Speaker compression linearization

[Leo11]

Hello everyone,
I open this thread to ask if it is possible to develop an additional correction function in Audiolense, combined with its convolver.
Measurements of audio equipment are now fashionable and many people are specializing in this, offering ever more comprehensive measurements.
Among these is Erin, who in his Audio Corner publishes many measures of speakers, including those of compression. I quote his site:

The below graphic indicates just how much SPL is lost (compression) or gained (enhancement; usually due to distortion) when the speaker is played at higher output volumes instantly via a 2.7 second logarithmic sine sweep referenced to 76dB at 1 meter. The signals are played consecutively without any additional stimulus applied. Then normalized against the 76dB result.
The tests are conducted in this fashion:
76dB at 1 meter (baseline; black)
86dB at 1 meter (red)
96dB at 1 meter (blue)
102dB at 1 meter (purple)
The purpose of this test is to illustrate how much (if at all) the output changes as a speaker’s components temperature increases (i.e., voice coils, crossover components) instantaneously.

Basically, the frequency response of the loudspeaker changes as the sound pressure emitted varies.
In Erin's measurements you can see speakers that change even 2 or 3 db.

Now, since Audiolense (and any other room correction software) creates correction filters to the specific sound pressure at which the measurements were made, listening at different volumes results in a suboptimal correction, even audibly.

So my idea is to compensate for this non-linearity as well.
In practice this means measuring sound pressure at several levels, then creating a filter that compensates according to the digitally regulated volume in the convolver.

Would it be feasible?

 

[jjazdk]

It is not feasible, as the compression effect is highly non-linear.
Also you listen to music, which in nature is dynamic (crest factor), and the heating of the voice coil is not comparable to a constant voltage log sweep.

If anything a correction would make it worse, as its estimate of the effect will often be wrong (unless it is a more advanced system that analyze the incoming signal and make the needed correction in real time, based on a complex model of each loudspeaker driver).

 

[Omid]

Hello everyone,
I open this thread to ask if it is possible to develop an additional correction function in Audiolense, combined with its convolver.
Measurements of audio equipment are now fashionable and many people are specializing in this, offering ever more comprehensive measurements.
Among these is Erin, who in his Audio Corner publishes many measures of speakers, including those of compression. I quote his site:

The below graphic indicates just how much SPL is lost (compression) or gained (enhancement; usually due to distortion) when the speaker is played at higher output volumes instantly via a 2.7 second logarithmic sine sweep referenced to 76dB at 1 meter. The signals are played consecutively without any additional stimulus applied. Then normalized against the 76dB result.
The tests are conducted in this fashion:
76dB at 1 meter (baseline; black)
86dB at 1 meter (red)
96dB at 1 meter (blue)
102dB at 1 meter (purple)
The purpose of this test is to illustrate how much (if at all) the output changes as a speaker’s components temperature increases (i.e., voice coils, crossover components) instantaneously.

Basically, the frequency response of the loudspeaker changes as the sound pressure emitted varies.
In Erin's measurements you can see speakers that change even 2 or 3 db.

Now, since Audiolense (and any other room correction software) creates correction filters to the specific sound pressure at which the measurements were made, listening at different volumes results in a suboptimal correction, even audibly.

So my idea is to compensate for this non-linearity as well.
In practice this means measuring sound pressure at several levels, then creating a filter that compensates according to the digitally regulated volume in the convolver.

Would it be feasible?

You are right that speakers have non linear distortion (related to heating during a constant voltage log sweep, as well as many other factors experienced during normal playback). I suspect these non linear changes are usually not audible as most tracks will span many degrees of music loudness, so every track should sound terrible.

I did worry about the same though, so I ran sweeps in REW, with AL filters in place, at several volume levels 90dB, 85 dB, .., 60 dB to see if the filter resulted in an inappropriate frequency correction or step response. It did not. All the curves were perfectly parallel to each other (freq and step resp).

That's a good thing too, because I don't think you could change the convolution process to vary the shape of the filter depending on the input signal amplitude (but others with better understanding of the math can correct me if I'm, wrong).