IMHO, this is pretty funky stuff, but also quite tough math...
It needn't be so difficult. There's a lot of jargon and domain-specific material in [1] and [2], but the principle is pretty straightforward.
There are two underlying ideas: (1) if a sequence is compressible at all, it's the consequence of some redundancy in the sequence, but (2) even a provably-optimal compression scheme must necessarily leave some residual structure in the output (compressed) sequence.
What's required is knowing where to look for the residual structure. In general, such residual structure will show up as patterns in the entropy of the compressed output sequence measured over multiple intervals. In other words, you slide through the sequence looking at the estimated entropy over several windows of different sizes. A particular kind of data will exhibit a characteristic distribution of values over those several windows.
To be honest, I've never actually seen this outside the
pencil+paper/matlab world, and chances are high you want something
simpler than this. However, if the odd chance is you are prepared to go
through the effort, I'd be anxious to find about if you were successful :)
Google on "Universal Prediction."
Frank
Cheers,
MB
[1] http://www.compressedsensing.com/
[2] http://www.dsp.ece.rice.edu/cs/random-filter-pub-03-web.pdf
[3] Zhi Tian; Giannakis, G.B., "Compressed Sensing for Wideband
Cognitive Radios," Acoustics, Speech and Signal Processing, 2007. ICASSP
2007. IEEE International Conference on , vol.4, no., pp.IV-1357-IV-1360,
15-20 April 2007
[4] Zhuizhuan Yu; Hoyos, S.; Sadler, B.M., "Mixed-signal parallel
compressed sensing and reception for cognitive radio," Acoustics, Speech
and Signal Processing, 2008. ICASSP 2008. IEEE International Conference
on , vol., no., pp.3861-3864, March 31 2008-April 4 2008
--
Martin Braun
Institut fuer Nachrichtentechnik
Universitaet Karlsruhe
http://www.int.uni-karlsruhe.de
_______________________________________________
Discuss-gnuradio mailing list
address@hidden
http://lists.gnu.org/mailman/listinfo/discuss-gnuradio