Compressed Sensing Options

[exoteric]

It seems some folks at Texas Instruments have made a 1-pixel digital camera that still manages to produce pretty impressive pictures by using compressed sensing.

It looks like compressed sensing has a lot of potential but how about the audio domain. Can compressed sensing be used to calculate higher definition from regularly sampled digital material, such as redbook CD material, or does the source material need to be sampled in a special "noisy" way?

If so, what software exists that one can use to do this?

If not, why not?

Lastly, on the somewhat related topic of sample rate conversion, what's the state of the art in sample rate conversion (primarily upsampling w/ interpolation) both real-time and ahead-of-time?

Thanks in advance.

[Paulhoff]

This sounds so much like the One Vacuum Tube that was to replace the CPU IC for home computers a few years ago.

Paul



This post has been edited by Paulhoff: Dec 11 2010, 15:55

[saratoga]

It looks like compressed sensing has a lot of potential but how about the audio domain.

Compressed sensing is used when you can't practically build an ADC needed for your requirements and are willing to spend however much time, effort and computational power to construct something equivalent to one. 96k/24bit ADCs cost about $3, so not much point.

Might be useful for things like spatially resolving a sound field for acoustic imaging though.

Can compressed sensing be used to calculate higher definition from regularly sampled digital material, such as redbook CD material, or does the source material need to be sampled in a special "noisy" way?

CDs are digital. You don't need an ADC to read one, you can just copy the 1s and 0s off of it.

Lastly, on the somewhat related topic of sample rate conversion, what's the state of the art in sample rate conversion (primarily upsampling w/ interpolation) both real-time and ahead-of-time?

Upsampling is easy, so I guess the state of the art is that it can be done really well

This post has been edited by saratoga: Dec 11 2010, 20:17

[greynol]

This was brought up a short while back but didn't get much attention:
http://www.hydrogenaudio.org/forums/index....showtopic=84222

EDIT: I see that saratoga was the lone respondent in the other thread, so I don't know if this post of mine will be all that helpful.

This post has been edited by greynol: Dec 11 2010, 21:22

[exoteric]

Upsampling is easy, so I guess the state of the art is that it can be done really well

May be easy, but then I wonder why almost noone can do it:

http://src.infinitewave.ca/

Some of these are real-time but I expect not all of them and so it puzzles me that most of them would be so drastically poor compared to the ideal.

[saratoga]

Upsampling is easy, so I guess the state of the art is that it can be done really well

May be easy, but then I wonder why almost noone can do it:

http://src.infinitewave.ca/

Unless I'm missing something nothing on that page mentions upsampling.

Some of these are real-time but I expect not all of them and so it puzzles me that most of them would be so drastically poor compared to the ideal.

Looking through that page, the vast majority of those provide excellent resampling. If you mean why don't the provide 150dB of alias rejection or something else absurdly useless, thats because theres a tradeoff between alias rejection and CPU time, and for 99% of applications alias rejection below -70dB or so is a waste of CPU time.