SANTOS, E. A.; http://lattes.cnpq.br/2512401304880772; SANTOS, Eline Alves.
Résumé:
Wireless sensor networks are usually composed of a large number of sensor nodes densely deployed to monitor an environment. In a distributed classification problem, each sensor node sends, based on its observation, a decision to the sensor node (fusion center) responsible for making the final classification decision. The decisions transmitted by the sensor nodes may be corrupted by noise, an alternative is to use error correcting codes. In previous works, a distributed classification fusion approach using error correcting
codes has been proposed, where each one of the M classes or hypotheses is associated to a codeword with blocklength N, where N is the number of sensor nodes, each sensor node sends only a symbol of the codeword associated with the hypothesis that corresponds to its observation. In this approach the codewords are obtained by random search in the set of binary strings of length N, where N is the number of sensors. In this work it is proposed the use of classical block codes, more specifically BCH (Bose, Chaudhuri e Hockquegueim) codes, to obtain these codewords. The proposed approach allows tailoring
decoding algorithms supported by well known algebraic decoding algorithms. In particular, with the new approach it is possible to avoid a massive table look-up-based decoding for a large number of hypotheses, what cannot be achieved with random selected codewords. It is showed by simulation that algebraic code-based classification performance is similar to the performance of previous random search-based classification.