Advanced techniques for automatic classification of digitally modulated communication signals

Automatic classification of the modulation type of a received signal is an indispensable step in many communication systems. It provides necessary information for data demodulation, information extraction and signal exploitation. In recent years, modulation classification is one of the most promising research areas and has found a variety of military and commercial applications.; In this research, a set of advanced techniques are proposed and investigated for automatic classification of digitally modulated signals. For inter-class classification at moderate to high signal-to-noise ratio (SNR) environment, we propose to use the wavelet transform to discriminate among quadrature amplitude modulation (QAM), phase shift keying (PSK) and frequency shift keying (FSK) signals. The wavelet transform can effectively extract the transient characteristics from different modulation types for simple identification. Then we focus on intra-class classification between binary PSK (BPSK) and quadrature PSK (QPSK) at moderate to low SNR environment. At low SNR environment, the performance of the classifier using wavelet transform degrades quickly, because the extracted features are masked by the noise and difficult to recognize. On the other hand, the decision theoretic technique that is based on likelihood function works well at all SNR environment. We developed the composite hypothesis tests to identify between BPSK and unbalanced QPSK signals, and to discriminate between BPSK and QPSK signals without prior knowledge of signal level. Furthermore, we applied the composite hypothesis testing approach to operate on antenna array outputs for the purpose of increasing the accuracy of BPSK and QPSK identification when only a short data record is available. The above decision theoretic based classifiers require some unknown parameters that must be estimated before the classification decision can be made. Hence, Cramer-Rao lower bound is derived to evaluate the performance of the proposed estimators in obtaining the unknown parameters.