Decoding a class of nonbinary codes using neural networks

A method for decoding RS (Reed Solomon) codes which used fixed decoding time and does not depend upon the number of errors present in messages is presented. The method is much simpler compared to existing decoding algorithms for RS codes. However, for large values of p (where p⩾0.2), the performance degrades significantly because for large noise levels the improvement obtained by error correction codes is less significant and is not necessarily worth the calculations and improved error correction techniques. This technique has not only resulted in improvement in signal to noise ratio and error correction capabilities but also saves the bandwidth. This savings was achieved by shortening the block length of RS code by omitting information symbols that do not reduce its minimum distance, and therefore any shortened RS code is also a maximum distance separable code. The use of neural network approaches improved the speed of computation, the simplicity, and robustness in addition to improvement in error correction and signal to noise ratio