Research On Classification For OFDM Signals Based On Neural Network

Digital modulation classification plays an important role in intelligent demodulation, electric surveillance,Software Defined Radio, etc. The researches had done mainly based on ideal environment or Additive White Gaussian environment only, rarely deal with the multi-path and Low SNR channel, which is the common case in radio communication. The Tree-based Neural Network is used often as it leads to a relatively simple neural-network-based scheme, and applied to many virtual situations because of its powerful competence in simulating the nonlinear relationship. However, how to design the configuration of the Decision Tree and select features in each tree node depend on the designer's experience. In this paper, a new neural network consisting of a Self-organizing Feature Map Neural Network and a Tree-based Neural Network is proposed. The Self-organizing Feature Map Neural Network is performed to extract the effective features firstly, and then the tree-based neural network is constructed by assembling several radial basis function neural network according to the result of the Self-organizing Feature Map Neural Network. To the problem of small sample set environment, a new Bootstrap technique is proposed to improve the performance of the classification system. Extensive simulations show that the new method has preferable performance in low SNR and multi-path environment.