We are much more sensitive to some types of distortion than others. And an untrained ear might pick a "euphonic" tube amp with 1% THD over a solid state amplifier with 0.001% distortion. And you could make the argument that picking the tube amp isn't even wrong, unless you're trying to choose an amplifier for laboratory or instrumentation purposes (in which case a listening test would mean nothing).
Remember, some musicians cherish their euphonic tube amplifiers. For a rock or blues guitarist, the distortion is part of the sound of the instrument; they even manipulate the distortion to make the tone go from "sweet" to that classic blues/rock wail. In fact, a lot of effort has been put into making solid state guitar amplifiers distort like tube amplifiers. And since wideband, high feedback amplifiers distort and recover from clipping way different than classic tube amplifiers (which you can clip with impunity from harsh distortion), this is not as easy as it might seem.
Remember, some musicians cherish their euphonic tube amplifiers. For a rock or blues guitarist, the distortion is part of the sound of the instrument; they even manipulate the distortion to make the tone go from "sweet" to that classic blues/rock wail. In fact, a lot of effort has been put into making solid state guitar amplifiers distort like tube amplifiers. And since wideband, high feedback amplifiers distort and recover from clipping way different than classic tube amplifiers (which you can clip with impunity from harsh distortion), this is not as easy as it might seem.
My rule of thumb is if an amp is at .005% or better at full power into 8 Ohms AND it clips cleanly AND has low full power IMD (so better than -90 dB) AND can drive a 2 Ohm load for at least 2 minutes, AND it is completely hum free AND its stable into any kind of capacitive load it's pretty much on the money from the engineering perspective. Funny, getting .005% or better THD is now easy with simulators, etc, but the other practical real-world stuff is not so easy. I guess you only find this stuff out when you build amps.
Maybe we could cook up a figure of merit scale to assess an amp?
I see far too much focus on distortion at the expense of other equally important aspects. I agree with PMA - DBT is the only way to assess subjective amplifier performance specifically.
Maybe we could cook up a figure of merit scale to assess an amp?
- Stability
- Load drive capability (Current)
- Noise (hum etc)
- Recovery from clipping
- THD
- IMD
I see far too much focus on distortion at the expense of other equally important aspects. I agree with PMA - DBT is the only way to assess subjective amplifier performance specifically.
The classic clipping circuit for guitar amplifiers consisted of a 12AX7 tube, carefully biased for symmetric clipping, driving a level control potentiometer. I saw circuits where they used a dual pot so turning it up increased clipping, but not volume. It's simple, foolproof, and makes a very distinctive (to my ears) effect on the sound - the classic wail of 1960s/1970s rock guitarists.
Solid state clipping circuits use diodes in the feedback path to simulate the sound, and probably other tricks. It works well I guess; what choice does a musician have nowadays unless they can find an old classic amp and have it serviced?
Solid state clipping circuits use diodes in the feedback path to simulate the sound, and probably other tricks. It works well I guess; what choice does a musician have nowadays unless they can find an old classic amp and have it serviced?
Last edited:
Stability and recovery from clipping go hand in hand. Extreme case is when an amplifier latches up.
Again, local feedback circuits do not suffer from this at all.
Again, local feedback circuits do not suffer from this at all.
We've just been discussing some of this over on another thread where the whole focus has been on THD to the point that the comp values have been tweaked because they lowered the distortion on the sim from 1 ppm to 500 ppb but there's yet to be a loop gain plot. We're still working on it 😉
There are quite a few ways to ensure clean recovery from clipping with no rail sticking or oscillation - but a lot of designers don't address this aspect seriously enough IMV.
There are quite a few ways to ensure clean recovery from clipping with no rail sticking or oscillation - but a lot of designers don't address this aspect seriously enough IMV.
Yes, there's a few tricks. It's a paramount issue to address, because if an amplifier latches up in service it's going to take the speaker with it.
I have been in a situation where I demo'd a feedforward amp to a friend and then noticed that one of the (wideband) speakers stopped working and turned out to be open voicecoil.
Fortunately, the speaker dealer was nearby, so we got a new driver, mounted it and all was well. For 5 minutes, when it stopped working again. Open circuit driver. That's when I got that sinking feeling in my stomach.
Jan
Fortunately, the speaker dealer was nearby, so we got a new driver, mounted it and all was well. For 5 minutes, when it stopped working again. Open circuit driver. That's when I got that sinking feeling in my stomach.
Jan
One advantage of a tube amps output transformer, no DC to the speaker. Or we could put a large cap in series with the output, and for closed box speakers it extends the low end too (when the value of the cap is right). Thanks for all your writings Jan!
I haven't had an oscilloscope for years. I have a circuit built on a very small board, one op amp that runs off a nine volt battery and a very short probe, that detects oscillation with an LED indicator. You can change a small capacitor and resistor on the board (they plug into a DIP-8 socket) and the circuit will tell you if there's a signal above 10 kHz, 100 kHz, 1 MHz, 10 MHz, etc just by changing the cap and/or resistor. It's mighty handy, especially since I probably couldn't see an oscilloscope trace but I can see an illuminated LED.
I read Jan's article when it was first published. The findings are unarguable but the assumption that the only variable is the frequency of the dominant pole is too simple.
An amplifier designed for 100dB of open-loop gain requires more stages than an amplifier designed for (say) 50dB of gain. The former needs a very low-frequency pole to use global feedback. The approach of designing for highest open-loop gain is suitable only for low frequency amplifiers. IMO, audio frequencies are high enough that one needs to choose the right amount of open-loop gain and feedback.
From reading the posts on this board, one can see that distortion is not even among the top problems. Much more problematic are stability (both electrical and thermal), hum, and plain old reliability. These areas separate the great from average designs.
Ed
An amplifier designed for 100dB of open-loop gain requires more stages than an amplifier designed for (say) 50dB of gain. The former needs a very low-frequency pole to use global feedback. The approach of designing for highest open-loop gain is suitable only for low frequency amplifiers. IMO, audio frequencies are high enough that one needs to choose the right amount of open-loop gain and feedback.
From reading the posts on this board, one can see that distortion is not even among the top problems. Much more problematic are stability (both electrical and thermal), hum, and plain old reliability. These areas separate the great from average designs.
Ed
From a point of view of error budgets, it is perfectly logical to design amplifiers for distortion levels well below audibility. When there is a certain distortion budget for the whole signal chain, you want to spend it on things that inherently distort a lot, rather than to waste it on amplifiers. If an amplifier then distorts so much you that can recognize it in a double-blind test by its distortion, it is way above its chunk of the budget.
Back in the day, tape decks and phono amps could introduce lots of distortion. It irks me just how awful many (most?) phono amps were back in the day. It's elementary to design and build a stellar phono amp, but unfortunately it's a few decades too late.
Not scientifically established but useful nonetheless:
• When you’re strapped for loop gain at 20kHz, limit low frequency loop gain to the same value.
• THD becomes higher but constant throughout the audio band.
• Colouration becomes less obvious and less annoying.
Bruno Putzeys, "An EE’s Guide to Survival Between Microphone and Voice Coil"
• When you’re strapped for loop gain at 20kHz, limit low frequency loop gain to the same value.
• THD becomes higher but constant throughout the audio band.
• Colouration becomes less obvious and less annoying.
Bruno Putzeys, "An EE’s Guide to Survival Between Microphone and Voice Coil"
Not so much in fact. It is the one and same non-linearity that creates both harmonic and intermodulation distortion. Could you show me a link to a literature that would proof high impact of IMD on audibility? Both HD and IMD are very poor measure of “sound quality” and audibility. You might read some papers of Earl Geddes, e.g.
What makes me quite sure? Tens of experiments with added nonlinear distortion to the original and results of DBT tests. Ear is very tolerant to nonlinear distortion on music, you may call it IMD if you like.
It is a closed chapter, to me, and I do not take serious any anecdotal stories not supported by DBT results. Not entering the same river we did about 10 years ago, it is over.
What makes me quite sure? Tens of experiments with added nonlinear distortion to the original and results of DBT tests. Ear is very tolerant to nonlinear distortion on music, you may call it IMD if you like.
It is a closed chapter, to me, and I do not take serious any anecdotal stories not supported by DBT results. Not entering the same river we did about 10 years ago, it is over.
IME if you listen to vocal harmonies, two people singing exact in balance, beat notes are created. They are not new frequencies, rather they are envelope amplitude modulations that result from adding frequencies. With even very low levels of distortion, the sound of beat notes changes. With some practice, IME one can get quite good at hearing the effect in the frequency band of the human voice.
Beyond that, just a reminder that a threshold of audibility is an estimate of the center of a bell curve distribution. It has historically been assumed that the distribution should be Gaussian. However, we now know that isn't always the case with real world data. The effect can be that outliers may be much more common than previously believed. More info on the subject at: https://www.edge.org/response-detail/11715
Beyond that, just a reminder that a threshold of audibility is an estimate of the center of a bell curve distribution. It has historically been assumed that the distribution should be Gaussian. However, we now know that isn't always the case with real world data. The effect can be that outliers may be much more common than previously believed. More info on the subject at: https://www.edge.org/response-detail/11715
People's audible perception varies greatly. Some people aren't aware of distortion they hear, while others hear stuff that isn't there.
I'm very empirical but if it doesn't sound right I go back and tweak. I have found that if you design and build to spec you're 97-100% on the bull's eye. Last speakers I built were 100% listenable first try; the pair before that required a few months of tweaking the crossover. Of course I learned a lot with the previous pair; it was the first pair I built using "modern" methods (simulation and manufacturer supplied T-S parameters). The crossover required incremental changes to make it "disappear." I just did the same thing with the crossover with the second design.
Electronics is way easier to hit the bulls eye first try; or maybe that's just my aptitude.
I'm very empirical but if it doesn't sound right I go back and tweak. I have found that if you design and build to spec you're 97-100% on the bull's eye. Last speakers I built were 100% listenable first try; the pair before that required a few months of tweaking the crossover. Of course I learned a lot with the previous pair; it was the first pair I built using "modern" methods (simulation and manufacturer supplied T-S parameters). The crossover required incremental changes to make it "disappear." I just did the same thing with the crossover with the second design.
Electronics is way easier to hit the bulls eye first try; or maybe that's just my aptitude.
Yers it is a simplification but the argument I was trying to debunk is that flat OL gain to 20kHz is somehow better. Such arguments are clearer when you only vary a single parameter. You are correct but it would not change the conclusions.
People often focus on the OL gain starting to roll of at say 10Hz, and that frightens them - how can an audio amp be any good with a roll off at 10Hz??
They go as far as to resistively load the Vas stage to squash the OL gain until it is as low everywhere as at 20kHz, thereby ruining the linearity below 20kHz.
Yet, when you grok the situation it isn't frightening at all.
Jan
Sure, no objections, but the race about THD+N < -130dB becomes quite pointless and it starts to be a marketing fight only. See ASR and the only important parameter SINAD at 1kHz/5W and product comparison charts based on this number. Naaah.