High Performance Methods for GPS GDOP Approximation using Neural Networks

要旨

In this paper, the Neural Networks (NNs)-based satellite geometry approximation for good or acceptable navigation satellite subset selection is presented. The approach is based on approximating the satellite Geometry Dilution of Precision (GDOP) factors utilizing the Radial Basis Function (RBF) NN, Multi-Layer Perceptron (MLP) NN, and Recurrent Neural Network (RNN) approaches. Without matrix inversion required, the NN-based approach is capable of evaluating all subsets of satellites and hence reduces the computational burden. This would enable the use of a high-integrity navigation solution without the delay required for many matrix inversions. The methods employed here are applicable regardless of the number of satellite signals being processed by the receiver. The tests results emphasize that total performance of MLP NN is better than RBF NN and RNN considering trade off between accuracy and speed for GPS GDOP approximation.