Channel Attention-Based Path Loss Prediction Model in Asymmetric Massive MIMO Systems

In asymmetric massive multiple-input multiple-output (MIMO) systems, the transmitting (Tx) and receiving (Rx) arrays are designed asymmetrically, resulting in nonreciprocal uplink (UL) and downlink (DL) propagation conditions and thus different parameters, e.g., path loss (PL). In this paper, we propose a novel channel attention-based PL prediction model. An image-based feature representation method of an asymmetric propagation environment is proposed. The efficient channel attention (ECA) module is added to a convolutional neural network (CNN) to enhance effective features and suppress ineffective features. With the proposed model, wireless propagation features and the beamwidth feature can be extracted from the three-channel images synthesized by the asymmetric propagation images, the user equipment (UE) propagation images, and environmental feature images. Simulation results illustrate that the proposed model outperforms the basic CNN model and the compared AI-based model.