A new magnetic compass calibration algorithm using neural networks

The magnetic compass can provide heading direction by measuring the Earth’s magnetic field. In practical applications, there usually exists an unwanted local magnetic field that will distort the magnetometer measurements; hence a calibration procedure is essential. Current calibration methods are limited by the inaccurate magnetometer error estimation when measurements are deteriorated by magnetic disturbances or large noises. This paper proposes a new compass calibration algorithm via modelling the nonlinear relationship between the compass heading and the true heading using neural networks. When an external heading reference is available, neural networks can be trained to properly model this nonlinear input–output pattern even in the presence of magnetic disturbances, and subsequently can be applied to convert the compass heading into the correct heading. The proposed algorithm does not require declination information and magnetometer biases and scale factor estimation. The simulation and field test results have verified the effectiveness and robustness of the proposed calibration method and have also shown that the calibration performance is proportional to the quality of the training data.