An error correcting neural network

The behavior of a particular class of neural networks and their application to error control coding are discussed. It is proved that the net is able to perform analog error correction without hanging up at some undesired states-an annoying event that occurs occasionally in Hopfield (1982) nets. The corresponding code is a balanced code which has a significantly higher rate than Hopfield's 1-in-N code. Accordingly, the product of balanced code has a higher rate and better error correcting ability than Hopfield's permutation N*N code. The application of the code to combined modulation and channel coding for a 4-ary channel is discussed.<
>