FIRST COMPARISON OF REMOTE CESIUM FOUNTAINS

Transcription

1 2001 EEE nternational Frequency Control Symposium and PDA Exibition FRST COMPARSON OF REMOTE CESUM FOUNTANS T. Parker*, P. Hetzel', S. Jefferts*, S. Weyers', L. Nelson", A. Bauc', J. Levine* *National nstitute of Standards and Tecnology 325 Broadway, Boulder, CO 80305, USA # P ysi kal isctecni sce B undesans talt, POB 3345, D38023 Braunscweig, GERMANY Abstract Te frequencies of te cesium fountain primary frequency standards at te National nstitute of Standards and Tecnology and te PysikaliscTecnisce Bundesanstalt ave been compared. Twoway satellite time and frequency transfer and GPS carrierpase were te principal frequencytransfer tecniques used to make te comparison. For te 15day interval in wic bot fountains were in operation te frequencies were compared wit an additional uncertainty due to te comparison process of only 6.2~10.'~. Te two standards agree witin teir stated onesigma uncertainties of 1.7~10l~. Key Words: atomic frequency standard, cesium fountain, comparison. ntroduction Since August of 2000 te cesiumfountain primary frequency standards at te National nstitute of Standards and Tecnology (NST) and te PysikaliscTecnisce Bundesanstalt (PTB) ave been in service on a discontinuous basis. Significantly, during te mont of August te two fountains were operated at nearly te same time. Te NST fountain, NSTF1, was in operation during te period MJD (Modified Julian Date) to 51794, and te PTB fountain, CSF, was operated over te period MJD to Tis gives a 15day overlap. Tere were also four additional periods of operation for te PTB standard, MJD to 51814, to 51839, to 51879, and to A second NSTF run occurred over te interval MJD to (February 2001). Since August te NST standard as accumulated 60 days of run time and te PTB standard 75 days. Tese evaluations ave all been reported to te Bureau nternational des Poids et Mesures (BPM), and details of te frequency standards ave been presented in oter papers [ 11, [ 21. Te nearly simultaneous operation of tese two standards provided an excellent opportunity for comparison of teir frequencies. Te best comparison comes from te August NSTFl run and te first two PTBCSF runs. However, all seven runs can also be compared wit somewat increased uncertainty. Te PTBCSF1 was operated for te first four runs at constant operational parameters, and te real measurement time comprised about 96.3 % of te nominal total measurement time of 1440 ours. During te operation of NSTF1 a range of atom densities was used, so it is not practical to sorten te comparison period to coincide exactly wit tat of te first interval for CSF1. Terefore, te comparison must be made by extending te PTB interval wit a stable (but not necessarily accurate) frequency reference. Tis can be done wit eiter EAL, wic is a free atomic scale calculated by te Bureau nternational des Poids et Measures (BPM), or te post processed NST maser ensemble, ATlE. Te fact tat te overlap of te two fountain runs is not exact means tat te stability of te frequency reference contributes to te uncertainty of te comparison. Te fift PTBCSF1 run was made wit a range of atom densities in a manner similar to tat of te NSTFl runs. Longdistance tecniques for frequency comparison must be used since te two primary frequency standards under discussion are separated by tousands of kilometers, and tis also adds an additional uncertainty to te comparison. Te extrapolation and frequencytransfer uncertainties are discussed in tis paper, and overall comparison uncertainties are calculated. Frequency Transfer Tree tecniques for time and frequency transfer were used for te fountain comparison in order to minimize te cance of a statistical aberration. Tese tecniques were Twoway Satellite Time and Frequency Transfer (TWSTR) [3], GPS carrierpase [4], and GPS commonview [5]. Te GPS commonview comparison was made wit data from te BPM publication Circular T, were corrections are made using precise orbits and measured ionosperic data. Te TWSTFT measurements followed te standard tree days per week (Monday, Wednesday and Friday) BPM scedule and were made at Kuband using a commercial communications satellite. Te twoway data used for te fountain compason were te same as tat reported to te BPM, except tat data comparing U.S. Government work not protected by US. copyrigt 63

2 UTC(N1ST) to te maser H2 at PTB was extracted. Te GPS carrierpase data comes from two dualfrequency, geodeticquality receivers located at NST and PTB [6]. Te TWSTm and carrierpase data bot give te time difference between UTC(N1ST) (wic is derived from a maser ensemble) and te maser H2. Te fountain frequencies can be related to tese two standards via internal measurements. At PTB te fountain directly measures te frequency of H2. At NST an internal measurement system is used to relate te frequency of te specific maser used as te fountain reference to UTC(N1ST). Te uncertainty of te NST internal measurement is well under 1~10l~ at 15 days. Te commonview GPS data relate te fountain frequencies to nternational Atomic Time (TA) via te reference clocks for eac standard. 860 E" 855 v v) 1 E M JD Figure 1. Time difference between UTC(N1ST) and H2 via te twoway metod. 10 v) F l P 6 v 0.1 Twoway (MD'S ) CarrierPase (MD'S Twoway (uneven data) TwoW ay (in tcrpolaicd data (day S) Figure 2. Time deviation of UTC(N1ST) minus H M JD Figure 3. TWST T minus GPS carrierpase for UTC(N1ST) H2 By differencing te data from bot transfer tecniques te clock noise can be removed, giving a clearer picture of te stability of te frequencytransfer processes, particularly in te long term. Figure 3 sows te time difference between te TWSTFT and carrierpase data for te UTC(N1ST) H2 link over a 100day period tat includes te August NSTF run and te first tree PTBCSF runs. Te data interval is one day and te twoway data ave been interpolated to matc tis interval. As expected, te longterm stability of tis data is better tan tat in Fig. 1, but tere is still some drift in te time difference. t is not clear weter tis drift comes from twoway, carrierpase, or bot, but in any case, it would constitute a real systematic error (rate offset) on te order of 5~10l~ in te frequency measurement if attributed to eiter one of te metods. t is important to note tat te final carrierpase solution is a combination of 3.5day analysis periods wit alfday overlaps. Terefore te solution is sensitive to te overlapping offsets of te consecutive data series as well as to corrections for jumps and gaps in te data. Tis could be 64

3 a possible cause of te drift in te time difference. Anoter possibility is seasonal temperature fluctuations affecting eiter system. Figure 4 sows te time deviation for tis data and Fig. 5 sows te Allan deviation. Te ox(z) and Allan deviation values at one day are bot biased low because of te interpolation of te twoway data, but in general te time deviation is flickerpase in nature at a level of about 200 ps. Te Allan deviation plot indicates tat te combined frequency uncertainty of TWSTFT and carrier pase is about 5~10l~ at 15 days. However, tis may be optimistic because bot te Allan deviation and time deviation statistics are based on te second difference of a time series, wic is insensitive to a rate (or frequency) offset. A sligtly larger uncertainty of 6~10l~ at 15 days is obtained using a firstdifference statistic tat is te RMS fractional frequency of te timeseries data in Fig. 3 [7]. Tis is more consistent wit te observed slope in te data. For te purposes of tis comparison we will assume tat te instabilities of TWSTFT and carrier pase are independent and tat tey contribute equally to te combined instability of 6~10~. Tis gives a frequencytransfer uncertainty of 4.2~10'~ at 15 days for eac of te two tecniques individually. For te uncertainty in te GPS commonview data we use te BPM estimate of 2~10''~ at 15 days, as stated in Circular T. Extrapolation MD'S A stable frequency reference must be used as a transfer standard since none of te fountain evaluations overlap perfectly. To estimate te uncertainty of comparisons wit dead time we ave used te metod of Douglas and Boulanger [8]. Te comparisons were made wit bot ATlE [9], [lo], a postprocessed scale based on a maser ensemble at NST, and EAL, te free atomic scale calculated at te BPM. Te stability caracteristics of tese scales are sown in Table 1 for z in units of 1 day. Table 1. Frequency Stability Caracteristics of AT 1E and EAL ': (day S) Figure 4. Time deviation of TWSTFT minus carrierpase for UTC(N1ST) H2 1 0 l ~ MD'S : 0 Allan Deviation A Total Deviation $11 fl l l Figure 5. Allan deviation of TWSTFT minus carrierpase for UTC(NST)H2. Te wite M and flicker FM noise caracteristics of ATlE were estimated from internal measurements made at NST, and te randomwalk FM noise level was estimated from measurements against EAL and CS2 (a termalbeam primary frequency standard at FTB). See [lo] for a more detailed discussion. Te stability caracteristics of EAL are tose publised by te BPM in Circular T. Te main advantage of ATlE is its muc lower wite FM noise. Tis is expected since EAL is affected by noise from GPS commonview. Comparison Results Te best candidates for comparison are te August NSTFl run and te first tree FTBCSF1 runs because tey ave te closest overlap. Te results are summarized in Tables 2 ad. Te stated uncertainties for te August NSTF1 run are ub = 1.5~10'~ and ua = 0.8~10'~. were ub is te systematic uncertainty and ua is te statistical uncertainty. Te combined uncertainty for tis NSTFl run is 1.7~10'~. (For te second NSTF1 evaluation te combined uncertainty is 1.8x10?) Since te uncertainties 65

4 ' for te first tree PTBCSFl runs were not all te same uncertainty due to te comparison process is ua(comp.), tey are listed individually in te tables. All uncertainties wic is obtained from ua(tt) and ua(dead) combined in tat are statistical in nature will be identified as 'a". Te quadrature. Finally, te uncertainty of te remote standard, frequency uncertainty due to te timetransfer process is u(remote), is calculated from te quadrature combination of identified as ua(tt) and ranges from 0.42~10l~ for ua(standard), ub(standard), and ua(comp.). n tis TWSTFT and carrierpase to ~.OX~O'~ for commonview discussion PTBCSF will be treated as te remote standard GPS. Te uncertainties due to dead time are identified as as seen from NST, altoug te roles could be reversed. ua(dead) and range from 0.46xlO'' to 2.9~10~. Te total Table 2. Comparison of NSTF wit PTBCSF1 MJD Table 2a Referenced to ATlE via twoway (units of lo") ua(tt) = 0.42 Table 2b Referenced to ATlE via GPS carrier pase (units of ua(tt) = L.... Table 2c Referenced to ATlE via commonview GPS from Circular T (units of ua(tt) = 2.0 _.T Table 2d Referenced to EAL, via commonview GPS from Circular T (units of ua(tt) = 2.0 PTB y(flcsfl) ~. ub.... t, i Runs 1.! CSFl _ CSFl 1 CSFl. (dead) t.. i(comp.)'. (remote), (w.... avg.) / (remote). i i L.... First......; j i 1.0 t._ j (1 run) ,.... i ,... /Second; ( i ' 2.5 ~ 3.2, (2.5(2runs)! L Tird i , ! i j (7) 2.3 (3 runs). _i..._..._.. i... L A. i.... t..... J... Te four tables compare data for: (a) TWSTFT wit ATlE used as a transfer standard, (b) GPS carrierpase wit ATlE used as a transfer standard, (c) GPS commonview wit ATlE used as a transfer standard, and (d) GPS commonview wit EAL used as a transfer standard. All uncertainties are one sigma. Te first row of data in eac table is for te first PTB run, wic overlapped te NST run by 15 days. n Tables 2 ac, te uncertainty ua(dead) in 66

5 tis row is small (but not zero) because te two runs overlapped, but were not te same lengt. Note tat ua(dead) is considerably larger wit EAL used as te frequency reference (Table 2d) since EAL as a muc larger wite FM noise level. Te observed values for te fractional frequency difference of te two standards, y(f1csfl), are 0.36~10" for comparison by twoway, and 0.03~ 10"' for comparison by carrierpase. ua(comp.) is 0.62~10~ for bot metods and u(remote), te uncertainty of PTBCSF1 as seen from NST, is 1.9~10.'~. Tis is only sligtly larger tan te stated combined uncertainty of CSF1, wic is 1.8~10'~ for tat run. Te two standards are in excellent agreement in te first run. Te comparison results using commonview GPS wit ATlE and EAL. are given in te first data rows of Tables 2c and 2d. Note tat ua(comp.) is muc larger. Te agreement between te two standards is not as good as wit twoway and carrier pase, but is still witin u(remote) for tese comparisons. Te second and tird runs of PTBCSF1 can also be compared to te August NSTF1 run, as sown in te second and tird rows of te tables. However, ua(dead) gets larger because te runs don't overlap at all. n general, te agreement in te second run is not as good as in te first. n te second run te frequency differences for te comparisons using ATlE wit twoway and ATlE wit commonview are large enoug tat te error bars of te two standards don't overlap. For ATlE wit carrierpase and EAL wit commonview tey do overlap. Te agreement in te tird run is muc better, even toug u(remote) is getting rater large, particularly wen EAL and commonview are used. A weigted average of te various runs can also be calculated. Te last two columns in rows 2 and 3 sow te weigted averages for y(f1csf1) and te corresponding uncertainties, uw(remote), for te first two runs, and all tree runs, respectively. (Results for te first run are repeated for clarity in te first row of tese columns even toug tere is no averaging.) Te weigting was based on te combined statistical uncertainties for eac case. t is questionable ow meaningful te weigted average is for all tree runs because it is very likely tat te errors due to te dead time are correlated between te second and tird runs; tus te question marks in te tables. However, te weigted average of te first two runs is meaningful, and it sows tat te fountains agree witin 1.2~10"~ for eiter te TWSTFT or carrierpase comparisons. Note tat uw(remote) for te weigted average of te first and second runs using tese two comparison tecniques, 1.8~10~~, is very close to te level of te stated uncertainty of te PTB CSFl. n principle, one could average te TWSTFT and carrierpase comparisons togeter to furter reduce ua(tt) by anoter factor of 1/42. However, tis would ave only a small impact, and may not be justified until te assumption of independence between TWSTFT and camerpase can be verified. 5 m \ L O F T o T A NSTFl 0 PTBCSF A NST7 0 PTBCS M JD (days) Figure 6. Fractional frequency deviation of AT 1E versus four primary frequency standards. 0! 0 Figure 6 sows te fractional frequency difference of ATlE (wit an arbitrary frequency offset removed) relative to four of te best primary frequency standards over a 500 day period. Te standards include te two fountains discussed in tis paper and two termalbeam standards, NST7 and PTBCS2. Te figure contains all available fountain evaluations (formal and informal) for tis period and provides a qualitative view of te relative frequencies of te four primary frequency standards. Te combined uncertainty for PTBCS2 is typically l2xlo5, and for NST7 ranges from 5 to 10x10". ndividual uncertainties are sown for te fountains. Te overall agreement between te four standards is quite good. Toug te scatter among te fountain data is large enoug tat a few of te error bars don't overlap, it is impossible to tell weter te fluctuations are from te standards or from ATlE. However, te scatter is statistically consistent at te onesigma level wit te uncertainties of te standards and te noise levels of ATlE [lo]. From te data in Fig. 6 te average frequency difference for te two NSTF1 evaluations since August 2000 (MJD 51764) and te five PTBCSF1 runs is y(f1csf1) = 0.2~10'~. Te uncertainty for tis comparison is difficult to calculate rigorously but u(remote) is estimated to be in te range of 2.0~10'~ to 2.6x1O5. Te sligt downward frequency drift in Fig. 6 is a caracteristic of ATlE. Figure 7 is a similar plot tat sows te fractional deviation d (in seconds) of te scale interval of TA (as obtained from Circular T) relative to te same four primary frequency standards. Note tat d = Av/v. Te details are somewat different, but te general relationsip of te four standards is about te same as tat in Fig. 6. Te average frequency difference for te two NSTF1 evaluations and te five FTBCSF runs using TA as te reference is 67

6 y(f1csf) = 0.3~10 ~, wit an estimated range for u(remote) of 2.6xlO to 3.3~10l~. e a 1.0 ~ ~ ~ ~ MJD (days) Figure 7. Fractional deviation of te duration of TA relative to four primary frequency standards. Conclusions Te cesium fountain primary frequency standards at PTB and NST ave been compared by tree different frequency transfer tecniques and wit two different stable frequency references. Tis variety of comparison tecniques was used in order to minimize te cance tat a statistical fluctuation in one tecnique migt give an unusually good or bad result. t as been demonstrated tat te uncertainties of te comparison process can be reduced to a nearly negligible level wit te use of TWSTFT or GPS carrierpase if te duration of te comparison is 15 days or longer. For te data in Table 2, only two cases out of te twelve individual comparisons and te four weigted averages exibit a frequency difference large enoug tat te uncertainty limits of te two standards don t overlap. Tis is entirely consistent wit onesigma uncertainties. n most cases te observed frequency differences are well witin te uncertainties of a single standard. Te best estimate of te frequency difference of te two standards is given by te weigted av erage of te first two PTB runs relative to te August NSTF1 run using eiter twoway or GPS carrierpase. Te frequency difference is l.2xlo, wic is well witin te stated combined uncertainties of te two standards, wic are 1.7~10 ~. A somewat less rigorous comparison using te two most recent NETFl evaluations and all five PTB CSFl runs also sows agreement well witin te estimated uncertainties References S. Weyers, R. Scrijder, and A. Bauc. Recent Results from PTB s Caesium Fountain CSFl, in Proc. 15 European Freq. And Time Forum, to be publised. S. Jefferts, D. M. Meekof, T. P. Heavner, T. E. Parker, Latest Results from NSTF1, a Laser Cooled Cesium Fountain Primary Frequency Standard, in Proc. 151 European Freq. And Time Forum, to be publised. D. Kircner, Twoway Time Transfer Via Communication Satellites, in Proc. of te EEE, vol. 79, pp , K. M. Larson and J. Levine, CarrierPase Time Transfer, in EEE Trans. on Ultrason.. Ferroelect., and Freq. Contr., vol. 46, no. 4, pp , W. Lewandowski and C. Tomas, GPS Time Transfer, in Proc. of te EEE, vol. 79, pp , L. Nelson, J. Levine and P Hetzel, Comparing Primary Frequency Standards at NST and PTB, in Proc EEE nternational Freq. Control Symp., pp ,2000. T. E. Parker, D. A. Howe and M. Weiss, Accurate Frequency Comparisons at te X~O ~~ Level, in Proc EEE nternational Freq. Control Symp., pp , R. J. Douglas and J. S. Boulanger, Standard Uncertainty for Average Frequency Traceability, in Proc. 11 European Freq. and Time Forum, pp ,1997. T. E. Parker, Hydrogen Maser Ensemble Performance and Caracterization of Frequency Standards, in Proc Joint Meeting of te European Freq. and Time Forum and te EEE nternational Freq. Control Symp., pp , T. E. Parker, Comparing and Evaluating te Performance of Primary Frequency Standards: mpact of Dead Time, in tis Proceedings. 68