Performance comparison of neural network and statistical discriminant processing techniques for automatic modulation recognition

摘要

Automatic classification of modulation types in analog and digital communications would be very useful for signal intercept and monitoring facilities and also allow realization of auto- switched multimode receivers. An earlier study demonstrated that linear discriminant analysis (LDA) could be successfully enhanced with multivariance analysis of variance (MANOVA) to isolate the effects of modulation class, noise, and also their interactions in an automatic modulation recognition (AMR) process for digital data signals. The current research develops an artificial neural network (ANN) solution for digital AMR using a nonlinear multilayer perceptron network. Results indicate that the ANN gives greatly improved performance over LDA and similar performance to MANOVA in the full range of signal/noise ratios.