Integration of GPS with Other Sensors and Network Assistance - Understanding GPS: Principles and Applications - page 486

I
Q

δρ

Receiver
correlator

Code loop
detector

Kalman
filter

Prefilter
(Optional)

ρ est

δ v est

v

−

+

Code loop
NCO

INS

u T

−

v s

(For each channel)

Figure 9.14 Integrated tracker/navigator block diagram.

tor for the set of satellite code phases and Dopplers). This essential observation does
not depend upon the inertial augmentation of GPS. Applications of this high-level
concept have therefore arisen in commercial applications of GPS [14, 15], where the
INS is absent. Such implementations are sometimes referred to as vector tracking,
since the individual tracking loops within the GPS receiver are no longer independ-
ent—they are coupled through their response to the position and velocity of the host
vehicle, as well as the common clock errors.
Returning to Figure 9.14, it can be seen that this architecture removes the con-
ventional code tracking loop, replacing it with a single loop that is closed through
the Kalman filter. A byproduct of this new architecture is a solution for the stability
problem: without the separate loop that produces an unmodeled measurement error
correlation for the Kalman filter, a well-designed filter will not cause stability prob-
lems in this aided configuration. Note the optional prefilter. The receiver correlator
outputs in-phase (I) and quadrature (Q) correlations at a rate typically ranging from
a few milliseconds to up to 20 ms. This is obviously an extremely high rate for
Kalman filter execution: one solution to this problem is to average the outputs of the
detector up to a more typical processing rate for a Kalman filter (e.g., once per sec-
ond). Recent applications of ultratight coupling have made use of reduced-order
Kalman prefilters to feed the Kalman tracking and navigation filter (the centralized
filter) in a federated filter architecture [16, 17]. Alternatively, a multirate mechani-
zation for the Kalman filter can be used, where the state propagation and update
occur at the highest rate at which the code loop detector output is generated, but
gain calculation, covariance propagation, and update (where the bulk of the Kalman
computations occur) are performed at a more typical lower rate (e.g., 1 Hz).
Simulations are used to compare the performance of the integrated, or tightly
coupled and partitioned, designs in [13]. Although the improvements in its response
to increasing noise levels are not significant (the first simulation case considered in
[13]), substantial improvements are realized when significant dynamics are com-
bined with near threshold noise levels (the second simulation case considered). The
results are somewhat intuitive: in the first case, since both designs are able to adapt
to an increasing noise level by lowering the effective aided receiver bandwidth, their
performance is quite similar. In the second simulation case, it is in fact the recogni-
tion of the receiver oscillator's g sensitivity by the integrated design that leads to the
substantial performance improvement. Recall from Section 9.2.1 that tracking the
dynamics of the local oscillator is a requirement that sets the floor on the aided