I see this stuff 1st hand. Very, very frustrating. Why? Because my customers are so afraid of getting their ***** sued off, that they will not look at solutions that even smell like something in a patent, even after we have gotten independent opinion that its not violating anyone's patent.
I am afriad its true. Patents are granted for crazy things.
Hi Godfrey,
This is a very good point, especially for those who prefer wide open loop bandwidth. It also underscores what I have said for a long time: that wide open loop bandwidth does not preclude high amounts of feedback, where in this case amount of feedback refers to that available at 20 kHz - which is what really counts the most.
BTW, another key point that you have covered is that when one wants wide open-loop bandwidth, one usually must restrict the low-frequency open-loop gain in some way with most amplifier topologies. Many do this by resistively loading the VAS, which is a bad idea because it makes the VAS work harder. As you have done, restricting the open-loop gain at low frequencies with feedback is the way to go.
Nice results!
Cheers,
Bob
Bode's Maximal Feedback uses flat "working frequency" gain - by managing the peaking where the flat slope meets the higher order roll off he gets an "extra" octave of flat loop gain for the same unity loop gain intercept
I've shown the distortion advantage of higher audio frequency loop gain in the TIM debates, it seems worthwhile to just let the gain increase as much as the Q's allow at low frequencies - this can involve a 1st order slope when Q output C rolls off gain for a while before conductance terms flatten it out
particularly when you consider that music spectral content is often claimed to have ~ 3kHz power bandwidth it appears to me worthwhile to keep decreasing distortion by letting loop gain rise at least that far down into the audio frequency range
and I've posted the 2nd order loop gain, "flat" audio frequency gain in a simplified sim before:
http://www.diyaudio.com/forums/soli...erview-negative-feedback-215.html#post1434254
(sinice the sim is only exploring input tanh distortion vs loop gain I use spice tricks to just synthesize the loop gain, not a "realistic" VAS, compensation network)
I've shown the distortion advantage of higher audio frequency loop gain in the TIM debates, it seems worthwhile to just let the gain increase as much as the Q's allow at low frequencies - this can involve a 1st order slope when Q output C rolls off gain for a while before conductance terms flatten it out
particularly when you consider that music spectral content is often claimed to have ~ 3kHz power bandwidth it appears to me worthwhile to keep decreasing distortion by letting loop gain rise at least that far down into the audio frequency range
and I've posted the 2nd order loop gain, "flat" audio frequency gain in a simplified sim before:
http://www.diyaudio.com/forums/soli...erview-negative-feedback-215.html#post1434254
(sinice the sim is only exploring input tanh distortion vs loop gain I use spice tricks to just synthesize the loop gain, not a "realistic" VAS, compensation network)
Last edited:
Hi jcx,
These are good points, especially in regard to allowing continuing loop gain increase as frequency decreases in the audio band. It's useful to bear in mind that while higher-order distortions, like the 7th harmonic, have their harmonic amplitudes increase at higher frequency due to falling negative feedback loop gain, higher-order IM products do not behave the same way. What we more often hear or perceive is IM products, not harmonics, especially when higher frequencies are involved.
A good example is the results of a CCIF IM test where 19kHz and 20 kHz signals are present together. In this case, the higher order odd-order IM products are at lower frequencies - 7th order IM products are at 16 kHz, benefitting from higher feedback loop gain in an amplifier where loop gain falls with increases in frequency. The situation is even more extreme with the more benign even-order products, which lie at much lower frequencies, such as the second order product which lies at 1 kHz. This component will be decreased by a much larger amount due to the largely increased loop gain at 1 kHz as compared to 20 kHz in an amplifier where loop gain rolls off at 6 dB/octave.
Cheers,
Bob
is this why very high gain opamps appear to perform very well in distortion measurements?
Hi Andrew,
Yes, I think this is a part of it. Even conventionally compensated op amps tend to have a much higher gain-crossover frequency than power amplifiers, and this helps a lot. They are also not usually being called upon to have huge voltage and current swings, like a power amplifier.
Interestingly, the 30-year-old 5534 has barely measurable distortion under reasonable conditions (I used it in my THD analyzer, for example, and got great performance), but most audiophiles don't think of it as a high-end performer in terms of sonics these days. It was not made with a fully complementary IC process, for example, so NPN-centric circuits had to be used for most of it. In particular, its output stage did not benefit from fast PNP transistors. Today's good op amps have a complementary bipolar process, sometimes even oxide-isolated. Also, the basically class-B output stage of most op amps probably affects their sound more than their measurements.
Nevertheless, there are superb op amps out there for audio.
Cheers,
Bob
try for low order "smooth" open loop nonliearity and apply lots of feedback
you might point to the amount of loop gain - I'd guess the 1st op amps to be considered even acceptable had 4 MHz GBW, the 5534 can be 20 MHz with de-comp - the lowly TL07x is very good if you put a buffer in the loop
and op amps are often used in lower total gain circuits in line level applications than power amps (phono pre amps should definitely use more than one op amp)
with today's faster complementary processes unity gain stable devices with 50+ MHz GBW and 10-100 Hz open loop fc are available with low distortion at 50mA output current
in my preferred composite op amp circuits I put several op amps inside the feedback loop for as much as 120 dB loop gain at 20 KHz, 100 Mhz CFA can make good output op amps (although they do have low open loop gains of 70-90 dB and consequent "flat loop gain" over audio frequencies)
you might point to the amount of loop gain - I'd guess the 1st op amps to be considered even acceptable had 4 MHz GBW, the 5534 can be 20 MHz with de-comp - the lowly TL07x is very good if you put a buffer in the loop
and op amps are often used in lower total gain circuits in line level applications than power amps (phono pre amps should definitely use more than one op amp)
with today's faster complementary processes unity gain stable devices with 50+ MHz GBW and 10-100 Hz open loop fc are available with low distortion at 50mA output current
in my preferred composite op amp circuits I put several op amps inside the feedback loop for as much as 120 dB loop gain at 20 KHz, 100 Mhz CFA can make good output op amps (although they do have low open loop gains of 70-90 dB and consequent "flat loop gain" over audio frequencies)
I certainly would not read it that way. I think that the point is that even MM RIAA preamps require enough equalized gain (typ 40 dB at 1 kHz, 60 dB at LF) that to get that gain with only one op amp would not leave enough loop gain for good distortion reduction by NFB.
Cheers,
Bob
http://www.diyaudio.com/forums/solid-state/45794-high-loop-gain-composite-op-amp-circuits.html
shows a low distortion diode clipper inside the feedback loop - you may not like the step recovery on clipping but I would usually try to size line level or headphone amplifier circuits made with these op amp techniques with considerable headroom to avoid clipping (structure system gain for >120 dB SPL output)
I believe low single digit uS clipping recovery details to unlikely to be audible especially if the clipping itself is really infrequent - few dyanmic tranducers will have as much as an octave more than 20 KHz 2nd order Fc
it is possible that better clipping recovery could be managed as promised by BJ Lurie's "nonlinear dynamic compensator" examples but I don't know of a systematic way to design such circuits - the greater access to internal stages in discrete circuits helps - as has been shown in a couple of amps here at diyAudio
phono preamps should show distortion, accuracy improvements with 2 properly chosen op amps in the feedback RIAA correction loop - you do have to shape the loop gain with local feedback around the amps to give linear stability with the RIAA feedback reducing to unity high frequency gain
typically a fast output op amp can have flat enough gain around the slower input op amp dominant pole loop gain intercept to avoid problems - fast unity gain buffers in the loop have long been popular but I see advantages to added loop gain at lower frequencies
a high current, fast, low distortion CFA output op amp allows driving very low RIAA feedback network Z at high frequency (~10-100 Ohms feedback R could be appropriate for MC) for low noise while eliminating the heavy loading thermal, ps pin common impedance errors of the input op amp
loop gain also has powerful linearizing effect on diff input linearity - 10x smaller input diff V from +20 dB audio frequency gain from a 2nd op amp in the loop gives ~ 100x less distortion from the 3rd order gm nonlinearity term that dominates in well balanced diff pair at the composit input
the same added 10x audio frequency gain means the input op amp swings 10x less V at its output too, as well as just having to drive the output op amp high Z positive input - almost certainly staying in "deep Class A" even with only 100-200 uA bias current
shows a low distortion diode clipper inside the feedback loop - you may not like the step recovery on clipping but I would usually try to size line level or headphone amplifier circuits made with these op amp techniques with considerable headroom to avoid clipping (structure system gain for >120 dB SPL output)
I believe low single digit uS clipping recovery details to unlikely to be audible especially if the clipping itself is really infrequent - few dyanmic tranducers will have as much as an octave more than 20 KHz 2nd order Fc
it is possible that better clipping recovery could be managed as promised by BJ Lurie's "nonlinear dynamic compensator" examples but I don't know of a systematic way to design such circuits - the greater access to internal stages in discrete circuits helps - as has been shown in a couple of amps here at diyAudio
phono preamps should show distortion, accuracy improvements with 2 properly chosen op amps in the feedback RIAA correction loop - you do have to shape the loop gain with local feedback around the amps to give linear stability with the RIAA feedback reducing to unity high frequency gain
typically a fast output op amp can have flat enough gain around the slower input op amp dominant pole loop gain intercept to avoid problems - fast unity gain buffers in the loop have long been popular but I see advantages to added loop gain at lower frequencies
a high current, fast, low distortion CFA output op amp allows driving very low RIAA feedback network Z at high frequency (~10-100 Ohms feedback R could be appropriate for MC) for low noise while eliminating the heavy loading thermal, ps pin common impedance errors of the input op amp
loop gain also has powerful linearizing effect on diff input linearity - 10x smaller input diff V from +20 dB audio frequency gain from a 2nd op amp in the loop gives ~ 100x less distortion from the 3rd order gm nonlinearity term that dominates in well balanced diff pair at the composit input
the same added 10x audio frequency gain means the input op amp swings 10x less V at its output too, as well as just having to drive the output op amp high Z positive input - almost certainly staying in "deep Class A" even with only 100-200 uA bias current
Last edited:
Hi jcx,
Thanks for describing this. It is quite a creative approach that appears to yield very good performance. I have to admit that I have not generally built circuits with two op amps in the loop in this fashion. I must admit that I also shy away from circuits where the phase lag goes beyond 180 degrees before the loop gain crossover. For things like RIAA preamps, I'm less creative, dividing the required gain across multiple closed loops.
Cheers,
Bob
It's finally happened, I've come across the impenetrable local oscillation of the outputs at... 50MHz.
It's only on the positive signals. This is a unique output stage in that the outputs are driven from a capacitive impedance, from the collectors of the drivers. Just an EF driven by collectors. I remember there as an RC netork across the B-C junction that orked most times? I thought it as in Cordell's book but I just can't find it. It is basically a snubber across the B-C I thought.
The drivers are 2SC4793/A1837, and the outputs are MJL0302/3281 or an unmatched mix, like suggested by Ostripper at one time. The oscillation begins hen I go over 20V rails. I already have 2.2R carbon comp base stoppers.
I kno there are a number of ays people have solved this problem in triple output stages but I liked the snubber option the best.
It's only on the positive signals. This is a unique output stage in that the outputs are driven from a capacitive impedance, from the collectors of the drivers. Just an EF driven by collectors. I remember there as an RC netork across the B-C junction that orked most times? I thought it as in Cordell's book but I just can't find it. It is basically a snubber across the B-C I thought.
The drivers are 2SC4793/A1837, and the outputs are MJL0302/3281 or an unmatched mix, like suggested by Ostripper at one time. The oscillation begins hen I go over 20V rails. I already have 2.2R carbon comp base stoppers.
I kno there are a number of ays people have solved this problem in triple output stages but I liked the snubber option the best.
Last edited:
Hi keantoken,
I came up with the snubber idea in my MOSFET power amplifier with error correction. I don't recall having used it in BJT output stages, although it might be useful there as well. I refer to them as gate zobel networks on page 226 of my book in the MOSFET chapter, but unfortunately do not show a schematic there with them. I cite as a reference my MOSFET amplifier paper. That paper is available on my website at CordellAudio.com - Home.
The basic idea is to damp out resonances that can be part of forming an oscillator topology, such as a Hartley or Colpitts. The technique may very well work with BJT arrangements as well. Don't rule out the possibility it is the drivers oscillating.
It was a little unclear whether the problem you saw was in simulation or prototype, but it sounds like the latter. In that case, Triples of any type can be susceptible to power supply wiring inductances and HF feedback through the power rails. When using ordinary thriple EFs, I often recommend some small-resistance R-C filtering in the rail as it travels back to the driver and thence the pre-driver (i.e., two LPF stages, often with resistances on the order of 1 ohm and 10 ohms, respectively). If the shunt capacitors are of a moderate value, the R-C combination not only acts as an LPF, but also acts as a Zobel on the rails.
Cheers,
Bob
Yes, this is a real prototype. I simulated it and then built it. I have a 40MHz signal generator so I have some indication hen simulations and reality don't converge.
Thanks for the info. I onder if I coul find the source of the oscillation by probing the air around the traces. I suppose this ould be a job better suited for an FET probe, unfortunately I don't have one.
Thanks for the info. I onder if I coul find the source of the oscillation by probing the air around the traces. I suppose this ould be a job better suited for an FET probe, unfortunately I don't have one.
Okay, I hooked up the 150p+47R B-C snubbers. I am totally amazed. The sound is so much more real no. I couldn't spot any oscillation on the scope. The output impedance still has the 50MHz spike, but no more signs of actual oscillation. That last spike seems really hard to get rid of.
Just letting you guys kno it orked.
Just letting you guys kno it orked.