The Statistical Analysis of Noise in Triaxial Sensor

This thesis demonstrates the statistical behaviour of noise in triaxial sensors, particularly magnetometers. The simple algorithm is proposed to support statistical tests. It is shown by statistical test that the sensor noise does not follow a Normal distribution as is commonly assumed. Beside noise, other sources of error corruption in magnetic field recording sensors are discussed. The combined effects of errors are demonstrated by a mathematical model containing translation, rotation and scaling factors. With this model, magnetometer readings are calibrated under the Maximum Likelihood Estimator scheme. The Least Square framework provides the unconstrained optimization problem with closed form solution. This solution can be used as initial guess for log-likelihood function. It guarantees that the algorithm converges to an accurate solution and reduces the number of iterations. Calibrated data is aligned to the sensor frame in order to be applied later in navigation applications, where the North finding problem is solved by true geomagnetic readings