Universal Nonhierarchical Automatic Modulation Recognition Techniques For Distinguishing Bandpass Modulated Waveforms Based On Signal Statistics, Cumulant, Cyclostationary, Multifractal And Fourier-Wavelet Transforms Features

Automatic Modulation Recognition (AMR) is one of the most important components in operation of cognitive software radio terminal by which the received signals are analyzed to determine the modulation formats that are present. This article describes five simple, universally applicable and computationally feasible AMR techniques, based on signal statistics, higher order statistics (cumulants), cyclostationary, multi-fractal and Fourier-Wavelet transforms features, suitable in software radio communications applications for distinguishing band-pass modulated waveforms. Eight representative transmitted signals are tested: 5 commonly employed commercial modulated waveforms, Quaternary Amplitude Shift Keying (QASK), Quaternary Frequency Shift Keying (QFSK), Quaternary Phase Shift Keying (QPSK), 16-Point Quadrature Amplitude Modulation (QAM-16 or QAM-4, 4), Gaussian Minimum Shift Keying (GMSK), and 3 military waveforms used in radar systems, Quaternary Linear Frequency Modulation (QLFM or 4-Chirp), Quaternary Pulse Width and Pulse Position Modulations (QPWM & QPPM). The received signals are processed to extract the signal statistics, cumulant, cyclostationary, multi-fractal and Fourier-Wavelet transforms features of the waveforms which are subsequently classified by a neural network to match with appropriate stored feature patterns. A correct modulation format is selected for a waveform that produces the highest matching output. Plots of correct classification probabilities for three best techniques and their combined 3-Best-AMR-Technique Majority-Selection-Rule scheme are generated which compares their relative performance for representative studied waveforms. The advantages and disadvantages of all five techniques are discussed.