Indian loumal of Radio & Space PhysicsVol. 26, August 1997, pp.228-236

Accurate on-line frequency calibration of a rubidium standard usingINSAT STFS broadcast in differential mode

A Sen Gupta', S Das/ & B S Mathur''Time & FrequencySection,NationalPhysicalLaboratory,New Delhi 110012

2ElectronicRegionalTest Laboratory(E),D. N. Block,Sector-V,Salt LakeCity, Calcutta700 091

Received 6 September 1996; revised received 17 April 1997

The Standard Time and Frequency Signal (STFS) broadcast via INSAT has been used to provide on-line frequency calibration traceable to the National Physical Laboratory (NPL), New Delhi. The methodconsists of tracking the phase of the received STFS relative to the local frequency standard. The slope ofthis phase variation gives the frequency offset. The actual experimental data used in this study consist ofthose collected simultaneously at the ERTL(East), Calcutta and at NPL. Although, individual data havesinusoidal diurnal residual errors due to satellite position prediction inaccuracy, these are mostlycommon mode errors that cancel in the differential mode, leading to the present high accuracy offrequency calibration of the order of a few parts in 10'2 over a measurement penod of few days.

1 IntroductionThe Electronic Regional Test Laboratones

(ERTL) under the STQC directorate of theDepartment of Electronics are mostly equipped WIthrubidium frequency standards as the inhousereference relative to which they perform all thefrequency calibrations. To establish traceability withthe national standard at the National PhysicalLaboratory (NPL), New Delhi, these rubidiumstandards are transported to NPL for frequencycalibration. The method of frequency calibrationused at NPL consists of tracking the phase of the 5MHz output of the rubidium device relative to thefrequency standard at NPL for about 10 days. Theslope of this variation then gives the frequencyoffset. While this calibration arrangement 'givesaccurate results for both short' term ('" few minutes)and longer ('" 1 day), it suffers from some basicdrawbacks. First, the rubidium clocks aretransported to and from NPL in shutdown or coldcondition. Thus, the frequency calibrationsperformed at NPL are not very strictly sustainedduring the transportation. Secondly, the overallduration of the whole process may turn out to beabout one month during which time the ERTL has todo without its frequency standard. Keeping theselimitations in mind it was proposed, in 1989, to usethe Standard Time and Frequency Signal (STFS)

broadcast via INSA T bemg transmitted by NPL.Primary advantage of this method is that it wouldprovide an on-line calibration which would removeboth the above mentioned limitations. Although theINSAT STFS has operated continuously sinceMarch, 1988, frequency calibration of ERTLrubidium clocks have been performed successfullyonly very recently. This IS due to the problemsencountered WIth the first generation STFSreceiving equipment available in the market and theprototype decoders developed at NPL for thispurpose. In the present paper, we present the firstresults of on-line frequency calibration of therubidium standard at ERTL(East), Calcutta, usingSTFS.

In the subsequent sections are discussed theINSA T STFS broadcast, explaining in particular thedifferential STFS concept. The baSIC concepts offrequency calibration are then briefly reviewed, Thisis followed by the experimental results anddiscussion on possibilities for future improvements.

2 INSA T STFS broadcastThe coded STFS are being broadcast on one of

the radio networking channels of INSA T sinceMarch 1988 (Ref 1).This service can be receivedall over the Indian subcontinent and can provide a

SEN GUPTA et al.: ON-LINE FREQUENCY CALIBRATJON USfNG INSAT STFS BROADCAST 229

time synchronization accuracy at about 10-20 IlS

level.'The STFS uplink to INSA T takes place from the

Delhi earth station (DES) at Sikandrabad (UP) about70 km east of Delhi. The channel has an uplink at5899.675 MHz and a downlink at 2599.675 MHz.'The RF bandwidth of the channel is 160 kHz andbaseband STFS is frequency modulated on thecarrier. The format of the time code which iscurrently operational is shown in Fig. 1. It consistsof a regular train of packets of 5 kHz sinusoidsoccuning at 100 pps. The packet width is binarymodulated in accordance with a code that carries thetime of day and instantaneous satellite positioninformation. The data transmission takes place at therate of 1 byte per second and the entire messagetakes one minute to be transmitted. The transmittedtime on STFS is derived from a cesium clock atDES which is kept synchronized to the NPL'satomic time scale through periodic portable clocktrips,

The receiving set-up is shown schematically inFig. 2. A chicken mesh antenna, 8 ft or larger indiameter, with a front end converter (FEC) mountedat its feed point, IS used for this purpose. The FEC

has to have a noise figure of 1.5 dB or better inorder to have an acceptable SIgnal to noise ratio. Thereceived signal is mixed with a quartz controlled LOof 2545 MHz, converting it into an IF of 54.675MHz which is then brought through a cable to afixed frequency FM receiver. The demodulatedoutput of this receiver IS the coded 5 kHz STFS. Thedecoding of the STFS is performed using a decoder

2599.675 MHz\4 8FT.• CHCKEN MESH

~~

~I'I I ~'RE:IVERSTfS

CORRECTED IPPSf. AMPUF1ER DECOOER[jrJ ".rn ••5 MHz 1 11 PI IS

OSCUATOR LOCAL254511lzfREQUENCYSTANDARD

ANnNNA IIOUNHD OONT [ND ANOCLOCK

Fig. 2--Schematic diagram of the receivmg setup for STFS atNPL and ERTL (East)

STFSFORM T·

88 89 90 91 92 93 94 95 96 97 98 99 0 I 2 3 4 5 6 7 I! 9 10 II 12 13 14

.~ '...: oj

1 BITSCONTD.

-PREAMBLE - Itp~ps= AAHEX ps I

MINUTEMARK

80DATA - SUCCESSIVEBYTE 1 BITS

10ms

I BIT - 7.5ms---+

I-J~rtJ\--2.5 m~OBIT

NO TIMEDATA 1ST

Ir, +--SATELLH y~ POSN. ---+HR MN XI X2 X3 X~i. ,', "3 Y4 ZI z: Z3 Z4

DATAFORMAT~~~~~-L~~~~~~~~_

o 14

!~ II SECONDS45 59

Fig. I--Fonnat of the transmitted STFS code via fNSAT

230 INDIAN J RADIO & SPACE PHYS, AUGUST 1997

which is microprocessor controlled and totallyautomatic", The basic function of this decoder is toextract accurately the 1 pulse per second (Pps),decode the time of day [in Indian Standard Time(1ST)] and the current satellite position coordinates. ~A 1 pps so obtained is, however, delayed from the ~

<4(

transmitted 1ST by an amount Tp equal to the sum of isthe propagation delays over the 'transmitter-to- r;;satellite', and the 'satellite-to-receiver' paths and a ~constant receiver delay. The co-ordinates of thetransmitter and the receiver are fixed andsupposedly known accurately. Thus, knowing thecurrent satellite position from the signal data it ispossible to compute Tp exactly. This computation ofTp is done by the microprocessor in the decoder inreal time and this information is fed to aprogrammable delay generator which then correctsfor the propagation delay. The output of the decoderis, thus, a corrected 1 pps which is synchronized tothe transmitted 1ST to an accuracy of about 10-20us. It has been observed that 1 pps so obtained isvery stable in the short term having jitter of less than1 us over 1-2 minutes. However, since the actualsynchronization accuracy finally obtained is only±1O-20 us, we have a built in time interval counterin the decoder which has a least count of only 1 us,This then limits the precision of the synchronizationmeasurements.

The primary cause for inaccuracy in the receivedtime through STFS is the satellite positionprediction which is carried out in the following way.The data on satellite orbit information are receivedfrom the INSAT master control facility (MCF) atHassan every 2-3 weeks. These data are then used asinput to a prediction programme' to generate hourlyvalues of the satellite position for the next 25 daysand loads the results into the STFS encoder at thetransmitting station. The encoder interpolatesbetween successive hourly values to transmitupdated satellite position every minute alongwiththe STFS.

Figure 3 illustrates typical results of timesynchronization using STFS broadcast as alreadydescribed. The data shown in Fig. 3 are residualtime errors, i. e. the difference between the NPLatomic time scale and the STFS time received andcorrected for propagation delay. As the cesiumclock at DES is synchronized with the NPL timescale, residuals in Fig. 3 represent inaccuracy of the

50r---~----~----~----~--~----,4030

8795 8805MJD- 40000

8815

Fig. 3-- Typical STFS monitoring data at NPL obtained usingthe Master cesium standard as the reference

STFS technique. It can be seen very clearly thatwhen new satellite orbital information was receivedfrom MCF, for example, on day 8792, the error issignificantly reduced. With passage of time,however, the satellite position prediction error andconsequently the inaccuracy increases. The nextinput of satellite orbital data on day 8811 againreduces the error.

The above level of accuracy of timesynchronization is generally adequate for most ofthe present users of the STFS broadcast. However,there exist a few users such as the Radio AstronomyGroup of TIFR, Bombay, who would require timesynchronization accuracy approaching a us or evenbetter. For such users we use a concept termeddifferential mode STFS (Ref. 4). In its basicessentials the method is simple and has been utilizedwith good success in the field of time transfer ascommon-view GPS technique" and passive TVtechnique via satellite'. It consists of the following:

The STFS time transfer is carried outsimultaneously at the given station and a referencestation, such as NPL, and the residual errorsrecorded. As observed before in Fig. 3, theinaccuracy in the time transfer is not a random jitterof the received signal, but systematic error in theform of a slow diurnal. This diurnal variation isprimarily a result of slight inaccuracy of the satelliteorbit modelling which results in error in thecomputed propagation delay. The distance of theground stations to the satellite is nearly 40000 kmcompared to a maximum of about 2000 km betweenany station in India. Thus the ray paths from any

SEN GUPTA et al.: ON-LINE FREQUENCY CALIBRATION USING INSAT STFS BROADCAST 231

two stations to the satellite subtend a small anglebetween them and most of the error is of commonmode. On subtracting the residuals recorded at NPLfrom those at the given station most of this commonmode inaccuracy can be cancelled out.

A simple theoretical expression for thedifferential STFS error between two stations A andB is given by"

t1 TA (DSTFS) = t1 TpA - t1 TpB

1 A A ••• (1)= - ( r A - rB).11S

c

where, rA and rB are unit vectors along the ray pathsfrom the satellite to the receiving station A and B,respectively. It has been shown by computermodelling that generally the differential error stayswithin 1 us over any point in India.

3 Frequency caHbrationIt is customary to characterize a high stability

oscillator in terms of its relative frequency offset (orerror), F, which is defined as?

F = (f - /0) = 6//0 /

... (2)

where,f is the frequency of the oscillator and fo is itsnominal frequency. We note that F is adimensionless quantity, being a ratio, and also thatthe F value for any frequency output obtained bymultiplication, division, etc. of the base oscillatorwould be the same. In view of this basic definition,it is found to be convenient to determine the relativefrequency offset of an oscillator by studying the 1pps clock derived from it. If we use a time transfertechnique, such as STFS, and monitor the clock driftfl.1 over a certain period of time T, the averagefrequency offset of the oscillator is then given by

[fl./[ = Ifl./I ... (3)

One has to take proper care to determine the sign ofthe offset by noting the direction of the clock drift.A clock which gains time has a positive F and onethat loses time has a negative F value. Gaining orlosing time are scientifically ambiguous terms; so itis necessary to define the concepts of clock offsetand drift. In algebraic notation we define the

difference II between a clock C and the referenceclock, say NPL, at a certain time instant TI as

... (4)

This expression implies that if a time intervalcounter is started with the 1 pps from C and stoppedwith 1 pps from NPL then the time interval readingis II' Further the drift of the clock C between times1'1 and Tz (Tz>TI) is M=(/2-/1)' Depending onwhether the drift fl.t is positive or negative, (theclock C is gaining or losing time) it has a positive ornegative frequency offset, respectively. The value ofoffset is the slope of the drift and is given by Eq.(2)with T=T2-TI•

4 Experimental results and discussionThe experiments on frequency calibration of the

rubidium standard at ERTL(East) using STFS werecarried out during 16-20 May 1994 in thepreliminary phase and more extensively during 11-24 July 1994. The experimental set-ups both atERTL and NPL were similar to that shown in Fig.2. The rubidium standard at ERTL and the mastercesium reference standard at NPL provided thelocal 1 pps clock time reference for the £TFSdecoder. The data were recorded automaticallyevery 10 min on a serial printer.

Figures 4-6 show the observations during thepreliminary phase. The x-axis is labelled in MeanJulian Days (MID) which corresponds to the periodindicated above. In Fig. 4 are shown the recordingsat ERTL which were made only between 10 AMand 6 PM during this phase. What appear as largediscontinuities in the slope in Fig. 4 are actually dueto the fact that the residuals are parts of a sinusoidalvariation as will be clear from an examination of theNPL data recorded simultaneously (shown in Fig.S). An overall positive drift is, however, clearlyvisible over the whole period indicating a frequencyoffset. One can also see a few outliers in Fig.4which appear to be correlated with each other. Theseabberations arise out of the receiver PLL going outof lock, once in a while, due to noise in the receivedsignal. The lock hunting process in the receivertakes it preferentially to one side. Now, in thisparticular case, it is just a matter of chance that theyappear correlated in Fig. 4. In Fig. S a small positivedrift is also visible but it clearly due to theinaccuracy of the satellite position prediction

232 INDIAN J RADIO & SPACE PHYS. AUGUST 1997

modelling, as the clocks at DES. Sikandrabad andNPL are synchronized. Finally, in Fig. 6 we haveinvoked the differential STFS concept bysubtracting the data in Fig. 5 from those in Fig. 4,thus eliminating most of the common mode errordue to satellite position prediction inaccuracy.

Barring a few outliers, all the data ofERTL(rubidium)-NPL, now fall close to a straightline indicatmg a linear positive drift. The slope ofthis drift is computed as 7.5 ± O.2xlO·\\ which is theaverage frequency offset of the ERTL rubidiumstandard during this penod.

201

'i,10 '\

0 \ "t

\\

-10 +"\

\ +

-20;

\ \'t \~ \'" "\:::I... -30 ; t

if, I ,.\ \ I

~-4°f • jf-< .

1/,

-50~,..

...JE-o::

-60~

~

~I

-70~

-8019488 9489 9490 9491 9492 9493 94\H

DAYS. MJD

Fig. 4-STFS recordings with rubidium standard as reference at ERTL (East)

20

10

+++.~0 ; ..,

-10

-20 ...

"". -30 ':::I...

if. Ic, -4Off-< Iif,I -50 ~~c,

-60 fz

-70 I-80L-

94118

...+ ;

+

-.• +.• +

.• +

+

+ ++

+ +

+.•.+#+

9489 94919490 9492 9493

DAYS. MJD

Fig. 5--STFS recordings using the Master cesium standard as reference at NPL

+'+; ;

+

+

9494

SEN GUPTA et al.: ON-LINE FREQUENCY CALIBRATION USING rNSAT STFS BROADCAST 233

Encouraged by this success, the experiment was residual, due to inaccuracy in satellite positionrepeated again during 11-24 July 1994. The modelling. The actual drift of the rubidium clock,observations for this phase are shown in Figs 7-9. using differential STFS, is illustrated in Fig. 9. AfterFor this period, the STFS monitoring at ERTL was recording data for one week at ERTL, during 11-18essentially continuous and the diurnal sinusoids in July, they were sent to NPL by speedpost. AnalysisFig. 7 are very clearly apparent. Figure 8 illustrates of these data at NPL, as shown in the graph of Fig. 9the STFS monitoring at NPL showing again a small for this period between MID 9544 and 9551, gave aartificial drift, apart from gradually growing diurnal frequency offset value of +6.3±O.15xl0-11

• This

0...

+ell::1... -10.....l +c,ZI -20

.....l +E- .•.e::

~~

-30

-409488 9489

20.-------~--------~------~--------._------_.------~

10...

++

+

1

9490 9493 94949491 9492

DAYS, MJD

Fig. 6-Frequency calibration of rubidium standard at ERTL (East) using differential STFS

110

120r-----~------~------~------~------,_------~----~

100

ell::1...

40

90

80

70

60

50

30 +

20~----~~----~------~------~------~------~----~9544 9546 9548 9550 9552 9554 9556 9558

DAYS, MJD

Fig. 7~TFS recordings using rubidium as frequency standard at ERTL (East)

234 INDIAN J RADIO & SPACE PHYS, AUGUST 1997

result was then communicated to ERTL. The finefrequency control of the rubidium clock wasadjusted at 1600 on 20 July to reduce its frequencyby this amount. The plot of Fig. 9 shows the effectof this adjustment for the subsequent period. Thenew value of frequency offset is now computed as

-0.3±O.1x10-11. This offset was sufficiently small sothat no further adjustment seemed necessary.

Certain features of the Differential STFSfrequency calibration process described so far needto be elaborated. We must remember that thereceived STFS is also a standard frequency traceable

40r-------,----,----~---___,__---__,__---.,..____--___;

30

20

\ t;10 A Ii A i \nAf\f'"=-

f V V t :0

C/l 'v i1t!\!tlt...E- -10C/l

V·V V VI-l -20CL..z

-30

-409544 9546 9548 9550 9552 95p4 9556 9558

DAYS, MJD

Fig. 8--STFS recordings using the Master cesium standard as reference at NPL

120+

+

110

+100+

••• 90=- +-lCL..ZI 80 ;t'...:I

e-ll::~

70 + +

I

60 .' .9544 9546 9548 9550 9552 9554 9556 9558

DAYS, MJD

Fig. 9---Frequency calibration of rubidium standard at ERTL (East) using differential STFS

SEN GUPTA et al.: ON-LINE FREQUENCY CALIBRATION USING INSAT STFS BROADCAST 235

to NPL. Therefore, if one measures the drift of say(ERTL-STFS), then its rate should yield the valueof frequency offset of the ERTL rubidium standard.The difficulty, however, lies in the fact that there aresinusoidal diurnal oscillations and, sometimes, asmall drift in the clock offset (ERTL - NPL) whichare artificially introduced due to inaccuracy in thesatellite orbit modelling. One way to get rid of thisdifficulty is to take a period spanning several daysso that one could fit a line through the middle of thesinusoids. This would necessitate the assumptionthat the frequency of the standard is constant overthis period, While this is a good assumption for ahigh quality atomic standard, such as a rubidium ora cesium as shown in Fig. 9, it may not hold goodfor a crystal oscillator. Also, if the observations arenot continuous but are in short segments duringdifferent times of the day, it may become difficult tofit an average line. The above difficulty is overcomeby using the differential STFS concept. It is thenpossible to determine accurately the frequencyoffset even over short durations. Typical value ofresidual r.m.s. jitter in the differential mode with thepresent system hardware is about 0.5 IlS. This wouldyield a frequency determination uncertainty of~0.3x 10-11 over a one-day measurement.

A typical illustration of the nature of frequency

-550

-560

-570

-580

-590

'" -600~...i -610Po.'z;I -620E

-630

-640

-6509358 9360 9362

variation of a high quality crystal oscillator is shownin Fig. 10. This is an HP crystal oscillator, belongingto Indian Telephone Industries (ITI), Naini, that wascalibrated at NPL using the differential STFS. Oneclearly observes the irregular frequency changesduring MID 9361-63 and again around 9368. Thesecould be due to temperature changes of theenvironment. The outliers, seemingly bunching onone side, are explained as before due to loss of lockin the decoder PLL.5 Summary and conclusions

In this paper the INSA T STFS broadcast by NPLand its use for accurate frequency calibration of highstability oscillators have been described. The mostadvantageous feature of this technique is that it canbe performed on line. It has been demonstrated thatby using the STFS broadcast in the differentialmode it is possible to get a calibration uncertainty ofa few parts in 1012 over a measurement period of 1-2days. We have also performed on-line calibration ofthe rubidium frequency standard at ERTL(East),Calcutta, for the first time using this technique. As aresult of determining the frequency offset accuratelyit has also been possible to set the rubidium standardfrequency very close to that of NPL. This techniquehas also been shown to be useful for studyingfrequency variations of crystal oscillators.

++

.•• +

++

II. + +

+t

9364 9366 9368 9370

DAYS, MJD

Fig. IO-Frequency calibration of HP crystal oscillator using differential STFS

236 INDIAN J RADIO & SPACE PHYS, AUGUST 1997

As has been mentioned in the previous section,the r.m.s. jitter in the data is .•0.5 JlS. This is not dueto the actual jitter in the received signal but due toan inherent limitation of the system hardware-namely, that the built-in time interval counter has aleast count of 1 us. The resolution of the timeinterval counter IS proposed to be improved in future 2to further improve the frequency calibrationaccuracy.

AcknowledgementsThe authors would like to record grateful thanks

to the Director, NPL, Director General, STQC andDirector, ER1L(East) for their support to this work.They are thankful to Shri N. K. Sen and Shri P. C.Banerjee of ER1L(East) for their keen interest andhelp. Co-operation received from the colleagues of

the Delhi Earth Station, Sikandrabad, time andfrequency section, NPL; and ER1L (East), Calcuttais gratefully acknowledged.

ReferencesSen Gupta A, Hanjura A K & Mathur B S, Proc IEEE(USA). 79 (1991) 973.Sen Gupta A & Hanjura A K, NPL Research Report No.NPL-91-A2008.

3 Sen Gupta A, IEEE Trans Instrum & Meas (USA), 42(1993) 480.

4 Allan D W & Weiss M, in Proceedings of the 34th AnnualFrequency Control Symposium, held at Fort Monmouth,New Jersy, USA, May 1980, p. 334.

5 Banerjee P, Saxena M & Mathur B S, IEEE Trans Instrum& Meas (USA). IM-36 (1987) 579.

6 Sen Gupta A & Mathur B S, IEEE Trans Instrum & Meas(USA), 83 (1997), in press.

7 Kamas G & Howe S, Time.and Frequency Users' Manual,NBS Special Publication 559, 1979.