无线知识驱动的大规模MIMO信道估计

Wireless knowledge driven channel estimation in massive MIMO

大规模多输入多输出(multiple-input multiple-output, MIMO)技术能够显著提升无线通信系统的能量效率和频谱效率,其关键前提是获取精确的信道状态信息,然而导频复用造成的导频污染问题严重影响信道估计精度,成为制约大规模MIMO系统性能的核心难题.本文提出无线知识驱动的大规模MIMO信道估计模型,基于信道海图技术挖掘用户在角度域空间中的近邻关系作为无线信道知识,设计基于无线知识驱动的卷积神经网络(convolutional neural network, CNN)信道估计器(WKDCNN)提高信道估计精度.研究表明,在不同信噪比(signal-to-noise ratio, SNR)下,天线数、导频污染程度、训练集大小对所提方法的信道估计性能均有重要影响.仿真结果表明, WKD-CNN在高信噪比区域的归一化均方误差相较于传统信道估计算法、基于多层感知机(multilayer perceptron, MLP)和无信道海图知识驱动卷积网络的信道估计方法均明显降低.

Massive multiple-input multiple-output(MIMO)technology can significantly improve the energy and spectral efficiencies of wireless communication systems.The key premise is to obtain accurate channel state information.However,pilot contamination caused by pilot reuse critically affects the channel estimation accuracy and is thus a core problem restricting the performance of massive MIMO systems.In this paper,a wireless knowledge-driven massive MIMO channel estimation model is proposed.Based on the channel chart technology,the neighboring relationship among users in the angular domain space is mined as wireless channel knowledge,and a convolutional neural network(CNN)channel estimator driven by wireless knowledge(WKD-CNN)is designed to improve the channel estimation accuracy.The research shows that under different signal-to-noise ratios,the number of antennas,the degree of pilot pollution,and the size of the training set significantly impact the channel estimation performance.The simulation results show that the normalized mean square error of WKD-CNN in the high signal-to-noise ratio region is reduced significantly compared with the conventional channel estimation algorithm,multilayer perceptron(MLP)-based channel estimation method,and CNN without channel charting.