一种利用机器学习优化MIMO窃听信道天线选择的方法

为保障多输入多输出窃听信道系统中信息传输的保密性，提出了一种基于机器学习的天线选择方案。首先利用机器学习解决分类问题准确率高、处理大数据高效这一优势，设计了基于奇异值分解的特征值提取、基于信干噪比的标签赋值方案，建立了k最近邻分类器和逻辑回归分类器选择最优天线最大化保密性能（可达保密速率和保密中断概率）。与传统天线选择方案相比，所提方案获得了几乎一致的保密性能，并且大幅降低了系统的选择复杂度和误比特率。

A Method for Optimizing Antenna Selection in MIMO Wiretap Channel by Machine Learning

For guaranteeing the confidentiality of information transmission in multiple-input multiple-output(MIMO) wiretap channels,a machine learning based antenna selection(AS) scheme is proposed.Firstly,the AS in MIMO wiretap channels is characterized as multiclass-classification problem.Then the advantages of machine learning is utilized to solve classification problem with high accuracy and efficient processing of big data,and the scheme is designed including singular value decomposition based feature extraction and signal-to-interference plus noise ratio(SINR) based label assignment.The k-nearest neighbors classifier and logistic regression classifier are established to select the optimal antenna to maximize the secrecy performance(the achievable secrecy rate and the secrecy outage probability).Compared with conventional AS scheme,the machine learning based scheme can almost acquire the same secrecy performance with decreasing bit error rate(BER) of the system and selection complexity by a large margin.