Interesting new DSP process.

This article is over a year old but I read it today and found it interesting.

I've been speaking with my cousin about audio DSP and he saw this and sent the link to me.

http://www.wired.com/maga.../02/ff_algorithm/all/1

I figured I would post it here rather than the coffee house because we've had some lively discussions about dsp here lately and I thought it might be interesting to see comments from as many folks as possible.

The article is very straight forward...  the gist of it is that mathematical reconstruction techniques may be getting so effective that there may be less need to manage the logistics of high resolution capturing (a.k.a. recording).

The mathematical method described seems to have so much potential that it seems to good to be true and I keep wondering if the article is some sort of practical joke. The article, however, seems pretty easy to understand.

Anyways, I'm hoping a few people get to see and read it and maybe share some thoughts.


best regards,
mike

I think it only has limited use in audio.

It's based on the idea that the original is sparse, or to put another way simple. "Noise" and any "unusual" events would be eliminated by the processing. What would happen with audio is the waveforms would be "smoothed" out, but not in the same way a low pass filter does.

For instance, this picture shows square waves constructed out of increasing numbers of odd harmonics:



I would expect it to be replaced with a "perfect" square wave (meaning an infinite number of odd harmonics) using compressed sensing. Which you would then low pass filter to 20kHz or whatever. But what if your original signal wasn't a perfect square wave, but one that only had a few harmonics, and that they didn't stop at 20kHz (or whatever the source's bandwidth limit was)?

You would also likely lose all noise, which might be a good thing in some cases, but what if it's from sibilance? Do you want 100% of the breath noise and consonants taken out of a vocal recording? 

It might be useful for removing huge amounts of noise, enharmonic distortion, and restoring higher harmonics from really poor recordings, like ancient disc pressings with gobs of surface noise and really limited bandwidth.

But I wouldn't expect it to be useful for high quality reproduction.

But of course, as always, I could be wrong.

[EDIT: Doing a little reading, it looks like it could actually be quite useful for restoration type work, even speech, and other specific purposes. But it's not clear that it can challenge traditional audio compression technology (MP3's, etc.) for general purpose audio.]