[Joe Bloggs]

 
A signal consists of multiple frequency. A clipped waveform has some DC component to it. An inductor (speakers) appearsto DC as just a wire. So the DC content is what destroys the speaker if it is large enough.

...no?

@dvw
 How much DC component? If you integrate an asymmetrical signal you might find some DC offset, I suspect that this should not be an issue in practice. Do you expect to have a very loud pulse waveform that has a crazy duty cycle?

If the original signal has no DC offset should we expect one after we clip it?

 

[StanD]

If the original signal has no DC offset should we expect one after we clip it?

In theory it's riding on a complex waveform (all of the music) so anything is possible, asymmetric clipping, albeit unlikely to be significant. In the big picture it probably averages out to nothing. I wouldn't be losing sleep over it, would you? I thought not.
 
I think we have to be careful not to seek out esoteric reasons or just any old theory and expect that to be the real reason for something or another. Some folks enjoy doing that, not me. I suspect that you would agree.

 

[dvw]

   
 

@dvw How much DC component? If you integrate an asymmetrical signal you might find some DC offset, I suspect that this should not be an issue in practice. Do you expect to have a very loud pulse waveform that has a crazy duty cycle?

A perfect symmetric square wave will have no DC component. However, clipping is caused by limitation of the voltage swing or current limited. In practice, you will need a perfect symmetry in plus and minus supply to maintain perfect symmetry. If not you created a DC offset. The amplitude of the DC depends on severity of the clipping and the duty cycle as well as the time period. The Telarc release of the 1812 overture has a warning that it might damage equipment because there is a huge cannon shot at the end. If you do not have adequate power, you will blow the speaker. In my case, I have glasses shaked off the table but no blown speaker. A reality check is today's music are very compressed and we do not have large variation in amplitude in today's recording. So unless you are playing the music very loud, you should not see this happens,
 
And also clipping also creates significant amount of high frequency component. But I don't know the impact of that.

 

[StanD]

  A perfect symmetric square wave will have no DC component. However, clipping is caused by limitation of the voltage swing or current limited. In practice, you will need a perfect symmetry in plus and minus supply to maintain perfect symmetry. If not you created a DC offset. The amplitude of the DC depends on severity of the clipping and the duty cycle as well as the time period. The Telarc release of the 1812 overture has a warning that it might damage equipment because there is a huge cannon shot at the end. If you do not have adequate power, you will blow the speaker. In my case, I have glasses shaked off the table but no blown speaker. A reality check is today's music are very compressed and we do not have large variation in amplitude in today's recording. So unless you are playing the music very loud, you should not see this happens,
 
And also clipping also creates significant amount of high frequency component. But I don't know the impact of that.

This is an unlikely scenario...I will sleep well tonight, unless it snows some more.

 

[castleofargh]

1/ measure the amp with an equivalent load at usual loudness/voltage to make sure there is no clipping.
2/ there is no 2/
 
this feels to me like the argument for tube amp about how clipping sounds nicer with a tube. it sure is true, but why would we go and use a system that clips in the first place?

 

[StanD]

  1/ measure the amp with an equivalent load at usual loudness/voltage to make sure there is no clipping.
2/ there is no 2/
 
this feels to me like the argument for tube amp about how clipping sounds nicer with a tube. it sure is true, but why would we go and use a system that clips in the first place?

+1
Clipping for a normal listener, as I said an "unlikely scenario," for a loud lead electric guitarist, very likely.

 

[pinnahertz]

 A perfect symmetric square wave will have no DC component. However, clipping is caused by limitation of the voltage swing or current limited. In practice, you will need a perfect symmetry in plus and minus supply to maintain perfect symmetry. If not you created a DC offset. The amplitude of the DC depends on severity of the clipping and the duty cycle as well as the time period. The Telarc release of the 1812 overture has a warning that it might damage equipment because there is a huge cannon shot at the end. If you do not have adequate power, you will blow the speaker. In my case, I have glasses shaked off the table but no blown speaker. A reality check is today's music are very compressed and we do not have large variation in amplitude in today's recording. So unless you are playing the music very loud, you should not see this happens,
 
And also clipping also creates significant amount of high frequency component. But I don't know the impact of that.

This is an unlikely scenario...I will sleep well tonight, unless it snows some more.

I have analyzed the effects of clipping extensively to determine its impact on DC offset, RMS power, high frequency content and how all of that might relate to the idea that clipping blows drivers. I can add more detail later when I'm not typing with two thumbs, but the short story is: clipping to an audibly objectionable extent doesn't do any of that. Moving up into clipping increases RMS energy more slowly than it would without clipping. In short, the entire clipping thing is myth.

 

[pinnahertz]

  A perfect symmetric square wave will have no DC component. However, clipping is caused by limitation of the voltage swing or current limited. In practice, you will need a perfect symmetry in plus and minus supply to maintain perfect symmetry. If not you created a DC offset.

This is not correct. Asymmetrical power supply rails do not necessarily cause a DC offset. In fact, usually not. For clipping to be consistently asymmetrical an amplifier would have to suffer a rather significant design flaw. Lacking that, musical signals are remarkably symmetrical. There are individual instruments and voices that have some asymmetry, but once combined into a mix with ambience, the result is average symmetry.  If an amp has severely asymmetrical rails and is AC coupled, hard clipping will momentarily shift the baseline.  But if the amp is DC coupled, that will not happen at all.  A momentary baseline shift is not DC in any case, it's AC, because it is changing, and thus won't ever have the high RMS equivalent of DC.

The amplitude of the DC depends on severity of the clipping and the duty cycle as well as the time period.

Actually, assuming the amp doesn't have a DC offset at idle, asymmetrical clipping (again, related to a design flaw) would cause a variable offset at worst. If the amp is DC coupled, even asymmetrical clipping won't cause a DC offset.

The Telarc release of the 1812 overture has a warning that it might damage equipment because there is a huge cannon shot at the end.

That was true.

If you do not have adequate power, you will blow the speaker.

That is not true. Driver damage is caused by two things: thermal damage caused by high RMS power and physical damage caused by over-excursion. Either one would be more likely with MORE amp power, not less.  And a clipping amp produces less power than the same input, same gain, but no clipping. 

And also clipping also creates significant amount of high frequency component. But I don't know the impact of that.

This is actually not true. The additional HF energy caused by clipping is minimal, and contributes virtually nothing to the total energy of the signal. There is no question that clipping should be avoided. However, the persistent misconception that clipping damages speakers is wrong for several reasons. Clipping does not impress DC, does not significantly raise HF content, and does not produce more power more quickly than a non-clipping amp. Rather, the reverse is true. If anyone still cares about this tangential subject I can post a graph of unclipped RMS vs clipped RMS power, which will make it obvious that an unclipped amp can and does produce far more damaging power than an clipping amp.

 

[Joe Bloggs]

I did some investigation and the results are ambiguous.

Firstly, the additional HF energy is indeed minimal within the audible range:
http://www.head-fi.org/t/823444/i-bought-the-hd600-whats-next-amps#post_12954966

Where it gets interesting is in supersonic frequencies. Square waves have infinite bandwidth and depending on the bandwidth of the amplifier and sharpness of the clipping, the frequency response of the clipped signal can extend way, way beyond the audible range and even the operable range of the tweeters. Here I take a sample music clip at 44.1kHz, resample it to 384kHz, boost it by 20dB (with attendant clipping) and analyse the frequency response before and after:



Looking at the orange tail it seems there is indeed quite a bit of excess supersonic power. How much, exactly? I amplified the same clip by 20dB again, clipping one copy while keeping the other unclipped (using floating point processing) and filtered out pretty much all content below 22kHz. The result is as follows:


Lest there is any confusion, the bottom waveform is from the clipped copy.

Per the FR analysis of the 1st graph, most of the energy of this excess waveform lie in the "low" frequencies just above 20kHz, frequencies the amplifier is most likely going to pass through and the tweeter respond to.

 

[StanD]

I did some investigation and the results are ambiguous.

Firstly, the additional HF energy is indeed minimal within the audible range:
http://www.head-fi.org/t/823444/i-bought-the-hd600-whats-next-amps#post_12954966

Where it gets interesting is in supersonic frequencies. Square waves have infinite bandwidth and depending on the bandwidth of the amplifier and sharpness of the clipping, the frequency response of the clipped signal can extend way, way beyond the audible range and even the operable range of the tweeters. Here I take a sample music clip at 44.1kHz, resample it to 384kHz, boost it by 20dB (with attendant clipping) and analyse the frequency response before and after:
 

Spoiler: Warning: Spoiler!

Looking at the orange tail it seems there is indeed quite a bit of excess supersonic power. How much, exactly? I amplified the same clip by 20dB again, clipping one copy while keeping the other unclipped (using floating point processing) and filtered out pretty much all content below 22kHz. The result is as follows:


Lest there is any confusion, the bottom waveform is from the clipped copy.

Per the FR analysis of the 1st graph, most of the energy of this excess waveform lie in the "low" frequencies just above 20kHz, frequencies the amplifier is most likely going to pass through and the tweeter respond to.

The legend in the lower graph seems a bit small to read the scale. So what are the actual levels delivered to the tweeter and how high does one have to crank up the volume to get anything significant?

 

[pinnahertz]

I did some investigation and the results are ambiguous.

Firstly, the additional HF energy is indeed minimal within the audible range:
http://www.head-fi.org/t/823444/i-bought-the-hd600-whats-next-amps#post_12954966

Where it gets interesting is in supersonic frequencies. Square waves have infinite bandwidth and depending on the bandwidth of the amplifier and sharpness of the clipping, the frequency response of the clipped signal can extend way, way beyond the audible range and even the operable range of the tweeters. Here I take a sample music clip at 44.1kHz, resample it to 384kHz, boost it by 20dB (with attendant clipping) and analyse the frequency response before and after:



Looking at the orange tail it seems there is indeed quite a bit of excess supersonic power. How much, exactly? I amplified the same clip by 20dB again, clipping one copy while keeping the other unclipped (using floating point processing) and filtered out pretty much all content below 22kHz. The result is as follows:


Lest there is any confusion, the bottom waveform is from the clipped copy.

Per the FR analysis of the 1st graph, most of the energy of this excess waveform lie in the "low" frequencies just above 20kHz, frequencies the amplifier is most likely going to pass through and the tweeter respond to.

Your level references are a bit ambiguous.  What you have to do is normalize the unclipped waveform to 0dBFS first, then apply gain.  BTW 20dB past clipping is absurd.  Never happens because nobody could actually do that.  Preamps would not be capable of driving an amp that hard, and it would sound so incredibly awful it wouldn't be tolerated for more than a few seconds.  Try +3dB (sounds horribly, but is entirely possible to do), on up to +10dB.  
 
As to the HF spectrum, thats great, but did you look at the level of the resulting HF above 20K?  Did you translate that to power applied?  DId you look at total RMS over time?  If you do you'll quickly see it's simply a non-issue, even at the absurd +20dB level. 

 

[Joe Bloggs]

As to the HF spectrum, thats great, but did you look at the level of the resulting HF above 20K?  Did you translate that to power applied?  DId you look at total RMS over time?  If you do you'll quickly see it's simply a non-issue, even at the absurd +20dB level. 

No but if you simply look at the bottom waveform graph it translates to waveforms approaching full scale with exclusively frequencies above 22.05kHz which seems pretty dangerous to me.

At somewhere close to 20dB past clipping granted but still.

 

[Joe Bloggs]

I did some investigation and the results are ambiguous.


Firstly, the additional HF energy is indeed minimal within the audible range:
http://www.head-fi.org/t/823444/i-bought-the-hd600-whats-next-amps#post_12954966


Where it gets interesting is in supersonic frequencies. Square waves have infinite bandwidth and depending on the bandwidth of the amplifier and sharpness of the clipping, the frequency response of the clipped signal can extend way, way beyond the audible range and even the operable range of the tweeters. Here I take a sample music clip at 44.1kHz, resample it to 384kHz, boost it by 20dB (with attendant clipping) and analyse the frequency response before and after:

 

Spoiler: Warning: Spoiler!

Looking at the orange tail it seems there is indeed quite a bit of excess supersonic power. How much, exactly? I amplified the same clip by 20dB again, clipping one copy while keeping the other unclipped (using floating point processing) and filtered out pretty much all content below 22kHz. The result is as follows:




Lest there is any confusion, the bottom waveform is from the clipped copy.


Per the FR analysis of the 1st graph, most of the energy of this excess waveform lie in the "low" frequencies just above 20kHz, frequencies the amplifier is most likely going to pass through and the tweeter respond to.

The legend in the lower graph seems a bit small to read the scale. So what are the actual levels delivered to the tweeter and how high does one have to crank up the volume to get anything significant?

Well it's relative to full scale, where one would assume that "full scale" here would be "pretty dang loud" given that it's clipping.

No nothing scientific here, move along...

 

[pinnahertz]

No but if you simply look at the bottom waveform graph it translates to waveforms approaching full scale with exclusively frequencies above 22.05kHz which seems pretty dangerous to me.
 

But what you are looking at there is a graph of very brief peak energy, not RMS.  Even so, most land below the .5 point, which would be 6dB below 0dBFS.  If you had a 100W amp, those peaks land at 25W (peak).   If you apply RMS analysis over a reasonably time window, I'm sure you'd find them many, many dB down, like at least another 10dB or more, placing them at 2W or so.   Remember, drivers are blown by heating.  RMS is "heating value".  You can't do this by just looking a peak energy.
 

At somewhere close to 20dB past clipping granted but still.

Please understand that 20dB past clipping is a ridiculous amount of clipping, and far, far beyond any real life situation.  As I said before, it's actually impossibly for a preamp to drive a power amp 20dB past clipping without the preamp clipping first.  And if you actually listen to the audio you get at that point, nobody would stand for that for a tiny instant.  What you are doing is saying "If you drive a car at 300mph into a brick wall, you totally destroy the car...and the wall".  Sure, but you can't actually do that.  
 
Drop your analysis to something realistic, like 5dB of clipping, run RMS analysis on the result, you'll find reality.  I took my analysis up from clipping threshold to +10dB in 1dB steps just to see what happened, and even that was past any real-life possibility. 

 

[StanD]

Fact or fiction?

http://www.head-fi.org/t/748067/official-schiit-magni-modi-2-uber-thread/2340#post_13364889