For the answer look up "Applying Physics Makes Auditory Sense" by Heerens & de Ru. Their theory of hearing (the ear physically distorts the incoming signal heavily) explains a lot of strange hearing phenomena that aren't fully explainable with previous theories, and everybody can recreate their tests as they published the details and even a piece of software to create the test signals they used.
Always delightful to read the musings of physicists about fields they know nothing what so ever about. Somewhat less delightful is the pungent egotism of the author in his self-published essay. Pity, using google search, there's no specialist in hearing perception that thinks there's any value to Heerens physics speculations.
Highly inappropriate of you to re-write my quotation. What I wrote was that I didn't understand it: "But I can't say as I really understand...".
B.
Last edited:
From what I read - which was very interesting - their results are not conclusive. Too much work by other researchers that contradict their sketchy theory has been done to simply discard it based on some preliminary studies. I am not saying that it is wrong, I think that this theory and the prevailing ones might be blended together in a plausible sense. But there are multitudes of cochlear models that show the traveling waves on the cochlea despite the hand-waving arguments given against it. The cochlea structure is a coupled fluid structural system, not a simple fluid system with a very high sound velocity. This model has been shown to be highly accurate and one cannot simply discard it.
I am not a cochlear expert, although I do know the prevailing theories, however, the fact that this paper was not accepted for publication, and I would suggest correctly so as it is incomplete, suggests that these theories should be taken as a hypothesis of a new paradigm that remain unproven.
Have the authors attempted to further their position as this paper is already 8 years old.
a worrying turn of events
Thanks for posting this.... It "rings true"(!) with my experience that the more accurate the time domain performance as shown in the CSD plots the more life like and natural the sound is.
I am particularly sensitive to muddy mid-range and low mid-range which make vocals sound artificial and distorted.
Its similar to the overblown bass the bass ball cap brigade pump out of their ported car audio sub-woofers at high SPL.... "more is better, louder is best man... "
Many designers in home and & pro audio loudspeaker industry are now locked in the same mind set...."some is good, more is better" so they use Transmission lines (the worst offender is PMC) or ports / passive radiators to add more quantity at the expense of accuracy.
The bass / low mid-range performance of these designs is a train wreck but the "emperors new clothes" syndrome backed by multi £million advertising wins the day.
A much more worrying marketing campaign PMC have used for over a decade is to buy the endorsement of universities / educational establishments audio & acoustic department.... Over the last 5 years PMC "donated" over £2.5 million (RRP) worth of loudspeakers to various universities, schools and private institutions to ensure that all the students studying music / recording / production and live sound get hooked on the PMC sound....
The next gen of musicians and producers are being deceived big time!
Thanks for posting this.... It "rings true"(!) with my experience that the more accurate the time domain performance as shown in the CSD plots the more life like and natural the sound is.
I am particularly sensitive to muddy mid-range and low mid-range which make vocals sound artificial and distorted.
Its similar to the overblown bass the bass ball cap brigade pump out of their ported car audio sub-woofers at high SPL.... "more is better, louder is best man... "
Many designers in home and & pro audio loudspeaker industry are now locked in the same mind set...."some is good, more is better" so they use Transmission lines (the worst offender is PMC) or ports / passive radiators to add more quantity at the expense of accuracy.
The bass / low mid-range performance of these designs is a train wreck but the "emperors new clothes" syndrome backed by multi £million advertising wins the day.
A much more worrying marketing campaign PMC have used for over a decade is to buy the endorsement of universities / educational establishments audio & acoustic department.... Over the last 5 years PMC "donated" over £2.5 million (RRP) worth of loudspeakers to various universities, schools and private institutions to ensure that all the students studying music / recording / production and live sound get hooked on the PMC sound....
The next gen of musicians and producers are being deceived big time!
My thinking as well. They didn't make it through the rigorous peer review process and we may assume this was for good reasons. Still I find their approach worthwile and it helped me to understand a lot of things much better. For example, why we can't judge a driver by listening to band limited multitone signal or noise. Even with a "distortionless" driver (say, headphones at pretty low plaback volume) it sounds heavily distorted, polluted by IMD products generated by my own hearing (which normally get masked by other signals at those frequencies with normal music, well, except for music with very sparse spectra like recorder flute quartets and such).
You know that's not true, right ?
If you have a resonance peak - whether mechanical or electrical, if you apply a complementary notch, where centre frequency, Q and amplitude match - and both resonance peak and notch are minimum phase, which is typically the case, the ringing is also eliminated in the time domain ?
The explanation is quite simple - the resonance rings in the time domain as it decays at the resonant frequency, a notch with the same centre frequency, Q and amplitude also rings the same amount but in reverse phase. So the ringing of the resonance and the ringing of the notch exactly cancel out so there is no ringing left over, and a CSD will show this.
That assumes the centre frequency, Q and amplitude are correct of course. If the centre frequencies don't match it will initially suppress the ringing but the two will drift out of phase causing a "tail" to appear after the initial decay, (which can be seen on a CSD if you look carefully) likewise if the Q is wrong the decay rate will be wrong which will cause an in phase or out of phase tail depending on whether the Q is too high or too low.
The way to think about it is that you're tuning the amplitude, ringing frequency and decay rate of the notch to precisely match the resonance that you're trying to eliminate.
Of course there are reasons that EQ might not be able to fix a resonance completely - one is if the driver is not minimum phase in the region of the resonance. Most conventional drivers are minimum phase but for example a dual cone driver is not minimum phase through the mechanical crossover region as it is really a two (or more) way system with a shared voicecoil.
Another reason EQ might not be usable is if the resonance manifests differently or not at all on a different axis - this would be the case if the resonance was really a diffraction artefact for example. Although a peak or dip caused by diffraction is minimum phase at a given point in space and can be corrected as such, the correction may make the response worse on another axis.
But for a mechanical resonance that is minimum phase and not significantly different on different axes you can most definitely correct it with EQ and not just flatten the frequency response but also eliminate the time domain ringing of that resonance.
Last edited:
Applying cut EQ (a frequency domain tool) to a time domain problem is only going to help you to not aggravate the problem. Now you have limited the playback systems ability to reproduce at the notch.
This may indeed sound somewhat better than letting your time domain issue run unabated but it’s not the fix.
To me this is analogous to prescribing opium to help a cancer patient. If only we could cure the cancer.
Barry.
This may indeed sound somewhat better than letting your time domain issue run unabated but it’s not the fix.
To me this is analogous to prescribing opium to help a cancer patient. If only we could cure the cancer.
Barry.
IF one could "pre-correct" a drivers CSD ringing issues by applying a "mirror image" correction curve this would prevent the driver from producing the distortion.... BUT this would also eliminate the music / signal at those frequencies.... Baby and bathwater!
Also there is currently no perfect Eq.... Fab Filter / Blue Cat / Waves etc in the Pro world or all the usual audiophile Trinnov / DEQX / Mini DSP hardware or various software packages still suffer from pre and or post ringing, rounding errors, group delay / phase distortion e
All the top mastering / production engineers disagree over which is the best Eq... It all comes down to the least damaging option subject to personal taste, studio equipment and room acoustics...
There is
All drivers "beam" therefore no driver can be Eq'd !
I agree ... If one corrects an on axis peak or trough in the frequency response it usually makes the response worse at ost / all other angles...
As all drivers "beam" ie fail miserably to generate a spherical sound wave (expanding bubble is a more accurate analogy) in an even 360 degree pattern, this a serious isue.
The vast majority of drivers / loudspeakers dont even attempt to do this, they aim to cover as little as possible of the bubble via " controlled directivity "
Ie aiming a small "wedge" of sound at a narrow sweet spot.
Using horns / wave guides / DSP beam forming, arrays / panel or angled divers etc. All these techniques fail to address the fundamental issue that all loudspeakers fail to reproduce the natural "sound bubble"....
In surface area terms, the typical "wedge" of sound is less than 5% of the bubble.
I believe the solution is to develop a technology which generates the whole bubble.
I agree ... If one corrects an on axis peak or trough in the frequency response it usually makes the response worse at ost / all other angles...
As all drivers "beam" ie fail miserably to generate a spherical sound wave (expanding bubble is a more accurate analogy) in an even 360 degree pattern, this a serious isue.
The vast majority of drivers / loudspeakers dont even attempt to do this, they aim to cover as little as possible of the bubble via " controlled directivity "
Ie aiming a small "wedge" of sound at a narrow sweet spot.
Using horns / wave guides / DSP beam forming, arrays / panel or angled divers etc. All these techniques fail to address the fundamental issue that all loudspeakers fail to reproduce the natural "sound bubble"....
In surface area terms, the typical "wedge" of sound is less than 5% of the bubble.
I believe the solution is to develop a technology which generates the whole bubble.
You quoted me but fail to see a correlation to my post. Any(!!) linear distortion problem can be perfectly un-EQ'd with the proper inversion. With FIR-filters you can do that down to very minute detail (not that this would really be required in practice.... but as long as your problem is 1-dimensional it really works well).
My point was that this correction can't address errors it is not aware of. If your driver has a +15dB peak at 3kHz with Q=10 it can be undone with its inverse, a -15dB 3kHz Q=10 dip. But any distortion still sees the full peak, for example the H3 component at 1kHz. If we are feeding only single sines to a speaker, we could also put a notch at 1kHz to stop it from generating distortion at 3kHz... but it doesn't help in practice as the complex intermodulation distortion products which are at quasi-random frequencies dominate the driver distortion big time (that is why single sine distortion plots are pretty useless) so there is no way to preprocess the input to avoid exiting resonances with distortion products.
With enough DSP power and the correct algorithms digital filters are flawless. If you cut corners in the implementation (like failing to use at least 64-bit floating point arithmetics, or using non-optimized unstable/oscillating IIR filters, etc), they're not. I have actually tested that with 64-bit floats even a very large FIR filter block applied to a 24-bit stream followed by it's inverse renders a bit perfect output way below 24 bit depth precisions (that is, the values before rounding are spot on the integer input with negegible errors). Given that data is converted to the frequency domain to speed up the convolution and then converted back to time domain (samples), this is even more remarkable.
Last edited:
Time and frequency domains are uniquely linked to each other, so you can perfectly treat a time domain problem in the frequency domain and vice versa. Actually, you can transform/switch between domains at will with no ill effects. That's how convolvers actually work: take a block of samples, transfer it to frequency domain (magnitude and phase at the FFT frequencies), process it, then transfer back to time domain.
I agree. Pre-correcting sounds to me like an adolescent power fantasy.... if I had a long enough lever, I could move the world. But joking aside, if you knew the motor errors, you would start fixing at the source, as is the routine procedure.
And the real answer is: motional feedback. How can you have a good system without feedback?*
B.
*only sealed boxes inherently provide some degenerative feedback
Last edited:
Only as long as your sensor has neglegible distortion and your phase margin is enough to have practical feedback at the problem frequency.
Klippel has shown that is is possible to nonlinearily pre-distort the input signal to a driver to correct it's (motor) distortion once you monitor driver current and voltage and have a proper adaptive (time-variant) nonlinear model of the driver. By this, distortion reduction to at least the same order of precisison can be realised without MFB.
Klippel has shown that is is possible to nonlinearily pre-distort the input signal to a driver to correct it's (motor) distortion once you monitor driver current and voltage and have a proper adaptive (time-variant) nonlinear model of the driver. By this, distortion reduction to at least the same order of precisison can be realised without MFB.
Any driver that doesn't operate on strict current drive has degenerative feedback. The lower the effective Qes the more local velocity feedback you have. But it only helps as long as the microphone voltage (back-EMF) is sufficiently linear, of course. Once BL starts to drop, the VC as a sensor measures too little voltage which causes the feedback to overcorrect, and the more feedback the more so...