Abstract:

Aiming at the problem of signal feature selection and classification, an algorithm of kernel principle component analysis (KPCA) and linear discriminant analysis (LDA) classifier is brought forward. According to the characteristic of communication signals, first the eliminating correlation and dimensionality reduction for these feature parameters are realized using a KPCA approach. Then the automatic recognition of modulation signals is designed by the LDA classifier. Simulation result shows that over a wide range of signaltonoise ratio scenarios, KPCA has a high performance in nonlinear classified feature. The classification accuracy based on KPCA+LDA is higher than principle component analysis (PCA)+ template matching. Through analysis it can also be seen that the KPCA+LDA algorithm is equal to the method of kernel Fisher discriminant analysis (KFDA).