GPS TIME STEERING
W. J. Klepczynski
U. S. Naval Obsservatory
H. F. Fliege 1
The Aerospace Corporation
D. W. Allan
National Bureau of Standards
The importance of the Global Positioning System (GPS) for global time transfer
makes it desirable to steer GPS time as closely as possible to the UTC rate.
Currently, GPS time is maintained to satisfy two system requirements. First, GPS
time is steered to within one microsecond of UTC(USNO) when the leap seconds
imposed on UTC since I960 have been removed. Second, the GPS Navigation
Message gives the offset UTC(USNO) - GPS time to users with an error not to
exceed 100 nanoseconds. User performance would be improved, however, if changes
in the GPS time rate were smaller and more gradually imposed than at present.
Three current developments are expected to improve GPS time steering
performance: the installation of a stable clock ensemble at the GPS Master Control
Station: improvement of supporting hardware; and application of control theory to
The Global Positioning System (GPS) for navigation and time transfer is currently in
the development phase, and must pass through a number of well-defined checkpoints
before becoming operational. Important concerns which are presently the focus of
management attention include the following:
1) the delay in deploying the full constellation of Block II Navstar satellites, in the
wake of the setback to the Space Transportation System;
2) the question whether the Block I satellites now operating will continue to be
useful, since several are long past the five year lifetime for which they were
3) the difficulty of testing and proving the ground control hardware and software
with the limited constellation of satellites now available.
Nevertheless, GPS has been astonishingly successful for its nonmilitary users in providing
an accurate and reliable means of global time transfer. For several years,
synchronization experiments between the US National Bureau of Standards (NBS) and
other laboratories, and the international timekeeping center at the Paris Observatory,
have been performed using the GPS with reported accuracy of 20 nanoseconds, using the
Report Documentation Page
OMB No. 0704-0188
Public reporting burden for the collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and
maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information,
including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington
VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to a penalty for failing to comply with a collection of information if it
does not display a currently valid OMB control number.
1. REPORT DATE
2. REPORT TYPE
3. DATES COVERED
00-00-1986 to 00-00-1986
4. TITLE AND SUBTITLE
GPS Time Steering
5a. CONTRACT NUMBER
5b. GRANT NUMBER
5c. PROGRAM ELEMENT NUMBER
5d. PROIECT NUMBER
5e. TASK NUMBER
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES)
U.S. Naval Observatory,3450 Massachusetts Avenue,
5f. WORK UNIT NUMBER
8. PERFORMING ORGANIZATION
9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES)
10. SPONSOR/MONITOR'S ACRONYM(S)
11. SPONSOR/MONITOR'S REPORT
12. DISTRIBUTION/AVAILABILITY STATEMENT
Approved for public release; distribution unlimited
13. SUPPLEMENTARY NOTES
Proceedings of the Eighteenth Annual Precise Time and Time Interval (PTTI) Applications and Planning
Meeting, Washington, DC, 2-4 Dec 1986
15. SUBIECT TERMS
16. SECURITY CLASSIFICATION OF:
c. THIS PAGE
17. LIMITATION OF
19a. NAME OF
Standard Form 298 (Rev. 8-98)
Prescribed by ANSI Std Z39-18
common view technique (Ref 1). The difference between UTC as provided by GPS and
UTC(USNO) is usually less than 40 nsec (Ref 2). Use of the GPS has replaced LORAN-C
as the transatlantic link between International Atomic Time (TAI) contribution, and
permitted the inclusion of contributors to TAI in Asia and Australia (Ref 3). The
usefulness of GPS, even in its unfinished state, has encouraged observers to make
recommendations for its further improvement, and this paper will review progress in
implementing those recommendations.
When the GPS is fully operational, GPS time (hereafter called Tgps) is to be related
to UTC (Ref 4) and is to satisfy three specifications on accuracy;
1) Tgps is to be maintained to plus or minus one microsecond of UTC time, after
leap seconds in UTC are removed. Specifically, since there are no leap seconds in Tgps,
the number of leap seconds in UTC accumulated since the GPS epoch of OH UTC on 6
January 1980, LS, is subtracted from the difference Tgps-UTC, so that the timing
Tgps - UTC - LS < 1 microsecond. (Ref 5)
2) The offset Tgps - UTC is to be supplied to authorized users with error not to
exceed 110 nanoseconds (nsec) standard deviation (1 sigma) (Ref 6). This requirement is
to be tightened to 100 nsec in revised editions of GPS Interface Control Documents.
3) The Standard Positioning Service (C/A code) will be made available to all users,
internationally, with a time equivalent 2 sigma accuracy of 250 nsec or better (Tgps), for
a user in a known location (Ref 7).
Notice that specifications (2) and (3), above, imply that in times when selective
availability is imposed, the authorized user can obtain UTC to better than 100 nsec (1
sigma), and the general user to 160 nsec (1 sigma) or better.
Of course, from a purely technical point of view, the above requirements are far
from pressing the state of the art in time and frequency control. Thus, the Committee
on Accuracy of Time Transfer in Satellite Systems reported;
The actual performance of the GPS system reference clock up to now is only
mediocre for a cesium clock compared to what is seen in similar clocks at the
USNO and in many terminals of the Defense Satellite Communications System
(DSCS)...The main reason for the poor performance of the GPS ground clocks has
been their adverse operational environment...
...a complete accounting of all clock rates and effects in the ground system
with respect to the DOD Master Clock will improve performance...
...progress...may make it possible to improve significantly the accuracy of
time transfer systems such as GPS. Yet such improvements might not be made
because the stated requirements at the time may not be tight enough to make them
obligatory ...In the absence of tighter requirements the system would continue to
operate at the current level despite the possibility for improvement. We think this
would be a mistake...(Ref 8)
We will show in this paper that many, if not all, user concerns are being met. Tgps is in
the process of becoming a much more stable measure of time than formerly, within the
framework of DOD policy. According to this policy, requirements are not driven by
capability, but by military necessity -- that is, not by US capability, but by the capability
of its potential adversaries.
GPS TIME SCALE: CURRENT PERFORMANCE
Figures 1 and 2 illustrate the performance of Tgps v/ith respect to UTC(USNO).
Figure 1 shows the offsets UTC(USNO) - Tgps from the USNO [aJGPSVl file, which
employs three day smoothing on differences measured between I 1TC extracted from the
navigation messages of all satellites and the USNO master clock, given the known
position of the USNO receiver. Figure 2 shows the UTC(USNO) - Tgps rate differences
from the same file. Thus, Figure 2 shows the first derivative of the function of Figure 1,
except for the effects of data smoothing.
All the specified requirements described above were met. However, four events
shown in Figure 1 caused the Tgps rate to change by large amounts, which would not have
happened in an ideal system.
1) An abrupt steer was imposed on day of year 40 (0 February) by operations
personnel, who observed that the Tgps rate was high and away from zero, and who
believed that the proper action was to reverse the rate.
At a performance analysis working group meeting at the GPS Master Control Station on
11 March, a procedure was proposed by which the US Naval Observatory would calculate
appropriate GPS steering parameters and post them in the USNO (3GPSD9 file for
operations use. This procedure eliminated large changes in the Tgps rate during April,
May, and June.
2) On 26 June, the Colorado Springs Monitor Station (MS) clock was acting as the
GPS master clock. On either 26 or 27 June, the master clock lost cesium control
and reverted to quartz crystal. The problem was detected on Monday, 30 June.
However, the role of master clock was switched from the Colorado to the Ascension
MS after the clock failure and before its detection. By the time the problem was
detected, the system software had mapped the defective clock rate to the new
clock, and it was too late to recover the system by re-estimating clock states via
the Kalman filter. The reason for the delay in detection of the problem is not fully
known, but at present we do not believe it was due to any deficiency in GPS system
. software. Aerospace, IBM, and the JPO are continuing to examine both software
and operator procedures to prevent similar events in the future.
The USNQ/MCS interface succeeded in returning the Tgps - UTC rate to zero, but a
series of communication line and monitor station hardware problems slowed the
recovery. Steering was suspended during July because of the large number of master
clock switches which these problems made necessary: 11 July, Ascension to Colorado; 14
July, Colorado to Hawaii; 21 July, Hawaii to Colorado; 22 July, Colorado to Kwajalein;
July, Kwajalein to Colorado. The ECEST software is designed to preserve GPS time and
rate against discontinuities during a master clock change, and the software worked very
well. The only large change in Tgps rate between events (2) and (3) occurred around 15
to 20 July, due to normal random behavior of the Hawaii MS clock. (The MS clocks are
not in a fully temperature controlled environment.)
3) The change in rate of the UTC(USNO) time scale of 12 nsec/day on 1 September
is reflected in Figure 1. Due to operational problems at the Master Control
Station (not clock related), it was not practical to steer GPS time to compensate
for the UTC change until 9 September.
4) Considering the fact that Tgps was already within 300 nsec of the permitted 1000
nsec limit from the norm, and that another clock failure might put us over the limit,
it was decided to steer by the maximum permissible rate to set a course toward
In retrospect, we see that two improvements would eliminate sharp changes in Tgps rate
and would permit steering of Tgps to within 50 nsec or less of the norm: first, a more
stable master clock, with more dependable supporting hardware, which would eliminate
the need for frequent clock switches; and second, clear procedures for controllers to
follow and appropriate training for system operators, to minimize the effects of clock-
related hardware failures.
USNO PRECISE TIME REFERENCE STATION
The U.S, Naval Observatory has installed a Precise Time Reference Station (PTRS) at the
GPS Monitor Station (MS) at Falcon AFS, CO which can serve, under normal
circumstances, as the GPS Master Clock. The system consists of an ensemble of cesium
beam frequency standards (clocks) which are coordinated by a data acquisition and
control system which is used for the monitoring of various systems and for the
communication and exchange of data in order to allow the setting of a station clock
which is kept synchronized in time and frequency to UTC(USNO) by the control of a
phase-microstepper (Ref. 9). The PTRS will serve as an interim system pending a
Hydrogen Maser Advanced Clock System (HMACS) which will constitute the Operational
Control System Advanced Clock to be installed at the GPSMS, Falcon AFS, CO by the
Naval Research Laboratory (NRL).
The present design of the Operational Control System (OCS) employs two high
performance Cesium Beam Frequency Standards. One drives the receiving equipment,
while the other is used as a spare. The hydrogen maser system will be based on the
design of the Precise Time Reference Station used by the USNO, which acts as a logical
intermediate system between the present system and the forthcoming Hydrogen Maser
Advanced Clock System.
Various forms of timing data are obtained by the PTRS, It consists of measured
differences of various pairs of clocks in the local ensemble and observed differences
between the local reference clock and GPS time as determined through a single channel
time transfer receiver. Through common-view measurements (Ref. 10) with USNO, the
drift rate and offset of the local reference clock at the GPSMCS with respect to
UTC(USNO) can be determined. Once these parameters are known adjustments can be
made to the phase-microstepper controlling the local reference clock. These
adjustments can be made automatically by the USNO ADS or provision can be made for
local station personnel to control the setting of the phase-microstepper because of
As of 21 Dec 1986, an algorithm to automatically control and steer a local reference
clock to UTC(USNO) has been implemented within the PTRS at Falcon AFS. Figure 3
shows the results of that steering. After automatic steering started, an anomalous
frequency change occurred in the local cesium frequency standard driving the PTRS
Local Reference Clock. The length of time it took the algorithm to compensate for this
change in frequency was long. The compensation could have been done much more
quickly, but the algorithm takes into consideration the operational constraints that are
currently imposed on the steering of GPS which were mentioned earlier.
Communication between the USNO and the GPSMCS is essential for exchange of data.
The daily exchange of data is sufficient to assure adequate control of the local reference
clock. In the event that there is loss of communication, then the local ensembel is used
as a flywheel to extrapolate UTC(USNO). The ensemble also monitors the short-term
performance of the local set of clocks. This allows the automatic identification of
poorly performing clocks -- thus improving system performance.
CONTROL THEORY EQUATIONS AND SIMULATED PERFORMANCE
The block diagram shown in Figure 4 illustrates the system. Our goal is to set the
filter function parameters in such a way as to drive the time difference to zero between
the output of the micro-stepper and the UTC(USNO MC) as seen at the Operational
Control Segment (OCS) via the GPS common-view time-transfer technique. This time
difference is given by
X (t) ="Xq (t) + ‘X S (t) - (Xm(t) + Xn (t)) (1)
In addition the parameters need to be set in such a way as to be insensitive to
system disturbances, and such that changes in frequency of the micro-stepper output are
less than 2 x 10-14/day, which is the OCS GPS requirement to prevent the Kalman filter
from propagating errors which may unduly perturb the navigation solutions for the users.
If Y(t-T) is the last frequency correction value set in the micro-stepper, then a
filtered estimate of an update frequency correction is given by
Y(t) = ^rr(Y(t-T) + (X(t)-X(t-'t))/'r) + l*X(t)/^ (2)
where the first term drives the syntonization and the second term drives the
synchronization. Tau "'ll" is the nominal time interval between measurements — in our
case it is typically one day. All we need to do is pick proper values of m and 1 for the
range of random variations we may encounter in the master clock, the OCS clock and in
the common-view the new output time added by the micro-stepper is given by
'X 3 (t+'t)=X 5 (t) - Y(t)/2 (3)
The common-view time transfer noise has been measured to be a white phase
modulation (PM) process with a standard deviation of about 1 to 2 ns. The random
variations of the master clock and the OCS clock have been measured to be a flicker
noise FM like process in the range of 1 to 4 x 10-14 for <TZ ("£), and 1 dayi X £ a few
Figures 5 and 6 are the measured and simulated freguency stabilities. We picked a
nominal worst case and a nominal best case for simulation purposes. The worst case
conditions are plotted in Figures 7 and 8 for the frequency stabilities of the micro-
stepper's steered output, x (t), and of the time amounts to less than 2 x 10-14, which is
the design goal; and the standard deviation of the residuals around a linear-least-squares
fit to the time difference output amounted to 9ns.
Figures 9 and 10 are the corresponding figures for the nominal best case
simulation. In this case the day to day rms steering correction amounts to 0.8 x 10-14,
and the standard deviation of the residuals around a linear-least-squares fit to the time
difference output amounted to 3ns.
Figure 11 illustrates the transient response to a beginning time and frequency
error. Through simulation and an empirical approach to setting the parameters, we were
able to find a single set which gave the above performance for the nominal worst case,
the nominal best case and for the transient response. The values we found were m = 2.5
and 1 = 0.4 with T = 1 day.
In order to improve GPS for the global distribution of precise time, a number of
modifications to existing procedures have been proposed and are in the process of being
implemented. Improved timekeeping hardware at the Master Control Station will
minimize switching of the designated GPS master clock among the various monitor
stations. This will help reduce the effect of master clock switching on steering when it
is being implemented by software techniques. The application of control theory to time
steering procedures will dampen any sudden changes in GPS master clock rate. The users
of precise time will benefit from these improvements in timing performance because the
time signals from GPS will be modeled with a higher degree of reliability than
1) "Accuracy of Time Transfer in Satellite Systems", CATTISS (Committee on
Accuracy of Time Transfer in Satellite Systems), National Academy Press, 1986,
2) Klepczynski, W. J. and Kingham, K. A., "Time From GPS", Proceedings of the First
International Symposium on Precise Positioning with the GPS, US Department of
Commerce, National Oceanic and Atmospheric Administration, May 1985, p879f,
3) Comite Consultatif Pour la Definition de la Seconde, Bureau International des Poids
et Mesures, Rapport de la lQe Session, pS45.
4) "Precise Time and Time Interval (PTTI) — Planning, Coordination, and Control",
Department of Defense Directive 5160.51, 14 June 1985.
5) "Navstar GPS Control Segment / US Naval Observatory Time Transfer Interfaces",
Interface Control Document- GPS-202, 21 November 1984, plO.
6) idem, p23.
7) Letter, Wayne H. Jones, Colonel, USAF, to Dr. Giacomo, Bureau International des
Poids et Mesures, 3 July 1985.
8) op. cit., Ref 1, pp 25, 41.
9) Wheeler, P.J., "Automation of Precise Time Reference Stations (PTRS)," 41-52,
Proceedings of the 15th Annual PTTI Applications and Planning Meeting, 1983.
10) Allan, D.W. and Weiss, M.A., 1985, "Accurate Time and Frequency Transfer Driving
Common-View of a GPS Satellite," Proceedings of the 34th Annual Frequency
Figure 2 - Differences (in nanoseconds/day) between the rate of UTC(USNO) and GPS
Time. The data represents the first derivative of the data in Figure 1, except for the
effects of smoothing.
UTC(USNO) - PTRS(Local Reference)
Figure 3 - Differences between UTC(USNO) and the Local Reference Clock of the USNO
Precise Time Reference Station (PTRS) in nanoseconds.
Figure 4 - Block Diagram of the Control Theory Steering Procedures.
UTCCUSNO HO - UTCCNBS) via egt. GPS C-V
SEPT, and OCT. 1986
Log TAU (Mcondi)
Figure 5 - Measured Frequency Stabilities between UTC(USNO, MC) and UTC(NBS)
Simulated USNO - OCS (1 /f FM at 2 and 4 E214)
T -’ ' --
1 S 6 7 a
Log TAU (••Condi)
Figure 6 - Simulated Frequency Stabilities between UTC(USNO) and GPS Time.
SIMULATED STEERED NICROSTEPPER OUTPUT UITH 2 AND 4 E-14 1/f FM
FOR HC AND OCS CLOCK C»-2,S. l-S.4)
Log TAU (seconds)
Figure 7 - Frequency Stability of the Steered Microstepper Output under worst case
SIMULATED ERROR SI6NAL UITH 2 AND 4 E-14 1/f FH
FOR HC RND OCS CLOCK (■-2.B. l-S.4)
Log TAU (seconds)
Figure 8 - Simulated Error Signal between UTC(USNO) and GPS Time for 1/f FM noise of
4x10-14 (worst case condition).
SIMULATED STEERED HICROSTEPPER OUTPUT HITH 1 AND 1 E-11 J/f FM
FOR MC AND OCt CLOCK Ca-2.E. l-S.IJ
0 v vr -
Log TAU (seconds)
Figure 9 - Frequency Stability of the Steered Microstepper Output under best case
SIMULATED ERROR SI6NAL UXTH 1 RND 1 E-11 1/f PM
FOR MC RND DCS CLOCK (»-2.E. l-S.1)
Log TAU (seconds)
Figure 10 - Simulated Error Signal between UTC(USNO) and GPS Time for 1/f FM noise
of 1x10-14 (best case condition).
SIMULATED SERVO ERROR SIGNAL FOLLOWING TRANSIENT
FREOUENCT AMD TIME STEFA
> 111 .
4GGSS. 4S7EB. 4SS6I. MIES. MSGS.
Figure 11 - Simulated response to a beginning time step and a frequency error