Automatic modulation classification (AMC) is an essential task in intelligent receivers. AMC research papers over multipath fading channels have two problems. The first problem is the Higher-order moment (HOM)based normalized channel...
moreAutomatic modulation classification (AMC) is an essential task in intelligent receivers. AMC research papers over multipath fading channels have two problems. The first problem is the Higher-order moment (HOM)based normalized channel coefficients estimator is not valid for some types of digital modulations. The second problem is poor classification accuracy. This paper solves these problems through significant steps, first, by finding a new HOM-based normalized channel coefficients estimator for all cases of digital modulation types, and second, by finding the mathematical forms of the transmitted signal's estimated normalized HOMs and Higherorder cumulants (HOCs), and third, by selecting the most discriminative estimated HOCs for AMC using features selection algorithms. The simulation results show that the classification accuracy of the selected estimated HOCs, for M-array Phase Shift Keying (MPSK) and M-array quadrature amplitude shift modulation (MQAM) classification is highly improved compared with reference papers. It's 100% for the three-tap multipath channel case, and Signal-to-noise ratio (SNR) values greater than 6 dB. Keywords Automatic Modulation Classification, Higher-order Cumulants, Multipath fading channel, Feature selection algorithms. The automatic modulation classification (AMC) algorithm determines the modulation type of the received signal. This task is somehow simple over the Additive white Gaussian noise (AWGN) channel, but it becomes more complicated over the multipath fading channel. Research papers [1-3] improved the AMC performance using HOCs over AWGN channels. Our work focuses on improving the AMC performance using HOCs but over multipath fading channels. Research papers classify the digital modulation types over multipath fading channels through two steps [4-6]: First, the normalized channel coefficients are estimated using the HOM of the received signal. Second, some of the normalized cumulants of the transmitted signal are estimated based on the previously estimated normalized coefficients. In , the author shows the performance accuracy of binary phase-shift keying (BPSK) and Quadrature Phase-Shift Keying (QPSK) classification, using the estimated normalized cumulant 42, ˆx C of the transmitted signal for the four-tap multipath case and SNR value of 10 dB is 89%. The performance accuracy of 4-quadrature amplitude modulation (4QAM), 16QAM, and 64QAM classification is 77%. In [5], the author shows the performance accuracy of BPSK and QPSK classification, using the estimated normalized cumulant 63, ˆx C of the transmitted signal for the four-tap multipath case, and SNR value of 10 dB is 91%. The performance accuracy of QPSK, 16QAM, and 64QAM classification is 45%. In [6], the author shows the performance accuracy of BPSK, QPSK, 8PSK, 16QAM, and 64QAM classification, using three normalized estimated cumulants 40, ˆx C , 41, ˆx C , and 42, ˆx C of the transmitted signal for the four-tap multipath case and SNR value of 10 dB is 82%. As a result, the AMC performance degrades for lower SNR values cases. So, the selected estimated HOCs in have poor performance for AMC over the multipath fading channel. Deep learning techniques have also been used to classify digital modulations over multipath channels . In [7], the author uses the convolutional neural network technique (CNN) to classify PSK, QAM, frequency-shift keying (FSK), Pulse-amplitude modulation (PAM), and analog modulation. The performance accuracy for SNR value of 10 dB is 90%.In [8], the author uses 60, y C and 63, y C of the received signal as input features to the stacked convolutional auto-encoder to classify BPSK, QPSK, 16-QAM, and 64-QAM over the multipath channel. The performance accuracy for the 5-tap multipath Rayleigh fading channel and SNR value of 10 dB is 91%. ' ' / 2 / 2 21 21 Estimate the normalized channel coefficients (Section 4)