Decoding complement-closed codes with coordinate-disjoint decomposition

Presents a multilayer neural net decoder for a group of linear block codes with complement-closedness (CC) property. Many well known linear block codes such as Hamming and Golay all fall into this group. A method called coordinate-disjoint decomposition (CDD) is introduced, and it is shown that this method leads to a minimised decoder structure. Coordinate-disjoint decomposition of Hamming (2 ^r-1, 2 ^r-r-1) and extended Hamming (2 ^r, 2 ^r-r-1) codes, Golay (24, 12) code, as well as quadratic residue (32, 16) code are illustrated, together with their corresponding decoder structures.<>