The GNSS integer ambiguities: estimation and validation
The GNSS integer ambiguities: estimation and validation
The GNSS integer ambiguities: estimation and validation
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
is maximum. <strong>The</strong> degrees of freedom qi may be different, <strong>and</strong> the level of significance<br />
αqi depends on λ(αqi , γ, qi), which is chosen as:<br />
λ(αqi , γ, qi) = λ(αm−n, γ, m − n) (2.81)<br />
This means that the non-centrality parameters <strong>and</strong> detection powers are chosen equal<br />
for both the detection <strong>and</strong> the identification test, implying that a certain model error<br />
can be found with the same probability by both tests. This approach is the B-method<br />
of testing, (Baarda 1968).<br />
In practice the identification step is started with the so-called data snooping, (Baarda<br />
1968), which means that each individual observation is screened for the presence of an<br />
outlier. In that case q = 1 <strong>and</strong> the matrix C reduces to a vector, denoted as c. <strong>The</strong> test<br />
statistic of equation (2.70) becomes:<br />
T 1 = (w) 2<br />
w =<br />
c T Q −1<br />
y ê<br />
<br />
c T Q −1<br />
y QêQ −1<br />
y c<br />
(2.82)<br />
<strong>The</strong> test statistic w has a st<strong>and</strong>ard normal distribution N(0, 1) under H0. An observation<br />
j is suspected to be biased if:<br />
|wj| ≥ |wi| ∀i <strong>and</strong> |wj| > N 1<br />
2 α(0, 1) (2.83)<br />
<strong>The</strong> final step in the DIA-procedure is adaptation, in which the model is corrected for one<br />
or more of the most likely model errors identified in the previous step. This may involve<br />
replacement of the erroneous data, e.g. by re-measurement, or a new null hypothesis is<br />
set up which takes into account the identified model errors. After this step, one has to<br />
make sure that the new situation will lead to acceptance of the null hypothesis, meaning<br />
that the DIA-procedure needs to be applied iteratively.<br />
2.5.3 <strong>GNSS</strong> quality control<br />
In this section a general testing procedure was described that can be applied in the case<br />
of linear <strong>estimation</strong> with normally distributed data, so that the parameter distribution<br />
is completely captured by the vc-matrix of the estimators. Unfortunately, this approach<br />
cannot be applied when <strong>integer</strong> parameters are involved in the <strong>estimation</strong> process, since<br />
these parameters do not have a Gaussian distribution, <strong>and</strong> therefore the parameter distribution<br />
itself is needed to obtain appropriate test statistics in order to validate the<br />
<strong>integer</strong> solution. This problem is the subject of section 3.5 <strong>and</strong> chapter 5.<br />
Least-squares <strong>estimation</strong> <strong>and</strong> quality control 25