Abstract
Blind modulation identification (BMI) of 6G aeronautical multi-beam satellite (MBS) systems is prominently challenging due to the intricacy of Shadowed-Rician (SR) fading, limited channel state information (CSI), and inter-beam interference (IBI). These factors degrade the statistical properties of modulation signals, rendering conventional terrestrial BMI techniques inapplicable to MBS systems. To remedy the above flaws, this paper proposes a physics-informed scattering transformation network (RCR-SCTNet) using symbol-level rotation correlation reconstruction for BMI of MBS signals. Initially, the rotation correlation reconstruction (RCR) strategy with partial CSI is designed to transform destructive interference into constructive interference, alleviating IBI, SR and limited CSI problems. Then, the constellation intensity matrix is constructed as the training data for RCR-SCTNet with the help of domain knowledge. Finally, the scattering transformation module is employed to capture salient features with the lowest trainable parameters, and the gradient centralization (GC) strategy is further integrated into RCR-SCTNet to achieve stable and efficient training. Experimental results demonstrate that our RCR-SCTNet performs best among previous BMI methods in the severe MBS channels. Further, it remains a great generalization, is robust and has low computation complexity.
Funding Information
  • National Natural Science Foundation of China (62201113, 62401011)
  • Natural Science Foundation of Anhui Province (22408085QF183, 2408085QF202)
  • Anhui Polytechnic University through Scientific Research Initiation Fund Program (S022024001, S022024002)