跨时间迁移的多源无线信号指纹融合定位方法

Fingerprinting localization of cross-temporal transferred and multi-source wireless signal fusion

针对传统指纹定位方法中无线信号覆盖范围有限以及信号时变特性难以处理的问题,提出利用多种无线信号进行指纹定位的方法,并采用测地线流核(geodesic flow kernel,GFK)的迁移学习方法减轻信号时变性对定位精度的不利影响。首先,通过对多源无线信号特征进行多轮随机采样融合,从而构建多个数据集,获得更丰富多样的指纹特征;其次,对不同时间的指纹特征进行迁移的GFK进行融合,使得迁移从2个领域扩展到多个领域;最后,使用多个数据集训练基分类器,并从多个基分类器得到最终的预测结果,从而提高位置分类器的泛化性能。实验结果表明,本方法的定位精度比传统方法更高。

To address the problem of limited coverage area of wireless signals and difficulty in dealing with time-varying characteristics of wireless signals in traditional fingerprinting localization,we propose a method of using multi-source wireless signals for fingerprinting localization,and the accuracy of positioning,which is affected by the time varying of signals,is mitigated by geodesic flow kernel.Firstly,we construct our datasets by a multi-round random sampling of multiple wireless signal sources,which provides a richer and more diverse fingerprint features.Secondly,we fuse geodesic flow kernels from fingerprint features of different times,so that we extend transferring methods from two domains to multiple domains.Finally,base classifiers are trained on multiple datasets,and the predicted position are obtained from all base classifiers,so as to elevate the generalization of the model.Simulation results show that the proposed method outperforms the traditional approaches in terms of positioning accuracy.