Deep Reciprocity Calibration for TDD Mmwave Massive MIMO Systems Towards 6G
Shu Xu,Zhengming Zhang,Yinfei Xu,Chunguo Li,Luxi Yang
DOI: https://doi.org/10.1109/twc.2024.3400616
2024-01-01
Abstract:Ideally, the bi-directional channel in time division duplex (TDD) millimeter wave (mmWave) massive multiple-input multiple-output (MIMO) systems exhibits reciprocity. However, the involvement of low-cost and non-ideal radio frequency (RF) chains disrupts this reciprocity. Consequently, prior to fully leveraging the advantage of channel reciprocity, it is essential to implement channel calibration. Despite numerous over-the-air calibration methods, such as Argos, the typical least square (LS) are proposed in the literature, none of their criteria directly focus on the calibration performance. To address this gap, we propose a novel deep learning based approach that aims to optimize the calibration performance and introduce device-level intelligence toward 6G networks. To be specific, two cascaded modules are designed in a model-assisted end-to-end manner. Firstly, we propose the double-CNN-based channel denoising module for joint bi-directional channel estimation by exploiting the characteristics of mmWave channel. Secondly, the deep calibration learning module is meticulously designed to obtain the calibration coefficients with the aid of assisted model. This traceable assisted model is established by leveraging the expert knowledge of calibration process, based on which the MetrNet and the CaliNet are designed. Numerical results demonstrate the superior performance of our proposed method compared to existing calibration methods. Particularly, additional simulations and analysis are conducted to verify the effectiveness of the two properly designed modules.