MANet: End-to-end MIMO wireless communication network

MIMO technology has emerged as a critical foundational technology for 5G and 6G communication systems due to the fast growth of mobile communication technology. However, for assuring optimal performance, precise channel estimate and efficient channel status information feedback are crucial. In this paper, we use deep neural networks to enhance the performance of MIMO channel estimation in OFDM systems. To build an end-to-end MIMO wireless communication system, we suggest a channel estimator based on MLP-mixer, with CSI feedback employing multi-head attention as the backbone network. In the channel estimation stage, Smooth_L1 serves as the loss function, cosine similarity serves as the loss function in the feedback stage, and NAdam serves as the training optimization method. Using simulation data from high-density and low-density pilot signals, performance analyses and comparisons with conventional estimators like LS and MMSE, as well as neural network techniques like CNNNet, were carried out. According to the results, the suggested MANet works well on OFDM MIMO communication systems when it comes to dense and sparse pilot estimate and CSI feedback.