Signal understanding of spectrum data using Bayesian network and neural network

This paper presents a method for automating the task of understanding signals of spectrum data. Probabilistic inferences are used for the purpose of decision theoretic fusion of multi-source data and knowledge, and a neural network is implemented within that as a subprocess that provides with pattern-specific concepts to the fusion model. The difficulties in using neural networks are: 1) vague transparency of the learned concepts; and 2) a screening problem of the training data to attain plausible learning. By restricting the concepts to be learned by the neural network only to the pattern-specific concepts and by joining it with another transparent probabilistic reasoning scheme, our proposing architecture could overcome the above problems.