Digital predistortion for speaker correction

[rtarbell]

Just an idea, does anyone know if there is a commercial product out that does this?

Take an audio power amplifier and driver

At the front end of the audio amplifier, place an ADC, DSP chip, and then a DAC (and then the input to the amp).

Place a measuring wire at the output amp/driver connection, this goes to the input of the ADC.

Using the DSP, apply an FM chirp over the frequency range of ~10Hz to 35kHz (enough to cover the audio range), and measure the response at the driver while bypassing the audio amplifier(ideally, the response will be flat, but we know this doesn't happen in real life).

Using the DSP processor and based on the driver frequency response, make a filter that acts as the inverse to the non-flat response of the driver. This will predistort the audio signal before it enters the amp.

==> Begin playing normal music into front end, where the input is now the ADC. The music should be predistorted so that the frequency response is very flat, and any non-linearities of the driver are mitigated.

Any thoughts?

 

[jcx]

not trivial but some people are working on it:

http://www.klippel.de/pubs/The Po...er Models.pdf


at: http://www.klippel.de/pubs/default.asp )

 

[blu_line]

probably better to do it static (once, at startup of your system) than do it dynamically !

1st of all, your AD and DA have an conversion delay of 1ms and you can add some us or ms for DSP processing.


grtz

Simon

 

[phase_accurate]

Not like your proposal but better i.e. it goes two steps further in that it corrects room interactions as well and it does not only correct in the frequency domain :

http://www.tactaudio.co.uk/Products/correction_systems.html

This one also uses digital correction of the drivers, FIR crossover and it can be tuned to your room as well:

http://www.klein-hummel.de/produkte/o500c/bilder/o500c_front_hq.jpg

Regards

Charles

 

[Jocko Homo]

I could be wrong, but B&O does something along this line. A powered speaker system, filled with their ICEpower modules.

Jocko

 

[jcx]

frequency reponse and room equalization are mathematically trivial compared to nonlinear distortion correction - not that even linear correction is truely simple

there is a PC software active crossover/loudspeaker fr/room correction thread in louspeakers

http://www.diyaudio.com/forums/showthread.php?s=&threadid=63078&highlight=

in fact search for room correction should give you plenty of reading

comercially the BeoLab 5 active loudspeaker integrates "room correction" (for bass) and voice coil heating nonllinear compensation internally in dsp

http://www.sawonline.com/licensees_bo_technical.shtml

 

[Christer]

For static correction, there isanother way. Rip a CD into your PC, predistort the content in software and burn the predistorted copy of it onto a CDR/CDRW. Not very practical to do with your entire CD collection perhaps, but it is a chep way to experiment with static corrections, since no extra hardware is required. You may have to write the software yourself though.

 

[Werner]

Static pre-compensation in the digital domain was done in the late 80s, early 90s by Essex University, with at least one commercial spin-off (Celestion?), and also by B&W (no commercial application).

Then in the late 90s Perpetual (name? anyway, Mark Schifter's company) promised something similar, but AFAIK never delivered.

 

[noah katz]

I like the Rhythmik Audio Directservo approach, which uses a sensing coil on the driver to correct driver response, including nonlinear mechanical and thermal compression.

http://www.rythmikaudio.com/servo_product.htm

I just wish they offered it with a driver larger than 12".

 

[dwk123]

As previous posts have illustrated, some aspects of this have been done. Linear distortion (ie frequency response) is 'pretty easy', except for the part about figuring out which part it is that you're supposed to correct

The really interesting problem (IMHO) is non-linear correction which is far from trivial, and I'm only vaguely aware of people attacking this.

I believe that non-linear correction is 'relatively straightforward' for electronics/amplifiers operating in their linear region as they can be modeled pretty well by a power-series expansion. This makes it feasible to produce a complementary function that pre-distorts the signal to create an undistorted output. I'm sure someone has done this, but I don't have any pointers

Speakers are significantly harder for several reasons, starting with the fact that many of the main distortion components are due to position, but the input signal is governing acceleration. ie the distortion is related to the 2nd integral of the signal. Since even numerical inaccuracies from rounding will eventually make the computation of position from the signal inaccurate, I suspect the only real way to tackle this is either :
- via position sensors (ie sense position directly)
- via servo/accelerometer (ie sense a variable directly related to the input).

Of course, *any* latency in the measurement will limit the bandwidth over which correction can be applied, and there is always latency.

So, it's a really interesting problem, but it's certainly WAY easier to simply get better speakers in the first place

 

[noah katz]

"Linear distortion (ie frequency response) is 'pretty easy',"

Except that it changes because voice coil temp changes the T-S parameters.

Tom Danley (inventor of the Contrabass and Tapped Hoen speakers) sauis recently on AVS forum that these changes become significant at about 10% of a drivers rated power output.

 

[noah katz]

chrisybee,

Huh? For the life of me I can't figure out what your web site is or what it offers?

 

[dmfraser]

Pre Distortion

The problem with this approach is that speakers change as they age. So unless a system can be calibrated every now and again, the pre-distortion will slowly become worse than useless.

The B&O speakers are set up to do exactly this. The catch is that they are hideously expensive. And they can only calibrate as good as the microphone.

This sort of trickery comes out every now and again and has not caught on yet.

I still think you are better off buying or building good speakers, with either minimal coloration or colorations you like and placing them properly in the room.

Then listen to the music and not to the speakers. You may think this is the same thing but I assure yiu it is not. If you listen to music, even fairly basic speakers work OK. If you listen to the speakers, you will never be satisfied.

 

[DSP_Geek]

"Linear distortion (ie frequency response) is 'pretty easy',"

Except that it changes because voice coil temp changes the T-S parameters.

Tom Danley (inventor of the Contrabass and Tapped Hoen speakers) sauis recently on AVS forum that these changes become significant at about 10% of a drivers rated power output.

Or about 10 dB down, not hard since the peak-to-average ratio of music is easily 5:1 for decently recorded pop, more like 10:1 or 20:1 for classical. In other words, that's not a big deal for home listening.

In any event, the dominant problem for bass EQ is the room, and that won't change if the voice coil heats up. Mid and high EQ is mostly a matter of controlling cone and rear chamber modes, once again unaffected by voice coil resistance. Linear correction is thus not a bad idea.


Francois.

 

[ghemink]

Not like your proposal but better i.e. it goes two steps further in that it corrects room interactions as well and it does not only correct in the frequency domain :

http://www.tactaudio.co.uk/Products/correction_systems.html

This one also uses digital correction of the drivers, FIR crossover and it can be tuned to your room as well:

http://www.klein-hummel.de/produkte/o500c/bilder/o500c_front_hq.jpg

Regards

Charles

http://www.deqx.com has a nice toy too.

Best regards

Gertjan