Deep Learning-Based Channel Estimation for One-Bit Multiuser Massive MIMO Systems

With the advancement of massive multiple-input multiple-output (Ma-MIMO) technology, channel estimation becomes increasingly challenging, especially for antennas that are equipped with only one-bit analog-to-digital converters (ADCs). Deep learning (DL) methods, including the current best-performing conditional generative adversarial network (cGAN), face significant challenges in accurately estimating the channel due to inadequately studied and designed loss functions.In this paper, we design a conditional generative adversarial network with a Transformer structure (TransGAN) to create more realistic channels, training two deep learning networks through adversarial training. TransGAN not only learns a mapping from the quantized received signal to the actual channel but also derives an adaptive loss function that enhances the network’s performance. Simulation results show that the TransGAN method designed in this paper is superior to the current DL method in terms of both performance and robustness.