Lightweight generative channel estimation with adaptive regularization in massive MIMO systems

Abstract

In massive multiple-input–multiple-output (MIMO) systems, as the number of transmit antennas increases, the number of transmit pilots increases for the channel estimation of user equipment. To reduce pilot overheads, generative-model-based channel estimation methods have been studied recently. In this paper, we propose a lightweight generative adversarial network (GAN)-based channel estimation scheme with adaptive regularization. Additionally, the proposed scheme groups users based on the channel collinearity matrix obtained using pre-acquired channel realizations for each user, and uses GANs trained on a per-user group basis. We design lightweight GAN architectures based on the locally centralized sparse characteristics of the beamspace channel. Here, we develop a novel objective function that adaptively determines the relative importance between the received pilots and the pre-trained generator. The simulation results show that the proposed scheme significantly improves performance in terms of the estimation accuracy and complexity when compared with conventional GAN-based channel estimation schemes. Moreover, the proposed scheme can achieve a flexible trade-off between the performance and the complexity of the training network.