Improvements in ALGOS - BIPM · 2013. 9. 5. · Improvements in ALGOS G. Panfilo Bureau...

Bureau International des Poids et Mesures 1

Improvements in ALGOS

G. Panfilo

Bureau International des Poids et Mesures

CCTF, Sèvres, 13-14 September 2012


• Clock prediction algorithm

• Quadratic model

• TT as frequency reference for drift evaluation

• Application and effect of the new prediction algorithm :

• EAL frequency drift is almost completely removed

• Effect on TAI and on the weights

• A new model for the weighting procedure

• The prediction is taken into account

• Effect on the UTC stability and on the weight

• Conclusion and discussion

Outline

Bureau International des Poids et Mesures 3 3

EAL prediction algorithm

Until July 2011 the linear prediction had been used to describe

the clock behaviour; the frequecy of the clocks is considered

constant during the calculation period; the frequency drift (or the

aging) is not included in the model.

UTC is calculated with more than 400 clocks of which:

- Caesium clocks 5071 (high performance tube): ~ 270

- H-masers Symmetricom/Sigma tau : more then 50

The ensemble of clocks shows very similar deterministic

signatures (frequency drift or aging).

As a consequence of that EAL shows a frequency drift respect to

TT of about:

-1.3x10-17/day


After test and simulations for a long period (5 years), after

publications and presentations[1,2,3] the new prediction

algorithm for EAL has been validated and implemented.

From August 2011 a quadratic prediction is officially used to

calculate UTC; the frequency drift is considered in the

model for all kind of clocks.

[1] G. Panfilo and E.F. Arias. ”Studies and possible improvements on EAL algorithm”. UFFC, Vol. 57, No.1,

January 2010, pp. 154-160.

[2] G. Panfilo, A. Harmegnies, L. Tisserand. ”A new prediction algorithm for the generation of International

Atomic Time”. Metrologia, 2012, 49, n°1, 49-56.

[3] G. Panfilo, A. Harmegnies, L. Tisserand. Report to CCTF. A new prediction algorithm for EAL.

4

EAL new prediction algorithm

The frequency drift affecting EAL is almost completely

removed.


The prediction algorithm for EAL

The main ideas of the new prediction algorithm are:

1. The use of the quadratic model to describe the frequency drift

2. The use of TT (Terrestrial Time, a time scale optimized for

frequency accuracy) as frequency reference to estimate the

frequency drift

2,1,,,

' ˆ2

1ˆ2

1ˆˆ1 kIipkkkIikIipIii ttCttttCttBath

kkkk


Simulation results

6

52000 53000 54000 55000

66

67

68

69

70

71

y(EAL)-y(TT)

MJD

No

rma

lize

d F

req

ue

nc

y 1

0-1

4

Linear Prediction

Quad. Pred. 2008

Quad. Pred. 2006

Two tests were performed on data starting in 2006 and in

2008 until 2010 to validate the new prediction algorithm.


Results: y(EAL)-y(PFS)

7

Starting from August 2011 the new algorithm is officially used in

UTC calculation:

Period of application of the new algorithm

The frequency drift

affecting EAL is almost

completely removed

55000 5600064.5

65

65.5

66

66.5

67

67.5

68

f(EAL)-f(PFS)

MJD

No

rmalized

Fre

qu

en

cy 1

0-14


y(EAL)-y(PFS)

8

54000 54500 55000 55500 56000 5650064.5

65

65.5

66

66.5

67

67.5

68

68.5f(EAL)-f(PFS) only fountains

MJD

No

rmalized

Fre

qu

en

cy 1

0-14

NIST-F1

PTB-CSF1

PTB-CSF2

SYRTE-FO1

SYRTE-FO2

SYRTE-FOM

IT-CSF1

NPL-CsF1NPL-CsF2

NICT-CsF1

NMIJ-F1


55400 55500 55600 55700 55800 55900 56000 56100 562000

1

2

3

4

5

6

7

No

rmalized

Fre

qu

en

cy 1

0-15

MJD

The effect on TAI

9

In the plot the fractional deviation d of the scale interval of TAI from

that of TT based on all available PFS measurements is reported .


The value of d is

decreasing to zero

with some noise


............f(TAI)-f(PFS)

10

Period of

application of the

new algorithm

The value of the difference f(TAI)-f(PFS) should be close to zero. Starting

from the application of the new prediction algorithm this value is

approaching zero with some noise.

54000 55000 56000-0.7

-0.6

-0.5

-0.4

-0.3

-0.2

-0.1

0f(TAI)-f(PFS)

MJD

No

rma

liz

ed

Fre

qu

en

cy

10-1

4


54000 54500 55000 55500 56000 565000

10

20

30

40

50

60

70

80

90

100

To

tal

we

igh

t in

pe

rce

nta

ge

te

rms

MJD

Cs-clock

H-maser

54000 54500 55000 55500 56000 565000

10

20

30

40

50

60

MJD

Nu

mb

er

of

H-m

asers

wit

h m

ax. w

eig

ht

Cs clock

H-maser

The effect on the weights

11

The weighting algorithm has not yet been changed. The new

prediction algorithm has no effect on the weights.



Weight

12 12

EAL: Weighting Algorithm

In the current weighting algorithm the weight attributed to

a clock reflects its long-term stability. The clocks with

deterministic signatures like frequency drift or aging are

de-weighted and considered “bad” clocks.

The H-masers have a small weight due to frequency drift

affecting the behaviour.

In the time scale algorithms clock weights are generally

chosen as the reciprocals of a statistical quantity which

characterizes their frequency stability, such as a frequency

variance (classical variance, Allan variance....)


A new proposed weight algorithm

After the implementation of the new prediction algorithm the

deterministic signatures are eliminated or minimized in the

time scale.

The main idea of the new weighting algorithm is that a good clock

is not a stable clock but a predictable clock[1].

In this new version of the weighting algorithm 1 year of the

difference between the predicted and real frequencies of the

atomic clocks is evaluated. A filter is used to give a bigger role

to the new measurement with respect to old ones.

[1] Levine J. “Introduction to time and frequency metrology”. Review of Scientific Instruments, Vol. 70, No.

6, pp.2567-2596, 1999.

13


Effect of the weight distribution – H-maser

14

A test was performed using 6 years of data: 2006-2011

By applying the new weighting algorithm for the H-masers:

Total weight: ~ 15% ~ 30%

Number of clocks at maximum weight: ~ 10 ~30

53500 54000 54500 55000 55500 560005

10

15

20

25

30

35

40

To

tal w

eig

ht

in p

erc

en

tag

e t

erm

s

MJD

H-ms orig weight

H-ms new weight

53500 54000 54500 55000 55500 560005

10

15

20

25

30

35

40

45

Nu

mb

er

of

H-m

asers

wit

h m

ax. w

eig

ht

MJD

H-ms orig weight

H-ms new weight


Effect on the weight distribution – Cs clocks

15

By applying the new weighting algorithm for the caesium clock:

Total weight: ~ 85% ~ 70%

Number of clocks at maximum weight: ~ 55 ~20

53500 54000 54500 55000 55500 5600055

60

65

70

75

80

85

90

95

To

tal w

eig

ht

in p

erc

en

tag

e t

erm

s

MJD

Cs orig weight

Cs new weight

53500 54000 54500 55000 55500 560000

10

20

30

40

50

60

Nu

mb

er

of

ce

siu

m c

lock

s w

ith

ma

x. w

eig

ht

MJD

Cs orig weight

Cs new weight


Results and effect on EAL

16

A test was performed using 6 years of data: 2006-2011 and by

using the quadratic prediction algorithm.

52000 53000 54000 55000

67

67.5

68

68.5

69

69.5

70

70.5

y(EAL)-y(TT)

MJD

No

rma

lize

d F

req

ue

nc

y 1

0-1

4

original weight

new weight


Stabilty of f(EAL)-f(TT)

17

By using the new weighting algorithm the long term stability of EAL is

not affected by the frequency drift of the H-Masers.

Due to short term correlation between EAL and TT we have only long

term information from this analysis.

105

106

107

108

10-17

10-16

10-15

10-14

Frequency Stability

Ov

erl

ap

pin

g A

lla

n D

ev

iati

on

Averanging Time, Seconds

Linear Prediction

original weight

new weight


Stability of EAL and TT vs USNO Rb fountain

18

The data coming from USNO Rb fountain MJD 55634-55924 are

used to evaluate EAL long term stability.

By using the new weighting

algorithm the short and

the long term stability of

EAL is improved.

simulated EAL (new weight)

simulated EAL (old weight)

105

106

107

10-16

10-15

10-14

Frequency Stability

Ov

erl

ap

pin

g A

lla

n D

evia

tio

n

Averanging Time, Seconds

Rb-EAL(new)

Rb-EAL(old)

Sigma/Tau

[d]

5 10 20 40 80

EAL 1.3E-15 8.7E-16 6.6E-16 3.9E-16 2.6E-16

EAL new 1.0E-15 6.9E-16 4.5E-16 2.5E-16 2.0E-16


Discussion and Conclusions - 1

Starting from August 2011 the new prediction algorithm has

been officially used for UTC.

The used mathematical model takes in to account the

treatment of the frequency drift.

No change in the weighting strategy has been implemented.

After 1 year of the application of the new algorithm the

positive effects are evident.

The frequency drift of EAL is almost completely removed.


Discussion and Conclusions - 2

The new prediction algorithm has no impact on the weights.

A revision of the weight algorithm has been presented to

increase the role of the H-masers in the time scale

ensemble.

The new weighting algorithm gives a good weight to the H-

masers.

Complementary studies are necessary to validate new

weighting algorithm.

20


In the framework of the Rapid UTC pilot project data clock

and time transfer data are available at 1 day interval.

At 1 day interval the dominant noise is given by the time

transfer system if the the H-masers are correctly used.

In this situation it can be interesting to test the effect of

the new weighting algorithm with 1 day interval data to

observe the short term contribution of the H-masers to the

ensemble.

21

Future developments – Rapid UTC

Improvements in ALGOS - BIPM · 2013. 9. 5. · Improvements in ALGOS G. Panfilo Bureau...

Documents

Transcript of Improvements in ALGOS - BIPM · 2013. 9. 5. · Improvements in ALGOS G. Panfilo Bureau...