Neither of these implementations is even remotely fast enough for real-time applications. There are a few open-source fast LDPC decoders that take advantage of modern processors. The real stand-out is the one by Bertrand Le Gal (https://github.com/blegal/Fast_LDPC_decoder_for_x86). I have done an extensive study of available decoders and that one is at least 3x faster than the nearest competitor (see here http://aff3ct.github.io/hof_ldpc.html). A paper coming out soon describes this decoder running at 1 Gbps on a server.
Creating a GR wrapper for it is a trivial exercise (took me about 2 days). In my wrapper I use dlopen to read in a dynamic library for a particular code. The dynamic libraries are built for all the available matrices.
Unfortunately it does not support arbitrary matrices and matrices cannot be set at runtime. Bertrand is very helpful and has created matrices for codes when I asked. We (or he) can put in a stand-alone application to convert from standard .alist format to his ‘C’ header files.
NOTE: we have also updated Bertrand’s code to run on an ARM NEON and it is quite fast (3 Mbps @ 1.2 GHz)
I am sorry that I cannot offer to make my code available to the community.
_______________________
Eugene Grayver, Ph.D.
Aerospace Corp., Principal Eng.
Tel: 310.336.1274
________________________
No comments:
Post a Comment