It was too long ago I did the tests... but agree the odd would be the one to key off of.
But with a good margin for error... like X10 reduction... I have assumed as my goal in designs or products I would purchase to be .001% thd to be assured of less than detectable for the entire system
It can be done today.... only with newest opamps and/or designs. But that still leaves the speakers and power amp as the weak links in the playback portion of the chain. Acoustic recordings would start with the mic preamp.... same problem... as speaker... microphone is limiting factor. It would be good to have an entire recording to playback chain measured... taking signal from mic to mixer to recorder thru mastering thru cd thru playback... what % distortion does the entire system of components add up to. Most likely would be above 0.1% is my guess. Newer almost all digital might offer best way to get entire system sounding more 'real'. Mic to ADC and at home DAC to digital amp. Leaves just mic and speaker as the limiting factor. They need to do better then.
--Richard
But with a good margin for error... like X10 reduction... I have assumed as my goal in designs or products I would purchase to be .001% thd to be assured of less than detectable for the entire system
It can be done today.... only with newest opamps and/or designs. But that still leaves the speakers and power amp as the weak links in the playback portion of the chain. Acoustic recordings would start with the mic preamp.... same problem... as speaker... microphone is limiting factor. It would be good to have an entire recording to playback chain measured... taking signal from mic to mixer to recorder thru mastering thru cd thru playback... what % distortion does the entire system of components add up to. Most likely would be above 0.1% is my guess. Newer almost all digital might offer best way to get entire system sounding more 'real'. Mic to ADC and at home DAC to digital amp. Leaves just mic and speaker as the limiting factor. They need to do better then.
--Richard
Last edited:
Dr. Lee has a degree in experimental psychology, so our procedures were well controlled.
Nobody is interested because those who have studied it find no advantage to better understanding it. As far as loudspeakers go, just stop worrying about nonlinear distortion. There is so much linear distortion that the nonlinear stuff just doesn't matter.
Academics, like Lidia, don't do "academic" studies. Someone needs to fund it and people don't fund projects whose results no one cares about.
Earl is right mics and speakers have low order distortions in general that vanish with level. Presence peaks, radiation patterns,etc. (so called linear distortions) have far more to do with what you are presented than any non-linear distortion mechanisms. Mic pre-amps with vanishing distortion of any kind are trivial to design.
Earl you are showing your age, there are several software tools that have complex math totally integrated in a way you don't even have to think about it. I don't know anyone in the scientific community that uses Fortran anymore except for a very narrow class of problems. I don't know of any circuit simulators or CAD programs written in it.
Last edited:
Seriously? Come on Mark, nothing of the sort happened. Unless you are of the view that those who couldn't hear a difference were the only ones not deluding themselves.
It is my understanding that Intel keeps FORTRAN alive because so much of their internal modeling software is in FORTRAN. Other compilers may have complex ability today, but FORTRAN has had it since its inception (or shortly thereafter.) This means many more complex number subroutines of proven accuracy are available in FORTRAN than just about any other language. Circuit simulations and CAD programs have never pushed the limits of complex number capability.
I believe that all FEA software is still in FORTRAN, or vast parts of it. Clearly no one today uses only one language for a complex app. I use at least two always. I never try and interface FORTRAN to the user. This is always done in VB.Net
That depends on what you expect. All the answers would be extremely expensive. An initial study several thousand dollars to 10's of thousands to start.
Low order and vanishing distortion at low levels? Distortion is quite high in many loudspeaker systems... multi-way. And at realistic playback level, many have audible issues... beyond freq response.
BTW - measuring up close to the driver with 1W is not realistic playback level ---- realistic levels further back in a large room even with high effec speakers needs 10-20W from my tests.
THx-RNMarsh
Last edited:
Maybe I'm misunderstanding you, yes massive FEA and other huge problems that benefit from parallelism and speed but I don't know of any complex functions not available to Mathematica, Matlab or Python (for instance). Could you name one?
Much of the algorithm development in the scientific community is done in non-compiled languages because the scientists time is more valuable than the C and Fortran coders.
and from Intel itself
For the fundamental intrinsic functions sure, many languages have complex functions, but in physics there are way more complex functions than any language has intrinsically. For example, say I wanted to know the complex eigenvalues for the Matheau Function. I can easily find existing subroutines that do this extremely obscure calculation in FORTRAN (I have these subroutines myself,) but I doubt that they exist in the public domain in Python. It's all about legacy.
Yes I realize now you are simply talking about existing canned software, the complex number aspect has nothing to do with that. I realize you tried the most hopelessly obscure thing you could think of.
Scilab is open source for Python http://www.fis.unipr.it/~coisson/Mathieu.pdf
Last edited:
Scipy/Numpy have an enormous number of those wonderful obscure mathematical functions. As mentioned before, I'm betting they're mostly compiled c modules. Sometimes less easy to get at, but it's open source and widely, widely, widely adopted at the academic level, where naive, idealistic people still exist and share things that they spent a lot of time developing. 
And where massive computational horsepower isn't necessarily always available, so people get rather crazy about code optimization (plus there are speed competitions out there).
And there's nothing preventing one from bolting on some fortran code into the middle of a python script as needed.
And where massive computational horsepower isn't necessarily always available, so people get rather crazy about code optimization (plus there are speed competitions out there).And there's nothing preventing one from bolting on some fortran code into the middle of a python script as needed.
Scipy and Numpy are transparent you just load the modules and use the functions. Once I got used used to the rich set of capabilities in the function arguments I never went back, pages of code turns into paragraphs and you can just understand it by reading it. The Analog Discovery, for instance, uses a separately compiled C++ .dll and you have to keep track of C variable typing in the arguments and return values in some cases but the functions just appear when you load the module.
I wasn't trying to give anyone a hard time, just pointing out how far the open source folks have come. Scilab is just another example of a huge collaborative open source effort all of which I support.
The use of the term "public domain" here is problematic. Any algorithm described 100yr. ago is by definition in the public domain, patent protection has certainly expired and copyright does not cover ideas disclosed 100yr. ago and not the author's original work.
Using "THD" (which has a well defined meaning) when you actually mean 'nonlinear distortion measured via harmonic distortion' is not helpful to meaningful debate.Markw4 said:
Most of the evidence is that anything which can be genuinely heard can be measured, although that does not necessarily mean that it is usually measured. Meaningful tests have to be on the basis 'does this or this sound the same or different?' because as soon as you ask 'does this sound better than this?' you are getting into preferences, which are notoriously unhelpful if the aim is high-fidelity sound reproduction. In either case the test must be unsighted, as otherwise all sorts of bias come into effect - especially for those who fondly believe themselves to be free of bias.
Do you actually mean "meaningless", or do you mean 'less meaning than some believe'? The fact that a carefully constructed signal with 20% THD can sound less distorted than another carefully constructed signal with much smaller THD does not prove that THD is "meaningless". It just shows that THD has be be treated with caution, especially when carefully constructed signals are being used. For most generally useful measures of almost anything it is possible to construct an artificial scenario which does not fit the measure and so gives misleading results.gedlee said:
I did not realise you were talking about speakers.AllenB said:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.