A Novel Blind High-Order Modulation Classifier Using Accumulated Constellation Temporal Convolution for OSTBC-OFDM Systems

Deep learning (DL)-based blind modulation classification (BMC) has flourished in the single carrier system for low-order modulation signals. However, the research of DL-based BMC has not been well explored in the OSTBC-OFDM system (Orthogonal Space-Time Block Coded-Orthogonal Frequency Division Multiplexing), especially high-order modulation types. This brief proposes a novel BMC algorithm for high-order modulation signals (e.g., 4096QAM), which is based on reconstruction processing and accumulation constellation temporal convolution BMC network (ACTC-BMCNet). First, we employ the blind zero-forcing (ZF) equalization algorithm to reconstruct the damaged signal and enhance the signal representation power. Then, the constellation accumulation strategy is executed on the raw constellation diagram (CD) to generate the accumulated CD (ACD) image for reducing constellation dispersion. After that, instead of traditional 2D convolution, the constellation temporal convolution enabled in ACTC-BMCNet is leveraged to reduce computational overhead and extract the more discriminative feature. Monte-Carlo experiments are performed on fourteen modulation signals, and results indicate that the proposed ACTC-BMCNet possesses a superior classification accuracy and a faster recognition speed than the DL-based benchmark classifiers.