Assessment of Discrete Stochastic Models of MEMS Inertial Sensors by Using the Allan Variance

Although MEMS (Micro-Electro Mechanical Systems) inertial sensors have key advantages as low cost and light weight, they are usually noisier than other types of inertial sensors. Thus, it is mandatory to compensate several types of errors from MEMS sensors to effectively use them in real-world applications. Most common errors are modeled as stochastic processes and numerically evaluated by using the Allan variance (AV) method. Stochastic models and AV are rarely assessed together in real-life scenarios and are taken for granted for inertial sensors of different qualities. In this work, it is proposed a methodology to evaluate the accuracy of both discrete stochastic models and the AV procedure. Discrete stochastic processes are simulated using the noise profile from a mid-range real inertial sensor. Finally, AV curves from both real and simulated sensors are compared. It is confirmed the validity of discrete stochastic models. On the other hand, it is exposed that the AV method can provide a coarse approximation to real errors.