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 3.5: Overview of all test statistics to validate the <strong>integer</strong> ambiguity solution as<br />
proposed in literature. <strong>The</strong> fixed solution is accepted if both the <strong>integer</strong> test <strong>and</strong> the<br />
discrimination test are passed.<br />
<strong>integer</strong> test discrimination test equation<br />
R1<br />
nˆσ 2 < Fα(n, m − n − p)<br />
R1<br />
nˆσ 2 < Fα(n, m − n − p)<br />
R1<br />
nˆσ 2 < Fα(n, m − n − p)<br />
ˇΩ2<br />
ˇΩ<br />
> Fα ′(m − p, m − p) (3.77)<br />
R2 > k (3.78)<br />
R1<br />
R2<br />
nˆσ 2 > Fα ′(n, m − n − p) (3.83)<br />
ˇΩ < χ 2 β (m − p, δ) ˇ Ω2 > χ 2 α(m − p) (3.84)<br />
R1<br />
nˆσ 2 < Fα(n, m − n − p) R2 − R1 > k (3.86)<br />
R1<br />
nˆσ 2 < Fα(n, m − n − p)<br />
R2−R1<br />
2 √ ˆσ 2 δ<br />
> tα ′(m − n − p) (3.87)<br />
An overview of all test statistics described in this section is given in table 3.5. It can<br />
be seen that there are three pairs of tests which have a similar ’structure’: the two<br />
ratio tests of (3.77) <strong>and</strong> (3.78); two tests (3.83) <strong>and</strong> (3.84) that consider the best<br />
<strong>and</strong> second-best <strong>integer</strong> solutions separately; <strong>and</strong> the two tests (3.86) <strong>and</strong> (3.87) that<br />
look at the difference d = R2 − R1. Note, however, that the discrimination tests of<br />
(3.78) <strong>and</strong> (3.86) have an important conceptual difference as compared to the other<br />
tests. <strong>The</strong>se two tests take only the float <strong>and</strong> fixed ambiguity solutions into account<br />
for discrimination, whereas the other tests take the complete float solution into account<br />
together with the fixed <strong>ambiguities</strong>.<br />
It would be interesting to know which of the approaches works best in all possible situations.<br />
For this purpose in Verhagen (2004) simulations were used for a two-dimensional<br />
example (geometry-free model with two unknown <strong>ambiguities</strong>). It followed that the ratio<br />
test (3.78) <strong>and</strong> the difference test (3.86) perform well, provided that an appropriate<br />
critical value is chosen. However, that is exactly one of the major problems in practice.<br />
It should be noted that in practice the ratio tests are often applied without the <strong>integer</strong> test<br />
(3.76), as it is thought that if one can have enough confidence in the float solution, the<br />
discrimination test is sufficient to judge whether or not the likelihood of the corresponding<br />
fixed ambiguity solution is large enough.<br />
In chapter 5 it will be shown that it is possible to give a theoretical foundation for some<br />
of the discrimination tests. <strong>The</strong> performance of the tests will also be evaluated, <strong>and</strong> it<br />
will be shown that one can do better by applying the optimal <strong>integer</strong> aperture estimator<br />
to be presented there.<br />
3.6 <strong>The</strong> Bayesian approach<br />
<strong>The</strong> Bayesian approach to ambiguity resolution is fundamentally different from the approach<br />
using <strong>integer</strong> least-squares, which is generally used in practice but lacks sound<br />
64 Integer ambiguity resolution