Soft vs hard clipping - I feel such a fool

[pdf64]

I seem to remember reading a report on a psychoacoustic test, where the conclusion was that the phase of the harmonic was significant in regard of its acceptability.
I think the whole odd/even thing is a distraction / irrelevent; rather if there's excessive high order content (which doesn't get filtered off by the speaker) then it will likely sound harsh.
But with a bad sounding circuit, the harmonic analysis is probably more of a symptom than a cause, rather it seems more likely that there's a bad design or failing component that's causing a problem, eg instability, latch up.

[Dave H]

I don't have a spectrum analyzer.

It looks like this Chuck. That's at only 5% THD. It does show symmetrical rounding of the peaks but I should do it again with harder clipping.

[pdf64]

Thanks Dave, what equipment etc are you using to get that?

[guitician]

Is that taken with a microphone or probe? The P-P circuit cancels some of the even harmonics. Bells have a tuned harmonic structure..
http://www.hibberts.co.uk/tuning.htm

Which would indicate that a perfect mix is really what the odd/even thing is about.

[Dave H]

what equipment etc are you using to get that?

It was a PicoScope 2204 USB oscilloscope and an ASUS netbook.

[eschertron]

I think the whole odd/even thing is a distraction / irrelevent; rather if there's excessive high order content (which doesn't get filtered off by the speaker) then it will likely sound harsh.

Since all even harmonics are by definition multiples of the root note (or fundamental) for most acoustic purposes they are the same note, just octaves higher. The odd numbered partials - especially higher order, as pdf64 says above - are overtones like the 7th, 9th, #11th, etc., that produce the clash we associate with dissonance. If there's enough acoustic energy in those partials, we hear 'buzzy' or 'harsh' tones. It's not so much 'presence' or 'absence', but the combined energy of the the higher partials that we judge.

The whole 'how we perceive overtones' thing has been done to death in academia, going back to before the advent of guitar amps! [citation needed]

[Dave H]

Since all even harmonics are by definition multiples of the root note (or fundamental) for most acoustic purposes they are the same note, just octaves higher.

Odd harmonics are also multiples of the root note . I think only the even harmonics in the series x2 x4 x8 x16 etc. are octaves, x6 for example is an octave above x3 which is a musical fifth and x10 is a major third.

Frequencies of Musical Notes

[eschertron]

Odd harmonics are also multiples of the root note . I think only the even harmonics in the series x2 x4 x8 x16 etc. are octaves, x6 for example is an octave above x3 which is a musical fifth and x10 is a major third.

Frequencies of Musical Notes

Yep. I misspoke. I should have said 'binary multiples' or some such thing. Thanks for being kind

The point I'm making is that in a triode stage if there's asymmetry we'll get a lot of the 2nd harmonic (octave), and not so much of everything else.

All of the low-order partials (as you point out, and others have too) are all what we consider consonant. Octaves, fifths, major-ish thirds. Even the now-accepted dominant 7th that makes all things jazz This takes us up the 8th partial. After that the overtones (odd series, since the evens are all upper-octave duplicates of what's already sounding) become more challenging to the ear, and when they are present with sufficient energy, tend to sound 'noisy' or 'dissonant'.

So when low-order partials dominate we hear "smooth". When high-order partials are emphasized we hear "harsh". And clipping at any gain stage will generate lots of high-order partials. If there's lots of energy in the low-order partials (with or without high-order partials present) then we perceive more of the fundamental because the low-order partials reinforce our notion of consonance.

When we generate harmonics (and we will in our guitar amps!) if the low-orders are not represented because we've tried to eliminate asymmetrical clipping, then we lose the smoothness of a naturally-decaying overtone series and get a bunch of higher-overtone 'buzz'. Not a judgement statement, and apologies if the word 'buzz' offends anyone.

[Dave H]

I found two more plots in the "archive"

The first one is the output of a Valve Junior (SE EL84) and the second one is a 12AX7. They both have about the same THD but the 12AX7 has more high order harmonics. Is that why the EL84 sounds better?

[Leo_Gnardo]

Yep. I misspoke. I should have said 'binary multiples' or some such thing. Thanks for being kind

The point I'm making is that in a triode stage if there's asymmetry we'll get a lot of the 2nd harmonic (octave), and not so much of everything else.

Pythagoras had it worked out a couple hundred years BC - it's integer fractions. Frequency ratio 2/1 of course that's an octave. 3/2 a fifth, 4/3 a fourth then we're on to major thirds, minor thirds then smaller intervals that are less ear-friendly but handy for jazz. Keep going and shortly you're out of the generally accepted "Western" scale zone but that hasn't stopped cultures in the mid-east where 17 intervals in an octave is one of the popular selections, and avant-garde composers who have sliced the octave into even smaller fractions for instance Harry Partch's 43 note scale.

[Chuck H]

...less ear-friendly but handy for jazz.

Thanks for the COFFEE ON MY KEYBOARD!

[Mark Hammer]

As I see it, the point is that with a non-inverting arrangement, the original signal is present alongside the clipped signal at the output of the clipping amp, whereas with an inverting arrangement, it is not.

Well now you have me baffled. How does the one consist of some sort of blend of two different signals, and the other does not?

[J M Fahey]

I found two more plots in the "archive"

The first one is the output of a Valve Junior (SE EL84) and the second one is a 12AX7. They both have about the same THD but the 12AX7 has more high order harmonics. Is that why the EL84 sounds better?

[ATTACH=CONFIG]35109[/ATTACH]

[ATTACH=CONFIG]35110[/ATTACH]

Thanks for posting and those graphs are very interesting.

They show way more data, "extra frequencies" , all kinds of amplitudes, so we should simplify them for useful analysis, take the most important parts of sound.

Very simplified analysis:
I consider (yes, pulled out of the blue ) that, say, -20dB harmonics (or components) can easily be heard and affect what we perceive in a significative way, -30dB do barely so and -40dB and below are lost in noise and general chaos.
Of course, any other scale can be used, it's arbitrary, but let's start with something.

Let's remember also that the EL84 is measured *after* a transformer, a band limiting device if I ever saw one (can't pass real highs, can't pass real lows, often has resonant peaks at or close to the audio band) and to boot, is driving a crazy impedance and full of resonances speaker, which to re-boot drops like a brick above 3000 to 4000Hz; while the poor 12AX7 probably has the scope probe clipped straight to its plate.

So EL84 shows strong 500Hz low order harmonics: -4 dB at 1K & 1K5 and some -10dB @ 2k Hz.
Close enough so musical relationship is clear and they are not considered mere "buzz".

2k5, 3k and 3k5 Hz are around -33/34 dB so barely contribute anything and all others -40dB or beyond: inaudible.
So EL84 will have "musical/fat/throaty" distortion (speaking in relative terms, of course).

While 12AX7 has very strong 1k5 , 2k and 2k5 harmonics which are "musical" under the above assumptions, it also has very strong and non decaying (logic would be that the higher components should be weaker and weaker, but that's not what we see in the graph) 2k5 , 3k , 3k5 , 4k , and still a last very listenable -20dB peak at 5k Hz
Not surprisingly, that 12AX7 will definitely be buzzy.

Amps like MBDR, which rely heavily on preamp distortion, cut highs all over the place to avoid that perception. (I perceive them as dull, though).

Uncompensated amps such as Soldano or Marshall JCM800 are perceived as buzzy or over bright.

[pdf64]

Well now you have me baffled. How does the one consist of some sort of blend of two different signals, and the other does not?

The gain of a non-inverting amp is 1 + (Rfb/Rin).
That means that original signal presented to the non-inverting input is present at its output, plus the amplified signal (if any, eg unity buffer).
Suggest that you have a play on your test bench with a TS clipping amp circuit.
Try it with a big input signal, eg 6V p-p, and the above may be apparent, ie the 6V p-p appears at the output with the 0.6V clipped signal added to it.

[Mark Hammer]

Yeah, but the original signal is always present at the output, whether inverting or non-inverting. The clipping merely adds harmonic components to the overall output signal which were not present in the original signal. Besides, unless the signal reaches the forward voltage of any clipping diodes (which is actually less time than one would think), all you're getting at the output is the original signal, plus whatever non-linearities and coloration the op-amp itself introduces.