The following resources are available:

The latest release...

• LDPC_v2.1 (gzipped tar, 166kB (without binaries)), (September 7th, 2006)

  This release brings the following improvements:
  - The codec supports any size of symbols (in previous releases, the symbol size had to be multiple of 4). Note that sometimes the packet size must still be a power of two (i.e. if multiple symbols are packed in the same packet).
  - Improved performances (6.3% relative improvement during encoding, 3.8% during decoding with LDPC-Staircase codes, with a 20,000 symbol object and 2/3 code rate).
  - Source code clarifications. In particular, the src/ldpc_profile.h file gathers all the codec compilation controls. Don’t hesitate to give a look at it and change when needed.
  - A new SPARSE_MATRIX_OPT_SMALL_INDEX mode reduces significantly the memory requirements of sparse matrices (the matrix size is decreased by 1/6 on 32-bit systems).
  - Tested on 64-bit architectures (Pentium D)/Linux.
  - This version is in line with IETF Internet-Draft: draft-ietf-rmt-bb-fec-ldpc-03.txt (available in the doc/ directory), and is compatible with release 2.0 (but not the 1.* releases).
  - There is no major change in the API. Yet an LDPCFecSession::GetMaxN() method has been added which can be useful in SPARSE_MATRIX_OPT_SMALL_INDEX mode.

Previous releases

• LDPC_v2.0 (gzipped tar, 130kB (without binaries)), (March 9th, 2006)

  This is a major release (hence the new major version number). It brings the following improvements:
  - added the LDPCFecScheme class that implements parts of the LDPC-Staircase /Triangle FEC Scheme, as defined in draft-ietf-rmt-bb-fec-ldpc-01.txt. It defines the notion of packet, i.e. the possibility to group several symbols in the same packet, which dramatically improves the codec efficiency when operating on small objects. But with large objects, there will be no change, a packet being composed of a single symbol. See the above Internet Draft for more information. Using this class makes the symbol(s) <=> packet mapping almost transparent to the user, the optimal number of symbols per packet being automatically defined.
  - New Pseudo Random Number Generator algorithm (much faster).
  - Changed a little bit the Triangle matrix creation algorithm (moved from j = i; to j = i-1). !!! Therefore LDPC_v2.0 IS NOT backward compatible with LDPC_v1.* codecs, but it can easily be made compatible by changing the above two points if needed !!!
  - Corrected an error in LDPCFecSession::DecodingStepWithSymbol() which used to create all parity symbols systematically when decoding was over, leading to a waste of ressource (CPU/Memory). Shame on us for not finding it sooner :-(
  - This version is in line with draft-ietf-rmt-bb-fec-ldpc-01.txt (available in the doc/ directory).
  - Added perf_tool2 which is an extended version of perf_tool, using the LDPCFecScheme class in addition to the LDPCFecSession class. This is also an example of how to use the LDPCFecScheme class in practice.
  - Added a memory management optimization PART_SUM_OPTIMIZATION for reduced memory requirements (src/ldpc_fec.h). Explanations are available C. Neumann, ``Large Scale Content Delivery applied to Files and Videos’’, PhD Dissertation, INPG (Institut National Polytechnique de Grenoble), France, December 2005. in section 8.2.2 (as well as many other technical points related to the LDPC FEC building block).
  - Updated the FAQ.
  
  Several major changes in the LDPCFecSession API (sorry in advance for the inconvenience):
  - LDPCFecSession methods are renamed so that it only uses the term "symbol". The term "packet" is now reserved to the LDPCFecScheme methods.
  - the "leftDegree" argument of LDPCFecSession::InitSession() has been moved at the last position, since it is only meaningfull in case of LDGM codes (do not change with LDPC-Staircase/Triangle codes).

The latest release...

• mcl_v2.99.6_snapshot (gzipped tar, 859KB without binaries), (May 25th, 2005)
  This is a major version with many changes for the ALC and FLUTE part:
    Creation of a C++ FLUTE API. The Flute application is now a small application built on top of the FLUTE API. Writting your own application on top of this FLUTE API enables a full and dynamic control of the FLUTE/ALC session, e.g. to add transmission cycles to the carousel, add/remove files to the carousel, and many other things. Extending the example Flute application to add new functionalities is both feasible and straightforward.
    Receivers are no longer limited by their RAM size (receiving files whose size is an order of magnitude larger than the host’s RAM is feasible). This is made possible by a dedicated virtual receive memory management scheme.
    IPv4 SSM (source specific multicast) is now supported.
    IPv6 SSM (source specific multicast) is now supported (Linux with 2.6.x kernels, tested with FC2 and FC3 distributions).
    Added automatic MCLv3-FLUTE versus MAD-FLUTE interoperability validation scripts.
• mcl_v2.99.5 (gzipped tar, 684KB without binaries), (December 21st, 2004)
  This is a major version with many changes for the ALC and FLUTE part. We regard it as a stable version.
  1/ ALC part:
  
    Largely improved FLUTE application:
  • FLUTE is now based on Xerces-c, the XML library of the Apache Software Foundation (view README for build instructions)
  • FLUTE runs on all platforms, including Windows, FreeBSD, Solaris, and iPAQ/Linux (!)
  • Many FLUTE internal improvements
  • Improved performances at a receiver (one data copy less thanks to the mcl_recvmsg() function)
  • Added a reception meter for each file sent within the session (unix versions only)
  • Added FLUTE man page (and updated FCAST’s man page too)

    IPv6 support on Windows platforms
    Congestion control protocol can now be selected at runtime (rather than preset at compilation time).
    Short LCT header fields (16 bits) for the TOI and TSI fields are now supported and used whenever possible.
    AIX, IRIX, HP-UX support added (thanks to Florian Geyer). Not tested on our side though.
    Many bugfixes
    Warning: -singlelayer argument of FCAST/FLUTE is replaced by the -cc0 argument (i.e. no congestion control)
  2/ NORM part:
  
    No changes.
  Dependancies: only tested with LDPC v1.7 and Xerces-C++ v2.6.0

Previous releases

TBD

NORM implementation is included in the following MCLv3 distribution (but not in the new recent ones...).