The GNSS integer ambiguities: estimation and validation
The GNSS integer ambiguities: estimation and validation
The GNSS integer ambiguities: estimation and validation
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Table 3.2: Two-dimensional example. Mean <strong>and</strong> maximum difference between success rate,<br />
Ps, based on simulations <strong>and</strong> the approximations. <strong>The</strong> success rate for which the maximum<br />
difference is obtained is given in the last row.<br />
LB bootstr. LB region UB ADOP UB region ADOP<br />
mean difference 0.005 0.018 0.001 0.018 0.004<br />
max. difference 0.010 0.105 0.003 0.065 0.009<br />
Ps 0.805 0.388 0.833 0.559 0.805<br />
Table 3.3: Approximated success rates using simulations (sim.), the lower bounds based on<br />
bootstrapping (LB bootstr.) <strong>and</strong> bounding the integration region (LB region), the upper<br />
bounds based on ADOP (UB ADOP) <strong>and</strong> bounding the integration region (UB region), <strong>and</strong><br />
the approximation based on ADOP (ADOP).<br />
example sim. LB bootstr. LB region UB ADOP UB region ADOP<br />
2-D 1.000 0.999 1.000 1.000 1.000 0.999<br />
06 01 0.818 0.749 0.698 0.848 0.942 0.765<br />
06 02 0.442 0.410 0.118 0.475 0.626 0.417<br />
10 01 0.989 0.976 0.988 0.999 0.992 0.978<br />
10 03 0.476 0.442 0.126 0.681 0.661 0.525<br />
though it is more useful to know how well the approximations work in practice. <strong>The</strong>refore,<br />
simulations were carried out for several geometry-based models, see appendix B. <strong>The</strong><br />
resulting lower <strong>and</strong> upper bounds are shown in table 3.3. <strong>The</strong> first row shows the results<br />
for the two-dimensional vc-matrix with f = 1.<br />
<strong>The</strong> results show that Kondo’s lower bound works very well for a high success rate, but in<br />
general the bootstrapped lower bound is much better. It can be concluded that Kondo’s<br />
lower bound seems to be useful only in a few cases. Firstly, to obtain a strict lower bound<br />
the precision should be high, so that the success rate is high. Even then, it depends<br />
on the minimum required success rate whether or not it is really necessary to use the<br />
approximation: if the bootstrapped success rate is somewhat lower than this minimum<br />
required success rate, Kondo’s approximation can be used to see if it is larger. <strong>The</strong><br />
minimum required success rate could be chosen such that the fixed ambiguity estimator<br />
can be considered deterministic. In this case, the discrimination tests as used in practice<br />
can be used, see section 3.5.<br />
An advantage of the bootstrapped success rate is that it is very easy to compute,<br />
since the conditional variances are already available when using the LAMBDA method.<br />
<strong>The</strong> computation of Kondo’s lower bound is more complex, since for high-dimensional<br />
problems the number of facets that bound the pull-in region can be very large.<br />
It is difficult to say which upper bound is best. For the examples with only four visible<br />
satellites the ADOP-based upper bound is better than the one obtained by bounding<br />
the integration region, but in the examples with more satellites the latter is somewhat<br />
better. All bounds are best in the case of high precisions, i.e. high success rates. All<br />
in all, one can have a little more confidence in the ADOP-based bound, since its overall<br />
performance, based on all examples, is slightly better. An advantage of the ADOPbased<br />
upper bound is that it is easy to compute, whereas using the upper bound based<br />
44 Integer ambiguity resolution