Model Selection Methods Based on Comprehensive Evaluation Criteria of AI for Air Interface

Previous research on AI-enabled communication has primarily focused on maximizing spectral efficiency, with limited attention to energy efficiency and model generalization. To address this, a trade-off factor is introduced to transform the multi-objective optimization problem into a single-objective problem, balancing spectral efficiency, energy efficiency, and generalization. This approach is applied to channel state information (CSI) feedback. In various channel scenarios, spectral and energy efficiencies corresponding to AI-based channel feedback accuracy are calculated, and a comprehensive objective value is determined through the trade-off of these metrics. Simulation results demonstrate that the optimal model can be selected from existing models by balancing these metrics, and the optimal model design can be further explored by fitting spectral efficiency as a function of computational complexity. The findings support green communication goals and are applicable to 6G AI-enabled systems.