Neural net decoders for some block codes

The paper presents simple neural circuits as an analogue decoder for plain balanced codes, a soft-decision decoder for the (16, 9, 6, 5, 4) error correcting balanced code, and a hard-decision decoder for the (47, 24) quadratic residue code. The authors study the dynamics of a class of very simple Hopfield nets and prove that they can be used as analogue decoders for plain balanced codes. This result is further extended so that the net is able to store constant weight binary vectors. With these nets as basic building structures, the authors then construct simple decoders for the (16, 9, 6, 5, 4) error correcting balanced code and the (47, 24) QR code, respectively.<>