TECHNICAL PAPERS

The Use of GPS Disciplined Oscillators

as Primary Frequency Standards for
Calibration and Metrology
Laboratories
Michael A. Lombardi
Abstract: An increasing number of calibration and metrology laboratories now employ a Global Positioning System
disciplined oscillator (GPSDO) as their primary standard for frequency. GPSDOs have the advantage of costing much
less than cesium standards, and they serve as self-calibrating standards that should not require adjustment or cali-
bration. These attributes make them an attractive choice for many laboratories. However, a few of their characteris-
tics can make a GPSDO less suitable than a cesium standard for some applications. This paper explores the use of
GPSDOs in calibration laboratories. It discusses how GPSDOs work, how measurement traceability can be established
with a GPSDO, and how their performance can vary significantly from model to model. It also discusses possible
GPSDO failure modes, and why a calibration laboratory must be able to verify whether or not a GPSDO is working
properly.

1. Introduction by many calibration and metrology laboratories and is some-

Quartz, rubidium, and cesium oscillators have historically been times used as their primary standard1 for frequency. These stan-
the three types of frequency standards [1, 2, 3] used by calibra- dards are quartz or rubidium oscillators whose frequency is
tion laboratories. Quartz oscillators are the least expensive controlled by signals broadcast from the GPS satellites.
choice; rubidium oscillators and cesium oscillators are atomic GPS, well known as a versatile, global tool for positioning and
devices that cost more, but require less adjustment and perform
much better over long time periods. In recent years, however, a
1 The term primary standard is sometimes reserved for a standard
fourth type of frequency standard, known as a Global Position-
whose value is accepted without reference to other standards that
ing System disciplined oscillator (GPSDO), has been acquired produce the same quantity. For example, cesium fountain standards
(such as NIST-F1 in the United States) are currently recognized as
Michael A. Lombardi true primary frequency standards because their uncertainty can be
Time and Frequency Division estimated by summing or combining the effects of their frequency
National Institute of Standards and Technology shifts, without comparing them to other standards. However, the term
325 Broadway, Boulder, CO 80305 USA primary standard is also commonly used to refer to the best stan-
dard available at a given laboratory or facility. It is in that sense that
Email: [email protected]
the term is used throughout this paper.

56 | MEASURE www.ncsli.org
 TECHNICAL PAPERS

navigation, is also the main system used

to distribute high accuracy time and fre-
quency worldwide. The GPS satellites
are controlled and operated by the
United States Department of Defense
(U.S. DoD). The GPS constellation
(Fig. 1) always includes at least 24 satel-
lites; with as many as eight operational
spares (31 satellites were usable as of
April 2008). These satellites orbit the
earth at a height of 20 200 km in six
fixed planes inclined 55 from the
equator. The orbital period is 11 hours
and 58 minutes (half the length of the
sidereal day), which means that each
satellite passes over a given location on
Earth four minutes earlier than it did on
the previous day. By processing signals
received from the satellites, even an inex-
pensive handheld GPS receiver can
determine its position with an uncer-
tainty of a few meters.
The GPS satellites carry atomic oscilla-
tors that are steered from U.S. DoD
ground stations to agree with
UTC(USNO), the Coordinated Universal
Time (UTC) time scale maintained by the
United States Naval Observatory
(USNO). UTC(USNO) and the National Figure 1. The GPS satellite constellation.
Institute of Standards and Technology
(NIST) time scale, UTC (NIST), are kept
20 ms
in close agreement and seldom differ +1
GPS Data Message
from each other by more than 20 ns. The
50 bps
average frequency offset between
1
UTC(USNO) and UTC(NIST) is nor-
mally a few parts in 1015 or less over a
+1 Repeating 1023 Chip Spread-Spectrum C/A Code (sent every millisecond
one month interval (see http://
1.023 Mbps
tf.nist.gov/pubs/bulletin/nistusno.htm
1
for NIST/USNO comparison data).
The GPS satellites currently broadcast
on two carrier frequencies: L1 at Carrier
1.57542 GHz, and L2 at 1.2276 GHz 1.57542 GHz

(future GPS satellites will add additional +1 1 +1

carrier frequencies). Each satellite broad-

casts a spread-spectrum waveform,
called a pseudorandom noise (PRN) Figure 2. The GPS signal structure (L1 carrier, C/A code, and data message).
code on L1 and L2, and each satellite is
identified by the PRN code it transmits. data message is also broadcast on both and to use the information contained in
There are two types of PRN codes. The carriers. [4, 5, 6] Nearly all of the these signals to control the frequency of
first type is a coarse acquisition (C/A) GPSDOs employed by calibration labo- a local quartz or rubidium oscillator. The
code with a chip rate of 1023 chips per ratories use the C/A code on the L1 satellite signals can be trusted as a refer-
millisecond (1.023 megabits/s). The carrier as their incoming reference signal ence for two reasons: (1) they originate
second type is a precision (P) code with (Fig. 2). from atomic oscillators, and (2) they
a chip rate of 10230 chips per millisec- must be accurate in order for GPS to
ond (10.230 megabits/s). The C/A code 2. How a GPSDO Works meet its specifications as a positioning
is broadcast on L1, and the P code is The basic function of a GPSDO is to and navigation system. To illustrate this,
broadcast on both L1 and L2. A 50 bit/s receive signals from the GPS satellites, consider that the oscillators onboard the

Vol. 3 No. 3 September 2008 MEASURE | 57

TECHNICAL PAPERS

GPS satellites receive clock corrections from earth-based

control stations once during each orbit (about once every 12 Reference
hours). The maximum acceptable contribution from the satellite Input Signal Phase Loop
Detector Filter
clocks to the positioning uncertainty is generally considered to
be about 1 m. Since light travels at about 3 108 m/s, the 1 m
requirement is equivalent to a time error of about 3.3 ns. Thus,
in order for the GPS system to meet its specifications, the satel-
lite clocks must be stable enough to keep time with an uncer- Voltage
tainty of less than 3.3 ns during the period between corrections. Controlled
Oscillator
This translates to a frequency stability specification near 6 10-14. Output Signal
The goal of the GPSDO designer is to transfer the inherent Phase Locked
to Reference
accuracy and stability of the satellite signals to the signals gen-
erated by the local quartz or rubidium oscillator.
The problem of transferring time and frequency from a master Figure 3. A phase locked loop (PLL).
oscillator to a local oscillator at a remote site has been of inter-
est for decades, and has been approached in various fashions by Steering software (controls
GPS
designers of disciplined oscillators. Many of the approaches Reference Phase
frequency, compensates for
aging, temperature, and other
environmental factors)
used to discipline oscillators are proprietary (some are Signal Detector
patented), and GPSDO manufacturers seldom disclose exactly Microcontroller
how their products work. However, there are a few basic con-
cepts that apply to most designs. Generally, the local oscillator
is controlled with one or more servo loops, with each loop
having a different time constant. [7] For example, one type of
servo loop is a phase locked loop, or PLL. In its basic form Voltage
Controlled
(Fig. 3), a PLL works by comparing the phase of a reference Disciplined
Oscillator
Frequency
input signal to the phase of a voltage controlled oscillator Output
(VCO). The phase detector then outputs the phase difference
between the two input signals to a loop filter, which in turn
sends a control voltage to the VCO. The control voltage changes Figure 4. Block diagram of a GPSDO that steers its local
the frequency of the VCO in a direction that reduces the phase oscillator.
difference between the VCO and the reference input signal. The
PLL is locked when the phase of the VCO has a constant offset Steering software (controls
GPS
relative to the phase of the input signal. [8] Reference Phase
frequency, compensates for
aging, temperature, and other
environmental factors)
In a GPSDO, the reference input signal to the PLL comes from Signal Detector
a GPS receiver. Most GPSDO manufacturers use a GPS receiver Microcontroller
built by a third party, because the cost of developing their own
receiver is usually prohibitive. GPS receivers designed for time
and frequency applications (sometimes called GPS timing
engines) have benefited from many years of research and devel-
Free Running
opment, and often cost less than $100 USD when purchased in Local
Frequency
Synthesizer Disciplined
quantity. These devices can track from 8 to 12 satellites, and Oscillator
Frequency
output a 1 pulse per second (pps) signal synchronized to Output
UTC(USNO). A simple GPSDO can be built by using a phase
detector to measure the difference between the 1 pps signal from
the GPS receiver and the signal from the VCO. The VCO is typ- Figure 5. Block diagram of a GPSDO that corrects the output of a
ically a 10 MHz oscillator, so its signal is divided to a lower fre- frequency synthesizer.
quency (often all the way down to 1 pps) prior to this phase
comparison. A microcontroller reads the output of the phase Adding software to the basic PLL design provides the loop
detector and monitors the phase difference. When the phase dif- with the ability to vary its time constant and to automatically
ference changes, the software changes the control voltage sent to adapt to different input parameters. For example, if a more
the VCO, so that the phase difference is held within a given stable VCO were used, the software could adapt the servo loop
range. Ideally, the software should smooth over the second-to- to use a longer time constant and make frequency corrections
second fluctuations of the GPS signals, reducing the amount of less often. Figure 4 shows a modified version of the basic PLL
phase noise and allowing the VCO to provide reasonably good where the loop filter is replaced with a microcontroller whose
short-term frequency stability. However, the software must allow software compensates not only for the phase and frequency
the GPS signals to control the VCO frequency in the longer term. changes of the local oscillator, but also for the effects of aging,
[9] temperature and other environmental parameters. [10]

58 | MEASURE www.ncsli.org
 TECHNICAL PAPERS

The quality of the local oscillator largely determines how

often steering corrections are needed. For example, a rubidium
oscillator of high quality might change its frequency due to
aging at a rate of less than 1 10-11 per month. [11] However,
if an inexpensive quartz oscillator is used, it might age 1000
times faster than a rubidium oscillator, so aging compensation
will be needed more often and the aging rate will be less pre-
dictable. A similar situation exists with temperature, where
rubidium oscillators tend to have much lower temperature coef-
ficients and respond in a more predictable fashion than quartz
oscillators to temperature changes. In spite of these differences,
some GPSDO manufacturers have designed adaptive algorithms
that can compensate for the aging and temperature changes of
a wide variety of local oscillator types [12], thereby doing a
remarkably good job with inexpensive quartz devices. Some Figure 6. GPS antennas installed on the roof of a calibration
laboratory.
algorithms even learn and then store the characteristics of the
local oscillator, allowing the local oscillator frequency to con-
tinue to be steered if the GPS input signal is temporarily lost.
This provides a GPSDO with holdover capability, a topic that is
discussed in more detail in Section 5.
Another type of GPSDO design does not correct the fre-
quency of the local oscillator. Instead, the output of a free
Figure 7. Back panel of a GPSDO with multiple output signals.
running local oscillator is sent to a frequency synthesizer. The
steering corrections are then applied to the output of the synthe-
sier (Fig. 5). Modern direct digital synthesizers (DDS) have GPSDO-7
Rubidium Local Oscillator
excellent resolution and allow very small frequency corrections Quartz Local Oscillator
GPSDO-6
to be made. For example, a 48-bit DDS can provide sub-micro-
GPSDO Under Test

hertz resolution at 10 MHz (1 Hz resolution at 10 MHz allows GPSDO-5

instantaneous frequency corrections of 1 10-13). In addition,

GPSDO-4
allowing the local oscillator to free run often results in better
performance than the VCO method, where unexpected shifts in GPSDO-3

the control voltage can produce unwanted adjustments in the GPSDO-2

output frequency. [10, 13]
As this discussion has illustrated, GPSDOs are sophisticated GPSDO-1

instruments, and a considerable amount of engineering effort 0 1 2 3 4 5 6 7 8

has gone into their design. However, they are still very easy for Allan deviation after 1 day of averaging (parts in 1013)
calibration laboratory personnel to install and use. The most dif-
ficult part of the installation is mounting a small antenna on a Figure 8. Comparison of the frequency stability of seven different
GPSDOs.
rooftop location (Fig. 6) with a clear view of the sky. The
antenna should be located relatively close to the lab so that ent models of GPSDOs produce different results, even when
signal loss along the antenna cable can be minimized. Once the operated in identical environments. Even so, when averaging for
GPSDO is installed, it will normally begin surveying its antenna periods of several days or longer, any GPSDO that is locked to
position as soon as it is turned on. The survey is a one-time the satellite signals should be inherently accurate (parts in 1013
process that typically lasts for several hours. When the antenna or better) and inherently stable. This is because the signals
survey is complete, the GPSDO is ready to use as a frequency broadcast by the GPS satellites are continuously steered to agree
and time standard. with Coordinated Universal Time (UTC), and GPSDOs that
Most GPSDOs produce 5 and/or 10 MHz sine wave signals simply follow the satellites will closely agree in both time and
for use as a frequency reference, and also produce 1 pps signals frequency with UTC.
for use as a time interval reference and for time synchronization From the point of view of a calibration laboratory, the most
to UTC. Figure 7 shows a portion of the back panel of a important specification of a GPSDO is probably frequency accu-
GPSDO. This particular model has a built-in distribution ampli- racy over a one day time period, because most frequency calibra-
fier with multiple 1 pps and 10 MHz outputs. tions last for one day or less. The frequency accuracy can be no
better than the stability, so a reasonably good metric to use when
3. GPSDO Performance evaluating a GPSDO is its frequency stability after one day of
The design characteristics and performance of GPSDOs can averaging, as estimated with the Allan deviation (ADEV). [17]
vary significantly, particularly over short averaging times. Figure 8 shows the estimated frequency stability at one day for
Several published studies [14, 15, 16] have shown how differ- seven different GPSDO models that were calibrated by NIST.

Vol. 3 No. 3 September 2008 MEASURE | 59

TECHNICAL PAPERS

UTC(NIST) during this same interval.

The frequency accuracy, as estimated
from the slope of the phase, was about
100 Nanoseconds Per Division

1 10-15. In sharp contrast, the fre-

quency of Device B was very loosely con-
trolled (the servo loop apparently has a
very long time constant). The phase plot
shows a very large peak-to-peak phase
variation of 588 ns, much larger than the
dispersion of the GPS timing signals.
Device A
During the first 40 days of the measure-
Device B ment, the rubidium oscillator inside
Device B was allowed to run with
minimal frequency correction, although
54340

54350

54360

54370

54380

54390

54400

54410

54420

54430

54440
there was clearly some compensation for
Modified Julian Dates (09/05/2007 to 11/23/2007, 80 days) the aging rate. The frequency accuracy
during this segment was about 150 10-15.
Figure 9. Phase comparison of two GPSDOs to UTC(NIST).
During the second 40 days, the slope of
the phase changed at least once every few
days, and the average frequency offset
1.0E-10
was just a few parts in 1015. This is
somewhat misleading, however, because
1.0E-11
the level of phase noise was much higher
than that of Device A.
Figure 10 shows the long-term fre-
1.0E-12 quency stability of both devices as esti-
Allan Deviation

mated with ADEV, for averaging times

ranging from 1 hour to about three
1.0E-13
weeks. Device A is more stable than
Device B at all averaging times by
roughly a factor of 10. Stability at one
1.0E-14
day, the key metric discussed earlier, is
Device A

Device B about 6 10-14 for Device A and about

1.0E-15 70 10-14 for Device B, representing the
1.0E+03 1.0E+04 1.0E+05 1.0E+06 1.0E+07 best and worst performance values
Averaging Time (Seconds) shown in Fig. 4.
Figure 11 shows the short-term fre-
Figure 10. Long-term frequency stability of two GPSDOs. quency stability of both devices for aver-
aging times ranging from 1 second to
The ADEV estimates at one day range (one-hour averages) from the 10 MHz 100 seconds. The two devices have
from about 7 10-13 to about 6 10-14. outputs of two GPSDOs, as compared to essentially equivalent stability out to
A stability of 1 10-13 or less after one UTC(NIST), for a period of 80 days. Both about five or six seconds of averaging,
day of averaging normally indicates a devices have rubidium local oscillators of before any of the steering loops are
device of very high quality, and was similar quality, and both cost approxi- implemented (as previously noted, the
achieved by two of the seven devices cal- mately $10,000 USD. During the test, rubidium local oscillators in the two
ibrated. As Fig. 8 indicates, the GPSDOs both GPSDO devices were connected to devices are similar). However, Device Bs
that employ a rubidium local oscillator the same GPS antenna using an antenna stability was more than a factor of two
(blue bars) do not always perform better splitter. The antennas position had previ- worse than Device A after 30 seconds of
than those that employ a quartz local ously been surveyed with an uncertainty averaging, as one of its servo loops
oscillator (gray bars), even though the of less than 1 m, and these precise coor- apparently has a short time constant and
rubidium based units typically cost sub- dinates were keyed into both units. had already begun steering. After 100
stantially more and have the technical The results show that the frequency seconds of averaging, both devices are
advantages discussed earlier. output of Device A was very tightly con- stable to about 1 10-12, but as Fig. 6
The performance differences between trolled. The peak-to-peak phase variation indicates, Device B was not able to
GPSDOs become more obvious when the over the entire 80 day period was just achieve this level of stability again until
received phase data are looked at closely. 38 ns, with most of this variation due to the averaging time reached about one
To illustrate this, Fig. 9 shows phase data the difference between UTC(USNO) and day.

60 | MEASURE www.ncsli.org
 TECHNICAL PAPERS

To be fair, the wide disparity between 1.0E-11

Device A and Device B probably comes
close to representing the two extremes of
GPSDO performance. Device B was
chosen for this example because of its
unusually loose steering of its local oscil-

Allan Deviation
lator, and Device A was chosen because
of its excellent all-around performance.
These examples are simply intended to
show that two different GPSDOs can
produce very different results, even when
connected to the same antenna and Device B

operated in the same environment. Even Device A

so, the frequency accuracy and stability of

1.0E-12
all GPSDOs should be less than 1 10-12 1 10 100
at one day, improving over longer inter- Averaging Time (Seconds)
vals. This level of performance exceeds
the measurement requirements of most Figure 11. Short-term frequency stability of two GPSDOs.
calibration laboratories.

Oscillator Type Rubidium Cesium GPSDO

4. Choosing Between a GPSDO
and a Rubidium or Cesium Frequency offset with 5 10-9 to 1 10-12 to 1 10-12 to
Standard respect to UTC(NIST) 5 10-12 5 10-14 5 10-14
(1 day average)
When a calibration laboratory decides
Stability at 1 5 10-11 to 5 10-11 to 1 10-10 to
which primary frequency standard to second 5 10-12 5 10-12 1 10-12
buy, it will likely be choosing between a Stability at 1 5 10-12 8 10-14 to 8 10-13 to
rubidium oscillator, a cesium oscillator, day 2 10-14 5 10-14
or a GPSDO. While many calibration Aging/year < 1 10-10 to None, by definition. None, the output is
laboratories now employ GPSDOs as 5 10-10 However, the frequency steered to compensate
of a cesium oscillator for aging and frequency
their primary standard, some calibration drifts by a small amount drift.
laboratories have excluded them from (typically by parts in
1017 over the course of a
consideration. Two of the chief reasons day).
for not selecting a GPSDO are (1) con- Phase noise -90 to -130 to -90 to
cerns about failures due to the loss of (dbc/Hz, 10 Hz -130 -136 -140
from carrier)
GPS reception (Section 5) and (2) con-
Life expectancy > 15 years 5 to 20 years; > 15 years
cerns about traceability (Section 6). 10 years is typical
Another concern is that some calibration
Produces an ontime No No Yes
laboratories prefer to have a standard pulse without being
whose frequency can be adjusted and synchronized to
another source?
controlled by calibration laboratory per-
Produce frequency Yes, with periodic Yes Yes
sonnel, such as a rubidium or cesium, accurate to within adjustment
rather than a GPSDO that is adjusted by 1 10-11 for 24
hours or longer?
signals from the satellites. In addition,
the short-term stability of some GPSDOs Cost (USD) $2,000 to $30,000 to $3,000 to
$10,000 $75,000 $15,000
can be poor when compared to that of
free running oscillators, due to the fre- Table 1. Typical performance characteristics of calibration laboratory primary frequency.
quency or phase steps that are intro-
duced when the local oscillator is steered specification sheets (at least several com- better than that of a standalone rubid-
to agree with the satellites. mercially available standards were ium, and, unlike the standalone rubid-
For now, well put aside the concerns reviewed in each category), and from the ium, the GPSDO will never require
of GPS failure modes and traceability (to results of measurements performed by adjustment. A GPSDO will normally cost
be discussed in Sections 5 and 6), and NIST. more than a standalone rubidium stan-
focus on the performance characteristics As Table 1 indicates, a GPSDO (which dard, but in most cases, the performance
of GPSDOs as compared to those of often has a rubidium inside) will outper- and convenience of the GPSDO will
rubidium and cesium oscillators (sum- form a standalone rubidium oscillator. easily justify the higher cost. Therefore
marized in Table 1). The specifications The long-term frequency accuracy and for most calibration laboratories a
listed in the table were obtained from stability of the GPSDO will be much GPSDO is probably a better choice, but

Vol. 3 No. 3 September 2008 MEASURE | 61

TECHNICAL PAPERS

some calibration laboratories still prefer Frequency

Time offset after
a standalone rubidium oscillator, due to GPSDO
Type
accuracy during
one week of
some of the concerns discussed earlier. Device one week of
holdover
holdover
The choice between a cesium standard
and a GPSDO is more difficult. The SI A Rubidium 80 10-12 42 s
second is defined as 9 192 631 770
B Rubidium 3 10-12 < 3 s
energy transitions of the cesium atom;
and thus cesium oscillators are intrinsic C Rubidium 1000 10-12 637 s
standards. This normally makes them the
D Quartz 300 10-12 82 s
preferred choice of frequency standard
for calibration laboratories with the Table 2. Holdover performance of four GPSDOs.
highest capabilities and most demanding
requirements. However, not all calibra- access to the roof. A cesium standard in place that verifies whether the device
tion laboratories can afford a cesium can be operated anywhere where elec- is working properly. This procedure
standard. They typically cost at least tric power is available. might involve periodically checking the
$30,000 USD per unit, and their beam Cesium standards are autonomous and front panel lights and indicators to verify
tubes eventually run out of cesium, typi- independent sources of frequency, whether or not the unit is locked, and
cally after about 10 years. [18] The cost which means they can operate without comparing the outputs of the GPSDO to
of replacing a beam tube is often about input from another source. A GPSDO other standards to check for abnormal
half the purchase price of the cesium can operate properly only where behavior. It might also involve using a
standard itself, so the cost of ownership signals from the GPS satellites are computer to monitor the number of
is much higher than that of a GPSDO. available, and will not meet the satellites being tracked, the received
Assuming that a calibration laboratory requirements of calibration laborato- signal strength (correlator-to-noise
can afford a cesium standard, should they ries that need an autonomous fre- ratio), the health of the local oscillator,
still save money by choosing a GPSDO as quency source. and so on. [19, 20]
their primary standard? There are several Based on these criteria, it seems that a GPSDOs can and do fail, particularly
pros and cons related to GPSDOs that certain percentage of calibration laborato- when the GPS signal is unavailable in a
should be considered before answering ries will require a cesium standard, and local area. There are many possible
this question. First the pros: will continue to purchase them in spite of failure modes that have been well docu-
A GPSDO costs much less than a their higher costs. Conversely, some cali- mented elsewhere [21], but the most
cesium standard to initially purchase, bration laboratories that can afford a likely cause of failure is probably RF
typically 50 % to 90 % less. It also cesium standard will undoubtedly choose interference and jamming (either inten-
costs less to own, because there is no a GPSDO as a lower cost alternative that tional or unintentional). GPS signals are
cesium beam tube to replace. This meets all of their requirements. In addi- very susceptible to interference due to
means that a calibration laboratory tion, some calibration laboratories will their low power levels. A receiver can
could buy two or more GPSDOs for operate both types of standards. A calibra- lose lock on a satellite due to an interfer-
less than the cost of a cesium standard, tion laboratory that already operates a ing signal that is only a few orders of
and use the additional standards for cesium standard as its primary standard magnitude more powerful than the
crosschecks and redundancy. might be wise to acquire a GPSDO as a minimum received GPS signal strength,
Unlike a cesium standard, a GPSDO secondary standard, or as a check stan- which is 160 dBW on earth for the L1
can recover time by itself (time-of-day dard, that they can use to ensure that their carrier, equivalent to 10-16 W. [22] One
and an on-time pulse synchronized to cesium standard is operating properly. jamming incident at NIST was caused
UTC). This is important if a calibra- by a GPS receiving antenna with a loose
tion laboratory needs time synchro- 5. GPSDO Failure Modes connector. The signals leaking from this
nization capability. As is the case with cesium standards, connector jammed other receivers whose
Cesium standards seldom require GPSDOs tend to be trusted unequivo- antennas were located 100 meters away.
adjustment, but a GPSDO will never cally, even when they have stopped [23]
require adjustment, since its frequency working. Because they work so well When the GPS signal is unavailable, a
is controlled by the signals from the without ever requiring adjustment, GPSDO continues to produce frequency
GPS satellites. GPSDOs tend to be checked even less but begins relying on its holdover capa-
often than cesium standards, with some bility. The holdover capability is provided
Now the cons: calibration laboratories allowing them to by either a free running local oscillator,
GPSDOs generally have poorer short- run for months or even years without any or a local oscillator that is steered with
term stability and higher phase noise attention. To guard against trusting the software that retains knowledge of its
than cesium standards. output of a failed device, calibration lab- past performance. There is no exact
GPSDOs require an outdoor antenna oratories that use a GPSDO as their answer as to how long GPSDOs can con-
that must be located in an area with primary standard must have a procedure tinue to meet the requirements of the cal-

62 | MEASURE www.ncsli.org
 TECHNICAL PAPERS

ibration laboratory in the absence of GPS

1700
signals. It depends entirely on the spe-
1600
cific model of GPSDO in use, and man- 1500
1400
ufacturers often do not provide guidance
1300
or holdover specifications. 1200

Time Error (Microseconds)

1100
A holdover experiment was conducted
1000
at the NIST laboratories in Boulder, Col- 900
orado in October 2006. [24] This simple 800
700
test consisted of removing the antennas 600
from four GPSDOs that had been contin- 500
400
uously running for weeks or months, and 300
leaving the antennas disconnected for a 200
100
week. The frequency accuracy of each 0
device was measured during the -100
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
outage, as well as the time offset after
Days (antenna disconnected at end of day 4)
one week of holdover (Table 2).
The NIST test was limited to four
devices that were available and was cer- Figure 12. Phase plot of rubidium GPSDO without holdover capability during a simulated
signal outage.
tainly not representative of the entire
GPSDO marketplace. All other things
being equal, a rubidium GPSDO should satellites have broadcast bad timing inition of traceability states, establishing
have better holdover capability than a information. Needless to say, it is impor- traceability requires maintaining an
quartz-based model, but this simple test tant for a calibration laboratory to be unbroken chain of calibrations that trace
showed that at least one rubidium-based able to verify that its GPSDO is working back to the International System (SI)
GPSDO (Device C) had no holdover properly, and especially to know whether units of measurement. Each calibration
steering algorithm in place. Figure 12 it has stopped working. in the traceability chain must have a
shows a phase plot of Device C before known and stated uncertainty. While this
and after its antenna was disconnected. 6. Establishing Measurement is a rigorous requirement, the process of
Device C almost immediately became a Traceability with a GPSDO establishing traceability with a GPSDO
free running oscillator with frequency The use of GPSDOs as primary stan- is no different than the process of estab-
accuracy near 1000 10-12, which is dards in calibration laboratories is now lishing traceability with a cesium stan-
typical of an unadjusted rubidium. The widely accepted by most, but their use is dard. For example, even though cesium
performance of a quartz-based GPSDO still questioned in some quarters. A few standards are intrinsic standards used to
without holdover capability would likely detractors claim that GPSDOs cannot be define the SI second, the uncertainty of
be 10 to 100 times worse than that of a used to establish measurement traceabil- the particular cesium device used by the
rubidium GPSDO when GPS signals are ity, but this is simply not true. In fact, calibration laboratory still must be quan-
unavailable. In sharp contrast, the fre- because the time and frequency outputs tified and known in order to complete
quency of Device B (Table 2) remained of a GPSDO are continuously steered to the traceability chain.
accurate to 3 10-12 during the week agree with UTC, they will have better How does a calibration laboratory
long outage, only three times worse than long-term accuracy and stability than any assign an uncertainty value to a GPSDO?
the 1 10-12 specification claimed by free running oscillator (including a There are at least three general
many GPSDO manufacturers when their cesium standard). Therefore, from a approaches that can be used:
device is working normally. technical viewpoint, a well designed 1. Send the GPSDO out periodically for
In addition to several incidents of GPSDO should be able to deliver trace- calibration. This is the traditional
jamming, NIST calibration customers able time and frequency measurements model for obtaining an uncertainty
have experienced GPS receivers failing as well or better than any standalone fre- value, widely used in most areas of
for other reasons, including: local oscilla- quency standard. In theory, a GPSDO is metrology. A calibration laboratory
tor failures; antennas falling off the roof a self-calibrating standard that never can send a GPSDO to its national
during high wind conditions; antenna requires adjustment, because the adjust- metrology institute (NMI), which is
cables being cut by repairmen; antenna ments are made automatically by the NIST in the United States, and have
cables being gnawed through by squirrels device itself, using information obtained it calibrated against the national
and other animals; and even one unusual from the UTC signals broadcast by the standard. [14, 15, 16, 25] Even
incident where a trespasser with a rifle satellites. then, however, traceability would be
used a GPS antenna for target practice. The key to establishing traceability established only at a given point in
Although the GPS system has proven to with a GPSDO is determining the meas- time, and would eventually have to
be exceptionally reliable, there have been urement uncertainty that should be be reestablished by another calibra-
rare instances where one or more GPS assigned to the GPSDO. As the VIM def- tion. For example, if a laboratory is

Vol. 3 No. 3 September 2008 MEASURE | 63

TECHNICAL PAPERS

applying for accreditation and an auditor were told that the dard can achieve frequency calibration and measurement capa-
uncertainty assigned to a GPSDO was obtained by a cali- bilities near 1 10-13 after one day of averaging, but must estab-
bration from five years ago, they would likely agree that the lish a procedure that verifies that the GPSDO is working
traceability chain was no longer valid. Therefore, calibra- properly and that the traceability chain is intact.
tion laboratories that rely solely on this approach will have
to periodically schedule and pay for repeat calibrations. 8. Acknowledgements
This is not an attractive option, because it negates one of The author thanks Tom Parker, Richard Fox, and David Smith
the chief advantages of owning a self-calibrating stan- of NIST for several helpful comments and corrections regarding
dard. this manuscript, and Andrew Novick of NIST for his contribu-
2. Assign an uncertainty, and then continuously verify that tions to the GPSDO measurements.
both the GPSDO and the GPS satellites are working prop- This paper is a contribution of the United States government
erly. A reasonable strategy for many calibration laboratories and is not subject to copyright.
is to assign a measurement uncertainty to their GPSDO
obtained from a previous calibration (see above), or from 9. References
the manufacturers specification sheet. To ensure that the [1] H. Hellwig, Frequency Standards and Clocks: A Tutorial Intro-
GPSDO is performing to this specification, the calibration duction, National Bureau of Standards Technical Note 616,
laboratory needs to develop a procedure that verifies that 72 p., June 1977.
the GPSDO is tracking satellites and working properly [2] L.L. Lewis, An Introduction to Frequency Standards, Proceed-
(Section 5). In addition, the calibration laboratory needs to ings of the IEEE, vol. 79, no. 7, pp. 927935, July 1991.
verify that the GPS satellites are working properly, because [3] J.R. Vig and A. Ballato, Frequency Control Devices, Ultrasonic
errors in the satellite broadcast could degrade the perform- Instruments and Devices, chapter 7, pp. 637701, Academic
ance of the GPSDO. To help calibration laboratories easily Press, 1999.
determine whether the satellites are working properly, NIST [4] M.A. Lombardi, L.M. Nelson, A.N. Novick, and V.S. Zhang, Time
and other NMIs compare the GPS signals to their national and frequency measurements using the Global Positioning
frequency standards, and publish the results on the Inter- System, Cal Lab: International Journal of Metrology, vol. 8, no.
net. The UTC(NIST) to GPS comparison results are 3, pp. 2633, July-September 2001.
updated daily and archived at: [5] T.E. Parker and D. Matsakis, Time and Frequency Dissemination:
http://tf.nist.gov/service/gpstrace.htm Advances in GPS Transfer Techniques, GPS World, vol. 15,
3. Have the GPSDO continuously measured and monitored by pp. 3238, November 2004.
a remote calibration service. NIST and other NMIs offer [6] J. Levine, Time and frequency distribution using satellites,
remote calibration services that make it possible for calibra- Report on Progress in Physics, vol. 65, pp. 11191164, July 2002.
tion laboratories to continuously compare a GPSDO to the [7] D.W. Allan, L. Fey, H.E. Machlan, and J.A. Barnes, An Ultra-
national frequency standard so that its uncertainty is known Precise Time Synchronization System Designed by Computer Sim-
at all times. NIST offers two remote calibration services ulation, Frequency, vol. 6, pp. 1114, January 1968.
that are suitable for continuous measurement of a GPSDO. [8] F.M. Gardner, Phaselock Techniques, John Wiley & Sons, New
The Frequency Measurement and Analysis Service (FMAS) York, NY, 1966.
can calibrate up to five frequency standards at once with an [9] B. Shera, A GPS-Based Frequency Standard, QST, pp. 3744,
uncertainty of 2 10-13 at one day. The measurement July 1998.
results can be viewed on the FMAS display, and calibration [10] Hewlett-Packard Company, HP SmartClock Technology, HP
reports are mailed to customers every month. [26] The Application Note 1279, 32 p., March 1998.
Time Measurement and Analysis Service (TMAS) can [11] J.A. Kusters, Determination of the Aging Rates of High Stability
measure a 1 Hz signal timing pulse from a single standard Rubidium Frequency Standards, Second IEEE International Con-
with an frequency uncertainty of 5 10-14 at one day. In ference on Frequency Control and Synthesis, pp. 8387, April
addition to this lower uncertainty, the TMAS has two other 1989.
advantages: it can measure the absolute timing accuracy of [12] B.M. Penrod, Adaptive Temperature Compensation of GPS Dis-
a GPSDO with an uncertainty of less than 15 ns (the FMAS ciplined Quartz and Rubidium Oscillators, 1996 IEEE Interna-
measures frequency only), and its customers can view their tional Frequency Control Symposium, pp. 980987, June 1996.
measurement results in real-time via the Internet. [27] Both [13] N.C. Helsby, GPS Disciplined Offset-Frequency Quartz Oscilla-
the FMAS and TMAS offer convenient, turnkey solutions to tor, Proceedings of the 2003 IEEE Frequency Control Symposium,
calibration laboratories, by providing continuous validation pp. 435439, May 2003.
of the frequency traceability chain. [14] J.A. Davis and J.M. Furlong, Report on the study to determine the
suitability of GPS disciplined oscillators as time and frequency
7. Summary and Conclusions standards traceable to the UK national time scale UTC(NPL),
GPS disciplined oscillators provide excellent performance at a National Physical Laboratory Report CTM 1, June 1997.
relatively low cost, and have gained widespread acceptance as [15] V. Pettiti and F. Cordara, Short-term characterization of GPS dis-
primary frequency standards in calibration and testing labora- ciplined oscillators and field trial of Italian calibration centers,
tories. Laboratories that employ GPSDOs as their primary stan- Proceedings of the Joint Meeting of the European Frequency and

64 | MEASURE www.ncsli.org
 TECHNICAL PAPERS

Time Forum and the IEEE International Frequency Control Sym- [21] US Department of Transportation, Volpe Center, Vulnerability
posium, pp. 404407, April 1999. Assessment of the Transportation Infrastructure Relying on the
[16] M.A. Lombardi, A.N. Novick, and V.S. Zhang, Characterizing the Global Positioning System, August 2001.
Performance of GPS Disciplined Oscillators with Respect to [22] ICD-GPS-200D, December 2004, available at:
UTC(NIST), Proceedings of the 2005 IEEE Frequency Control www.navcen.uscg.gov/gps/geninfo/IS-GPS-200D.pdf
Symposium and Precise Time and Time Interval (PTTI) Meeting, [23] M.A. Lombardi and A.N. Novick, Effects of the Rooftop Environ-
pp. 677684, August 2005. ment on GPS Time Transfer, Proceedings of the 2006 Precise
[17] IEEE, IEEE Standard Definitions of Physical Quantities for Fun- Time and Time Interval (PTTI) Meeting, pp. 449465, December
damental Frequency and Time Metrology - Random Instabilities, 2006.
IEEE Standard 1139-1999, March 1999. [24] M.A. Lombardi, Comparing LORAN Timing Capability to
[18] J.A. Kusters, L.S. Cutler, and E.D. Powers, Long-Term Experi- Industrial Requirements, Proceedings of the 2006 International
ence with Cesium Beam Frequency Standards, Proceedings of the Loran Association (ILA) Meeting, 14 p., October 2006.
1999 IEEE Frequency Control Symposium and European Fre- [25] N. Goldovsky, GPS Receiver Calibration: Why Is It Necessary?,
quency and Time Forum (EFTF), pp. 159163, April 1999. Cal Lab: International Journal of Metrology, vol. 14, no. 1, pp.
[19] J. Stone, L. Lu, and P. Egan, Calibrating Laser Vacuum Wave- 3643, January-March 2007.
length with a GPS-based Optical Frequency Comb, Measure: The [26] For information about the NIST FMAS, see:
Journal of Measurement Science, vol. 2, no. 4, pp. 2838, Decem- http://tf.nist.gov/service/fms.htm
ber 2007. [27] For information about the NIST TMAS, see:
[20] M. Cunavelis, Verification Philosophy of GPS Based Frequency http://tf.nist.gov/service/tms.htm
Standards, Proceedings of the 2000 National Conference of Stan-
dard Laboratories (NCSLI), July 2000.

Vol. 3 No. 3 September 2008 MEASURE | 65