I found a small error in your circuit: the value of Rlocal1, and Rlocal2 are slightly different:
-Rlocal1 = {1e6/(a+1u)}
-Rlocal2 = {1e6/(a+1m)}
I changed the second one to be the same as the first one - see the attached files.
Of course, this doesn't change the results: there are practically no differences in the output distortions with/without the feed back resistors.
thanx, I changed one on the fly when I could see a little I(Rlocal) in @1, @2 steps doing a "sanity check" plot just to be sure the R was really in/out of the circuit when I thought
both the same is a little more logical - but being "out of the circuit" by 1000x or 1e6 x shouldn't matter
just to clarify - my position isn't that the JC-3 inner feedack R can never have any "benefical" effect whatsovever on the circuit distotion performance - if you look at my simplified sim you will see that I only claimed the "lowering drive impedance" at the internal node to reduce (PIM) distortion heuristic was flawed, since the inner loop feedback doesn't "perform" in that respect I see no reason to avoid the (large) added output stage Verror reduction by using the loop gain in the global loop
I think there is also going to be little improvement in the input stage performance at audio frequencies when you consider that you can have the same feedback, less the ~10% loading drop of the output stage by using just global feedback
excepting possible (single harmonic?) cancellation interactions which may arise with all stages distortions modeled - clearly more output stage distortion does give more inteaction possibilities
both the same is a little more logical - but being "out of the circuit" by 1000x or 1e6 x shouldn't matter
just to clarify - my position isn't that the JC-3 inner feedack R can never have any "benefical" effect whatsovever on the circuit distotion performance - if you look at my simplified sim you will see that I only claimed the "lowering drive impedance" at the internal node to reduce (PIM) distortion heuristic was flawed, since the inner loop feedback doesn't "perform" in that respect I see no reason to avoid the (large) added output stage Verror reduction by using the loop gain in the global loop
I think there is also going to be little improvement in the input stage performance at audio frequencies when you consider that you can have the same feedback, less the ~10% loading drop of the output stage by using just global feedback
excepting possible (single harmonic?) cancellation interactions which may arise with all stages distortions modeled - clearly more output stage distortion does give more inteaction possibilities
Last edited:
I thought I would have been forgotten by now. I try to keep a low profile these days.
I don't really have a 'dog' in this fight. I find the models only approximate, especially because they do not emulate the original parts that I first used in 1974, when I made the closest version to this specific design as a horn tweeter amplifier with 108 or so SPL output at 1W. The amp only worked in this project at 4KHz and above.
Looking at the QUALITY of the substitutions, I find that the output devices are not necessarily more linear than what I originally used, and the greatest improvement of substituting the Toshiba devices for the Siliconix devices that were originally used was slightly more open loop gain.
The REASON for adding the secondary loop was to make an amp design based on the principles used in the earlier (1973) Lohstroh-Otala design originally developed at Philips Research Labs. So instead of using resistors to 'throw away' the excess gain, I simply fed it back to the input, in order to LOWER THE DRIVE IMPEDANCE to the complementary darlington output stage, that ideally runs class A, and at the BETA PEAK of the output devices. This cannot be a 'bad thing' to do, as the alternative was to throw away the extra gain. ONLY with this early prototype, did I use a single pair of power output transistors, the 2N5884-6, that are more linear than higher voltage versions of essentially the same devices. The actual JC-3 used 2 complementary pairs of output devices running at 1A per output pair. How on earth, any significant higher order terms could be generated in this design is a mystery to me, as it is virtually class A in every stage, at any working voltage or current (for the most part). Where is the 'kink' that will create the higher harmonics?
Looking at the QUALITY of the substitutions, I find that the output devices are not necessarily more linear than what I originally used, and the greatest improvement of substituting the Toshiba devices for the Siliconix devices that were originally used was slightly more open loop gain.
The REASON for adding the secondary loop was to make an amp design based on the principles used in the earlier (1973) Lohstroh-Otala design originally developed at Philips Research Labs. So instead of using resistors to 'throw away' the excess gain, I simply fed it back to the input, in order to LOWER THE DRIVE IMPEDANCE to the complementary darlington output stage, that ideally runs class A, and at the BETA PEAK of the output devices. This cannot be a 'bad thing' to do, as the alternative was to throw away the extra gain. ONLY with this early prototype, did I use a single pair of power output transistors, the 2N5884-6, that are more linear than higher voltage versions of essentially the same devices. The actual JC-3 used 2 complementary pairs of output devices running at 1A per output pair. How on earth, any significant higher order terms could be generated in this design is a mystery to me, as it is virtually class A in every stage, at any working voltage or current (for the most part). Where is the 'kink' that will create the higher harmonics?
The resistors cause almost 50dB more sensitivity to what is left of the crossover distortion. Even if you reduce it a lot some might come through. It's not rocket science the circuit has 80dB of gain out to a few kHz, but the resistors limit this to 50 (~34dB). What's left is a simple voltage loop. Secondly the simple thirds of the non-linear C multiply by being "in the loop".
Last edited:
I would like to point out to everyone that we did not always design amps in exactly the same way that IC's are made today.
Back in the 1960's we had somewhat limited devices, BUT we got BETTER spec sheets on the ones that we could get, so we knew more about the devices, and we cared more about optimizing the devices for lowest distortion.
Also, in the 1960's, we normally used the 'current drive' model of circuit amplification, rather than rely on Vbe deviations. This made us more sensitive to an output stage's distortion from beta nonlinearities, AND this happens when the output stage is CURRENT DRIVEN, which it essentially is, in the JC-3 without the additional feedback resistors. This is one of the reasons that we were so generous with the quiescent currents of each stage, often using a TO-5 transistor where today a TO-92 package might be used. This meant that the current change was less and the beta remained close to optimum. Today, especially in IC's, I can easily believe that higher order distortion artifacts can be easily generated, and the only reason they are not apparent is that they are suppressed by the negative feedback, and the noise floor of the the test equipment, but this was not the case with the Levinson JC-2 or the JC-3.
Today, I do not use the multi-loop feedback approach, because I use complementary mosfet drivers where the driver transistors for the output once were. This allows (to a greater extent) the allowing of a high impedance drive impedance from the previous stage. Instead of trying to force the open loop frequency by reducing loop feedback, I just make faster circuits, with output devices that are about 10 times faster than what I had in 1974. I do not achieve 20KHz open loop bandwidth, but I do have a higher open loop bandwidth than many amplifier designs, and I like it that way.
Back in the 1960's we had somewhat limited devices, BUT we got BETTER spec sheets on the ones that we could get, so we knew more about the devices, and we cared more about optimizing the devices for lowest distortion.
Also, in the 1960's, we normally used the 'current drive' model of circuit amplification, rather than rely on Vbe deviations. This made us more sensitive to an output stage's distortion from beta nonlinearities, AND this happens when the output stage is CURRENT DRIVEN, which it essentially is, in the JC-3 without the additional feedback resistors. This is one of the reasons that we were so generous with the quiescent currents of each stage, often using a TO-5 transistor where today a TO-92 package might be used. This meant that the current change was less and the beta remained close to optimum. Today, especially in IC's, I can easily believe that higher order distortion artifacts can be easily generated, and the only reason they are not apparent is that they are suppressed by the negative feedback, and the noise floor of the the test equipment, but this was not the case with the Levinson JC-2 or the JC-3.
Today, I do not use the multi-loop feedback approach, because I use complementary mosfet drivers where the driver transistors for the output once were. This allows (to a greater extent) the allowing of a high impedance drive impedance from the previous stage. Instead of trying to force the open loop frequency by reducing loop feedback, I just make faster circuits, with output devices that are about 10 times faster than what I had in 1974. I do not achieve 20KHz open loop bandwidth, but I do have a higher open loop bandwidth than many amplifier designs, and I like it that way.
You bias the output at 2 amps and deliver 8 amps peak, this leaves residual error.
I'm struck by the fact that in the last several months we've dissected designs like the AD797 and the JC-3 designed decades ago and still relevant. In fact, both are still Plenty Good Enough to Great even with today's perspective. Makes me wonder how fast technology *really* changes.
Yeah, the Internet is wonderful, but cell phones are the spawn of Satan, so ya have to average.
Thanks,
Chris
Yeah, the Internet is wonderful, but cell phones are the spawn of Satan, so ya have to average.
Thanks,
Chris
The original amp had a total power supply voltage of +/- 24V max. The output would blow up, if more was used. The original load was well defined at 8 ohms. Peak output near clipping was not often anticipated, because that meant 120SPL from the tweeter above 4KHz, hardly normal from a monitor speaker.
It is true that the JC-3, in its original form could exceed 4A peak, but it would be almost instantaneous, and difficult to detect by any human ear. Later designs with more than 50V power supplies tend to use multiple transistors, optimized to cross over to class B operation with minimal 'glitch'. In fact, this is one thing that I work hard to do properly. To emulate this, the emitter resistors, (Re) would have to be reduced to .05ohm, and 4 output pairs would be used to get the 2A quiescent current.
Driving this simple emulation to maximum output, rather than nominal output, is not appropriate.
It is true that the JC-3, in its original form could exceed 4A peak, but it would be almost instantaneous, and difficult to detect by any human ear. Later designs with more than 50V power supplies tend to use multiple transistors, optimized to cross over to class B operation with minimal 'glitch'. In fact, this is one thing that I work hard to do properly. To emulate this, the emitter resistors, (Re) would have to be reduced to .05ohm, and 4 output pairs would be used to get the 2A quiescent current.
Driving this simple emulation to maximum output, rather than nominal output, is not appropriate.
how many ways do I have to say it?
there is the 3rd way - recognize that either of you 2 options above are simply wrong, a false choice since they do not achieve the stated goal of distortion reduction via lower Z of the inner feedback point in this circuit topology
whatever the validity of your and Otala's Phillip's work, probably to more complex amps - it doesn't apply to the JC-3
the point of my sims is that the RESULT of the inner feedback R in this circuit is not lower distortion via lower impedance at the inner feedback point, the closed loop Z of the inner node is not reduced by the inner feedback R compared to the same total feedback used in the global loop only
Bode sensitivity analysis and my "full" sims do not show reduced distortion with the inner feedback R with injected error current at the Darlington input node - the Ierror sensitivity is the same with or without the inner feedback R of the JC-3 circuit
in my sims I adjust the outer loop feedback to compensate for the removal of inner feedback R and the nominal AC loss thru the loaded output
when this is done the closed loop gain of the amp doesn't change when the feedback is moved
my view is that you still get "the same" sensing of the (V)error caused by current injection at the Darlington's base thru the global feedback to well beyond audio frequencies, nearly the same "information" about the inner node still reaches the input stage suming node from either the 2x 1 Meg R or an added ~450 K R in parallel with the outer loop feedback
there is the 3rd way - recognize that either of you 2 options above are simply wrong, a false choice since they do not achieve the stated goal of distortion reduction via lower Z of the inner feedback point in this circuit topology
whatever the validity of your and Otala's Phillip's work, probably to more complex amps - it doesn't apply to the JC-3
the point of my sims is that the RESULT of the inner feedback R in this circuit is not lower distortion via lower impedance at the inner feedback point, the closed loop Z of the inner node is not reduced by the inner feedback R compared to the same total feedback used in the global loop only
Bode sensitivity analysis and my "full" sims do not show reduced distortion with the inner feedback R with injected error current at the Darlington input node - the Ierror sensitivity is the same with or without the inner feedback R of the JC-3 circuit
in my sims I adjust the outer loop feedback to compensate for the removal of inner feedback R and the nominal AC loss thru the loaded output
when this is done the closed loop gain of the amp doesn't change when the feedback is moved
my view is that you still get "the same" sensing of the (V)error caused by current injection at the Darlington's base thru the global feedback to well beyond audio frequencies, nearly the same "information" about the inner node still reaches the input stage suming node from either the 2x 1 Meg R or an added ~450 K R in parallel with the outer loop feedback
Last edited:
But could you get the same distortion levels with only 20 dB of feedback around the ENTIRE LOOP, by reducing the Gm of the input stage further, perhaps, rather than putting the extra gain back into the input from the driver stage? Or would a passive pair of resistors to ground be a better method to get 20dB of feedback, overall. Please remember, I am paralleling Lohstroh-Otala, in this design example.
Lohstroh on circuit simulation . . .
"During my later work in digital research I used circuit simulation in order to understand the principal properties of a digital circuit. Setting all the parasitic capacitances and/or resistances to zero for the circuit except in one place gave you good insight. You can do the same for an analog circuit. Set all capacitances in the device models to zero and use only a single capacitor in the circuit; for instance, the Miller capacitor of the input stage. The amplifier’s dynamic behavior will now only be determined by that one capacitor.
Excite the amp with a step impulse, possibly with a sine wave superimposed on it, check the signals at all important nodes, and then re-introduce all the parasitics step by step to find out where some limiting factors come from. This might be time consuming, but you learn more from it than by including all parasitics right away. In this way, you can 'play' with a circuit and look at open-loop bandwidth and nonlinear behavior such as IMD and TIM, to mention just a few." Jan Lohstroh
from Jan Didden's My interview with Dr. Lohstroh
"During my later work in digital research I used circuit simulation in order to understand the principal properties of a digital circuit. Setting all the parasitic capacitances and/or resistances to zero for the circuit except in one place gave you good insight. You can do the same for an analog circuit. Set all capacitances in the device models to zero and use only a single capacitor in the circuit; for instance, the Miller capacitor of the input stage. The amplifier’s dynamic behavior will now only be determined by that one capacitor.
Excite the amp with a step impulse, possibly with a sine wave superimposed on it, check the signals at all important nodes, and then re-introduce all the parasitics step by step to find out where some limiting factors come from. This might be time consuming, but you learn more from it than by including all parasitics right away. In this way, you can 'play' with a circuit and look at open-loop bandwidth and nonlinear behavior such as IMD and TIM, to mention just a few." Jan Lohstroh
from Jan Didden's My interview with Dr. Lohstroh
Attachments
Hi,
Try dissecting a Harman Kardon Citation 2!
I have "The Internet" on my "Cellphone" (we call it a mobile) and a library that includes the legendary John Curl Collection, all of Self's books (need a laugh occasionally and sometimes a reference), Robert Cordell's book, dozens of old tube books, hundreds of datasheets, several deacdes worth of Audio Electronics Mags like "Valve", "Sound Practices", Glass Audio, Audio Express, Speaker Builder, Klang + Ton, Hobby HiFi and tons of Audio Comics mainly for reference and tons several gigabyte of Martinist, Rosicrucian and other Esotheric literature, the whole Baha'i library (means most of the worlds major religeous writings, almost all decent science fiction published up to the 90's and some newer stuff and a ton of other good books.
Yes, you are right. The Spawn of Lucifer. Ex Lucis Ad Lucem.
Ciao T
Try dissecting a Harman Kardon Citation 2!
I have "The Internet" on my "Cellphone" (we call it a mobile) and a library that includes the legendary John Curl Collection, all of Self's books (need a laugh occasionally and sometimes a reference), Robert Cordell's book, dozens of old tube books, hundreds of datasheets, several deacdes worth of Audio Electronics Mags like "Valve", "Sound Practices", Glass Audio, Audio Express, Speaker Builder, Klang + Ton, Hobby HiFi and tons of Audio Comics mainly for reference and tons several gigabyte of Martinist, Rosicrucian and other Esotheric literature, the whole Baha'i library (means most of the worlds major religeous writings, almost all decent science fiction published up to the 90's and some newer stuff and a ton of other good books.
Yes, you are right. The Spawn of Lucifer. Ex Lucis Ad Lucem.
Ciao T
Last edited:
Which brings up the old joke. That's wonderful and those two suitcases attached to it? They are the battery!
So has it occurred to any of the modelers that the two feedback resistors can only be close in value and not exactly the same. For 1974 I would expect two 1% resistors to match to .3% or so. I don't think that is a disadvantage in this circuit.
- Status
- Not open for further replies.