Towards Practical Antenna Selection Based on Multilabel CNN for Large‐Scale MIMO System in an Indoor Scenario

Abstract

Large‐scale multiple input multiple output (MIMO), also known as massive MIMO, serves as a transformative technology that effectively addresses the surging need for data‐intensive applications in the realm of 5G mobile networks and beyond. With a multitude of antennas at its disposal, a large‐scale MIMO base station possesses the capability to concurrently improve by orders of magnitude system spectral and energy efficiencies. Nevertheless, hardware cost and power consumption arise as serious challenge. To circumvent this latter, we investigate the potential of multilabel convolutional neural network (ML‐CNN) to develop an antenna selection (AS) model that based on the available channel state information (CSI) at the transmitter, it dynamically selects the optimal subset of antennas in real‐time. We evaluate the model performance using real indoor channel measurements under three different antenna array configurations. The obtained results demonstrate that the proposed ML‐CNN approach performs similarly to the convex relaxation‐based method, a mathematical technique often used for AS problems that provides optimal solutions with high computation cost, with the advantage of significantly reduced computation time. Moreover, it exhibits strong resilience across various antenna array configurations. Additionally, assessing the ML‐CNN's performance in scenario with imperfect channel estimation confirms its robustness.