It's perfectly correct. But it would be more general if the "Nonlinear Model" block was shown as many parallel blocks each of a different order of nonlinearity. And in its most general sense, each of those blocks would have a transfer function in frequency. It this most general sense, the block "Linear Model" would just be the leg for the n=1 term, its slope being the system gain.
"Defects" here would be all the left over "junk" that doesn't get captured in the above. Basically this would be the stuff that was not correlated to anything else.
"Defects" here would be all the left over "junk" that doesn't get captured in the above. Basically this would be the stuff that was not correlated to anything else.
You weren't told that, at least not by me.
Each leg will pass different orders, but each odd order passes some small amount of each lower odd order and each even order passes a small amount of each lower even order.
So, for example, a 5th order transfer characteristic as a separate leg would yield a first order (linear) much smaller contribution and a smaller third order contribution and a fifth order contribution. Using say input = 100 Hz, and the value of the fifth order is -.01, the output would be (guessing at some numbers) .998 @ 100 Hz (a drop of a fraction of a dB,) .006 @ 300 Hz and -.01 @ 500 Hz. (plus and minus just being phase inversion.)
Take a sine(w*t) and raise it to the 5th power and you will see exactly how this comes about.
Have you looked at http://gedlee.azurewebsites.net/Papers/Distortion_AES_I.pdf
Each leg will pass different orders, but each odd order passes some small amount of each lower odd order and each even order passes a small amount of each lower even order.
So, for example, a 5th order transfer characteristic as a separate leg would yield a first order (linear) much smaller contribution and a smaller third order contribution and a fifth order contribution. Using say input = 100 Hz, and the value of the fifth order is -.01, the output would be (guessing at some numbers) .998 @ 100 Hz (a drop of a fraction of a dB,) .006 @ 300 Hz and -.01 @ 500 Hz. (plus and minus just being phase inversion.)
Take a sine(w*t) and raise it to the 5th power and you will see exactly how this comes about.
Have you looked at http://gedlee.azurewebsites.net/Papers/Distortion_AES_I.pdf
Last edited:
this block diagram confuses me i took it literally
It's not literal. It's an lumped equivalent model. Lumped models are always approximations, although they may give very accurate results. They may even give extremely accurate results in some cases and for some purposes. But, they are models, not the exactly reality.
This model is says in each of the boxes for distortion, you could put a mathematical equation (or maybe a lookup table or something) that describes the distortion added by processes in each lumped box (just divided up into three categories for our modeling purposes). The diagram also shows that you can add up the results of the different kinds of distortion to get the final distorted output. In other words, the diagram is gives instructions for doing calculations of predicted distortion based some some models or some empirical measurements.
The real physical speaker is different, but it's complicated. Having a model that breaks things down into categories and math actually makes it easy to analyze in some ways.
Last edited:
It could in some cases, but that would be a more complex model than we are using at the moment.
You might want to go back and look at my last post on the last page: http://www.diyaudio.com/forums/mult...stortion-speaker-drivers-111.html#post5261627
There would need to be line or arrow in the model going from the linear distortion box to the nonlinear distortion box in order to represent what you suggested in your post quoted above.
Last edited:
Distortion test were carried in a research facility.
Best drivers by far for mid/woofers are made by: Harbeth, Dynaudio (in the more expensive models)
For tweeters, the beryllium versions from scan-speak, usher, seas and many others are all in top league.
For woofers, it is a whole lot of problems, the room itself will generate distortion and many manufacturers make good woofers, you need many, big, and $$$ to have low distortion figures.
Integration with the box and filters are more important imo than getting the best drivers which are relatively better by 5 to 10 db of reduction in distortion.
Best drivers by far for mid/woofers are made by: Harbeth, Dynaudio (in the more expensive models)
For tweeters, the beryllium versions from scan-speak, usher, seas and many others are all in top league.
For woofers, it is a whole lot of problems, the room itself will generate distortion and many manufacturers make good woofers, you need many, big, and $$$ to have low distortion figures.
Integration with the box and filters are more important imo than getting the best drivers which are relatively better by 5 to 10 db of reduction in distortion.
Last edited:
Almost !
Linear distortion affects the fundamental - non linear distortion generates harmonics.
Linear distortion does also affect the harmonics that belong to the original Input signal.
But the harmonic (i.e. nonlinear) distortion that is caused by the speaker itself can indeed be influenced by its linear distortion. Just imagine a resonant peak around 1.5 kHz of about 10 dB. The example driver has a given third order THD (k3) of one % at low frequencies. If we send 500 Hz into this driver the k3 will now be in the order of around 3%.
Regards
Charles
I'm not trying to be funny, isn't that obvious, the speaker doesn't know they are harmonics, or am I missing something?
IMHO there is no place for integration here, these guys are dissecting the distortion parameter of the loudspeaker driver.
This is a lab discussion.