Noise Reduction by Data Fusion in a Multisensor System of Replicated MEMS Inclinometers

A Data Fusion (DF) approach for noise reduction in a system of multiple MEMS inclinometers is presented. The outputs of four inclinometers (ST IIS2CLX), that are nominally identical and set with the same operative conditions, have been acquired for 48 h consecutively. Each acquired dataset has been studied separately employing the Overlapping Allan VARiance (OAVAR) analysis to identify the Velocity Random Walk (VRW) and the Bias Instability (BI) noise contributions. The DF approach based on ensemble averaging across samples has been then applied combining the four acquired datasets, thus creating new datasets DFn, that contain averaged output data of n inclinometers at each acquisition time. The VRW and BI noise contributions of the DFn datasets have been identified through the OAVAR analysis and compared with the noise contributions of single inclinometers. Experimental results have shown a reduction of the noise variance σ2n for both the BI and VRW with a factor 1/n or, equivalently, the noise deviation σn with a factor $1/\sqrt n $, in good agreement with theoretical expectations.