TransNet: Full Attention Network for CSI Feedback in FDD Massive MIMO System

Channel state information (CSI) is a key aspect of massive multi-input multi-output (MIMO) system. It depicts important properties of transmission channels such as scattering, fading, the attenuation of power with distance, etc. The quality and cost of CSI feedback between user equipment (UE) and base station (BS) play vital roles in the quality of the whole communication system. In this letter, a new deep learning (DL) method based on Google’s famous Transformer architecture is presented for CSI feedback in frequency division duplex (FDD) massive MIMO system. Simulation results show that the presented inception network named TransNet outperforms other DL methods on the quality of CSI feedback.

I. Introduction

Massive multi-input multi-output (MIMO) is one of the core technologies of the current 5G and the following 6G communication system. It is a method to increase sector throughput and capacity density by using large numbers of antennas at base station (BS). Channel state information (CSI) depicts crucial characters for signal transmission in a specific channel. In order to improve the communication quality of the whole wireless system, it is important for both transmitters and receivers to attain complete and accurate CSI. In time-division duplex (TDD) system, the downlink and uplink channels are divided by different time slots and have little difference in fading characteristics. It is useful to use channel reciprocity to conduct a back channel estimation. However, the same approach is not suitable for frequency division duplex (FDD) system, where the uplink and downlink work at different frequencies. As a result, user equipment (UE) of FDD system always needs to send CSI of downlink channels as complete and accurate as possible to BS by feedback link.