Low-Resolution Massive MIMO Channel Estimation With LSTM Attention-Based CBDNet

Channel estimation of a massive multi-input multi-output (MIMO) system that utilizes a one-bit analog-to-digital converter (ADC) is a foremost challenge. Traditional deep learning (DL) approaches have been recently employed to circumvent this problem; however, they are limited to noise levels. Unlike the existing works, we use a DL-based denoise architecture for channel estimation from one-bit received signals, improving the estimation performance as the signal-to-noise ratio (SNR) increases. The model leverages a dual-branch architecture to estimate and remove noise from input data. We propose a DL model: a long short-term memory (LSTM) attention-based convolutional blind denoising network (LA-CBDNet) comprising the noise estimation subnetwork and the non-blind denoising subnetwork. The noise estimation subnetwork comprises convolutional layers estimating the noise map, followed by an LSTM to converge the noise estimation. The non-blind subnetwork comprises accompanying attention and LSTM layers estimating the noise matrix. The numerical results demonstrate that our model performs better than benchmark approaches for varying SNRs and base station (BS) antennas. In addition, it outperforms the comparative methods for different pilot lengths and number of users.