Integrated Channel Estimation and Phase Optimization in RIS-Assisted MIMO Network: An End-to-End Deep Learning Approach

Downlink channel estimation and achievable rate optimization are critical challenges in current RIS-assisted MIMO systems. The two issues are studied separately: deep learning is used for channel estimation, and convex optimization is employed for rate maximization. But the step-by-step execution will lead to an increase in computational complexity and algorithmic complexity due to the separate handling of each task. To overcome these limitations, we propose an end-to-end integrated deep learning framework that jointly executes channel estimation and optimizes phase configuration to improve system performance. Specifically, a two-stage deep neural network architecture is introduced, which is trained using a multi-task optimization strategy. Moreover, the optimal network structure was explored by adjusting the scale of each stage and the training weight ratio. The simulation results show that the proposed algorithm significantly reduces computational overhead while maintaining high system performance.