Automatic Modulation Classification in Non-Cooperative OFDM Systems Under Time-Varying Channels

Non-cooperative orthogonal frequency division multiplexing (OFDM) systems over time-varying channels suffer from inter-carrier interference (ICI), deep fading, carrier frequency offset (CFO), and phase offset (PO), which severely degrade conventional automatic modulation classification (AMC) performance. To tackle this issue, we propose a bidirectional chamfered distance-based AMC method (RCD-AMC). First, a regularized subband-smoothed recursive difference division (RAM-SCDD) preprocessing is introduced. It employs an SNR-dependent regularization factor and a local subband smoothing mechanism to cancel CFO/PO effects and mitigate noise spikes induced by channel variations, yielding a stable non-negative spectral quotient sequence. Second, an RCD feature extractor is developed, which leverages bidirectional matching errors and median aggregation to suppress down-order misclassification that plagues conventional error vector magnitude (EVM) at low SNR. Third, a fuzzy support vector machine (FSVM) driven by feature confidence is constructed, where matching residuals are mapped to sample memberships, and a differential penalty scheme adaptively curbs the influence of low-quality samples on decision boundaries. Simulation results demonstrate that RCD-AMC achieves superior classification accuracy and cross-channel generalization in both homogeneous and heterogeneous time-varying channel scenarios, while maintaining low computational complexity—effectively overcoming the performance degradation of traditional feature-based AMC under dynamically varying channel conditions.