Cycle-Slip Detection and Repair

in Integrated Navigation Systems

A. Lipp ', X. G u

ABSTRACT Acronyms:
In order to use pure carrier-phase measurements, the
DGPS Differential GPS
initial ambiguities of the measurements have to be sol- GPS Global Positioning System (USA)
ved. Once obtained, these ambiguity solutions can be GNSS Global Navigation Satellite System (generic)
lost by shadowing, short-time loss of lock on one or GLONASS GLObal NAvigation Satellite System (Russia)
LSQ Least SQuares estimate
several satellites and other effects, commonly referred INS Inertial Navigation System
to as cycle-slips. SA Selective Availability
Restarting the ambiguity resolution algorithm is a
time- and computation-intensive task and should be
Symbols:
avoided as long as possible. In order to compensate
for such short interruptions, an algorithm would be C speed of light
desirable which detects cycle-slips and is able to cor- D doppler measurement
rect their magnitude for the individual measurements. e doppler residuals
k number of visible satellites
In this paper an algorithm is presented which de- n carrier phase ambiguity
tects, using the redundancy inherent in the GPS- number of cycle-slips
measurements, any cycle-slips. The residuals between pseudorange
satellite position
the doppler measurement and a calculated doppler ba- user position
sed on a least-squares solution show significant chan-
ges whenever cycle-slips occur. 6t measurement epoch length
A ~ E receiver clock offset
Determining in which channel the cycle slip occurred AtE receiver clock drift
and calculating its size is possible using external infor- fr INSmeasurement error
mation, for instance an INS-system. Instead of using x carrier wavelength
accumulated carrier phase
the least-squares solution, the short-term stability of
the INS-system is utilized to generate the calculated
doppler. Indices:
Results will be presented for such an integrated system
showing the effectiveness in correcting even multiple E receiver
I index (measurement)
simultaneous cycle-slips down to the loss of one cycle. r receive-time
The effects of decreasing satellite redundancy on the res residual
s satellite
effectiveness of the algorithm and the influence of the S transmi t-time
information supplied by the INSsystem on the quality I x-axis of a local horizontal
of the cycle-slip repair are investigated. Since GPSand Y y-axis coordinate system
INSsensors work in a closed loop for correction of the Z,H z-axis (vertical axis)
cycle-slips, error detection at this stage is essential for
reliable operation of the integrated navigation system.

'Institute of Flight Guidance and Control 'Deutsche Aerospace AG,

Technical University of Braunschweig Airborne Systems
Hans-Sommer-Str. 66 Sedanstr. 10
D-38106Braunschweig, GERMANY D-89070Ulm, GERMANY

681

0-7803-1435-2194 $3.00 0 1994 IEEE

 INTRODUCTION sian GLObal NAvigation Satellite System (GLONASS).
Position Determination The methods described in the following for GPS are
therefore valid in principle for both systems.
Position determination with Global Navigation Satel-
lite Systems (GNSS)is performed using propagation- Carrier-Phase Measurements
time measurements of electromagnetic waves. These The use of carrier-phase measurements is indispensa-
measurements are multiplied with the speed of light ble for high-precision measurements with satellite na-
to obtain the pseudorange. Together with known sa- vigation systems. All algorithms using code informa-
tellite positions and neglecting error sources, the user tion only are limited to a range accuracy of about 0.5-
position can in principle be calculated using the folle 1 m due to code noise (for GPS, even for narrow cor-
wing equation: relator receivers). Range measurements using carrier-
phase information, on the other hand are limited to
only 0.5-3 mm by noise as shown in [l].
For several satellites a nonlinear system of equations In order to utilize the carrier-phase information some
results, with i running from 1 to the number of availa- differencing in time has to be performed: On carrier-
ble satellites, k. smoothed pseudorange algorithms the changes in
Navigation in space can only be performed if four or phase between measurement epochs are utilized to re-
more satellites are available to obtain the three posi- duce code-noise in a filter process, while in algorithms
tion variables and the user clock offset. If the num- employing pure carrier information a mechanism to
ber of satellites exceeds four, the additional ones can eliminate the ambiguity has to be implemented. This
be used to improve the quality and integrity of the is either performed by direct differencing (Triple Dif-
solution. Available are basically three types of mea- ference Techniques) or by employing statistic methods
surements: Code (Pseudorange), Phase (Carrier) and over several epochs to reduce the search area for am-
Doppler-Frequency (derivative of carrier phase). For biguity solutions to a minimum [2], [3].
the code measurement an internal code generated wi- Cycle-Slips
thin the receiver is correlated with the received signal.
Displacement of the internal code for maximum cor- In any case, short interruptions of signal reception or
relation results in the desired propagation-time. This other causes can result in sudden jumps of the carrier-
information is ambiguous by the period of the code phase signal from one epoch to the next, commonly
(lms), but due to the long code length the ambiguity called cycle-slips. In order to correct them, the mea-
can usually be solved by using a position estimate of surement process can be stopped and restarted with a
about 300 km accuracy. A similar process occurs for new ambiguity solution - a time-intensive and someti-
the carrier-phase information as shown in Figure 1. mes impossible method, resulting in the loss of several
Here the wavelength is only few centimeters, so that measurement epochs. Cycle-slips therefore should be
ambiguity plays an important role in the use of this in- detected and corrected before influencing the naviga-
formation. The third measurement variable, doppler tion solution and before changing filter-states in code-
frequency, results from the frequency difference bet- smoothing systems.
ween received and generated signal. In this paper a solution is presented within an inte-
grated navigation system. This system consists of the
Ges-receiver and an inertial navigation system.
--- Total phase I
value jizz CYCLE-SLIP
OCCURRENCE
The phase difference between received and generated
signals shown in Figure 1 can be measured by a phase-
locked loop. The initial integer count of wavelenghts is
unknown and commonly denoted as integer ambiguity.
Clearly any signal interruption or failure of the recei-
ver to correctly follow the incoming waveform with its
internal tracking loops will cause measurement errors.
Re&m Similarily the failure to correctly detect and count an
I Receiver Tmcscale
Time
I- ~ - incoming waveform due to poor signal-to-noise ratio
will result in lost cycles. In the following only the in-
Figure 1: Carrier Phase Measurement teger part of the lost measurement will be referred to
as cycle-slip - the noninteger part is measured by the
The basic measurement process is identical for both receiver and usually interpreted as noise on the phase
GNSS systems available at the moment, the Ameri-
measurement.
can Global Positioning System (GPs) and the Rus-

682
 GPS - Satellite from the knowledge of satellite position and a user
position based on INS-data. The difference provides
error information for the Kalman filter, which then
models INS-errors to improve the quality of the navi-
gation solution [7].
Also shown are the cycle-slip detection and repair pro-
cess with a least-squares estimation delivering data to
the detection algorithm. If missed cycles are detec-
B&Aigle ted, the path of the GPs-data is routed through the
GPS - Satellite
repair algorithm before being passed to the Kalman
filter. This ensures that only GPs-data are used for
detection, avoiding a feedback loop from the Kalman
filter via the navigation solution that could lead to in-
stabilities on errors in the detection process. For both
detection and repair differential corrected GPS data
should be used to aquire a sufficient basic accuracy
PI
CYCLE-SLIP
DETECTION
Most GPS receivers have some type of detection flag
Figure 2: Antenna Shadowing indicating loss of reception for more than one epoch.
Thus only interruptions of less than one epoch have
to be considered for an additional detection algorithm.
In a moving vehicle such as a car or an airplane both Such short losses will be indicated by a variety of re-
of these factors are commonly present: Obstructions ceivers - unfortunately with widely varying accuracy.
in the signal path are caused by buildings and aircraft
For stationary receivers the expected change in range
parts above or near the antenna (Figure 2), while miG
from one epoch to the next (which is only due to sa-
neuvering causes dynamic influences. tellite motion) can be calculated and subtracted from
Cycle-slips can occur separately or at a common epoch the actual, differential corrected measurement in or-
in the different channels of a parallel receiver and they der to detect cycle-slip occurrence [9]. A measurement
range from approximately 0.2 m (one wavelength) to variable is therefore defined as the difference between
several tens of meters, with the lower values occurring phase measurements of adjacent epochs, signifying the
more often. change in range:
In this context "detection" of cycle-slips shall describe D = O(tl + 6t) - O ( t l )
the test whether cycles are lost at all, while "repair"
describes the process of finding the affected channel(s) In the following this difference will be termed "dopp-
and determining the number of cycles lost in each af- ler" measurement since it can be seen as the average
fected channel. Adding the lost cycles to the ambi- of the instantaneous doppler frequency between the
guity value before the cycle-slip leads to the correct two epochs. This definition allows the method to be
phase measurement value. extended to receiver types without continuous doppler
INTEGRATEDNAVIGATION
SYSTEM frequency output.
For moving receivers the procedure described above
The Inertial Navigation System (INS) can calculate
position, speed and attitude of a moving vehicle from is impossible, since the velocity vector of the receiver
and therefore its doppler component is not known with
acceleration and rotation rates. It has good dynamic
behaviour but a long term drift. Satellite navigation sufficient accuracy, as shown in Figure 4.
system data are sensitive to dynamic influences but The algorithm described in the following uses an ex-
stable in their long-term behaviour and thus comple- tension of the doppler prediction method useful for a
ment the information from the INSvery well [4]. The moving receiver as well. Like many other algorithms
combined system has a very good dynamic behaviour for integrity monitoring it needs some redundancy in
and excellent long term stability [5],[6]. the measurement - more than four usable doppler mea-
Figure 3 shows the integration process with a Kalman surements. If this is the case, a least-squares (LsQ)
approximation of aircraft speed can be retransformed
filter estimating sensor errors of the INS-SyStem on a
range correction basis. Differential-corrected range in- into approximate doppler values. The residuals e bet-
formation is compared with synthetic ranges derived ween approximate and actual doppler give an indica-
tion of the single receiver channel relative to the total
 DGPS Range
Calculation

Variables: PAitiOll
21 Position Velocity
41 Velocity Anitude
1 Acceleration
Q Attitude
Code Measurement
* 9 Phase Difference
** Receiver Clock Offset
6 f Receiver Clock Drift
p Phase Residual
5 AttitudeError
Modifiers:
A Estimate
6 Difference
Figure 3: Integrated Navigation System

detection criteria must be defined from statistical ex-

amination of measurements made with the target sy-
stem. A detection level above 3a of the carrier-phase
noise leads to more than 99% detection probability
assuming a normal distribution of the measurements.
Whenever cycle-slips occur, these residuals rise shar-
ply for most channels, even unaffected ones, since the
approximation is biased by the faulty measurement
(see also Figure 8). For slips in one channel only,
the affected channel will show the greatest difference,
Figure 4: Definition of Doppler Residuals while several faulty channels can bias the approxima-
tion in such a way that even an unaffected channel
might show the largest deviation. Therefore identi-
doppler estimate. fication of the affected channel cannot be performed
Besides cycle-slips other effects produce changes in reliably with techniques using GPs-data only.
the doppler measurement as well. These are mainly The algorithm also fails if only four measurements are
due to atmospheric effects, SA, multipath and receiver available - the LSQ-estimation will not show residual
noise. For high-precision measurements, where cycle- errors due to insufficient measurement redundancy. In
slip detection is most useful, differential techniques this case other information must supplement the mea-
are usually employed and eliminate all errors common surements, the most obvious being the receiver clock.
to the receivers, such as atmospheric errors and SA Since receiver clocks do not run absolutely synchonous
almost completely [SI. Multipath effects change rela- to Gps-time, their offset and drift is usually calcula-
tively slow from one epoch to the next (incidentally ted together with the navigation solution. This infor-
making them so hard to detect), making their influ- mation, filtered over several epochs, serves to build a
ence negligible for doppler calculation. The doppler clock model theat can be used to estimate clock drift
residuals e thus are usually small and depend mostly in parallel to the LSQ-calculation. The redundancy
on receiver noise as Figure 5 shows for a typical flight essential to cycle-slip detection is again sufficient as
test. Figure 6 shows. Here drift calculated from a simple
Since receiver noise depends on the type of receiver the linear clock model is compared to the drift resulting

. .
from the LSQ-estimation of DGPSvelocity calculation. Drift of a Rubidium Oscillator
i t [ds]
Cycle slips of one respective four cycles were injected
into one measurement at times A and B, while measu-
rements from four visible satellites were utilized. The
influence of the cycle-slip is transformed into the clock
-1
estimate and can be evaluated. Caution has to be -2
exercised in taking this approach, since many receiver
manufacturers reset or adjust the clock in more or less
frequent intervals, making it harder to reliably predict
clock behaviour. & [,,,S/~I Drift of a Quartz Oscillator

I* l B IC

Figure 7: Oscillator Sensitivity to Acceleration

check for the appearance of cycle-slips, but this would

not exploit the capabilities of the integrated system
to its fullest extent: The additional data can be used
to indenpendantly check the detection algorithm and
even repair cycle-slips in real-time.
REPAIR
CYCLE-SLIP
Figure 5: Doppler Residuals without Cycle-Slips
As noted previously, slips in a single channel can be
identified by the largest doppler discrepancy. Re-
testing with a subset of channels can lead to the

"t
40
size of the cycle slip by better doppler approxima-
tion (Method using hypothesis testing as described
in [lo]). Concurrent cycle-slips in several channels
will not reliably be identified, since the vector sum
of the error may point in a direction of an unaffec-
ted measurement, leading to erroneous cycle-slip si-
Figure 6: Clock Drift Change with Cycle-Slips
zes. Through the iterations with all possible subsets of
measurements this technique is also very computation-
For quartz oscillators acceleration sensitivity poses intensive and ill-suited for real-time applications.
another challenge to clock modeling. Figure 7 graphs
clock drift as described above and its change relative At this point the short-term stability of the INS-
to GPs-time to demonstrate this. In an experiment re- system can be used to an advantage: Starting from the
ceivers containing rubidium and quartz oscillators re- last uncorrupted measurement at time t l the position
spectively were mounted on a common platform and and speed are extrapolated using the INSinformation
rotated by 90' at the times A and B, and by 180' to the current epoch t 2 . A geometric range between
a time C. The result shows significant jumps for the satellite and user antenna R and the geometric change
quartz oscillator due to the change in direction of the in range 6R can now be calculated and compared with
gravity vector. The slow oscillations present in both the measured values RM and ~ R MThe . equation has
graphs are due to slight multipath influence. to be corrected by the uncertainty of the calibrated
INSmeasurement €1 and the receiver clock drift A ~ E
For reliable detection of single cycle slips the unmode- during the time difference between the two epochs.
lable changes in clock drift should not exceed few ns/s.
An atomic oscillator is therefore highly recommended The following equations result, with n*X denoting the
if this precision is to be maintained even under bad range difference due to cycleslips and c the speed of
reception conditions. light:
The integration of GPS- and INSinformation can also 6R = R(t2) - R(t2)
be used to generate a synthetic doppler in order to ~ R M= R ~ ( t 2- )R ~ ( t 2 )

685
The number n of cycles missing in each channel can
now be calculated for each channel independent of the
satellite constellation. Important for this process is
the quality of the receiver clock model and the INS
calibration. Both terms increase with the time dif-
ference since the last uncorrupted measurement and
show the short term nature of this algorithm.
VALIDATION
EXPERIMENTS
Unfortunately almost no reference systems exist to va-
lidate GPs-data in an airplane environment at the pre-
cision levels needed to detect single cycle-slips. The-
refore the algorithms were tested in a simulation en-
vironment until a suitable reference method could be Figure 8: Result without Cycle-Slip Repair
defined. A pattern of sinusoidal movement in the east-
west direction was programmed with position in the
other axes held fixed for reference. A typical constel- data-bus system behaviour and syncronization varian-
lation was used with six satellites usable for naviga- ces between INSand GPS systems. As Figure 9 shows,
tion. The motion was programmed to contain conti- a small time-tag error can be tolerated, while a larger
nous acceleration/deceleration of up to 0.5g, since ac- error does not allow full compensation of all cycle-slips
celeration phases are those where cycle-slips are most any more. With such a large timing error the aircraft
likely to be encountered and those where conventional model created using the INS-data does not describe
algorithms show the highest error rates. Single and the motion well enough to guarantee 1 cycle accuracy.
multiple cycle-slips were injected especially in phases In the next modeling step several typical errors of an
of high acceleration and with sizes varying from one INS-system were introduced into the simulation. First
(0.2m) to 200 cycles (40m). In order to test detection sensor drift and misalignment errors were modeled for
over several epochs, multiple slips in adjacent epochs an INS-system calibrated by GPs-measurements on a
were implemented as well. Cycle slips larger than 200 range basis. The INSdata in Figure 10 were calculated
cycles can easily be detected using plausibility checks with a typical gyro drift and an accelerometer offset.
with assumed maximum values of aircraft dynamics The data now exhibit a characteristic sawtooth form
and were therefore not employed in this investigation. resulting from residual sensor drift reset to a base va-
Simulation output were a set of data files identical in lue by the GPS measurement in each epoch. For the
format to real DGPS receiver measurements and con- INS-system employed in the flight tests this leads to
taining all relevant GPS error sources. An INSoutput a position drift of less than 1 cm/s and a velocity
data file was also generated, containing time stamp, error of less than 5 mm/s2 due to unmodelable errors.
position, velocity and acceleration information. Barometric altitude was used to stabilize the vertical
channel.
They were read using the on-line software with the
detection and repair algorithm and a list of detected With this error model the repair algorithm perfor-
cycle-slips was generated and compared with the ori- med very similar to the behavior observed for time-
ginal values. tag errors: As long as INS errors do not exceed the
size of 0.8 cycles during one GPS epoch flawless cor-
As can be seen from Figure 8 the cycle slips show rections are obtained even during phases of high ac-
clearly in the result without a repair process and lead celeration. Larger errors lead to slow failure, with
to significant deviations in the vertical channel ( d H corrections being one or two cycles off the true value.
altitude offset; VH vertical speed). The doppler resi- For the strapdown-INS system used a t the Institute
duals D,,, show large spikes, while the speed in the of Flight Guidance and Control the residual errors are
direction of motion V, does not seem to be influenced safely below the 0.8 cycle limit if the errors are calibra-
due to the scaling of the graph. ted by DGPSevery second, allowing reliable cycle-slip
With the repair process using an ideal INS-signal, all corrections.
cycle-slips were detected with the correct size and no Important is the cutoff of the INSfilter adaptation al-
false alarms occurred. In order to further test the gorithm while cycle-slips are detected. Otherwise the
robustness of the algorithm, errors were now imple- feedback of erroneous corrections will rapidly detune
mented in the INS-data. First the time-tag was modi- the sensor models and make further calculations im-
fied to simulate changes in timing due to the aircraft

686
 ..
aoi
an
4Bl

Figure 10: Drift of Calibrated INS-System

to the number of cycle slips for each satellite summed

over the entire flight. Test results up to now, using
0.01
flights with low to medium dynamics and receivers
em using narrow corelator technology, did not result in a
401 cycle slip being detected with the carrier phase solu-
tion and no cycle slip was detected with the algorithm
described in this paper.
Figure 9: Results with Cycle-Slip Repair Therefore, until flights are performed using different
receiver types and other antenna mounting locations,
possible. A simple method of testing whether repair it can be concluded that the false-alarm rate of the
algorithm is very low and the simulation results show
was successful or not must therefore be devised for
reliable operation. The outcome of this test will ge- correct detection and repair of all cycle slips even un-
nerate a criterion for turning on the filter-adaptation der high dynamic influences if high quality INS sy-
algorithm again. stems are employed.

One option for such a test is a comparison between CONCLUSIONS

cycle-slip detection and repair algorithms. If the de- Detection and repair of cycle-slips is important to all
tection algorithm gives no indication of cycle-slips in a applications using carrier-phase information. Cycle-
subsequent epoch, although the repair process shows slips often occur in situations with high dynamics and
significant correction values, the corrections must be in areas with buildings where precise navigation is of
due to erroneous sensor models. The importance of special importance. Detection of cycle-slips can reli&
having detection algorithms independent of the repair bly be performed using information derived from mea-
process is stressed once again by this example. surement redundancy and a priori information about
It can be seen that repair of cycle-slips is dependent the receiver and its clock. For real-time applications
on three main factors: accurate timing correlation bet- repair of cycle-slips is of utmost importance, neces-
ween the different sensors, INSsensor and filter model sitating the use of information external to the GPS-
quality and epoch duration of the GPSreceiver. Consi- system. One option shown to be feasible is the inte-
derations for a specific implementation may thus trade gration with an Inertial Navigation System employing
accuracy requirements to a certain extent between the its short term accuracy. This approach is sensible,
different systems. since INS-systems are widely used for aircraft navi-
gation where high-precision position determination in
Final validation of the detection and repair process real-time is important. Repair of cycle-slips is then li-
comes from flight test data in comparison to an al- mited to the error model precision of the INS-system.
gorithm using pure carrier phase measurements with For currently available systems this precision is ade-
an ambiguity search method. The ambiguities were quate to correct even single cycle-slips if the time cor-
solved for a stationary position before and after the relation between GPS and INS data is known. Since
flight and the results compared. Their difference leads the basic quality of the INSerror model depends on the
 accuracy of the GPSmeasurements [ll],high-precision [lo] TEUNISSEN, P.J.G. (1991): On the Minimal Detectable
DGPSposition data during the time before the cycle- Biases of GPS Phase Ambiguity Slips, Proceedings: First
International Symposium for Real Time Differential A p
slip occured also determine repair accuracy. The high plications of the Global Positioning System, Sept. 1991,
quality of the receivers used in the flight tests up to Vol. 1, Braunschweig
now resulted in no occurrence of cycle-slips under low
[ll] TIEMEYER, M.E.; Lu, G.; S C H ~ N Z EG.
B.; CANNON, R,
to normal aircraft dynamics. Further tests using diffe- (1994): Ambiguity Resolution of GPS L1 Carrier-Wave
rent receivers and higher dynamics will be performed Measurements for High Precision Aircraft Navigation,
to complete validation of the results of this algorithm. Proceedings: Position, Location and Navigation Sympo-
sium (PLANS’94), April 11-15, 1994; Las vegas
Acknowledgement
We would like to thank the German Research Society
( DFG)for their support through the Special Research
Project (SFB) 212 ”Safety in Aviation” and the Ger-
man Space Agency (DARA)for their partial financial
support of this research projekt.

References
LANDAU, H.; EULER, H.J. (1991): The Key to Decimeter-
Level Differential GPS Navigation : Carrier Phase Am-
biguity Determination, Proceedings: First International
Symposium for Real Time Differential Applications of the
Global Positioning System, Sept. 1991, Vol. 1, Braun-
schweig

Gu, X.; LIPP, A. (1994): DGPSPositioning Using Carrier

Phase for Precision Navigation, Proceedings: Position,
Location and Navigation Symposium (PLANS’94), April
11-15, 1994; Las Vegas

LACHAPELLE, G.; CANNON, M.E.; Lu, G. (1992): High

Precision GPS Navigation with Emphasis on Carrier Phase
Ambiguity Resolution, Marine Geodesy, vo1.15, No.4,
1992, pp. 2 53-269

Gu, X. (1991): A Real Time Differential GPS System for

Aircraft Landing, Proceedings: First International Sym-
posium for Real Time Differential Applications of the Glo-
bal Positioning System, Sept. 1991, Vol. 1, Braunschweig

TIEMEYER, B.; S C H ~ N Z EG.

R , (1992): Integrated Preci-
sion Navigation System, Paper pres. on the 54th Meeting
of the AGARDGuidance and Control Panel, ’Integrated
and Multi-Function Navigation’, 14-15 May 1992, Ottawa,
Canada

JACOB, T. (1989): Integrated Flight Guidance Systems

using Differential-GPs for Landing Approach Guidance,
AGARD Guidance and Control Panel, 1989, Lisboa

VIEWEG,S.; TIEMEYER, B. (1993): Modelling and Ca-

libration of DSNSand Inertial Sensors in a n Integrated
Navigation System, Paper pres. on the 2nd International
Symposium on Differential Satellite Navigation Systems
DsNs’93, 29 Mar.-2 Apr.1993, Amsterdam

Gu, X.; TIEMEYER, B.; LIPP, A. (1993): High Precision

Navigation with Wide Area DGPS,Paper pres. on the 2nd
International Symposium on Differential Satellite Naviga
tion Systems DsNs’93, 29 Mar.-2 Apr.1993, Amsterdam

LICHTENEGGER, H.; HOFFMANN-WELLENHOF, B. (1989):

GPS-Data Preprocessing for Cycle-Slip Detection, In:
BOCK,Y.; LEPPARD,N. (EDS): Global Positioning Sy-
stem: An Overview, Springer, New York Berlin Heidel-
berg London Paris Tokyo Hong Kong

688

_ . .” . ..
..- .- - . .. .