To quote a Wireless World article from 1955:-
"Even harmonics have to be higher than the odd before
they are noticeably discordant. The lowest odd harmonic
that sounds definitely discordant is the 7th, but the
lowest discordant even harmonic is the 14th."
So it looks like the 7th harmonic distortion level could be an indicator of amplifier worthiness. Personally, I think high levels of this could be what we perceive as harshness.
I wonder though if the significance of intermodulation distortion levels are vastly underestimated by using 2-tone testing. Music with its multitude of frequencies would be generating intermodulation products right across the audible frequency band bearing no relationship to the music; i.e. generating a noisy 'hash' that masks the subtle details that differentiate good amps from bad. This, I would relate to a 'boring' sound lacking subtlety.
All guesswork of course. But I know what I like.
"Even harmonics have to be higher than the odd before
they are noticeably discordant. The lowest odd harmonic
that sounds definitely discordant is the 7th, but the
lowest discordant even harmonic is the 14th."
So it looks like the 7th harmonic distortion level could be an indicator of amplifier worthiness. Personally, I think high levels of this could be what we perceive as harshness.
I wonder though if the significance of intermodulation distortion levels are vastly underestimated by using 2-tone testing. Music with its multitude of frequencies would be generating intermodulation products right across the audible frequency band bearing no relationship to the music; i.e. generating a noisy 'hash' that masks the subtle details that differentiate good amps from bad. This, I would relate to a 'boring' sound lacking subtlety.
All guesswork of course. But I know what I like.
You want a spectral regrowth test? Now THAT's a can of worms. Put in a noise signal with a X Hz bandwidth, centered at Y Hz, and calculate the amount of noise power produced at (Y-X) and (Y+X) combined, in an X Hz bandwidth. That's the way they test the linearity of cell phone amplifiers, in the 2 GHz range. A 2 tone test will give a similar figure of merit, even if you can't use it to directly calculate ACPR.
But spectral regrowth is exactly what happens, and is what really matters to reproducing music. It's a PITA to measure at low frequencies. Actually easier at RF.
But spectral regrowth is exactly what happens, and is what really matters to reproducing music. It's a PITA to measure at low frequencies. Actually easier at RF.
I went to a piano several decades ago and played the various harmonic distortion notes along with the fundamental, and learned that only the second harmonic is not necessarily damaging. The third is good if the chord is a major chord, but causes dissonance if the chord is minor. Above the 3rd it was hit and miss as to whether the distortion product note was musically pleasing, in conjunction with the first few harmonics. The higher up you went, the less often the note was a good thing. Try it for yourself.
I agree with your second part, where you point out that the accumulation of low distortion amounts with the complexity of actual music becomes significant, as opposed to what typical test methods show. I think it applies to both I.M. and Harmonic distortion.
If you want that "fizzle" sound, go for more higher order harmonic distortion products. The heavy metal guitarists basically turn their waveforms into squarewaves, which are rich in odd harmonics that extend way out, but sound to me like fingernails on a blackboard. It can get hard to hear the fundamental note through all the distortion.
If you want a more pure and natural sound, all the triodes I've looked at on a spectrum analyzer (6SN7 for example, and many others) give you a rapidly rolling off distortion spectrum, with the 2nd harmonic being higher than the 3rd, and not a lot after that. That's as good as it gets IMO. I think that's why many hard-core audiophiles are doing everything with single ended triodes.
A slightly unbalanced push-pull output circuit can work much like a triode too, if the crossover distortion issue is under control. Some topologies do this better than others. Tubes are never perfectly matched so you don't have to actually unbalance the circuit if there's only a small amount of negative feedback going on. With tubes you don't need much feedback, if any. Transistors generally need a lot of feedback since they're a lot less linear. The best reason to have maybe 12dB of negative feedback in a tube circuit (for Hi-Fi), is so the amps output impedance will be well below an ohm, so it doesn't let the woofer ring at it's resonant frequency (often around 60HZ), and so you don't throw off the calibration of a passive crossover network.
Your explanation of phase margin is one of the better ones I've seen. Not sure about 2 pole rolloff rates though, since they will approach having 180 degrees of phase shift sooner in frequency than a one pole. The idea is to roll off the open loop gain down to one before the phase shift gets near 180 degrees. Done with a single pole, you'll have a significantly better phase margin when the loop gain gets to one. It's hard to explain this to most people without drawing graphs. Phase shift happens substantially before the amplitude response shows significant rolloff. The rest of what you said is right on.
And a good one like Neumann KH120 (5" woofer) or ME Geithain 904 (6.5") below 0.5% at 90dBspl and above 100Hz.
The reason you may have two different distortion specs is probably to do with at what output power the distortion is quoted at. At lower output, you can expect lower powers.
On my e-Amp, I spec'd it at 180 W into 8 Ohms because at that output, distortion was about 10 ppm while the amp is quite capable of producing 220 W at 0.1%.
You are unlikely to notice any difference. Richard Marsh's advice is good in this regard.
If crossover distortion is a concern, then the THD number at full power is meaningless. Look for the THD numbers at 1W and below.
Is there a fundamental design difference between an amp whose THD increases approaching full power as opposed to one that decreases?
Hi,
THD+N is easy to measure by simply nulling methods.
Its the de facto industy standard, and irrefutable.
Fact is though its nearly totally subjectively meaningless.
1% noise and a perfect amplifier will sound vastly superior
to an amplifier with 0.1% noise and grim 0.9% distortion.
That speakers produce more distortion is true, but they
have great difficulty producing the sort of grim distortions
anplifiers can, but again THD+N is the way speakers are
measured for distortion, though obviously N is meaningless.
However because speaker distortion is higher its often plotted
usefully up to say the 7th harmonic, to see trends in the even
and odd order harmonics, but for an amplifier you need to look
at the harmonic spray up to at least the 50th for bad designs.
(Of course the 50th makes no real sense for 1KHz being 50KHz,
but underbiasing causes a spray of harmonics for 1KHz that
don't particularly reduce up to 30KHz, so what about 300Hz?)
Understand the number for what it is and isn't.
THD+N will be dominated by N at low levels, if THD is low,
but its fairly easy to average out noise, pointless at high level.
So the way things work the power that causes say 5%
THD+N distortion just says its clipping, nothing else.
rgds, sreten.
THD+N is easy to measure by simply nulling methods.
Its the de facto industy standard, and irrefutable.
Fact is though its nearly totally subjectively meaningless.
1% noise and a perfect amplifier will sound vastly superior
to an amplifier with 0.1% noise and grim 0.9% distortion.
That speakers produce more distortion is true, but they
have great difficulty producing the sort of grim distortions
anplifiers can, but again THD+N is the way speakers are
measured for distortion, though obviously N is meaningless.
However because speaker distortion is higher its often plotted
usefully up to say the 7th harmonic, to see trends in the even
and odd order harmonics, but for an amplifier you need to look
at the harmonic spray up to at least the 50th for bad designs.
(Of course the 50th makes no real sense for 1KHz being 50KHz,
but underbiasing causes a spray of harmonics for 1KHz that
don't particularly reduce up to 30KHz, so what about 300Hz?)
Understand the number for what it is and isn't.
THD+N will be dominated by N at low levels, if THD is low,
but its fairly easy to average out noise, pointless at high level.
So the way things work the power that causes say 5%
THD+N distortion just says its clipping, nothing else.
rgds, sreten.
Last edited:
For low harmonics like 2nd and 3rd it has been suggested that 0.1% is very roughly the limit of audibility. This applied to ClassA both push/pull and single ended and to both Valve/Tube and Solid State.
For crossover distortion of push/pull ClassAB, again in both Valve/Tube and Solid State, where the distortion is high level (relative to the fundamental), short duration, higher harmonics, the audibility is probably less than 0.01%.
One must differentiate between low order harmonic distortion and cross-over distortion. They are, audibly, very different. One can sound nice, the other never sounds nice.
For crossover distortion of push/pull ClassAB, again in both Valve/Tube and Solid State, where the distortion is high level (relative to the fundamental), short duration, higher harmonics, the audibility is probably less than 0.01%.
One must differentiate between low order harmonic distortion and cross-over distortion. They are, audibly, very different. One can sound nice, the other never sounds nice.
Last edited:
Can you give an example of cross-over distortion? In what audible/practical form can it be described? Sibilance? Poor transient performance? FR?
In an extreme case, it sounds like fuzz or static. As it is reduced it is not immediately noticeable as such, but can result in a loss of clarity even at low levels. If you delibrately under bias a class B amplifier you can get an idea what it "sounds" like in the extreme. If you don't want to tinker with a working amp, put a pair of inverse-parallel diodes in series with your speaker wire. It will sound horrible at low levels, and reduce as you crank it up (but still be a bit "fuzzy"). The "fuzz" is high odd-order intermodulation distortion. You will note that it sounds "worse" with music than it does with a single tone.
Crossover manifests itself in vocal tones that are noise based. Like "S", "F", and "T" letters (but is not limited to !!). When, indeed you'd underbias an amp to the extreme, you'll notice these to be much more pronounced/sharp, contrary to the original sound. This is clearly detectable. Small amounts of X-over will cause irritation over longer periods of listening and aren't immediately distinguishable.
P.S. I rated this thread 5 stars. Nice, honest objective discussion and experience sharing going here 🙂
P.S. I rated this thread 5 stars. Nice, honest objective discussion and experience sharing going here 🙂
Last edited:
As if crossover distortion wasn't bad enough, with it's spectral reach in many cases, it also causes the feedback system to go into a frenzy by effectively disconnecting it for a tiny amount of time and then reconnecting it, so then the feedback system has to re-stabilize and get back on track way faster than it can. That is likely to amplify the effects of any crossover distortion glitch.
Serious distortion research requires a good spectrum analyzer, so you can separate out noise, but more importantly to see the distortion spectrum shape and how far up the significant harmonic products extend in frequency. It's also very important to look at I.M. distortion products, not just because they can cause "difference beat frequency" products in the audio frequency range, but because their "sum" products can cause slewing related distortions, especially in high feedback circuits.
In guitar land there's good distortion and bad, and many variations in between. I've learned a lot about this by trying to produce the best sounding guitar amp for the blues musician (me) who wants to be on the soft edge of distortion, but never so distorted to the point where it turns all fizzy to the point where you can barely tell what note are being played. It's actually been a very interesting journey to learn about all the different distortion mechanisms, and how they can work together to create a good sound, or a "fingernails on the blackboard" horrible scratchy and harsh sound. I think it's made me a better Hi-Fi audio engineer to learn about this.
Serious distortion research requires a good spectrum analyzer, so you can separate out noise, but more importantly to see the distortion spectrum shape and how far up the significant harmonic products extend in frequency. It's also very important to look at I.M. distortion products, not just because they can cause "difference beat frequency" products in the audio frequency range, but because their "sum" products can cause slewing related distortions, especially in high feedback circuits.
In guitar land there's good distortion and bad, and many variations in between. I've learned a lot about this by trying to produce the best sounding guitar amp for the blues musician (me) who wants to be on the soft edge of distortion, but never so distorted to the point where it turns all fizzy to the point where you can barely tell what note are being played. It's actually been a very interesting journey to learn about all the different distortion mechanisms, and how they can work together to create a good sound, or a "fingernails on the blackboard" horrible scratchy and harsh sound. I think it's made me a better Hi-Fi audio engineer to learn about this.
Both yours and MagicBox's descriptions are effects I believe I've heard in the past. So how do you differentiate these from poor transient performance? Or can you without bench testing?
Where is THD , there is IMD to and that is not meaningless.
Besides it is much easier to measure THD ,than IMD.
opa with less than 0.00005%
Hi
For my experience with opa, I prefers generaly opa with less than 0.00005%, with high slew rate. Sometime a high slew rate can compense a little more distortion.
They are specially better for the high, especialy with classical music (violin...) with detailled drivers (TD4001 / HD800). I have tested a lot of op amp sometime with A/B test.
I have tested BURSON, DEXA, MUSES, they can have a more natural sound, but they are dirty for the high.
And it's the same for the amp...
Hi
For my experience with opa, I prefers generaly opa with less than 0.00005%, with high slew rate. Sometime a high slew rate can compense a little more distortion.
They are specially better for the high, especialy with classical music (violin...) with detailled drivers (TD4001 / HD800). I have tested a lot of op amp sometime with A/B test.
I have tested BURSON, DEXA, MUSES, they can have a more natural sound, but they are dirty for the high.
And it's the same for the amp...
I said THD as a number. From IMD you could see interesting things from the graph, not from a number.🙂
A single number THD without any other information is pretty useless. Usually these default to the THD measurement at 1K and as such conveniently 'hides' any THD performance for the rest of the frequency range. A full THD vs Frequency graph is much more telling, but if it has to be a single number, I'd want it to be the THD20K spec, which, here on DIY is used as the benchmarking standard.
- Status
- Not open for further replies.