See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/255027322

Carrier Tracking Loop using the Adaptive Two-Stage Kalman Filter for High
Dynamic Situations

Article  in  International Journal of Control Automation and Systems · December 2008

CITATIONS READS

36 468

3 authors, including:

Gyu-In Jee Jong-Hwa Song

Konkuk University Konkuk University
209 PUBLICATIONS   3,399 CITATIONS    13 PUBLICATIONS   129 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Anti-jamming&spoofing simulator View project

Fault-Tolerant Navigation System Design View project

All content following this page was uploaded by Jong-Hwa Song on 23 February 2015.

The user has requested enhancement of the downloaded file.

948 International Journal of Control, Automation, and Systems, vol. 6, no. 6, pp. 948-953, December 2008

Carrier Tracking Loop using the Adaptive Two-Stage

Kalman Filter for High Dynamic Situations
Kwang-Hoon Kim, Gyu-In Jee and Jong-Hwa Song

Abstract: In high dynamic situations, the GPS carrier tracking loop requires a wide bandwidth to
track a carrier signal because the Doppler frequency changes more rapidly with time. However, a
wide bandwidth allows noises within the bandwidth of the tracking loop to pass through the loop
filter. As these noises are used in the numerical controlled oscillator (NCO), the carrier tracking
loop of a GPS receiver shows a degraded performance in high dynamic situations. To solve this
problem, an adaptive two-stage Kalman filter, which offers the NCO a less noisy phase error, can
be used. This filter is based on a carrier phase dynamic model and can adapt to an incomplete
dynamic model and a quickly changed Doppler frequency. The performance of the proposed
tracking loop is verified by several simulations.

Keywords: Adaptive Kalman filter, carrier tracking loop, GPS receiver, high dynamic.

1. INTRODUCTION proposed the carrier tracking loop using the Kalman

filter, which offers the NCO a less noisy phase error.
In a GPS receiver, the input is the GPS signal and a The linear Kalman filter or the extended Kalman filter
phase lock loop (PLL) must track this signal. However, was used and these filters were based on a carrier
the GPS signal is a bi-phase coded signal. The carrier phase dynamic model [2-6]. These Kalman filters
and code frequencies change due to the Doppler effect, estimated the carrier phase error between the
which is caused by the motion of the GPS satellite as incoming signal and the NCO by using an estimate of
well as from the motion of the GPS receiver. In order the Doppler frequency of a carrier phase dynamic
to track the GPS signal, the C/A code information model. However, under high dynamic situations, the
must be removed. As a result, it requires two PLLs to Kalman filter may not track the Doppler frequency
track a GPS signal. One is to track the C/A code and because the Doppler frequency changes more rapidly.
the other is to track the carrier frequency [1]. This As such, this paper proposes a new carrier tracking
paper only considers the latter. loop using an adaptive two-stage Kalman filter based
The tracking performance of a PLL is affected by on a carrier phase dynamic model. This adaptive filter
several factors, such as the signal-to-noise ratio (SNR), was proposed by Kim and coauthors [7] and could
dynamic stress, thermal noise, and so forth. However adapt to an incomplete dynamic model and a quickly
specially, a PLL does not give a good performance changed Doppler frequency. Section 2 shows the
under high dynamic situations. In high dynamic concept of the carrier tracking loop using an adaptive
situations, the Doppler frequency changes more Kalman filter. Section 3 provides the adaptive two-
rapidly with time. Therefore in such cases, the carrier stage Kalman filter. The derivation of a filter equation
tracking loop of a GPS receiver requires a wide is omitted. Finally, the performance of the proposed
bandwidth to track a carrier signal. However, a wide tracking loop is verified by several simulations in
bandwidth allows noises within the bandwidth of the Section 4.
tracking loop to pass through the loop filter. As these
noises are used in the numerical controlled oscillator 2. CARRIER TRACKING LOOP DESIGN
(NCO), the carrier tracking loop of a GPS receiver
shows a degraded performance under high dynamic 2.1. Costas carrier tracking loop
situations. To solve this problem, several researchers A GPS receiver can track incoming GPS signals by
__________ using the PLL. The PLL replicates the incoming
Manuscript received October 16, 2007; revised June 12, signal’s frequency and phase. A Costas loop is one
2008; accepted September 9, 2008. Recommended by among these phase lock loops used for carrier phase
Editorial Board member Poo Gyeon Park under the direction recovery. Normally, a Costas carrier tracking loop is
of Editor Young Il Lee.
used in GPS receivers because the 50Hz navigation
Kwang-Hoon Kim, Gyu-In Jee, and Jong-Hwa Song are
with the Department of Electronic Engineering, Konkuk message data modulation signal remains after the
University, Hwayang-dong, Gwangjin-gu, Seoul 143-701, carrier and code signals have been wiped off the
Korea (e-mails: {kwanghun, gijee, hwaya}@konkuk.ac.kr). incoming GPS signal [8]. Fig. 1 shows a block
 Carrier Tracking Loop using the Adaptive Two-Stage Kalman Filter for High Dynamic Situations 949

Pre-integration  ∆t 5 ∆t 4 ∆t 3 
Filter
SI  
 20 8 6 
Loop  ∆t 4 ∆t 3 ∆t 2 
Input
signal
NCO
Filter
Discriminator
+ Qa  , (2)
900 shift  8 3 2 
SQ  ∆t 3 
∆t 2
Pre-integration  ∆t 
Filter  6 2 

Fig. 1. The Costas loop. where the system matrix and the spectral intensity in
diagram of a Costas loop. continuous time are
1 ∆t ∆t 2 2 Qθ 0 0
2.2. Carrier phase dynamic model  
As mentioned in the introduction, several re- Φ k ,k −1 = 0 1 ∆t  , Qc =  0 Qd 0  . (3)
searchers proposed the carrier tracking loop using the 0 0 1   0 0 Qa 
Kalman filter based on a carrier phase dynamic model  
[2-6]. A carrier phase model is as follows [6]. An Arctan discriminator is used in this paper. The
measurement is obtained from this discriminator
θe,k  1 ∆t ∆t 2 2 θe,k −1   ∆t 
    
∆t   f d ,k −1  −  0  f dNCO
output θemea
,k −1 = arctan SQ S I .( )
SQ and S I are
 f d ,k  = 0 1 ,k −1 + Wn ,
 f a,k  0 0  ouputs of the pre-integration filters.
   1  f a,k −1   0 
 The measurement can be modeled by θ eave as
(1)
where θ e,k is the carrier phase error between the ∆t
1
θeave ∫ θe,k −1 +  fd ,k −1 − fd ,k −1  τ
NCO
=
incoming signal and the output of the NCO, f d ,k is ∆t 0 (4)
the Doppler frequency in the incoming signal, f a,k 1 
+ f a,k −1τ 2  dτ
is the change rate of Doppler frequency caused by the 2 
acceleration along the line of sight (LOS) between the 1 1
= θe,k −1 + f d ,k −1 − f dNCO 
, k −1  ∆t + f a ,k −1∆t 2
satellite and the receiver, f dNCO,k is the Doppler 2 6
frequency replicated by the NCO and ∆t = 0.001 is As a result, the measurement equation is as follows.
the update period and pre-integration time.
θe,k −1 
Wn = [Wθ Wd Wa ] is a noise vector which
T
 ∆t∆t 2    ∆t NCO
θemea
,k −1 = 1   f d ,k −1  − f d ,k −1 + Vk −1
consists of discrete Gaussian white noise sequences.  2 6  2
 f a,k −1 
Assuming that these noise sequences are independent  
and the spectral intensity of the continuous time white (5)
noise process of these noise sequences are Qθ , Qd where Vk ~ ( 0, Rk ) is a Gaussian white noise
and Qa , respectively. Qθ is caused by the receiver sequence.
clock phase bias. Qd is caused by the receiver clock
2.3. Costas loop using an adaptive filter
frequency drift. Qa is caused by the acceleration The carrier tracking loop of a GPS receiver requires
along the LOS. The covariance Q which corresponds a wide bandwidth to track a carrier signal under high
to the discrete noise process can be calculated as dynamic situations. However, a wide bandwidth
follows [6]. allows noises within the bandwidth of the tracking
∆t loop to pass through the loop filter. As a result, the
Q = ∫ Φ k ,k −1Qc ΦTk ,k −1dt carrier tracking loop shows a degraded performance
0
under high dynamic situations. To solve this problem,
 ∆t 3 ∆t 2  the Kalman filter can be used as indicated in Fig. 2.
 0
 ∆t 0 0   3 2  This Kalman filter estimates the output of discrim-
 ∆t 2  inator and a Doppler frequency. That is, the Kalman
= Qθ  0 0 0  + Qd  ∆t 0 filter gives the loop filter the less noisy carrier phase
 2  error. Of course, this filter is based on a carrier phase
 0 0 0   0 0 0 dynamic model. However, a Doppler frequency of a
 
  carrier phase dynamic model is rapidly changed under
950 Kwang-Hoon Kim, Gyu-In Jee and Jong-Hwa Song

Pre-integration known. Nevertheless, in most cases, they are unknown.

Filter If this information is incomplete, the performance of
SI
the TKF may be degraded or diverged. To solve this
Loop problem, the TKF has to be adapted to an environment
Input NCO AKF Discriminator
signal Filter
of incomplete random bias information. This problem
900 shift SQ was solved by the adaptive two-stage Kalman filter
Pre-integration (ATKF), which was proposed by Kim and coauthors
Filter
[7]. Furthermore, the adaptive two-stage nonlinear
filter expanded from the ATKF was applied to the
Fig. 2. Costas loop using an adaptive Kalman filter.
INS-GPS loosely coupled navigation system with
unknown fault bias and showed a good performance
high dynamic situations. Thus, to estimate a Doppler [17]. The concept and the derivation of the ATKF are
frequency, an adaptive Kalman filter noted in Section explained in another paper [7]. As such, this section
3. Equations (6-8) are given for this adaptive Kalman only briefly introduces the ATKF. Consider the
filter. following linear discrete-time stochastic system
represented by
 θe,k  1 −∆t   θe,k −1 
 NCO  =    NCO  xk +1 = Φk xk + Bk bk + wkx , (9a)
 f d ,k  0 1   f d ,k −1 
(6) bk +1 = Ak bk + wkb , (9b)
 ∆t 2   f
+  ∆t  d ,k −1  + Wθ 0
, zk = H k xk + Dk bk + vk , (9c)
2  f  

 a ,k −1   0 0 
 0 0  where xk is the n × 1 state vector and zk is the
 f d ,k  1 ∆t   f d ,k −1  Wd 0  m × 1 measurement vector. Φk and H k are state
f =  + , (7) transition matrix and observation matrix respectively.
 a,k  0 1   f a,k −1   0 Wa 
bk is the p × 1 bias vector with unknown magnitude.
 ∆t   θe,k −1  All matrices have the appropriate dimensions. The
θemea
,k −1 = 1 −  
 2   f dNCO
, k −1 
 noise sequence wkx , wkb , and vk are zero mean
(8)
 ∆t ∆t 2   f d ,k −1  uncorrelated Gaussian random sequences with
+   + Vk −1.
 2 6   f a,k −1   x  x T  x
  wk   w j   Qk 0 0
     
E   wkb   wbj   =  0 Qkb 0  δ kj , (9d)
 
3. ADAPTIVE TWO-STAGE   vk   v    0 0

Rk 
KALMAN FILTER     j   
  
 
The well-known Kalman filtering has been widely where Qkx > 0, Qkb > 0, Rk > 0 and δ kj is the
used in many industrial areas. This Kalman filtering
technique requires complete specifications of both Kronecker delta. The initial states x0 and b0 are
dynamical and statistical model parameters of the assumed to be uncorrelated with the white noise
system. However, in a number of practical situations, processes wkx , wkb and vk . Assume that x0 and
these models contain parameters that may deviate
b0 are Gaussian random variables with
from their nominal values by unknown constant or

( )( )
unknown random bias. To solve this problem, a new  T
procedure for estimating the dynamic states of a linear E [ x0 ] = x0* , E  x0 − x0* x0 − x0*  = P0x > 0, (9e)
 
system in the presence of unknown constant bias was
( )( )
 T
suggested by Friedland [9]. This filter is called the E [b0 ] = b0* , E  b0 − b0* b0 − b0*  = P0b > 0, (9f)
two-stage Kalman filter (TKF). Many researchers  
have contributed to this problem. Ignagni and Keller
( )( )
 T 
suggested the alternate derivation of Friedland’s bias E  x0 − x0* b0 − b0*  = P0xb . (9g)
filtering technique in the case of unknown constant  
bias [10,11]. Recently, the TKF to consider not only a Definition 1: A discrete-time adaptive two-stage
constant bias but also a random bias has been Kalman filter is given by the following coupled
proposed through several papers [12-16] with the difference equations when the information of the
representative optimal TKF being proposed by Hsieh linear stochastic system given by (9) is incomplete [7].
[15]. This optimal TKF assumes that the dynamic


model and the noise covariance of a random bias are xk ( − ) = xk ( − ) + U k bk ( − ) , (10a)
 Carrier Tracking Loop using the Adaptive Two-Stage Kalman Filter for High Dynamic Situations 951

xk ( + ) = xk ( + ) + Vk bk ( + ) , (10b) Also, the initial conditions are

Pkx* ( − ) = Pkx* ( − ) + U k Pkb* ( − )U kT ,

( )
−1
(10c) V0 = P0xb P0b , x0 ( + ) = x0* − V0b0* , b0 ( + ) = b0* , (14a)
Pkx* ( + ) = Pkx* ( + ) + Vk Pkb* ( + )VkT , (10d)
P0b* ( + ) = P0b , P0x* ( + ) = P0x − V0 P0bV0T . (14b)
where Ak and Qkb
are partially known. The The ATKF of Definition 1 is based on the two-stage
modified bias free filter is Kalman filter of Heish [15] and the adaptive fading
Kalman filter [7]. This ATKF gives a solution for the
xk ( − ) = Φ k −1 xk −1 ( + ) + uk −1 , (11a) system with the unknown random bias on the
Pkx* ( − ) = λkx Φ k −1Pkx−*1 ( + ) ΦTk −1 + Qkx−1  , (11b) assumption that the stochastic information of the
  random bias is incomplete. The carrier tracking loop
−1
K kx* = Pkx* ( − ) H kT  H k Pkx* ( − ) H kT + Rk  , (11c) under high dynamic situation must track unknown
  Doppler frequency. Thus, this paper treats unknown
( )
Pkx* ( + ) = I − K kx* H k Pkx* ( − ) , (11d) Doppler frequency as the unknown random bias of
(9b). The ATKF can track unknown Doppler
ηkx = zk − H k xk ( − ) , xk ( + ) = xk ( − ) + K kx*ηkx , (11e) frequency, which changes more rapidly with time.
Ckx = H k Pkx* ( − ) H kT + Rk , (11f) 4. SIMULATION
k
1
Ckx = λkx Ckx = ∑ ηixηixT ,
M − 1 i = k − M +1
(11g) As mentioned in the introduction, the performance
of a GPS tracking loop with a low PLL bandwidth is
{
λkx = max 1, trace Ckx ( ) ( )} ≥ 1,
trace Ckx (11h) better than that with a high PLL bandwidth, because a
high bandwidth allows noises within the bandwidth of
and the bias filter is the tracking loop to pass through the loop filter and
these noises degrade the tracking performance.
bk ( − ) = Ak −1bk −1 ( + ) , (12a) Therefore, the general method is to use a low band-
width. However, a GPS tracking loop with a low PLL
Pkb* ( − ) = λkb  Ak −1Pkb−*1 ( + ) AkT−1 + Qkb−1  , (12b)
  bandwidth may not track a fast Doppler frequency. As
b* b* T  x* T a result, a GPS tracking loop uses a high PLL
K k = Pk ( − ) N k H k Pk ( − ) H k + Rk
 bandwidth in the case of high dynamic situations. But,
−1
(12c) if the PLL loop using the ATKF is considered in a
+ N k Pkb* ( − ) N kT  ,
 GPS tracking loop, we can expect that a GPS tracking
b* b* b* loop will show a good performance even though a low
Pk ( + ) = I − K k N k Pk ( − ) ,
  (12d)
  bandwidth is used. Of course, in the case of a high
ηkb = zk − H k xk ( − ) − N k bk ( − ) bandwidth, the general PLL loop and the PLL loop
(12e) with the ATKF demonstrate a similar performance
= ηkx − N k bk ( − ) , because two PLL loops can track a Doppler frequency.
To verify the performance of a carrier tracking loop
bk ( + ) = bk ( − ) + K kb*ηkb , (12f) using an ATKF, the 3rd order PLL with and without
Ckb = H k Pkx* ( − ) H kT + Rk + N k Pkb* ( − ) N kT , (12g) the ATKF are simulated under high dynamic situation
k
as follows. Simulation conditions and PLL
1
Ckb = λkb Ckb = ∑
M − 1 i =k − M +1
ηibηibT , (12h)
Table 1. Simulation conditions.

{
λkb = max 1, trace Ckb ( ) ( )} ≥ 1
trace Ckb (12i)
L1 Carrier Frequency
Code Frequency
1575.42 MHz
1.023 MHz
Signal-to-Noise Ratio -20dB
with the coupling equations
Intermediate Frequency 1.4054 MHz
N k = H k U k + Dk , Vk = U k − K kx* N k , (13a) Sampling Frequency 17.5 MHz
−1  Simulation Time 1572 msec

U k = U k  I − λkb Qkb−1  Pkb* ( − )   , (13b)
    Table 2. PLL configuration.
U k = ( Φk −1Vk −1 + )
Bk −1 Ak−−11 , (13c) Order of the PLL Three
uk = (U k +1 − U k +1 ) Ak bk ( + ) , (13d)
Discriminator Arctan
PLL Bandwidth 7Hz, 15Hz
Qkx = Qkx + U k +1QkbU k +1. (13e) Pre-integration Time 1 msec
952 Kwang-Hoon Kim, Gyu-In Jee and Jong-Hwa Song

configuration are shown in Table 1 and 2. For Fig. 4(a) shows that the 3rd order PLL with the ATKF
example, the GPS L1 signal (1575.42 MHz), -20dB gives a small Doppler frequency error as compared
SNR, the 3rd order PLL, an Arctan discriminator, and with Fig. 3(a). But as time goes by, the Doppler
1ms pre-integration time are used in simulations. frequency error of Fig. 4(a) becomes similar to that of
Simulations are performed by MATLAB software. Fig. 3(a). Fig. 4(b) shows that a discriminator gives
User trajectory is a half-circle with initial velocity and less noisy output than Fig. 3(b).
centripetal acceleration, which are 50m/s and 100m/s2,
respectively. 4.2. Low bandwidth
Secondly, the bandwidth of the PLL is selected as
4.1. High bandwidth 7Hz. Fig. 5(a) shows that the 3rd order PLL without
To easily track a signal, firstly the bandwidth of the the ATKF gives a large Doppler frequency error
PLL is selected as 15Hz. Fig. 3(a) shows that the 3rd during initial interval and also cannot track one
order PLL without the ATKF tracks all channels but channel. A discriminator of Fig. 5(b) gives more noisy
gives a large Doppler frequency error during the output. On the other hand, Fig. 6. shows the result of
initial interval. Fig. 3(a) also presents that the Doppler the 3rd order PLL with the ATKF. Fig. 6(a) indicates
frequency error is about 40 degrees as more time that the 3rd order PLL with the ATKF gives a small
passes. Here the Doppler frequency error means a Doppler frequency error as compared with Fig. 5(a)
difference between a true Doppler frequency and an and also tracks all channels without fail. Fig. 6(a)
estimated Doppler frequency replicated by the NCO. shows that the Doppler frequency error is about 25
Fig. 3(b) indicates that a discriminator gives more degrees as time goes by. Fig. 6(b) presents that a
noisy output during initial interval. On the other hand, discriminator gives less noisy output than Fig. 5(b).

(a) Doppler frequency error (a) Doppler Frequency Error

150 300 =

100 200
50
100
0
0
-50
-100
-100
-200
-150
-200 -300

-250 -400

-300 -500
-350 -600 =
0 500 1000 1500 0 500 1000 1500

(b) Discriminator output (b) Discriminator output =

1.5 1.5

1 1

0.5 0.5

0 0

-0.5 -0.5

-1 -1

-1.5 -1.5
=
0 500 1000 1500 0 500 1000 1500

Fig. 3. The 3rd PLL without the ATKF (BW:15Hz). Fig. 5. The 3rd PLL without the ATKF (BW: 7Hz).
(a) Doppler Frequency Error
(a) Doppler Frequency Error =
100 150
80 100
60
50
40
20 0
0
-50
-20
-40 -100

-60 -150
-80
-200 =
-100 0 500 1000 1500
0 500 1000 1500

(b) Discriminator output (b) Discriminator output =

1.5 1.5

1 1

0.5 0.5

0
0

-0.5
-0.5
-1
-1
-1.5 =
-1.5 0 500 1000 1500
0 500 1000 1500

Fig. 4. The 3rd PLL with the ATKF (BW: 15Hz). Fig. 6. The 3rd PLL with the ATKF (BW: 7Hz).
 Carrier Tracking Loop using the Adaptive Two-Stage Kalman Filter for High Dynamic Situations 953

As a result, the 3rd order PLL with the ATKF gives a estimator,” IEEE Trans. on Automatic Control,
small Doppler frequency error and a good tracking vol. 26, pp. 746-750, 1981.
performance as compared with the 3rd order PLL [11] J. Y. Keller, L. Summerer, and M. Darouach,
without the ATKF. “Extension of Friedland’s bias filtering
technique to discrete-time system with unknown
6. CONCLUSIONS inputs,” International Journal of Systems
Science, vol. 27, no. 12, pp. 1219-1229, 1996.
The carrier tracking loop of a GPS receiver demon- [12] J. Y. Keller and M. Darouach, “Optimal two-
strates a degraded performance under high dynamic stage Kalman filter in the presence of random
situations. The reason is that a carrier phase error bias,” Automatica, vol. 33 no. 9, pp. 1745-1748,
generated from a discriminator is a noisier signal than 1997.
that generated from low dynamic situations and this [13] M. N. Ignagni, “Separate-bias Kalman estimator
signal is used as an input of the loop filter. To solve with bias state noise,” IEEE Trans. on Automatic
this problem, this paper proposes a new carrier Control, vol. 35, pp. 338-341, 1990.
tracking loop using an adaptive two-stage Kalman [14] A. T. Alouani, P. Xia, T. R. Rice, and W. D. Blair,
filter, which can track a quickly changed Doppler “On the optimality of two-stage state estimation
frequency. A new carrier tracking loop gives a smaller in the presence of random bias,” IEEE Trans. on
Doppler frequency error than the conventional PLL. Automatic Control, vol. 38, pp. 1279-1282, 1993.
Also, this new carrier tracking loop shows a good [15] C. S. Hsieh and F. C. Chen, “Optimal solution of
performance under a low bandwidth as compared with the two-stage Kalman estimator,” IEEE Trans.
the conventional PLL. on Automatic Control, vol. 44, pp. 194-199,
1999.
REFERENCES [16] M. N. Ignagni, “Optimal and suboptimal
[1] J. B. Tsui, Fundamentals of Global Positioning separate-bias Kalman estimator for a stochastic
System Receivers: A Software Approach, 2nd bias,” IEEE Trans. on Automatic Control, vol. 45,
edition, John Willey and Sons, Inc., 2000. pp. 547-551, 2000.
[2] M. L. Psiaki, “Smoother-based GPS signal [17] K. H. Kim, J. G. Lee, and C. G. Park, “Adaptive
tracking in a software receiver,” Proc. of ION two-stage extended Kalman filter for the fault
GPS, pp. 2900-2913, 2001. tolerant INS-GPS loosely coupled system,”
[3] M. L. Psiaki and H. Jung, “Extended Kalman IEEE Trans. on Aerospace and Electronic
filter methods for tracking weak GPS signals,” Systems, to be published.
Proc. of ION GPS, pp. 2539-2553. 2002.
[4] H. Jung, M. L. Psiaki, and S. P. Powell, Kwang-Hoon Kim received the Ph.D.
“Kalman-filter-based semi-codeless tracking of degree from the School of Electrical
weak dual-frequency GPS signals,” Proc. of ION Engineering and Computer Science,
Seoul National University, in 2006.
GPS, pp. 2515-2523. 2003.
His research interests include Kalman
[5] N. I. Ziedan and J. L. Garrison, “Bit synchro- filtering, the GNSS/ INS integration
nization and Doppler frequency removal at very system, and the GNSS signal
low carrier to noise ratio using a combination of processing algorithm.
the Viterbi algorithm with an extended Kalman
filter,” Proc. of ION GPS, pp. 2515-2523, 2003.
[6] P. Lian, G. Lachapelle, and C. Ma, “Improving Gyu-In Jee received the Ph.D. degree
tracking performance of PLL in high dynamics in Systems Engineering from Case
applications.” Proc. of ION NTM, pp. 1042-1052, Western Reserve University in 1989.
His research interests include Indoor
2005.
GPS positioning, Software GPS
[7] K. H. Kim, J. G. Lee, and C. G. Park, “Adaptive receiver, GPS/Galileo baseband FPGA
two-stage Kalman filter in the presence of design, and the IEEE 802.16e based
unknown random bias,” International Journal of wireless location system.
Adaptive Control and Signal Processing, vol. 20,
no. 7, pp. 305-319, 2006.
[8] D. Kaplan, Understanding GPS: Principle and Jong-Hwa Song is a Ph.D. student in
Applications, Artech House, Inc., 1996. the Dept. of Electronic Engineering,
[9] B. Friedland, “Treatment of bias in recursive Konkuk University, Seoul, Korea. His
filtering,” IEEE Trans. on Automatic Control, research interests include the GPS anti-
jamming, software GPS receiver, the
vol. 14, pp. 359-367, 1969.
GNSS/INS integration system, and the
[10] M. N. Ignagni, “An alternate derivation and GNSS signal processing algorithm.
extension of Friedland’s two-state Kalman

View publication stats