David W.

Allan
Allan’s TIME
www.allanstime.com
www.ItsAboutTimeBook.com
NIST Time & Frequency Seminar 2016

1
Pioneering work of
Mandelbrot and Voss
• Developed FRACTILES
• Demonstrated mathematically and pictorially that the
world can be well modeled by self-similar and flicker-
noise like processes.
• Prior to their work, these and other noise processes
had been shown as useful noise models for clocks.
• This work has very important implications for
metrology (non-stationary statistics).

2
The variations of a clock’s output
can be well modeled as:
• SYSTEMATICS: time offset; frequency offset;
frequency drift; its sensitivity to the environment, etc.
PLUS
• RANDOM VARIATIONS: white noise, flicker noise,
random-walk noise, etc.
• On top of these is measurement noise

3
4
When characterizing a clock, you
need a reference.
• You either need to know the characteristics of the
reference, or that it’s performance is better than the
clock being measured.
• You can characterize it in the frequency-domain –
typical for short-term: Look at the Fourier frequencies
of its variations; or
• You can characterize it in the time-domain – typical for
long-term – as we will show here.
• Do both and learn more 

5
There is Gold in the Long-term; time-
domain analysis can help get the gold!

6
The time-domain techniques work for clocks, telecom
systems, navigation devices, and in many natural processes
(IEEE special issue)
• Random variations tend to be dominant in the short-term
and systematic variations in the long-term, but if the
systematics are subtracted then the random are
dominant.
• Determining the kind of random-noise allows optimum
prediction and estimation of the systematics.
• Optimally estimating the systematics and subtracting
them can enhance the performance enormously – as we
will show.

7
Example:“Hidden Value”
of Quartz
• Historically, the high-performance of quartz-crystal oscillators
has been overlooked:
• frequency offset & drift can be effectively removed
• temperature & acceleration transduced to frequency
• When systematic effects are compensated, Quartz
performance improves dramatically:
• Two examples of optimally measuring the systematics and
removing them:
1) NBS LIRTCU removed systematics; achieved ~10-12 stability with
calibration data available for 1 s every 12 h (1966); two NBS
papers on this topic
2) 5¢ Quartz Resonator (like in your wristwatch) is stable at 10
parts per billion (~10-8) over 140 days (1997). I will show you a
fascinating experiment in this regard.
8
Atomic clock Example: the HP 5071-A
cesium-beam clock
• Dr. Len Cutler and his team at HP
- They optimized the short-term stability
- They considered every systematic; measured it,
and removed it automatically to obtain long-term
stability and accuracy.
- The first atomic-clock to be “booted up!”
• Obtained the (gold) best long-term frequency stability
performance for commercial Cs. Ideal white-noise FM
over six decades +
• ~85% of the 300+ clocks making up TAI and UTC are
5071-A cesium clocks; now owned and produced by
Microsemi (the 4th owner).
9
In 1965, major comparison of H-masers and Cs
clocks; wrote AT1
• Resulted in 12 author paper: R.F.C. Vessot, H. Peters, J. Vanier, R.E.
Beehler, D. Halford, R. Harrach, D.W. Allan, D.J. Glaze, C.S. Snider, J.A.
Barnes, L.S. Cutler, and L. Bodily
• Used Allan variance to characterize and compare clocks
• AT1 time-scale algorithm generates time for NBS/NIST; optimizes
Allan variance for ensemble and estimates performance of each
clock. It is still ticking! Algorithm removes systematics and optimally
weights each clock.
- software clock better than best clock
- best clock cannot take over; is optimized
- worst clock enhances output
- robust: rejecting clocks with mal-performance
- optimizes performance for both short-term and long-
term stability performance 10
• Classical variance or Standard deviation does not converge for several
important noise processes in clocks, navigation, communication
systems, and in nature.
• Classical variance cannot distinguish the different important kinds of
noise relevant to clocks, oscillators, to navigation and communication
systems, and in nature.

11
Standard Deviation
is like
black and white 12
Characteristics of Useful Measures

• Theoretically Sound
• Easy to use and intuitive
• Relates to real situations
• Yields useful spectral information for design engineers
• Useful diagnostic tool
• Optimum smoothing, estimation, and prediction
• Communicates to the Manager (Decision Makers)

13
HISTORICAL PERSPECTIVE
• 1964 IEEE & NASA Short-term Stability Symposium
• 1966 IEEE Proceedings, Feb. Special Issue: “Frequency Stability”
• 1971 IEEE I&M, Characterization of Frequency Stability
• 1974 NBS Monograph 140
• 1981 Frequency Control Symposium: Modified “Allan Variance” with
Increased Oscillator Characterization Ability
• 1988 IEEE Standard 1139-1988: Standard Terminology for Fundamental
Frequency and Time Metrology
• Late 1980s: development of TVAR (Time Variance for telecom.)
• 1990 NIST Technical Note 1337, Characterization of Clocks and
Oscillators
• 1997 ITU HANDBOOK: “Selection and Use of Precise Frequency and
Time Systems,” Radiocommunication Bureau
• 1997 Hewlett Packard Application Note 1289, The Science of Timekeeping
• 2000-2010 Additional variance work at NIST by Dave Howe’s group gives
additional degrees of freedom and provides efficiency & tighter confidences
– http://tf.nist.gov/general/publications.htm
• Handbook of Frequency Stability Analysis, by W. J. Riley, NIST Special
Publication, SP 1065 (2007); also available at www.wriley.com
• 2016 www.ItsAboutTimeBook.com with many novel variance applications 14
 April 2016
IEEE UFFC
Special Issue
Celebrating the
50th Anniversary
Of the
Allan Variance
This Issue
Could be
A launch pad
For a large
Number of
Other
applications 15
Phase and Frequency Deviations

y(t) 
   0

 (t)
0 2 0
16
Defining Phase and Time Deviations

 (t )
x (t ) 
2 

17
In summary we

18
19
 20

Adapted from Tutorial on Quartz Crystal Resonators and Oscillators by John R. Vig

20
Definitions of:
• Synchronization – Two clocks are synchronized if they read the same
time (accounting for propagation delays and relativistic effects)
• Syntonization – Two clocks are syntonized if they are running at the
same rate in a particular frame of reference (their times can be very
different)

21
 FLIP OF A COIN IS A RANDOM UNCORRELATED PROCESS:
(white noise spectrum)

Sy ( f )  f 0
INTEGRATING THESE FLIPS GENERATES A RANDOM-WALK PROCESS:
(heads = one step forward)
(tails = one step backward)
AFTER “N” FLIPS OF A COIN, WILL BE (on average) √N AWAY FROM THE ORIGIN:

2
Sx ( f )  f
Since: y = dx/dt, or
t
x(t )   y(t ' )dt '
0
Similarly, taking a first difference is like a derivative and
turns a random-walk process into a white-noise process.
x xi 1  xi
yi  
  22
POWER-LAW SPECTRA:  
Sy ( f )  f Sx ( f )  f

23
STATISTICAL THEOREM
• The optimum estimate of the mean of a process with
a white-noise spectrum is the simple mean.
• HENCE:
• For white PM, the optimum estimate of the phase or the
time is the simple mean of the independent phase or time
readings.
- For White FM, the optimum estimate of the frequency is
the simple mean of the independent frequency readings,
which is equivalent to the last time reading minus the first
time reading divided by the data length, if there is no dead-
time between the readings.

24
25
y(t)

26
This ratio is called the bias function, B 1 (N); can
be used to ascertain kind of noise:

27
28
Sir Michael James Lighthill,
(23 January 1924 – 17 July 1998)
• He was a British applied mathematician, known for his
pioneering work in the field of aeroacoustics. (jet
engine silencing – allowed airports near cities)
• Held Lucasian Professor of Mathematics or Newton’s
chair before Stephen Hawking.
• His book, An Introduction to Fourier Analysis and
Generalised Functions (Cambridge Monographs on
Mechanics) was key to my master’s thesis, p.43! I am
standing on the shoulders of Giants with him, Jim
Barnes, Len Cutler, and several others over the years!

29
30
31
32
In 1981 we learned how to modulate the bandwidth in the software;
fundamental breakthrough

1
fs 
1
fs 
n o

n o
33
Equation 34 in 1965 Master’s thesis shows bandwidth
dependance
• Once we learned how to modulate the bandwidth in
the software, this resolved the ambiguity problem
for 𝜎 y 2(𝝉) for white PM and flicker PM and gave us
MVAR.
• In the late 1980s, it also allowed us to write a time
variance (TVAR) for the telecom community to help
them fulfill their needs: TVAR = 𝝉 2 MVAR/3;
normalized so that TVAR = classical variance for
white PM for 𝝉 = 𝝉 o like AVAR is normalized for
white FM.

34
35
Variance & Deviation nomenclature

• SQUARE SQUARE ROOT

• 𝜎 y 2(𝝉) = AVAR 𝜎 y(𝝉) = ADEV
• Mod. 𝜎 y 2(𝝉) = MVAR Mod. 𝜎 y(𝝉) = MDEV
• 𝜎 x 2(𝝉) = TVAR 𝜎 x 2(𝝉) = TDEV

36
37
38
39
Time-keeping equation:
systematics + random

This equation is idealistic but useful;

in practice what is measured is the
time or frequency difference
Between two clocks
40
Clock-time Keeping Ability or
predictability: x p and 𝝉 is
prediction time.
________________________________________________________________________

x p ( )  k   y ( )
k  1 for white FM and random walk FM

k  1 / ln 2  1.2 for flic ker FM

k  1 / 3 for white PM and flic ker PM

41
Prof. Ashby and 3 watches, so that we could
do 3-cornered hat and show 5ms time
predictability of a 5 cent crystal!

42
 First, look at the data
1. Raw data: daily times of the 3 clocks over 140 days
taken at 1400 UTC each day in Professor Ashby’s
living room.
2. Plot with mean frequency, yavg , subtracted:
(xend – xbeginning )/data length = yavg , since all the
middle terms cancel.
3. Plot with frequency drift subtracted: quadratic fit is
optimum because of white-noise PM on time
residuals.
4. Knowing the kind of noise and characteristics of the
measurement system allows us to ascertain the
source of the instabilities.
43
3 watches synchronized initially
after 140 days, watch #1 read -170s,
watch #2 read +70s, and watch #3 read -120s.
Frequency inaccuracies are about +/- 1 x 10-5

44
 After subtracting the mean frequency from each, the
peak-to-peak variations reduced: 240 seconds to about 1.7 s.
Note: 2 and 3 are strongly correlated in the long-term. (temp.?)

45
 Subtract least-squares quadratic (optimum for white PM)
Peak-to-peak residuals of about 1s/140 days = 8.3 x 10 -8 (temp.?)

46
Because clocks 2 and 3 are highly correlated in the long-term, probably due to
temperature, subtracting them removes this correlation and the difference
shows us the potential of 5 cent quartz oscillators without temperature
perturbations showing predictabilities of 0.5 s over 140 days –
an improvement of 500:1 over the raw data

47
Understanding the 3 stop-watches
measurement model allows us to sort
out the kind and size of each error
component
Let:
S = precision tripping shutter + delay from the telephone
source of UTC via WWV
L = LED readout precision 0.01s (> or = 3 ms)
C = quartz tuning fork instabilities
Our goal: to characterize 5¢ quartz crystal.

48
Characterizing 3 stop-watch measurement system
x1 = xS + xL1 + xC1 𝜎 x1(𝝉 = 1 day) = 48 ms
x2 = xS + xL2 + xC2 𝜎 x2(𝝉 = 1 day) = 52 ms
x3 = xS + xL3 + xC3 𝜎 x3(𝝉 = 1 day) = 54 ms
Since S is common: xi – xj = xLi – xLj + xCi – xCj.
Taking the variance yields:
𝜎 2ij = 𝜎 2Li + 𝜎 2Lj + 𝜎 2Ci + 𝜎 2Cj - 2𝜎 2Ci,Cj
Mod. sigma-tau diagram shows for 𝝉=1 day 𝜎 2Li >> 𝜎 2Ci
So use 3-cornered hat equation: 𝜎2i = ½(𝜎 2ij + 𝜎 2ik – 𝜎 2jk );
yields 26 ms, 40 ms, & 36 ms as LED noise (should be 3ms).
Now 𝜎 2i = 𝜎 2S + 𝜎 2Li + 𝜎 2Ci, at 𝝉=1 day ignore clock noise
And we know LED noise; therefore, calculate 𝜎S for each i.
Yields 40 ms, 33 ms, & 40 ms; average 𝜎S = 38 ms. 49
50
51
52
World record (2013) for stability
NIST ytterbium optical lattice clock

53
54
55
56
57
58
59
60
Effect of fm on TDEV
2
Effect of fm on TDEV
10
White PM + Modulation
-1
1 Empirical Fit
10

0
10

-1
10
x()

-2
10

 o sin 3 ( fm  )
-3
10
 x ( )  1.4 x pp
-4  sin( fm  o )
10

0 1 2 3 4
10 10 10 10 10
sample time, 

61
62
63
In summary: AVAR useful in characterizing all passive
atomic clocks; MVAR useful in characterizing masers,
T&F propagation and navigation systems; TVAR
useful for time trasfer instabilities and
telecommunication systems.

AVAR
MVAR
TVAR
64
As an example:
Advanced GPS Common-View
• ACV is still undeveloped (an Opportunity)
• Robin Giffard and I did these experiments in the
1990s between HP Palo Alto, CA and USNO in
Washington D.C.
• Potential of excellent inexpensive high precision, high
accuracy, long-term remote frequency stability
comparison technique

65
66
67
68
I spent 32 years at NBS/NIST, and it has been my privilege to
stand on the shoulders of giants --some of the finest people
you would want to work with; It’s always great to come back

• Also, it is inspiring to see the excellent scientists who have
come on board the Division since I retired in ‘92.
• This 3rd of July marks 56 years that I have been in the time
and frequency field and the advancements have been
remarkable; in my life time, a billion times improvement in
frequency accuracy.
• The people who come to this seminar always seem to be the
finest ilk as well.
• In 2014, I wrote a book: It’s About Time and have a new
(2016) addition of it this year: www.ItsAboutTimeBook.com
69
The book, in laymen’s English,
includes:
• Student tutorials on these three variances (AVAR, MVAR, and
TVAR), and how time is kept in the world and how GPS works.
What is TIME, and where are we in time?
• A new Unified Field Theory introducing diallel-field lines with
experiments modifying gravity and showing a universal energy
field coupling everything.
• We have done seven experiments to date – all validating this new
UFT and showing quantum states in this field.
• We address – using data and a new approach to the scientific
method – life’s major questions:
• We also deal with the secular and materialistic encroachment on
society and give some solutions to the world wide problem of
moral decay. “It’s About Time!”
70
Diallel-field line coupling of the planets to the
sun shown in sunspot activity

71
 It’s About Time!
Chapter 20: Man’s Time, How GPS works and how to make it as
much as 100 times better and to meet the accuracy comparison
goals of the world’s best clocks with some interesting
applications of these three variances;
Chapter 21: UFT experiments;
Chapter 22: Develop Transcendental Time;
Chapter 23: Where are we in Time? And I show it is the most
exciting time in the history of the planet. In the last century
science and religion have been at odds. This century is bringing
forth exciting data that brings harmony to science and religion –
countering the materialistic encroachment.

72
Some of the information in the book can be found
on our web sites:
• http://www.allanstime.com/UFT_private/ ; here are
two scientific papers.
• www.ItsAboutTimeBook.com contains 150 pages of
appendices for the book including The Science of
Timekeeping
• Chapter 21 describes not only the UFT experiments,
but one experiment that has never been published
before -- showing quantum states in the diallel-field
structure.
• http://www.allanstime.com/Spiritual/In_Touch_wit
h_Eternity.htm

73
Time
Involves
Measurements
Extraordinaire
74