Application of Learning Vector Quantization Neural Network for Digital Modulation Classification

This paper deals with the problem of classification of a group of frequently used modulations, named BPSK, QPSK, BASK, QASK, BFSK and QFSK. A new hierarchical algorithm of digital modulation classification is proposed, using learning vector quantization neural network as classifier and wavelet transform together with entropy and Fourier transform of the received signal as features. Moreover, a new method is proposed to initial the weight vectors of learning vector quantization neural network, in which two main problem of this neural network is solved and also its convergence is accelerated and facilitated. Although most of the feature-based methods of modulation classification need an optimization algorithm in order to find the most suitable features or the best amount of classifier’s parameters to get an appropriate result, simulation results show that the proposed algorithm can separate all the six mentioned modulations and achieves a high accuracy even at negative levels of signal-to-noise ratio, without the help of any optimization algorithm