Time-delayed multiple linear regression for de-noising MEMS inertial sensors. (June 2019)

• Record Type: Journal Article
• Authors: Gonzalez, Rodrigo; Catania, Carlos A.
• Is Part Of: Computers & electrical engineering. Volume 76(2019)
• Page Start: 1
• Page End: 12
• Publication Date: 2019-06
• Subjects: MEMS -- Inertial sensors -- De-noising -- Multiple linear regression -- Supervised learning; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2019.02.023
• Languages: English
• ISSNs: 0045-7906

Abstract

Highlights: De-noising of MEMS inertial sensors is mandatory for real-world applications. Previous works assumed such errors cannot be compensated by linear models. MEMS sensors manufacturers claim nonlinearity in MEMS sensors is negligible. A linear method such as TD-MLR reduces significantly white noise in MEMS inertial sensors. It is statistically demonstrated that TD-MLR shows a similar performance to nonlinear models in addition to present a lower complexity. Abstract: Compensation of errors from MicroElectroMechanical Systems (MEMS) inertial sensors is mandatory for real-world applications. Recently, several machine learning methods focused on dealing with nonlinear behaviors have been proposed to increase MEMS inertial sensors performance. However, manufacturers of MEMS inertial sensors claim that nonlinearity in these devices is negligible in most cases. This article provides a rigorous analysis of the viability of Time Delayed Multiple Linear regression technique (TD-MLR) for decreasing white noise observed in MEMS inertial sensors. TD-MLR is evaluated on four MEMS inertial measurement units and compared to two well-known methods with different complexity levels, i.e., Moving Average (MA) filtering and the Multi Layer Perceptron (MLP). A strong statistical framework is applied on the three methods to guarantee that optimal models are obtained during the adjustment phase. Experimental results show that no significant statistical differences exist between TD-MLR and MLP. In addition, TD-MLR proves to be a remarkable improvement over MA.