Automatic Modulation Classification for Adaptive OFDM Systems Using Convolutional Neural Networks With Residual Learning

Automatic modulation classification (AMC) is becoming a promising technique for future adaptive wireless transceiver systems. The existing blind modulation classification (BMC) methods for orthogonal frequency division multiplexing (OFDM) fail to achieve the required performance by using statistical-based methods. Thus, the modulation classification research community is trying to adopt the deep learning (DL) method to improve the modulation classification accuracy. However, most of the existing DL methods for AMC of OFDM that involve the extraction of statistical features from the signal do not work for adaptive transceiver systems where the signal parameters are changed dynamically. In this paper, we design and implement AMC for adaptive OFDM systems by using a convolutional neural network (CNN) with residual learning. The proposed AMC can identify the modulation format of the received OFDM signal with different number of subcarriers, randomized carrier frequency offset (CFO), symbol timing offset (STO), phase offset, and unknown channel state information. We use residual learning to mitigate the effect of varying CFO, STO, and AWGN noise on the received OFDM signal. A larger pool of modulation schemes such as binary phase-shift keying (BPSK), quadrature PSK (QPSK), offset QPSK,
$\pi $
/4-QPSK, minimum shift keying, 8-PSK, 16-quadrature amplitude modulation (QAM), and 64-QAM are being considered for the proposed AMC for OFDM system in a dynamic environment. The performance and complexity of the proposed AMC are compared with the existing statistical feature-based and DL-based approaches. The proposed AMC for the OFDM system is also verified on the real-time data set generated from the universal software radio peripheral testbed setup.