Using Linear Interpolation to Reduce the Training Samples for Regression Based Visible Light Positioning System
Yu-Chun Wu, Ke-Ling Hsu, Yang Liu, Chong-You Hong, Chi‐Wai Chow, Chien-Hung Yeh, Xin-Lan Liao, Kun-Hsien Lin, Yiyuan Chen
Abstract
We put forward and experimentally demonstrate a second order machine-learning (ML) based visible-light-positioning (VLP) system using simple linear interpolation algorithm to reduce the training samples required in the ML algorithm. Algorithms of the second order regression ML model using 2,430 training samples; and using the reduced training samples of 570 with and without the proposed linear interpolation are compared and discussed. We can observe that the positioning accuracy of using training samples of 570 with the proposed interpolation can have similar performance when compared with using 2,430 training samples. The training samples are reduced by ~76.5%. Here, off-the-shelf LED lamps and low bandwidth electrical and optical components are employed; and the system is cost-effective. Good quality on-off keying (OOK) identifier (ID) signals are retrieved after frequency down-conversion from 20 kHz, 40 kHz and 60 kHz without and with optical background noises respectively.