Neural network error corrector for binary messages on hydro-acoustic channels

An application of neural networks for the identification and correction of transmission errors in binary messages is described. The network is used as a classifier of detected hydroacoustic signals. It converts the signals into one of a possible alphabet of symbols. The algorithm used is a Hamming-type neural network classifier associated with the transmission of a Hamming code. This system can detect and correct all transmission errors if the number of errors is less than or equal to half the Hamming distance between transmitted symbols minus one. Symbols to be transmitted are chosen and associated to messages, assuring that bit-to-bit nonsimilarities result on the prescribed Hamming distance. The auto-associative error correcting scheme can be used to generate a teaching signal to a supervised learning equalizer tracking the channel nonstationary characteristics. The proposed system is intended for use in hydroacoustic communication applications and is undergoing sea tests.<
>