Temperature Variation Effects on The Stochastic Performance of Smartphones Sensors Using Allan Variance and Generalized Method of Wavelet Moments

This paper investigates the Generalized Method of Wavelet Moments (GMWM) method for stochastic modeling of low-cost Micro-Electro-Mechanical-Systems (MEMS) inertial sensors, and compares the results with the most widely used Allan Variance (AV) results. The purpose is to investigate the stochastic characteristics of MEMS sensors and their changes with temperature changes, and the pros and cons of the AV method in identifying low-cost MEMS sensor error sources and computing the error parameters quantification. Both AV and GMWM are used to identify the stochastic error sources and their quantitative models for sensors in two smartphones under four temperature points. The outcomes show that for smartphone sensors, one needs to consider the variation of not only deterministic errors, but also stochastic error parameters when temperature changes. Thus, a temperature dependent stochastic model of the sensor drift has the potential to enhance the sensor performance. Compared with AV, GMWM require computational load, but provide two benefits: (a) It has the ability to check the existence of outliers inside the data set which could be done by comparing the classical calculated Wavelet Variance (WV) by another robust one, and (b) It can provide all possible error combinations, build number of candidate models and give a rank to indicate the optimal one using manual and automatic ranking techniques.