Joint Learning of Channel Estimation and Beamforming for Cell-Free Massive MIMO Systems

In cell-free massive multiple-input multiple-output (MIMO) systems, beamforming is a crucial technique to enhance system performance. Nevertheless, practical challenges arise in obtaining accurate downlink channel state information (CSI), which plays a crucial role in effective beamforming design. Moreover, the limited fronthaul capacity hinders system performance improvement. In this letter, we propose a model-driven deep learning-based framework to obtain the optimal beamforming solution with unknown downlink CSI, comprising the neural network (NN) module and the beamforming recovery (BR) module. Specifically, the NN module jointly trains two neural networks with different functions to obtain downlink CSI and key features extracted from the beamforming matrix. According to the output of the NN module, the BR module recovers the beamforming matrix during the deployment phase. The simulation results demonstrate that the performance of the proposed method is close to that of the numerical algorithm and outperforms benchmark approaches, but the computational complexity is significantly reduced.