一种基于机器学习分类器的UWB异常信号识别方法

UWB abnormal signal recognition based on machine learning classifier

针对超宽带定位技术中非视距信号与多径信号影响超宽带定位精度的问题,在不考虑信道冲激响应这一重要特征前提下,分别从测试与训练场景相同和异同两方面对比研究支持向量机、随机森林和多层感知器3种机器学习方法的低成本超宽带异常信号识别性能。通过将多径信号与非视距信号分离,完成视距信号、非视距信号和多径信号的数据多分类。对比性实验结果表明,测试环境与训练环境相同和异同两种情况下的随机森林分类器表现最优,识别准确度可达92.52%、74.82%;在缺失信道冲激响应这一重要特征条件下,机器学习依然可以较好地识别非视距信号和多径信号,为低成本室内定位提供一种思路。

Aiming at the problem that NLOS signals and multipath signals affect the positioning accuracy of UWB in ultra wideband positioning technology,without considering the important feature of channel impulse response,the paper compares the performance of low-cost UWB abnormal signal recognition based on three machine learning methods,namely,Support Vector Machine,Random Forest and Multilayer Perceptron,from the same or different aspects of test and training scenarios.By separating multipath signals from NLOS signals,data multi classification of LOS signals,NLOS signals and multipath signals has been completed.The comparative experimental results show that the Random Forest classifier performs best under the same or different test and training environment,and the recognition accuracy can reach 92.52%and 74.82%;In the absence of channel impulse response,machine learning can still better identify NLOS signals and multipath signals,providing new ideas for low-cost indoor positioning.