? If most users do need it, fitting it in the module avoids stocking and stuffing a little part. They don't have to use it: if the iso is in the module you still need an FB pin and they can take naked output there. But if most users don't need it, they pay an extra dime anyway.

On something as beefy as the 990, it is tough because worst-case the iso-resistor could eat 2W of MHz squeal, so it has to be large. If I were making a discrete 5534 replacement, a 1/4W resistor would do OK (though proportionally still a tight fit on an 8-DIP header).
 
I don't have any worthwhile info to contribute at this point, but I just wanted to pipe up and say that this thread rocks! This is The Lab at its best... lots of solid information.
 
PRR- I know about all the reasons one would want a straight wire with pure resistance in the feedback network for the 990. I've been around a little. :wink: Unfortunately, when posting at 1am after a few :guinness::guinness::guinness:, I apparently didn't come across as coherantly as I would have liked. :oops:

NYD- This is definitely the best, most friendly, and most altruistic analog design community I've ever been a part of, and that includes rooms of scary-good professionals at a Fortune 100 and tables of engineers at IEEE conferences.

More rambling thoughts:

The issue of load protection, especially in high speed amps, can be very vexing, expecially for those of us that want zero phase shift from DC to light with a GBW above 500MHZ in a well-characterized SO8 for $2.95. (yes, I actually like SMT :grin:) The post-feedback node placement of the isolator fixes a multitude of sins by those implementing the circuit, especially in less than ideal situations. Clobbering the NFB loop with filters to slow the amp to a crawl rounds off the transients we know and love. Trying to eliminate all those nasty parasitic Cs will drive a person to madness. And forget trying to maintain a characteristic impedance from 20Hz to 20kHz on FR4 :shock: Personally, I'm a much bigger fan of trying to use the bulk capacitance of PCB planes than to eliminate it. If you can't beat 'em, join 'em, right?

Coming back to the 990 and the twin-servo schematic, it seems to me that the isolator was chosen and placed more as a belt-and-suspenders solution. Looking at the schematic:
The second stage that drives the output transformer will see no less than 1K4 ohms to ground, with a small amount of capacitive reactance (800pF at 1K) from the primary alone. The load resistance will be reflected back to the primary in series with that 1K4 R, making for a pretty darn friendly load, no? I can see the use of a small series resistor, but when driving a transformer with it's own L and C and R, why add more inductance? There's an inherant RLC bandpass in the transformer itself, isn't there? C6 (470pF across the 2nd stage feedback network) should drop the gain close to unity well before the reactive portion transformer's reactance (the high frequency non-R-dominated part) becomes significant.

Oh, Steve and I are chatting offline about the 990B/C differences. I'll ask him about sharing things once we're all done. He did some neat tricks..

Since it's lunchtime, I'm going to mull these questions and more over some food and walk away from the computer for a while. I can't wait to see the insightful and pithy replys :grin:

Thanks again guys

-dave
 
Wow! There is a lot to clarify in the last group of posts.

History: (I was there as a first party witness)--

Deane's work on the 990 was done in the late 1970's. He had already designed an amplifier called the 918 for Pacific Recorders and Engineering in the San Diego Area and the LM394 supermatch pair had just come out and he wanted to make a "universal gain block" with it.

It is important to realize that in those days DC coupling of circuits was very rare -- the sonic degradation due to multiple electrolytic caps in the signal path had not yet come to light. Deane concentrated on the AC performance (both small signal and large signal amplitude and phase response) of the 990. The paper mentions that DC performance can be enhanced, but he did not elaborate. The resistors feeding the two current sources and the diode as part of the current mirror was his original circuit.
It was NOT done for cost savings. Transistors were not that expensive.
Those of us making the original 990 including me and JH decided to make the current mirror out of two thermally coupled transistors for better matching. The original paper mentions this. JH and I also used better parts than the original circuit, since we both were building a no compromise opamp.

The 990-A : Deane added a single transistor switch to turn on the current sources only when there was a positive voltage with respect to ground. This prevented accidental destruction of the 990 when one supply was lost.
Resistors still fed the current sources.

990-B: Was a 990 circuit modification that I developed. The main goal was to make a 990 that would work optimally with all voltages without changing resistors. I figured that if the 990 A already had an extra transistor, why not make it a current source to feed the voltage reference strings of the other two current sources? I developed a temperature compensated current source with an extra diode, a zener and other parts that turn on the 990 at about +9 Volts and allows the 990 to work from bipolar 12 to bipolar 24 with no parts changes and keeps the other two current sources doing 3.2 mA and 5.4 mA over that wide voltage range.
Only about 100 990-B's were made by me, because I immediately developed the 990-C.

990-C: Was developed by Deane and me again during the development of the 990 Twin Servo circuit. The input offset of a 990 B was about 600 to 1000uV typical. By adding a string of two 1N914B/1N4448 diodes between the collector/base of Q10 and the Collector of Q1 the Voltage across the Collector and Emitter of Q1 became much closer to the VCE of Q2. This dramatically reduced the Input Offset Voltage and allowed it to be made very small with small balancing adjustments of the 301 Ohm current mirror resistors. My 990-C's are guaranteed to have <100uV of Vios, but I adjust them in test to less than 20uV.

The 990-C still has high Input bias requirements like all 990's and an input bias compensation scheme should be applied when using in an all DC coupled circuit like the twin servo.

The 990-C is not a selected version of a 990-B. It is a 990-B with 2 diodes added to the circuit and selected 301 Ohm current mirror resistors.
The 990-C requires a different PCB.

The Pin Layout of my TSS-990-A,B,C were all the same and were based on the Pinout used by the 990's built by Pacific Recorders and Engineering.
The Load isolator was included on the module because it was usually used and to make it compatible with the PRE module.

Remember that Deane Jensen made transformers and that one of the most popular uses of the 990 was to drive audio output transformers.
Output transformers have the lowest distortion when driven by a 0 Ohm source. Yet amplifiers benefit from impedance isolating their outputs from the delaying effect of capacitance. The isolator fixes both. It is under 1 Ohm at low frequencies where the transformer cares and it is about 19 Ohms at 1 MHz and eventually reaches 39 Ohms at higher frequencies isolating the feedback loop from load capacitance and enhancing stability.

The feedback pin on my 990 gives access to the 990 output before the load isolator and the output pin is after the isolator. Feedback components should be connected only to the FB pin.

So sorry for the long post. Hope it adds clarity to the discussion.
 
[quote author="NewYorkDave"]I don't have any worthwhile info to contribute at this point ... ! This is The Lab at its best... lots of solid information.[/quote]
yes indeed
love your work guys !
:thumb:
with great stuff like this, I think I'll stick around a while longer.
:thumb: :thumb:

and much greatings and welcomes to Steve H.
:cool:
 
[quote author="Steve Hogan"]So sorry for the long post[/quote]

Don't apologize! This is exactly the kind of stuff we want to read around here... You can't beat first-hand information.
 
I was re-reading some earlier posts and came up with more to say --

Re: the Twin Servo schematic linked from the Jensen website.

The word "BASIC" in the title is very important, as it is not complete. I did not do this schematic, it was done by Dale Roche, one of the other Jensen engineers. This circuit does not have the bias compensation circuitry that I believe will be necessary to prevent the 990's input bias current from being drawn through the pot. The DC flowing in the pot can cause it to scratch. I do not believe this circuit as drawn was ever breadboarded, and I would certainly do that before laying out artwork. All of the power supply decoupling caps are also missing. I would use 2 each 1000uF/35 Volt Panasonic FM series caps (or your favorite) as close to each 990's power supply pins as possible. Note that the output load isolator shown in that schematic complements the one on my 990 module. If you use a JH 990-C you should use two isolators.

If you can't afford the output transformer, just run it single ended with the output terminal of the 990 directly connected to pin 2 of the XLR. The output impedance is low and with the second load isolator from ground to pin 3 of the XLR, the single-ended output has balanced impedance to ground and should work fine into typical balanced or unbalanced equipment. I would highly recommend biting the bullet, however, and getting the JT-16-A/B input transfomers. At Jensen, I always told customers that you only cry once when buying a Jensen -- that's when you write the check! Its sound quality is a big part of what makes a Jensen Twin Servo among the very best preamps available. I made major internal improvements to he JT-16-A/B in the early 1990's and it is the best mic input xfmer I know.


The JT-11-BM output transformer is of bifilar construction. Primary and Secondary are wound simultaneously with two primary and secondary wires side by side the entire length of the winding. Bifilar output transfomers (the only ones anyone should use at the send end of a line) have enormous capacitances between primary and secondary. At high frequencies they are a dead short high side primary to high side secondayr and from low side primary to low side secondary. As a result the line capacitance to ground is free to reduce the phase margin of the line driver opamp. The output transformer does NOT substitute for a load isolator. The load isolator(s) are especially helpful when driving the somewhat squirrelly transformer load.
 
[quote author="Steve Hogan"]Re: the Twin Servo schematic linked from the Jensen website....The word "BASIC" in the title is very important, as it is not complete.[/quote]
I thought as much Steve. The diagram/layout (I do them in parallel) I'm working on has both local decoupling and on-card re-regulation/filtering. I have a pile of low ESR 4700uV 36V radials, so I'm using those. Should be plenty, especially with the soft-start supply I'm working on. I'll be posting the diagram later on once it's approaching something that will actually work.
[quote author="Steve Hogan"]The JT-11-BM output transformer is of bifilar construction....As a result the line capacitance to ground is free to reduce the phase margin of the line driver opamp.[/quote]
That explains a lot. I was under the assumption that it was a singular set of windings, not bifilar, so I completely missed the interaction of primary/secondary capacitance. Mea culpa. That much capacitance will certainly need isolation from the feedback network. The ringing at higher frequencies would be enormous, and likely catestrophic if left undamped to oscillate.
Thanks for that amazing history of the 990. Wonderful reading. My interest in..."budget transformers" is more due to impatience than anything else :grin: I have a timescale in my head for getting the project truly finished, but dammit, I'm going to want to use them before then. :twisted:
Posts like that one are never too long Steve. We'll never turn down freely contributed knowledge, especially when it comes from the proverbial horses' mouth! Having participants like you, JH, PRR, NYD, and the rest I'm forgetting right now is really, truly, amazing. The transfer of knowledge is more valuable than a lot realize.

Samuel- JH's bias comp. network is in the design as of now. I suppose I should use that in both stages, in spite of the fact that only the second stage is actually driving a transformer, in the interests of matching the bias currents throughout?

OK guys, my brain is getting sore. I'm going to take an evening to let all of this soak in. I sure hope this topic keeps going though...I don't think we've had a better discussion here in a while!

Thanks a million y'all

-dave

edited thanks to my inability to spell, and PRR's watchful eyes.
 
> JH's bias comp. network ...I suppose I should use that in both stages, in spite of the fact that only the second stage is actually driving a transformer, in the interests of matching the bias currents throughout?

As Steve says: DC error also makes the pots scratchy. In my world, <1uA flowing into a 1K pot is no big deal, but then I'm a low-tech pretty tolerant guy. After paying for a couple 990s and that iron and pots, I'd be inclined to use a bias-current compensation scheme, if it were well proven in audio. (An unproven or IC-maker suggested bias comp scheme may make more audio trouble than it cures.)

In fact the output offset voltage, AD706-servoed, probably won't give a large E-I transformer much trouble. Pot-scratch may be more important.

If the odd pot is a problem, figure a switch. Now pot-scratch is less of a problem, since switch contacts can carry 1uA quite nicely, and any offset-shift is masked by the shift in audio level.

Yeah, calculating the switch resistors is a pain, especially since you have two gain-set networks. But a switch also frees you from the "ganging" of a dual-pot. I'll ask Steve and Dan: does it make sense to stagger the gains, leave the input amp at fairly high gain as long as possible? That gives lowest noise and gets the output stage into a high-NFB condition as soon as possible (though neither are really an issue with the 990 and these impedances).

> The input offset of a 990 B was about 600 to 1000uV typical. By adding a string of two 1N914B/1N4448 diodes between the collector/base of Q10 and the Collector of Q1 the Voltage across the Collector and Emitter of Q1 became much closer to the VCE of Q2. This dramatically reduced the Input Offset Voltage

Huh. I had to look at that, but you are (of course) right. Reading down from V+, Q10 collector would be 2Vbe, Q3 collector would be 3.5Vbe. About a 1V difference. If Q1 Q2 had infinite collector impedance, no problem. But counting on my fingers, if the Mu of Q1 Q2 is 1,000, then this causes about 1mV (1000uV) error.
 
[quote author="PRR"]Watch yer language![/quote]
Sheesh....I can't seem to engineer or spell right today. I wonder what the trifecta will be? :oops:
As Steve says: DC error also makes the pots scratchy. In my world, <1uA flowing into a 1K pot is no big deal, but then I'm a low-tech pretty tolerant guy. After paying for a couple 990s and that iron and pots, I'd be inclined to use a bias-current compensation scheme, if it were well proven in audio.
I'm not at all a fan of scratchy pots personally, so I was planning on including the bias comp. no matter what. I was more curious if there was a better reason. All I can figure is it'll reduce the DC offset, which is always a good thing.
In fact the output offset voltage, AD706-servoed, probably won't give a large E-I transformer much trouble. Pot-scratch may be more important.
Well, it won't exactly be pot-scratch, since I'm planning on some rotary switches and a pile of 1% resistors, but the pops aren't nice either. Plus, it's just 'good housekeeping' as one of my professors used to say.
Yeah, calculating the switch resistors is a pain, especially since you have two gain-set networks. But a switch also frees you from the "ganging" of a dual-pot. I'll ask Steve and Dan: does it make sense to stagger the gains, leave the input amp at fairly high gain as long as possible?
Funny, that was the first question in a half-completed post I tossed after I started to reply to yours. Great minds run in the same rut? I'm not sure where te diminishing returns are as far as that goes. I think JH uses matched gains in his preamps, but I won't presume to know until he himself replies.

I've started on an excel sheet to do the resistor calculating, but so far it's turning into a disaster. I'm thinking that someone somewhere already did the work, and as a fellow LABer he or she would gladly contribute either the calculator or the values directly to the common good. Search really didn't help, but I'm tired and may have missed a critical word. I'm thinking 31 steps. Differential gains are making my head hurt a little. I obviously need a little more time to let this all sink in. Don't let that fool anyone tho...keep it going!

-dave
 
Ignoring some minor (0.1% or 0.01dB) servo-insert details. and possible fraction-dB error at very high gain:

The gain is IIG * (1+Rf1/Rs1) * (1+Rf2/Rs2) * OIG, where

IIG= Input Iron Gain, about 1:2 or +6dB for JT-16
OOG= Output Iron Gain, about 1:0.9 or -1dB for JT-11 loaded in 600 ohms

TS-gain.gif


Conditions:

Rf+Rs should not be "too low", where 75 ohms is way-low and 600-1,000 ohms is more comfortable.

Rf1||Rs1 should be "low" compared to the 600~800 ohm source impedance at JT-16. (A similar issue arises for Rf2||Rs2; actually Rf2||Rs2 can be higher than Rf1||Rs1 by about 1+Rf1/Rs1 so it tends to be no problem.)

(1+Rf1/Rs1) and (1+Rf2/Rs2) should not be less than 2 (I think) to keep the 990 stable.

In the published plan, Rf1+Rs1 is 2,253 ohms, worst-case Rf||Rs is 563 ohms. A somewhat lower impedance might be measurably better, but pot-values are limited, and anyway the gain at this worst-case condition is so low that noise is moot.

Gain: (1+Rf1/Rs1) is minimum 1+1,210/1,043= 2.16 = 6.7dB, maximum gain is 1+2,210/43.2= 52.15 = 34.3dB, and both nets have the same gain.

So suggested minimum gain is 6dB+6.7dB+6.7dB+(-1dB)= 18.4dB, max is 6dB+34.3dB+34.3dB+(-1dB)= 76.7dB

Taking IIG and OIG (transformers) as fixed gain, the two amps give gain of 12.4dB to 70.7dB total, 6.2dB to 35.3dB each.

Total range of gain is 76.7dB-18.4dB= 58.3dB

> I'm thinking 31 steps.

Yo can get a 31-throw 2-pole switch?

I assume 31 switch positions which is 30 steps. 58.3dB/30= 1.9433dB per step. I'll assume you like "2". And that is really 1dB per step in each stage.

If you want a dead-simple calculation, find a "Switched potentiometer" calculator, tell it you want a total resistance of 2K or so, and 1dB steps from -6dB to -35dB. (Or 0dB to -35dB and throw out the top settings.) (For the above calc, tell it 36 steps and -36dB, fixed slope.) Normalized to 2K ohms: 591 to FB, 122, 109, 97, 86.6,... 5.46, 4.87, 4.34 to ground. (No, that dang Java App won't let you cut/copy the values to text... I'll let your fingers do the transcribing.)
 
If Steve Hogan says the JT-11-BM is bifilar, it is bifilar.

On the other hand, (and perhaps what I may have said on RAP), there is also the "JT-11-BMQ", which is a variation of the normal JT-11-BM. I use the "Q" version in the M-1, M-2 and Jensen Twin Servo mic preamps that I manufacture. It's a long story regarding the JT-11-BMQ, and Steve Hogan would know the precise details. My basic recollection is that Deane Jensen wanted to try a QUAD-filar version of the basic bifilar JT-11-BM. Beyond that, I don't know anything about the details and methods of the quadfilar windings.

Apparently there was little or no improvement in performance with the quadfilar winding method, and the winding method was considerably more expensive (or something like that). Whatever the reasons, Deane decided it was not worth the effort. But the DC resistance of the windings was slightly different, so Deane decided to go back to the bifilar winding method for the JT-11-BMQ, but with a slightly different number of turns on the windings compared to the basic JT-11-BM (still a 1:1 ratio). This would make the bifilar "Q" transformers the most closely matched in performance to the early quadfilar ones in case anyone wanted to add channels to an early Jensen Twin Servo, etc.

I'm sure I missed a point or two, but I think that gives a general idea. Thanks.

John Hardy
The John Hardy Co.
www.johnhardyco.com
 
> QUAD-filar version of the basic bifilar JT-11-BM. ... Apparently there was little or no improvement in performance

Worth a try, but for a simple 2-winding transformer I don't think it would make a big difference. With two windings and perfect bifilar, dang near all the flux has to cut the other winding. 50% at one wire diameter, so for a winding of more than a few layers you have 90% coupling, maybe 97% coupling with the number of layers used for a 600Ω winding. This on top of the 99.9% coupling in the iron even with crummy winding technique. That suggests bandwidth of what, 10,000 * 30 = 300,000? 10Hz-3MHz? Or 3Hz-1MHz if you like bass.

Perhaps the idea was to lay the interwinding capacitance differently, since that is the big drawback of bifilar. At the moment I'm not seeing a way to make it better, or not even 0.7X which is only a 1.2X shift in top-resonance.

At an extreme: hammer the wires to ribbon as wide as the window and spiral-wind. 2,000 turns of 1"x0.0005" ribbon as two 1,000-turn windings. Need some magic-thin insulation, and the fine foil would be sure to rip or wrinkle. And the capacitance would be fat.

I thought the big McIntosh tube outputs were wound many-filar, since they had to couple plates to cathodes very well to avoid glitching, and couple tubes to load/feedback to get performance to justify the great cost. But I've never dared to open one and see. (CJ?!) I've suspected that the output/FB windings were not all that intimate with the tube windings, that the -filar was mostly about keeping the tubes coupled plate to grid and push to pull.
 
I have enjoyed interacting with you all --Thank you for your kind comments.

Once again I must jump in and set the record completely straight re: JT-11-BM and the differences between the JE-11-BMQ and JT-11-BMQ Output transformers.

When Deane was working on developing the Jensen Twin Servo Mic Preamp Circuit, the concept was to have an absolutely no-holds barred set of transformers. We were making some custom output transformers for a company in Oregon who makes very nice PC controlled Distortion Measuring equipment. That transformer used a twisted quad of wire in which the relative positions of the two primary wires and the two secondary wires were precisely controlled. The twisting of the wire made the high frequency coupling of the primary to the secondary very tight. The leakage inductance was very small and as a result the frequency response extended into the MegaHertz with almost no ringing at all.
The twisted wire took up a lot of room in the bobbin, however, and necessitated a lower total number of turns, making the transformer have a bit less low frequency headroom. The JE-11-BMQ used twisted quad wire connected as a bifilar output transformer.
In the early 1990's, Jensen changed to computer controlled Swiss-made winding machines. These ultra precision machines with precisely controlled tensions could lay in the wires of a bifilar winding so precisely that there were no crossovers at all. If you look at a JT-11-BM or BMQ which is made on the new equipment you will see red wire through the nylon bobbin on one side and green wire showing through on the other. The JE-series output transformers were never that precise -- they were made on much cruder winding machinery. The high frequency response was so much more consistant on the new equipment, that we decided to abandon the twisting of the wires on the new JT-11-BMQ. Instead we put on more turns and adjusted the wire sizes, etc. to have exactly the same losses in the audio band as the old JE-11-BMQ which was made with twisted quad wire. The result was a transformer which had almost the same hf characteristics and better low frequency characteristics as the old part, but was enormously less hassle to build. The identical losses made it compatible with older JH modules. The JT-11-BM has more turns of finer wire so it has a bit higher DC resistance and higher losses, but it has the same precision bifilar winding techniques as the JT-11-BMQ.

Re: the gain setting of the Twin Servo Mic Preamp. The Jensen-published "BASIC" circuit that has been discussed here has the pot connected differently than traditional Jensen circuit suggestions. Deane always designed the preamps with a fixed Feedback Resistor along with a fixed Feedback capacitor which was chosen to give the amplifier the same bandwidth over the entire gain range. A high gain limit resistor (43.2 Ohms in the traditional twin servo) was chosen and was connected close to the -Input of the opamp to isolate the -Input of the opamp from HF peaking caused by stray capacitance from the -input to ground. This is especially important if the gain controls must be several inches away from the 990. The variable gain resistor was placed between ground and the gain limiting shunt resistor. The variable resistor can be a pot, a switched set of resistors, relays with an R2R attenuator, anything, really.

A fixed (1K) resistor for the feedback resistor allows me to use a really high quality resistor in that location which maximizes the sound quality. I use the Caddock TF020R 1K for that critical location with great results.

I have recently modified my twin servo mic preamp cards with a relay that switches out the 2nd stage for studio situations where the twin servo has just too much gain even with it turned down all the way. I use a pad with my twin servos that is optimized for the input impedance of the JT-16-A/B, not because the transformer will overload, but because the input levels are so hot that less gain is necessary.

There has been some discussion re: how to set the gain in the two stages.
The general idea that Deane had and I believe still makes sense is to use equal gains in both stages. At low gains, the signal is so hot that noise is not going to be a problem, even though there might be some academic noise improvement by running the first stage with higher gain. The lowest distortion comes with each amplifier running with the minimum gain needed to accomplish the overall gain of the two stages, so we chose to run both stages with the same gain. The Twin Servo topology generally never asks a single 990 to do more that 30dB of gain, where a single stage would have to do 40 or 50 or even 60 dB because the input transformer has only 5.6dB of Gain. The two-stage approach sounds much better (especially when high gains are required) because the 2 amplifiers each run with lots of feedback (whoops, dirty word?) and very, very low distortion.
 

Latest posts

Back
Top