Analysis of Multiple Prediction Techniques of Received Signal Strength to Reduce Surveying Effort in Indoor Positioning

Abstract

Received Signal Strength is the measure of attenuation of electromagnetic signals emitted by the access point, reaching the receiver after traveling some distance. This work used the attenuation of Wireless Local Area Network signals propagated through the air for the purpose of indoor positioning. Previous research had shown some problems such as indoor mapping requires human effort and are time-consuming. Furthermore, received signal strength for different indoor conditions may vary such that constant calibration and new acquisition for unknown indoor locations is required. An approach to reduce manual acquisition is by employing prediction algorithms. In this work, an analysis on prediction techniques used predict the RSS is analyzed in the context on indoor positioning. First, to determine the optimum training size for the models, the models are given different training size. Then the models are evaluated based on the similarity of signal pattern predicted and the error between the predicted signal and real signal. In conclusion, the random function model showed best estimation for signal for most of the tested signal received at certain distances from the transmitter. The optimum training size found for all the prediction models are 1100 out of 1200 data. It is also found that for a very noisy data set, the minimum training size for best result are at 900 out of 1200. Bayesian Support Vector Regression outperforms other models in terms of root mean square error.