Characterization of MEMS Accelerometer self-noise by means of PSD and Allan Variance analysis Antonino D’Alessandro, Giovanni Vitale, Salvatore Scudero, Roberto D’Anna, Antonio Costanza National Earthquakes Centre Istituto Nazionale di Geofisica e Vulcanologia Rome, Italy antonino.dalessandro@ingv.it Adriano Fagiolini, Luca Greco University of Palermo Istituto Nazionale di Geofisica e Vulcanologia Palermo, Italy Abstract—In this paper, we have studied the sources of error of a low-cost 3-axis MEMS accelerometer by means of Power Spectral Density and Allan Variance techniques. These techniques were applied to the signals acquired from ten identical devices to characterize the variability of the sensor produced by the same manufacturer. Our analysis showed as identically produced accelerometer have somehow variable behavior in particular at low frequency. It is therefore of paramount importance before their use in Inertial Navigation or Earthquakes Monitoring System, a complete characterization of each single sensors. Keywords—Micro Electro-Mechanical Systems, low-cost Accelerometer, self-noise, Power Spectral Density, Allan Variance, Inertial Navigation System, Earthquakes Monitoring System. I. INTRODUCTION Micro Electro-Mechanical Systems (MEMS) are a highly enabling technology with a huge commercial potential. In the 90s, MEMS accelerometers revolutionized the automotive airbag system and are today widely used in laptops, tablet, games controllers and mobile phones. Thanks to their great commercial success, the research and development of MEMS technology actively continues all over the world. Due to their versatility, MEMS accelerometers are increasingly being used in a wide field of sciences, including physics, engineering, and medicine. Recent advances in the construction of MEMS have made possible to manufacture small and light accelerometers, with very good performance, employable in the realization of Inertial Navigation Systems (INS) or on Earthquakes Monitoring System (EMS). An INS is a self-contained navigation technique in which measurements provided by accelerometers and gyroscopes are used to track the relative position and orientation of a given object knowing its starting point, orientation and velocity [1], [2]. An INS generally contains three accelerometers, which are commonly arranged with their sensitive axes perpendicular each to one another, to measure vector accelerations. Given the ability to measure the acceleration of the moving body, it is possible to calculate the change in velocity and position by performing successive mathematical integrations of the acceleration with respect to time. Indeed, to calculate the position of the device, the accelerometers signals are double integrated. An EMS is a network of sensors able to record accelerations generated by an earthquake. Nowadays, the sensitivity and the dynamic range of low-cost MEMS accelerometers are such as to allow the recording of earthquakes of moderate magnitude even at a distance of several tens of kilometers [3], [4]. Due to their low cost and small size, MEMS accelerometers can be easily installed in urban areas in order to achieve dense urban seismic networks, which can be used for monitoring, earthquake early warning and optimization of civil protection interventions[5], [6]. In Japan, California, Taiwan, and Italy, EMS consisting entirely of MEMS sensors are being developed like the Home Seismometer Network [7] managed by the Japan Meteorological Agency, the Quake-Catcher Network (QCN, [8], [9]) initiated by the University of Stanford, the Community Seismic Network (CSN, [10]) managed by the California Institute of Technology, the Palert in Taiwan [11] and the MEMS network [12] by the Istituto Nazionale di Geofisica e Vulcanologia (INGV) in Italy. It is clear that the international community of seismology is focusing on that technology that could revolutionize in a short time the way to monitor earthquakes. However, low-cost MEMS accelerometers are affected by different types of noise, which could degrade the accuracy of an INS or the detection threshold of an EMS. It is therefore essential, before their use in INS, EMS or other applications, a complete characterization of the sensors self-noise. These errors consist of deterministic and stochastic parts [13], [14]. The deterministic part includes constant biases, scale factors, axisnonorthogonality, axis misalignment and so on, which are removed from raw measurements by the corresponding calibration techniques. The stochastic part contains random errors which cannot be removed from the measurements and should be modeled as stochastic processes. In this paper, we have studied the main sources of error of a low-cost 3-axis MEMS accelerometer by means of Power Spectral Density (PSD) and the Allan Variance (AV) techniques. These techniques were applied to the signals acquired from ten identical devices to characterize the variability of sensors produced by the same manufacturer. After a short introduction to the method and noise type, we describe the equipment and experiment and the main results of the signals analysis.