If you can hear something but not measure it, you are not measuring the right thing.
But back to distortion, there is a test on Klippel's website. Who are Klippel? They make measurement systems and software that is used by just about every loudspeaker driver manufacturer. https://www.klippel.de/company/about-us.html and https://www.klippel.de/company/references.html
And now the listening test - what is the minimum distortion you can hear?
https://www.klippel.de/listeningtest/
But back to distortion, there is a test on Klippel's website. Who are Klippel? They make measurement systems and software that is used by just about every loudspeaker driver manufacturer. https://www.klippel.de/company/about-us.html and https://www.klippel.de/company/references.html
And now the listening test - what is the minimum distortion you can hear?
https://www.klippel.de/listeningtest/
Hi edbarx,
Welcome to the cruel facts of life. There is no linear amplifying device yet, so therefore distortion is added to every signal. The good news is that between careful circuit design to linearize active devices' operating points, and feedback, we can often reduce distortion down into the noise floor.
Pretty amazing - isn't it?
Now if the distortion products are lower than the noise floor, they practically do not exist. Wouldn't you agree?
Welcome to the cruel facts of life. There is no linear amplifying device yet, so therefore distortion is added to every signal. The good news is that between careful circuit design to linearize active devices' operating points, and feedback, we can often reduce distortion down into the noise floor.
Pretty amazing - isn't it?
Now if the distortion products are lower than the noise floor, they practically do not exist. Wouldn't you agree?
Hi anatech, yes, as you wrote, distortion can be reduced using negative feedback. Negative feedback relies on 'comparing' two voltages keeping the difference very low. This makes the voltages much more dependent on the input signal. The secret to make the voltages at the input virtually equal is to have a very high voltage gain without feedback. This minimises the voltage error between the inputs.
I remember my sister buying me five books named "Elements of Electronics". The author is F.A. Wilson. Besides that, I have other books on the subject of electronics. One of them is by Sedra and Smith. The title is "Microelectronic Circuits" if my memory serves me right.
I remember my sister buying me five books named "Elements of Electronics". The author is F.A. Wilson. Besides that, I have other books on the subject of electronics. One of them is by Sedra and Smith. The title is "Microelectronic Circuits" if my memory serves me right.
Hi edbarx,
True, however there is no way to amplify anything without some distortion. So we accept this as a fact of life. We make our forward path as linear as reasonably possible (without getting really silly about it).
The trick is to not distort the input signal before it reaches the diff pair (or whatever). If you can do that, then you have something worthwhile to compare the output to. Feedback is merely a correction signal, not some evil entity. The signal on the inverting input is a scaled down version of the output (making certain your divider doesn't distort the signal!). If your difference amplifier is matched properly, it will subtract the two signals leaving you with a scaled down forward signal, and inverted distortion component scaled properly to pre-distort the signal in a complimentary way. That way the forward path delivers a very low distortion, low noise higher voltage (or current) input signal. That's what we want.
At the end of the day, if you apply a signal to the amplifier and end up with the desired amplified signal with distortion products too low to detect, isn't that what matters? Whether you use 200dB of feedback (not possible now) or 3 dB, it doesn't matter to fidelity. If you can achieve the same output impedance with low feedback - then GREAT! Any other arguments are pointless given the same noise and distortion levels, plus performance.
True, however there is no way to amplify anything without some distortion. So we accept this as a fact of life. We make our forward path as linear as reasonably possible (without getting really silly about it).
The trick is to not distort the input signal before it reaches the diff pair (or whatever). If you can do that, then you have something worthwhile to compare the output to. Feedback is merely a correction signal, not some evil entity. The signal on the inverting input is a scaled down version of the output (making certain your divider doesn't distort the signal!). If your difference amplifier is matched properly, it will subtract the two signals leaving you with a scaled down forward signal, and inverted distortion component scaled properly to pre-distort the signal in a complimentary way. That way the forward path delivers a very low distortion, low noise higher voltage (or current) input signal. That's what we want.
At the end of the day, if you apply a signal to the amplifier and end up with the desired amplified signal with distortion products too low to detect, isn't that what matters? Whether you use 200dB of feedback (not possible now) or 3 dB, it doesn't matter to fidelity. If you can achieve the same output impedance with low feedback - then GREAT! Any other arguments are pointless given the same noise and distortion levels, plus performance.
That's strictly speaking not a meaningful statement since the noise-floor is measured in V/√Hz and the signal in V. Given a narrow enough bandwidth (i.e. enough analysis time, stable enough oscillator, big enough FFT) you can always pick out any distortion products from the noise - this is the working principle of the lock-in amplifier (there's a technology that is indistiquishable from magic if ever there was one).
However if we agree to limit the analysis time to a few seconds you can talk about signal being below the noise floor.
Oh for crying out loud Mark! This isn't a professional paper to be peer reviewed. I am trying to explain this to everyone, even people not versed in electronics.
So, do you think I got the idea across, or would you like to debate silly things?
So, do you think I got the idea across, or would you like to debate silly things?
I agree with several posts above that if the distortion products are low enough then they don't matter - regardless of the relative phases/amplitudes of the harmonics. I won't claim to know exactly how far down they need to be. But as a specific example I usually listen at 70 dB, so if all distortion products are 80 dB down at -10 dB SPL then I doubt they matter. Even if I had superior hearing and can hear such low levels (which I am pretty sure I don't!) the background noise in my home will be much higher.
I am guessing the fancy metrics that take into account the amplitudes and phases of the different distortion products are primarily going to matter (at least if they accurately capture 'sound quality') when distortion is higher. Back to the OP, when I have simulated and measured spectra from crossover distortion (anecdotally - have not done a serious study) I find that the second harmonic tends to be roughly 90 degrees out of phase with the fundamental. This implies there is hysteresis, so the Gedlee metric doesn't apply anyway and you can save yourself the trouble of trying to figure out what it is and how to use it.
jason
I am guessing the fancy metrics that take into account the amplitudes and phases of the different distortion products are primarily going to matter (at least if they accurately capture 'sound quality') when distortion is higher. Back to the OP, when I have simulated and measured spectra from crossover distortion (anecdotally - have not done a serious study) I find that the second harmonic tends to be roughly 90 degrees out of phase with the fundamental. This implies there is hysteresis, so the Gedlee metric doesn't apply anyway and you can save yourself the trouble of trying to figure out what it is and how to use it.
jason
From a practical physics point of view, that is strictly speaking a VERY meaningful statement.
It's called order of magnitude.
Meaning that other variables become meaningless when something else is much much bigger.
In this case totally masked as well.
How meaningful a statement is, depends on the context, not if some kind of special measurements system can measure it somehow.
When we are interested in order of magnitude of meters, we also don't give the resolution down to 3*10^−12 m in 99.8% time of the cases. Only in those cases when it's necessary.
Which by definition implies you first have to figure out if it's even meaningful or not to go that far.
Audio has always been very weird this way, since people have a believe that extremely small resolutions are obligatory.
Distorsion at 0.1% should be enough for very high fidelity given that there s no speakers that can approach
even remotely such a linearity, so even a digital source will provide no benefit for this parameter, only for the lower
noise floor and eventually a larger bandwith.
even remotely such a linearity, so even a digital source will provide no benefit for this parameter, only for the lower
noise floor and eventually a larger bandwith.
Hi wahab,
It depends entirely on what the distortion is. Numbers without a spectrum really don't mean much.
Some speakers at low power levels can be lower distortion as well.
It depends entirely on what the distortion is. Numbers without a spectrum really don't mean much.
Some speakers at low power levels can be lower distortion as well.
For anyone who wold like to experience the audibility of different harmonics:
REW offers the possibility to add individual harmonic waves with selectable intensities to a fundamental sine wave (in sine generator tool).
I found it very revealing how much difference the harmonic number (multiplier) makes, also the frequency of fundamental and eventual masking of different harmonics.
REW offers the possibility to add individual harmonic waves with selectable intensities to a fundamental sine wave (in sine generator tool).
I found it very revealing how much difference the harmonic number (multiplier) makes, also the frequency of fundamental and eventual masking of different harmonics.
wahab - Music peaks in energy around 50-100Hz and is 20-40dB down by 5KHz. This is why high harmonics can be audible even in small amounts.
Ed
Ed
Just have to love this forum and how a basic question on crossover distortion can turn into a discussion on audibility of phase/envelopes and signals below the noise floor. I say this in a good way - I fond it this slightly unstructured approach to discussions interesting and very instructive.
In the section on distortion mechanism in Randy Slone's book on Power Amplifiers he show a test circuit and show some scope pictures together with FFT results. Not sure if they are simulations or actual measurements.
If I am to summarize my interpretation of this section of his book it is as follows.
a) Crossover distortion at low levels (1V RMS with no bias on his test circuit) is dominated by lower order harmonics and because (most) amps have increasing open-loop gain with lower frequencies the feedback mechanism will effectively suppress most of these making THD at low signals levels better than what immediate impression would make one think.
b) Crossover at higher output levels (4V RMS with no bias) tend to generate more of the high-order harmonics. Even with feedback there is typically less open-loop gain to help "combat" these high-order distortion products.
c) The end result is that typical THD percentage at lowest listening levels will be approximately 3 times that of full power (EDIT: assuming proper biasing)
d) The un-linear effect of the crossover distortion will affecting signals up to approximately to about 7 Volts swing (although I can not find proof of this claim)
e) EF stages is better than CF in terms of crossover distortion.
f) crossover distortion should not be confused with switching distortion. Although also appearing in the crossover region this is a different animal that must be mitigated elsewhere in the circuit.
Attached is the circuit he used for reference. He used 0 volt as bias (which I guess is class-C) to maximize the effects of distortion on and the scope traces and FFT plots.
In the section on distortion mechanism in Randy Slone's book on Power Amplifiers he show a test circuit and show some scope pictures together with FFT results. Not sure if they are simulations or actual measurements.
If I am to summarize my interpretation of this section of his book it is as follows.
a) Crossover distortion at low levels (1V RMS with no bias on his test circuit) is dominated by lower order harmonics and because (most) amps have increasing open-loop gain with lower frequencies the feedback mechanism will effectively suppress most of these making THD at low signals levels better than what immediate impression would make one think.
b) Crossover at higher output levels (4V RMS with no bias) tend to generate more of the high-order harmonics. Even with feedback there is typically less open-loop gain to help "combat" these high-order distortion products.
c) The end result is that typical THD percentage at lowest listening levels will be approximately 3 times that of full power (EDIT: assuming proper biasing)
d) The un-linear effect of the crossover distortion will affecting signals up to approximately to about 7 Volts swing (although I can not find proof of this claim)
e) EF stages is better than CF in terms of crossover distortion.
f) crossover distortion should not be confused with switching distortion. Although also appearing in the crossover region this is a different animal that must be mitigated elsewhere in the circuit.
Attached is the circuit he used for reference. He used 0 volt as bias (which I guess is class-C) to maximize the effects of distortion on and the scope traces and FFT plots.
Attachments
Last edited:
Agree on this point but distorsion, even crossover one, is still harmonic, intermodulation contain differences of the existing frequencies but it s already present in any musical signal, that s why i stated the 0.1% number for the sum of all distrorsions.
So far i made some tests and something like 0.3% can be heard assumindg that the fundamental is a sine signal at 1KHz and harmonics being odd.
For crossover THD i did tests at low level using a very basic low NFB amp, that is a differential + one transitor VAS + a pair of lazy darlingtons like the BDW93/BDW94 and on the verge of bias at say 1-2V output level crossover THD is clearly audible but that s with very low NFB.
Amplifiers have generaly very low distorsion with 50-100Hz signals, besides even such musical signals are never pure sines, there is always harmonics in any of this signal, FI a bass drum with a hard hitting sound contain a lot of high frequencies components that are not forcibly harmonic related to the fundamental.
Last edited:
Electronic music often contains clean pure tones at any frequency, including deep bass. On typical laptop speakers these can appear silent as there's often no bass-extension at all below 100Hz and the source has no harmonics...
You'd have to check. It's really 6 and some small fraction. (Edit, My bad.. Missed about 100 newer posts.)
Also, depending on the sampling rate, dithering could be more or less effective as the noise from it can be shaped so it stays above some minimum frequency.
Hi Mark,
Hi wahab,
In all my years of testing, solid state equipment worst case is at low power and low impedance (4R standard test). The distortion number always tends to improve with increasing power up to near clipping in a normal amplifier. Now, that test circuit isn't normal, but rather designed to showcase his point. Other issues will raise their ugly heads in a less well designed amplifier with increasing power. But do your job right and worst case performance is generally low power, low impedance. That's why the standard tests are done at 1 watt, 8 ohms (some amps can't drive 4 ohm loads well). Plus it puts all measurements on an equal playing field so you can compare them.
One thing for sure. Equipment that doesn't do well at 1 watt sure isn't going to get better at higher powers! A tube amplifier will perform less well as the power goes up. Nature of the beast, and low open loop gain means less negative feedback to correct the situation.
I am not saying high power tests are pointless. What they are is very good at determining if there are other issues with an amplifier. But they are also specific to a certain amplifier power output. Similarly, testing full power is more of a test of the power supply and AC mains than it is for an amplifier. That is unless protection kicks in, then it's a test of the protection circuit, not the amplifier circuit.
That boils down to "know what you are testing". It also means that if you are reading reports, understand what other factors may be coming into play.
Usually not. A pure sine is boring as heck, and musicians using electronic instruments almost always add harmonics, distortion and / or compression to get an interesting sound. Spent time in recording studios and repaired MI equipment under warranty. I have yet to see anyone use a pure sine wave except as an effect (maybe - never seen it done that I can remember). Even then I wouldn't trust it is a pure sine!
Hi wahab,
In all my years of testing, solid state equipment worst case is at low power and low impedance (4R standard test). The distortion number always tends to improve with increasing power up to near clipping in a normal amplifier. Now, that test circuit isn't normal, but rather designed to showcase his point. Other issues will raise their ugly heads in a less well designed amplifier with increasing power. But do your job right and worst case performance is generally low power, low impedance. That's why the standard tests are done at 1 watt, 8 ohms (some amps can't drive 4 ohm loads well). Plus it puts all measurements on an equal playing field so you can compare them.
One thing for sure. Equipment that doesn't do well at 1 watt sure isn't going to get better at higher powers! A tube amplifier will perform less well as the power goes up. Nature of the beast, and low open loop gain means less negative feedback to correct the situation.
I am not saying high power tests are pointless. What they are is very good at determining if there are other issues with an amplifier. But they are also specific to a certain amplifier power output. Similarly, testing full power is more of a test of the power supply and AC mains than it is for an amplifier. That is unless protection kicks in, then it's a test of the protection circuit, not the amplifier circuit.
That boils down to "know what you are testing". It also means that if you are reading reports, understand what other factors may be coming into play.