At lower volumes it becomes harder to hear distortion . I regularly listen to my system at
10mw peak. This is the same as a small radio playing low. I overbiased the outputs to 125ma so here it's class A. The 110db s/n is more important.
In my system tube distortion swamps any amp distortion .0007% at 1w
10mw peak. This is the same as a small radio playing low. I overbiased the outputs to 125ma so here it's class A. The 110db s/n is more important.
In my system tube distortion swamps any amp distortion .0007% at 1w
Crossover distorsion increases with lower volume, so it will be more audible with low volume. But then again "low volume" is relative...
The problem with THD being bad (as in badly correlated with "bad sound") is that no alternative has caught on. The Gedlee Metric is the only alternative I've even heard of, until now - Google gave me this article as the second link for gedlee metric, and it's on the gedlee.com site (and I thought I'd read everything there about this). It discusses alternatives proposed in 1937 and 1950 (and the basic reason they weren't adopted - measuring levels of individual harmonics wasn't practical for most electronics labs). One thing these all have in common is weighting higher harmonics more strongly than the lower ones in a distortion figure. This makes good sense as the higher harmonics are easier to hear.
We should call such a thing WHD, for Weighted Harmonic Distortion.
http://www.gedlee.com/Papers/THD_.pdf
We should call such a thing WHD, for Weighted Harmonic Distortion.
http://www.gedlee.com/Papers/THD_.pdf
Hmmm, distortion numbers can give a general indication of quality. But as it has been hinted at, you really want to see the spectrum. THD and IMD considered together give a good picture.
I have never heard an amplifier or signal stage with high distortion (2nd or 3rd) sound good. If you get distortion products below audibility it will sound good, not surprisingly the circuit design is also good and stable.
I have never heard an amplifier or signal stage with high distortion (2nd or 3rd) sound good. If you get distortion products below audibility it will sound good, not surprisingly the circuit design is also good and stable.
@benb. Thanks for the above link.
Dr. Earl Geddes thoughts may be helpful here https://www.diyaudio.com/community/...rl-geddes-of-gedlee-audio.371621/post-6636235 (transcript of interview here)
and a quick summary of Dr. Earl Geddes THD thoughts are here https://www.diyaudio.com/community/...istortion-speaker-drivers.294787/post-4787082 ...
Dr. Earl Geddes thoughts may be helpful here https://www.diyaudio.com/community/...rl-geddes-of-gedlee-audio.371621/post-6636235 (transcript of interview here)
and a quick summary of Dr. Earl Geddes THD thoughts are here https://www.diyaudio.com/community/...istortion-speaker-drivers.294787/post-4787082 ...
Well Mark,
You have a noise floor in the room. That compresses your dynamic range quite a lot.
I can easily measure -130 dB below 1 watt and do it routinely. Lower if I am very careful with cabling and grounds. I would say that anything below -100 dB from there is all imagination. Same for a 1 volt level. A lot of equipment shows 120 Hz hum peaks between -80 dBV and -100 dBV, and that is never complained about. Now that is a steady state tone, easily perceived. Try distortion with music already rich in distortion added as effects or in the instrument. Then (and I haven't done this), imagine what the distortion levels are in the loudspeaker. I'm using an RTX 6001.
What I am saying is that I (and most others these days) can easily see distortion products below what the human body can perceive. Some equipment performs at very low distortion levels (I'm talking power amplifiers too). So when you examine a spectrum with single (THD) and again with dual tones (IMD 19K & 20K), you do get a very good picture of performance.
The days of reporting a needle indication of THD are over. That isn't a complete picture. I'll report it simply because people are used to it, but I also give them the spectrum display. So when I speak of THD and IMD, I am referring to the entire spectrum displays, not a pointer.
Interestingly, I had the lowest distortion at a bias level of around 5 mA in a Symasym using MJW0302 and MJW0281 outputs. 8 R, 1 watt. It seemed happiest there over a wide power range. The crossover notch was gone. It was monitored using the output from an HP 339A into an oscilloscope locked to the output sine. So, I ran it at 5 mA, at high bias levels I was seeing supply noise more than notch or amplifier distortion. Supplies were bipolar 37 volts-ish as I recall.
You have a noise floor in the room. That compresses your dynamic range quite a lot.
I can easily measure -130 dB below 1 watt and do it routinely. Lower if I am very careful with cabling and grounds. I would say that anything below -100 dB from there is all imagination. Same for a 1 volt level. A lot of equipment shows 120 Hz hum peaks between -80 dBV and -100 dBV, and that is never complained about. Now that is a steady state tone, easily perceived. Try distortion with music already rich in distortion added as effects or in the instrument. Then (and I haven't done this), imagine what the distortion levels are in the loudspeaker. I'm using an RTX 6001.
What I am saying is that I (and most others these days) can easily see distortion products below what the human body can perceive. Some equipment performs at very low distortion levels (I'm talking power amplifiers too). So when you examine a spectrum with single (THD) and again with dual tones (IMD 19K & 20K), you do get a very good picture of performance.
The days of reporting a needle indication of THD are over. That isn't a complete picture. I'll report it simply because people are used to it, but I also give them the spectrum display. So when I speak of THD and IMD, I am referring to the entire spectrum displays, not a pointer.
Interestingly, I had the lowest distortion at a bias level of around 5 mA in a Symasym using MJW0302 and MJW0281 outputs. 8 R, 1 watt. It seemed happiest there over a wide power range. The crossover notch was gone. It was monitored using the output from an HP 339A into an oscilloscope locked to the output sine. So, I ran it at 5 mA, at high bias levels I was seeing supply noise more than notch or amplifier distortion. Supplies were bipolar 37 volts-ish as I recall.
So, -100dB from peak levels? If so, I could argue that the actual number is probably closer to -60dB. Or, I could argue that the actual number is probably below -110 dB. IME, it depends. If a brass band can be inaudible if buried at -60dB then how could anyone possibly hear anything below that? OTOH, if quite a few people (less than half probably) can hear 16-bit quantizing distortion at -93dBFS (calculated) when peak levels are set to as high as slightly above 0dBFS, then -100dB might be cutting it too close for some smaller number of individuals.
What's the difference between a brass band at -60dB and quantizing distortion (or TPDF dither, if used) down around -93dBFS? It would seem to do with signal correlation. Distortion and or noise more stronly correlated with an audio signal being listened to tends to be more audible than a non-correlated signal (brass band).
I would rather not have keep pointing this out, but PMA once put on a listening test with recordings of non-inverting unity-gain audio opamp buffers. He took measurements, of course. I sorted the files in order of distortion under double blind conditions, except one file I gave up on due the difficultly and prolonged concentration required. I sent him my sort list to him by PM before he announced which opamp was which and what distortion numbers were. He mentioned me in a a thread post for my accomplishment. Yeah, it was very hard, but IIRC I was using a Benchmark DAC-1, a 20-year old Bryston 4-B, and a pair of Yamaha NS-10 monitors. Probably with my current setup it would be much easier to hear lower levels of distortion.
IIRC, the audio opamps PMA used included LM4562 and TL-072. Some of the distortion numbers for opamps were under your -100dB number, not to mention that the sound of the opamps was muddied up by PMA's less that perfect data converters.
So, IME -100dB is not a definitive number. We may disagree on that, but unless you can find some scientific research to show I have to be wrong, then I think we may have to continue to disagree as to whether the number expressed in your opinion is correct or not.
Overall statement and observed artifact
in op is about right though.
with low bias at lower power levels
THD would be higher.
And also noted with mosfets at high
frequency. More bias needed to be linear.
with typical differential input / feedback and
bias around 20 to 30ma per device.
low level THD should be non existent.
likely the huge mystery to slew rate as well.
Even though the basic calculated slew rate
needed is typically much lower than expected.
if a amp differential is correcting a lot of distortion.
the signal coming in doesnt need much slew rate.
The rather ugly waveform it is correcting, does.
in op is about right though.
with low bias at lower power levels
THD would be higher.
And also noted with mosfets at high
frequency. More bias needed to be linear.
with typical differential input / feedback and
bias around 20 to 30ma per device.
low level THD should be non existent.
likely the huge mystery to slew rate as well.
Even though the basic calculated slew rate
needed is typically much lower than expected.
if a amp differential is correcting a lot of distortion.
the signal coming in doesnt need much slew rate.
The rather ugly waveform it is correcting, does.
Please don't loose track on context.
In sense of just SNR, it really depends what kind of system we're talking about.
Not from an audibility point of view, but also from a system sensitivity point of view.
So with a big line array with quite some high sensitive compression drivers, I can tell from experience (although you can also calculate it), that 100dB SNR isn't gonna make it.
So in this case the absolute noise figure is important. With digital systems with a digital volume control, this point is fixed, since it depends on the noise levels of the DAC.
In all systems, this depends on the noise levels of the power amplifier as well (since we have volume control before the power amplifier).
Also you want to have a bit of overhead, just in case. That doesn't have to be an extreme amount, but something like 12dB is nice to have.
So, in practice this often leads to something like 108-115dB SNR.
That doesn't mean we always need or hear all of that. In-room noise is often already around 30-40dB(A).
But going by case-by-case basis, isn't a very practical approach.
Keep in mind that a 16 bit systems, so CD players and music that is derived from there, have a hard limit at 98dB.
So going lower with THD in that sense, will just mean it will be covered up by all the noise. The masking effect will also be enormous at that level.
That being said, we also don't want to add more noise/distortion.
So to following stage needs to be at least 3dB better than 98dB.
With modern systems, there is music available that has been mixed and recorded with better quality.
But you have to really search for it.
From a practical sense, most 16 bit systems won't provide full potential. So sometimes that's also a reason to go for a better system. Just because all of a sudden the entire circuit is done better.
True, but a noise floor is not a hard floor like a concrete slab. Its more like the top of cloud. Its well known that humans can detect signals to some extent buried in a noise floor. That is exactly the reason dither is applied before truncation rather than after. Some of the lower level bit information (below the truncation threshold) when previously summed with noise then sometimes rises above the truncation threshold so that a extra bit or two of music can still be audible despite being in the noise. Works best with modern adaptive noise-shaped dither algorithms though. Also, dither works a little differently from the way stochastic resonance works, but in a way there is some partial similarity.
Last edited:
Hi b_force,
I was only trying to point out you have a restricted acoustic dynamic range. Average quiet noise floor is around 40 dB, max SPL for most people runs 110 dB, so we'll call it 120 dB. That gives an amazing 80 dB dynamic range which most folks simply do not have. Speaker efficiency and available power plays a role, but I doubt many will exceed 120 dB SPL no matter how much power and efficiency they have. My own systems range from 86 dB/watt to 99 dB/watt. So can we agree we have an optimistic dynamic range of 80 dB?
I went to electronics measurements to exclude the efficiency argument for the rest. I think I made that point, so we can leave it.
For digital sources, you have an encoded noise floor that is generally higher than todays DACs. I worked in recording studios, and in analogue days, only a Studer using DOLBY SR exceeded digital. I forget the exact figure, but it was impressive to the zero. Never mind with analogue we had headroom on tape and nice compression. With digital, when you run out of one's and zero's, it's over. Hard clip - so you must consider peaks and not average zero like we used to do.
16 bit systems may theoretically be 98 dB SNR, but they never got there. ENOB and other system noise interferes big time.
The human body can only detect some low level of sound. I forget what that was now, but I don't know if we have an 80 dB range as a biological system. Of course it is frequency dependent.
I think what I am trying to point out is that we have practical limitations as a human, and our measurement capabilities have long ago surpassed this. It still needs interpretations, you can't just parrot a number back. You have to look at the spectrum of signals plus distortion. Once the electronics passes this mark, you have to look at the transducers (speakers these days), signal sources and reproduction equipment (tape, record, FM radio or digital source). Each has its evils. However, we can measure and quantify everything these days, and it does correlate to how people perceive it when it's all good. I have predicted how something will sound based on the measurements I got and it was accurate.
I'm not about to engage in a debate with anyone, it only serves to make someone feel better they can debate something. I'll just say, try it. Educate yourself and come to your own conclusions. I lived as an audio tech from the 1970's up to now and progressed through investigations and test equipment the entire way. I've seen and looked into each fad, measured and listened to countless wire, connector and component types. Most of this is smoke and noise to make someone an expert or to sell product (or often both). It fortunately boils down to the science and physics, and this shouldn't surprise anyone. Worked in live concerts, acoustic and amplified, recording studios and reproduction industries. Pretty wide exposure.
I was only trying to point out you have a restricted acoustic dynamic range. Average quiet noise floor is around 40 dB, max SPL for most people runs 110 dB, so we'll call it 120 dB. That gives an amazing 80 dB dynamic range which most folks simply do not have. Speaker efficiency and available power plays a role, but I doubt many will exceed 120 dB SPL no matter how much power and efficiency they have. My own systems range from 86 dB/watt to 99 dB/watt. So can we agree we have an optimistic dynamic range of 80 dB?
I went to electronics measurements to exclude the efficiency argument for the rest. I think I made that point, so we can leave it.
For digital sources, you have an encoded noise floor that is generally higher than todays DACs. I worked in recording studios, and in analogue days, only a Studer using DOLBY SR exceeded digital. I forget the exact figure, but it was impressive to the zero. Never mind with analogue we had headroom on tape and nice compression. With digital, when you run out of one's and zero's, it's over. Hard clip - so you must consider peaks and not average zero like we used to do.
16 bit systems may theoretically be 98 dB SNR, but they never got there. ENOB and other system noise interferes big time.
The human body can only detect some low level of sound. I forget what that was now, but I don't know if we have an 80 dB range as a biological system. Of course it is frequency dependent.
I think what I am trying to point out is that we have practical limitations as a human, and our measurement capabilities have long ago surpassed this. It still needs interpretations, you can't just parrot a number back. You have to look at the spectrum of signals plus distortion. Once the electronics passes this mark, you have to look at the transducers (speakers these days), signal sources and reproduction equipment (tape, record, FM radio or digital source). Each has its evils. However, we can measure and quantify everything these days, and it does correlate to how people perceive it when it's all good. I have predicted how something will sound based on the measurements I got and it was accurate.
I'm not about to engage in a debate with anyone, it only serves to make someone feel better they can debate something. I'll just say, try it. Educate yourself and come to your own conclusions. I lived as an audio tech from the 1970's up to now and progressed through investigations and test equipment the entire way. I've seen and looked into each fad, measured and listened to countless wire, connector and component types. Most of this is smoke and noise to make someone an expert or to sell product (or often both). It fortunately boils down to the science and physics, and this shouldn't surprise anyone. Worked in live concerts, acoustic and amplified, recording studios and reproduction industries. Pretty wide exposure.
Mark,
People can detect periodic signals near a noise floor. You're right, it isn't a hard floor, but it doesn't extend very far down past that "Soft surface" if you will. When you talk music, the signals are not periodic. Noise, that mixes in nicely. Have fun separating that.
"Could be" arguments only really confuse the issue. The basic truth is still there. You can't avoid it, and using one specific case that isn't often found in normal listening does not advance any point of view.
People can detect periodic signals near a noise floor. You're right, it isn't a hard floor, but it doesn't extend very far down past that "Soft surface" if you will. When you talk music, the signals are not periodic. Noise, that mixes in nicely. Have fun separating that.
"Could be" arguments only really confuse the issue. The basic truth is still there. You can't avoid it, and using one specific case that isn't often found in normal listening does not advance any point of view.
Good 24-bit systems can certainly get there, including when reproducing well digitized 16-bit content. IME most of the limitations have been in the DACs.
BTW, the number is 96dB isn't it? 6dB per bit times 16-bits.
Why not mitigate the crossover, instead of playing with these (probably) meaningless measurements?
I am looking forward to experiment with the various biasing schemes that eliminate switching (for example, maintaining a minimum current of 20mA on both outputs at all times). In terms of heatsinking requirements it will not demand much more than usual AB biasing.
I am looking forward to experiment with the various biasing schemes that eliminate switching (for example, maintaining a minimum current of 20mA on both outputs at all times). In terms of heatsinking requirements it will not demand much more than usual AB biasing.
I read it, thanks. It could be a useful addition to the article the inclusion of the spice models used.
edit: I see now that no spice models were used, only the equations for BJTs and MOSFETs.
Thanks,
Alex
Last edited:
Hi Alexandre,
Look at the Nikko Alpha 220 and 230 bias circuits.
The measurements are not meaningless at all. The spectrum displays are the most meaningful and really tell all. Previously my point has simply been that we can these days measure far past what we can hear on any day with any material in any system. So we can see exactly what impairments a system may have. The argument that someone can hear what cannot be measured died well over a decade ago, that's all.
Look at the Nikko Alpha 220 and 230 bias circuits.
The measurements are not meaningless at all. The spectrum displays are the most meaningful and really tell all. Previously my point has simply been that we can these days measure far past what we can hear on any day with any material in any system. So we can see exactly what impairments a system may have. The argument that someone can hear what cannot be measured died well over a decade ago, that's all.
Hi Mark,
My main points are that the theoretical performance is never near the real performance for digital systems. Analogue filtering and signal conditioning also impact performance. Then there is the source of your signal. Even when recorded at 32 bits, when you manipulate the signal you end up with artifacts (in the digital realm).
But in the end, we are arguing about stuff that we, as humans, can no longer hear. It's past our abilities. That's when things work well. Talking about earlier material and real systems we once again come into issues you can hear. Easily measured.
My main points are that the theoretical performance is never near the real performance for digital systems. Analogue filtering and signal conditioning also impact performance. Then there is the source of your signal. Even when recorded at 32 bits, when you manipulate the signal you end up with artifacts (in the digital realm).
But in the end, we are arguing about stuff that we, as humans, can no longer hear. It's past our abilities. That's when things work well. Talking about earlier material and real systems we once again come into issues you can hear. Easily measured.