There are several advantages of
performing the position calculation
on the server (handset-assisted AGNSS)
[9]. One of the advantages is
that there is opportunity to include
additional range measurements
that the server can obtain from the
network into the position calculation.
This can improve both the yield and
accuracy of the location solution.
When the handset is in an area where
its view of the sky is obscured it may
not be able to lock onto the minimum
number of satellites to perform a
position calculation. In this case
the other range measurements may
permit a position to be calculated
when otherwise there would be none.
In the case where there is more than
the minimum number of satellites in
view, additional measurements provide
increased measurement redundancy
and hence improve yield and accuracy
by making the position calculation less
susceptible to bad measurements.
The accuracy of range measurements
depends on the radio technology and
their sources. Range measurements
are generally either calculated from
signal strengths measured from
a known source in combination
with a propagation model, or are
determined from timing measurements
associated with known signals
arriving from a known source They
could be Timing Advance (TA)
in a GSM network, Uplink Time
Difference of Arrival (UTDOA)
time difference measurements,
WiFi or Digital TV (DTV) ranging
signals. When incorporating these
range measurements into the
position calculation algorithm,
the accuracy of the measurement
needs to be considered so that it
doesn't make the location worse.
Some results of position calculation
accuracy with simulated range
measurements are shown in Figure 3
and Figure 4. Here the case where there
are not enough satellite measurements
to do a complete GPS solution is
considered. GPS data was collected
over a complete day from a GPS
receiver on the roof of the laboratory
and consists of 80,657 distinct epochs.
In order to simulate a situation where
only three satellites are in view (which
is the case in Figure 3), three satellites
are randomly selected from each epoch
and range measurements are introduced
with varying levels of inaccuracy.
These simulated range measurements
show how the accuracy of the overall
position calculation is influenced by
errors in the range measurements
across a whole day of GPS data.
The position calculation function
used for experiments is a parametric
weighted least squares implementation
where range measurements are
considered as another input to
the solution with appropriate
weighting in the stochastic model.
Figure 3 shows the effect of the
error in the position calculation
when there are only three satellites
in view of the receiver and there are
different numbers of simulated range
measurements of varying inaccuracies.
It shows that when there are only
3 satellites and 1 accurate range
measurement there is less than 10
metres of error over a complete day
of data for 67% of the data and just
over 10 for 95%. As the inaccuracy
of the range measurement increases the distance of the calculated location
from the ground truth increases. For
example, GSM TA has accuracy in
the vicinity of 550 metres. So that
would result in a position accuracy
of over 100m at the 67% level.
Similarly, Figure 4 shows the results
of using only two satellites and 2,3
and 4 range measurements. In this
case the position accuracy follows a
similar curve and the accuracy of the
location is highly dependant on the
accuracy of the range measurements.
.gif)
