2024, Chapter 1
The Most Abused Audio Terms, Part 4: Feedback
Let’s start with a mind-blower:
All your audio electronics have feedback.
All of them. Every electronic product you love listening to. Op-amp products. Discrete products. Class A and Class AB and Class D and Class S and H and G…all have feedback.
Even ones labeled “no feedback”…
have feedback!
Shocked?
Yeah, I thought you would be.
And I’m sure some of you are saying, “No, wait a minute, I only buy no-feedback products! I know they’re no-feedback because the marketing told me so!”
And some are protesting, “Hold on, you guys make no-feedback products, don’t you?”
And some are wondering, “Why does everything have feedback?”
Simple: Every audio electronic product has feedback because if it doesn’t, it probably sucks so much it isn’t a viable product.
Huh? What?
Has Jason fallen to the cult of measurement?
Has he joined the ranks of feedback maximalists who say feedback is great, lots is better, and mostest is bestest?
In short, no.
It’s just that feedback is a complex subject. It has real pluses and real minuses and lots of gray area in implementation. And because it’s complex and misused, it’s gotten a bit of a stink in some circles—specifically, among some audiophiles who think All Feedback is Bad…and, at the same time, it has been portayed as the One True Savior in other corners of audiodom, most particularly amongst some designers who have stated, pretty much, “you can never have too much feedback!”
How can feedback be vilified and celebrated at the same time? And why is the best use of it somewhere in that scary hard-to-define middle?
And what about all this about everything having feedback? That can’t be true, can it?
Well, let’s first back up and define feedback, for the guys out there who are shaking their heads and saying, “What are all these nerds talking about?”
Feedback 101
Want to make a signal bigger?
Of course you do. Whether you’re starting with 0.5mV (5/10,000 of a volt) from a MC cartridge or 2V from a DAC, you need a whole lot more gain to drive a speaker—gain in terms of both voltage and current.
So you want an audio amplifier.
Now, when we’re designing this audio amplifier, we can choose to use a whole lot of different discrete amplification devices: BJTs (“transistors”), JFETs, MOSFETs, tubes, etc. And, if the current requirements are small, you could even choose an integrated circuit, an op-amp.
The problem is: every amplification device is nonlinear.
As in, it’s not a perfect “valve.” So it will create distortion. And noise. And it most likely will have too much gain to be usable. It may not even have enough bandwidth to cover the full audio band.
In audio, negative feedback is a way of trading gain for linearity.
Apply negative feedback, and you get:
- Higher bandwidth
- Lower distortion
- Lower noise
- Lower output impedance
And most amplification devices have plenty of gain to trade. Op-amps can have 100-130dB of gain open-loop. (And with an open-loop bandwidth of 10-100Hz, oh boy they need to trade.) A simple single-transistor undegenerated circuit can have 40dB of gain. A 2-stage discrete amp can have 60-80dB of gain.
Aaannndd…you really can’t use that much gain. So trading some of it sounds like it makes a lot of sense. But there are some gotchas, which we’ll get to later.
Negative feedback has been likened to “comparing the input and the output,” which calls to mind weird electronic structures that exist outside of the amplifier, considering it with cool and calculating eyes like Well’s Martians.
In reality, negative feedback is inherently part of the circuit. There’s no invisible arbiters existing outside of the Holy Topology. The topology itself contains the feedback.
Feedback In classic control systems, it looks like this.
Or, in a simple circuit (op-amp), it looks like this:
In a discrete design, it can look like this:
“Huh? Wait a sec!” some of you cry. I don’t see the feedback here. Where’s the feedback?”
Gotcha. There are actually several types of feedback being used in that diagram…and it can still be labeled “no feedback” for the purpose of marketing.
Mind blown?
Cool. Let's deconstruct this.
Stayin’ Local
The first two diagrams above show negative feedback being applied to an entire circuit. The feedback loop goes from the system output back to the input, encompassing the entire topology.
This is known as “overall negative feedback.”
And, when amplifiers claim “no feedback” or “zero feedback,” this is what they mean:
There’s no overall feedback from the output all the way back to the input.
“Oh hell wait a sec you can have feedback buried inside an amp?” you yell. “Feedback that's feedback but you call it no feedfback? That’s sneaky AF!”
Well, yes and no.
Yes, you can have feedback inside an amp topology—local feedback. There are many kinds of local feedback, from degeneration to local loops to complementary pairs and more. All practical amplifiers, especially no-feedback amps, will have local feedback.
But no, it’s not sneaky. Local feedback is usually necessary to make an amplifier really usable—as in, with high enough linearity and low enough gain that it’s a real product that effectively provides the gain and performance you need.
Without local feedback, and without an overall feedback loop, an audio power amp might have 60-80dB of gain. That’s 1,000-10,000x gain. No matter how loud you listen, or how much you love the opening scene of Back to the Future, that’s wayyyyyyyyy tooooo much gain.
So, internally, you trade some gain for linearity with local feedback…
…and then you can choose to have an overall feedback loop as well.
Let’s look at some of the local feedback options:
Degeneration:
Sounds horrible, doesn’t it? “You degenerate! How dare you use degeneration!”
In reality, this is probably the most commonly used form of feedback, and one that has pretty much 100% positive effects. Both Doug Self and Bob Cordell have written extensively about degeneration and its benefits. In short, it:
- Reduces the gain of the first stage of an amplifier
- Increases the linearity of that same first stage
- Compensates for variability in discrete devices
Degeneration is used in all of our products. And, I’ll bet it’s used in every product labeled “no feedback” as well, because it’s a great way to enhance performance with really no downsides.
“Wait a sec,” someone says. “Is that really feedback?”
In short, yes, this is really feedback. It meets all the criteria for feedback. It is just so localized that it doesn’t look like feedback. And it is so necessary—especially in amps with no overall feedback—that it’s used pretty much everywhere.
Or, in other words: yes, Virginia, your no-feedback amp has feedback.
Re: (No Boom):
Like those resistors in the output stage? They’re doing something similar to degeneration in the input stage—they’re compensating for variations in the output devices (sorry, guys, discrete devices do vary…simulators beware), and they’re usually providing some degree of thermal stability as well.
Also usually found in no-feedback power amps. But yeah, still feedback.
Thermal Stability (Also No Boom):
Oh this is a weird one, right? Yeah, these resistors aren’t technically needed in a diamond buffer, but try getting by without them. You may find it goes up in smoke due to thermal runaway. A different kind of feedback, but still feedback.
And, by the way, current feedback amps like Vidar, Magni, etc absolutely depend on these resistors.
Complementary Pairs:
This is one of my favorites.
Complementary pairs, or Complementary Feedback Pairs, are one of the best ways to get a simple, high-performance circuit from very few discrete components.
- Pair an NPN and PNP transistor, and suddenly you have a much higher-performance device.
- Stack two transistors (BJT, JFET, MOSFET) on top of each other with one acting as a current source, and suddenly the distortion cancels, creating a much higher-performance device.
Heck, a 2-transistor buffer can reach -116-118dB THD+N into light loads. That’s 2 transistors, 2 resistors. Near state of the art performance. No feedback! (LOL)
Local Loops:
Looks like overall feedback, but not overall, right?
Exactly. It’s a negative feedback loop that doesn’t go all the way to the output. Very common in “no feedback” amps, though I’d consider this one kinda sneaky if it went all the way to the drivers, as in this example.
This kind of local loop is super-useful, especially with complex amplifiers that may need some additional compensation options.
“Compensation?” you ask.
Yeah, more on that later, in the downsides of feedback.
Error Correction:
Another local loop, this time around the output stage—and, oh yeah, not technically feedback.
This is error correction, or feedforward, as the breathless pundits told us was the One True Path circa 2016 or so. Actually, this is Hawksford error correction, which predates the pundits of 2018 by about 30 years.
Aaaaaand…for additional confusion, this is
not technically feedback.
The equations are totally different. It’s not feeding back the entire signal, it’s feeding back
only the difference between input and output.
This is a fun one for a no-overall-feedback amp, because it can be both no-feedback AND have low output impedance.
DC Servo:
Here’s a fun one: feedback only at very low frequencies.
Why would someone want to do that? To eliminate DC at the output of an amplifier, without trimming (and possible drift over time) or coupling capacitors. DC servos are widely used to both eliminate DC, and provide correction for input DC, in DC-coupled amps.
Is it feedback? Absolutely. Can it be used in a “no-feedback” amp? Also absolutely.
So is that it for sneaky feedback?
Oh no, there are tons of other techniques. Get Bob Cordell’s book and look into his chapter on how to create a no-overall-feedback amplifier, and he’ll get into much more detail, including stuff like instrumentation front ends.
And, here’s the thing to keep in mind: You can use every single one of these local techniques in an amp—
and still claim “no feedback!”
Because what you really mean is “no overall feedback,” and that can be absolutely true.
But even in overall feedback, there are flavors.
Let’s talk about two of them.
Current vs Voltage
You may have heard of amps billed as “current feedback” amplifiers. Heck, you’ve heard it from us, because the vast majority of amps we do are current feedback.
Current feedback is kinda weird, though. Most of the amplifiers on the planet—discrete or op-amp—are voltage feedback amps.
Why?
I can be snarky and say, “Well, it’s because the Lin topology, which evolved into Blameless, and is used as the foundation for pretty much every amp you see in every audio power amp book, is voltage feedback, and so most people start with voltage feedback, it’s what they know, why try anything else, and there you go you have all phones as rounded rectangles and you have pretty much all amps as voltage feedback.”
But it’s more than that. Voltage feedback and current feedback both have their pluses and minuses, and so they both have their place in the world. In brief:
Voltage feedback:
Advantages:
- More gain—allows for more feedback and better numbers
- Higher power supply rejection—you can have a noisier power supply and it will ignore it better
- High impedance negative input terminal means easy overall negative feedback
Disadvantages:
- Lower bandwidth—not affected by feedback
- Slower slew rate, dominated by how much current available to drive Miller capacitance
- More prone to asymmetric slew and ringing—more difficult to compensate
Fun fact: the diagram above also shows
nested loops--a local loop inside an overall loop.
Current feedback:
Advantages:
- Higher bandwidth—can be “tuned” with feedback
- Faster slew rate, not affected by front end current
- Very easy to compensate
Disadvantages:
- Lower gain—lower feedback and worse THD numbers
- Lower power supply rejection—getting good PSRR is more work
- Low impedance feedback terminal
In general, a voltage feedback amp will do better in terms of both THD and noise, which shouldn’t be surprising because it has more gain and can use more feedback. This shouldn’t be a surprise, because it has two stages of voltage gain, versus the single stage of a current feedback amp.
Aside: it’s amazing Magni Unity does as well as it does, given the inherent disadvantages of a current feedback amp. It’s uncomfortably close to state of the art measurements, with 100dB less feedback.
So why stick with current feedback when voltage feedback gives better numbers?
In some applications, it’s a matter of speed—rise time needs to be faster than practical for voltage feedback. But these are extreme applications, not audio. In audio, the choice of current feedback is a more personalized one.
“Well, hey you said you like current feedback better because it sounds better,” someone says. And yes, that’s my personal opinion. It’s also the result of much internal blind listening (yes, double-blind, yes, level-matched, no, not DBX, no, not worthy of MIT-level publication).
In addition, we also like the overall simplicity of current feedback, the lack of excessive gain, and its easy-to-compensate nature. Anyone who’s fought with an unstable voltage feedback amp knows exactly what I’m talking about.
Which is a great segue, because I’m sure more than a few of you have been wondering:
“Why all this talk of negative feedback?
Is there positive feedback?”
Well, yes, but you don’t want it. Positive feedback creates an oscillator. As in, a circuit that produces a single frequency constantly, forever, as soon as you turn it on. Sometimes this is a good thing—you may want an oscillator to produce a sine at a particular frequency.
Most of the time, it is a bad thing. Heck, you know it’s a bad thing if you’ve ever grabbed the mic of an old-skool PA system and had it shriek you out of the auditorium. That’s positive feedback.
And it’s a very very bad thing if you’re talking about a power amp that can source tens of amps that is suddenly oscillating full scale and torching itself and the speakers it’s connected to.
Huh? What do I mean? And why is this a good segue?
It’s a good segue because compensating amps is all about making sure their open-loop gain goes below 1 by the time the output phase shift has reached 180 degrees.
Huh?
Okay, let's see if we can explain this without math:
- Negative feedback: feedback that’s inverted, or 180 degrees out of phase, with the input signal. This subtracts from the input. Safe.
- Positive feedback: feedback that’s in phase with the input signal, or 0 degrees out of phase. This adds to the input. No bueno.
Now, all amps have phase shift, and the phase shift increases with frequency. So your negative feedback, that safe 180-degrees-out-of-phase feedback, slowly changes. At 10kHz, it might be 160 degrees off. At 100kHz, it might be 90 degrees. At 1MHz, it might be 0.
And, if the amp has positive gain (more than a gain of 1) at 1MHz, it suddenly becomes a power oscillator and things go bang.
So, you compensate the amp (manage poles and zeroes, or, in English, place compensation capacitors at strategic places), in order to ensure the amp never has positive gain when phase shift has turned it into a potential oscillator.
With a voltage feedback amp, it’s not unusual to have complex and nested compensation with half a dozen capacitors or capacitors and resistor combos to get it stable. That previously-mentioned Magni Unity? One capacitor. Done.
Yeah, we like current feedback.
But voltage feedback (and no overall feedback) also have their place.
Feedback Gotchas and Best Practices
Now, some of you aren’t happy. You’re crossing your arms and saying, “You said there were pros and cons to feedback. You’ve talked about the pros. And you’ve talked about the kinds of feedback. What about the cons? What’s the catch?”
Yep yep, I hear you. So here are the catches when you use feedback:
- Gain and phase margin are now critical. Amplifier stability is now something you have to pay much, much more attention to if you want to have a solid, reliable, versatile product.
- Transient response can suffer. Even if you have a stable amplifier, it may still ring, or, in engineering terms, exhibit underdamped characteristics.
- Compensation may entail trade-offs between gain and phase margin and transient response. You want a stable amplifier, and you’d like it to have rapid rise times, but sometimes not all of those are in the cards.
- Using too little feedback can cause different kinds of distortion that aren’t pleasant—re-entrant distortion. To be honest, this isn’t common.
- Using too much feedback can mask problems with the open-loop circuit that should be fixed, like running too low of a Class AB bias.
- On some monumentally slow circuits, feedback can cause transient distortion that is also very unpleasant. This is, like re-entrant distortion, not super common.
- The more complex the feedback, the more difficult it is to get the amplifier stable. Imagine an amp with local loops, a global loop, and error correction. That’s a lot of complexity to manage.
But these are just the technical catches. One of the biggest catches is that, simply, feedback is considered by some to be undesirable. They say things like “Feedback squashes dynamics, amps with feedback sound lifeless,” and “no feedback is the only way to go.”
However, at the same time, feedback is considered by others to be a 100% positive thing. “There’s no such thing as too much feedback,” they say. “More feedback is better, and most is best.”
Yes. There are
Feedback Minimalists and
Feedback Maximalists.
In reality, both oversimplify. At least in my opinion.
Feedback Minimalists sometimes don't:
- Acknowledge that “no feedback” amps actually contain plenty of feedback
- Realize that “no overall feedback” is really what these amps are
- Understand that making an amp with low output impedance usually requires some loop feedback—or will be very expensive because it will need many paralleled output devices
- Learn that types of feedback may matter—it’s not one size fits all
- Accept that there’s a place for feedback, properly used
Feedback Maximalists sometimes need to:
- Be honest about their assertions: “more feedback is better for better measurements” is typically what they mean
- Acknowledge what feedback might be hiding—nonlinearities in the overall circuit or topology
- Provide enough information about transient performance to assure the negative effects of feedback are managed
- Understand that low- and no-overall-feedback amps can get very close to, or even exceed, the measured performance of high-feedback designs
- Accept that there are other ideologies in audio amplifier design
“So what does Jason think?” someone asks.
Sigh.
As usual, I’m gonna irritate everyone.
Because I think both the drive to zero feedback and the drive to maximum feedback are both a bit bonkers. Both oversimplify the problem.
With zero overall feedback, and minimal local feedback (no large local loops, for instance), it’s very difficult to produce a usable amplifier with decent measurements and low output impedance, at least not without having it cost a beeeeeelion dollars because you have to use 32 paralleled pairs of output devices per channel.
At the same time, with the maximum amount of feedback to aim at the maximum measurements, it’s going to be realllly tempting to do everything you can to get more gain, even if it sacrifices open-loop linearity, and it’s gonna be reaaallly reallllllly tempting to use every feedback technique—local, error correction, nested, and overall—to really get those numbers.
My approach is in the middle:
- Make the open-loop gain stage as linear as possible. This will include degeneration, and, at times, compound pairs or cascoding.
- Use as little overall feedback as possible to reach the design goals. Most of our stuff has 6-30dB of overall feedback. Some have zero. But zero is not a goal in itself.
- Consider local loops and nested loops for difficult topologies. As things get more complex, and power output increases, local or nested loops can improve performance.
Some devices deviate quite a bit from this template—see the appendix below—but that’s the general idea.
Feedback on Feedback: The Good, Bad, and Ugly
I opened this chapter as “The Most Abused Audio Terms, Part 4.”
And “feedback” is certainly abused—usually as in an implication that it’s something best avoided.
In fact, “no feedback” has become a badge of honor amongst some, but as discussed above, it’s really an oversimplification. All audio amplifiers contain some forms of feedback. “No feedback” usually refers only to the lack of overall loop feedback.
Aside: I mean, heck, Magni Heresy and Magni Heretic both qualify as “no overall feedback” amplifiers, but considering the 120dB of feedback in the local voltage gain loop and 130dB in the local output buffers, calling it “no feedback” seemed beyond the pale.
So, yeah. “No feedback,” when seen as a marketing claim, should immediately raise some questions.
- Do they mean “no overall feedback?” Usually yes, but it’s worth asking about.
- Do they mean “no overall feedback, and we try to limit the amount of feedback used, because we’re really most interested in creating very linear gain stages that don’t need a lot of feedback,” because, again, that’s frequently what they’re trying to say.
- Or do they mean, “Yeah, this is a pure tube amplifier that doesn’t use degeneration and has bypassed cathodes so technically it’s running full-out, full-gain, all the device can give, because we need it for a phono stage, oh yeah and by the way you better get used to matching these tubes really close, because gain variance will be a real thing, but hey if you really want no feedback we’re as close as you can get, enjoy,” because yeah, maybe that’s a thing too. I haven’t seen it personally, but who knows? Audio is a wonderful space, full of designers with different ideas.
The reality is that feedback isn’t something that should be considered 100% “good” or 100% “bad.” There are great amps with feedback and terrible amps with no feedback.
“So tons and tons of feedback is the ugly,” somebody says, referencing the title of this section.
Well…not necessarily. You may want something with very high loop gain and a lot of feedback—an op-amp—as the error amplifier on a multiplying DAC. Or you may need the same kind of device due to limited design size or power dissipation. Those devices may perform very well in those applications.
No. The only ugly is insisting that there’s One True Path.
The reality is that feedback isn’t evil. It’s very, very useful. It’s everywhere. And it’s usually required for a truly great product, even if it is “only” local.
The other reality is that moar and more feedback isn’t the singular answer either. It can hide real problems and promote numbers-chasing that doesn’t really advance the state of the art.
The real challenge is figuring out which type of feedback to use where—and how to use it best. At least that’s how I see it. But, as I’ve said before: I may be crazy!
I hope you enjoyed this foray into feedback!
Appendix 1: Schiit Products and Feedback
Magni: Voltage Feedback, Overall
Magni 2: Voltage Feedback, Overall
Magni 2U: Voltage Feedback, Overall
Magni 3: Current Feedback, Overall
Magni+: Current Feedback, Overall
Magni Unity: Current Feedback, Overall
Magni Heresy: No Overall Feedback (!), Local Voltage Feedback Front End and Output Stage
Magni Heretic: No Overall Feedback (!), Local Voltage Feedback Front End and Output Stage
Magnius: No Overall Feedback (!), Local Voltage Feedback Front End, Local Current Feedback Output Stage.
Midgard: Current Feedback, Mixed Mode Halo™ Current Sensing
Asgard: No Overall Feedback
Asgard 2: No Overall Feedback in high gain
Asgard 3: Current Feedback, Overall
Valhalla: No Overall Feedback
Valhalla 2: No Overall Feedback
Jotunheim: Current Feedback, differential Nexus™
Jotunheim 2: Current Feedback, differential Nexus™
Mjolnir 1: No Overall Feedback
Mjolnir 2: No Overall Feedback in high gain
Mjolnir 3: No Overall Feedback, Switchable to Overall Feedback
Ragnarok 1: No Overall Feedback, Local Front End Loop
Ragnarok 2: Current Feedback, differential Nexus™
Folkvangr: Voltage Feedback, Overall
Rekkr: Current Feedback, Overall
Gjallarhorn: Current Feedback, Overall
Aegir: Current Feedback, Overall
Aegir 2: Current Feedback, Mixed Mode Halo™ Current Sensing
Vidar: Current Feedback, Overall
Vidar 2: Current Feedback, Overall
Tyr: Current Feedback, differential Nexus™
Loki Mini: Current Feedback, Overall
Loki Mini+: Current Feedback, Overall
Lokius: Current Feedback, Overall
Loki Max: Current Feedback, differential Nexus™
Mani: Voltage Feedback, Overall
Mani 2: Voltage Feedback, Overall
Skoll: No Overall Feedback
Saga: No Overall Feedback
Saga+: No Overall Feedback
Saga S: No Overall Feedback
Freya: No Overall Feedback
Freya+: No Overall Feedback (differential buffer), Voltage Feedback (overall, tube)
Freya S: Current Feedback, differential Nexus™
Kara: Current Feedback, differential Nexus™
Appendix 2: Links to "Most Abused Audio Terms" Chapters
The Most Abused Audio Terms: Class A
The Most Abused Audio Terms, Part 2: Balanced
The Most Abused Audio Terms, Part 3: Discrete
Appendix 3: Links to Audio Books You Should Own If You're Reaaaaaallly Interested In How This Stuff Works
Bob Cordell, Designing Audio Power Amplifiers
Douglas Self, Small Signal Audio Design
