Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.compression > #1267

Searching for the best perfomance LDPC or TC for competition with our new approach

Show all headers | View raw

We have implemented some new approach to forward error correction (Correction Trees) and it essentially outperform all state-of-art methods we could find - we would like to ask if someone knows some better ones?
For now there is only correction for Binary Search Channel (each bit has independently given probability to be flipped), but this channel is extremely difficult to get near Shannon's limit, making it perfect to compare different approaches.
Here is comparison of output BER to input BER for different rate 1/2 or 1/3 codes (LDPCs come mainly from MacKay's library): http://dl.dropbox.com/u/12405967/comaprison.jpg
The others give up at about 7.5%, while ours (triangles) provides similar correction for 9.25%. The used parameters were chosen for online correction on standard PC - using dedicated hardware we could further reduce output BER.
Here is implementation and link to arxiv paper: https://indect-project.eu/correction-trees/

ps. I apology for placing it on comp.compression, but it's the closest to error correction and most of concepts involved has appeared thanks of working on entropy coder (Asymmetric Numeral Systems).

Back to comp.compression | Previous | Next | Find similar | Unroll thread

Thread

Searching for the best perfomance LDPC or TC for competition with our new approach Jarek Duda <dudaj@interia.pl> - 2012-04-28 01:43 -0700

csiph-web