Now, where did this 'open loop' bandwidth come from? In the tube days, it was simply the limitation of the AMOUNT of global negative feedback that was practical, in the great majority of audio applications, due to excess LOW FREQUENCY phase shift in the output transformers and coupling caps. This was hard to work around, so people settled for about 20dB or 10 times reduction in open loop distortion, in general. They had to get the rest of the specifications from design, either by improving the high frequency response and phase of the the output transformer or any other high frequency extension technique, so that they would have extended high frequency bandwidth. This usually gave a 5K-10K open loop bandwidth, by default, given 20dB of feedback. Nobody that I knew really thought about it much, but they could hear differences in different amps, due to many other factors. In any case, the reduction of open loop bandwidth in audio design was primarily due to the IC op amp industry, that originally made op amps for servo and analog computer applications, but gradually got increased interest in audio applications, as we better understood the convenience and versatility of the op amp design concept, then taught more in college classes, and as tube design faded from the classroom. More later.
PIM distortion was hinted at by Matti Otala in my working with him in the late 1970's, as he was NOT happy with just using SLEW RATE and a class A input stage as the only real predictor of problems with high frequency audio reproduction. He wanted to have a category called DIM. TIM would be included as a subset of DIM. This way, if we found other dV/dt related effects, we could study them separately, but put them under DIM. For example, non-linear capacitance, where impedance is so high at low frequencies that it is not very important, but with high rate of change of the signal, perhaps very significant.
In any case, this is when Matti Otala introduced a purely theoretical model of PIM distortion at an AES convention, BUT he could NOT get it published in the JAES, because it was considered too controversial. At least that is what he personally told me.
While this sort of distortion was addressed by others and dismissed as unimportant, much as TIM was, the decade before, we were still with the problem of making better quality audio equipment. Why didn't IC's quite do what they seemed to imply from measurements, for example? Why could we make world class successful audio products IF we paid attention to details, including hi open loop bandwidth, low or no global negative feedback, and pristine construction method and parts, yet our overall measurements might not be much different, and even somewhat worse than an IC based design with the comprises necessitated by automated assembly, which would be classed as sounding 'not quite right'.
In any case, I spent a decade, armed with 'what worked' to make custom products, ignoring the PIM debate, but still holding 'high open loop bandwidth' as important (for some reason) as it worked for me. More later.
In any case, this is when Matti Otala introduced a purely theoretical model of PIM distortion at an AES convention, BUT he could NOT get it published in the JAES, because it was considered too controversial. At least that is what he personally told me.
While this sort of distortion was addressed by others and dismissed as unimportant, much as TIM was, the decade before, we were still with the problem of making better quality audio equipment. Why didn't IC's quite do what they seemed to imply from measurements, for example? Why could we make world class successful audio products IF we paid attention to details, including hi open loop bandwidth, low or no global negative feedback, and pristine construction method and parts, yet our overall measurements might not be much different, and even somewhat worse than an IC based design with the comprises necessitated by automated assembly, which would be classed as sounding 'not quite right'.
In any case, I spent a decade, armed with 'what worked' to make custom products, ignoring the PIM debate, but still holding 'high open loop bandwidth' as important (for some reason) as it worked for me. More later.
Hi John
I see myself as more "objectivist" than "subjectivist" (in the sense that those words are regularly abused on the forum).
And yet the wide open-loop bandwidth design philosophy makes perfect sense to me. Here's why:
It seems to me that for good reproduction, it might be a good idea to aim for a consistent sonic signature throughout the audio band (or as much of it as possible), and that this would require distortion characteristics to be consistent throughout the audio band.
OTOH I have a sneaking suspicion that distortion higher up (e.g. in the upper midrange) is audibly more objectionable than distortion down in the bass.
Either way, we need clean reproduction at the higher frequencies.
Another important point is that high-order distortion is more objectionable than low-order distortion. There seems to be general agreement about that.
~~~~~~~~~~~~~~~~~~~~~~~~
Given the above, I find it a little bizarre that anyone would make a conscious decision to implement feedback in such a way that:
A) Distortion of the upper midrange is at least 10 times worse than distortion of the bass.
B) At most audio frequencies, the (harmless) 2'nd harmonic is suppressed about 10dB harder than the (obnoxious) 7'th harmonic.
And yet this is what most folks do.
~~~~~~~~~~~~~~~~~~~~~~~~
Even those that care more about the distortion spectrum @ 1kHz than anything else tend to appreciate that the audibly objectionable components of that spectrum are out at 5kHz, 7kHz, 9kHz etc.
I would have thought it obvious that adding an extra 20dB of feedback at 500Hz (compared to the level at, say 5kHz) isn't going to help clean that up.
In fact, increased feedback (at low frequencies only) requires increased open-loop gain, often at the expense of worse open-loop linearity (at all frequencies). In those cases it's actually going to make the audible nasties even nastier.
~~~~~~~~~~~~~~~~~~~~~~~~
To me there's no black magic here, just common sense telling me that high feedback combined with low open-loop bandwidth is somewhere between pointless and a bad idea.
When looking at (or thinking about) an amp design, I tend to ask myself "Is the performance at high frequencies good enough?"
If the answer is "yes", then surely the performance at lower frequencies is also good enough, so no extra feedback is required down there.
If the answer is "no", then to fix it, more attention needs to be paid to the open-loop linearity and bandwidth. It certainly won't help to sacrifice linearity, bandwidth, and perhaps even slew-rate in order to get more feedback at the opposite end of the spectrum.
[/rant]
Cheers - Godfrey
I see myself as more "objectivist" than "subjectivist" (in the sense that those words are regularly abused on the forum).
And yet the wide open-loop bandwidth design philosophy makes perfect sense to me. Here's why:
It seems to me that for good reproduction, it might be a good idea to aim for a consistent sonic signature throughout the audio band (or as much of it as possible), and that this would require distortion characteristics to be consistent throughout the audio band.
OTOH I have a sneaking suspicion that distortion higher up (e.g. in the upper midrange) is audibly more objectionable than distortion down in the bass.
Either way, we need clean reproduction at the higher frequencies.
Another important point is that high-order distortion is more objectionable than low-order distortion. There seems to be general agreement about that.
~~~~~~~~~~~~~~~~~~~~~~~~
Given the above, I find it a little bizarre that anyone would make a conscious decision to implement feedback in such a way that:
A) Distortion of the upper midrange is at least 10 times worse than distortion of the bass.
B) At most audio frequencies, the (harmless) 2'nd harmonic is suppressed about 10dB harder than the (obnoxious) 7'th harmonic.
And yet this is what most folks do.
~~~~~~~~~~~~~~~~~~~~~~~~
Even those that care more about the distortion spectrum @ 1kHz than anything else tend to appreciate that the audibly objectionable components of that spectrum are out at 5kHz, 7kHz, 9kHz etc.
I would have thought it obvious that adding an extra 20dB of feedback at 500Hz (compared to the level at, say 5kHz) isn't going to help clean that up.
In fact, increased feedback (at low frequencies only) requires increased open-loop gain, often at the expense of worse open-loop linearity (at all frequencies). In those cases it's actually going to make the audible nasties even nastier.
~~~~~~~~~~~~~~~~~~~~~~~~
To me there's no black magic here, just common sense telling me that high feedback combined with low open-loop bandwidth is somewhere between pointless and a bad idea.
When looking at (or thinking about) an amp design, I tend to ask myself "Is the performance at high frequencies good enough?"
If the answer is "yes", then surely the performance at lower frequencies is also good enough, so no extra feedback is required down there.
If the answer is "no", then to fix it, more attention needs to be paid to the open-loop linearity and bandwidth. It certainly won't help to sacrifice linearity, bandwidth, and perhaps even slew-rate in order to get more feedback at the opposite end of the spectrum.
[/rant]
Cheers - Godfrey
Hi Godfrey,
You have done a very good job of articulating the arguments favoring wide open loop bandwidth. However, most of these arguments are somewhat intuitive. Let me point out a couple of things about how amplifiers work with large and small open loop bandwidth. Some of these coincide with your observations. Most of what follows is covered in my new book. You can find a table of contents on my website at CordellAudio.com - Home.
1. All else equal, amplifiers with high and low open loop bandwidth will have the same amount of NFB and distortion characteristics at 20 kHz. The difference is that the amplifier with low open loop bandwidth has higher amounts of NFB at low frequencies.
2. Distortion generally rises as frequency increases with all amplifiers, but it does so more with amplifiers that have low open-loop bandwidth, all else remaining equal. However, this is not because distortion is higher at high frequencies with low OLBW, but rather because it is lower at low frequencies with low OLBW due to the higher freedback at low frequencies.
3. The size of the error signal presented to the input stage is about the same in the two amplifiers at high frequencies, but it is much lower for the amplifier with low OLBW and higher feedback at low frequencies. This means that input stage distortion is about the same in both cases, but is much lower in the amplifier with low OLBW.
4. The higher amount of open loop gain in an amplifier with low OLBW does not come at the expense of more circuitry or less linearity in the circuitry. It is a natural result of feedback compensation. In fact, most amplifiers that have high open loop bandwidth achieve it by deliberately resistively loading the VAS. This makes the VAS work harder at all frequencies and actually degrades open-loop linearity.
5. Most people focus on harmonic distortion, where the higher-order frequencies occur at higher frequencies, as you pointed out with your 1 kHz THD example where the obnoxious 7th harmonic is at 7 kHz. However, we don't listen to sinusoids. The reality is that most perceived distortion results from intermodulation products that are often at lower frequencies than the original signal components. The spit on a cymbal is not due to supersonic distortion components, but rather due to IM products of the cymbals HF components that are reflected down into the middle of the audio band. In this case, larger amounts of NFB at lower frequencies actually do reduce the amplitude of IM components at the lower frequencies.
6. Actual measurements of TIM via the DIM test usually show that the amplifier with low open loop bandwidth and higher feedback at low frequencies has lower distortion, again, all else equal.
7. Similarly other tests, like 19+20 kHz CCIF, also reveal lower amounts of distortion for amplifiers with low open-loop bandwidth and higher amounts of NFB at low frequencies, all else remaining equal.
Cheers,
Bob
Hi Bob,
no wonder, the result of 19+20kHZ CCIF is almost same as 1kHz THD FFT of 2nd order and 3rd order components, provided there is no extra SR issue or output stage cross-conduction at 20kHz.
Hi Bob
Wow, thanks for the detailed reply!
I go along with most of what you say, and think your point 5 about IM products being reflected down to lower frequencies than the original signal components is particularly worth taking seriously.
I'd like to play devil's advocate and take issue with your points 3 & 4 though.
I agree that amplifiers with high and low OLBW will have about the same amount of feedback at high frequencies, and thus about the same size of error signal at the input to the first stage. However I don't agree that this implies that the input stage distortion will be the same.
Let's take as an example an amplifier of similar topology to Doug Self's Blameless, and try to improve it by doubling the loop gain while reducing the open-loop bandwidth from 1kHz to 500Hz.
The easiest (and perhaps only) way to double the open-loop gain is to halve the amount of emitter degeneration in the first stage LTP. At the same time, the compensation capacitor will have to be doubled.
The input stage now has to deliver double the current, with half the local feedback compared to the original. We can thus expect the input-stage distortion above 1kHz to be about 8 times worse, while the amplifier's maximum slew rate will be reduced to half it's previous value.
I will concede that as long as the input-stage distortion remains much lower than the output-stage distortion, the overall measured distortion of the "improved" amp won't be significantly worse than the original at high frequencies, and will be about 6dB better below about 500Hz.
Cheers - Godfrey
Wow, thanks for the detailed reply!
I go along with most of what you say, and think your point 5 about IM products being reflected down to lower frequencies than the original signal components is particularly worth taking seriously.
I'd like to play devil's advocate and take issue with your points 3 & 4 though.
I agree that amplifiers with high and low OLBW will have about the same amount of feedback at high frequencies, and thus about the same size of error signal at the input to the first stage. However I don't agree that this implies that the input stage distortion will be the same.
Let's take as an example an amplifier of similar topology to Doug Self's Blameless, and try to improve it by doubling the loop gain while reducing the open-loop bandwidth from 1kHz to 500Hz.
The easiest (and perhaps only) way to double the open-loop gain is to halve the amount of emitter degeneration in the first stage LTP. At the same time, the compensation capacitor will have to be doubled.
The input stage now has to deliver double the current, with half the local feedback compared to the original. We can thus expect the input-stage distortion above 1kHz to be about 8 times worse, while the amplifier's maximum slew rate will be reduced to half it's previous value.
I will concede that as long as the input-stage distortion remains much lower than the output-stage distortion, the overall measured distortion of the "improved" amp won't be significantly worse than the original at high frequencies, and will be about 6dB better below about 500Hz.
Cheers - Godfrey
Well thought out Godfrey, something similar happens as well when we use an active load, instead of a resistive load on the input stage. Usually the second stage becomes more non-linear with the active load driving it. Also, most designs, especially IC's do not remain EQUAL, and current saving class B output stages replace the class A output stages, because the heavy feedback will tend to hide the less accurate transfer function.
However, Bob is right that merely adding low frequency gain, ALL ELSE BEING EQUAL, should be OK. However, there is still the problem of PIM as first expressed by Matti Otala, and brought forth in a more realistic way by Barrie Gilbert, 15 years later. This is what we should concentrate on next.
However, Bob is right that merely adding low frequency gain, ALL ELSE BEING EQUAL, should be OK. However, there is still the problem of PIM as first expressed by Matti Otala, and brought forth in a more realistic way by Barrie Gilbert, 15 years later. This is what we should concentrate on next.
Hello John,
I would like to correct some of the details in this post. The amp you made for ML was a 25 watt, pure class A design. After Sandy Berlin booted Mark (the man) out of the company around 1980, the company had a succession of forgettable products.
Then starting around 1990 or so, they had a string of "home runs". First they had the No.20 power amp, which was a pure class-A 100 wpc amp. It was ungodly hot, but it sounded quite good. They followed that up with the No.23 power amp, which was a 200 wpc class-AB design. This caused a big problem for the marketing department.
The market was shifting to higher and higher powers, so they had to respond. But going beyond 50 wpc pure class-A becomes impractical. (I doubt that the No. 20 was really what it claimed to be.) And now they were in the distinctly uncomfortable position where the top-of-the-line amp was half the power of the second-position amp.
Plus they had introduced the No.30/31 transport and DAC as "reference" pieces. So they needed a "reference" power amp that was more powerful than the No.23. And they also needed a way to gracefully exit from the whole "class-A is better" mantra they had been preaching for years.
The solution was simple -- employ a sliding-bias quasi-class-A design that put out 300 watts. But here's where the tale gets sordid. They stole the output stage and bias circuit from an AES paper of at least a dozen years prior -- written by engineers from Sansui! Of course, they claimed this design as their own, which was a bald-faced lie.
None of this has anything at all to do with Mark Levinson the man. Except possibly that it takes a snake (Sandy Berlin) to deal with a snake (Mark Levinson).
Hmmm.
The VAS "works harder"???
And this "degrades open-loop linearity"???
By this logic, we could never build a successful power amp, as the eight ohm load of the loudspeaker (or worse yet four, or two, or even one!!!) would cause the output stage to "work harder" and "degrade open-loop linearity".
I would submit that it is certainly possible to build a gain stage that performs just dandy when presented with a load besides the nasty compensation capacitor found in nearly all conventional feedback designs.
I agree Charles. I also seem to do pretty well, distortion measurement wise, with resistive loading. If I have to make choice, I will choose higher open loop bandwidth every time, yet we STILL have not evaluated WHY this is so on this recent portion of the thread.
It must also be understood as to why I have been so slow in bringing this topic up. Without previous grounding in circuit design and its tradeoffs, it is easy to smooth over PIM distortion as being unimportant. Yet, we still hear differences, and this distortion has only been lightly addressed.
It reminds me of what was considered necessary as to slew rate regarding phono cartridges, back in the 1970's. Lz and Vy tried to find maximum slew rate with a SHURE phono cartridge. Of course, they didn't find any, because the Shure they tested with, had a 4 pole low pass filter built into it, after about 20KHz. Had they looked at the time at an Ortofon moving coil cartridge, available for more than 15 years, and sold in the same shops as the Shure cartridge, they might have had different results. More later.
It must also be understood as to why I have been so slow in bringing this topic up. Without previous grounding in circuit design and its tradeoffs, it is easy to smooth over PIM distortion as being unimportant. Yet, we still hear differences, and this distortion has only been lightly addressed.
It reminds me of what was considered necessary as to slew rate regarding phono cartridges, back in the 1970's. Lz and Vy tried to find maximum slew rate with a SHURE phono cartridge. Of course, they didn't find any, because the Shure they tested with, had a 4 pole low pass filter built into it, after about 20KHz. Had they looked at the time at an Ortofon moving coil cartridge, available for more than 15 years, and sold in the same shops as the Shure cartridge, they might have had different results. More later.
Last edited:
Seems you've misconstrued what Bob's saying - he'd not disagree that lower speaker impedances are worse for amps. As amp designers we really have no control over what speaker's connected. We do though have total control over whether we load the VAS with a resistor to decrease the LF OL gain. Bob's argument is that loading the VAS in such a way has no benefit - indeed it has the drawback of making the VAS work harder. In this, he's in agreement with Doug Self.
I doubt Bob would disagree. I'd ask though 'why bother?'. Why put resources into something unncessary?
And just exactly what do you mean by "work harder"?
Well, I guess if "good sound" is "something unnecessary", there wouldn't be much point, would there? So don't bother.
In this case - increase its dissipation, make it run hotter, make it deliver more current.
Ah, you not only misconstrue Bob, you also do the same to me.
Does it indeed sound better with a resistor? I wasn't aware that it did, so please enlighten me.
And is that such a bad thing? We're not talking about fan cooling or liquid cooling or something impractical. I use TO-92 parts here with no troubles at all.
In contrast, look at how bad the problem is with a typical feedback amp. Douglas Self's website has an excerpt from his book that shows the problem caused by loading with a compensation capacitor:
Distortion In Power Amplifiers
Scroll down to Section 5.4 and see that the load due to the capacitor drops below 100 ohms at 100 kHz. So which is it? Loading the VAS is bad, or loading the VAS is good as long as you are using a lot of feedback?
Try it some time and let me know what you think.
I'm not claiming its a bad thing per se. I'm saying there's no evidence for any commensurate benefit for doing it. So - given no evident reason for adding the resistor, I won't add it.
You're missing the point again. You missed it when you critiqued Bob, and you missed it in your earlier post to me. So I'll run through it again.
Loading the VAS is not bad in itself. Its only sub-optimal when there's no corresponding benefit to doing it. That cap has the benefit of linearising the VAS - its still worth it even though it comes with downsides. A loading resistor de-linearises it and no evident upside to this.
Engineering is always about trade-offs. As engineers we're seeking an optimisation, given certain resources. Thinking in terms of 'bad' and 'good' things won't lead to optimal solutions, its religious not scientific or engineering thinking. The resistive loaded VAS is sub-optimal. Sub-optimal is not 'bad' its merely sub-optimal.
Any clearer now?
If you describe the differences in sound with it and without it, then I may well. Until such time, I will probably remain unpersuaded. Do you have any rationale for why it might sound different?
I could say, "Feedback sucks." But for some reason I don't think that will convince you. Never mind. As you were. Carry on. Self and the distortion analyzers have all the answers.
So in your estimation, feedback is bad? If so you're entitled to your opinion.
You're quite correct in this case and for reasons I've already set out. If you're saying 'feedback is bad' (and you haven't clarified that you are, so its just a working hypothesis at this stage) then you're making a religious statement, not an engineering or scientific one. I'm an engineer and find religious statements fundamentally flawed.
Somehow I find that unconvincing too.
As an engineer, Abrax... what kind of engineer are you? How many audio amps and preamps have you designed and are available for independent evaluation? I think this is a fair question, as you infer that we are not engineers, and therefore are somehow flawed in our reasoning. Of course, MY definition of 'engineer' might be different than yours. Does a piece of paper saying 'Engineer' make you an expert in this specific area?
- Status
- Not open for further replies.