Is distortion in electronics meaningful ?

[Harmonie]

"Meaningful", yes.

"Audible", probably not. .....

many visitors forget that most of us "geeks" turn off the technical side and just enjoy listening to (or making) the music without analyzing each atom along the way. My processor probably measures horribly compared to the stand-alone DACs, and my amps are certainly not state of the art, but it sounds good (enough) to me.

Wish there were more DonH56, RestorerJohn aso ...

 

[sergeauckland]

"Meaningful", yes.

"Audible", probably not. But it depends on the frequency of the distortion, system gain and volume, and so forth.

I remember when -80 dB (0.01%) THD was excellent and considered easily 10x lower than what anyone could hear. Now -100 dB is not enough.

As for ASR, it is nice to have a place to discuss technical aspects, though we don't get too deep too often (I'm as guilty as anyone, too tired for deep technical discussions at the end of a long day), but I think many visitors forget that most of us "geeks" turn off the technical side and just enjoy listening to (or making) the music without analyzing each atom along the way. My processor probably measures horribly compared to the stand-alone DACs, and my amps are certainly not state of the art, but it sounds good (enough) to me.

Speaking personally, I can turn off 'Geek Mode' when I'm listening to music because I know how well my equipment works, what it does well and perhaps what it doesn't do as well (especially when listening to LPs) so I don't need analysis when listening for pleasure. It's nice, however, to have the knowledge to turn on Geek Mode when something isn't sounding right, or when choosing or setting up equipment.

S

 

[Racheski]

Uncorrelated it should RSS -- root-sum-square -- so calculate each distortion term, square it, add all the terms, take the square root, and calculate the new distortion. So if the speakers are at -60 dBc and amp at -80 dBc then the distortion is (10^-60*2/20 + 10^-80*2/20) = 0.00000101 (using voltages) and converting back to dBc (20*log) the result is -59.96 dBc -- about 0.04 dB reduction increase from the amp. A DAC at -130 dBc is in the mud...

 

[Blumlein 88]

Was your amp near clipping and did you measure the distortion difference at the speakers? That’s where it would be most audible, right?

Distortion and noise were measured in the resulting file. As per Don above, 80 db down with the speaker being 60 db down will not matter effectively. And no my amps were not close to clipping.

 

[DonH56]

<typo in my previous post>

Fixed, thanks, vacillated on how I wanted to word it and naturally left it in an indeterminate wrong state.

 

[DonH56]

So let’s use a real edge case: DAC is a phone at 90 dB, speaker midrange THD is at 2.5% distortion and amp is at 1% (before dynamics). What RSS dB does that give you?

The math is outlined above... The speaker and amp distortion indicates levels near clipping.

-90 dBc = 0.003162% (approx) so is a non-player.
1% (-40 dB) and 2.5% (-32.04 dB) are close enough that the 1% will contribute a little.

I get about 2.7% net, about -31.40 dB, using all three terms.

 

[MediumRare]

The math is outlined above... The speaker and amp distortion indicates levels near clipping.

-90 dBc = 0.003162% (approx) so is a non-player.
1% (-40 dB) and 2.5% (-32.04 dB) are close enough that the 1% will contribute a little.

I get about 2.7% net, about -31.40 dB, using all three terms.

Thanks Don.

So my takeaways are that:

A. we are well into "audible" territory with very common measures (based on @amirm 's reviews). This is audible generally when a system is pushed hard (initially for transients/other high dynamics moments), but that is enough to be noticeable, isn't it?
B. Clearly, speakers are the biggest contributor.
C. Once you're in "audible" territory, the cumulative contribution of the source, amps, and anything else in the chain will have the *potential* to make it worse.
D. We need better tools to self-test audibility of distortion during peaks; steady tones are not the most likely use case. @pkane, can your tool help us with that?

 

[MediumRare]

Distortion and noise were measured in the resulting file. As per Don above, 80 db down with the speaker being 60 db down will not matter effectively. And no my amps were not close to clipping.

So, I think that is the nut of the conceptual issue. No criticism of you intended in any way. Certainly there is no audible difference between 0.01% and 0.001%. But what are the conditions under which the signal distortion does reach an audible state, even for a fraction of a second? When each element of the chain is at its worst (cones pushed, soft clipping, impedance+phase angle, etc.). That's the issue we really need to draw out to fully understand the requirements for true "transparency".

 

[DonH56]

Thanks Don.

So my takeaways are that:

A. we are well into "audible" territory with very common measures (based on @amirm 's reviews). This is audible generally when a system is pushed hard (initially for transients/other high dynamics moments), but that is enough to be noticeable, isn't it?
B. Clearly, speakers are the biggest contributor.
C. Once you're in "audible" territory, the cumulative contribution of the source, amps, and anything else in the chain will have the *potential* to make it worse.
D. We need better tools to self-test audibility of distortion during peaks; steady tones are not the most likely use case. @pkane, can your tool help us with that?

I don't know about (A). And I mean that literally; I am not sure just how much distortion is really audible. When you push a system hard, speakers are most likely to be the main distortion source unless your amplifier is severely clipping, and again that may be only on very loud peaks. For sine waves, if I flip back and forth I can hear 1% pretty readily, 0.1% barely better than guessing, at low levels via headphones that may or may not be contributing. Listening to music 1% is almost impossible to tell (by me), except in special cases where I know the music and what to listen for, and even then is tough. Certainly (well, maybe not "certainly", but pretty certain for me) 1% distortion at the peak of a cymbal hit, drum strike, or gun shot/explosion seems unlikely to be audible. If I inject 1% into a two-tone or multitone test I can hear it due to IMD products thus a little easier to pick out than for single tones.

For (C), it depends upon the relative levels of distortion. Ideally your sources, preamp, amp, and speakers all clip together to maximize dynamic range (at least at the upper end). Electronics, at least SS, tend to quickly go from completely inaudible to audible distortion (due to feedback keeping it low until that point) -- look at any amp THD chart and notice how quickly THD rises when you approach and exceed its power limits.

I also do not know the integration period of the ear/brain system. Can we really hear clipping if it is just on a peak for a fraction of a second? How large a fraction? Not something I know.

I am not sure we need better tools. You can use any AWG to generate tone bursts to listen for clipping. That is what I used to do, anyway, back when I was playing around with this stuff in the primordial past. Since we know the low-level stuff is unlikely to be a problem, you could just adjust the signal into the amp, but the problem is going to be tolerating high SPLs -- both your ears and the speaker's drivers. A test to destruction when it's my ears or speakers being destroyed is not a practical methodology, at least for me. If I do basic sweeps or single-tone tests to perhaps 100~110 dB, assuming within the capabilities of my amp and speakers, and the distortion is about 1% or even a few percent, then I'm good.

For that matter, when I was piddling with subwoofers, I found 10% or more distortion was actually pleasing to some folk -- it made the bass/sub "richer" or fuller. Why? My conjecture is that at say 60 Hz the fundamental is hard to hear, so when you add a 120 Hz component, there is "more bass" so people like it.

 

[MediumRare]

I don't know about (A). And I mean that literally; I am not sure just how much distortion is really audible. When you push a system hard, speakers are most likely to be the main distortion source unless your amplifier is severely clipping, and again that may be only on very loud peaks. .... Certainly (well, maybe not "certainly", but pretty certain for me) 1% distortion at the peak of a cymbal hit, drum strike, or gun shot/explosion seems unlikely to be audible.

So, I think you just convinced me that it is audible. Here's a perfectly good speaker and check the distortion peaks.

Certainly (well, maybe not "certainly", but pretty certain for me) 1% distortion at the peak of a cymbal hit, drum strike, or gun shot/explosion seems unlikely to be audible. If I inject 1% into a two-tone or multitone test I can hear it due to IMD products thus a little easier to pick out than for single tones.

I also do not know the integration period of the ear/brain system. Can we really hear clipping if it is just on a peak for a fraction of a second? How large a fraction? Not something I know.

That transient cymbal strike at the same time as the bass pluck and bass drum hit are, IMO, the MOST likely to be heard. After all, what is "graininess"? My guess is very transient distortion. Surely we can measure these things if we go looking for them specifically.

 

[pkane]

D. We need better tools to self-test audibility of distortion during peaks; steady tones are not the most likely use case. @pkane, can your tool help us with that?

Of course, that’s what it’s for!

 

[Blumlein 88]

One thing sometimes overlooked is playback level. Our ear seems best able to hear distortion at levels of 75 db or so. Around 80 db the ear tries to limit its sensitivity to protect against loud sounds, and with very loud sounds the ear is distorting enough to cover distortion in the sound.

I've read the account of some engineers at a pro amp company a few years back. They did blind testing of their amps. They let people set levels and found when it wasn't obvious they'd turn it up to hear things better. They found anyone who exceeded 80 db SPL to do poorly and those turning it up past 85 db even for a short time meant they couldn't hear rather large amounts of distortion for quite a while afterwards. They obtained more consistent and more discriminating results when they limited average spl levels to 75 db and didn't allow the testee to go above that. Unfortunately they didn't say what levels were audible from their testing.

An anecdotal bit. I'd worked around some machinery in an environment where the noise level was 105 db all the time. You can walk thru for a for a couple minutes safely, but you wouldn't want to work in there for long. It was all a cacophony. Put some headphones on or better some ear plugs in that knocked the level down to about 85 db or slightly lower and there were all kinds of intricate details to the sound. Heard loud it didn't even have any resemblence to the sound attenuated. You could easily hear differences in these machines that let you find parts about to go out or wear out. Also your ability to hear direction was good with headphones and nearly useless within the raw sound environment where everything sounded like it was coming from everywhere.

 

[Wes]

Audible .NE. annoying or negative in any way

But, it is conceivable that distortion could have negative effects on the listener even if not Audible...

It is not demonstrated that distortion in electronics is fungible with distortion in a transducer...

and... distortion is a catch-all term for a number of phenomena, some of which may be pleasing if inaccurate

Taking all the above into account, just focus on the speakers.

 

[o2so]

IMO you are hung up on the speaker performance a bit too much. THD+N is not like "everything in any unit below the worst unit doesn't matter". THD+N actually adds tones to the signal at each stage of the process. It's both additive and multiplicative. In other words, the unwanted (added) tones carry over to the next link in the audio chain and then are in turn are multiplied AND added to.

Consider all the amp reviews here which show IMD rising at HF, plus THD rising as the amp clips. A lot of discussion here has been about dynamics requiring 10x the sustained power level for peaks. The cumulative effect across a signal chain can absolutely be audible, especially if your high-end hearing is good (HF noise & clipping) and, in an extreme case, you are using sensitive IEMs.

We call ourselves "Science" but the truth, IMO, is there is still a lot of speculation on what is audible and what isn't. Amir will tell you training enhances your ability to detect the flaws and he DOES hear differences at the end of the chain - unless the FULL chain AFTER addition/multiplication is still "transparent". If Redbook requires 96 dB to be transparent before reaching your speakers that's a tall order.

The real benefit of our approach is to filter out the provable BS and snake-oil and provide visibility of the data. How you interpret the data, your price/value preferences, and your taste for shaping the signal is up to you.

Thank you. My original question was exactly about how to estimate the cumulative effect and, in that, what was the relative contribution of speakers vs electronics. It has now been clarified that electronics, unless very bad, are contributing so little to distortion that one could almost forget about it.

 

[o2so]

Uncorrelated it should RSS -- root-sum-square -- so calculate each distortion term, square it, add all the squared terms, take the square root, and calculate the new distortion. So if the speakers are at -60 dBc and amp at -80 dBc then the distortion is (10^-60*2/20 + 10^-80*2/20) = 0.00000101 (using voltages) and converting back to dBc (20*log) the result is -59.96 dBc -- about 0.04 dB reduction in performance from the amp. A DAC at -130 dBc is in the mud...

Thanks, this is exactly what I was asking. But I think you meant 0.04dB reduction from the speakers, not the amp.

 

[DonH56]

Thanks, this is exactly what I was asking. But I think you meant 0.04dB reduction from the speakers, not the amp.

I'm wording-challenged. In this case, the speakers alone were at -60 dBc, so the amp's -80 dBc additional distortion only reduced the speaker's performance by 0.04 dB. In other words, the amp's contribution to the net distortion was only 0.04 dB (nil).

 

[bobbooo]

The math is outlined above... The speaker and amp distortion indicates levels near clipping.

-90 dBc = 0.003162% (approx) so is a non-player.
1% (-40 dB) and 2.5% (-32.04 dB) are close enough that the 1% will contribute a little.

I get about 2.7% net, about -31.40 dB, using all three terms.

Could the contribution of the whole audio reproduction chain's distortion (smartphone + amp + speaker) to the net distortion of the audio reproduction + production distortion be calculated i.e. including typical recording mic distortion?

 

[AnalogSteph]

Could the contribution of the whole audio reproduction chain's distortion (smartphone + amp + speaker) to the net distortion of the audio reproduction + production distortion be calculated i.e. including typical recording mic distortion?

Basically yes, but it would be depending on a gazillion things, including:
* Recording level, mic type(s), mic preamp type and gain setting
* playback level

At least the recording side gear tends to be reasonably well-behaved and predictable, with dominant low-order distortion and well-defined 2nd/3rd order intercept points.

Besides, adding distortion components is always tricky business (well, complex business, to be more specific - they've got an amplitude and a phase). You could have two components with asymmetries going in opposite directions, which would create even-order harmonics of opposite phase. When cascading both, those would tend to cancel each other out - so total distortion would be less than for both individually. General calculations will only result in a worst-case estimate (both in phase).

 

[DonH56]

Could the contribution of the whole audio reproduction chain's distortion (smartphone + amp + speaker) to the net distortion of the audio reproduction + production distortion be calculated i.e. including typical recording mic distortion?

Sure, if you knew all the parameters, but in practice, no... Beyond all the equipment to make the recording (mics, preamps, line amps, mixers, recorders, etc.) you have mixing (probably using a digital audio workstation, DAW), mastering (ditto), perhaps additional production steps, etc. And you'd need to know signal levels (amplitude and frequency) at each point along the process, the noise and distortion characteristics of each component (plus whatever is intentionally added in mixing and mastering), etc. And the RSS approach does not always apply perfectly -- some components may add more linearly, others may actually cancel, etc. A rather daunting task.

Just estimating typical and peak distortion and noise levels in your own playback system is a lot of work, and that is a fairly well bounded system (except at the speaker, where most of us do not know the real impedance, sensitivity, and distortion characteristics of our speakers).

 

[FeddyLost]

I think there are at least 2 reasons
1) currently, with availability of great measuring DAC chips, mediocre designer can get "decent" numbers with converter, performing bad in real world use case just because high SINAD on test bench now can be measured even in Realtec integrated codec.
So, to be sure that designer made some efforts and maybe even listened his converter, is good idea to buy something better, than just "already non-audible SINAD".
2)AFAIK, distortions are mostly multiplying on each stage
Think about real signal as about some weird multitone. So, it's like sending multitone into DAC, then (with higher noise shelf and more tones) into power amp and then (again with higher shelf and even more tones) into speaker driver. It's the best case, without preamps and passive crossovers.

And I can't say for sure that distortion is totally not correlated. At least in case of amp and speakers, usually in LF voltage swing and woofer displacement are very correlated.

So, I'd look at equipment with best test results in your budget. Just to be on safe side.