Search

The effect of geometry on the relationship between measurement error and position determination error is described by geometric dilution of precision (GDOP). It is illustrated that a subset with lowest GDOP will result in lowest error. Since the global positioning system (GPS) GDOP computation based on complicated transformation and inversion of measurement matrices is a time consuming procedure, the neural network (NN) is used as an approximator or classifier for GDOP data. The back propagation (BP) is a most common method to train a feed-forward NN. However, in many applications including the GPS GDOP classification, it cannot train an NN with an acceptable speed and accuracy. Therefore, in this paper, a new approach to classify the GPS GDOP by using scaled conjugate gradient algorithm (CGA) to train a feed-forward NN and principal component analysis (PCA) is proposed. Scaled CGA is a powerful tool to train an NN, which is widely used in many applications that need to a high speed. PCA is a well-known method to reduce and optimize the dimensions of the data. PCA is applied on entire dataset in order to have some few uncorrelated and informative features. The results show that the scaled CGA with PCA has better performance than the scaled CGA without PCA and also, scaled CGA without PCA has better performance than the basic BP.