Using Machine Learning and Light Spatial Sequence Arrangement for Copying Positioning Unit Cell to Reduce Training Burden in Visible Light Positioning (VLP)

Machine learning (ML) can improve the positioning accuracy in visible-light-positioning (VLP) system. To reduce the training time and complexity, the first step is to divide the whole positioning area into many positioning unit cells. The second step is to train one positioning unit cell; and then copy the 'trained' unit cell model to other un-trained 'target' unit cell. Here, we show that just copying and applying the 'trained' ML model to other unit cells will produce high positioning errors. We propose and demonstrate a light spatial sequence arrangement (LSSA) scheme together with second order linear regression (LR) ML algorithm to copy a 'trained' unit cell to a 'target' unit cell. A practical test-bed is constructed. By applying the proposed scheme, the average positioning error is significantly reduced by 90.71%, while the training burden is significantly reduced since there is no need of training in the 'target' unit cell.