Professional Documents
Culture Documents
D.-J.Jwo
Abstract: For a GPS receiver, decreasing the receiver tracking loop bandwidth reduces the
probability of loss of lock if there are no vehicle dynamics. However, reduced bandwidth increases
tracking errors due to dynamics. Beyond a certain limit it causes a serious degradation in the
dynamic tracking performance. Therefore, there is involvement of a tradeoff between two
opposing considerations: narrow tracking loop bandwidths are desired for filtering noise due to
thermal effects, but wide tracking loop bandwidths are desired to permit tracking of vehicle
dynamics. Optimal tracking loop bandwidths, which yield the minimum errors in a certain
dynamics environment, are first investigated. The linear Kalman filter is employed as the optimal
estimator. The covariance for the arbitrary gain model is solved and applied to the sensitivity
analysis for investigating error growth due to incorrect noise level estimate. Theoretical results are
verified by numerical simulation, and results from both approaches are in very good agreement.
I H(J'LII)
1' = [ ~ ( j w ) ~ ( - j o ~ ) l (4)
Fig. 1 Rloclc diagram ?t'.yimpl$ed GPS receiver tmclcing loop,^ 191
Analytical solution for the tracking loop bandwidths, by
implementing eqn. 3, might be tedious. R. S. Philips' table
shown in Fig. 2. Thc closed-loop transfer function of the of contour integrals can be employed for simpler computa-
tracking loop is tion
Earlynzinus-late power
Dot-product
r B,d
b
Fig. 2 iMulhematicul modeh of'linenri.wd tracking loop where otDLL is the standard deviation of tracking error in
a In terms of original circuit parameters units of PRN chips, B, is thc single-sidcd code tracking-
0 In terms of natural frequency loop bandwidth in Hz, T is thc predetection integration
242 IEE Pmc.-Radar: Soiior h'avig., h l , 145, No. 4,Airgzist 2001
u(s) = a,s" + a,,- + + a"
,.Fl ' ' '
c:,
I , =-
2a"(l ,
time (PIT) in scconds, U' is the early-to-late correlator For the DLL the thermal noise is indepcndent of
spacing normalised with respect to one chip tracking-loop order; for the PLL, the thermal noise jitter
is not directly dependent on the loop order, too [4]. The
U' = I for time-shared tau-dithcrcd early-late correlator C/NOs of the GPS with good signal power typically range
= f for dedicated early and late correlator from 35-55 dB-Hz, so tracking errors generally run on the
lower end of the range. For the nominal range of C/No
and clno is the carricr-to-noise ratio value (35 dB-Hz and above) the squaring loss is usually consid-
ered negligible. Typically, the unaided code tracking-loop
clno = (SNR)(BL)(ratio-Hz) bandwidths are in the range of 1-4 Hz and unaided carrier
CllY, = 10 log,o(c/no) (dB-Hz) tracking-loop bandwidths are in the range of 5-1 5 Hz. Fig.
5 shows one-sigma thermal jitters of the early-minus-late
The second term in parentheses is known as the 'squaring noncoherent DLL and the Costas-type PLL, respectively.
loss', Most GPS receivers employ a noncohercnt delay- The maxiinum bandwidths allowed for maintaining lock in
lock loop (DLL) for code tracking and a Costas-type various GIN, environments arc shown in Fig. 6.
phase-lock loop (PLL) for tracking the Doppler-shifted
carrier. The thermal noise jitter for the commonly used
tau-dithered early-late DLL with d = I , is
3.2 Dynamic stress errors
A small steady-state crror is usually desired and is consid-
ered as the criterion of good tracking performance. The
transfer function representing the tracking crror is
- [x(+-
c/no
1
2Tc/no
(degrees) (Xb)
The loop order is sensitive to the same order of dynamics,
e.g. first order to velocity stress, second order to accelera-
tion stress, and third order to jerk stress. The first-order
loop is suitable for a user position that varies in a random
walk manner (white noise velocity); the second-order loop
is suitable for a user velocity that varies in a random walk
The GPS has a 50 Hz navigation data message bit rate; the manncr (white noise acceleration); the third-order loop is
predetection integration timc i s usually the period of a suitable for a user acceleration that varies in a random walk
navigation data bit, 20 ins. The wavclcngths A, for L , and manner (white noise jerk). For example, the dynamic
L, carricrs, due to the link frequencies 1575.42 and condition resulting in a second-order loop tracking error
1227.6 MHz, are 0.193 and 0.2442 m, respectively. i s constant acceleration. There is no steady-state transient
IEE Pmc.-Rudur: Sonar Nuvavig.. Vol. 148, ,!Vo 4,Azrg~ist2001 243
Loop order 1st order 2nd order (active lag-lead) 3rd order
~
K
s+K
BL K
-
4
Q),
5 + w,
BL a2 +I ah2 + a2 - b
4 4a 4(ab - 1)
Fig. 4 Loop.filtrr.s, closed-loop tmn.Tfir,functions, and equivulent noise bandwidths E , (11~)for different ordim uf receiver tracking loops
error for a second-order loop with constant velocity input 3.2.7 Code tracking: Peak errors caused by a range
and is optimal for constant acceleration input. The limita- rate, range acceleration, and jerk step in the third-order
tion exists during the transient period when going from no loop can be approximated by the steady-state response of a
acceleration to constant acceleration. first-, second-, and third-order loop caused by a range rate,
From eqn. 9 the error transfer functiols due to user rangc acceleration, and jerk step function, respectively. The
dynamics are three-sigma steady-state errors can be written as (in units
of chips)
d2z/dt2 7(')
-
-
CO;
= w; (second order) (13h)
=A f
_ . (second order)
4
=-
47 (third order) ( 14c)
m,l
a
where Afand c$~, are in units o f Hz and cycles. From the
20 -
rule-of-thumb 14, 51, the three-sigma code noise should not
exceed 1/2 chip of thc chip length and the carrier noise
should not cxcccd 45' of the wavelength for maintaining
16-
... ............
. \\ threshold signal lock. Therefore the onc-sigma code noise should not
exceed 1/6 chip of thc chip length and the carrier noise
m
U
should not exceed 15" of thc wavelength, i.e.
12 -
0- and
r
4-
opLL= d G nf + o: + dJ3 -< I5 (degrees) (1 6)
Po=
1000
0
0
0
1000
0
:]
1000
function and handwidth for thc second-order loop are
Based on these assumptions the optimal bandwidth respectively. The minimum covariance and Kalinan gain
B, = 12.2258 Hz, and consequently, thc minimum matrices, respectively, are
measurement noise R = 105.2581nz2 are obtained
(explained in Section 5 ) . Fig. 7 shows the typical profiles
of Kalman gains and error variances for a third-order GPS
receiver tracking loops.
r
8-
H(.r) =
+
K , s 2 +K2s Ki
-
bf,,,.? awts
~ + + (U; ,
BL =
+
ab2 a2 - h
w,
1200 4(ab - 1)
respectively. The minimum covariancc and Kalinan gain
1 000
80C
N
E
ai
e
._m
60C
i3
40C
20c
time, s
b
-BL -
_ 1
- (first order)
Wn 4
--
-
a' +I (second order)
-
4a
+
ab2 a2 - b
(third order)
=0.3024 , ) (first order) (240)
4(ah - 1) ~
2c/no I+--
Tcln, :A
If defining the parameter CI = B,/cri,,, the dynamic stress
errors with minimum mean square error can then be
obtained utilising a and eqn. 14
A!
= - = 0.25-
A? (first order) (1 9a)
*<,
c01, BL =""I54 ( 2,
' (second order) (24b)
'fx 0.2812-A !
~
- $ (i)3$= x 0.5787-Af
52
(third order) (19c)
-
P
U
2
'? 60
80
? 0.01 ~
c I\ / Top
tion parameters; it is a priori knowledge of message
statistics that permits threshold reduction. The adaptive
loop performs selection of the desired bandwidth from a
m table of precomputed values based on the noise level C,",j
9 estimate, and control logic to modify the bandwidths and
2 0.00; prevent transients. As shown before, received C/No is the
key parameter in system performance analysis. In some
25 situation, there is an incorrect estimate of noise level; in
k' some applications, there is no need to adjust the loop to
o'ool0
30
35 40, 45.
\ \
I
attain exactly the best performance. The sensitivity analy-
sis usually involves in how the incorrect parameters
0 2 4 6 8 10 (departure from the design point) selected influence on
bandwidth, Hz the error growth.
a
The filter gain in the error covariance eqn. 17 does not
appear explicitly. Another form of error covariance propa-
gation is described now. The error covariance relationships
for a discrete filter with the same structure as the Kalman
filter, but with an arbitrary gain matrix are written as
2'ol
1.5
1 .o
LB
,
which can be described in a single differential equation for
20 25 30 35
the continuous filter
CIN,, dB.Hz
b
P = (F - KH)P + P(F -K + +
H ) ~ G Q G ~ K R K ~ (29)
Fig. 9 Total tracking err07 peTformance and optimal handwidth for
minimum errors and handwidth lower hound for maintaining lock in
various carrier-to-noise ratio environments.for third-order PLL (L, carrier) This equation defines the error covariance for the filter with
a general filter gain matrix K , which can be solved for the
covariance of an arbitrary gain model. The sensitivity
Using the same procedure for deriving the optimal PLL analysis can be conveniently implemented by using this
loop bandwidth, we have representation. Taking the partial derivative of P, with
respect to K, using aPm/aK=0 for a minimum gives the
same result as eqn. 17.
If the fixed-gain matrix K of a filter has been designed
for particular values of Q and R , the steady-state error
covariance will vary linearly with the actual process noise
spectral density or measurement error spectral density. If
the actual noise variances are assumed fixed and the design
and values of Q and R are varied, quite different curves result.
Any deviation of the design variances, and consequently K ,
from the correct values will cause an increase in the filter
error variance (a consequence of the optimality of the
filter). More information on sensitivity analysis is
contained in [ 111.
The error covariance for the filter with arbitrary gain
model, can be obtained by using eqn. 29
First order
Figs. 8 and 9 provide examples on the tracking perfor-
mance for the third-order DLL and PLL, respectively.
Since the carrier-tracking accuracy is excellent even at P = -2KPfK2R+Q
wide bandwidths, its velocity measurements can be used
to aid the code-tracking loop. Using carrier-aided code Second order
techniques makes the dynamic stress error for the code-
tracking loop a small-order effect that usually can be
assumed negligible. If inertial information is delivered to
the tracking loops, the required minimum bandwidth can
be reduced without the penalty of increasing dynamics
errors [6, 9 , 101.
248 I&? Proc.-Radul; Sonar Navig., Vol. 148, No. 4, August 2001
incorrect estimate of the CIN, (i.e. design CIN,, deviating
from the actual ratio). The optimal bandwidth
(BL)DLLapfimum is first determined by a given estimate of
CjN,. With knowledge of CjN, and the related optimal
bandwidth, the measuremcnt noise variance R , and conse-
quently the Kalman gain matrix K , are determined. An
incorrect estimate of GIN, gives an incorrect gain matrix
and results in error growth.
The information flow chart for the sensitivity analysis is
shown in Fig. 10. Procedures for theoretical approach are
By setting P = 0, the solutions are found to be as follows. After input of design and actual C,”, and
dynamics information, the optimal bandwidths for design
P, =-
Q + K*R (first order) (33) and actual environments, i.e. (B,),,,, and (B,),,,, , are
2K subsequently determined by eqns. 24 (for DLL) and 26 (for
PLL). Once the optimal bandwidths are determined, noise
information, i.e. Rdcsignand R,,,, can be estimated from
KtQ + K2Q + K:R eqns. 7 (for DLL) and 8 (for PLL). Consequently the
optimal gain matrices, i.e. KdZsjgll
and K,,,, are determined
(second order) (34) by eqns. 17 and 18. Finally, the error covariance is
determined by eqns. 33 and 35. Error variance obtained
by actual measurement noise R,,, with K,,, (based on
(third order) ( 3 5 ) correct estimate on C,”,) has a minimum value while error
variance obtained by actual measurement noise R,.,,, with
where
P I ] , = K:K2K3R - KlKiR +K:KjR +KIQ
P12, = KI(K.K,R
P13m
+ KI Q>
( K j R + Q)(K1K2 - K3)
350L
300
+
P , ~ , = K;K,R K:Q K:R K ~ Q + +
p23 oo = K2 (K2KiR + K? Q)
P33, = K & R + K ~ K ~ Q + K I K ~ Q K&Q
- 5 I \ \ -
3.5
design meas. noise, calculated by
eqns. 7 for DLL
actual meas. noise, -
& 8 for PLL
hue %
;; 3.0
I
N
E
I 12 2.5 ai
.-
design gain matrix Kalman gain matrix J
&rue 2.0
s!
z
1.5
I I
IEE Prac.-Radar. Sonar Navig., Yol. 148. No. 4, August 2001 249
Kdesrgn(duc to originally incorrect estimate on C/N,) theoretical and simulation results are in very good agree-
results in error growth. The numerical siinulation essen- ment. Implementation on the other orders of loops can be
tially follows the same procedures as theoretical approach done without difficulty based on the same procedure.
except that random numbers generated by Matlab are
employed. The error variances are determined by analysing
8 References
the output data from the filter. Figs. 11 and 12 give
example on the sensitivity for the third-order DLL (C/A 1 SPILKER, J.J.: ‘Fundamental of signal tracking theory’ in PARKTN-
code) and PLL ( L , carrier), respectively. The theoretical SON, B.W, SPILKER, J.J., PENINA, A., andENGA, P. (Eds.): ‘Global
positioning system: theory and applications’, vol. 1 (AIAA, Washington
results, represented by solid lines, and the simulation UC, 1996), Chap. 7
results, represented by symbols, are in very good agree- 2 VAN DIEREDONCK, A.J.: ‘GPS receiver’ in PARKINSON, B.W.,
ment. SPILKER, J.J., PENINA, A., and ENGA, P. (Eds.): ‘Global positioning
system: theory and applications’, vol. 1 (AIAA, Washington DC, 1996),
Chap. 8
3 BRAASCH, M.S., and VAN DIEREDONCK, A.J.: ‘GPS receiver
7 Conclusions architecture and measurements’, Pmc. IEEE, 1997, 87, (I), pp. 48-64
4 WARD, PW: ‘Satcllite signal acquisition and tracking’ in KAPLAN,
Analysis of optimisation and sensitivity of GPS receiver E D . (Ed.): ‘Understanding GPS: - principlcs
. and anplications’
.. (Artech
House, MA, 19Y6), Chap.-5
tracking loops has been presented. The maximum band- 5 WARD, P.W.: ‘GPS receivcr RF interfcrence monitoring, mitigation, and
width threshold is essentially governed by the thermal analysis techniques’, Nuvig. J: Ins(. Nuvig., 1994-1995,41, (4), pp. 367-
791
noise jitter; however, the minimum bandwidth threshold 6
~ , I
250 IEE Proc.-Radac Sonar Navrg., Vol. 14X, No. 4, Augu:urt 2001