Maximum Likelihood Decoding Using Neural Nets

Abstract

Maximum likelihood decoders are presented here for plain constant weight codes† as well as the (24, 12) Golay code. Both codes have very nice arithmetic structures. Constant weight codes can be decoded by a class of simple Hopfield nets. The (24,12) Golay code consists of 8 sybcodes with identical structures. We store one of the subcodes in a three layered net and use it to decode the entire code. Compared with traditional methods, the neural net decoders have faster response and simpler hardwares.

Indexing terms:

• Maximum likelihood decoders
• Neurals nets