Fuzzy processing
GPS accuracy has a statistical distribution, which is dependent on two important factors. The expected accuracy will vary with the error in the range measurements as well as the geometry or relative positions of the satellites and the users. The Geometric Dilution of Precision (GDOP) indicates how much the geometric relationship of the tracked satellites affects the estimate of the receiver's position, velocity and time. The errors in the range measurements used to solve for position may be magnified by poor geometry. For the same range measurement errors, lower GDOP causes more accurate estimates [3].
The GDOP and the sum of Signal to Noise Ratio (SNR) are used as input fuzzy variables to the fuzzy processing unit. Fuzzy system output is defined as Reliable Factor [4]. The main block diagram of fuzzy processing is shown in Fig.1.
Fig.1. The main block diagram of fuzzy processing
GDOP and SNR are divided into three and four segments for partition the rule space, respectively. R.F. is fuzzified with a singleton membership function [4]. The membership functions are shown in Fig.2.
Fig.2. Membership functions: (a) GDOP, (b) SNR and (c) R.F.
The approximate reasoning method is used for the inference process and the center of area method is employed for the defuzzification [4]. There are twelve rules in the rule base, which are shown in Table 1.
| R.F. | SNR |
| S | MS | MB | B |
| GDOP | S | S | MS | MB | B |
| M | S | S | MS | MB |
| B | S | S | S | MS |
Neural network prediction
In this special application, the patterns of fuzzy selected position errors (dx, dy and dz) are fed into neural network. This neural network is trained by such patterns. Then it finds the ability to predict later errors [5,6]. Because of linear correlation between x,y and y,z and also x,z , the neural network should be of such kinds that their input variables include patterns of positioning errors conjointly [7]. Neural network outputs are values of predicted errors of x component (ˆdx(n+1)), y component (ˆdy(n+1)) and z component (ˆdz(n+1)). The
structure of this topology is shown in Fig.3. Back-Propagation (BP) algorithm is used to train the neural network.
Fig.3. The structure of proposed neural network
GPS accuracy has a statistical distribution, which is dependent on two important
factors. The expected accuracy will vary with the error in the range measurements as well as the geometry or relative positions of the satellites and the users.
|
