|
Recently error correcting codes are used for decreasing error
probability of any communication systems, improving
communication quality and prolonging the executing time of
communication systems without error. Error correcting codes
have become a standard technology and been used wide spreadly.
Because of the parallel processing ability of the Artificial
Neural Network, it can perform massive parallel processing. As
a result, the problems which are complicated and real-time
processing can be solved. Traditional decoding algorithms is
either too complex or too slow, thus it limits the efficiency
of communication systems and feasibility of some error
correcting codes. In this thesis, we can replace traditional
error correcting decoder by either Perceptron, Back-propagation
network or Hamming network. The new decoder resulted is called
neural-net decoder. The neural-net decoder can be divided into
total neural-net decoder and partial neural-net decoder if the
complexity and suitability are taken under consideration.
According to the size of message bits or syndrome bits for (n,
k) linear block codes, the goals of real-time and low-
complexity probably can be achieved by using total neural-net
decoder or partial neural-net decoder. In this thesis, we
propose the utilization of Perceptrons in partial neural-net
decoder and the utilization of modified Hamming networks for
total neural-net decoder. The results of computer simulation
show that it is able to achieve 100% error correcting rate if
there are only limited t error bits in the transmission. So
neural-net decoders can be applied on traditional decoders
indeed.
|