Application of neural network technique in implementation of constant weight codes

Constant weight codes are very useful in high speed optical data links and spread spectrum systems. The error correction/detection depends on the length of the code (amount of redundancy added) and minimum distance of the code. The current technique, utilizing lookup tables for error correction/detection, is a problem as far as high throughput is concerned. A scheme utilizing a neural network approach is proposed. It results in better error correction and retransmission capability than does the current technique and can be used efficiently in frequency hopped systems. The neural network approach is much faster, and the decoding process has been reduced to simple vector-matrix multiplication