The development of a GPS/MEMS INS integrated system utilizing a hybrid processing architecture

Integrated INS/GPS systems have evolved into small lowcost systems with the advent of MEMS technology. This has expanded the scope of these systems to include the civil community which can now use Global Positioning Systems (GPS) in single point positioning (SPP) mode integrated with Micro-Electro-Mechanical Systems (MEMS) Inertial Measurement Units (IMUs) for everyday applications. Most civil or recreational applications do not require a high degree of accuracy, but the performance of current MEMS based IMUs do not yet meet these loose requirements for general navigation applications due to their noisy measurements and poor stability. The traditional method for INS/GPS integration has been the Kalman Filter. Newer methods using artificial intelligence have proposed replacing the Kalman Filter entirely, but because of processing requirements and a dependence on suitable training data the Kalman Filter has remained at the forefront of INS/GPS integration. The architecture proposed in this paper uses a hybrid combination of Kalman Filtering and neural networks to overcome the disadvantages of both stand-alone methods. A two layered feedforward backpropagation neural network is used to learn how the Kalman Filter residual errors behave during GPS outages. Once this network is trained it can be used in prediction mode to provide compensation to the navigation Kalman Filter drifts in the north, east and up directions.