Trial for upgrading of Japan Standard Time

Transcription

1 ATF2015 Trial for upgrading of Japan Standard Time Y. Hanado, F. Nakagawa and K. Imamura National Institute of Information and Communications Technology (NICT), Japan

2 Contents Introduction : How to make our timescale Trial for upgrading : Another way for making UTC(NICT) Development of distributed JST system

3 Flow of making Japan Standard Time H Maser ( Signal source ) AOG ( u phase stepper ) Steering to trace UTC Actual signals for UTC(NICT) ( 5MHz,1pps ) Steering to trace NET JST Various services 18 clocks ( for NET ) 24ch DMTD & TI counter ( measure the time differences of all HM, clocks, AOGs ) NICT Ensemble Time scale (NET) ( NET exists only in a computer)

4 Detail of the system

5 Feature 1 : Combination of 5MHz + 1pps data 5MHz 100 smpl. average / 1sec DMTD C Clk1 Clk2 5MHz measurement by DMTD high precision risk of cycle slip Clk1 Clk2 DMTD B DMTD A Clk1 Clk2 Clk1 Clk2 Time Diff. 1 period 200ns?? Cycle slip may occur after a long-term lack of data. time Initial phase by reliable 1pps data TI counter 1PPS One shot data / 1h Clk1 Clk2 Frequency + by precise x 5MHz data = Clk1 Clk2 measurement data 1pps measurement by TIC precision is not so high reliability of phase data 1pps measurement is less precise, but more reliable in phase determination.

6 Feature 2 : Automatic data selection procedure 4 devices measure the same data, but Only one data is used for a NET calculation. 5MHz 100 smpl. average / 1sec DMTD C Program for measured data processing Simultaneous data comparison Anomaly detection / data selection Clk1 Clk2 DMTD B DMTD A (ex.) If only data (B) is largely different, it is due to the anomaly of device (B). If all three data are strange, it is due to the anomaly of clock itself. Bad data is judged and removed by using Majority Rule. TI counter 1PPS One shot data / 1h If at least one device works well, measurement data is automatically obtained.

7 Detail of the system

8 NET : Parameters for NET NET is the weighted average of prediction errors of clocks. Measurement data interval NET ( t ) NET ( t 1 ) w ( t ) ( t ) k < 1 hour clock rate estimation by linear fit < 30 days k phase i i k i k 1 prediction error accumulation of Weighting for clocks < 1/ y ( ) = 30 days NET calculation interval < 1 hour NET time

9 NET : Check of clock anomaly Clock anomaly check (2009~) If y i (t) y i (t 30days) > limit, then w i (t) = 0. limit = 1E 14 ~ 1.6 x average of all y ( = 30days) phase time Sudden rate change of clock increases the prediction error and causes a bent in NET. NET UTC (ns) 120 days limit = 1 E 14. NET (before) NET (after) Weight of a bad clock Weight NET ( t ) NET ( t 1 ) w ( t ) ( t ) k k i i k i k

10 Stability of 5071As and NET 13 Allan deviation log y ( ) 14 NET clocks (5071A) Average time (s)

11 Detail of the system

12 UTC(NICT): Realization of NET NET is the reference of UTC(NICT). Adjusting the AOG frequency has 2 steps; (1) To trace NET (regularly steered), (2) To trace UTC (not regularly). HM AOG UTC(NICT) Freq. adj. Allan deviation log y ( ) NET HMs Average time (s) H maser 5071A Time Diff. NET AOG (free run) AOG (steered) UTC (2) By judgement of person in charge. (1) Additional freq. to AOG : f ad = k * (x AOG / T) + y HM x AOG : Phase offset (AOG NET) y HM : Rate offset (HM NET) Estimation y HM : 10 days Steering interval T : 8 hours Coeff. moderation k : 1/10

13 Behavior of UTC(NICT) UTC UTC(NICT) ( ns ) Adj. freq. to trce UTC ( x 1E 15) Allan deviation (log scale) UTC(NICT) (=HM4+AOG) HM3 NET HM3 HM4 15 HM Year Install the Anomaly check ( ) Change the AOG steering Parameters ( ) Averaging time (sec) NET UTC,UTC(NICT) UTC: 2011~2015 HM : 2015.Jn~Mar

14 Contents Introduction : Outline of our activity How to make our timescale Trial for upgrading : Another approach for making UTC(k) Development of distributed JST system

15 Another approach for making UTC(k) Allan deviation log y ( ) UTC(NICT) (=HM4+AOG) HM4 NET Average time (s) Delay of UTC : ~ 45 days in maxim. => HM needs NET in middle term. Delay of NET : 1hour De trend of HM : linear fit (by 10days) Delay of UTCr : ~ 9days in maxim. => UTC(k) by HM + UTCr? H maser 5071A Time Diff. NET Last UTC AOG (free run) UTC(NICT) (steered AOG) Time Diff. UTC H maser Last UTCr AOG (free run) UTC(k) (steered AOG) UTC

16 UTC(k) by using HM + UTCr? f ad = x AOG / T + y HM T= 1 day X AOG = [ UTC(k) UTCr] = [ UTC(k) HM ] + [ HM UTCr] = [ UTC(k) HM ] + [ HM UTCr] est + [ HM HM est ] 1 Measurement 2Estimation Error 60 days Quadratic fitting using the past [HM UTCr]s. H maser y HM is also estimated from this quadratic fitting. Time Diff. UTC(k) (steered AOG) 2 1 T AOG (free run) UTC y HM * T x AOG

17 Preliminary results of the simulation Allan deviation (log scale) 13 UTCr UTC(NICT) ( ns ) 14 HM3 (de drift by 30days) HM4 (de drift By 60 days) UTC(k) via HM4 15 UTC(NICT) UTC(k) via HM3 UTC(NICT) UTC(k) via HM Averaging time (sec) Year UTC(k) by [HM + UTCr] shows better stability than current UTC(NICT). Appropriate span for Quadratic fitting depends on each H maser.

18 Contents Introduction of current status : Outline of our activity How to make our timescale Trial for upgrading : Another way for making UTC(NICT) Development of distributed JST system

19 Concept Current JST is generated from only the clocks at NICT HQ (Tokyo). Distributed system for Japan Standard Time generation Sub station NICT Kobe Master station NICT HQ JST Features NET from all distributed clocks linked via satellites Multiple generation of NET by common database at remote sites Merits Robustness : Distribution of clocks and multiplex NETs decrease a risk of stopping a timescale. Scalability : Each station can generate UTC(k) via any local clock at the station by using any NET by common database.

20 Procedure Process 1. Remote stations share all data of local and remote clocks. 2. Each station independently calculates NET from all clocks. Clock ensemble for each NET is common, but weight of each clock may be different at each site. For a check, NETs should be compared with each other. B NET(A),(B),, should be almost same, because they use the same clock ensemble, but different a little, because clocks include different link errors. NET(B) D NET(D) 3. NET at the Master station is the origin of JST, in principle. Labo. Clocks Remote Clocks C NET(C) A NET(A)

21 Required actions Construction of sub station at Kobe has started since * Rooms, Basic measurement systems * Clocks * GPS link * Generation of UTC(Kobe) Installed ( x 4, HM x 1) Regular operation has started Test has started by HM + AOG + GPS, tracing UTC(NICT) * Dissemination system Completed B A

22 Summary Current UTC(NICT) and Japan Standard Time is based on the Ensemble timescale of A (NET), Rate: 30d, Weight: 1/Adev(30d), Rate check: 1e 14, Calc.: Every hour is realized by [H maser + AOG], HM NET rate: 10d, Mod.Coeff.: 1/10, Steering: every 8h has 2 features in data processing. (1) 1pps & 5MHz combination, and (2) automatic data selection. Trial for upgrading is going on; UTC(k) from [HM + UTCr] shows a good possibility. UTCr is highly effective for estimating HM s drift. Distribution system of JST is under development. For improving robustness and scalability.