Automatic modulation classification with genetic backpropagation neural network

Abstract

Automatic modulation classification of digital signals plays an important role in civilian and military applications. The challenge focuses on the efficiency under low signal noise ratio (SNR) and compatibility with new types of digital modulations. In this paper, we propose a high-efficiency classification system for both the classical digital modulations and the binary offset carrier (BOC) and its derivative modulations. In detail, the classical digital modulations are ASK, PSK and FSK, and the new kind of signals are BOC, composite binary offset carrier (CBOC) and alternative binary offset carrier (AltBOC). Our system consists of two parts: feature extraction and classification algorithm. For feature extraction, we extract a suitable combination of signal statistical characteristics and instantaneous characteristics to provide better ability to distinguish different modulation signals. First, we preprocess the signal using the Hilbert transform to get the analytic expression. Then, four instantaneous parameters and four statistical parameters are used to represent the features of signal based on the expression. For classification algorithm, we investigate a genetic backpropagation neural network (BPNN). Genetic algorithm (GA) is used to design the architecture of BPNN to find the best value for the number of hidden layers and the number of neurons in each layer. This approach eliminates the human factor and improves the efficiency and accuracy of network. The simulation results demonstrate that our system shows high classification accuracy and high speed for the researched digital modulation signals at low SNR of 3dB.