Ensemble Learning Aided Large Communication Model for Multi-Scenario Nonlinear Distortion

In this paper, we propose the large communication model (LCM), a deep neural network receiver designed specifically for orthogonal frequency-division multiplexing (OFDM) systems. Inspired by the Mixture of Experts (MoE) model, LCM incorporates ensemble learning within the Comm-Trans Net framework to address the challenges posed by various nonlinear distortions in wireless communication environments. By utilizing ensemble methods, LCM achieves robust adaptation to diverse distortion scenarios without requiring specific prior domain knowledge of specific distortion types. This architecture enables LCM to dynamically adapt to complex distortion environments while maintaining high performance. Extensive evaluations demonstrate that LCM consistently outperforms traditional OFDM receivers, highlighting its effectiveness across a wide range of distortion conditions. Our findings emphasize the reliability and adaptability of LCM in maintaining near optimal communication performance.