On blind deconvolution and classification of digitally modulated unknown QAM and PSK signals

Abstract

In this dissertation, a new automatic modulation type recognition system for classifying digitally modulated unknown QAM and PSK signals is proposed. The proposed system consists of the blind deconvolution and the classification stages. First, for the blind deconvolution, we employ the linear prediction to estimate the transmission channel characteristics and remove the effects of channel convolution. It removes the inter-symbol interferences resulting from the previous samples. Using the proposed method, we do not need a priori knowledge of statistical properties of transmitted data. Second, we propose new classification parameters, the variance of constellation magnitude ratios and the mean of mod 2π phase differences, for classifying QAM and PSK signals. Our proposed parameters are based on the statistical characterization of constellations and have the following advantages. The ratio of magnitudes does not depend on the absolute sizes of amplitudes. Therefore, no problems arise from the imprecise amplitude of the recovered signals. Our classification does not require the training experiments, therefore implementations are relatively easy. We derive the classification parameters in a Gaussian channel and find the optimal decision regions for the classification. Seven different types of QAM constellations - 4QAM, 8QAM, 16-square QAM, 16-cross QAM, 32QAM, 64QAM and 128QAM - and three different types of PSK constellations - BPSK, QPSK and 8PSK - are tested and the classification accuracy for each constellation type is investigated in various SNR environments. The effectiveness of our proposed scheme is tested for the PSTN voice telephony channel environment. From the simulation results, a good classification accuracy is obtained when SNR of the input signals is higher than that of required for the constellation types under test.