Volume 1, Issue 5, October 2011

Comparing GPS GDOP Approximation Accuracy using Recurrent Wavelet Neural Network and ARMA Modeling [Download Paper]

S. Tafazoli, M. R. Mosavi and N. Rahemi


Abstract—A GPS receiver must be capable of determining visible satellites and then choose four satellites among them which
have  the  best  geometric  arrangement  to  minimize  the  observation  error.  Usually,  the  best  arrangement  is  provided  by minimizing the Geometric Dilution of Precision (GDOP) parameter. The most correct method of calculating GDOP uses inverse matrix  for  all  combinations  and  selects  the  lowest  ones.  However,  the  inverse  matrix  method,  especially  when  there  are  so many  visible  satellites,  imposes  a  huge  calculation  load  on  the  processor  of  the  receiver.  In  this  paper,  the  GPS  GDOP approximation accuracy based on Recurrent Wavelet Neural Network (RWNN) and Auto Regression Moving Average (ARMA) models has been compared. These methods provide a realistic calculation approach to determine GPS GDOP without any need
to calculate inverse matrix. The simulation results demonstrate that ARMA modeling has greater accuracy than RWNN modeling for selecting an optimal subset of satellites.