theuseofgpsdisciplinedoscillators ...the gps satellite constellation. figure2.the gps signal...

56 | MEASURE www.ncsli.org

TECHNICAL PAPERS

1. IntroductionQuartz, rubidium, and cesium oscillators have historically beenthe three types of frequency standards [1, 2, 3] used by calibra-tion laboratories. Quartz oscillators are the least expensivechoice; rubidium oscillators and cesium oscillators are atomicdevices that cost more, but require less adjustment and performmuch better over long time periods. In recent years, however, afourth type of frequency standard, known as a Global Position-ing System disciplined oscillator (GPSDO), has been acquired

by many calibration and metrology laboratories and is some-times used as their primary standard1 for frequency. These stan-dards are quartz or rubidium oscillators whose frequency iscontrolled by signals broadcast from the GPS satellites.GPS, well known as a versatile, global tool for positioning and

The Use of GPS Disciplined Oscillatorsas Primary Frequency Standards forCalibration and MetrologyLaboratoriesMichael A. Lombardi

Abstract: An increasing number of calibration and metrology laboratories now employ a Global Positioning Systemdisciplined oscillator (GPSDO) as their primary standard for frequency. GPSDOs have the advantage of costing muchless 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 ofGPSDOs in calibration laboratories. It discusses how GPSDOs work, how measurement traceability can be establishedwith a GPSDO, and how their performance can vary significantly from model to model. It also discusses possibleGPSDO failure modes, and why a calibration laboratory must be able to verify whether or not a GPSDO is workingproperly.

Michael A. Lombardi

Time and Frequency DivisionNational Institute of Standards and Technology325 Broadway, Boulder, CO 80305 USAEmail: [email protected]

1 The term “primary standard” is sometimes reserved for a standardwhose value is accepted without reference to other standards thatproduce the same quantity. For example, cesium fountain standards(such as NIST-F1 in the United States) are currently recognized astrue primary frequency standards because their uncertainty can beestimated by summing or combining the effects of their frequencyshifts, without comparing them to other standards. However, the term“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 thatthe term is used throughout this paper.

MEASURE | 57Vol. 3 No. 3 • September 2008

TECHNICAL PAPERS

navigation, is also the main system usedto distribute high accuracy time and fre-quency worldwide. The GPS satellitesare controlled and operated by theUnited States Department of Defense(U.S. DoD). The GPS constellation(Fig. 1) always includes at least 24 satel-lites; with as many as eight operationalspares (31 satellites were usable as ofApril 2008). These satellites orbit theearth at a height of 20 200 km in sixfixed planes inclined 55° from theequator. The orbital period is 11 hoursand 58 minutes (half the length of thesidereal day), which means that eachsatellite passes over a given location onEarth four minutes earlier than it did onthe previous day. By processing signalsreceived from the satellites, even an inex-pensive handheld GPS receiver candetermine its position with an uncer-tainty of a few meters.The GPS satellites carry atomic oscilla-

tors that are steered from U.S. DoDground stations to agree withUTC(USNO), the Coordinated UniversalTime (UTC) time scale maintained by theUnited States Naval Observatory(USNO). UTC(USNO) and the NationalInstitute of Standards and Technology(NIST) time scale, UTC (NIST), are keptin close agreement and seldom differfrom each other by more than 20 ns. Theaverage frequency offset betweenUTC(USNO) and UTC(NIST) is nor-mally a few parts in 1015 or less over aone month interval (see http://tf.nist.gov/pubs/bulletin/nistusno.htmfor NIST/USNO comparison data).The GPS satellites currently broadcast

on two carrier frequencies: L1 at1.57542 GHz, and L2 at 1.2276 GHz(future GPS satellites will add additionalcarrier frequencies). Each satellite broad-casts a spread-spectrum waveform,called a pseudorandom noise (PRN)code on L1 and L2, and each satellite isidentified by the PRN code it transmits.There are two types of PRN codes. Thefirst type is a coarse acquisition (C/A)code with a chip rate of 1023 chips permillisecond (1.023 megabits/s). Thesecond type is a precision (P) code witha chip rate of 10230 chips per millisec-ond (10.230 megabits/s). The C/A codeis broadcast on L1, and the P code isbroadcast on both L1 and L2. A 50 bit/s

data message is also broadcast on bothcarriers. [4, 5, 6] Nearly all of theGPSDOs employed by calibration labo-ratories use the C/A code on the L1carrier as their incoming reference signal(Fig. 2).

2. How a GPSDO WorksThe basic function of a GPSDO is toreceive signals from the GPS satellites,

and to use the information contained inthese signals to control the frequency ofa local quartz or rubidium oscillator. Thesatellite signals can be trusted as a refer-ence for two reasons: (1) they originatefrom atomic oscillators, and (2) theymust be accurate in order for GPS tomeet its specifications as a positioningand navigation system. To illustrate this,consider that the oscillators onboard the

Figure 1. The GPS satellite constellation.

Figure 2. The GPS signal structure (L1 carrier, C/A code, and data message).

Carrier1.57542 GHz

1.023 Mbps

50 bps

20 ms+1

–1

–1

+1

+1 –1 +1

Repeating 1023 Chip Spread-Spectrum C/A Code (sent every millisecond

GPS Data Message


TECHNICAL PAPERS

GPS satellites receive clock corrections from earth-basedcontrol stations once during each orbit (about once every 12hours). The maximum acceptable contribution from the satelliteclocks to the positioning uncertainty is generally considered tobe about 1 m. Since light travels at about 3 × 108 m/s, the 1 mrequirement 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-tainty of less than 3.3 ns during the period between corrections.This translates to a frequency stability specification near 6 × 10-14.The goal of the GPSDO designer is to transfer the inherentaccuracy 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

oscillator to a local oscillator at a remote site has been of inter-est for decades, and has been approached in various fashions bydesigners of disciplined oscillators. Many of the approachesused to discipline oscillators are proprietary (some arepatented), and GPSDO manufacturers seldom disclose exactlyhow their products work. However, there are a few basic con-cepts that apply to most designs. Generally, the local oscillatoris controlled with one or more servo loops, with each loophaving a different time constant. [7] For example, one type ofservo loop is a phase locked loop, or PLL. In its basic form(Fig. 3), a PLL works by comparing the phase of a referenceinput signal to the phase of a voltage controlled oscillator(VCO). The phase detector then outputs the phase differencebetween the two input signals to a loop filter, which in turnsends a control voltage to the VCO. The control voltage changesthe frequency of the VCO in a direction that reduces the phasedifference between the VCO and the reference input signal. ThePLL is locked when the phase of the VCO has a constant offsetrelative to the phase of the input signal. [8]In a GPSDO, the reference input signal to the PLL comes from

a GPS receiver. Most GPSDOmanufacturers use a GPS receiverbuilt by a third party, because the cost of developing their ownreceiver is usually prohibitive. GPS receivers designed for timeand frequency applications (sometimes called “GPS timingengines”) have benefited from many years of research and devel-opment, and often cost less than $100 USD when purchased inquantity. These devices can track from 8 to 12 satellites, andoutput a 1 pulse per second (pps) signal synchronized toUTC(USNO). A simple GPSDO can be built by using a phasedetector to measure the difference between the 1 pps signal fromthe GPS receiver and the signal from the VCO. The VCO is typ-ically a 10 MHz oscillator, so its signal is divided to a lower fre-quency (often all the way down to 1 pps) prior to this phasecomparison. A microcontroller reads the output of the phasedetector and monitors the phase difference. When the phase dif-ference changes, the software changes the control voltage sent tothe VCO, so that the phase difference is held within a givenrange. Ideally, the software should smooth over the second-to-second fluctuations of the GPS signals, reducing the amount ofphase noise and allowing the VCO to provide reasonably goodshort-term frequency stability. However, the software must allowthe GPS signals to control the VCO frequency in the longer term.[9]

Adding software to the basic PLL design provides the loopwith the ability to vary its time constant and to automaticallyadapt to different input parameters. For example, if a morestable VCO were used, the software could adapt the servo loopto use a longer time constant and make frequency correctionsless often. Figure 4 shows a modified version of the basic PLLwhere the loop filter is replaced with a microcontroller whosesoftware compensates not only for the phase and frequencychanges of the local oscillator, but also for the effects of aging,temperature and other environmental parameters. [10]

Figure 3. A phase locked loop (PLL).

PhaseDetector

LoopFilter

VoltageControlledOscillator

ReferenceInput Signal

Output SignalPhase Lockedto Reference

Figure 4. Block diagram of a GPSDO that steers its localoscillator.

GPSReferenceSignal

PhaseDetector

DisciplinedFrequencyOutput

Steering software (controlsfrequency, compensates foraging, temperature, and otherenvironmental factors)

Microcontroller

VoltageControlledOscillator

Figure 5. Block diagram of a GPSDO that corrects the output of afrequency synthesizer.

GPSReferenceSignal

PhaseDetector

DisciplinedFrequencyOutput

Steering software (controlsfrequency, compensates foraging, temperature, and otherenvironmental factors)

Microcontroller

Free RunningLocalOscillator

FrequencySynthesizer


TECHNICAL PAPERS

The quality of the local oscillator largely determines howoften steering corrections are needed. For example, a rubidiumoscillator of high quality might change its frequency due toaging at a rate of less than 1 × 10-11 per month. [11] However,if an inexpensive quartz oscillator is used, it might age 1000times faster than a rubidium oscillator, so aging compensationwill be needed more often and the aging rate will be less pre-dictable. A similar situation exists with temperature, whererubidium oscillators tend to have much lower temperature coef-ficients and respond in a more predictable fashion than quartzoscillators to temperature changes. In spite of these differences,some GPSDOmanufacturers have designed adaptive algorithmsthat can compensate for the aging and temperature changes ofa wide variety of local oscillator types [12], thereby doing aremarkably good job with inexpensive quartz devices. Somealgorithms even “learn” and then store the characteristics of thelocal 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 isdiscussed 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 freerunning local oscillator is sent to a frequency synthesizer. Thesteering corrections are then applied to the output of the synthe-sier (Fig. 5). Modern direct digital synthesizers (DDS) haveexcellent resolution and allow very small frequency correctionsto be made. For example, a 48-bit DDS can provide sub-micro-hertz resolution at 10 MHz (1 µHz resolution at 10 MHz allowsinstantaneous frequency corrections of 1 × 10-13). In addition,allowing the local oscillator to free run often results in betterperformance than the VCO method, where unexpected shifts inthe control voltage can produce unwanted adjustments in theoutput frequency. [10, 13]As this discussion has illustrated, GPSDOs are sophisticated

instruments, and a considerable amount of engineering efforthas gone into their design. However, they are still very easy forcalibration laboratory personnel to install and use. The most dif-ficult part of the installation is mounting a small antenna on arooftop location (Fig. 6) with a clear view of the sky. Theantenna should be located relatively close to the lab so thatsignal loss along the antenna cable can be minimized. Once theGPSDO is installed, it will normally begin surveying its antennaposition as soon as it is turned on. The survey is a one-timeprocess that typically lasts for several hours. When the antennasurvey is complete, the GPSDO is ready to use as a frequencyand time standard.Most GPSDOs produce 5 and/or 10 MHz sine wave signals

for use as a frequency reference, and also produce 1 pps signalsfor use as a time interval reference and for time synchronizationto UTC. Figure 7 shows a portion of the back panel of aGPSDO. This particular model has a built-in distribution ampli-fier with multiple 1 pps and 10 MHz outputs.

3. GPSDO PerformanceThe design characteristics and performance of GPSDOs canvary significantly, particularly over short averaging times.Several published studies [14, 15, 16] have shown how differ-

ent models of GPSDOs produce different results, even whenoperated in identical environments. Even so, when averaging forperiods of several days or longer, any GPSDO that is locked tothe satellite signals should be inherently accurate (parts in 1013

or better) and inherently stable. This is because the signalsbroadcast by the GPS satellites are continuously steered to agreewith Coordinated Universal Time (UTC), and GPSDOs thatsimply “follow” the satellites will closely agree in both time andfrequency with UTC.From the point of view of a calibration laboratory, the most

important specification of a GPSDO is probably frequency accu-racy over a one day time period, because most frequency calibra-tions last for one day or less. The frequency accuracy can be nobetter than the stability, so a reasonably good metric to use whenevaluating a GPSDO is its frequency stability after one day ofaveraging, as estimated with the Allan deviation (ADEV). [17]Figure 8 shows the estimated frequency stability at one day forseven different GPSDO models that were calibrated by NIST.

Rubidium Local Oscillator

Quartz Local Oscillator

Figure 8. Comparison of the frequency stability of seven differentGPSDOs.

Figure 6. GPS antennas installed on the roof of a calibrationlaboratory.

Figure 7. Back panel of a GPSDO with multiple output signals.G

PS

DO

Und

erTe

st

Allan deviation after 1 day of averaging (parts in 1013)0 1 2 3 4 5 6 7 8

GPSDO-7

GPSDO-6

GPSDO-5

GPSDO-4

GPSDO-3

GPSDO-2

GPSDO-1


TECHNICAL PAPERS

The ADEV estimates at one day rangefrom about 7 × 10-13 to about 6 × 10-14.A stability of 1 × 10-13 or less after oneday of averaging normally indicates adevice of very high quality, and wasachieved by two of the seven devices cal-ibrated. As Fig. 8 indicates, the GPSDOsthat employ a rubidium local oscillator(blue bars) do not always perform betterthan those that employ a quartz localoscillator (gray bars), even though therubidium based units typically cost sub-stantially more and have the technicaladvantages discussed earlier.The performance differences between

GPSDOs become more obvious when thereceived phase data are looked at closely.To illustrate this, Fig. 9 shows phase data

(one-hour averages) from the 10 MHzoutputs of two GPSDOs, as compared toUTC(NIST), for a period of 80 days. Bothdevices have rubidium local oscillators ofsimilar quality, and both cost approxi-mately $10,000 USD. During the test,both GPSDO devices were connected tothe same GPS antenna using an antennasplitter. The antenna’s position had previ-ously been surveyed with an uncertaintyof less than 1 m, and these precise coor-dinates were keyed into both units.The results show that the frequency

output of Device A was very tightly con-trolled. The peak-to-peak phase variationover the entire 80 day period was just38 ns, with most of this variation due tothe difference between UTC(USNO) and

UTC(NIST) during this same interval.The frequency accuracy, as estimatedfrom the slope of the phase, was about1 × 10-15. In sharp contrast, the fre-quency of Device B was very loosely con-trolled (the servo loop apparently has avery long time constant). The phase plotshows a very large peak-to-peak phasevariation of 588 ns, much larger than thedispersion of the GPS timing signals.During the first 40 days of the measure-ment, the rubidium oscillator insideDevice B was allowed to run withminimal frequency correction, althoughthere was clearly some compensation forthe aging rate. The frequency accuracyduring this segment was about 150 × 10-15.During the second 40 days, the slope ofthe phase changed at least once every fewdays, and the average frequency offsetwas just a few parts in 1015. This issomewhat misleading, however, becausethe level of phase noise was much higherthan that of Device A.Figure 10 shows the long-term fre-

quency stability of both devices as esti-mated with ADEV, for averaging timesranging from 1 hour to about threeweeks. Device A is more stable thanDevice B at all averaging times byroughly a factor of 10. Stability at oneday, the key metric discussed earlier, isabout 6 × 10-14 for Device A and about70 × 10-14 for Device B, representing thebest and worst performance valuesshown in Fig. 4.Figure 11 shows the short-term fre-

quency stability of both devices for aver-aging times ranging from 1 second to100 seconds. The two devices haveessentially equivalent stability out toabout five or six seconds of averaging,before any of the steering loops areimplemented (as previously noted, therubidium local oscillators in the twodevices are similar). However, Device B’sstability was more than a factor of twoworse than Device A after 30 seconds ofaveraging, as one of its servo loopsapparently has a short time constant andhad already begun steering. After 100seconds of averaging, both devices arestable to about 1 × 10-12, but as Fig. 6indicates, Device B was not able toachieve this level of stability again untilthe averaging time reached about oneday.

Device A

Device B

Figure 9. Phase comparison of two GPSDOs to UTC(NIST).

Device A

Device B

Figure 10. Long-term frequency stability of two GPSDOs.

100

Nan

ose

cond

sP

erD

ivis

ion

Modified Julian Dates (09/05/2007 to 11/23/2007, 80 days)

5434

0

5435

0

5436

0

5437

0

5438

0

5439

0

5440

0

5441

0

5442

0

5443

0

5444

0

Alla

nD

evia

tion

1.0E-10

1.0E-11

1.0E-12

1.0E-13

1.0E-14

1.0E-15

Averaging Time (Seconds)

1.0E+03 1.0E+04 1.0E+05 1.0E+06 1.0E+07


TECHNICAL PAPERS

To be fair, the wide disparity betweenDevice A and Device B probably comesclose to representing the two extremes ofGPSDO performance. Device B waschosen for this example because of itsunusually loose steering of its local oscil-lator, and Device A was chosen becauseof its excellent all-around performance.These examples are simply intended toshow that two different GPSDOs canproduce very different results, even whenconnected to the same antenna andoperated in the same environment. Evenso, the frequency accuracy and stability ofall GPSDOs should be less than 1 × 10-12

at one day, improving over longer inter-vals. This level of performance exceedsthe measurement requirements of mostcalibration laboratories.

4. Choosing Between a GPSDOand a Rubidium or CesiumStandard

When a calibration laboratory decideswhich primary frequency standard tobuy, it will likely be choosing between arubidium oscillator, a cesium oscillator,or a GPSDO. While many calibrationlaboratories now employ GPSDOs astheir primary standard, some calibrationlaboratories have excluded them fromconsideration. Two of the chief reasonsfor not selecting a GPSDO are (1) con-cerns about failures due to the loss ofGPS reception (Section 5) and (2) con-cerns about traceability (Section 6).Another concern is that some calibrationlaboratories prefer to have a standardwhose frequency can be adjusted andcontrolled by calibration laboratory per-sonnel, such as a rubidium or cesium,rather than a GPSDO that is adjusted bysignals from the satellites. In addition,the short-term stability of some GPSDOscan be poor when compared to that offree running oscillators, due to the fre-quency or phase steps that are intro-duced when the local oscillator is steeredto agree with the satellites.For now, we’ll put aside the concerns

of GPS failure modes and traceability (tobe discussed in Sections 5 and 6), andfocus on the performance characteristicsof GPSDOs as compared to those ofrubidium and cesium oscillators (sum-marized in Table 1). The specificationslisted in the table were obtained from

specification sheets (at least several com-mercially available standards werereviewed in each category), and from theresults of measurements performed byNIST.As Table 1 indicates, a GPSDO (which

often has a rubidium inside) will outper-form a standalone rubidium oscillator.The long-term frequency accuracy andstability of the GPSDO will be much

better than that of a standalone rubid-ium, and, unlike the standalone rubid-ium, the GPSDO will never requireadjustment. A GPSDO will normally costmore than a standalone rubidium stan-dard, but in most cases, the performanceand convenience of the GPSDO willeasily justify the higher cost. Thereforefor most calibration laboratories aGPSDO is probably a better choice, but

Oscillator Type Rubidium Cesium GPSDO

Frequency offset withrespect to UTC(NIST)(1 day average)

5 × 10-9 to5 × 10-12

1 × 10-12 to5 × 10-14

1 × 10-12 to5 × 10-14

Stability at 1second

5 × 10-11 to5 × 10-12

5 × 10-11 to5 × 10-12

1 × 10-10 to1 × 10-12

Stability at 1day

5 × 10-12 8 × 10-14 to2 × 10-14

8 × 10-13 to5 × 10-14

Aging/year < 1 × 10-10 to5 × 10-10

None, by definition.However, the frequencyof a cesium oscillatordrifts by a small amount(typically by parts in1017 over the course of aday).

None, the output issteered to compensatefor aging and frequencydrift.

Phase noise(dbc/Hz, 10 Hzfrom carrier)

-90 to-130

-130 to-136

-90 to-140

Life expectancy > 15 years 5 to 20 years;10 years is typical

> 15 years

Produces an ontimepulse without beingsynchronized toanother source?

No No Yes

Produce frequencyaccurate to within±1 × 10-11 for 24hours or longer?

Yes, with periodicadjustment

Yes Yes

Cost (USD) $2,000 to$10,000

$30,000 to$75,000

$3,000 to$15,000

Table 1. Typical performance characteristics of calibration laboratory primary frequency.

Device B

Device A

Figure 11. Short-term frequency stability of two GPSDOs.

Alla

nD

evia

tion

1.0E-11

1.0E-12

Averaging Time (Seconds)

1 10 100


TECHNICAL PAPERS

some calibration laboratories still prefera standalone rubidium oscillator, due tosome of the concerns discussed earlier.The choice between a cesium standard

and a GPSDO is more difficult. The SIsecond is defined as 9 192 631 770energy transitions of the cesium atom;and thus cesium oscillators are intrinsicstandards. This normally makes them thepreferred choice of frequency standardfor calibration laboratories with thehighest capabilities and most demandingrequirements. However, not all calibra-tion laboratories can afford a cesiumstandard. They typically cost at least$30,000 USD per unit, and their beamtubes eventually run out of cesium, typi-cally after about 10 years. [18] The costof replacing a beam tube is often abouthalf the purchase price of the cesiumstandard itself, so the cost of ownershipis much higher than that of a GPSDO.Assuming that a calibration laboratory

can afford a cesium standard, should theystill save money by choosing a GPSDO astheir primary standard? There are severalpros and cons related to GPSDOs thatshould be considered before answeringthis question. First the pros:• A GPSDO costs much less than acesium standard to initially purchase,typically 50 % to 90 % less. It alsocosts less to own, because there is nocesium beam tube to replace. Thismeans that a calibration laboratorycould buy two or more GPSDOs forless than the cost of a cesium standard,and use the additional standards forcrosschecks and redundancy.

• Unlike a cesium standard, a GPSDOcan recover time by itself (time-of-dayand an on-time pulse synchronized toUTC). This is important if a calibra-tion laboratory needs time synchro-nization capability.

• Cesium standards seldom requireadjustment, but a GPSDO will neverrequire adjustment, since its frequencyis controlled by the signals from theGPS satellites.

Now the cons:• GPSDOs generally have poorer short-term stability and higher phase noisethan cesium standards.

• GPSDOs require an outdoor antennathat must be located in an area with

access to the roof. A cesium standardcan be operated anywhere where elec-tric power is available.

• Cesium standards are autonomous andindependent sources of frequency,which means they can operate withoutinput from another source. A GPSDOcan operate properly only wheresignals from the GPS satellites areavailable, and will not meet therequirements of calibration laborato-ries that need an autonomous fre-quency source.Based on these criteria, it seems that a

certain percentage of calibration laborato-ries will require a cesium standard, andwill continue to purchase them in spite oftheir higher costs. Conversely, some cali-bration laboratories that can afford acesium standard will undoubtedly choosea GPSDO as a lower cost alternative thatmeets all of their requirements. In addi-tion, some calibration laboratories willoperate both types of standards. A calibra-tion laboratory that already operates acesium standard as its primary standardmight be wise to acquire a GPSDO as asecondary standard, or as a check stan-dard, that they can use to ensure that theircesium standard is operating properly.

5. GPSDO Failure ModesAs is the case with cesium standards,GPSDOs tend to be trusted unequivo-cally, even when they have stoppedworking. Because they work so wellwithout ever requiring adjustment,GPSDOs tend to be checked even lessoften than cesium standards, with somecalibration laboratories allowing them torun for months or even years without anyattention. To guard against trusting theoutput of a failed device, calibration lab-oratories that use a GPSDO as theirprimary standard must have a procedure

in place that verifies whether the deviceis working properly. This proceduremight involve periodically checking thefront panel lights and indicators to verifywhether or not the unit is locked, andcomparing the outputs of the GPSDO toother standards to check for abnormalbehavior. It might also involve using acomputer to monitor the number ofsatellites being tracked, the receivedsignal strength (correlator-to-noiseratio), the health of the local oscillator,and so on. [19, 20]GPSDOs can and do fail, particularly

when the GPS signal is unavailable in alocal area. There are many possiblefailure modes that have been well docu-mented elsewhere [21], but the mostlikely cause of failure is probably RFinterference and jamming (either inten-tional or unintentional). GPS signals arevery susceptible to interference due totheir low power levels. A receiver canlose lock on a satellite due to an interfer-ing signal that is only a few orders ofmagnitude more powerful than theminimum received GPS signal strength,which is –160 dBW on earth for the L1carrier, equivalent to 10-16 W. [22] One“jamming” incident at NIST was causedby a GPS receiving antenna with a looseconnector. The signals leaking from thisconnector jammed other receivers whoseantennas were located 100 meters away.[23]When the GPS signal is unavailable, a

GPSDO continues to produce frequencybut begins relying on its holdover capa-bility. The holdover capability is providedby either a free running local oscillator,or a local oscillator that is steered withsoftware that retains knowledge of itspast performance. There is no exactanswer as to how long GPSDOs can con-tinue to meet the requirements of the cal-

Table 2. Holdover performance of four GPSDOs.

GPSDODevice Type

Frequencyaccuracy duringone week ofholdover

Time offset afterone week ofholdover

A Rubidium 80 × 10-12 42 µs

B Rubidium 3 × 10-12 < 3 µs

C Rubidium 1000 × 10-12 637 µs

D Quartz 300 × 10-12 82 µs


TECHNICAL PAPERS

ibration laboratory in the absence of GPSsignals. It depends entirely on the spe-cific model of GPSDO in use, and man-ufacturers often do not provide guidanceor holdover specifications.A holdover experiment was conducted

at the NIST laboratories in Boulder, Col-orado in October 2006. [24] This simpletest consisted of removing the antennasfrom four GPSDOs that had been contin-uously running for weeks or months, andleaving the antennas disconnected for aweek. The frequency accuracy of eachdevice was measured during the“outage”, as well as the time offset afterone week of holdover (Table 2).The NIST test was limited to four

devices that were available and was cer-tainly not representative of the entireGPSDO marketplace. All other thingsbeing equal, a rubidium GPSDO shouldhave better holdover capability than aquartz-based model, but this simple testshowed that at least one rubidium-basedGPSDO (Device C) had no holdoversteering algorithm in place. Figure 12shows a phase plot of Device C beforeand after its antenna was disconnected.Device C almost immediately became afree running oscillator with frequencyaccuracy near 1000 × 10-12, which istypical of an unadjusted rubidium. Theperformance of a quartz-based GPSDOwithout holdover capability would likelybe 10 to 100 times worse than that of arubidium GPSDO when GPS signals areunavailable. In sharp contrast, the fre-quency of Device B (Table 2) remainedaccurate to 3 × 10-12 during the weeklong outage, only three times worse thanthe 1 × 10-12 specification claimed bymany GPSDO manufacturers when theirdevice is working normally.In addition to several incidents of

jamming, NIST calibration customershave experienced GPS receivers failingfor other reasons, including: local oscilla-tor failures; antennas falling off the roofduring high wind conditions; antennacables being cut by repairmen; antennacables being gnawed through by squirrelsand other animals; and even one unusualincident where a trespasser with a rifleused a GPS antenna for target practice.Although the GPS system has proven tobe exceptionally reliable, there have beenrare instances where one or more GPS

satellites have broadcast bad timinginformation. Needless to say, it is impor-tant for a calibration laboratory to beable to verify that its GPSDO is workingproperly, and especially to know whetherit has stopped working.

6. Establishing MeasurementTraceability with a GPSDO

The use of GPSDOs as primary stan-dards in calibration laboratories is nowwidely accepted by most, but their use isstill questioned in some quarters. A fewdetractors claim that GPSDOs cannot beused to establish measurement traceabil-ity, but this is simply not true. In fact,because the time and frequency outputsof a GPSDO are continuously steered toagree with UTC, they will have betterlong-term accuracy and stability than anyfree running oscillator (including acesium standard). Therefore, from atechnical viewpoint, a well designedGPSDO should be able to deliver trace-able time and frequency measurementsas well or better than any standalone fre-quency standard. In theory, a GPSDO isa self-calibrating standard that neverrequires adjustment, because the adjust-ments are made automatically by thedevice itself, using information obtainedfrom the UTC signals broadcast by thesatellites.The key to establishing traceability

with a GPSDO is determining the meas-urement uncertainty that should beassigned to the GPSDO. As the VIM def-

inition of traceability states, establishingtraceability requires maintaining anunbroken chain of calibrations that traceback to the International System (SI)units of measurement. Each calibrationin the traceability chain must have aknown and stated uncertainty. While thisis a rigorous requirement, the process ofestablishing traceability with a GPSDOis no different than the process of estab-lishing traceability with a cesium stan-dard. For example, even though cesiumstandards are intrinsic standards used todefine the SI second, the uncertainty ofthe particular cesium device used by thecalibration laboratory still must be quan-tified and known in order to completethe traceability chain.How does a calibration laboratory

assign an uncertainty value to a GPSDO?There are at least three generalapproaches that can be used:1. Send the GPSDO out periodically for

calibration. This is the traditionalmodel for obtaining an uncertaintyvalue, widely used in most areas ofmetrology. A calibration laboratorycan send a GPSDO to its nationalmetrology institute (NMI), which isNIST in the United States, and haveit calibrated against the nationalstandard. [14, 15, 16, 25] Eventhen, however, traceability would beestablished only at a given point intime, and would eventually have tobe reestablished by another calibra-tion. For example, if a laboratory is

Figure 12. Phase plot of rubidium GPSDO without holdover capability during a simulatedsignal outage.

Tim

eE

rro

r(M

icro

seco

nds)

1700

1600

1500

1400

1300

1200

1100

1000

900

800

700

600

500

400

300

200

100

0

-100

Days (antenna disconnected at end of day 4)

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16


TECHNICAL PAPERS

applying for accreditation and an auditor were told that theuncertainty assigned to a GPSDO was obtained by a cali-bration from five years ago, they would likely agree that thetraceability chain was no longer valid. Therefore, calibra-tion laboratories that rely solely on this approach will haveto periodically schedule and pay for repeat calibrations.This is not an attractive option, because it negates one ofthe chief advantages of owning a “self-calibrating” stan-dard.

2. Assign an uncertainty, and then continuously verify thatboth the GPSDO and the GPS satellites are working prop-erly. A reasonable strategy for many calibration laboratoriesis to assign a measurement uncertainty to their GPSDOobtained from a previous calibration (see above), or fromthe manufacturer’s specification sheet. To ensure that theGPSDO is performing to this specification, the calibrationlaboratory needs to develop a procedure that verifies thatthe GPSDO is tracking satellites and working properly(Section 5). In addition, the calibration laboratory needs toverify that the GPS satellites are working properly, becauseerrors in the satellite broadcast could degrade the perform-ance of the GPSDO. To help calibration laboratories easilydetermine whether the satellites are working properly, NISTand other NMIs compare the GPS signals to their nationalfrequency standards, and publish the results on the Inter-net. The UTC(NIST) to GPS comparison results areupdated daily and archived at:http://tf.nist.gov/service/gpstrace.htm

3. Have the GPSDO continuously measured and monitored bya remote calibration service. NIST and other NMIs offerremote calibration services that make it possible for calibra-tion laboratories to continuously compare a GPSDO to thenational frequency standard so that its uncertainty is knownat all times. NIST offers two remote calibration servicesthat are suitable for continuous measurement of a GPSDO.The Frequency Measurement and Analysis Service (FMAS)can calibrate up to five frequency standards at once with anuncertainty of 2 × 10-13 at one day. The measurementresults can be viewed on the FMAS display, and calibrationreports are mailed to customers every month. [26] TheTime Measurement and Analysis Service (TMAS) canmeasure a 1 Hz signal timing pulse from a single standardwith an frequency uncertainty of 5 × 10-14 at one day. Inaddition to this lower uncertainty, the TMAS has two otheradvantages: it can measure the absolute timing accuracy ofa GPSDO with an uncertainty of less than 15 ns (the FMASmeasures frequency only), and its customers can view theirmeasurement results in real-time via the Internet. [27] Boththe FMAS and TMAS offer convenient, turnkey solutions tocalibration laboratories, by providing continuous validationof the frequency traceability chain.

7. Summary and ConclusionsGPS disciplined oscillators provide excellent performance at arelatively low cost, and have gained widespread acceptance asprimary frequency standards in calibration and testing labora-tories. Laboratories that employ GPSDOs as their primary stan-

dard can achieve frequency calibration and measurement capa-bilities near 1 × 10-13 after one day of averaging, but must estab-lish a procedure that verifies that the GPSDO is workingproperly and that the traceability chain is intact.

8. AcknowledgementsThe author thanks Tom Parker, Richard Fox, and David Smithof NIST for several helpful comments and corrections regardingthis manuscript, and Andrew Novick of NIST for his contribu-tions to the GPSDO measurements.This paper is a contribution of the United States government

and is not subject to copyright.

9. References[1] H. Hellwig, “Frequency Standards and Clocks: A Tutorial Intro-

duction,” National Bureau of Standards Technical Note 616,72 p., June 1977.

[2] L.L. Lewis, “An Introduction to Frequency Standards,” Proceed-ings of the IEEE, vol. 79, no. 7, pp. 927–935, July 1991.

[3] J.R. Vig and A. Ballato, “Frequency Control Devices,” UltrasonicInstruments and Devices, chapter 7, pp. 637–701, AcademicPress, 1999.

[4] M.A. Lombardi, L.M. Nelson, A.N. Novick, and V.S. Zhang, “Timeand frequency measurements using the Global PositioningSystem,” Cal Lab: International Journal of Metrology, vol. 8, no.3, pp. 26–33, July-September 2001.

[5] T.E. Parker and D. Matsakis, “Time and Frequency Dissemination:Advances in GPS Transfer Techniques,” GPS World, vol. 15,pp. 32–38, November 2004.

[6] J. Levine, “Time and frequency distribution using satellites,”Report on Progress in Physics, vol. 65, pp. 1119–1164, July 2002.

[7] D.W. Allan, L. Fey, H.E. Machlan, and J.A. Barnes, “An Ultra-Precise Time Synchronization System Designed by Computer Sim-ulation,” Frequency, vol. 6, pp. 11–14, January 1968.

[8] F.M. Gardner, Phaselock Techniques, John Wiley & Sons, NewYork, NY, 1966.

[9] B. Shera, “A GPS-Based Frequency Standard,” QST, pp. 37–44,July 1998.

[10] Hewlett-Packard Company, “HP SmartClock Technology,” HPApplication Note 1279, 32 p., March 1998.

[11] J.A. Kusters, “Determination of the Aging Rates of High StabilityRubidium Frequency Standards,” Second IEEE International Con-ference on Frequency Control and Synthesis, pp. 83–87, April1989.

[12] B.M. Penrod, “Adaptive Temperature Compensation of GPS Dis-ciplined Quartz and Rubidium Oscillators,” 1996 IEEE Interna-tional Frequency Control Symposium, pp. 980–987, June 1996.

[13] N.C. Helsby, “GPS Disciplined Offset-Frequency Quartz Oscilla-tor,” Proceedings of the 2003 IEEE Frequency Control Symposium,pp. 435–439, May 2003.

[14] J.A. Davis and J.M. Furlong, “Report on the study to determine thesuitability of GPS disciplined oscillators as time and frequencystandards traceable to the UK national time scale UTC(NPL),”National Physical Laboratory Report CTM 1, June 1997.

[15] V. Pettiti and F. Cordara, “Short-term characterization of GPS dis-ciplined oscillators and field trial of Italian calibration centers,”Proceedings of the Joint Meeting of the European Frequency and


TECHNICAL PAPERS

Time Forum and the IEEE International Frequency Control Sym-posium, pp. 404–407, April 1999.

[16] M.A. Lombardi, A.N. Novick, and V.S. Zhang, “Characterizing thePerformance of GPS Disciplined Oscillators with Respect toUTC(NIST),” Proceedings of the 2005 IEEE Frequency ControlSymposium and Precise Time and Time Interval (PTTI) Meeting,pp. 677–684, August 2005.

[17] IEEE, “IEEE Standard Definitions of Physical Quantities for Fun-damental Frequency and Time Metrology - Random Instabilities,”IEEE Standard 1139-1999, March 1999.

[18] J.A. Kusters, L.S. Cutler, and E.D. Powers, “Long-Term Experi-ence with Cesium Beam Frequency Standards,” Proceedings of the1999 IEEE Frequency Control Symposium and European Fre-quency and Time Forum (EFTF), pp. 159–163, April 1999.

[19] J. Stone, L. Lu, and P. Egan, “Calibrating Laser Vacuum Wave-length with a GPS-based Optical Frequency Comb,”Measure: TheJournal of Measurement Science, vol. 2, no. 4, pp. 28–38, Decem-ber 2007.

[20] M. Cunavelis, “Verification Philosophy of GPS Based FrequencyStandards,” Proceedings of the 2000 National Conference of Stan-dard Laboratories (NCSLI), July 2000.

[21] US Department of Transportation, Volpe Center, VulnerabilityAssessment of the Transportation Infrastructure Relying on theGlobal Positioning System, August 2001.

[22] ICD-GPS-200D, December 2004, available at:www.navcen.uscg.gov/gps/geninfo/IS-GPS-200D.pdf

[23] M.A. Lombardi and A.N. Novick, “Effects of the Rooftop Environ-ment on GPS Time Transfer,” Proceedings of the 2006 PreciseTime and Time Interval (PTTI) Meeting, pp. 449–465, December2006.

[24] M.A. Lombardi, “Comparing LORAN Timing Capability toIndustrial Requirements,” Proceedings of the 2006 InternationalLoran Association (ILA) Meeting, 14 p., October 2006.

[25] N. Goldovsky, “GPS Receiver Calibration: Why Is It Necessary?,”Cal Lab: International Journal of Metrology, vol. 14, no. 1, pp.36–43, January-March 2007.

[26] For information about the NIST FMAS, see:http://tf.nist.gov/service/fms.htm

[27] For information about the NIST TMAS, see:http://tf.nist.gov/service/tms.htm

theuseofgpsdisciplinedoscillators ...the gps satellite constellation. figure2.the gps signal...

Documents