Professional Documents
Culture Documents
Abstract. A probabilistic justi®cation is given for using where y is the given GPS data vector of order m, a and
the integer least-squares (LS) estimator. The class of b are the unknown parameter vectors respectively of
admissible integer estimators is introduced and classical order n and o, and e is the noise vector. The matrices A
adjustment theory is extended by proving that the and B are the corresponding design matrices. The data
integer LS estimator is best in the sense of maximizing vector y will usually consist of the `observed minus
the probability of correct integer estimation. For global computed' single- or dual-frequency double-dierence
positioning system ambiguity resolution, this implies (DD) phase and/or pseudorange (code) observations
that the success rate of any other integer estimator of the accumulated over all observation epochs. The entries of
carrier phase ambiguities will be smaller than or at the vector a are then the DD carrier phase ambiguities,
most equal to the ambiguity success rate of the integer expressed in units of cycles rather than range. They are
LS estimator. The success rates of any one of these known to be integers, a 2 Z n . The entries of the vector
estimators may therefore be used to provide lower b will consist of the remaining unknown parameters,
bounds for the LS success rate. This is particularly such as for instance baseline components (coordinates)
useful in case of the bootstrapped estimator. and possibly atmospheric delay parameters (tropo-
sphere, ionosphere). They are known to be real-valued,
b 2 Ro .
Key words. Integer LS á GPS ambiguity resolution á The procedure which is usually followed for solving
Ambiguity success rate the GPS model of Eq. (1) can be divided into three steps
(for more details we refer to e.g. Teunissen 1993 or de
Jonge and Tiberius 1996). In the ®rst step we simply
disregard the integer constraints a 2 Z n on the ambigu-
ities and perform a standard adjustment. As a result we
1 Introduction obtain the (real-valued) estimates of a and b, together
with their variance±covariance matrix
Ambiguity resolution applies to a great variety of global 2 3 2 3
positioning system (GPS) models currently in use. These a^ Qa^ Qa^b^
range from single-baseline models used for kinematic 4 5; 4 5
2
positioning to multi-baseline models used as a tool for b^ Qb^
^a Qb^
studying geodynamic phenomena. An overview of these
and other GPS models, together with their application This solution is referred to as the `¯oat' solution. In the
in surveying, navigation and geodesy, can be found in second step the `¯oat' ambiguity estimate a^ is used to
textbooks such as those of Leick (1995), Parkinson and compute the corresponding integer ambiguity estimate
Spilker (1996), Hofmann-Wellenhof et al. (1997), Strang a. This implies that a mapping F : Rn 7! Z n , from the n-
and Borre (1997) and Teunissen and Kleusberg (1998). dimensional space of real numbers to the n-dimensional
Despite the dierences in application of the various GPS space of integers, is introduced such that
models, it is important to understand that their ambi- a F
^
a
3
guity resolution problems are intrinsically the same.
That is, the GPS models on which ambiguity resolution Once the integer ambiguities are computed, they are
is based can all be cast in the following conceptual frame used in the third step to ®nally correct the ¯oat estimate
of linear(ized) observation equations of b. As a result we obtain the `®xed' solution
y Aa Bb e
1 b b^ ÿ Qb^ ÿ1
a ÿ a
^a Qa^
^
4
588
The ambiguity residual
^ a ÿ a is thus used to adjust the 2 A class of integer estimators
¯oat solution so as to obtain the ®xed solution.
It is of course not enough to compute the ®xed so- There are many ways of computing an integer ambiguity
lution and be done with it. We can always compute such vector a from its real-valued counterpart a^. To each such
a solution, whether it is of good quality or not. We method belongs a mapping F : Rn 7! Z n from the n-
therefore still need to address the question whether we dimensional space of real numbers to the n-dimensional
have enough con®dence in the computed integer ambi- space of integers. Once this map has been de®ned, the
guity solution. After all, unsuccessful ambiguity resolu- integer ambiguity vector follows from its real-valued
tion, when passed unnoticed, will all too often lead to counterpart as a F
^ a. Due to the discrete nature of
unacceptable errors in the positioning results. We Z n , the map F will not be one-to-one, but instead a
therefore need to have a way of knowing how often we many-to-one map. This implies that dierent real-valued
can expect the computed ambiguity solution to coincide ambiguity vectors may be mapped to the same integer
with the correct, but unknown, solution. Is this 9 out of vector. We can therefore assign a subset Sz Rn to each
10 times, 99 out of a 100, or a higher percentage? It will integer vector z 2 Z n :
certainly never equal 100%. After all, the integer am-
biguities are computed from the data: they are therefore Sz fx 2 Rn j z F
xg; z 2 Zn
5
subject to uncertainty just like the data are.
The subset Sz contains all real-valued ambiguity vectors
In order to obtain such a description, we require the
that will be mapped by F to the same integer vector
probability distribution of the integer ambiguities
z 2 Z n . This subset is referred to as the pull-in region of z
(Teunissen 1997). This distribution will be a probability
(Jonkman 1998; Teunissen 1998a). It is the region in
mass function, due to the integer nature of the ambi-
which all ambiguity ¯oat solutions are pulled to the
guities. Of this probability mass function, the proba-
same ®xed ambiguity vector z.
bility of correct integer ambiguity estimation is of
Having de®ned the pull-in regions, we are now in a
particular interest. This probability will be denoted as
position to give an explicit expression for the corre-
P
a a. It describes the frequency with which one can
sponding integer ambiguity estimator. It reads
expect to have a successful ambiguity resolution. It
equals the expected ambiguity success rate. This prob- X
1 if a^ 2 Sz
ability depends on three contributing factors: the a zsz
^
a with sz
^
a
6
0 otherwise
functional model (the observation equations), the sto- z2Z n
chastic model (the distribution and precision of the
Since the pull-in regions de®ne the integer estimator
observables) and the chosen method of integer ambi-
completely, we can de®ne a class of integer estimators by
guity estimation. Changes in any one of these will aect
listing properties of these pull-in regions. In this section
the success rate.
we introduce three properties for which it seems
In this contribution the choice of integer ambiguity
reasonable that they are possessed by the pull-in regions.
estimator will be considered. In general, we would like to
It seems reasonable to ask of the pull-in regions that
have the highest success rate possible. We therefore
their union covers the n-dimensional space completely
would like to know which integer ambiguity estimator
maximizes P
a a. For this purpose we ®rst need to [
S z Rn
7
introduce a class of candidate integer estimators. Such a z2Z n
class of admissible integer estimators is introduced in
Sect. 2. In constructing this class, we are led by practical Otherwise we would have gaps, in which case not every
considerations such as the following: the estimator a^ 2 Rn could be assigned to a corresponding integer
should map any ¯oat solution to a unique integer so- ambiguity vector.
lution and when the ¯oat solution is perturbed by an Another property that we require of the pull-in-re-
integer amount, the integer solution should be perturbed gions is that any two distinct regions should not have an
by the same integer amount. In Sect. 3 we give three overlap. Otherwise we could end up in a situation where
examples of integer estimators which belong to this class a ¯oat solution a^ 2 Rn cannot be assigned uniquely to
of admissible estimators. They are the `rounding' esti- a single integer vector. For the interior points of two
mator, the `bootstrapped' estimator and the integer distinct pull-in regions we therefore require
least-squares (LS) estimator. \
Section 4 contains the main result of this contribu- Sz1 Sz2 ;; 8 z1 ; z2 2 Z n ; z1 6 z2
8
tion. It is proven that the integer LS estimator has the
largest ambiguity success rate of all admissible estima- We allow the pull-in regions to have common bound-
tors. The integer LS estimator is therefore the best es- aries, however. This is permitted if we assume zero
timator in the sense of maximizing the probability of probability that a^ lies on one of the boundaries. This will
correct integer estimation. Any other integer ambiguity be the case when the probability density function (pdf)
estimator, such as for instance the `rounding' estimator of a^ is continuous.
or the `bootstrapped' estimator, will have a smaller The third and last property asked is that the integer
success rate. GPS ambiguity resolution will therefore be map F possesses the property that F
x z
less successful when integer estimators other than the LS F
x z; 8 x 2 Rn ; z 2 Z n . This property is also a
estimator are used. reasonable one to request. It states that when the ¯oat
589
solution is moved by an integer amount z, the corre- of the de®nition are satis®ed, since ± apart from ties in
sponding integer solution is moved by the same integer rounding ± any ¯oat solution a^ 2 Rn gets mapped to
amount. This property allows one to use the `integer a unique integer vector. The third condition is also
remove±restore' technique: F
^ a ÿ z z F
^
a. It satis®ed since rounding admits the integer remove±
therefore allows us to work with the fractional parts of restore technique, that is, x ÿ z z x; 8 x 2 R; z 2 Z.
the entries of a^, instead of with its complete entries, Since componentwise rounding implies that each
which may sometimes be large numbers. real-valued ambiguity estimate a^i ; i 1; . . . ; n, is map-
The integer remove±restore property implies that ped to its nearest integer, the absolute value of the
dierence between the two is at most 1/2. The pull-in
Sz1 z2 fx 2 Rn j z1 z2 F
xg regions SR;z that belong to this integer estimator are
fx 2 Rn j z1 F
x ÿ z2 F
x ÿ z2 g therefore given as
fx 2 Rn j z1 F
y; x y z2 g
SR;z \ni1 x 2 Rn j j xi ÿ zi j 12g ; 8 z 2 Z n
11
Sz1 z2 ; 8 z 1 ; z2 2 Z n :
They are n-dimensional cubes, centred at z 2 Z n , all
Hence, it means that the pull-in regions are translated having sides of length one.
copies of one another. This third property may therefore
also be stated as
3.2 Integer bootstrapping
Sz z S0 ; 8 z 2 Zn
9
with S0 being the pull-in region of the origin of Z n . Another relatively simple integer ambiguity estimator is
Integer ambiguity estimators that possess all three of the bootstrapped estimator (Blewitt 1989; Dong and
the above stated properties form a class. This class will Bock 1989). The bootstrapped estimator can be seen as a
be referred to as the class of admissible integer ambi- generalization of the previous estimator. It still makes
guity estimators. It is de®ned as follows. use of integer rounding, but it also takes some of the
correlation between the ambiguities into account. The
P bootstrapped estimator follows from a sequential con-
De®nition. The integer estimator a z2Z n zsz
^
a is said
to be admissible if ditional LS adjustment and it is computed as follows.
S If n ambiguities are available, we start with the ®rst
1. z2Z n Sz Rn ambiguity a^1 , and round its value to the nearest integer.
T Having obtained the integer value of this ®rst ambiguity,
2. Sz1 Sz2 ;; 8 z1 ; z2 2 Z n ; z1 6 z2
the real-valued estimates of all remaining ambiguities
3. Sz z S0 ; 8 z 2 Z n are then corrected by virtue of their correlation with the
®rst ambiguity. Then the second, but now corrected,
Various integer estimators exist that belong to this class.
real-valued ambiguity estimate is rounded to its nearest
As the de®nition shows, one way of constructing
integer. Having obtained the integer value of the second
admissible estimators is to choose a subset S0 such that
ambiguity, the real-valued estimates of all remaining
its translated copies cover Rn without gaps and overlaps.
n ÿ 2 ambiguities are then again corrected, but now by
In two dimensions this can be achieved, for instance,
virtue of their correlation with the second ambiguity.
by choosing S0 as the unit square centred at the origin.
This process is continued until all ambiguities are
considered. The components of the bootstrapped esti-
mator aB are given as
3 Examples of admissible estimators
aB;1 ^
a1
In this section three dierent admissible integer estima-
tors are considered. All three of them have been in use, aB;2 ^ a2 ÿ ra^2 a^1 rÿ2
a2j1 ^ a^1
^ a1 ÿ aB;1
in one way or another, for GPS ambiguity resolution. ..
They are the `rounding' estimator, the `bootstrapped' .
estimator and the LS estimator. h X
nÿ1 i
aB;n ^
anjN a^n ÿ ra^n a^ijI rÿ2
a^ijI
^
a ijI ÿ
aB;i
12
i1
3.1 Integer rounding
where the shorthand notation a^ijI stands for the ith LS
ambiguity obtained through a conditioning on the
The simplest way to obtain an integer vector from the
previous I f1; . . . ;
i ÿ 1g sequentially rounded am-
real-valued ¯oat solution is to round each of the entries
biguities.
of a^ to its nearest integer. The corresponding integer
The bootstrapped estimator is admissible. The ®rst
estimator reads therefore
two conditions of the de®nition are satis®ed, since ±
aR
^ an T
a1 ; . . . ; ^
10 apart from ties in rounding ± any ¯oat solution gets
mapped to a unique integer ambiguity vector. The in-
where `[]' denotes rounding to the nearest integer. This teger remove±restore technique again applies. To see
estimator is clearly admissible. The ®rst two conditions this, let a0B be the bootstrapped estimator which corre-
590
Teunissen PJG (1997) Some remarks on GPS ambiguity resolution. Teunissen PJG (1998c) Success probability of integer GPS ambi-
Arti®c Sat 32(3): 119±130 guity rounding and bootstrapping. J Geod 72: 606±612
Teunissen PJG (1998a) On the integer normal distribution of the Teunissen PJG, Kleusberg A (eds) (1998) GPS for Geodesy, 2nd
GPS ambiguities. Arti®c Sat 33(2): 49±64 edn. Springer, Berlin Heidelberg New York
Teunissen PJG (1998b) A class of unbiased integer GPS ambiguity
estimators. Arti®c Sat 33(1): 3±10