Class D amps can be made to run without feedback as well. And really no feedback, no degeneration at all. Distortion can be reasonable with the 2nd and 3rd being predominant, and output impedance being only a few milliohms.
Guys
Could it be possible after this long thread to summarize some guidelines by using feedback when and how?
Best regards
Could it be possible after this long thread to summarize some guidelines by using feedback when and how?
Best regards
Last edited:
Sure, I'll take a punch.
Global feedback--rule of thumb:
None is OK. Some is bad. A lot is great.
If you have a plan for low distortion and controlled DC operating point without global feedback, then go for it. If you need to apply just a little feedback, then you are worse off than if you had applied none. If you have an enormous open loop gain and apply tons of feedback, this is best.
Doug Self shows this on pages 64-68 of his 6th edition Audio Power Amplifier Design. A tiny amount of feedback causes intermodulation of the harmonics you already have. A lot more feedback suppresses all harmonics and intermodulation. If you cannot provide at least 40 dB of loop gain across the audio band, that's not enough feedback. More is better.
In my designs, it's my goal to still have 60 dB of loop gain at 20 kHz, which is a GBW product of 20 MHz. It is not easy. Two-pole compensation is probably the only way to do it in a conventional LTP-VAS_OPS amplifier.
Of course, the amplifier must be designed for high open-loop linearity and compensated with ample slew rate of Vout amplitude times (2*pi*20kHz) times 10. For a 100W amplifier, that is 40*126,000*10 = 50 V/us.
Global feedback--rule of thumb:
None is OK. Some is bad. A lot is great.
If you have a plan for low distortion and controlled DC operating point without global feedback, then go for it. If you need to apply just a little feedback, then you are worse off than if you had applied none. If you have an enormous open loop gain and apply tons of feedback, this is best.
Doug Self shows this on pages 64-68 of his 6th edition Audio Power Amplifier Design. A tiny amount of feedback causes intermodulation of the harmonics you already have. A lot more feedback suppresses all harmonics and intermodulation. If you cannot provide at least 40 dB of loop gain across the audio band, that's not enough feedback. More is better.
In my designs, it's my goal to still have 60 dB of loop gain at 20 kHz, which is a GBW product of 20 MHz. It is not easy. Two-pole compensation is probably the only way to do it in a conventional LTP-VAS_OPS amplifier.
Of course, the amplifier must be designed for high open-loop linearity and compensated with ample slew rate of Vout amplitude times (2*pi*20kHz) times 10. For a 100W amplifier, that is 40*126,000*10 = 50 V/us.
I've designed a few of those. Deadtime has to be very low, and turn-on/turn-off of outputs must be very fast. There is only 6 dB PSRR, and EMI is very high.
The only issue of global negative feedback is the real possibility of amplifier instability. By the latter, I specifically mean, amplifier self oscillations which everyone visiting these fora knows can easily destroy expensive output stages.
Having read how convinced Douglas Self is of the benefits of global negative feedback and his vast experience in amplifier design, I choose his opinion.
PS:
One comment from Douglas Self which made me lough was when he wrote: 'musical' amplifiers are those which are often found to distort most, and are often the ones overpriced. I like the way Self rejects opinionism where audio is involved to stick with scientifically known facts.
Having read how convinced Douglas Self is of the benefits of global negative feedback and his vast experience in amplifier design, I choose his opinion.
PS:
One comment from Douglas Self which made me lough was when he wrote: 'musical' amplifiers are those which are often found to distort most, and are often the ones overpriced. I like the way Self rejects opinionism where audio is involved to stick with scientifically known facts.
I can confirm every word of this.
If the amp is a class D and self-oscillating, you can put quite a lot of feedback on it as the idea is to exceed the phase margin of the amp anyway, and to use the oscillation as the switching frequency.
Douglas seems to have missed a point or you didn't include it in your paraphrasing. A key point in a musical sounding amplifier is that the distortion is consistent at all frequencies. Its one thing to have nice low distortion at 100Hz; its quite another to have the same figures at 10KHz.
Zero feedback amps can have the same distortion figures across the band and that's part of why they can be musical despite having large THD numbers. But it must be kept in mind that such amps are exhibiting lower ordered harmonics (2nd, 3rd and 4th) as their primary distortion components.
If the 2nd is the primary component, then the amp is expressing a quadratic nonlinearity. If the 3rd, then a cubic nonlinearity. But they are musical to the ear since it treats the 2nd and 3rd the same way and these harmonics will mask higher orders if in sufficient quantity.
Feedback got a bad rap in the audiophile world because it adds higher ordered harmonics and IMD of its own (see Crowhurst; this has been an issue for a long time). Its really only been recently that amps have been around that are able to have so much feedback that the amp can clean up the distortion caused by the feedback itself. For this you need a good 60dB of loop gain as a minimum since you'll need in excess of 35dB to do this.
That is why no feedback can be OK (but you get a high output impedance), a bit is bad (bright and harsh due to audible higher ordered harmonics) and a lot is good. The issue is frequency poles in the amp and the accompanying phase margin which must not be exceeded to make a stable amplifier. The Benchmark is an example of this. But Bruno Putzeys showed that you can get around this with a class D amp and use the oscillation for a good purpose.
All true, atmasphere.
I will quibble a bit about THD of high frequency signals. You can only hear IM products of signals above 10 kHz. If those drop down into lower frequencies, then the loop gain at those frequencies will reduce them that amount.
Of course, better to never have the IM products in the first place. It's only a minor quibble, and ultimately, you are right.
I will quibble a bit about THD of high frequency signals. You can only hear IM products of signals above 10 kHz. If those drop down into lower frequencies, then the loop gain at those frequencies will reduce them that amount.
Of course, better to never have the IM products in the first place. It's only a minor quibble, and ultimately, you are right.
Yes, you're not likely to hear a harmonic of 10KHz! Its not so much about the harmonics as it is the behavior of the amp.
I agree. Because it is difficult and it can not be a simple design, not many designer put effort to make such design. And I think not many DIY'er built such amplifier although some amplifier in this forum have similar approach.
I like an amplifier which have almost same THD at all audio frequency and almost same THD at all level below clipping, and have monotonic harmonic profile. More better speaker and more better recording, more better the sound. Some amplifier sound like "live" or believable sound and some don't.
Some amplifiers which have high distortion, can make bad speaker and bad recording sound adequate or better. Because not all distortion sound bad, some distortion sound pleasing.
Global feedback is a huge benefit for voltage regulator and oscillator circuits. Musical or not, amplifiers typically reproduce only a small portion of the signal content. How are signal losses measured and incorporated into the concept of distortion?
That is a difficult statement to justify. A typical high-quality audio thingy (amp, preamp, DAC, whatever) reproduces a signal with less than 10 parts per million error. Are you claiming that 99.999% fidelity is only a small portion of the signal content?
At this point, only speakers and microphones have fidelity errors worth mentioning. I guess phono cartidges, too, if that's what you're into.
24-bit/192 kHz signal & DAC (THD < 1 ppm) + minimal preamp (THD below 1 ppm) + amp (THD below 10 ppm) is a damn clean signal. The best loudspeakers are starting to approach 1,000 ppm (0.1%), but until recently have been 5,000 to 10,000 ppm or worse.
Russ,
Are you sure you aren't ignoring the stuff we don't usually measure? Stereo channel imperfection correlations. Dynamically modulated noise floors and or dynamic noise skirts around signal frequencies in dacs due to convolution with clock phase noise, etc? It isn't only about harmonic distortion evaluated with stationary test signals. Figure of merit, sure. Actual sound quality, not necessarily.
Are you sure you aren't ignoring the stuff we don't usually measure? Stereo channel imperfection correlations. Dynamically modulated noise floors and or dynamic noise skirts around signal frequencies in dacs due to convolution with clock phase noise, etc? It isn't only about harmonic distortion evaluated with stationary test signals. Figure of merit, sure. Actual sound quality, not necessarily.
Actually you can get a pretty good handle on things if you know that the IMD is low and the THD is constant, favoring lower ordered harmonics, over all frequencies. If these aspects are properly controlled the noise floor isn't going to be messy.
For DACs? ESS, for one. That is, they say they have. Are they liars you think?
With respect to analog preamps, which may have more to do with the feedback topic, Rob Watts describe noise floor modulation due to RF incursions from the power mains.
With respect to feedback in particular sometimes feedback loops are cut by HF/RF magnetic fields that don't show up well on an FFT. They can surely affect sound though and it is in some cases likely a dynamic modulation between audio signal and noise effects that is being produced. Since it is dynamic, it may not show up very well on a typical audio FFT.
My point is that we don't always measure what matters out in the field/real-world. We get used to thinking in terms of how we measure and it becomes easy to believe that's all there is and all there can be.
With respect to analog preamps, which may have more to do with the feedback topic, Rob Watts describe noise floor modulation due to RF incursions from the power mains.
With respect to feedback in particular sometimes feedback loops are cut by HF/RF magnetic fields that don't show up well on an FFT. They can surely affect sound though and it is in some cases likely a dynamic modulation between audio signal and noise effects that is being produced. Since it is dynamic, it may not show up very well on a typical audio FFT.
My point is that we don't always measure what matters out in the field/real-world. We get used to thinking in terms of how we measure and it becomes easy to believe that's all there is and all there can be.
Last edited: