Local Ensemble and Other Reference

Transcription

1 Clock Steering Model According to Local Ensemble and Other Reference Time Scale Calvin S.Y. Lin National Standard Time and Frequency Lab., Taiwan 1

2 What and Why Outline The character of commercial atomic clocks UTC(k) of APMP NMIs Phase Lock Loop vs. clock steer Steer model of UTC(TL) Summary and suggestion 2

3 What is clock steer: What and Why Adjust the frequency or phase offset of a clock Why need clock steer: To generate the definition of SI second Frequency offset as close as 0 To fit the UTC time scale Phase offset as close as 0 Goal: To generate a time scale as close as UTC 3

4 Osc. Locked by a Reference Simple Phase Lock Loop f ref Phase Low Pass V-Control f out detector Filter Osc. Feedback control Frequency Divider Dominated by f ref, Oscolliator, and phase detector f Q: How we steer an atomic clock? 4

5 UTC(k) Steer Loop Master Clock UTC, UTCr, TA(k), PFS TWSTFT, GNSS, TIC. Control Strategy Micro phase stepper UTC(k) Distribution amplifier

6 Facilities in NMIs Reference clock and phase, frequency stepper A stable clock Internal or external micro phase stepper Reference Time Scale Should be more accuracy or stable than master clock UTC, UTCr, Primary frequency Standard Localclocksensemble clocks ensemble: cesium clocks+ hydrogen masers Time comparison: TW, GNSS, TIC 6

7 Phase data of a Coc Clock Phase x ( t ) = x o Frequency offset + y t Frequency Drift D t 2 + σ ( t ) x Higher order term Tpicall Typically, We neglect the higher order term of cesium clock and hydrogen maser 7

8 Phase data of Cesium Clock(5071a) Coc frequency offset removed 8

9 Frequency offset of 5071a 9

10 Frequency Offset of PTB Lab type Cs 10

11 Characters of 5071a x( t) = x o + y t D t 2 + σ ( t) Phase change from ±50 ~ ±150 ns/2 years (frequency offset removed) Frequency offset of 5071a are about ±30 ns/day Frequency Drift are about 3~15 ns/day/life time Commercial Cesium Clocks (5071a) have drift Drift of Lab type cesium (thermal bean, fountain.) < 1 ns/day/5 years In perfect case, we can use single 5071a to maintain a time scale within ±50 ns accuracy for all its life time x 11

12 UTC UTC(k) UCUC()of APMP NMIs 12

13 Stability ty of UTC(k) UC()of APMP NMIs 13

14 Time and Freq. APMP NMIs Phase change of UTC(k) from 4000 ns to 4000 ns in 1500 days More than 60% NMIs keep in ±100 ns Long term stability from ~10 15 Short term stability (τ = 5 days) from ~ Can be improved by a good clock steer strategy At lease as good as a single commercial clock 14

15 Time Stability ty of Clocks Coc 15

16 Frequency Stability ty of Clocks Coc τ -1/2 τ -1 16

17 Frequency Offset of Hydrogen Maser

18 Characters acte sof 5071a (2) White FM noise dominate when τ < 30 days Each 5071a looks alike the same Random walk or clock drift dominate when τ > 30 days All showed different drift CS2365 is almost zero drift 18

19 Characters acte sof H maser (imaser 3000) Much more stable than cesium clock whenτ< 5 days Flicker PM noise dominate when τ< 1 day Random walk noise and clock drift dominate when τ > 5 days Drift of Hydrogen maser is predictable We can use a script to compensate it 19

20 Master Clock Coc (1) Choose a predictable clock as master clock predictable means stable choose a stable clock as master clock Long term drift can be removed according to reference time scale choose a short term stable clock as master clock Hydrogen maser > 5071a high performance tube > 5071a standard tube> othercesiumclock clock 20

21 Master Clock Coc (2) Steer clock using external micro phase stepper Step resolution of 5071a is 1e 15, about 115 ps/day Stepresolution of somemicrophase micro steppers can achieve 1e 19 A good external micro phase stepper can reduce noise,

22 Reference eee Time Scale, UC UTC Get UTC UTC(k) and frequency offset from ftp://ftp2.bipm.fr/pub/tai/publication p// p p /p / /p Announce 8 th ~ 12 nd day every month UTC 0h point every 5 days Official time scale 22

23 Reference eee Time Scale, UC UTC Uncertainty of UTC(k) point base on TAI link, up to 20 ns ua from 0.3 ~10 ns, ub from 5~20 ns Windows period up to 45 days Date h UTC SEP 29 OCT 4 OCT 9 OCT 14 OCT 19 OCT 24 OCT 29 Uncertainty/ns Notes MJD ua ub u Laboratory k [UTC-UTC(k)]/ns AOS (Borowiec) APL (Laurel) (1) AUS (Sydney) BEV (Wien) BIM (Sofiya) BIRM (Beijing) BY (Minsk)

24 Reference eee Time Scale, UC UTCr BIPM official product computed since 1 July Get UTCr UTCr(k) from ftp://tai.bipm.org/utcr, announce every Wd Wednesdayd UTC 0h point everyday y ua, ub did not announce (the same as UTC?) Windows periods up to 6 days 24

25 Reference eee Time Scale, TA(k) () Average white and Flicker noise of clocks Better short term performance than a single clock Allan deviation of ensemble 1/ N Filter out bad clocks, more robust than a single clock Calculate by ourselves No windows periods UTC (ALGOS), USNO (A.1), NIST (AT1). 25

26 Reference eee Time Scale, PFS Primary Frequency Standard Lab type thermal beam Cesium clock Cesium fountain Much more accurate and stable than commercial clock UTC(PTB)

27 Steer Strategy (1) Proportional control Control effort bases on phase offset Optimized for best accuracy Derivate control Control effort bases on frequency offset Optimized for best stability 27

28 Steer Strategy (2) Parameters of control effort measurement and control period Follow long term or short term of reference time scale Noise of measurement, master clock, and reference time scale Phaseandfrequency and offset Do nothing If offset < noise Upper limitof control effort Avoid damping and overshoot 28

29 UTC(TL) UC( )Steering Use UTC as reference time scale Monitor UTCr for mid term checking 2 Local ltime scales TA 0 (TL), TA 1 (TL) as short term reference Daily P control by TA 1 (TL) Monthly D control by UTC 29

30 TA 0 (TL): 2 Local Time Scales free run time scale, 13~ a cesium clock ensemble TA 1 (TL): Adjust offset of TA 0 (TL) according to UTC every month. TA 2 (TL): Combination of hydrogen maser and cesium clock average of 2 nd difference of phase offset between HMs and TA 1(TL) (optimum for random walk FM noise.) (terminated because only 1 normal HM at TL since 2010) 30

31 TA 0 (TL) 0( ) = Δ Δ + = N i i i i t t d t t t x t w N t TL t Ens 1 )] ( ) ( ) ( [ ) ( 1 ) ( ) ( Δ Δ = N t t b t t b i i i e t w ) ( ) ( 2 2 ) ( σ σ (t Δt) All d i ti f h l k = Δ i t t b i e 1 ) ( σ σ i (t-δt) : Allan deviation of h-maser vs. clock i 31

32 Comparison of Weighting eg gpolicies oces w i = N e i= 1 2 bσ i e 2 bσ i w i = N 1/ σ k = 1 2 i 1/ σ 2 k TL weighting Inverse square weighting 32

33 Comparison of Weighting eg gpolicies oces 33

34 Comparison of Weighting eg gpolicies oces No significant difference between 3 weighting processes TL uses inverse exponential weighting process More gentle than traditional weighting No cutting head upper limitof weight Clock drift will be average out (in a large number clock ensemble) As a filter, to eject bad clock automatically 34

35 TAI TA 0 0( (TL) free run time scale, 13~ a cesium clock ensemble 35

36 TA 1 1( (TL) TA 1 (TL) UTC(TL) UTCr(TL) Adjust phase and frequency offset of TA 0 (TL) according to UTC every month Replenish short term data of UTC 36

37 Steering of UTC(TL) UC( )(1) Master clock: hydrogen maser with AOG 110 micro phase stepper Drift calculated by TA 1 (TL), removed by script program Monitor TA 1 (TL) UTC(TL) every day. Steer TA 1 (TL) every month by UTC TA 0 (TL) Adjust by UTC 0 1 TA 1 (TL) Steer H maser UTC(TL) UTC(TL) : follows UTC at long term, follows TA 1 (TL) at mid term, hydrogen maser itself at short term 37

38 Steering of UTC(TL) UC( )(2) TDev(τ=5 days) of TA 1 (TL) 0.5 ns TDev of HM0057 TA 1 (TL) UTC(TL) < 0.5 ns do nothing TA 1 (TL) UTC(TL) > 0.5 ns proportional control TDev (τ=5 days) of HM ns (drift removed) Upper limit of fractional frequency step 0.3 ns/day We trust our hydrogen maser because we have just few clocks in clock ensemble e [ TA ( TL) UTC ( TL) /100 ps]

39 UTC(TL) UC( )and duc( UTCr(TL) 39

40 Summary and dsuggestion (1) Master Clock Choose a short term stable clock, h maser>cesium clock Choose a linear clock to be master clock We can predict because its linear Linearity of phase offset (cesium clock), linearity of frequency offset (hydrogen maser) Micro phase stepper Choose a low noise one High resolution 40

41 Summary and Suggestion (2) Reference Time Scale Long term: UTC Short term: PFS, TA(k) () Mid term: UTCr, TA(k) For UTC Windows period of UTC is up to 45 days Uncertainty of UTC is up to 20 ns Use the last point of UTC UTC(k) for p control use monthly points for d control

42 For UTCr Summary and dsuggestion UTCr does not match UTC, inconsistency up to 5 ns Windows period of UTCr is up to 7 days Use as a backup reference time scale For ensemble < 5 clocks TA(k) may be dominated by 1 or 2 clocks, Add, remove clocks or a bad clock will cause TA(k) unstable. Use 3 cornered hat to monitor the master clock Use UTCr as mid term reference time scale 42

43 Summary and dsuggestion (3) For ensemble > 5 clocks: clocks ensemble with > 5 Cs may have the same time deviation with HM at 5 10 days average A local TA(k) to reinforce the mid term stability of UTC(k) Can use UTCr as backup reference time scale Steer Strategy: Optimum for accuracy proportional control Optimum for stability derivate control Check Circular T every month to adjust your policy 43

44 Thank You and Questions!!! Calvin S.Y. Lin