Handbook of Frequency Stability Analysis

W.J. Riley Hamilton Technical Services Beaufort, SC 29907 USA

COPYRIGHT NOTICE

Copyright Notice
© 2007 Hamilton Technical Services All Rights Reserved

No part of this document may be reproduced or retransmitted in any form or by any means, electronically or mechanically, including photocopying or scanning, for any purpose other than the purchaser’s personal use, without the express written permission of Hamilton Technical Services. The information in this document is provided without any warranty of any type. Please report any errors or corrections to the address below. Handbook of Frequency Stability Analysis 03/07/2007 Available from www.lulu.com as ID # 508588 Hamilton Technical Services 650 Distant Island Drive Beaufort, SC 29907-1580 USA Phone: 843-525-6495 Fax: 843-525-0251 E-Mail: bill@wriley.com Web: www@wriley.com

ii

ABOUT THE AUTHOR

About the Author
William J. Riley, Jr. Proprietor Hamilton Technical Services 650 Distant Island Drive Beaufort, SC 29907 Phone: 843-525-6495 Fax: 843-525-0251 E-Mail: bill@wriley.com Web: www.wriley.com

Mr. Riley has worked in the area of frequency control his entire professional career. He is currently the Proprietor of Hamilton Technical Services, where he provides software and consulting services in that field, including the Stable program for the analysis of frequency stability. Bill collaborates with other experts in the time and frequency community to provide an up-to-date and practical tool for frequency stability analysis that has received wide acceptance within that community. From 1999 to 2004, he was Manager of Rubidium Technology at Symmetricom, Inc. (previously Datum), applying his extensive experience with atomic frequency standards to those products within that organization, including the early development of a chip-scale atomic clock (CSAC). From 1980-1998, Mr. Riley was the Engineering Manager of the Rubidium Department at EG&G (now PerkinElmer), where his major responsibility was to direct the design of rubidium frequency standards, including highperformance rubidium clocks for the GPS navigational satellite program. Other activities there included the development of several tactical military and commercial rubidium frequency standards. As a Principal Engineer at Harris Corporation, RF Communications Division in 1978-1979, he designed communications frequency synthesizers. From 19621978, as a Senior Development Engineer at GenRad, Inc. (previously General Radio), Mr. Riley was responsible for the design of electronic instruments, primarily in the area of frequency control. He has a 1962 BSEE degree from Cornell University and a 1966 MSEE degree from Northeastern University. Mr. Riley holds six patents in the area of frequency control, and has published a number of papers and tutorials in that field. He is a Fellow of the IEEE, and a member of Eta Kappa Nu, the IEEE UFFC Society, and served on the PTTI Advisory Board. He received the 2000 IEEE International Frequency Control Symposium I.I. Rabi Award for his work on atomic frequency standards and frequency stability analysis.

iii

During his tenure at NIST Jim made many significant contributions to the development of statistical tools for clocks and frequency standards. and in 1966 a Ph. closed-captioning was developed (which received an Emmy award) and the speed of light was measured. Colorado after a long struggle with Parkinson’s disease. In 1992. He was survived by a brother. simulation and practical understanding of clock noise. Jim worked for Austron. now the National Institute of Standards and Technology (NIST). Also. He received a Bachelors degree in engineering physics from the University of Colorado. January 13. While division chief. 5 and 6) were developed under his leadership. Jim joined the National Bureau of Standards. iv . Colorado. a Masters degree from Stanford University. and two grandchildren. three children. Rabi Award from the IEEE Frequency Control Symposium “for contributions and leadership in the development of the statistical theory. After retiring from NIST in 1982. three primary frequency standards (NBS 4. Jim Barnes died Sunday. 2002 in Boulder. he received the Distinguished PTTI Service Award. In 1995. Jim also received an MBA from the University of Denver. Jim received the NBS Silver Medal in 1965 and the Gold Medal in 1975.ieeeuffc. a pioneer in the statistics of frequency standards. Note: This biography is taken from his memoriam on the UFFC web site at: http://www. Jim was the first Chief of the Time and Frequency Division when it was created in 1967 and set the direction for this division in his 15 years of leadership. Jim was a Fellow of the IEEE. in physics from the University of Colorado. Barnes (1933-2002).asp?page=barnes.DEDICATION Dedication This handbook is dedicated to the memory of Dr. Jim received the I.org/fcmain.D. Barnes was born in 1933 in Denver. After graduating from Stanford. and the application of this understanding to the characterization of precision oscillators and atomic clocks”. James A. James A.I.

ACKNOWLEDGMENTS Acknowledgments The author acknowledges the contributions of many colleagues in the Time and Frequency community who have contributed the analytical tools that are so vital to this field. Howe in carrying on that tradition today at NIST.A.A. Weiss and C. the techniques of frequency stability analysis have advanced greatly during the last 45 years.A. encouragement. Barnes and D. and D. v . Greenhall. In particular. he wishes to recognize the seminal work of J.W. Allan in establishing the fundamentals at NBS. supporting the orders-of-magnitude progress made on frequency standards and time dissemination. I especially thank David Howe and the other members of the NIST Time & Frequency Division for their support. Together with such people as M.A. and review of this Handbook.

During the same period. vi . This Handbook summarizes the techniques of frequency stability analysis. while primary standards and the time dissemination networks that support them have improved by several orders of magnitude.PREFACE Preface The author has had the great privilege of working in the time and frequency field over the span of his career. significant advances have been made in our ability to measure and analyze the performance of those devices. and be manufactured by the tens of thousands. bringing together material that I hope will be useful to the scientists and engineers working in this field. I have seen atomic frequency standards shrink from racks of equipment to chip scale.

... Time Error Prediction.. 11 TIME DOMAIN STABILITY ....................................... Noise Identification Using B1 and R(n)....................................2................................ Hadamard Variance................................................................... MVAR........ AVAR......................................2..2........... 47 5....1.. 53 vii .......................................... 41 5................. 3.................. CONFIDENCE INTERVALS .... Integrated Phase Jitter and Residual FM ... MTIE . NewThêo1........... 13 5......................12......................... Overlapping Hadamard Variance...................................................... Standard Variance .......................... 29 5................................................3....................................................................................... 1 FREQUENCY STABILITY ANALYSIS...........................9.. Modified Total Variance ...............1................. 31 5........ 8 STABILITY MEASURES ................2.............. ThêoBR and ThêoH ............................... 14 5.......................3........................................................ 50 5.......... 3 DEFINITIONS AND TERMINOLOGY ............................................................................................................................................. NOISE MODEL . Thêo1 EDF................1.................................................. NOISE IDENTIFICATION ..............2............................................ 46 5.............. 3 BACKGROUND ......................... Simple Confidence Intervals ... 51 5......5............ 3.......2........................................17........20..... TOTVAR and TTOT EDF...................10..........................................................................................19............................................................................................................... Thêo1 ...........................................5.......................................... 44 5............................2............. 8 DIFFERENCED AND INTEGRATED NOISE ...2...........................2........ Modified Allan Variance............................................... 21 5.............................................. 50 5............................ Overlapping Allan Variance .1............................. 33 5....8..... 13 5.11.......................................................... 52 5................................................................. VARIANCES ..................... 3..........4...................................2........................13.3...................................2.............................2. Power Law Noise Identification.................................................................... 45 5.............................................................................................................5........... 52 5...........7..... 42 5. 24 5..............................6............................... Overlapping Samples ............4......................................................................................15................................2.............................. 27 5............................................................14....................................................................................... DEGREES OF FREEDOM ........... 10 4 5 STANDARDS...................................................1............2..............16.................................... 37 5....................................2............................................................................18............................... 33 5....2............. Modified Hadamard Variance ..............................2............................. Total Variance ..............1..............4..............2....................2................... 46 5....... 23 5.. 26 5.....3...........2...........................4..................................................... 8 GLOSSARY ............................................................TABLE OF CONTENTS Table of Contents 1 2 INTRODUCTION ............................. 22 5........... 18 5.............................. SIGMA-TAU PLOTS ......4............4........................................ 25 5................. TVAR and HVAR EDF ........2.......3...............5...............................1...........................................2........................................................ 17 5.. 3 2.................4.................. 7 3....2........... 19 5................... Dynamic Stability...... 7 POWER LAW NOISE .....................................................4....5................... MTOT EDF ...................................... TIE rms ........................ 43 5............................. 3.............................. Time Variance..............2...........................3.............................. Allan Variance .....2........... Time Total Variance.... Chi-Squared Confidence Intervals ......................................................... 39 5....2. 47 5......... Hadamard Total Variance ...................................

.....4..............................6.....................................11................................. 59 5............................19.............. POWER LAW DOMAIN CONVERSIONS................................5...............2................. 58 5.................................. DRIFT ANALYSIS METHODS......................... 67 5........... 6............. 59 5................................ PHASE DRIFT ANALYSIS ............................................................ 71 5..................................................................................................1.............................................. 79 POWER SPECTRAL DENSITIES .................................. BIAS FUNCTIONS .................................................................. 91 8............................................6............................8.........6................... viii ................................................................................... 69 5.............................. 86 EXAMPLE OF DOMAIN CONVERSIONS ............6....... 58 5.....................21............... 8................................................... 53 5...1...........................................3....................................................................................... 82 MULTITAPER PSD ANALYSIS ........ 6....................... 58 5..........................16...........................9.....................................5............................ 67 5.................... 6............. FREQUENCY DRIFT...................................14....8............................................................................. 65 5............................................................... 68 5.......... 82 PSD NOTES ................................... 9.........................................5...5.............. 60 5........................................................ UNEVENLY SPACED DATA ......................................................................................................................... 75 6 FREQUENCY DOMAIN STABILITY ........................ MTOT BIAS FUNCTION ...........................10................................ 82 PSD AVERAGING ....... 9.... WHITE NOISE GENERATION...........................................................5..22.5.... AND SINUSOIDAL COMPONENTS ................2.... 64 5... 56 5........... 95 HETERODYNE METHOD .... 67 5........ THÊO1 BIAS . 99 9................ The Lag 1 Autocorrelation..................................... 58 5..........7..................................................... 54 5.............................................................................. 70 5.......... 97 MEASURING SYSTEM SUMMARY ......................................... TVAR BIAS FUNCTION .. 8............................................. 9................................3.................................................................................... 79 6................ 92 FLICKER NOISE GENERATION .... 7 NOISE SPECTRA...... PARSIMONY ...... Noise ID Algorithm...................................4...................1......... 86 8 NOISE SIMULATION........................ 85 7......... TRANSFER FUNCTIONS ...............2..................3........... 92 FLICKER WALK AND RANDOM RUN NOISE GENERATION ....................................................................... B3 BIAS FUNCTION ........................................................... 92 9 MEASURING SYSTEMS......................... DEAD TIME ........12....................20..... FREQUENCY DRIFT ANALYSIS .........................2..............5................... 92 FREQUENCY OFFSET..............................................4....................................................................................................25....................... FREQUENCY OFFSET........................................................... 7.. HISTOGRAMS ........... 95 TIME INTERVAL COUNTER METHOD ................ 6........4.................................... 80 PHASE NOISE PLOT .............................1......................................................................................... ALL TAU . 84 DOMAIN CONVERSIONS .. FREQUENCY DRIFT ................17........... 8.................. The Autocorrelation Function.................................................26. B1 BIAS FUNCTION .............................................. 96 MEASUREMENT PROBLEMS AND PITFALLS .... 9....................... R(N) BIAS FUNCTION ...... 58 5...... Noise Identification Using r1................................HANDBOOK OF FREQUENCY STABILITY ANALYSIS 5................. 81 SPECTRAL ANALYSIS......................3.............. 59 5..............................................................24.... 9. 6................................................................................................................................... ENVIRONMENTAL SENSITIVITY .......... 82 PSD WINDOWING .............................................................................................7....... 72 5...........13............................................................................ 6........................................ B2 BIAS FUNCTION ................. 54 5...........................................................................................................................................18..............................................15.................................................................... 96 DUAL MIXER TIME DIFFERENCE METHOD ..................................................................... 6............................................................... 98 DATA FORMAT .........................

.....................................................7..................................... 114 CASE STUDIES................... 11....................................1................. 135 INDEX ......................................................... 11............. 99 TIMETAGS ..... 10..15.......................9...... 9...............................................................7......................................................... 118 DETECTION OF PERIODIC COMPONENTS .............. FLICKER FLOOR OF A CESIUM FREQUENCY STANDARD ...................... 100 ARCHIVING AND ACCESS ....................... 12................TABLE OF CONTENTS 9......... 110 REPORTING .......... 10. 106 GAPS................................................................................................................................................................................... 108 DRIFT ANALYSIS ..................................................................... 117 PHASE NOISE IDENTIFICATION .......................................................... 125 TEST SUITES............................................................ 126 NBS DATA SET .......................................................................................... 135 ix ............................................................................................... SOFTWARE VALIDATION.....................................................................3.... 10........................... 108 SPECTRAL ANALYSIS .. 12................. 131 BIBLIOGRAPHY................. 109 VARIANCE SELECTION ..... 120 STABILITY UNDER VIBRATIONAL MODULATION ...........................6.................. 10................ 123 COMPOSITE AGING PLOTS ....... 109 THREE-CORNERED HAT............ 125 12.........11........................................ 122 WHITE FM NOISE OF A FREQUENCY SPIKE .............................................. 107 PHASE-FREQUENCY CONVERSIONS ............. 128 13 14 GLOSSARY ..... 12...............................................................................13...................6......................................... 11.................... 127 IEEE STANDARD 1139-1999....................2...... 10....5............................................................. 105 PREPROCESSING .......... 11 DATA PRECISION ..... 10..................................9. 123 12 SOFTWARE .................................................................................8................................................................................................ 126 1000-POINT TEST SUITE ............. 10 DATA QUANTIZATION ....... 107 ANALYSIS OF DATA WITH GAPS ........ 115 11. 10.......................................................... 10........... 11.... 115 HADAMARD VARIANCE OF A SOURCE WITH DRIFT ......................................................... 108 VARIANCE ANALYSIS .................4................................................ 103 10..............................5... 9...................................................................................... 10..........................................................................................8............................ 10.......................7....... 10..................14....................... 108 DATA PLOTTING ............................................................... 106 GAP HANDLING .................... JUMPS AND OUTLIERS ........... 10............................................................. 10................ 100 ANALYSIS PROCEDURE ..........2.......... 11.......3...................................................................................................................1................... 12..................4............10.............................................................................................................5..................................................2..4.................................. 11................................................................................ 10................................................................................................. 108 OUTLIER RECOGNITION ........................3.......................... 107 UNEVEN SPACING ..1.....12.........................................

HANDBOOK OF FREQUENCY STABILITY ANALYSIS x .

Throughout these analyses. [3] for this subject are given below.L. Two general references [2]. 2. Department of Commerce. D.W.gov/timefreq/general/pdf/554. D. If this is your first exposure to this field. where specialized statistical variances have been developed to characterize clock noise as a function of averaging time.W.A. It is often important to separate deterministic factors such as aging and environmental sensitivity from the stochastic noise processes. measuring systems and data formats. post processing steps. pp. it is worthwhile to remember R. analyze stability and report results. Beaufort. NIST Technical Note 1337. Some of the examples use the Stable32 program [1].W Allan. Hamming’s axiom that “the purpose of computing is insight. you may find it helpful to scan the sections to gain some perspective regarding frequency stability analysis.A.1 INTRODUCTION 1 Introduction This handbook describes practical techniques for frequency stability analysis. D. SC 29907. Also on the NIST web site at http://tf. It covers the definitions of frequency stability.S. Hamilton Technical Services. 1 . which is a commercially available tool for studying and performing frequency stability analyses. 3. One must always be aware of the possibility of outliers and other measurement problems that can contaminate the data.com. Howe and F. This handbook can be used both as a tutorial and as a reference. I strongly recommend consulting the references as part of your study of this subject matter.A. preprocessing steps. on Freq.gov/timefreq/general/pdf/868. Barnes. 1-47. March 1990. "Characterization of Clocks and Oscillators". May 1981.pdf. U. Symp. The emphasis is on time domain stability analysis. Proc. National Institute of Standards and Technology.wriley. The Stable32 Program for Frequency Stability Analysis. The analyst should feel free to use his intuition and experiment with different methods that can provide a deeper understanding. D. analysis tools and methods. Suggested analysis procedures are recommended to gather data. Methods are presented to perform those calculations. and reporting suggestions. http://www.nist. Allan and J. not numbers”. preprocess it.pdf. identify noise types and determine confidence limits.B Sullivan.nist. 35th Annu..Walls (Editors). Examples are included for many of these techniques. Howe. "Properties of Signal Sources and Measurement Methods''. D. Contrl. References for Introduction 1. http://tf.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS 2 .

and the introduction of the 2-sample (Allan) variance in 1966 [2] marked the beginning of this new era. yet complete. 2. and their dependence on time and environmental conditions. In particular. has units of seconds. respectively.. This period also marked the introduction of commercial atomic frequency standards. The data sampling or measurement interval. where the index i refers to data points frequency domains. and the use of the LORAN radio navigation system for global precise time and frequency transfer. heterodyne period measurements with electronic counters. not a population of such devices. It is often best to characterize and remove deterministic factors (e.. A stability analysis may be concerned with both the stochastic (noise) and deterministic (systematic) properties of the device under test. the fluctuations of those values. quantitative and standardized description of the phase and frequency of the source. increased emphasis on low phase noise.g. The goal of a time domain stability analysis is a concise.1. which was summarized in a special issue of the Proceedings of the IEEE in 1966 [3]. ∆f/f. Computation time is generally not as much of a factor. and low noise phase noise measurements with doublebalanced diode mixers). Environmental effects are often best handled by eliminating them from the test conditions. which is the (finite) population under analysis. where m is the averaging factor). new statistics became available that were better suited for common clock noises than the classic N-sample variance. The analysis may show that this is not true. A common problem for time domain frequency stability analysis is to produce results at the longest possible analysis interval in order to minimize test time and cost. in which case the data record may have to be partitioned to obtain meaningful results. t. The subsequent advances in the performance of frequency sources depended largely on the improved ability to measure and analyze their 3 .SECTION 2 FREQUENCY STABILITY ANALYSIS 2 Frequency Stability Analysis The time domain stability analysis of a frequency source The objective of a frequency is concerned with characterizing the variables x(t) and stability analysis is to y(t). It is accomplished frequency fluctuations of a with an array of phase and frequency data arrays. and better methods were developed for high resolution measurements (e. xi and frequency source in the time and yi respectively. Both are commonly called "phase" to distinguish them from the independent time variable. A frequency stability analysis is normally performed on a single device. It is also assumed that the frequency reference instability and instrumental effects are either negligible or removed from the data. The analysis interval or period. The xi values have units of time in seconds. the phase (expressed in units of time error) and the characterize the phase and fractional frequency. and the yi values are (dimensionless) fractional frequency.g. The output of the device is generally assumed to exist indefinitely before and after the particular data set was measured. may be a multiple of τ0 (τ = mτ0. including their nominal values. frequency drift and temperature sensitivity) before analyzing the noise. loosely called “averaging time”. equally spaced in time. τ0. τ. The x(t) time fluctuations are related to the phase fluctuations by φ(t) = x(t)·2πν0. It is also generally assumed that the stochastic characteristics of the device are constant (both stationary over time and ergodic over their population). Background The field of modern frequency stability analysis began in the mid 1960’s with the emergence of improved analytical and measurement techniques. A seminal conference on short-term stability in 1964 [1]. where ν0 is the nominal carrier frequency in Hz.

These advances also mean that the field of frequency stability analysis has become more complex. The error bars show the improvement in statistical confidence for the same data set.HANDBOOK OF FREQUENCY STABILITY ANALYSIS stability. while the extension to longer averaging time provides better long-term clock characterization without the time and expense of a longer data record. An example of the progress that has been made in frequency stability analysis from the original Allan variance in 1966 through Thêo1 in 2003 is shown in the plots below. It is the goal of this handbook to help the analyst deal with this complexity. Original Allan Overlapping Allan Total Thêo1 Overlapping & Thêo1 4 .

Feb. 5 . Vol.SECTION 2 FREQUENCY STABILITY ANALYSIS This handbook includes detailed information about these (and other) stability measures. 2. References for Frequency Stability Analysis 1. Proceedings of the IEEE-NASA Symposium on the Definition and Measurement of Short-Term Frequency Stability. D. Allan. 2. 54. 54. IEEE. No. Proc. NASA SP-80. 1966. "The Statistics of Atomic Frequency Standards''. Proc. No. pp. IEEE. Vol. 1964. 3. Special Issue on Frequency Stability.W. Feb. 221-230. Nov. 1966. 2.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS 6 .

has units of seconds. and the type and magnitude of the noise. Phase noise is fundamental to a frequency stability analysis. . These data are equivalent. The analysis or averaging time. The x(t) time fluctuations are related to the phase fluctuations by φ(t) = x(t) ⋅ 2πν0 . 7 . we define the fractional frequency as y(t ) = where 1 dφ dx ∆f ν ( t ) − ν 0 = = = 2πν 0 dt dt . like most Specialized definitions and others. For the analysis of frequency stability. where the index i refers to data points in time. ν0 f x ( t ) = φ ( t ) / 2πν 0 . along with other factors such as aging and environmental sensitivity. where m is the averaging factor). Noise Model A frequency source has a sine wave output signal given by [1] V ( t ) = V0 + ε ( t ) sin 2πν 0t + φ ( t ) where V0 = nominal peak output voltage ε(t) = amplitude deviation ν0 = nominal frequency φ(t) = phase deviation. Both are commonly called "phase" to distinguish them from the independent time variable. may be a multiple of τ0 (τ = mτ0.SECTION 3 DEFINITIONS AND TERMINOLOGY 3 Definitions and Terminology The field of frequency stability analysis. 3. ∆f/f. The x values have units of time in seconds. τ. For precision oscillators. xi and yi.1. analysis is an array of equally spaced phase (really time error) or fractional frequency deviation data arrays. determine the stability of the frequency source. respectively. and the y values are (dimensionless) fractional frequency. The instantaneous frequency is the derivative of the total phase: ν (t ) = ν 0 + 1 dφ 2π dt . we are concerned primarily with the φ(t) term. has its own specialized definitions and terminology are used for terminology. The data sampling or measurement interval. where ν0 is the carrier frequency in hertz. The basis of a time domain stability frequency stability analysis. and conversions between them are possible. t. τ0.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS

3.2.

Power Law Noise

It has been found that the instability of most frequency sources can be modeled by a combination of power-law noises having a spectral density of their fractional frequency fluctuations of the form Sy(f) ∝ f α, where f is the Fourier or sideband frequency in hertz, and α is the power law exponent. Noise Type White PM Flicker PM White FM Flicker FM Random Walk FM Flicker Walk FM Random Run FM

α
2 1 0 -1 -2 -3 -4

Examples of the four most common of these noises are shown in the table below:

3.3.

Stability Measures

The standard measures for frequency stability in the time and frequency domains are the overlapped Allan deviation, σy(τ), and the SSB phase noise, £(f), as described in more detail later in this handbook.

3.4.

Differenced and Integrated Noise

Taking the differences between adjacent data points plays an important role in frequency stability analysis for performing phase to frequency data conversion, calculating Allan (and related) variances, and doing noise identification using the lag 1 autocorrelation method [2]. Phase data x(t) may be converted to fractional frequency data y(t) by taking the first 8

SECTION 3

DEFINITIONS AND TERMINOLOGY

differences xi+1 - xi of the phase data and dividing by the sampling interval τ. The Allan variance is based on the first differences yi+1 - yi of the fractional frequency data or, equivalently, the second differences yi+2 - 2yi+1 + yi of the phase data. Similarly, the Hadamard variance is based on third differences xi+3 - 3xi+2 + 3xi+1 - xi of the phase data. Taking the first differences of a data set has the effect of making it less divergent. In terms of its spectral density, the α value is increased by 2. For example, flicker FM data (α = -1) is changed into flicker PM data (α = +1). That is the reason that the Hadamard variance is able to handle more divergent noise types (α ≥ -4) than the Allan variance (α ≥ -2) can. It is also the basis of the lag 1 autocorrelation noise identification method whereby first differences are taken until α becomes ≥ 0.5. The plots below show random run noise differenced first to random walk noise and again to white noise.
15000 Random Run Noise 50 40 30 10000 Amplitude Amplitude 20 10 0 -10 -20 0 -30 Random Walk Noise

5000

0

200

400 Point #

600

800

1000

0

200

400 Point #

600

800

1000

Original Random Run (RR) Noise

Differenced RR Noise = Random Walk (RW) Noise
White Noise

6

4

2 Amplitude

0

-2

-4

-6

0

200

400 Point #

600

800

1000

Differenced RW Noise = White (W) Noise The more divergent noise types are sometimes referred to by their color. White noise has a flat spectral density (by analogy to white light). Flicker noise has an f-1 spectral density, and is called pink or red (more energy toward lower frequencies). Continuing the analogy, f-2 (random walk) noise is called brown, and f-3 (flicker walk) noise is called black, although that terminology is seldom used in the field of frequency stability analysis. Integration is the inverse operation of differencing. Numerically integrating frequency data converts it into phase data (with an arbitrary initial value). Such integration subtracts 2 from the original α value. For example, the random run data in the top left plot above was generated by simulating random walk FM data and converting it to phase data by numerical integration. 9

HANDBOOK OF FREQUENCY STABILITY ANALYSIS

3.5.

Glossary

See the Glossary chapter at the end of this handbook for brief definitions of many of the important terms used in the field of frequency stability analysis.

References for Definitions and Terminology
1. 2. "IEEE Standard Definitions of Physical Quantities for Fundamental Frequency and Time Metrology - Random Instabilities", IEEE Std 1139-1999, July 1999. W.J. Riley and C.A. Greenhall, “Power Law Noise Identification Using the Lag 1 Autocorrelation”, Proceedings of the 18th European Frequency and Time Forum, University of Surrey, Guildford, UK, 5 - 7 April 2004.

10

2. 4. Sydnor (Editor). General Specification For.I. MIL-PRF-55310. Oscillators. Report 580. IEEE-Std-1139 contains definitions. Crystal. July 1999. as shown in the the field of frequency references below [1]-[5]. 11 . measurement methods. Guide to the Expression of Uncertainty in Measurement. 1986. These standards define terminology. 142-150. In particular. stability analysis. pp. 1995. “The Selection and Use of Precise Frequency Systems”.R. "Characterization of Frequency and Phase Noise". 1995.Random Instabilities". International Standards Organization. ITU-R Handbook. International Consultative Committee (C. recommendations. etc. means for characterization and specification. R.L. 3.C. "IEEE Standard Definitions of Physical Quantities for Fundamental Frequency and Time Metrology . and examples for the characterization of frequency stability. IEEE Std 1139-1999.).SECTION 4 STANDARDS 4 Standards Standards have been adopted for the measurement and Several standards apply to characterization of frequency stability. ISBN 92-67-10188-9. References for Standards 1. 5.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS 12 .

function of time. Sigma-Tau Plots The most common way to express the time domain stability of a frequency source is by means of a sigma-tau plot that shows some measure of frequency stability versus the time over which the frequency is averaged. This analysis generally uses nd measures are based on the some type of variance. a 2 moment measure of the statistics of the phase or fluctuations. which is based on the variations around the average value. as shown on the following log s vs. but are different on a Mod sigma plot. which is often used to distinguish between them. Log sigma versus log tau plots show the dependence of stability on averaging time. a form of Time domain stability time series analysis [1]. log τ plots. 5. The power law noises have particular slopes. and other variances have been developed that provide a better characterization of such devices. which dependence shows the properties of the noise. A key aspect of such a characterization is the dependence of the variance on the averaging time used to make the measurement. 13 . is not convergent.SECTION 5 TIME DOMAIN STABILITY 5 Time Domain Stability The stability of a frequency source in the time domain is based on the statistics of its phase or frequency fluctuations as a function of time. and show both the stability value and the type of noise. the standard variance. and α and µ are related as shown in the table below: Noise W PM F PM W FM F FM RW FM α 2 1 0 -1 -2 µ -2 ~ -2 -1 0 1 The log σ versus log τ slopes are the same for the two PM noise types. µ. For many divergent noise types commonly frequency fluctuations as a associated with frequency sources.1.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS Sigma Tau Diagram -9 White PM or Flicker PM σy(τ) ∼ τ α Sy(f) ∼ f µ = -α-1 µ/2 log σy(τ) -11 -13 τ -1 Freq Drift White Flicker FM FM RW FM τ +1 -15 0 2 τ -1/2 4 τ log τ 0 τ 8 +1/2 10 6 Mod Sigma Tau Diagram -9 White PM Mod σy(τ) ∼ τ Sy(f) ∼ f Flicker PM α µ′/2 log -11 -3/2 Mod τ σy(τ) -13 µ′ = -α-1 Freq Drift White FM Flicker RW FM FM τ -15 0 -1 τ +1 τ 2 -1/2 4 τ log τ 6 0 τ 8 +1/2 5. summed. with more details to follow. Variances Variances are used to characterize the fluctuations of a frequency source [2. say. the length of rods around a nominal value. The variations from the mean are squared. Several statistical variances are available to the frequency stability analyst. and divided by one less than the number of measurements. and there are several versions of it that provide better statistical confidence. The Allan variance is the most common time domain measure of frequency stability. and this section provides an overview of them.2. These are second-moment measures of scatter. 3]. 14 . this number is called the “degrees of freedom”. much as the standard variance is used to quantify the variations in. can distinguish between white and flicker phase noise.

and the Hadamard variances to d = 3. Overlapping types provide better confidence than classic Allan variance.0 on the ADEV plots below. Total types provide better confidence than corresponding overlapping types. One obtains unbiased estimates of this variance from available phase data by computing time averages of the differences squared. The Allan variances correspond to d = 2. The modified Allan deviation can be used to distinguish between white and flicker PM noise. on the MDEV plots. The Hadamard variance can better handle frequency drift and more divergence noise types. and several versions of it are also available. a multiple of τ0.most widely used – first choice Used to distinguish W and F PM Based on modified Allan variance Rejects frequency drift. The newer Total and Thêo1 variances can provide better confidence at longer averaging factors.5 and –1. and handles divergent noise Better confidence than normal Hadamard Better confidence at long averages for Allan Better confidence at long averages for modified Allan Better confidence at long averages for time Better confidence at long averages for Hadamard Provides information over nearly full record length Hybrid of Allan and Thêo1 variances All are second moment measures of dispersion – scatter or instability of frequency from central value. Hadamard types also converge for FW and RR FM noise. which use dth differences of averaged phase samples. Thêo1 (Theoretical Variance #1) provides stability data out to 75 % of record length. These give respectively the overlapped estimator and non-overlapped estimators of the stability.0. 15 . Some are quite computationally intensive. Modified types have additional phase averaging that can distinguish W and F PM noises. For example.SECTION 5 TIME DOMAIN STABILITY and can describe time stability. and modified variances. but they can be distinguished as –1. τ. which use dth differences of phase samples. The corresponding variances are defined as a scaling factor times the expected value of the differences squared. Variance Type Standard Allan Overlapping Allan Modified Allan Time Hadamard Overlapping Hadamard Total Modified Total Time Total Hadamard Total Thêo1 ThêoH • • • • • • • • • • • Characteristics Non-convergent for some clock noises – don’t use Classic – use only if required – relatively poor confidence General Purpose . There are two categories of stability variances: unmodified variances. the W and F PM noise slopes are both ≈ -1. All are usually expressed as deviations. especially if results are wanted at all (or many) analysis intervals (averaging times). Time variances based on modified types. All are normalized to standard variance for white FM noise. The usual choices for the increment between estimates (the time step) are the sample period τ0 and the analysis period τ. respectively. All except standard variance converge for common clock noises.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS ADEV MDEV W PM F PM The Hadamard deviation may be used to reject linear frequency drift when a stability analysis is performed. the simulated frequency data for a rubidium frequency standard in the left plot below shows significant drift. but it has lower confidence for a given τ. Notice that. For example. a sign of linear frequency drift. Allan deviation plots for these data are shown in the right hand plots for the original and drift-removed data. the Allan deviation plot has a +τ dependence at long τ. However. the Hadamard deviation for the original data is nearly the same as the Allan deviation after drift removal. without drift removal. 16 .

The standard variance is usually expressed as its square root. 1970. Stein. 5.A.2. E. the standard deviation. 2. Rutman.1. 3. and y= 1 N ∑y i =1 N i is the average frequency.” Precision Frequency Control. Gerber and A. 1048-1075. 1 N s = ∑ yi − y N − 1 i =1 2 b g 2 . Standard Variance The classic N-sample or standard variance is defined as The standard variance should not be used for the analysis of frequency stability.E. S.M. vol. as shown in the figure below. New York: Academic Press. “Frequency and Time – Their Measurement and Characterization.SECTION 5 TIME DOMAIN STABILITY References for Variances 1.P. Ballato (eds. 2.). Box and G.” Proceedings of the IEEE. 17 . 66(9). pp. G. “Characterization of Phase and Frequency Instabilities in Precision Frequency Sources: Fifteen Years of Progress. Vol. Jenkins.R. It is not recommended as a measure of frequency stability because it is nonconvergent for some types of noise commonly found in frequency sources. J. s. 1985. Time Series Analysis: Forecasting and Control. where the yi are the N fractional frequency values. San Francisco: Holden-Day. 1978.

which is not stationary for the more divergence noise types. In the context of frequency stability analysis. and can describe time stability.5 2. the standard variance is used primarily in the calculation of the B1 ratio for noise recognition. There are several versions of the Allan variance that provide better statistical confidence. pp. 1987. as described for the Allan variance below. “Should the Classical Variance be used as a Basic Measure in Standards Metrology?”. The original non-overlapped Allan. 646-654. That problem can be solved by instead using the first differences of the fractional frequency values (the second differences of the phase). Similar to the standard variance. . is the standard time domain measure of frequency stability [1. D. The problem with the standard variance stems from its use of the deviations from the average. or 2-sample variance.2. Reference for Standard Variance 1. Allan Variance The Allan variance is the most common time domain measure of frequency stability.0 Standard or Allan Deviation 2. IM-36. 5. AVAR..0 10 100 Sample Size (m=1) 1000 The standard deviation (upper curve) increases with the number of samples of flicker FM noise used to determine it.W. but has the advantage of being convergent for most types of clock noise. Meas. IEEE Trans. while the Allan deviation (lower curve and discussed below) is essentially constant.5 1.0 1. it is a measure of the fractional frequency fluctuations.2. Allan. It is defined as 18 The original Allan variance has been largely superseded by its overlapping version.HANDBOOK OF FREQUENCY STABILITY ANALYSIS Convergence of Standard & Allan Deviation for F FM Noise 3. Instrum. can distinguish between white and flicker phase noise. 2].

but at the expense of greater computational time. as shown in the diagram and formulae below.2. The overlapping samples are not completely independent. but is often estimated as ±σy(τ)/√N. 19 . Other variances (e.SECTION 5 TIME DOMAIN STABILITY σ 2 (τ ) = y M −1 1 2 ∑ yi+1 − yi . The choice of overlapping samples applies to the Allan and Hadamard variances. the Allan variance may be calculated as σ 2 (τ ) = y 1 2( N − 2)τ 2 N −2 ∑ i =1 xi+ 2 − 2 xi+1 + xi . 2( M − 1) i =1 where yi is the ith of M fractional frequency values averaged over the measurement (sampling) interval. but has the advantage.g. τ. for more divergent noise types such as flicker noise. 2 where xi is the ith of the N = M+1 phase values spaced by the measurement interval τ. The result is usually expressed as the square root. but do increase the effective number of degrees of freedom.3. whereby the calculation is improve the confidence of a stability performed by utilizing all possible combinations of estimate. the Allan deviation. Overlapping samples don’t apply at the basic measurement interval. Overlapping Samples Some stability calculations can utilize (fully) Overlapping samples are used to overlapping samples.. ADEV. which should be as short as practical to support a large number of overlaps at longer averaging times. The use of overlapping samples improves the confidence of the resulting stability estimate. σy(τ). the data set. The confidence interval of an Allan deviation estimate is also dependent on the noise type. Note that these y symbols are sometimes shown with a bar over them to denote the averaging. The Allan variance is the same as the ordinary variance for white FM noise. of converging to a value that is independent on the number of samples. In terms of phase data. total) always use them. 5.

m =3 1 2 Non-Overlapping Samples 3 4 1 2 3 4 5 Overlapping Samples Non-Overlapped Allan Variance: Stride = τ = averaging period = m⋅τ0 Overlapped Allan Variance: Stride = τ0 = sample period σ y2 τ = b g 2b M1 − 1g ∑ b y i =1 M −1 i +1 − yi g 2 σ y2 τ = bg 1 2 2m M − 2m + 1 M − 2 m+1 j + m−1 j =1 i= j b g ∑ ∑ by i+ m − yi g 2 The following plots show the significant reduction in variability. hence increased statistical confidence. obtained by using overlapping samples in the calculation of the Hadamard deviation: Non-Overlapping Samples Overlapping Samples Hσ y τ = 2 b g 6b M1− 2g ∑ b y i =1 M −2 i+2 − 2 yi +1 + yi g 2 Hσ y τ = 2 bg 1 6m2 M − 3m + 1 M −3 m+1 j + m−1 j =1 i= j b g ∑ ∑ by i+2 m − 2 yi + m + yi g 2 20 .HANDBOOK OF FREQUENCY STABILITY ANALYSIS Averaging Factor.

σ²y(τ). can be first integrated to use this faster formula. or AVAR. where m is the averaging factor and τ0 is the basic measurement interval. even though the additional overlapping differences are not all statistically independent. Sample variances are distributed according to the expression χ = 2 df ⋅ s2 σ2 . the overlapping Allan variance can be estimated from a set of N = M+1 time measurements as σ 2 (τ ) = y 1 2( N − 2m)τ 2 N −2 m ∑ i =1 x i + 2 m − 2 xi + m + xi . σy(τ). The term AVAR has come to be used mainly for this form of the Allan variance.2. that makes maximum use of a data set by forming all possible overlapping samples at each averaging time τ. they nevertheless increase the number of degrees of freedom and thus improve the confidence in the estimation. 21 . V W 2 This formula is seldom used for large data sets because of the computationally intensive inner summation. It can be estimated from a set of M frequency measurements for averaging time τ = mτ0.SECTION 5 TIME DOMAIN STABILITY 5. The result is usually expressed as the square root. and using that to establish single. Analytical methods are available for calculating the number of degrees of freedom for an estimation of overlapping Allan variance. by the expression M − 2 m+1 1 σ (τ ) = ∑ 2 m2 ( M − 2m + 1) j =1 2 y The overlapped Allan deviation is the most common measure of timedomain frequency stability. based on Chi-squared statistics. the Allan deviation. where χ² is the Chi-square. The confidence interval of an overlapping Allan deviation estimate is better than that of a normal Allan variance estimation because.or double-sided confidence intervals for the estimate with a certain confidence factor. and df is the number of degrees of freedom (not necessarily an integer). ADEV. yi. df is determined by the number of data points and the noise type. Overlapping Allan Variance The fully overlapping Allan variance. is a form of the normal Allan variance. In terms of phase data. R S∑ T i= j j + m−1 yi + m − yi U. σ² is the true variance. s² is the sample variance. 2 Fractional frequency data.4. For a particular statistic. and ADEV for its square root.

S. Greenhall.. 2. W 2 i The result is usually expressed as the square root. 2. "Does Allan Variance Determine the Spectrum?". 623-627. Vol. C. C. where m is the averaging factor and τ0 is the expression M − 3 m+ 2 1 Modσ y (τ ) = ∑ 4 2m ( M − 3m + 2) j =1 2 Use the modified Allan deviation to distinguish between white and flicker PM noise.S Department of Commerce. the modified Allan deviation. J. J. "Characterization of Frequency Stability". MVAR. Allan.A. 1970. 6. 5. et al.HANDBOOK OF FREQUENCY STABILITY ANALYSIS References for Allan Variance 1. The modified Allan variance is the same as the normal Allan variance for m = 1.A. and has the advantage of being able to distinguish between white and flicker PM noise. NBS Technical Note 394. 2. 105-120. It includes an additional phase averaging operation. D. No. D. IEEE. No. It is estimated from a set of M frequency measurements for averaging time τ = mτ0. "Allan Variance". 221-230. 1966. by the R FG y S∑ H∑ T j + m−1 i + m−1 i= j k =i k +m − yk IJ U . IEEE Trans. May 1971.W. Mod σy(τ). "Characterization of Frequency Stability". June 1997. basic measurement interval. pp. Instrum. Feb. U. pp. Symp. pp. KV W 2 In terms of phase data.. “Spectral Ambiguity of Allan Variance”. June 1998. Barnes. National Bureau of Standards. 7. 3. NIST Technical Note 1318. the modified Allan variance is estimated from a set of N = M+1 time measurements as Modσ 2 y bg 1 τ = 2 2 2 m τ N − 3m + 1 N −3 m+1 b g ∑ j =1 R x S∑ T j + m−1 i= j i+2 m − 2 xi + m U +x V . Vol. Cont. The confidence interval of a modified Allan deviation determination is also dependent on the noise type. IM-47.2. IEEE Trans. IM-20. Mod σ²y(τ). pp. Greenhall. 1997 Intl. Instrum. 4. Modified Allan Variance The modified Allan variance. 358-365. but is often estimated as ±σy(τ)/√N. Meas. "The Statistics of Atomic Frequency Standards''. Proc. Meas. Department of Commerce. National Institute of Standards and Technology. 5. “Variances Based on Data with Dead Time Between the Measurements”. Allan.A.5. 22 . No. 3.A. Freq. Proc. is another common time domain measure of frequency stability [1].W.. Vol. Allan's TIME. Barnes. U. 1990. 54. Oct.

IM-33. IM-46. Greenhall. pp. 4. No. Instrum. May 1981. Proc. 696-703. No.SECTION 5 TIME DOMAIN STABILITY References for Modified Allan Variance 1. C.2. Symp.A. 470-474.. May 1995. Barnes. Lesage and T. 3. TDEV is MDEV whose slope on a log-log plot is transposed by +1 and normalized by √3. on Freq. pp. IEEE Trans. pp. In simple terms.. Allan and J. 332-336. "A Modified Allan Variance with Increased Oscillator Characterization Ability''. D. Greenhall. “Characterization of Frequency Stability: Analysis of the Modified Allan Variance and Properties of Its Estimate”. pp. Contrl. It is defined as Use the time deviation to characterize the time error of a time source (clock) or distribution system. Symp. June 1997. Time Variance The time Allan variance. is a measure of time stability based on the modified Allan variance [1]. Dec. It is particularly useful for measuring the stability of a time distribution network. TVAR. as shown in the following figure: 23 . Meas. 5. IEEE 1995 Freq. Vol. "Estimating the Modified Allan Variance". IEEE Trans. σ²x(τ) = (τ²/3)·Mod σ²y(τ). with square root TDEV.A.W. Instrum. 346-353. "The Third-Difference Approach to Modified Allan Variance". Vol.A. C... 3.6. 2. 4. The time Allan variance is equal to the standard variance of the time deviations for white PM noise. 1984. It can be convenient to include TDEV information on a MDEV plot by adding lines of constant TDEV. Meas. Contrl. Proc. Ayi. 35th Annu. P.

For consistency. Allan. 107-116. pp. the frequency offset. 667-678. where ∆T is the total time error. Davis. and noise considerations. N..W. In that case. May 1991. 2. as shown in the following expression: The time error of a clock can be predicted from its time and frequency offsets.A.g. temperature) as expected during subsequent operation. D. ∆f/f is the sum of the initial and average environmentally induced frequency offsets.. Contrl. 44th Annu. on Freq.. • Initial Syntonization The effect of an initial frequency (syntonization) error.W. The initial syntonization should be performed. Symp. pp. 24 . J. ∆f/f . Without occasional resyntonization (frequency recalibration). and measurement noise. "New Inexpensive Frequency Calibration Service From NIST''. Symp. • Initial Synchronization The effect of an initial time (synchronization) error. Allan.. Levine.HANDBOOK OF FREQUENCY STABILITY ANALYSIS References for Time Variance 1. Time Error Prediction The time error of a clock driven by a frequency source is a relatively simple function of the initial time offset. and noise. M. D. the finite measurement resolution. 5.g. and it is usually necessary to generate a timing error budget. Proc. under the same environmental conditions (e. The measurement noise can be estimated by the square root of the sum of the Allan variances of the clock and reference over the measurement interval. frequency drift. the syntonization error is subject to uncertainty due to the frequency reference. is a linear time error. D. June 1990.7. ∆T = To + (∆f/f) ⋅ t + ½ D ⋅ t2 + σx(t). Weiss and J.D. Hironaka. the measurement and tuning resolution. M. clock error prediction is seldom easy or exact. Because of the many factors. and D. units of dimensionless fractional frequency and seconds should be used throughout. Okayama. and assumptions involved. is a constant time offset due to the time reference. plus the effect of noise.L. Weiss. Proc. To is the initial synchronization error. "A Frequency-Domain View of TimeDomain Characterization of Clocks and Time and Frequency Distribution Systems''. Therefore.g. conditions. Contrl. and the subsequent frequency drift. to the greatest extent possible. Jespersen.A. and σx(t) is the (rms) noise-induced time deviation. To.. it can be important to have a means for periodic clock syntonization (e. 45th Annu. and their variability. The measurement resolution and noise depends on the averaging time. frequency aging can cause this to be the biggest contributor toward clock error for many frequency sources (e. quartz crystal oscillators and rubidium gas cell standards). GPS or cesium beam standard). D is the frequency drift (aging rate). on Freq.2.

Allan and H. making it particularly useful for the analysis of rubidium atomic clocks [6. For the purposes of time-domain frequency stability characterization. the Hadamard variance is defined as: Hσ 2 τ = y bg b 1 6 M −2 g∑ i =1 M −2 yi+ 2 − 2 yi+1 + yi .SECTION 5 TIME DOMAIN STABILITY • Environmental Sensitivity After initial syntonization. with the overlapping version providing better estimates for this statistic. it converges for the Flicker Walk FM (α = -3) and Random Run FM (α = -4) power-law noise types. This requires a good understanding of the both the device and its environment. As a spectral estimator. Hellwig. overlapping. the Hadamard transform has higher resolution than the Allan variance. and the Hadamard total variance offering improved confidence at large averaging factors. 2 where yi is the ith of M fractional frequency values at averaging time τ. “Time Deviation and Time Prediction Error for Clock Specification.2. Because the Hadamard variance examines the second difference of the fractional frequencies (the third difference of the phase variations). Characterization. It has also been used as one of the components of a time-domain multivariance analysis [5]. environmental sensitivity is likely to be the largest contributor to time error. which was adapted by Baugh as the basis of a time-domain measure of frequency stability [3]. 25 . Hadamard Variance The Hadamard variance is a three-sample variance Use the Hadamard variance to similar to the two-sample Allan variance that is characterize frequency sources with commonly applied for the analysis of frequency divergent noise and/or frequency drift. Reference for Time Error Prediction D.W. it is important to separate the deterministic environmental sensitivities from the stochastic noise. modified.476τ respectively [4].8. stability data that has highly divergent noise (α < −2) or linear frequency drift. the most important advantage of the Hadamard variance is its insensitivity to linear frequency drift. 5. and total versions of the Hadamard variance. The Hadamard [1] variance is based on the Hadamard transform [2]. March 1981.2337N-1τ-1 -1 and 0. It is also unaffected by linear frequency drift. and is related to the third structure function of phase noise [8]. When performing a frequency stability analysis. Environmental frequency sensitivity obviously depends on the properties of the device and its operating conditions. 7]. For frequency data. and Application”. since the equivalent noise bandwidth of the Hadamard and Allan spectral windows are 1. There are normal.

For frequency data.9. The result is usually expressed as the square root. the Hadamard deviation. 2 where xi is the ith of N = M+1 phase values at each measurement time. but the confidence interval of the estimation is better. Like the Allan variance. an inner averaging loop over m frequency values is necessary. Computation of the overlapping Hadamard variance is more efficient for phase data. 2 where xi is the ith of N = M+1 phase values at averaging time τ.2. where the averaging is accomplished by simply choosing the appropriate interval. the Hadamard variance is usually expressed as its square-root. the overlapping Hadamard variance can be estimated from a set of N = M+1 time measurements as: Hσ 2 τ = y bg b 1 6 N − 3m τ 2 N −3m g ∑ i =1 x i + 3 m − 3x i + 2 m + 3 x i + m − x i . averaging time τ the overlapping Hadamard variance uses all 3-sample combinations [9]. Analytical methods are available for 26 . It can be estimated from a set of M frequency measurements for averaging time τ = mτ0 where m is the averaging factor and τ0 is the basic measurement interval. HDEV. Even though not all the additional overlapping differences are statistically independent. by the expression: 1 Hσ τ = 2 6 m M − 3m + 1 2 y bg M −3 m+1 b g ∑ j =1 R y S∑ T j + m−1 i= j i+2 m − 2 yi + m U +y V. Hσy(τ).HANDBOOK OF FREQUENCY STABILITY ANALYSIS For phase data. the Hadamard deviation. HDEV or Hσy(τ). Overlapping Hadamard Variance In the same way that the overlapping Allan variance The overlapping Hadamard variance makes maximum use of a data set by forming all provides better confidence than the possible fully overlapping 2-sample pairs at each non-overlapping version. In terms of phase data. W 2 i where yi is the ith of M fractional frequency values at each measurement time. the Hadamard variance is defined as: Hσ 2 τ = y bg 1 2 6τ N − 3 b g∑ i =1 N −3 xi+3 − 3xi+2 + 3xi+1 − xi . The expected value of the overlapping statistic is the same as the normal one described above. 5. they nevertheless increase the number of degrees of freedom and thus improve the confidence in the estimation.

MHVAR. which can extend from 1 to ⎣N/4⎦. which can be modeled as a power law function S ∝ f α. the various power law noises correspond to particular slopes µ. and df is the number of degrees of freedom (not necessarily an integer). and MHVARs using higher-order differences: 27 . The df is determined by the number of data points and the noise type. MAVAR. The Mod σ 2H(τ) log-log slope µ is related to the power law noise exponent by µ = –3 –α. In terms of phase data. Sample variances are distributed according to the expression: χ²(p. df) =(df · s²) / σ². This is an unbiased estimator of the modified Hadamard variance. a modified version of the Hadamard variance can be defined [15] that employs averaging of the phase data over the m adjacent samples that define the analysis τ = m⋅τ0. perhaps in cases where it was necessary to distinguish between short-term white and flicker PM noise in the presence of more divergent (α = -3 and –4) flicker walk and random run FM noises. Clock noise (and other noise processes) can be described in terms of power spectral density.or double-sided confidence intervals for an overlapping Hadamard variance estimate with a certain confidence factor. 5. df). where f is Fourier frequency and α is the power law exponent. where it was found to be slightly better for that purpose than the modified Allan variance. and m is the averaging factor. HVAR. MAVAR. MHVAR. and σ²max = (s2· df) /χ²(1-p. Given the df. and that same theory can be used to establish reasonable single.SECTION 5 TIME DOMAIN STABILITY calculating the number of degrees of freedom for an overlapping Allan variance estimation. σ² is the true variance. Modified Hadamard Variance By similarity to the modified Allan variance.2. MHVAR could also be useful for frequency stability analysis. the three-sample modified Hadamard variance is defined as N − 4 m+1 Modσ 2 H bτ g = ∑ j =1 R x − 3x S∑ T j + m−1 i= j i i+ m + 3x i + 2 m − x i + 3 m 6m2τ 2 N − 4m + 1 U V W 2 . based on Chi-squared statistics. MHVAR was developed in Reference [15] for determining the power law noise type of Internet traffic statistics. Expressions for the equivalent number of χ2 degrees of freedom (edf) required to set MHVAR confidence limits are available in [2]. when there were a sufficient number of data points. s² is the sample variance. df).10. When a variance such as MHVAR is plotted on log-log axes versus averaging time. The modified Hadamard variance concept can be generalized to subsume AVAR. where N is the number of phase data points xi at the sampling interval τ0. the confidence limits around the measured sample variance are given by σ²min = (s2 · df) /χ²(p. where χ² is the Chi-square value for probability p and degrees of freedom df.

The allowable power law exponent for convergence of the variance is equal to α > 1 – 2d. d = 2 corresponds to MAVAR. nonoverlapped AVAR and HVAR variances are given by setting m = 1.d τ = bg ∑ j =1 d !m2τ 2 N − ( d + 1)m + 1 R FG d IJ ( −1) x U V S ∑ ∑HkK W T j + m−1 d i= j k i + km k =0 2 ’ where d = phase differencing order. so the second difference Allan variances can be used for α > -3 and the third difference Hadamard variances for α > -5. 28 .HANDBOOK OF FREQUENCY STABILITY ANALYSIS N −( d +1) m+1 Modσ 2 H . d = 3 to MHVAR. higher-order differencing is not commonly used in the field of frequency stability analysis. Confidence intervals for the modified Hadamard variance can be determined by use of the edf values of Reference [16]. The unmodified.

June 2006.11. 4. Contrl. 8. Contrl. C. pp. Symp. Frequency Measurements and Control. The work on total variance began with the realization that the Allan variance can "collapse" at long averaging factors because of symmetry in the data. pp. but offers improved reflection at both ends. Proc. December 2003 5. 6. E. Total Variance The total variance. Proc. Riley. “Statistical Study of Phase Fluctuations and Oscillator Stability”. Annu. 3. January 1969. S. December 1996.T. 201-213. TOTVAR. pp. 222-225. 57. S. "Frequency Modulation Analysis with the Hadamard Variance". is a relatively new The total variance offers improved statistic for the analysis of frequency stability. "Uncertainty of Stability Variances Based on Finite Differences". 10. K.3. R. Boileau and B. London. Greenhall and W. 7-8. "Relating the Hamamard Variance to MCS Kalman Filter Clock Estimation". 9. 1. "Pulsar-Appropriate Clock Statistics". pp. E. 5/7/99.3. "A Multi-Variance Analysis in the Time Domain".A. "Extended Variances and Autoregressive Moving Average Algorithm for the Measurement and Synthesis of Oscillator Phase Noise".J. No.. Tome 30. J. 1994. No. Riley. J. T. June 1989. IEEE Trans. Meas. This expression for the overlapping Hadamard variance was developed by the author at the suggestion of G. 24th PTTI Meeting. It is confidence at large averaging factor similar to the two-sample or Allan variance. Proc. “Improved Estimation of the Hurst Parameter of Long-Range Dependent Traffic Using the Modified Hadamard Variance”. Ann. pp. Symp. 16. pp. Bregni and L. 5. Jacques Saloman Hadamard (1865-1963). pp. Andrews.A. on Freq. An early idea was to shift the data by 1/4 of the record length and average the two resulting Allan variances. Hutsell.T.T. 225-236. IM-25. Symp. Private communication. S. French mathematician. Rutman... December 1996. 413-424. June 1971. Baugh. 28th PTTI Meeting. 1975. Matsakis and F. The next step was to wrap the data in a circular fashion and calculate the average of all the Allan variances at 29 . Pratt. June 1977. 43rd Annu. 2. 12. March 1976.N. Proceedings of the IEEE ICC. C. Greenhall to W. Section 3. D.K. "Hadamard Transform Image Coding". Visr and J.331-335. Chapman & Hall.291-301.2.38-67. W. Walter. Chronos Group. "Operational Use of the Hamamard Variance in GPS". B. "Processus a Accroissements Stationnaires". des telecom. 14. Vol.. 13. Contrl. No. 15. Proc. Roberts. 66-75. 211-212. Dieter and S. July-Aug. IEEE. Proc. Kane and H.SECTION 5 TIME DOMAIN STABILITY References for Hadamard Variance 1. and by extending the data set by has the same expected value. Josties. 291-302. on Freq. Jmoda. 1. Instrum. Picinbono. December 1992. pp. 1977 IEEE International Freq. 27th PTTI Meeting. December 1995. "Oscillator Specifications: A Review of Classical and New Ideas". Wan. Hutsell.C. pp. 28th PTTI Meeting. Hutsell. pp. confidence at long averaging times [1-5]. Proceedings of the 35th Annual Precise Time and Time Interval (PTTI) Systems and Applications Meeting. 7. Picinbono.J. Proc. 11. ISBN 0-412-48270-3.

flicker FM and random walk FM noise by the expression b(T/τ)-c. need to be applied for flicker and random walk FM noise. The original data are in the center. The original data are in the center of x* with i =1 to N and x*=x. where T is the record length.500. and the extended data for j = 1 to M-1 are equal to y*1-j = yj and y*M+1 = yM+1-j. Totvar can also be defined for frequency data as Tot var(τ ) = M −1 2 1 ∑ y*i+ j+1 − y*i+ j . and c=0. το. A further improvement of the total variance concept was to extend the data by reflection.481 and 0. 2( M − 1) i =1 where the M = N-1 fractional frequency values. TOTDEV. where y*I = yi for i = 1 to M. exactly decomposes the classical standard variance [6]. This technique is very effective in improving the confidence at long averaging factors but requires end matching of the data. respectively. The number of equivalent χ² degrees of freedom for TOTVAR can be estimated for white FM. 1.168 and 0. where a=0. first at one end of the record and then at both ends. σtotal(τ). 2 2τ ( N − 2) i=2 Tot var(τ ) = where τ = mτο. The result is usually expressed as the square root.358. the edf for a total deviation estimate is the same as that for the overlapping ADEV with the number of χ² degrees of freedom increased by 2. The reflected portions added at each end extend from j = 1 to N-2 where x*1-j = 2x1-x1+j and x*N+j = 2xN-xN-j. This latest technique.HANDBOOK OF FREQUENCY STABILITY ANALYSIS every basic measurement interval. called TOTVAR. where b=1.222 and 0. When calculated by use of the doubly reflected method described above.927. y. the expected value of TOTVAR is the same as AVAR for white and flicker PM or white FM noise. For white and flicker PM noise. 30 . the total deviation.750. Bias corrections of the form 1/[1-a(τ/T)]. gives a very significant confidence advantage at long averaging times. and is an important new general statistical tool. respectively. and the N phase values x measured at τ = το are extended by reflection about both endpoints to form a virtual sequence x* from i = 3-N to i = 2N-2 of length 3N-4. 0. TOTVAR is defined for phase data as N −1 2 1 ∑ x *i − m − 2 x * i + x *i + m . measured at τ = το (N phase values) are extended by reflection at both ends to form a virtual array y*.

a linear trend (frequency offset) is removed from the subsequence by averaging the first and last halves of the subsequence and dividing by half the interval. First. Symp. D.A. 279-290. are shown in the diagram below: 31 . Lainson.A. Dec.J. D. Cont. pp. "Simulation Study Using a New Type of Sample Variance. June 1995. D.A. C. "Total Variance: a Progress Report on a New Frequency Stability Characterization. Howe and K. March 1997. D. 1997 PTTI Meeting. 1997. D. xi) with sampling period το that are to be analyzed at averaging time τ=mτ0. even reflection. D. Finally. D. Freq. an Estimator of LongTerm Frequency Stability". 3. MVAR. August 11. 91-99. "Effect of Drift on TOTALDEV.B. MTOT is computed from a set of N-3m+1 subsequences of 3m points." Proc." Proc.B. Then the offset-removed subsequence is extended at both ends by uninverted. "An Extension of the Allan Variance with Increased Confidence at Long Term. 6. Greenhall.A. 5. Lainson." Proc. Howe and K. It uses the same phase averaging technique as MVAR to distinguish between white and flicker PM noise processes. Greenhall. pp. Percival and D." Proc. 39-48. Symp. Howe.97-105. Cont.. 1997 PTTI Meeting. Peppler.12.A. 1997 European Frequency and Time Forum. pp. is another new The modified total variance statistic for the analysis of frequency stability. Modified Total Variance The modified total variance. 883-889. Howe and C.. Dec. 5. 1995 PTTI Meeting. D. and modified Allan and total variances. these steps are repeated for each of the N-3m+1 subsequences. 2001 Intl. Cont.A. These steps.2. "Total Variance as an Exact Analysis of the Sample Variance". 321-329. 2. 1997. Freq.A." Proc. Howe. A calculation of MTOT begins with an array of N phase data points (time deviates. has the same expected value. Freq. It is combines the features of the similar to the modified Allan variance. 7.available on the Hamilton Technical Services web site). 8. pp. “Definitions of Total Estimators of Common Time-Domain Variances”. Howe. Dec. Howe and D.J.SECTION 5 TIME DOMAIN STABILITY References for Total Variance 1. 1995. 1998. but offers improved confidence at long averaging times. pp. Howe and T. Next the modified Allan variance is computed for these 9m points. similar to those for MTOT. MTOT. 1996 Intl. "Methods of Improving the Estimation of Long-term Frequency Variance. Proc. June 2001.A. 4. "Total Variance. 1995 IEEE Int.A.. (to be published . Percival. but acting on fractional frequency data. Symp. calculating MTOT as their overall average. pp. June 1996. Proc.

even reflection to 9m points. τ0. the MTOT process requires three nested loops: • • • An outer summation over the N-3m+1 subsequences. 31st PTTI Meeting. the outer summation consists of only one term. but the inner summation has 6m terms. i = 1 to N 3m N-3m+1 Subsequences: xi i=n to n+3m-1 Linear Trend Removed: 0 xi xi = xi . Reference for Modified Total Variance D.c i ⋅ i. its linear trend is removed. 267-276. 32 . ci = freq offset 9m Extended Subsequence: Uninverted. and the prefix 0 denotes that the linear trend has been removed. At the largest possible averaging factor. An inner summation over the 6m unique groups of m-point averages from which all possible fully overlapping second differences are used to calculate MVAR.HANDBOOK OF FREQUENCY STABILITY ANALYSIS Phase Data x i . where _ _ _ zn(m) = x n(m) . Overall. averaging factor. m = N/3. A loop within the inner summation to sum the phase averages for three sets of m points. Vernotte. The 3m-point subsequence is formed. thus providing a sizable number of estimates for the variance.2x n+m(m) + x n+2m(m) 2 2 2 Calculate Mod σy (τ) for Subsequence: 2 Computationally. and number of points. "Generalization of the Total Variance Approach to the Modified Allan Variance". this can be expressed as: Mod Tot var τ = b g 2bmτ g b N − 3m + 1g 1 2 0 N − 3 m+1 ∑ n =1 R1 S 6m T n + 3 m−1 i = n −3m ∑ 0 zi m # bgU V W 2 where the 0zi#(m) terms are the phase averages from the triply-extended subsequence. and it is extended at both ends by uninverted. pp. The final step is to scale the result according to the sampling period. N. Dec. Proc. m. Even Reflection: 0 xn-l = xn+l-1 # 0 0 xi # 0 xn+3m+l-1 = # 0 xn+3m-l 1 ≤ l ≤ 3m 9 m-Point Averages: 6m 2nd Differences: mod σy (τ) = 1/2τ ⋅ 〈 z n (m) 〉 .A. Howe and F. 1999.

these steps are repeated for each of the N-3m+1 subsequences. HTOT. The Hadamard total variance. it rejects linear frequency drift while offering improved confidence at large averaging factors. handling more divergent noise types. is a similar measure of time stability. Time Total Variance The time total variance is a measure of time stability based on the modified total variance. First. is a total version of the Hadamard variance. Then the drift-removed subsequence is extended at both ends by uninverted. a linear trend (frequency drift) is removed from the subsequence by averaging the first and last halves of the subsequence and dividing by half the interval. Finally.14. An HTOT calculation begins with an array of N fractional frequency data points. TTOT.SECTION 5 TIME DOMAIN STABILITY 5. It is defined as σ²x(τ) = (τ²/3)·Mod σ²total(τ). These steps are shown in the diagram below: 33 . based on the modified total variance. Hadamard Total Variance The Hadamard total variance combines the features of the Hadamard and total variances by rejecting linear frequency drift.2. Next the Hadamard variance is computed for these 9m points. even reflection. and providing better confidence at large averaging factors. HTOT is computed from a set of N-3m+1 subsequences of 3m points. The time total variance. calculating HTOT as their overall average. yi with sampling period το that are to be analyzed at averaging time τ=mτ0.2. 5.13. As such.

An inner summation over the 6m unique groups of m-point averages from which all possible fully overlapping second differences are used to calculate HVAR. averaging factor. the outer summation consists of only one term. and it is extended at both ends by uninverted.2yn+m(m) + yn+2m(m) 2 2 Calculate Had σy (τ) for Subsequence: 2 Then Find HTOT as Average of Subestimates Computationally. ci = freq drift 9m Extended Subsequence: Uninverted. this can be expressed as: TotalHσ y ( m. 34 . Overall. i = 1 to N i 3m N-3m+1 Subsequences: yi i=n to n+3m-1 Linear Freq Drift Removed: 0 yi yi = yi . m = N/3. even reflection to 9m points.HANDBOOK OF FREQUENCY STABILITY ANALYSIS Fractional Frequency Data y. The final step is to scale the result according to the sampling period. At the largest possible averaging factor. m. The Hadamard total variance is a biased estimator of the Hadamard variance. its linear trend is removed. Even Reflection: 0 yn-l = yn+l-1 # 0 0 yi # 0 yn+3m+l-1 = # 0 yn+3m-l 1 ≤ l ≤ 3m 9 m-Point Averages: 6m 2nd Differences: Had σy (τ) = 1/6 ⋅ 〈 zn (m) 〉 . and number of points. N. the HTOT process requires three nested loops: • • • An outer summation over the N-3m+1 subsequences. A loop within the inner summation to sum the frequency averages for three sets of m points. thus providing a sizable number of estimates for the variance. The 3m-point subsequence is formed. drift-removed subsequences.ci ⋅ i. N ) = 2 N −3m+1 1 1 n +3m−1 ( ∑ H i ( m) ∑ 6( N − 3m + 1) n=1 6m i= n−3m b g) 2 . τ 0 . where _ _ _ zn(m) = yn(m) . so a bias correction is required that is dependent on the power law noise type and number of samples. but the inner summation has 6m terms. where the Hi(m) terms are the zn(m) Hadamard second differences from the triply extended. τ0.

SECTION 5

TIME DOMAIN STABILITY

The following plots shown the improvement in the consistency of the overlapping Hadamard deviation results compared with the normal Hadamard deviation, and the extended averaging factor range provided by the Hadamard total deviation [10].

Hadamard Deviation

Overlapping Hadamard Deviation

35

HANDBOOK OF FREQUENCY STABILITY ANALYSIS

Hadamard Total Deviation

Overlapping & Total Hadamard Deviations A comparison of the overlapping and total Hadamard deviations shows the tighter error bars of the latter, allowing an additional point to be shown at the longest averaging factor. The Hadamard variance may also be used to perform a frequency domain (spectral) analysis because it has a transfer function that is a close approximation to a narrow rectangle of spectral width 1/(2⋅N⋅τ0), where N is the number of samples, and τ0 is the measurement time [3]. This leads to a simple expression for the spectral density of the fractional frequency fluctuations Sy(f) ≈ 0.73 ⋅τ0 ⋅Hσ2y(τ) / N, where f = 1/ (2⋅τ0), which can be particularly useful at low Fourier frequencies. 36

SECTION 5

TIME DOMAIN STABILITY

The Picinbono variance is a similar three-sample statistic. It is identical to the Hadamard variance except for a factor of 2/3 [4]. Sigma-z is another statistic that is similar to the Hadamard variance that has been applied to the study of pulsars [5]. It is necessary to identify the dominant power law noise type as the first step in determining the estimated number of chi-squared degrees of freedom for the Hadamard statistics so their confidence limits can be properly set [6]. Because the Hadamard variances can handle the divergent flicker walk FM and random run FM power law noises, techniques for those noise types must be included. Noise identification is particularly important for applying the bias correction to the Hadamard total variance.

References for Hadamard Total Variance
1. 2. 3. 4. 5. 6. D.A. Howe, et al., "A Total Estimator of the Hadamard Function Used For GPS Operations", Proc. 32nd PTTI Meeting, pp. 255-267, Nov. 2000. D.A. Howe, R. Beard, C.A. Greenhall, F. Vernotte, and B. Riley, “Total Hadamard Variance: Application to Clock Steering by Kalman Filtering”, Proc. 2001 EFTF, pp. 423-427, Mar. 2001. Chronos Group, Frequency Measurement and Control, Section 3.3.3, Chapman & Hall, London, ISBN 0-412-48270-3, 1994. B. Picinbono, "Processus a Accroissements Stationnaires", Ann. des telecom, Tome 30, No. 7-8, pp. 211-212, July-Aug, 1975. D.N. Matsakis and F.J. Josties, "Pulsar-Appropriate Clock Statistics", Proc. 28th PTTI Meeting, pp. 225-236, December 1996. D. Howe, R. Beard, C. Greenhall, F. Vernotte, and W. Riley, "A Total Estimator of the Hadamard Function Used For GPS Operations", Proc. 32nd PTTI Meeting, Nov. 2000, pp. 255-268.

5.2.15.

Thêo1
Thêo1 is a 2-sample variance with improved confidence and extended averaging factor range.

The Thêo1 statistic is a two-sample variance similar to the Allan variance that provides improved confidence, and the ability to obtain a result for a maximum averaging time equal to 75 % of the record length. Thêo1 [1] is defined as follows:
Theo1( m, τ 0 , N ) =

N − m m/ 2−1 1 1 [( xi − xi−δ + m/ 2 ) + ( xi+ m − xi+δ + m/ 2 )]2 ∑∑ 2 ( N − m)( mτ 0 ) ⋅ 0.75 i=1 δ =0 ( m / 2 − δ )

,

where m = averaging factor, τ0 = measurement interval, and N = number of phase data points, for m even, and 10 ≤ m ≤ N - 1. It consists of N - m outer sums over the number of phase data points –1, and m/2 inner sums. Thêo1 is the rms of frequency differences averaged over an averaging time τ = 0.75 (m-1) τ0.

37

each with two phase differences. associated with a Thêo1 value is τ = 0.30 0.70 -1. as shown in the following table: 38 .74 c 0.00 0. τ.87 -1. Theo1 Schematic for n=11. m=10 i =1 to n-m = 1 .53 -1 1.HANDBOOK OF FREQUENCY STABILITY ANALYSIS A schematic for a Thêo1 calculation is shown in the figure below. T. Note that the effective tau for a Thêo1 estimation is τ = 0.85 0.40 Empirical formulae have been developed [1] for the number of equivalent χ2 degrees of freedom for the Thêo1 statistic. and it therefore requires the application of a bias correction. This example is for eleven phase samples (N = 11) at the largest possible averaging factor (m = 10). Thêo1 is a biased estimator of the Allan variance.09 0. δ = 0 to m/2 -1 = 4 x[ ] index = 1 0 1 δ 2 3 4 2 3 4 5 6 7 8 9 10 11 The single outer summation (i = 1 to 1) at the largest possible averaging factor consists of m/2 = 5 terms. Thêo1 has the same expected value as the Allan variance for white FM noise. A total of 10 terms contribute to the Theo1 statistic at this largest-possible averaging factor. where τ0 is the measurement interval. These terms are scaled by their spans m/2 δ = 5 thru 1 so that they all have equal weighting.79 0. Avar. Noise RW FM F FM W FM F PM W PM Thêo1 Bias Parameters Alpha a b -2 2.00 0. where m is the averaging factor and the constants a. where τ0 is the measurement interval.14 0. for all noise types except white FM noise.05 0 1.75·m·τ0. b and c are given in the table below.82 2 0. The averaging time. but provides many more samples that provide improved confidence and the ability to obtain a result for a maximum τ equal to ¾ of the record length. Reference [2] contains the preferred expression for determining the Thêo1 bias as a function of noise type and averaging factor: Thêo1 Bias = Avar/Thêo1 = a + b/mc .00 1 0.75·m·τ0.

86( N + 1)( N − 4r / 3)IJ FG r IJ H K H r + 114K N −r .374Nr + 12.3K H . N ). NewThêo1 was used in Reference [2] to form a composite AVAR/ NewThêo1 result called LONG. ThêoBR and ThêoH are expected value as the Allan variance.τ . 2 2 2 2 3 3 3/ 2 3/ 2 2 1/ 2 Thêo1 EDF Formulae EDF where r = 0. ThêoBR and ThêoH Because Thêo1 has the same bias-corrected NewThêo1. 5.9r K H ( 4.75m. FG 4.2.6r(4. τ 0 . Nr K H r + 2. Several forms of this composite statistic have evolved. to the New Thêo1 and ThêoH statistics described below. combine Thêo1 and AVAR results into a composite stability analysis. N n + 1 Thêo1( m = 12 + 4i. and because it versions of Thêo1 that provide bias covers a different (larger) range of averaging removal and combination with the factors. N ) Q n 0 0 i=0 0 denotes the floor function.4N − 2 IJ FG c4. τ .3K FG 0. it is useful to Allan variance. and with the condition τ0 ≤ T/10.5 OPFG r IJ N N r QH r + 5. . N ) = where n = MM N − 3PP and N 30 Q LM 1 ∑ Avar( m = 9 + 3i.SECTION 5 TIME DOMAIN STABILITY Noise RW FM F FM W FM F PM W PM .2 K F 4. NewThêo1.387r I FG r IJ GH br + 36. which has been superseded by ThêoH (see below). ThêoBR [3] is an improved bias-removed version of Thêo1 given by 39 . to matching the Thêo1 points on-average to the AVAR points at those averaging times where both are available. The NewThêo1 algorithm of Reference [2] provides a method of automatic bias correction for a Thêo1 estimation based on the average ratio of the Allan and Thêo1 variances over a range of averaging factors: NewThêo1( m. LM 41N + 0.4N − 1) + 114r hIJ H 2. .8 − 31N + 6.798N − 6.τ .4N − 1) − 8. N ) OPThêo1( m.4 N − 3) K FG 2N − 13Nr − 35r IJ FG r IJ .6g ( N − r ) JK H r + 0.16. from simply plotting the two. 10 to N-2 versus 1 to (N-1)/2.

but it can determine an unbiased estimate of the Allan variance over the widest possible range of averaging times without explicit knowledge of the noise type. U | |. τ . τ . N ) Q n 0 0 i=0 0 denotes the floor function. N n + 1 Thêo1( m = 12 + 4i. N ) for k | 0. N ). N ) = where n = MM N − 3PP and N6 Q LM 1 ∑ Avar( m = 9 + 3i. τ 0 . m even | | W It is the best statistic available for estimating the stability of a frequency source at large averaging factors.τ . ThêoH is a hybrid statistic that combines ThêoBR and AVAR: RAvar( m. An example of a ThêoH plot is shown in the figure below: 40 . ThêoBR is computationally intensive for large data sets.τ . N ) for 1 ≤ m < k | τ | ThêoH( m. V ≤ m ≤ N − 1.τ .τ .75τ T 0 0 0 0 0 where k is the largest available τ ≤ 20% T. N ) = S |ThêoBR( m.HANDBOOK OF FREQUENCY STABILITY ANALYSIS ThêoBR( m. N ) OPThêo1( m.

particular time interval. References for Thêo1. n. Howe and T.N.A. T. The data are a set of 1001 simulated phase values measured at 15-minute intervals taken over a period of about 10 days. D. ThêoBR and ThêoH 1. MTIE is the overall maximum of this time interval error over the entire data set: MTIE (τ ) = Max1≤ k ≤ N − n { Maxk ≤i≤ k + n ( xi ) − Mink ≤i≤ k + n ( xi )} where n = 1.. unpublished private communication. 3. Howe.A. β. This statistic is very commonly used in the telecommunications industry.A.SECTION 5 TIME DOMAIN STABILITY ThêoH is a composite of AVAR and bias-corrected ThêoBR analysis points at a number of averaging times sufficiently large to form a quasi-continuous curve.. It is calculated by moving an n-point (n = τ/τo) window through the phase (time error) data and finding the difference between the maximum and minimum values (range) at each window position.N. Tasset. while Thêo1 is able to extend the analysis out to nearly a week. MTIE is a measure of the peak time deviation of a clock and is therefore very sensitive to a single extreme value. it is necessary to express it in terms of a probability level. 2]. 2. D. Proc. Howe and T. Peppler. thus providing significantly more information from the same data set. High-Confidence Statistic that Improves on the Allan Deviation”. 233-238. pp. but has been the subject of recent theoretical work [1. “Thêo1: Characterization of Very Long-Term Frequency Stability..A.2. 4. D. " Very Long-Term Frequency Stability: Estimation Using a Special-Purpose Statistic". "ThêoH". S322-S331. Tasset. that a certain value is not exceeded.. 41 . The AVAR results are able to characterize the stability to an averaging time of about two days. Because of the peak nature of the MTIE statistic.17. transient or outlier. Metrologia 43 (2006). unpublished private communication. N-1 and N = number of phase data points. “ThêoH: A Hybrid. 5. The relationship between MTIE and Allan variance statistics is not completely defined. Proceedings of the 2003 IEEE International Frequency Control Symposium . Tasset and D. NewThêo1. Howe. July 2004. 5.K. 2004 EFTF. October 2003. T. The analysis requires less than five seconds on a 1 GHz Pentium processor. and can become very long for large data sets (although faster algorithms are available – see Reference 4 below). is a commonly used in the telemeasure of the maximum time error of a clock over a communications industry. MTIE. MTIE MTIE is a measure of clock error The maximum time interval error. May 2003.. "A Practical Thêo1 Algorithm". The time required for an MTIE calculation increases geometrically with the averaging factor.2.N.

β. 49-60. pp.2. University College. " Measurement of Maximum Time Interval Error for Telecommunications Clock Stability Characterization”. pp. % 95 90 80 kβ 1.. Oct. 7. 3. Meo. Meas. 5. Meas. 1997. P. pp. 6. 313-326. MTIE can be approximated by the relationship MTIE(τ. 5. G. 5. "Simulating the Wander Accumulation in a SDH Synchronisation Network". Dec. Dec. 1996. IM-46. 30th Annu. P. References for MTIE 1. . β) = kβ⋅√(h0⋅τ) = kβ⋅√2⋅σy(τ)⋅τ . “The Range Covered by a Random Process and the New Definition of MTIE”. IEEE Trans. “Synopsis of Timing Measurement Techniques Used in Telecommunucations”. TIE rms. Dec. IEEE Trans. 2000. and ho is the white FM power-law noise coefficient. MTIE is computationally intensive for large data sets [7]. Dec. 1996. Leschiutta. Dec. Dodone and S. Bregni. 1998. 4. Vol. S. "Fast Computation of Maximum Time Interval Error by Binary Decomposition". Instrum. Bregni and S. Instrum.HANDBOOK OF FREQUENCY STABILITY ANALYSIS For the case of white FM noise (important for passive atomic clocks such as the most common rubidium and cesium frequency standards). and is not as easily related to such clock noise processes as TDEV [1]. Vol. Master's Thesis. No.77 1. 1284-1294. PTTI Meeting.. Ivens.39 The maximum time interval error (MTIE) and rms time interval error (TIE rms) are clock stability measures commonly used in the telecom industry [3. 119-123. Bregni.59 1. 6. No. 1992. MTIE is determined by the extreme time deviations within a sliding window of span τ. 2. Zampetti. I&M. No. Travella and D. pp. β. as given in the table below.18. is another clock statistic commonly used by the telecommunications industry. 6. M. IEEE Trans.J. Proc. A. 28th Annu. pp. 5]. “The Range Covered by a Clock Error in the Case of White FM”. 49. " Clock Stability Characterization and Measurement in Telecommunications”. PTTI Meeting. Vol. 24th PTTI Meeting. 1240-1244. Proc. Travella. where kβ is a constant determined by the probability level. pp. UK. London. IM-45. TIE rms The rms time interval error. TIE rms is defined by the expression TIErms = 42 1 N −n ∑ xi + n − xi N − n i=1 b g 2 . 900-906. Proc. November 1997. Maccabruni.

The power law model for phase noise spectral density (see section 6.SECTION 5 TIME DOMAIN STABILITY where n = 1. Integrated Phase Jitter and Residual FM Integrated phase jitter and residual FM are other ways of expressing the net phase or frequency jitter by integrating it over a certain bandwidth. IEEE Trans. 6. Dec. K is amplitude in rad2. It is therefore similar in behavior to TDEV.. dB2). 46. although the latter properly identifies divergent noise types. pp. No. "Clock Stability Characterization and Measurement in Telecommunications"... Vol..2.2. Meas. dB1) and f2. These can be calculated from the amplitudes of the various power law terms. 5. where Sφ is the spectral density of the phase fluctuations in rad2/Hz.19. Given two points on the plot (f1.. 1997. and x is the power law exponent. Reference for TIE rms S. . K = 10H 10 FG dB 1 − x log f1 IJ K . Instrum. N-1 and N = # phase data points. The integrated phase jitter can then be found over this frequency interval by ∆φ 2 = ∆φ 2 = z f2 f1 Sφ ( f ) ⋅ df = z f2 f1 K ⋅ f ⋅ df K x +1 x +1 ( f 2 − f1 ) for x ≠ −1 x +1 2 ∆φ = K ⋅ (log f 2 − log f1 ) for x = −1 It is usually expressed as ∆φ in rms radians. It can be represented as a straight line segment on a plot of Sφ(f) in dB relative to 1 rad2/Hz versus log f in hertz. 43 . TIE rms is approximately equal to the standard deviation of the fractional frequency fluctuations multiplied by the averaging time.1) can be written as Sφ ( f ) = K ⋅ f x . the values of x and K may be determined by x= and dB1 − dB2 10 ⋅ (log f1 − log f 2 ) . f is the modulation frequency. 1284-1294. Bregni. For no frequency offset.

The value of Sφ(f) in dB can be found from the more commonly used £(f) measure of SSB phase noise to carrier power ratio in dBc/Hz by adding 3 dB. changes in noise level or type. An example of a DAVAR plot is shown below. The integrated frequency jitter or residual FM is therefore ∆f 2 = ∆f 2 = z f2 f1 Sν f ⋅ df = bg z f2 f1 K ⋅ f x + 2 ⋅ df K x +3 x +3 ( f 2 − f1 ) for x ≠ 3 x+3 2 ∆f = K ⋅ (log f 2 − log f1 ) for x = −3 It is usually expressed as ∆f in rms hertz. Rohde. It is able to detect variations in clock stability (noise bursts. J. U. Englewood Cliffs. Digital PLL Frequency Synthesizers. .2. Riley. and Sy(f) is the spectral density of the fractional frequency fluctuations (see section 6. References for Integrated Phase Noise and Residual FM 1. The ratio of total phase noise to signal power in the given integration bandwidth is equal to 10 log ∆φ2.20. 44 . "Integrate Phase Noise and Obtain Residual FM". August 1979. thereby showing changes (nonstationarity) in clock behavior versus time. The results of a dynamic stability analysis are presented as a 3D surface plot of log sigma versus log tau or averaging factor as a function of time or window number.) that would be difficult to see in an ordinary overall stability analysis. 78-80. This example is similar to the one of Figure 2 in Reference [2]. pp. pp. 5. Prentice-Hall. Microwaves. 2. showing a source with white PM noise that changes by a factor of 2 at the middle of the record. 1983.HANDBOOK OF FREQUENCY STABILITY ANALYSIS Similarly. the spectral density of the frequency fluctuations in Hz2/Hz is given by Sν f = ν 0 ⋅ S y ( f ) = f 2 ⋅ S ∆φ f = K ⋅ f x + 2 2 bg bg . The total integrated phase noise is obtained by summing the ∆φ2 contributions from the straight-line approximations for each power law noise type.L. etc. Dynamic Stability A dynamic stability analysis uses a sequence of sliding time windows to perform a dynamic Allan (DAVAR) or Hadamard (DHVAR) analysis.1). where ν0 is the carrier frequency in hertz. 411-418. W.

"Tracking Nonstationarities in Clock Noises Using the Dynamic Allan Variance". pp. For example.0x10-11. thereby finding an estimate of its expected value.3. 2005 Joint FCS/PTTI Meeting. 5. This section describes the use of χ² statistics for setting the confidence intervals and error bars of a stability analysis. It is always a good idea to include such a confidence limit in reporting a statistical 45 . one could state with 95 % confidence that the actual value does not exceed (say) 1. 2. 2003 Joint FCS/EFTF Meeting. The confidence limits of a variance estimate depend on the variance type. L. Confidence Intervals It is wise to include error bars (confidence intervals) on a stability plot to indicate the degree of statistical confidence in the numerical results. depending on the noise type and size of the data set. 239-244.SECTION 5 TIME DOMAIN STABILITY References for Dynamic Stability 1. the number of data points and averaging factor. Galleani and P. Proc. That determination should be accompanied by an indication of the confidence in its value as expressed by the upper and (possibly) lower limits of the statistic with a certain confidence factor. Proc. "The Characterization of Clock Behavior with the Dynamic AllanVariance". the statistical confidence factor desired. if the estimated value of the Allan deviation is 1. and the type of noise. Tavella. Tavella. L. Galleani and P.2x10-11. It is generally insufficient to simply calculate a stability statistic such as the Allan deviation.

or (equivalently) error bars on a plot.99 0. Simple Confidence Intervals The simplest confidence interval approximation. with no consideration of the noise type. to apply to the confidence interval: Noise Type Random Walk FM Flicker FM White FM Flicker PM White PM Kn 0. If you are unfamiliar with the basics of confidence limits. Even though those confidence limits or error bars are themselves inexact. Thêo1. χ² statistics apply to the classical standard variance. For frequency stability analysis. That noise type is then be used to determine a multiplicative factor.1.77 0. (4) use the inverse χ² distribution to find the normalized confidence limit(s). the emphasis is on various variances.75 0. (3) determine the number of equivalent χ² degrees of freedom (edf). (2) choose an appropriate confidence factor (e.3. Chi-Squared Confidence Intervals Chi-squared statistics can be applied to calculate single and double-sided confidence intervals at any desired confidence factor. they should be included to indicate the validity of the reported result. averaging factor. where N is the number of frequency data points used to calculate the Allan deviation. A more accurate determination of this confidence interval can be made by considering the noise type. where the noise type of the previous averaging factor is used). upper and lower numeric bounds. Hadamard. which can be estimated by the B1 bias function (the ratio of the standard variance to the Allan variance). Kn. using the number of MVAR degrees of freedom for the particular noise type.) used for frequency stability analysis. and (5) multiply those by the nominal deviation value to find the error bar(s). Most stability plots show ±1σ error bars for its overlapping Allan deviation plot.HANDBOOK OF FREQUENCY STABILITY ANALYSIS result. 5. noise type estimates are made at each averaging factor (except the last. Strictly speaking. and number of data points. 95 %).99 5.2. total. During the Run function. it is recommended that an introductory statistics book be consulted for an introduction to this subject.3.87 0. which can be shown as an upper numeric limit.g. Sample variances are distributed according to the expression 46 . A good introduction to confidence limits and error bars for the Allan variance may be found in Reference [1]. but they have been found applicable to all of the other variances (Allan. whose confidence limits (variances of variances) are treated with chi-squared (χ²) statistics. The basic idea is to (1) choose an single or double-sided confidence limits (upper or upper and lower bounds). These calculations are based on a determination of the number of degrees of freedom for the estimated noise type. etc. sets the ±1σ (68 %) error bars at ±σy(τ)/√N. The error bars for the modified Allan and time variances are also determined by Chi-squared statistics.

p. 47 . and thereby set the statistical limit(s) on the variance. 5. (2) determining the dominant power law noise type of the data. determine the edf from the variance type. the progression from the original two-sample (Allan) variance to the overlapping. In general. based on χ² statistics. Degrees of Freedom The equivalent number of χ² degrees of freedom (edf) associated with a statistical variance (or deviation) estimate depends on the variance type. and the Hadamard variance (HVAR) is found by a combined algorithm developed by C.A.4. σ² is the true variance. the number of data points.SECTION 5 TIME DOMAIN STABILITY χ2 = edf ⋅ s2 σ2 . σ 2 min = s2 ⋅ edf edf 2 2 . calculate the corresponding χ² value. Highly correlated data have a smaller edf than does the same number of points of uncorrelated (white) noise. TVAR and HVAR EDF The equivalent number of χ² degrees of freedom (edf) for the Allan variance (AVAR). An edf determination therefore involves (1) choosing the appropriate algorithm for the particular variance type. for many of its variance functions. total. AVAR. For double-sided limits. s² is the sample variance. and σ max = s ⋅ 2 χ ( p. The edf is determined by the number of analysis points and the noise type. the modified Allan variance (MVAR) and the related time variance (TVAR).or double-sided confidence intervals with a selectable confidence factor. MVAR. and Thêo1 variances has provided larger edfs and better confidence. where χ² is the Chi-square. and edf is the equivalent number of degrees of freedom (not necessarily an integer). edf ) χ (1 − p. Procedures exist for establishing single. edf ) 2 5. and (3) using the number of data points to calculate the corresponding edf. based on its generalized autocovariance function [2]. Greenhall.or double-limited confidence factor. The general procedure is to choose a single. The noise type matters because it determines the extent that the points are correlated. noise type and number of analysis points.4. and the type of noise involved.1.

m 100 Overlapping ADEV EDF for N=100 100 α= 2 1 EDF 0 10 Full EDF Algorithm vs m with α as Parameter -1 -2 1 1 10 Averaging Factor. m 100 48 .HANDBOOK OF FREQUENCY STABILITY ANALYSIS Overlapping ADEV EDF for W FM Noise 1000 N= 500 250 Full EDF Algorithm vs m with N as Parameter 100 EDF 100 50 25 10 10 1 5 1 10 Averaging Factor.

9 For m = 1 49 . The equivalent # of χ2 degrees of freedom (edf) for the fully overlapping Allan variance (AVAR) can be estimated by the following approximation formulae for each power law noise type: Power Law Noise Type W PM AVAR edf.SECTION 5 TIME DOMAIN STABILITY Overlapping EDF for N=100 and W FM Noise 100 Variance Type: EDF 10 Allan Hadamard Modified Full EDF Algarithm vs m with Variance Type as Parameter 1 1 10 Averaging Factor. m 100 This method for estimating the edf for the Allan. where N = # phase data points. modified Allan and Hadamard variances supersedes the following somewhat simpler empirical approximations (which may still be used).3N − 4. m = averaging factor = τ/τ0 F PM W FM F FM b N + 1gb N − 2mg 2b N − mg L F b N − 1gIJ lnFG b2m + 1gb N − 1gIJ OP exp Mln G N H 2m K H 4 K Q LM 3b N − 1g − 2b N − 2g OP 4m N Q 4m + 5 N 2m 2b N − 2g 2 2 2 1/ 2 2.

22 0.36 5.HANDBOOK OF FREQUENCY STABILITY ANALYSIS RW FM LM b N − 2g OPLM b N − 1g − 3mb N − 1g + 4m OP b N − 3g MN m PQMN PQ 2 2 2 5N 2 4m N + 3m b g For m >1 The edf for the modified Allan variance (MVAR) can be estimated by the same expression as the overlapping Hadamard variance (see below) with the arguments changed as follows (valid for -2 ≤ α ≤ 2): MVAR and TVAR edf for N.17 0. Greenhall based on its generalized autocovariance function. as shown in the following table: Power Law Noise Type White FM Flicker FM Random Walk FM TOTVAR edf Coefficients b c 1.3. τ is the averaging time. where 1/edf = (1/p)(a0-a1/p). with the coefficients as follows: Power Law Noise Type W FM F FM RW FM FW FM RR FM HVAR edf coefficients a0 a1 7/9 1/2 1. The HVAR edf is found either as a summation (for small m cases with a small number of terms) or from a limiting form for large m.93 0.00 0. The edf for the fully overlapping Hadamard variance (HVAR) can be found by an earlier algorithm also developed by C.2.53 1. and b and c are coefficients that depend on the noise type as shown in the following table: 50 .50 0 1.62 31/30 17/28 1.A. τ is the averaging time.54 5. m and α-2. MTOT EDF The edf for the modified total variance (MTOT) is given by the same formula b(T/τ) . where T is the length of the data record. where T is the length of the data record.06 0.30 0.c. TOTVAR and TTOT EDF The edf for the total variance (TOTVAR) and the related total time variance (TTOT) is given by the formula b(T/τ) .4.4. and b & c are coefficients that depend on the noise type. m and α = MVAR edf for N+1.c.

10 1.31 5.4.9τ ⎠ ⎝ ⎠ Flicker FM Random Walk FM 51 .85 0. Thêo1 EDF The equivalent number of χ² degrees of freedom (edf) for the Thêo1 variance is determined by the following approximation formulae for each power low noise type.1N x + 0.4τ ⎞ edf = ⎜ ⎟ ⎟⎜ (4.5 ⎤ ⎛ τ 3/ 2 ⎞ edf = ⎢ − ⎟ ⎥ ⎜ 3/ 2 τ Nx ⎣ ⎦ ⎝ τ + 5. m = averaging factor = τ/τ0 Flicker PM White FM ⎛ 0.86 ( N x + 1) ( N x − 4 ⋅τ ) ⎞ ⎛ τ ⎞ 3 edf = ⎜ ⎟⎜ ⎟ Nx −τ ⎝ ⎠ ⎝ τ + 1. τ=0.4 N x − 3) 2 ⎝ 2.20 0. Power Law NoiseType White PM Thêo1 edf.374 N xτ + 12.40 1. where N = # phase data points.10 1.3 ⎠ 2 2 ⎛ 4.798 N x2 − 6.4 N x − 1) − 8.4.2 ⎠ ⎛ 2 N x2 − 1.387τ ⎞ ⎛ τ ⎞ edf = ⎜ ⎟⎜ ⎟ (τ + 36.SECTION 5 TIME DOMAIN STABILITY Power Law Noise Type White PM Flicker PM White FM Flicker FM Random Walk FM MTOT edf Coefficients b c 1.6τ (4.3 ⎠ ⎡ 4.20 1.1N x + 6.4 N x − 2 ⎞ ⎛ (4.90 2.6)1/ 2 ( N x − τ ) ⎝ ⎠ ⎝ τ + 0.8 3.50 0.5τ ⎞ ⎛ τ 3 ⎞ edf = ⎜ ⎟⎜ 3 ⎟ N xτ ⎝ ⎠ ⎝ τ + 2.3 N xτ − 3.14 ⎠ ⎛ 4.75m.75 0.4 N x − 1) + 11.

A. Peppler.A. Allan and J.A. 1044-1049. Proc. 32nd PTTI Meeting. D. “A Total Estimator of the Hadamard Function Used for GPS Operations”. FWFM or RRFM) of the spectral density of the fractional frequency fluctuations. et. private communication. 12. May 2003. pp. "Properties of Signal Sources and Measurement Methods''.A. May 1995.. Noise Identification Identification of the dominant power law noise type is often necessary for setting confidence intervals and making bias corrections during a frequency stability analysis. Instrum. For example. pp. Howe. Greenhall. pp. March 2000. D. Riley. Symp. IEEE Std 1139-1999. Proc. November 2000. Proc. D. C. 9. 2003 PTTI Meeting. May 1999. “Estimation of Very Long-Term Frequency Stability Using a Special-Purpose Statistic”. and Time Forum.A.A. July 1999. D.W. 2. Symp. 1999 Joint Meeting of the European Freq. 1093-1099. 6.A. 3. J. C. to perform a frequency stability analysis. 11. 5. "IEEE Standard Definitions of Physical Quantities for Fundamental Frequency and Time Metrology . Proc.. Howe. Howe. Proc. Barnes. 6. private communication.K.A. 7. 35th Annu. and it is essential to know the dominant noise type to correct for bias in the newer Total and Thêo1 52 . Vol. 1997 PTTI Meeting. Barnes and D. Contrl. pp. 39-48. K. Contrl. “Degrees of Freedom of the Estimate of the Two-Sample Variance in the Continuous Sampling Method. pp.A. Meas. 1989.5. D. Howe. Yoshimura. Allan. 5. NIST Technical Note 1318. The most effective means for power noise identification are based on the B1 and R(n) functions and the lag 1 autocorrelation. 10. Sy(f) = hαf α (α = 2 to –4). December 1997. 255-268. Symp. Available on line at NIST web site. Greenhall. D. Power Law Noise Identification It is often necessary to identify the dominant power law noise process (WPM. 15. May 1981.W.. December 1991. 40. 8. 5.. D. Greenhall and W. Al. D.HANDBOOK OF FREQUENCY STABILITY ANALYSIS References for Confidence Intervals 1.. “Uncertainty of Stability Variances Based on Finite Differences”.Random Instabilities".1. April 1999. pp. 1-47. 91-99. 14.5. Proc. No. pp. 1990. "Recipes for Degrees of Freedom of Frequency Stability Estimators". 11th European Freq. Greenhall. Variances Based on Data with Dead Time Between the Measurements". March 2000. Howe. Greenhall. private communication. 346-353. Contrl. IEEE 1995 Freq. FPM. Proc. Proc. Howe.A. ". 4.A. "Methods of Improving the Estimation of Long-Term Frequency Variance". RWFM. Symp. "Total Variance Explained".A.A. "Estimating the Modified Allan Variance". Howe and T. and Time Forum and the IEEE Freq. C. Contrl. IEEE Trans.. D. "Total Variance: A Progress Report on a New Frequency Stability Characterization". WFM. 994-999.A.” IEEE Transactions IM-38. 2003 Joint Meeting of the European Freq. on Freq. March 1997. pp. C. and Time Forum and the IEEE International Freq. Howe & C. 13. FFM. knowledge of the noise type is necessary to determine the equivalent number of chi-squared degrees of freedom (edf) for setting confidence intervals and error bars. December 2003.

and the R(n) function [3]. This method cannot distinguish between W and F PM at unity averaging factor. 5. which is the ratio of the N-sample (standard) variance to the two-sample (Allan) variance. The B1 dependence on µ is used to determine the power law noise type for –2 ≤ µ ≤ 2 (W and F PM to FW FM).SECTION 5 TIME DOMAIN STABILITY variances.5. the autocorrelation at lag k is defined as 53 . The B1 function has as arguments the number of frequency data points. either manually or by fitting a line to it. for which RR FM noise can apply. r (which is set to 1). More specifically.3. The most common method for power law noise identification is simply to observe the slope of a log-log plot of the Allan or modified Allan deviation versus averaging time. . During a stability calculation. and that works for a single τ point over the full range of –4 ≤ α ≤ 2 is based on the Barnes B1 function. For a B1 corresponding to µ = -2. Calculate B1(N. r=1. the B1 ratio can be applied to frequency (rather than phase) data. the dead time ratio. particularly if bias corrections and/or error bars must be applied. For the Hadamard variance. Noise Identification Using B1 and R(n) A noise identification algorithm that has been found effective in actual practice. Calculate the standard and Allan variances for the applicable τ averaging factor. 5. µ 2( N − 1)(1 − 2 ) Determine the expected B1 ratios for a = -3 through 1 or 2. While the noise type may be known a priori or estimated manually. µ. the α = 1 or 2 (F PM or W PM noise) ambiguity can be resolved with the R(n) ratio using the modified Allan variance. 3. Resolve an α = -3 or –4 ambiguity by applying B1 to frequency data. This obviously requires at least two stability points. and adding 2 to the resulting µ. it is desirable to have an analytic method for power law noise identification that can be used automatically as part of a stability analysis algorithm. (m =3. Set boundaries between them and find the best power law noise match. Resolve an α = 1 or 2 ambiguity with the modified Allan variance and R(n). a = -4). thereby showing the degree by which its value at one time is similar to its value at a certain later time. The Autocorrelation Function The autocorrelation function (ACF) is a fundamental way to describe a time series by multiplying it by a delayed version of itself. 4. N.5. it is desirable (or necessary) to automatically identify the power law noise type at each point. 6. The overall noise B1/R(n) noise identification process is therefore: 1.2. 5. and the power law t-domain exponent. N (1 − N ) µ The boundaries between the noise types are generally set as the geometric means of their expected values. There is little literature on the subject of power-law noise identification. which is the ratio of the modified Allan to the normal Allan variances. µ) = 2.

the process is stationary and has a lag 1 autocorrelation equal to ρ1 = δ / (1-δ) [6].5. The results can be rounded to determine the dominant noise type or used directly to estimate the noise mixture.5. σz2 is its variance. and E denotes the expected value. If z is a τ-average of frequency data y(t). respectively.4. For frequency data. and white FM noise has ρ1 = 0. For those more divergent noises. µ is its mean value. and the noise type can therefore be estimated from δ = r1 / (1+r1). –1/3 and zero. if z is a τsample of phase data x(t). corrected for the differencing.25. t ∑(z t =1 N − z) 2 where z is the mean value of the time series and N is the number of data points [4]. The properties of this power law noise identification method are summarized in the table below. 5. for which the expected values are –1/2.HANDBOOK OF FREQUENCY STABILITY ANALYSIS ρk = E[( zt − µ )( zt + k − µ )] σ z2 where zt is the time series. Noise Identification Using r1 An effective method for identifying power law noises using the lag 1 autocorrelation [5] is based on the properties of discrete-time fractionally integrated noises having spectral densities of the form (2 sin π f )-2δ. white PM noise has ρ1 = -1/2.25. first differences of the data are taken until a stationary process is obtained as determined by the criterion δ < 0. 54 . where α is the usual power law exponent f α. then α = p + 2.5. then α = p. including difficult cases with mixed noises. FPM and WFM are then used. where round (2δ) is 2δ rounded to the nearest integer and d is the number of times that the data is differenced to bring δ down to < 0. The more divergent noises have positive r1 values that depend on the number of samples. and tend to be larger (approaching 1). and the same criteria as for WPM. 5. The autocorrelation is usually estimated by the expression rk = 1 N N −k t =1 ∑(z 1 N t − z )( zt + k − z ) . For frequency data. the data are differenced until they become stationary. The noise identification method therefore uses p = -round (2δ) –2d. the lag 1 autocorrelation is able to easily identify white and flicker PM noise. The Lag 1 Autocorrelation The lag 1 autocorrelation is simply the value of r1 as given by the expression above. For the more divergent noises. flicker PM noise has ρ1 = -1/3. For δ < ½. and white (uncorrelated) FM noise. thereby determining the noise type at that averaging time. It has excellent discrimination for all common power law noises for both phase and frequency data.

7 ≈1 ≈1 -1/2 -1/3 0 ≈0.SECTION 5 TIME DOMAIN STABILITY Lag 1 Autocorrelation for Various Power Law Noises Noise Type W PM α 2 Phase Data* x(t) d=0 ACF of Phase Data F PM 1 W FM 0 F FM -1 RW FM -2 * The differencing operation changes the appearance of the phase data to that shown 2 rows higher.7 -2/3 -3/5 -1/2 -1/3 -2/3 -3/5 -1/2 -1/3 -3/4 -5/7 -2/3 -3/5 † 0 0 -1/2 -2 ≈1 ≈1 ≈1 Shaded values are those used for noise ID for the particular noise and data type. r1† Type d=0 d=1 d=2 α x(t) y(t) x(t) y(t) x(t) y(t) 2 1 0 -1 0 ≈0.7 -1/2 -1/3 0 ≈0. Lag 1 Autocorrelation for Various Power Law Noises and Differences Noise Lag 1 Autocorrelation. 55 .

.25 Or d >= dmax) p = −2(δ + d ) Done = True Else z1 = z2 − z1 . τ. The dmax parameter should be set to 2 or 3 for an Allan or Hadamard (2 or 3-sample) variance analysis. The output is p. Noise ID Algorithm The basic lag 1 autocorrelation power law noise identification algorithm is quite simple.HANDBOOK OF FREQUENCY STABILITY ANALYSIS 5. respectively. and the maximum order of differencing dmax. Done = False.5. zN of phase or frequency data. respectively. and it may therefore be necessary to decimate the phase data or average the frequency by the appropriate averaging factor before applying the noise identification algorithm. The algorithm is fast. requiring only the calculation of one autocorrelation value and first differences for several times. consistently identifying pure power noise for α = 2 to –4 for sample sizes of about 30 or more. z N −1 = z N − z N −1 N = N −1 d = d +1 End If End While Note: May round p to nearest integer The input data should be for the particular averaging time. d = 0 While Not Done 1 z= N r1 = i =1 ∑z i =1 i N N i ∑(z r1 1 + r1 N −1 − z )( zi+1 − z ) i ∑(z i =1 − z )2 δ= If d >= dmin And (δ < 0. the fractional result provides an indication of their ratio. an estimate of the α of the dominant power law noise type. the minimum order of differencing dmin (default = 0). For a mixture of adjacent noises. and generally identifying the dominant type of mixed noises when it is at least 10 % larger than the others. 56 . and may be rounded to an integer (although the fractional part is useful for estimated mixed noises)... The alpha result is equal to p+2 or p for phase or frequency data..…. The lag 1 autocorrelation method yields good results. It is independent of any particular variance. The inputs are a vector z1. of interest. It can handle all averaging factors.6. and (optionally) the value of d.

May 1992. A-6. F. 1992 IEEE Frequency Control Symposium. Seconds Frequency Stability and Noise Analysis of Two Hydrogen Masers References for Noise Identification 1.A.L. Ferrelectrics and Freq. Greenhall. Proc. 52. P. Time Series Analysis. D. Time Series: Theory and Methods.A. discontinuities. The algorithm tends to produce jumps in the estimated alpha for mixed noises when the differencing factor. IEEE Trans. σy(τ) 2 10 Linear Frequency Drift Removed Noise Type Noise Type WPM FPM WFM FFM RWFM 1 10-12 0 10-13 -1 10-14 -2 10-15 Stability -3 10-16 0 10 Lag 1 ACF Noise ID Extends to # Analysis Points ≥ 10 FWFM 10 1 10 2 10 3 10 4 10 5 Averaging Time. Ultrasonics. “Power Law Noise Identification Using the Lag 1 Autocorrelation”. for Variances Based on Finite Samples of Processes with Power Law Spectral Densities”. 4. d. Ultrasonics. Holden-Day. Vol. Contrl.9). The lag 1 autocorrelation method for power law noise identification is a fast and effective way to support the setting of confidence intervals and to apply bias corrections during a frequency stability analysis.J. 2007 2. Ferroelectrics and Freq. 5. Kasdin and T. where N is the number of data points.A. at the expense of higher variability when a lower d would have been sufficient. Aug.2. “Tables of Bias Functions. (13. pp. This can be avoided by using the same d for the entire range of averaging times. pp. Vernotte. Howe. B1 and B2. R. 1969. Brockwell and R. Characterization of Clocks and Oscillators. “Discrete Simulation of Power Law Noise”. Greenhall. changes (although the alpha value when rounded to an integer is still consistent). N. Howe and F. 1990. 8. NIST Technical Note 1337. 8. “ThêoH and Allan Deviation as Power-Law Noise Estimators”. Contrl. Beard. 274-283. Jan. Riley and T. Proc. Box and G.. California. W. Springer-Verlag. Howe. Barnes.W.J. J. 2005. 1976. J. Riley and C. D. Oakland. 2nd Edition. Peppler. New York. Sec. G.B Sullivan. Forecasting and Control. Davis. 3. 7. Eq. ISBN 0-8162-1104-3. 57 . Acceptable results can be obtained from the lag 1 autocorrelation noise identification method for N ≥ 32.A. Walter. 1991. IEEE Trans.J.SECTION 5 TIME DOMAIN STABILITY Before analysis. 1253-1261. "Enhancements to GPS Operations and Clock Evaluations Using a "Total" Hadamard Deviation". D..A.K. Jenkins. and deterministic components. C.L. D. McGee and D. Allan. as shown in the example below: α SAO VLG11B H-Maser S/N SAO26 vs SAO18 -11 Overlapping Allan Deviation. April 2004. 18th European Frequency and Time Forum. 6. No. W. Chapter 2. Feb. Walls (editors).A. the data should be preprocessed to remove outliers. NBS Technical Note 375.τ.

The product of the B2 and B3 bias functions is used for dead time correction. τ) / σ²(2. 5.2. τ) / σ²(N.2).7. as discussed in section 5. τ = averaging time. r . and R(n). T. 58 . τ) . and µ = exponent of τ in Allan variance for a certain power law noise process: B1(N. T. the ratio of the modified Allan variance to the normal Allan variance. R(n) Bias Function The R(n) function is the ratio of the modified Allan variance to the normal Allan variance for n = number of phase data points.2.6.7. 5.2). the exponent of the power law noise type: R(n) = Mod σ²y(τ) / σ²y(τ) The R(n) bias function is useful for performing power law noise identification by comparing its actual value to those expected for the various noise types (see section 5. 5.HANDBOOK OF FREQUENCY STABILITY ANALYSIS 5. Note: R(n) is also a function of α. B2 Bias Function The B2 bias function is the ratio of the 2-sample (Allan) variance with dead time ratio r = T/τ to the 2-sample (Allan) variance without dead time (r = 1): B2(r . τ) .2). µ) = σ²(N. 5.2. M. In particular. µ) = σ²(2.9. M. τ) / σ²(2. the ratio of the standard variance to the Allan variance. r . and the B2 and B3 bias functions are used to correct for dead time in a frequency stability measurement. T. µ) = σ²(N. where T = time between measurements. Bias Functions Several bias functions are defined and used in the analysis of frequency stability. B3 Bias Function The B3 bias function is the ratio of the N-sample (standard) variance with dead time ratio r = T/τ at multiples M = τ/τ0 of the basic averaging time τ0 to the N-sample variance with the same dead time ratio at averaging time τ: B3(N. The B1 bias function is useful for performing power law noise identification by comparing its actual value to those expected for the various noise types (see section 5. T.10. T. B1. τ) .8. as defined below. τ. B1 Bias Function The B1 bias function is the ratio of the N-sample (standard) variance to the 2-sample (Allan) variance with dead time ratio r = T/τ. are used for the identification of power law noise types (see section 5.

TOTVAR is biased low by about 24 % for RW FM noise. a = 3/4 = 0. and for white FM noise.74 0. Note that the effective tau for a Thêo1 estimation is t = 0.27 1.31 This bias factor should be used to correct all reported MTOT results. For flicker and random walk FM noise.30 0. 5.85 0. where a = 1/3⋅ln2 = 0.87 c -1.09 0. This bias function should be used to correct all reported TOTVAR results.82 0.79 W FM F PM W PM 0 1 2 1.30 1. b and c are given in the table below. The ratio of the expected value of TOTVAR to AVAR is given by the expression B(TOTAL) = 1 . At the maximum allowable value of τ = Τ/2.53 -1. and T is the record length.75·m·t0.00 0.17 1.05 0. as shown in the following table: Noise W PM F PM W FM F FM RW FM Bias Factor 1.00 0. where t0 is the measurement interval.750 for random walk FM noise.12. 5.40 59 . MTOT Bias Function The MTOT statistic is a biased estimator of the modified Allan variance.06 1. Thêo1 Bias The Thêo1 statistic is a biased estimator of the Allan variance.13.481 for flicker FM noise.SECTION 5 TIME DOMAIN STABILITY 5.a⋅(τ/Τ).00 0. where m is the averaging factor and the constants a. Noise Alpha RW FM F FM a -2 -1 b 2. 0 < τ ≤ Τ/2 . The Thêo1 bias factor (the ratio of the expected value of Thêo1 to AVAR) depends on both noise type and averaging factor: Thêo1 Bias = AVAR/Thêo1 = a + b/mc .11. TVAR Bias Function The TOTVAR statistic is an unbiased estimator of the Allan variance for white and flicker PM noise.70 1. The MTOT bias factor (the ratio of the expected value of Mod Totvar to MVAR) depends on the noise type but is essentially independent of the averaging factor and number of of data points.14 0. TOTVAR is biased low as τ becomes significant compared with the record length.

It can have a significant effect on the results of a stability analysis. Howe. May 1999. C. May 2000. F. 39-48. 11. 8. 3. 994-999.. or because of a deliberate wait between measurements. R. Riley. Greenhall. C.A.A. Contrl. "Estimating the Modified Allan Variance". Vernotte and W. November 2000. "IEEE Standard Definitions of Physical Quantities for Fundamental Frequency and Time Metrology .A. private communication. Howe. 9. Howe. private communication. 1990.g. 346-353. Meas. Contrl. 4. Symp. D.A.W.A. Proc.Random Instabilities".A.A. Proceedings of the 32nd Annual Precise Time and Time Interval (PTTI) Systems and Applications Meeting. July 1999. pp. 5. 7. "Total Variance Explained". 6. Howe & C. pp. IEEE Trans. Greenhall. Greenhall. pp. NIST Technical Note 1318. Dead time can occur in frequency measurements and can significantly affect a subsequent stability analysis. pp. 11th European Freq. private communication.A. 91-99. No. C. pp. and Time Forum and the IEEE Freq. "Methods of Improving the Estimation of Long-Term Frequency Variance". "Recipes for Degrees of Freedom of Frequency Stability Estimators". December 1991. 2. IEEE Std 1139-1999. 10. Barnes and D. 1999 Joint Meeting of the European Freq. 40. D. Proc. pp.HANDBOOK OF FREQUENCY STABILITY ANALYSIS References for Bias Functions 1. Methods are available to correct for dead time and thus obtain unbiased results.A. 255-268.14. Symp. J. Greenhall. and Time Forum. C. C.. once per hour). D. March 1997.A Greenhall. frequency data taken for 100 seconds. "A Total Estimator of the Hadamard Function Used for GPS Operations". Proc. D. March 2000. D. Instrum. 1997 PTTI Meeting. Greenhall.. 5. Allan. Vol. April 1999. 1093-1099. Proc. IEEE 1995 Freq.. "Variances Based on Data with Dead Time Between the Measurements". December 1997. 60 . "Total Variance: A Progress Report on a New Frequency Stability Characterization". May 1995. Howe. 6. especially for the case of large dead time (e. Beard. Dead Time Dead time can occur in frequency measurements because of instrumentation delay between successive measurements.

one can determine only the average frequency and its drift. It is recommended that these issues be avoided by making measurements with zero dead time. Those bias corrections are made by use of the product of B2 and B3. Restricting the dead time corrections to Allan deviations is a conservative approach based on the B2 and B3 definitions. Frequency Stability Plots for Common Power Law Noises with Large Measurement Dead Time (r = T/τ = 36) Simulated Data Sampled for τ = 100 Seconds Once Per Hour for 10 Days Nominal 1x10-11 Stability at τ = 100 Seconds Shown by Lines Plots Show Stability of Simulated Data Sets for Continuous. Dead time correction is problematic for data having multiple noise types. the ratio of the variance with distributed dead time to the variance with all the dead time at the end. Requiring manual noise selection avoids the problem of noise identification for biased data having the wrong sigma-tau slope. lowers the maximum allowable averaging factor. Dead time that occurs at the end of a measurement can be adjusted for in an Allan deviation determination by using the Barnes B2 bias function [1]. Otherwise. measurement dead time reduces the confidence in the results. Those bias functions depend critically on the power law noise type. These corrections are particularly important for non-white FM noise with a large dead time ratio. Sampled and Dead Time-Corrected Data White PM (µ = -2) √B2 = 0. without this correction. The power law noise type must be known in order to calculate these bias functions. as shown in the figures below. the ratio of the two-sample variance with dead time ratio r = T/τ to the two-sample variance without dead time. and prevents proper conversion of frequency to phase. Moreover. no information is available about the behavior of the device under test during the dead time. Simulated periodically sampled frequency data with distributed dead time for various power law noise processes shows good agreement with the B2 and B3 bias function corrections. it is necessary to also use the B3 bias function [2]. When such data are used to form frequency averages at longer tau.SECTION 5 TIME DOMAIN STABILITY Dead time corrections can be applied by dividing the calculated Allan deviation by the square root of the product of the Barnes B2 and B3 bias ratios.82 at AF = 1 61 . In addition to introducing bias.

82 at AF = 1 White FM (µ = -1) √B2 = 1.HANDBOOK OF FREQUENCY STABILITY ANALYSIS Flicker PM (µ = -2) √B2 = 0.00 at AF = 1 62 .

31 at AF = 1 These simulations show that the B2 and B3 bias corrections are able to support reasonably accurate results for sampled frequency stability data having a large dead time. Moreover.SECTION 5 TIME DOMAIN STABILITY Flicker FM (µ = 0) √B2 = 1. when the power law noise type is known. and limits the allowable averaging factor. The relatively small number of data points reduces the confidence of the results. The slope of the raw sampled stability plot does not readily identify the noise type. and mixed noise types would make correction difficult. Sparsely sampled data are therefore not recommended for the purpose of stability analysis. the infrequent measurements provide no information about the behavior of the clock during the dead time.92 at AF = 1 RW FM (µ = 1) √B2 = 7. 63 . and prevent a proper conversion of frequency to phase. however.

NIST Technical Note 1318.15. the data may be analyzed by use of the average timetag spacing as the analysis tau. where the data spacing is either one or two days.77x10-12 sec at τ=1 day (TDEV = MDEV 64 .A. An example of unevenly spaced data is two-way satellite time & frequency transfer (TWSTFT) measurements made on Monday.A. Unevenly Spaced Data Unevenly spaced phase data can be handled if they have associated timetags by using the individual timetag spacing when converting them to frequency data. in effect placing the frequency data on an average uniform grid. The TWSTFT data are simulated as 256 points of white PM noise with an Allan deviation (ADEV) level of σy(τ) = 1x10-11 at 1-day. 5. January 1969.W. tau variations of ± 10% will yield reasonable results as long as the exact interval is used for phase to frequency conversion. Allan. B1 and B2. Wednesday. The corresponding TDEV is 5. 2. for Variances Based on Finite Samples of Processes with Power Law Spectral Densities". Barnes. "Tables of Bias Functions.HANDBOOK OF FREQUENCY STABILITY ANALYSIS References for Dead Time 1. if the tau differences are reasonably small. "Variances Based on Data with Dead Time Between the Measurements". J. and Friday of each week. A composite plot of the TWSTFT TDEV results is shown above. Then. J. NBS Technical Note 375. While completely random data spacing is not amenable to this process. Barnes and D. 1990.

One can expect a normal (Gaussian) distribution for a reasonably sized data set. “Noise Analysis of Unevenly Spaced Time Series Data”. 2. pp.33 ≈ 1. The green line shows that the –0. C. If the missing points are replaced by linearly interpolated phase values. the longer-tau TDEV estimates are lowered to the correct level.SECTION 5 TIME DOMAIN STABILITY divided by √3).5 slope of the TDEV plot for W PM noise. but cannot apply standard techniques to analyze the complete data set. as shown in curve C. References for Unevenly Spaced Data 1. For a normal distribution. The adjusted method of using frequency data converted from phase data by using individual tau values adjusted for the timetag spacing is recommended because it does not use interpolation. and uses the best frequency estimates. and a different (e.5. the average tau. it is not possible to estimate the noise at shorter times. bimodal) distribution can be a sign of a problem. In that case. The TWSTFT data are sampled once on Monday. and can provide insight regarding them. because the true sampling interval is τavg. and their TDEV is shown in curve B. 33. Hackman and T.33 days (depending on the final fractional week that is included). 5. “Noise Characterization of Irregularly Spaced Data”. does not present results at unrealistically low tau. It is interesting that. Vol. These factors are smaller for other non-white PM and FM noise processes. 65 . As they point out in that paper. Histograms A histogram shows the amplitude distribution of the phase or frequency fluctuations. and the resulting TDEV is shown in curve D. Parker. One could obtain reasonable results for the shorter averaging times. Proceedings of the 12th European Frequency and Time Forum. None of the results is in good agreement with the nominal simulation..E. Another situation is data that are taken in bursts. especially for an uncorrelated white noise process. and is similar to that of Tavella and Leonardi. is close to 2. τavg. the (integrated) TDEV results are not (at least for a white PM noise process).g. Note that these time stability plots include points at all possible tau values. perhaps averaging those results to obtain better statistical confidence. Wednesday and Friday of each week. Leonardi. pp. Tavella and M. the standard deviation is approximately equal to the half-width at half-height (HWHA = 1. as shown in Figure 1 of Reference [1].177s) . If the sampled phase data are converted to frequency data using their timetag differences to determine the individual measurement intervals. They further suggest that the higher level of the estimated noise is related to the ratio of the true and interpolated sampling times (≈2. 209-214. P. although the converted frequency results are different depending on whether the average or individual (more correct) taus are used. the TDEV becomes highly distorted. 1996. the best approach is probably to analyze the segments separately. Metrologia.33 days.16. These sampled data therefore have an average tau of 7/3 = 2. By applying a correction factor of √2. The result with the linearly interpolated phase points is particularly bad for τ<τavg. It is essentially identical to that for the sampled phase data shown in curve B.33) and the √τ dependence of TDEV. as shown in curve A. 457-466. March 1998.

5 2.0 -1.5 -1.30 0.15 0.5 0 -1.5 1.177σ ±1σ Half-Amplitude 2 f(x)=(1/√2π)e -(x /2) 0.05 0 -3. σ=1) 0.0 -0.35 0.10 0.0 x (In units of σ) An example of a histogram for a set of white FM noise is shown below: 66 .0 1.HANDBOOK OF FREQUENCY STABILITY ANALYSIS Standardized Normal Distribution (µ=0.177σ +1.20 0.25 f(x) 0.0 -2.5 3.0 2.40 0.5 -2.

often by the method of least squares.. Frequency drift is the systematic change in frequency due to all effects. for a quartz crystal oscillator. while frequency aging is the change in frequency due to effects within the device.19. linear. etc. and its suitability may be judged by the degree to which it produces white (i. using either phase or frequency data. Drift Analysis Methods Several drift methods are useful for phase or frequency data as described below. Therefore. The term drift refers to the systematic change in frequency due to all effects. The difference between the first and last points of the phase data: y(t) = slope = [ x(end) − x(start) ] / (M-1).).17.e. which can be judged by the randomness of the residuals. Frequency Offset It is often necessary to estimate the frequency offset from either phase or frequency data. The average of the first differences of the phase data (optimum for white FM noise): y(t) = slope = [x(t+τ)−x(t)]/τ. aging refers to a change in the resonant frequency of its quartz crystal resonator. where M = # phase data points.SECTION 5 TIME DOMAIN STABILITY 5. Thus. 5. 67 . A least squares linear fit to the phase data (optimum for white PM noise): x(t) = a + bt. log. and it is often necessary (and usually advisable) to characterize and remove this systematic drift before analyzing the stochastic noise of the source. Frequency drift is generally analyzed by fitting the trend of the frequency record to an appropriate mathematical model (e.g. 3. Frequency offset is usually calculated from phase data by either of three methods: 1.18. drift is the quantity that is usually measured. The model may be based on some physical basis or simply a convenient equation. while drift would also include the effects of its external environment. but it is generally done under constant environmental conditions to the greatest extent possible so as to approximate the aging of the device. 2. This method is used mainly to match the two endpoints. The best method depends on the quality of the fit. Frequency Drift Most frequency sources have frequency drift. where slope = y(t) = b. while aging includes only those effects internal to the device. 5.. uncorrelated) residuals.

respectively. 3…. respectively . N. slope = y'(t) = 2c. 2.HANDBOOK OF FREQUENCY STABILITY ANALYSIS Data Phase Phase Phase Phase Phase Phase Freq Freq Freq Freq Method Quadratic Fit Avg of 2nd Diffs 3-Point Fit Linear Avg of 1st Diffs Endpoints Linear Fit Bisection Fit Log Fit Diffusion Fit Noise Model W PM RW FM W & RW FM Frequency Offset Frequency Offset Frequency Offset W FM W & RW FM Stabilization Diffusion 5. b and c coefficients have units of sec. This continuous model can be expressed as xn = a + τ0bn + τ0²cn² for n =1. Phase Drift Analysis Three methods are commonly used to analyze frequency drift in phase data: 1. The fit coefficients can be estimated by the following expressions [1]: FG H $ F b = G B∑ x + τ D ∑ nx + τ H F $ c = G C ∑ x + τ E ∑ nx + τ H N N n =1 N n =1 N n 0 n n =1 N n =1 N n 0 n n =1 n =1 2 $ a = A∑ xn + τ 0 B∑ nxn + τ 0 C ∑ n 2 xn / G n =1 N 2 0 2 0 IJ K I E ∑ n x J / Gτ K I F ∑ n x J / Gτ K N 2 n n =1 N 2 n n =1 0 2 0 . sec/sec and sec/sec2. and τ0 is the sampling interval for discrete data.20. and the frequency drift slope and intercept are 2c and b. A least squares quadratic fit to the phase data: x(t) = a + bt + ct². where the A-F terms are as follows 68 . where the a. where y(t) = x'(t) = b + 2ct.

The frequency drift intercept and slope are a and b. The average of the second differences of the phase data: y(t) = [x(t+τ)−x(t)]/τ. middle and end of the phase data: slope = 4[x(end)−2x(mid)+x(start)]/(Mτ)². where M = # data points. 5. It is the equivalent of the bisection method for frequency data. τ0 is the sampling interval for the discrete data. That continuous model can be expressed as yn = a + τ0bn for n =1. D = 12 2 N + 1 8 N + 11 / N + 1 N + 2 b b gb g b gb g A quadratic fit to the phase data is the optimum model for white PM noise. Frequency Drift Analysis Four methods are commonly used to analyze frequency drift in frequency data: 1. The fit coefficients can be estimated by the following expressions: 69 . A three-point fit at the start. 2. A least squares linear regression to the frequency data: y(t) = a+bt. 3. This method is optimum for random walk FM noise. 3….SECTION 5 TIME DOMAIN STABILITY g B = −18b2 N + 1g C = 30 A = 3 3N N + 1 + 2 b g F = 180 / b N + 1gb N + 2g G = N b N − 1gb N − 2g E = −180 / N + 2 2. and have units of sec and sec/sec. y(t) = a + bt. M where M is the number of frequency data points.21. Linear frequency drift can be estimated by a linear least squares fit to the frequency data. where a = intercept. b = slope = y'(t). respectively. slope = [y(t+τ)−y(t) ]/τ = [x(t+2τ)−2x(t+τ)+x(t)]/τ².

A linear fit to the frequency data is the optimum model for white FM noise. 70 . where slope = y'(t) = ab/(bt+1). 1983.A. 1992. and are a simple form of spectral analysis.22. pp. This bisection method is optimum for white and random walk FM noise. an all tau analysis is computationally intensive and can therefore be slow. "Confidence on the Three-Point Estimator of Frequency Drift". References for Drift 1. cyclic variations are often the result of interference between the sampling rate and some periodic instability (such as environmental sensitivity). J. Proc. 2. Barnes. Weiss and C. pp. A log model of the form (see MIL-O-55310B) that applies to frequency stabilization: y(t) = a·ln(bt+1).HANDBOOK OF FREQUENCY STABILITY ANALYSIS M M M $ b= M ∑ ny n − ∑ n ∑ y n n =1 M∑n n =1 M 2 F I − G ∑ nJ H K M n =1 n =1 n =1 2 and: $ a= ∑ yn n =1 M M − b⋅ ∑n n =1 M M . Proc.A. Dec. 1991. The frequency averages over the first and last halves of the data: slope = 2 [ y(2nd half) − y(1st half) ] / (Nτ). Such a “many tau” analysis can be orders of magnitude faster and yet provide the same information. Inc. 551-582. J. 5. where slope = y'(t) = ½·b(t+c)-1/2. "The Measurement of Linear Frequency Drift in Oscillators''. "The Analysis of Frequency and Time Data". Hackman. PTTI Meeting. 3. 2. however. For most purposes. Dec. 3. 451-460.. In particular. Austron. M. it is not necessary to calculate values at every tau. A diffusion (√t) model of the form y(t) = a+b(t+c)1/2. 24th Annu. Dec. All Tau Stability calculations made at all possible tau values can provide an excellent indication of the variations in the results. where N = number of points.A. However. PTTI Meeting. 4. 15th Annu. Barnes. but instead to do so at enough points to provide a nearly continuous curve on the display device (screen or paper).

It is often possible to control the environmental conditions sufficiently well during a stability run so that environmental effects such as temperature and supply voltage are negligible.SECTION 5 TIME DOMAIN STABILITY All Tau Stability Plot Many Tau Stability Plot 5. it can be very difficult to distinguish between those different mechanisms for phase or frequency variations.24. Those factors should be measured individually. Environmental Sensitivity Environmental sensitivity should be treated separately from noise when performing a stability analysis. However. Determining how well those factors have to be controlled requires knowledge of the device’s environmental sensitivities. if possible. over the largest reasonable 71 .

Test Methods for Electronic and Electrical Component Parts. P. No. 1990. its expected value is independent of the number of data points. “Environmental Factors and Hydrogen Maser Frequency Sensitivity”. More information will be found in the following references. T.745-751.” Proceedings of the 22nd Annual Precise Time and Time Interval (PTTI) Applications and Planning Meeting. MIL-0-55310. “Physical Origin of the Frequency Shifts in Cesium Beam Frequency Standards: Related Environmental Sensitivity. 8. C. the Allan deviation has the important advantage that. “Environmental Sensitivities of Precision Frequency Sources. Jorgensen.25. Gagnepain. MIL-STD-202. References for Environmental Sensitivity 1. 9. pp. “Physics of Systematic Frequency Variations in Hydrogen Masers. This is not just a matter of elegance. 2. 1990. 35. temperature affects the phase delay through a cable. 241-249. et al. 6. Walls and J. Because environmental sensitivity is highly dependent on device and application. No. pp. 3. H. 5. Many of the techniques used in the analysis of frequency stability. 72 .” Proceedings of the 22nd Annual Precise Time and Time Interval (PTTI) Applications and Planning Meeting.J. it does not receive detailed consideration in this handbook. “The Physics of Environmental Sensitivity of Rubidium Gas Cell Atomic Frequency Standards. pp.E. Winter 1988-89.M. Oscillators. because of its convergence characteristics for all common clock noises. General Specification for Military Specification.. IEEE Std 1193. IEEE Trans. 11. “General Considerations in the Metrology of the Environmental Sensitivities of Standard Frequency Generators. 2. which can cause confusion. a temperature ramp produces a rate of change of phase that mimics a frequency change in the source. Breakiron. pp.” Proceedings of the 1992 IEEE Frequency Control Symposium. Some environmental factors affect phase and frequency differently.J. “Special Relativity and Intersatellite Tracking”. 1990. Parker.” Proceedings of the 22nd Annual Precise Time and Time Interval (PTTI) Applications and Planning Meeting. 12. MIL-STD-810. Navigation. Hellwig.HANDBOOK OF FREQUENCY STABILITY ANALYSIS excursions to minimize the effect of noise. pp. Mattison. Environmental sensitivity can best be determined by considering the physical mechanisms that apply within the unit under test. For example. IEEE Guide for Measurement of Environmental Sensitivities of Standard Frequency Generators.” IEEE Transactions UFFC-39. 1992. however. Parsimony In any measurement or analysis. pp. Useful information about the environmental sensitivity of frequency sources can be found in the references below. “Environmental Sensitivities of Quartz Crystal Oscillators.S.. 816-830. 419-440. For example. Crystal. pp. Military Specifications and Standards. Environmental Test Methods and Engineering Guidelines. 429-442. L. Vol. pp. W. 7. no. UFFC-46. 4.A. 4.” IEEE Transactions IM-39. 10. et al. 1992. 453-464. 1990. 5. 3. the additional parameters may be implicit or arbitrary and thereby cause confusion or error if they are ignored or misunderstood. however. Riley. F. E. it is desirable to minimize the number of extraneous parameters. 1999. 301-306. Dynamically. 441-452. Audion.L.

g. Phase to Tau Accuracy Is measurement frequency interval known conversion exactly? Is it the same for each point? Frequency to Tau See above phase Initial phase Set to zero. Drift Removal Remove or not Is drift It is generally wise deterministic? to remove Is its cause deterministic drift known? before noise analysis Model White residuals Model may have are sign that physical basis (e. Outlier removal is an especially difficult case. Normalize Use to emphasize See “Remove frequency noise rather than average frequency frequency offset above” Analysis Drift Estimation Model Smallest residuals Can be critical. A listing of some of these nonparsimonious parameters is given in the table below: Table I. limit – can hide deeply in algorithm Remove Model Noise type Not necessarily average simple arithmetic frequency mean of frequency.SECTION 5 TIME DOMAIN STABILITY do require that certain parameters be chosen before they can be performed. drift removal requires the choice of a model to which the data will be fit (perhaps using the criterion of white residuals). Doesn’t affect subsequent stability analysis. Known physics can help choose. Generally conversion arbitrary. especially for predictions. where judgment often enters into the decision as to whether or not a datum is anomalous. Non-Parsimonious Parameters in Frequency Stability Analysis Type Process Parameter Criterion Remarks PreOutlier Number of Apply best Use of MAD-based processing Removal sigmas judgment robust statistics is recommended.. 73 . Generally uncritical Convergence For iterative fit. model is diffusion process) appropriate. For example.

noise type Time available for measurement. Critical for these biased estimators Resolution. drift Visibility Convention Generally uncritical – can hide deeply in algorithm Generally uncritical Process choice affects results As many as possible Avoid As large as confidence allows Generally unambiguous Hadamard Deviation Commonly used in GPS control operations. fill.HANDBOOK OF FREQUENCY STABILITY ANALYSIS Convergence limit Frequency Estimation Gaps Allan Deviation Noise model Skip. color Type – Parametric or nonparametric. Must remove outliers. Total Deviation Thêo1 Less commonly used Generally unambiguous Affects calc time AVAR or HVAR Personal preference Analysis tools available Knowledge of analyst Uncritical Dynamic Stability Spectral Analysis Clarity 74 . omit. mesh. Property of measuring system Confidence degrades Noise type significant affect on results See above Easier handling of clock with drift or divergent noise See above Better confidence at long tau. number of windows Data properties: noise. distribution. or exclude Number of data points Dead time Maximum AF Noise ID method to support error bars All Allan deviation parameters To use rather than Allan deviation or separate drift removal All Allan deviation parameters To use rather than Allan deviation Noise ID method to support bias removal Window and step size Variance type Viewpoint. Windowing – Bias reduction For iterative fit Lowest residuals for known noise type Number.

binominally weighted Hadamard variance. maximum at some confidence level Notation Smoothing – variance reduction Presentation – plot or fit Table vs. Transfer Functions Variances can be related to the spectral density of the fractional frequency fluctuations by their transfer functions. For example. etc. The transfer function is determined by the statistic’s time-domain sampling pattern. environmental conditions. Clarity Insight Clarity Clarity. Plot Time or frequency Include or not Remove or not Cost/risk tradeoff State outlier & drift removal.SECTION 5 TIME DOMAIN STABILITY Presentation Form Domain Error Bars Reference noise subtraction Nominal vs. Sy(f) is the spectral density of the fractional frequency fluctuations. and fh is the upper cutoff frequency (determined by hardware factors). zero dead-time. this relationship is σ 2H τ = bg z fh 0 S y f ⋅ HH f bg b g df 2 . for the Hadamard variance. The transfer functions for the most common variances used in frequency stability analysis are shown in the table below: Variance Allan Magnitude Squared Transfer Function |H(f)|² Hadamard F sin πτf IJ sin πτf 2G H πτf K F sin πτf IJ sin πτf 2G H πτf K 2 2 2 4 4 75 . correspondence with requirements Clarity If affects results May be specified Tradeoff vs resolution Use both Use both Use best Error bars recommended Only if similar to unknown Usually nominal at 1sigma confidence Disclose choices Judgment of analyst 5. HH(f) is its transfer function.26. where σ²Η (τ) is the three-sample.

The squared magnitudes of these transfer functions are shown in the plots below: Transfer Function of Allan Variance 2-Sample (N=2).HANDBOOK OF FREQUENCY STABILITY ANALYSIS For πτf << 1. where n is an integer). the transfer function of the Allan variance behaves as (πτf)2.6 0. Binomially Weighted Coefficients (1.4 0..2 0 0 1 2 3 4 5 x=πτf 6 7 8 9 10 Allan Variance Transfer Function of Hadamard Variance 3-Sample (N=3).8 |HA(f)| 2 0.2. Zero Dead-Time (T=τ). As a 76 . indicating that it is convergent for power law processes as low as α = -4 (Random Run FM).1) 8 6 |HH(f)| =2 (sin(πτf)/πτf) (sin (πτf) 2 4 2 4 |HH(f)| 2 4 2 0 0 1 2 3 4 5 x=πτf 6 7 8 9 10 Hadamard Variance These responses have their peaks where the frequency is one-half the sampling rate. while the transfer function of the Hadamard variance behaves as (πτf)4. at f = n/τ.e.2 1. Zero Dead-Time (T=τ) 1. and nulls where it is a multiple of the sampling rate (i. indicating that it is convergent for power law processes Syα down to as low as α = -2 (Random Walk FM).0 ⏐HA(f)⏐ =2(sin(πτf)/πτf) (sin (πτf) 2 2 2 0.

. "Oscillator Specifications: A Review of Classical and New Ideas". 4. Symp. respectively [5]. R. Metrologia 43 (2006).291-301. 43rd Annu. “ThêoH: A Hybrid. 1977 IEEE International Freq. References for Transfer Functions 1. Contrl. D. 5.G. December 1977. S322-S331. Similar transfer functions exist for the modified. Proc. Freq. “A Direct Time-Domain Measure of Frequency Stability: The Modified Allan Variance”. 77 . 66. 6. Howe and C. Meas.411 τ-1 and 0. 9. Proc. D. 38-41. 331-335.476 τ . J. pp. pp. Wan. Visr and J. J.A. E. 39-48.. Proc. Howe. June 1989. "Extended Variances and Autoregressive Moving Average Algorithm for the Measurement and Synthesis of Oscillator Phase Noise". Roberts. the Hadamard variance has slightly higher resolution than the Allan variance. High-Confidence Statistic that Improves on the Allan Deviation”. Rutman.A. June 1977. 29th Annual Precise Time and Time Interval (PTTI) Meeting. pp. 2. Symp. No. Instrum. 3. Rutman. K.SECTION 5 TIME DOMAIN STABILITY spectral estimator. IEEE Trans.A. Vol. Greenhall.. since the equivalent noise bandwidths of the Hadamard and Allan spectral -1 windows are 0. Contrl. "Characterization of Phase and Frequency Instabilities in Precision Frequency Sources". pp. total and Thêo1 variances. IEEE. Wiley. September1978. IM-26. “Total Variance: A Progress Report on a New Frequency Stability Characterization”.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS 78 .

The most commonly encountered noise spectra are White (f 0) Flicker (f −1) Random Walk (f −2) Flicker Walk (f −3) . Spectral stability measures are directly related to the underlying noise processes. where: Sy(f) f h(α) α = = = = one-sided power spectral density of y. Noise Spectra The random phase and frequency fluctuations of a frequency source can be modeled by power law spectral densities of the form Sy(f) = h(α)f α . Examples of these noise types are shown in the figure below. Hz intensity coefficient exponent of the power law noise process.1. the fractional frequency fluctuations. 79 . 6. Frequency stability can also be characterized in the frequency domain in terms of a power spectral density (PSD) that describes the intensity of the phase or frequency fluctuations as a function of Fourier frequency.SECTION 6 FREQUENCY DOMAIN STABILITY 6 Frequency Domain Stability Frequency domain stability measures are based on power spectral densities that characterize the intensity of the phase or frequency fluctuations as a function of Fourier frequency. and are particularly appropriate when the phase noise of the source is of interest. 1/Hz Fourier or sideband frequency.

The spectral characteristics of the power law noise processes commonly used to describe the performance of frequency sources are shown in the following table: Spectral Characteristics of Power Law Noise Processes Noise Type α β µ µ' η White PM 2 0 −2 −3 −1 Flicker PM 1 −1 −2 −2 0 White FM 0 −2 −1 −1 1 Flicker FM −1 −3 0 0 2 Random Walk FM −2 −4 1 1 3 6. where Sx(f) = PSD of the time fluctuations. The relationship between these is Sφ(f) = (2πν0)² · Sx(f) = (ν0/f) ²· Sy(f) and £(f) = 10 · log[½ · Sφ(f)] .HANDBOOK OF FREQUENCY STABILITY ANALYSIS Power law spectral models can be applied to both phase and frequency power spectral densities. Phase is the time integral of frequency. α<2 µ' = − (α+1. rad²/Hz and its logarithmic equivalent £(f). α<3 η = − (α−1). hertz. dBc/Hz. sec²/Hz. α<3 . where ν0 is the carrier frequency. so the relationship between them varies as 1/f²: Sx(f) = Sy(f)/(2πf)² . These frequency-domain power law exponents are also related to the slopes of the following time domain stability measures: Allan variance Modified Allan variance Time variance σ²y(τ) Mod σ²y(τ) σ²x(τ) µ = − (α+1). Two other quantities are also commonly used to measure phase noise: Sφ(f) = PSD of the phase fluctuations. Power Spectral Densities Four types of power spectral density are commonly used to describe the stability of a frequency source: 80 .2. The power law exponent of the phase noise power spectral densities is β = α−2.

SSB Phase Noise Diagram -60 RW α = -2 FM Sy(f) ∼ f α β α = -1 -µ β = α -2 -80 Sφ(f) ∼ f £(f) = 10·log10 [½· Sφ(f)] -40 Flicker α = -1 FM -100 £(f). t. Both are commonly called "phase" to distinguish them from the independent time variable. PSD of Time Fluctuations Sx(f) The power spectral density of the time fluctuations x(t) in units of sec²/Hz is given by Sx(f) = h(β) · f β = Sy(f)/(2πf)².SECTION 6 FREQUENCY DOMAIN STABILITY PSD of Frequency Fluctuations Sy(f) The power spectral density of the fractional frequency fluctuations y(t) in units of 1/Hz is given by Sy(f) = h(α) · f α. Hz for various power law noise processes. where ν0 = carrier frequency. Fourier frequency. dBc/Hz versus log f. The time fluctuations are related to the phase fluctuations by x(t) = φ(t)/2πν0. -120 dBc/Hz -140 -30 Slopes in dB/decade White α=0 FM Flicker PM α = 1 White PM -20 -10 α=2 -160 0 1 0 log f. SSB Phase Noise £(f) The SSB phase noise in units of dBc/Hz is given by £(f) = 10 · log [½ · Sφ(f)]. Hz. 6.3. Hz 2 3 4 5 Figure 6. where f = sideband frequency. £(f). Hz. This is the most common function used to specify phase noise. Phase Noise Plot The following diagram shows the slope of the SSB phase noise.3 SSB Phase Noise Plot 81 . where: β = α−2. PSD of Phase Fluctuations Sφ(f) The power spectral density of the phase fluctuations in units of rad²/Hz is given by Sφ(f) = (2πν0)² · Sx(f).

Without averaging. The averaging factor improves the PSD standard deviation by the square root of the averaging factor. variance and spectral resolution. one or the other is usually preferred. Spectral analysis is most often used to characterize the short-term (< 1 second) fluctuations of a frequency source as a plot of phase noise power spectral density versus sideband frequency.4. Modern instrumentation is tending to merge these approaches by digitizing the signal waveform at high sampling rates.7. 3]. The tradeoff in this averaging process is that each section of the data is shorter. An example of seven of these functions for order J=4 is shown below. 6. in practice. and there is much technical literature on that subject [2. or as the amplitude and phase at discrete frequencies.6. Spectral Analysis Spectral analysis is the process of characterizing the properties of a signal in the frequency domain. either because of measurement and/or analysis convenience. yielding a result with coarser frequency resolution that does not extend to as low a Fourier frequency. This can be accomplished by dividing the data into sections. Some sort of filtering or averaging is usually necessary to provide less noise in the PSD results. PSD Averaging Without filtering or averaging. 82 . performing an FFT analysis on each section separately. PSD Windowing Data windowing is the process of applying a weighting function that falls off smoothly at the beginning and end to avoid spectral leakage in an FFT analysis. For the former. While. The most common windowing types are Hanning. Without windowing.5. spectral analysis can aid in the identification of periodic components such as interference and environmental sensitivity. For the latter. each spectral result has only two degrees of freedom. 6. bias will be introduced that can severely restrict the dynamic range of the PSD result. Spectral analysis can thus be applied to both noise and discrete components for frequency stability analysis. thereby allowing FFT analysis at relatively high Fourier frequencies. Nevertheless. while a time domain analysis is most often used to provide information about the statistics of its instability over longer intervals (> 1 second). there are many pitfalls and subtleties associated with spectral analysis that must be considered if meaningful results are to be obtained. More data provides finer frequency resolution. Multitaper PSD Analysis The multitaper PSD analysis method offers a better compromise between bias. for white noise. Time domain data can be used to perform spectral analysis via the Fast Fourier Transform (FFT). or because the results are more applicable to a particular application. spectral analysis complements statistical analysis in the time domain. Hamming and Multitaper. not lower noise (while the data sampling time determines the highest Fourier frequency). either as a power spectral density for noise. The classic Hanning and Hamming windows can be applied more than one time. 6. the variances of the PSD results are always equal to their values regardless of the size of the time domain data set. in principle.HANDBOOK OF FREQUENCY STABILITY ANALYSIS 6. Averaging is accomplished by applying a set of orthogonal windowing (tapering) functions called discrete prolate spheroidal sequences (DPSS) or Slepian functions to the entire data array. and then averaging those values to obtain the final PSD result. time and frequency domain analyses provide equivalent information.

0 4. where N is the number of data points sampled at time interval t. which affects the variance.0 3.5 4. 83 .0 2.0 # Windows 1-3 1-4 1-5 1-6 1-7 1-8 1-9 The resolution BW is given by 2J/Nt. A higher J permits the use of more windows without introducing bias. which provides more averaging (lower variance) at the expense of lower spectral resolution. which affects the resolution bandwidth. An adaptive algorithm can be used to weight the contributions of the individual tapers for lowest bias. These multiple tapering functions are defined by two parameters.5 3. It is therefore particularly useful for examining discrete components along with noise.4. The higher windows have larger amplitude at the ends that compensates for the denser sampling at the center.3 Slepian SPSS Taper Functions The 1st function resembles a classic window function. J. as shown in the following table: Order J 2. The multitaper PSD has a flat-topped response for discrete spectral components that is nevertheless narrower than an averaged periodogram with the same variance.SECTION 6 FREQUENCY DOMAIN STABILITY Figure 6.5 5. and the number of windows. the order of the function. while the others sample other portions of the data.

2.. Walden. The amplitude of discrete components should be increased by the log of the BW (Fourier frequency spacing in hertz). 1988. Contrl. AISN Software. 2007. References for Spectral Analysis 1. McGee and D. Lees and J. Ultrasonics. 3. IEEE . M. ISBN 0-521-43541-2. Vol.H. “ThêoH and Allan Deviation as Power-Law Noise Estimators”. Flannery. S.. "A C-Subroutine for Computing Multi-Taper Spectral Analysis". J. Ferroelectrics and Freq. B. 1999. Help File for the AutoSignal program for spectral analysis. which is a negative number for typical sub-hertz bandwidths. the value of n is reduced by the averaging factor. IEEE Trans. 5. 195-236. Press.J. 1982. D. 1993.A. Amplitude corrections need to be made for the noise response of the windowing functions. "Spectrum Estimation and Harmonic Analysis".T. pp.8. Vetterling. Teukolsky and W. Howe. pp. D. Park. The PSD fit is a least-squares power law line through octave-band PSD averages [6]. Computers in Geosciences . PSD Notes A carrier frequency parameter applies to the Sφ(f) and £(f) PSD types. Vol.A. Proc. J. Thomson. Sept. Cambridge University Press.T. Percival and A. 9.P.HANDBOOK OF FREQUENCY STABILITY ANALYSIS 6. 21. Inc. The number of Fourier frequency points is always the power of 2 greater than or equal to one-half of the number of time domain data points. Spectral Analysis for Physical Applications. 84 . Feb. 70. 4. and should include the quantitative display of power law noise in common PSD units. No. 1055-1096. Chapter 12. Cambridge University Press. perhaps with fits to integer power law noise processes. ISBN 0-521-35465-X. and the highest Fourier frequency is 1/2t. W. For characterizing frequency stability. a spectral analysis is used primarily for the analysis of noise (not discrete components). The spacing between Fourier frequency points is 1/nt.B. If averaging is done. 6. n. Numerical Recipes in C. 1995.

Time domain frequency stability is related to the spectral density of the fractional frequency fluctuations by the relationship σ 2 M . The equivalent expression for the modified Allan variance is Modσ y (τ ) = 2 2 N 4π 2τ 0 2 z fh S y ( f )sin 6 (πτf ) f 2 sin 2 (πτ 0 f ) 0 df . σ²y(τ). The latter method can be applied measures of frequency stability. 85 . Conversions between these domains may be made by stability measurements may be numerical integration of their fundamental easier to perform in one domain than relationship. in the frequency domain. where |H(f)|2 is the transfer function of the time domain sampling function. The transfer function of the Allan (two-sample) time domain stability is given by H ( f ) = 2 sin 4 (πτf ) / (πτf )2 2 .SECTION 7 DOMAIN CONVERSIONS 7 Domain Conversions The stability of a frequency source can be specified The stability of a frequency source and measured in either the time or the frequency can be specified and measured in domain. The general conversion from time to frequency domain is not unique because white and flicker phase noise have the same Allan variance dependence on τ. The main lobe of the σy(τ) and Mod σy(τ) responses occur at the Fourier frequency f = 1/2τ. in the time domain. T . only when the dominant noise process decreases toward higher sideband frequencies. Conversions are possible on a power law spectral model for the noise between these generally equivalent processes involved. One domain is often Allan variance. Examples of these stability measures are the either the time or the frequency domain. Otherwise. the more fundamental method based on numerical integration must be used. and the preferred to specify the stability spectral density of the fractional frequency because it is most closely related to fluctuations. and therefore the Allan variance can be found from the frequency domain by the expression σ y2 τ = 2 bg z fh 0 Sy( f ) sin 4 (πτf ) df (πτf )2 . When performing any of these conversions. or by an approximation method based the other. Similarly. Sy(f). it is necessary to choose a reasonable range for σ and τ in the domain being converted to. the particular application. τ = S y ( f ) ⋅ H ( f ) ⋅ df 0 b g z ∞ 2 .

Then. For this example. The σy(1)white FM noise fit parameter is 1.2a. the same data are analyzed in the frequency domain with an £(f) PSD. and τ0 is the basic measurement time. The number of points is chosen as an even power of 2 for the frequency data. where the hα terms define the level of the various power law noises. and the time domain properties of this noise are analyzed by use of the overlapping Allan deviation. This factor applies only to white and flicker PM noise. 1 1038 + 3 log e 2πf hτ 3 fh + h1 + h2 2 2 2 2τ 2π τ 2π τ 2 b g .2b. we generate 4097 phase data points of simulated white FM noise with a 1second Allan deviation value σy(1) = 1x10-11 and a sampling interval τ = 1 msec. close to the desired value. 7. Power Law Domain Conversions Domain conversions may be made for power law noise models by using the following equation and conversion formulae: σ y 2 τ = h−2 bg b2π g τ + h 2 6 −1 2 log e 2 + h0 .08e-11. The above conversion formulae apply to the ThêoH hybrid statistic as well as to the Allan variance.HANDBOOK OF FREQUENCY STABILITY ANALYSIS 7. a power law domain conversion is done. Example of Domain Conversions This section shows an example of time and frequency domain conversions.1. and those results are compared with those of the spectral analysis. 86 . the other power spectral density types are examined.038 + 3·ln(2πfhτ0)/4π² E = 3fh /4π² . a set of simulated power law noise data is generated. Next.2. Noise Type RW FM F FM W FM F PM W PM σ²y(τ) A · f 2 · Sy(f) · τ 1 B · f 1 · Sy(f) · τ 0 C · f 0 · Sy(f) · τ −1 D · f −1 · Sy(f) · τ −2 E · f −2 · Sy(f) · τ −2 Sy(f) A−1 · τ −1 · σ²y(τ) · f −2 B−1 · τ 0 · σ²y(τ) · f −1 C−1 · τ 1 · σ²y(τ) · f 0 D−1 · τ 2 · σ²y(τ) · f 1 E−1 · τ 2 · σ²y(τ) · f 2 where A = 4π²/6 B = 2·ln2 C = 1/2 D = 1. Finally. and fh is the upper cutoff frequency of the measuring system in hertz. First. The generated set of simulated white FM noise is shown as frequency data in Figure 7. and their overlapping Allan deviation is shown in Figure 7.

2 dBc/Hz and a slope of –19. as shown in Figure 7.6 dB/decade.1. the SSB phase noise to signal ratio in a 1 Hz BW as a function of sideband frequency.SECTION 7 DOMAIN CONVERSIONS Figure 7.2a Simulated W FM Noise Data Figure 7. 87 .2c. as shown in the table below. in close agreement with the expected values of -80 dBc/Hz and -20 dB/decade. The expected PSD values that correspond to the time domain noise parameters used to generate the simulated power-law noise can be determined by the power law domain conversion formulae of section 7. The fit parameters show an £(1) value of -79. f.2b Overlapping Allan Deviation The power spectrum for the phase data is calculated by use of a 10 MHz carrier frequency and a £(f) power spectral density type.

Burgoon and M. S. pp.00e-3. the spectral density of the time fluctuations in sec2/Hz. pp. J. 5. 160-165.1977. pp. 240-244.514-519. Meas.00000e-11 Time Domain: Power Type RWFM FFM WFM FPM WPM All Law Noise: dB/dec PSD -40 None -30 None -20 -80. Greenhall. Cont. 52.0 -10 None 0 None -80.. 9. C.R. J. “Practical Calculation Methods of Spectral Density of Instantaneous Normalized Frequency Deviation from Instantaneous Time Error Samples”.00000e+01 Sigma Type: Normal Tau (Sec): 1. 2. Contrl. pp. 88 . Egan. 7. 2. Vol. Proc. the spectral density of the phase fluctuations in rad2/Hz. 3. 28th Freq. Asmar and K. June 1997. June 1978.HANDBOOK OF FREQUENCY STABILITY ANALYSIS Domain Calculation Results Frequency Domain: PSD Type: L(f).C. Thomson. IEEE Trans. "Does Allan Variance Determine the Spectrum?".J. F. No. "Relations Between Spectral Purity and Frequency Stability”.. and fo = 10 MHz are shown in the following table.F. Dec. 6. “An Efficient Algorithm to Compute Allan Variance from Spectral Density”.00000e+00 0. Rutman. 9th PTTI Meeting. 523-548. pp. IEEE Trans. pp. 358-365. R.A. dBc/Hz SB Freq (Hz): 1. March 1991. A. 5th European Freq. June 1988. Gagnepain. The expected value of all these quantities for the simulated white FM noise parameters with σy(1) = 1. 37. Chi. 449-455. 1997 Intl. Oudrhiri. Freq.0 The other types of PSD that are commonly used for the analysis of frequency domain stability analysis are Sφ(f).... pp.00000e+00 1. W. No. F. 4. Sept. 2005. Groslambert and J. “The Mechanics of Translation of Frequency Stability Measures Between Frequency and Time Domain Measurements”. Sx(f). and Time Forum. References for Domain Conversion 1. Vernotte. Instrum. Proc.00000e-11 0. Contrl. 14681472. Fisher.00000e+00 0. τ = 1. the spectral density of the fractional frequency fluctuations in units of 1/Hz. 32th Freq. Proc. “Limitations on the Use of the Power-Law Form of Sy(f) to Compute Allan Variance”. “Conversion Between Time and Frequency Domain of Intersection Points of Slopes of Various Noise Processes”. Symp. and Sy(f).00e-11.00000e+00 1. Symp.00000e+00 Carrier(MHz): 1. Proc. UFFC. May 1974.00000e-03 Avg Factor: 1000 Type RWFM FFM WFM FPM WPM All Mu +1 0 -1 -2 -2 Sigma 0. Symp. Vol.

03e-24 1.SECTION 7 DOMAIN CONVERSIONS PSD Type £(f).70 -23.96 -1.2d Sφ(f) Power Spectral Density 89 .2c £(f) Power Spectral Density Figure 7.40e-22 Fit Exponent -1.96 -1.066e-24 2e-22 Log Value same -7. dB/decade -20 -20 -20 0 Fit Value -79.3 -21. rad2/Hz Sx(f). sec2/Hz Sy(f). 1/Hz Data Type Phase Phase Phase Frequency Simulated Value -80 2e-8 5.2 1.04 Parameter These spectral densities are shown in the plots below: Figure 7.7 Slope. dBc/Hz Sφ(f).19e-8 3.96 -0.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS Figure 7.2f Sy(f) Power Spectral Density 90 .2e Sx(f) Power Spectral Density Figure 7.

noise recognition.SECTION 8 NOISE SIMULATION 8 Noise Simulation It is valuable to have a means of generating simulated power law clock noise having the desired noise type (white phase. flicker phase. The noise type and time series of a set of simulated phase data are shown below: Noise Type Random Walk FM α=-2 Random Run Noise Phase Data Plot Flicker FM α=-1 Flicker Walk Noise White FM α=0 Random Walk Noise Flicker PM α=1 Flicker Noise 91 . A good method for power-law noise generation is described in Reference 8. frequency offset. and modeling. and perhaps a sinusoidal component. frequency drift. This can serve as both a simulation tool and as a way to validate stability analysis software. Allan deviation. flicker frequency and random walk frequency noise). particularly for checking numerical precision. white frequency.

Flicker Noise Generation Flicker noise is more difficult to generate because it cannot be described exactly by a rational transfer function. White noise can be numerically integrated and differenced to transform it by 1/f2 and f2. The most common methods involve linear filtering by RC ladder networks [5]. and subtract 6 [4]. The FFT method can produce noise having any integer power law exponent from α = -2 (RW FM) to α = +2 (W PM) [7]. Frequency Offset. and convert it to a new set of RR FM phase data. 8. For example. [3] to produce a white spectrum with Gaussian deviates. 8. Drift. One popular technique is to first generate two independent uniformly distributed random sequences [1]. frequency drift.3. and combine them using the BoxMuller transform [2]. 8. and Sinusoidal Components Beside the generation of the desired power law noise. 92 . add them. and a sinusoidal component in the simulated clock data. [8].4. Another method is to generate 12 independent random sequences uniformly distributed between 0 and 1. White Noise Generation White noise generation is straightforward. This will via the central limit theorem produce a Gaussian distribution having zero mean and unit variance.1.2. first generate a set of RW FM phase data. to produce simulated noise having any even power law exponent. to generate RR FM noise. or by FFT transformation [7]-[9]. Then treat this RW FM phase data as frequency data.HANDBOOK OF FREQUENCY STABILITY ANALYSIS White PM α=2 White Noise 8. it is desirable to include selectable amounts of frequency offset. respectively. Flicker Walk and Random Run Noise Generation The more divergent flicker walk FM (a = -3) and random run FM (a = -4) power law noise types may be generated by using the 1/f² spectral property of a frequency to phase conversion. and much effort has been devoted to generating it [5]-[9].

Bregni.T. 10. Proc. "The Analysis of Frequency and Time Data". S. Proceedings of the 34th Annual Precise Time and Time Interval (PTTI) Systems and Applications Meeting. "FFT-Based Methods for Simulating Flicker FM". Proc.S Department of Commerce. F. Communications World. 1995. "Large Sample Simulation of Flicker Noise". 10. pp. L. 5. V.P Box and M. Teukolsky and W. Muller. "The Generation and Recognition of Flicker Noise". 8. 2.A. G.. Proc. Barnes and D. Vol. 7. ACM. May 1992.W. 1192-1201. “A Note on the Generation of Random Normal Deviates”. 461. National Bureau of Standards. Allan.A. Contrl. Greenhall and J. Kasdin and T. J. 1991. Symp. W. Park and K. S.1958. Proceedings of the IEEE. U. Barnes.A. 274-283. C.P. "Random Number Generators: Good Ones are Hard to Find''.A. PTTI Meeting. pp. B. 4. S. 3. 24th Annu. U.E. p. Barnes. 19th Annu. 5660. J. 203-217. Flannery. Editor: N. Greenhall. Dec. J.J.H.SECTION 8 NOISE SIMULATION References for Noise Simulation 1. Dec. pp. 9. Vol. of IEEE GLOBECOM '95. 54.W. 2. 1992. Inc. Annals Math. NBS Report 9284. Numerical Recipes in C.K. Press. 29. 93 . 1987 and Proc. Valtriani. February 1966. Walter. Nov. pp. 11. Bregni. Austron. Vetterling. 1988.A. WSES Press. “A Statistical Model of Flicker Noise”.A. "Discrete Simulation of Power Law Noise". N. Cambridge Univ. 6. No. Veghini. Comm. June 1967. C. 2001. Barnes. Dec. Miller. 1992 IEEE Freq. pp. S. "Generation of Pseudo-Random Power-Law Noise Sequences by Spectral Shaping". Mastorakis... Stat. 176-178. 31.K.E. Press. Cambridge. pp. 2002.A. 610-611. "Simulation of Clock Noise and AU-4 Pointer Action in SDH Equipment". PTTI Meeting. No.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS 94 .

take many forms. and determines how long data can 95 . the scatter in the data when the two inputs are driven coherently by the same source. high-resolution multichannel clock measuring systems with associated archival databases. for high performance devices. The resolution of a digital frequency. The most important attribute of a frequency measuring system is its resolution. requires 1 picosecond/second (pp1012) or better resolution. period or time interval counter is determined mainly by its speed (clock rate) and the performance of its analog interpolator (if any).1 Block Diagram of a Time Interval Counter Measuring System This measurement method is made practical by modern high-resolution interpolating time interval counters that offer 10 digit/second resolution. as described below: 9. the ratio of the RF input to the IF beat frequencies. The performance of the measuring system also depends on the stability of its reference source. Three methods are commonly used for making precise time and frequency measurements. A low noise ovenized quartz crystal oscillator may be the best choice for a reference in the short term (1-100 seconds). one of which may be adequate resolution and a low noise considered to be the reference.1. and cesium beam tube devices at longer averaging times. Noise is another important consideration for a high-performance measuring system whose useful resolution may be limited by its noise floor. The resolution is not affected by the division ratio. Time Interval Counter Method The time interval counter method divides the two sources being compared down to a much lower frequency (typically 1 pulse/second) and measures their time difference with a high resolution time interval counter: Figure 9.SECTION 9 MEASURING SYSTEMS 9 Measuring Systems Frequency measuring systems are instruments that A frequency measuring system with accept two or more inputs. That resolution generally improves linearly with the averaging time of the measurement. while a active hydrogen maser generally provides excellent stability at averaging times out to several days. These systems can clock measurements. They can be custom built or bought from several organizations specializing in such systems. and compare their floor is necessary to make precision relative phase or frequencies. which sets the minimum measurement time. and it can be enhanced by preceding it with a mixer that improves the resolution by the heterodyne factor. and more elaborate hardware than a counter. from the direct use of a frequency counter to elaborate low-noise. which.

and measures the period of the resulting audio-frequency beat note.2. requires a priori knowledge of the sense of the frequency difference.8 days before a 1 pps phase spillover occurs after being initially set at the center. 9. Even worse. for example. A disadvantage of this approach is that a stable offset reference is required at exactly the right frequency. Thus a period counter with 100 nanosecond resolution (10 MHz clock) can resolve clock phase changes of 1 picosecond. using a time interval counter to measure the relative phase of the beat signals from a pair of mixers driven from a common offset reference: 96 . Heterodyne Method The heterodyne method mixes (subtracts) the two sources being compared.9999 MHz Hz offset reference will produce a 100 Hz beat signal whose period variations are enhanced by a factor of 10 MHz/100 Hz = 105.3. be measured for only about 5. For example. and often has dead time between measurements.2 Block Diagram of a Heterodyne Measuring System This heterodyne technique is a classic way to obtain high resolution with an ordinary period counter.HANDBOOK OF FREQUENCY STABILITY ANALYSIS be taken before a phase spillover occurs (which can be hard to remove from a data set). combines the best features of the first two. in effect. mixing a 10 MHz source against a 9. it can measure only frequency. It is based on the principle that phase information is preserved in a mixing process. 9. The measurement resolution is increased by the heterodyne factor (the ratio of the carrier to the beat frequency). A source having a frequency offset of 1x10-6 can. Figure 9. Drift in the trigger point of the counter can be a limitation to this measurement method. Dual Mixer Time Difference Method The third method.

3 Block Diagram of a Dual Mixer Time Difference Measuring System This dual mixer time difference (DMTD) setup is arguably the most precise way of measuring an ensemble of clocks all having the same nominal frequency. because its effect cancels out in the overall measurement process. When expanded to multiple channels by adding additional buffer amplifiers and mixers. or the zero-crossings can simply be time tagged. The measuring system resolution is determined by the time interval counter or time-tagging hardware.10 Hz offset oscillator (providing a 5x106/10 = 5x105 heterodyne factor). the zero-crossings should be coincident or interpolated to a common epoch. and the beat note is time tagged with a resolution of 100 nsec (10 MHz clock). Instruments using those techniques are available that automatically make both time and frequency domain measurements. Measurement Problems and Pitfalls It can be difficult to distinguish between a bad unit under test and a bad measurement.4. they usually cause results that are worse than expected. An early commercial version is described in Reference [3]. and time tagging the zerocrossings of the beat notes for each channel. and a newer technique is described in Reference [8]. For example. nor must it have particularly low noise or high accuracy.SECTION 9 MEASURING SYSTEMS Figure 9. Cross-correlation methods can be used to reduce the DDS noise. the measuring overall system resolution is 10-7 /5x105 = 0. if two 5 MHz sources are mixed against a common 5 MHz . It is nearly impossible for a measurement problem to give better than correct results. Additional counters can be used to count the whole beat note cycles to eliminate their ambiguity. and the mixer heterodyne factor. this arrangement allows any two of the clocks to be intercompared. A direct digital synthesizer (DDS) can be used as the offset reference to allow measurements to be made at any nominal frequency within its range. For best cancellation. 9. When problems occur in time-domain frequency stability measurements. The offset reference need not be coherent. and there is considerable justification in saying that the best results are the 97 . Multichannel DMTD clock measuring systems have been utilized by leading national and commercial metrology laboratories for a number of years [1-5].2 psec.

Think about all the environmental sensitivities of the unit under test. while not a measurement problem per se. One common problem is hum that contaminates the measurements due to ground loops.g. Teflon dielectric is an especially bad choice for temperature stability. electromagnetic. must also be considered. while foamed polyethylene is much better. aliased "beats" occur in the frequency record. Measuring System Summary A comparison of the relative advantages and disadvantages of these methods is shown in the following table: Table 9. Inspection of the raw data can show this. it is important to consider the mechanical rigidity of the test setup (e.g. and not necessarily coherent with it. In general.5 Comparison of Time and Frequency Measurement Methods Method Advantages Disadvantages Divider & Time Interval Counter Provides phase data Covers wide range of carrier frequencies Easily expandable at low cost Modest resolution Not suitable for short tau 98 . 1x10-15 = 1 psec/1000 seconds). Two possible exceptions to this are (1) misinterpretation of the scale factor. etc. Because the measurement interval is usually much longer than the period of the power line frequency. magnetic. as long as the operating conditions of the unit under test and the reference are undisturbed. Lack of stationarity (changes in the source itself).g. It is difficult to maintain the integrity of a measuring system over a long period. Stories abound about correlations between elevators moving and cars coming and going ("auto-correlations") that have affected clock measurements. Phase jumps are a problem that should never be ignored.. All sorts of other mechanisms (electrical. gaps in the data record may be acceptable. Slow periodic fluctuations will show up more distinctly in an all tau (rather than an octave tau) stability plot. In fact. acoustic.. 9.) exist that can interfere with time domain frequency measurements.5. and the best cure is often isolation transformers in the signal leads. this is a wise precaution to take in all cases. Examination of the raw phase record is critical because a phase jump (frequency impulse) appears strangely in a frequency record as a white FM noise characteristic [10]. Even a few degrees of temperature variation will cause the phase of a highstability source to "breathe" as it passes through 100 feet of coaxial cable.3 mm). injection locking of one source to another due to inadequate isolation. thermal. Some large phase jumps are related to the carrier period (e. This includes the electrical length (phase stability) of the connecting cables. Think about what can have changed in the overall test setup. where picoseconds matter (e. the more devices available and the more measurements being made. Be alert to day-night and weekly cycles that indicate human interference. and (2) inadvertent coherency (e. barometric. vibrational. a malfunctioning digital frequency divider). but.HANDBOOK OF FREQUENCY STABILITY ANALYSIS correct ones. An uninterruptible power system is indispensable to maintain the continuity of a long run. and control as many of them as possible.g. 1 psec = 0. In high-precision measurements. the easier it is to sort things out.

and it must be represented with sufficient precision to convey those variations after removal of fixed offsets (see section 10. particularly to identify anomalies or to compare several sources. and a system using a dual-mixer time interval measurement is recommended. An example of highly quantized frequency data is the random telegraph signal shown below. Nevertheless. A low-noise reference not only supports better measurement precision but also allows measurements to be made faster (with less averaging).. no time tags are necessary. Although these data have a non-Gaussian amplitude distribution (their histogram consists of two spikes).1 below). 99 . highly quantized data can still contain useful information. but most software will reject lines that start with a # or some other non-numeric character. Time tagging is generally required for archival storage of clock measurements. Nevertheless.4 below). It is desirable that the measurement control.6. The inclusion of comments and headers can pose problems.g. the use of timetags is recommended (see section 9. Double-precision exponential ASCII numeric format is recommended for ease of reading into most analysis software. the random occurrences of the two levels produce a white FM noise characteristic. especially after they are combined into longer averaging times.7. Data Format A one-column vector is all that is required for a phase or frequency data array. 9. and analysis functions be separated to provide robustness and networked access to the data. The recommended unit for phase data is seconds. An automated high-resolution multi-channel clock (phase) measuring system with a high-performance (e. with comma or space-delimited fields and one data point per line. Data Quantization The phase or frequency data must be gathered with sufficient resolution to show the variations of interest. but one that can pay off in better productivity. 9.SECTION 9 MEASURING SYSTEMS Resolution enhanced by heterodyne factor Mixer and Period Counter Provides direct frequency data Usable for short tau Expandable at reasonable cost High resolution. while frequency data should be in the form of dimensionless fractional frequency. low noise Provides phase data Dual Mixer Time Difference Offset reference noise & inaccuracy cancels No fixed reference channel No phase data No frequency sense Requires offset reference Single carrier frequency Single carrier frequency Relatively complex It is preferable to make continuous zero dead time phase measurements at regular intervals. data storage. but a single vector of extracted gap-filled data is sufficient for analysis. hydrogen maser) reference is a major investment. Because the data points are equally spaced.

.2. Based on the astronomical Julian Date. It is widely used.8.000.5. 9. particularly for identifying the exact time of some anomaly. Archiving and Access There is no standard way to archive and access clock data. is easily converted to other formats. temperature) from a data acquisition system. The MJD (including the fractional day) can be obtained from the C language time() function by dividing its return value by 86400 and adding 40587. it is sufficient to simply save the raw phase or frequency data to a file. 100 . and can be usefully applied to the analysis of these data. purely numeric. provide security and robust data integrity. Timetags are highly desirable for frequency stability measurements. At the other extreme. multichannel clock measuring systems may require an elaborate database to store a large collection of data. For some purposes.g. and serve the archived data via a network. the number of days since noon on January 1. the MJD is the Julian Date .9. is non-ambiguous over a two-century (1900-2099) range. and is free from seasonal (daylight saving time) discontinuities Analysis software can easily convert the MJD into other formats such as year. The preferred timetag is the Modified Julian Date (MJD) expressed as a decimal fraction and referenced to UTC.HANDBOOK OF FREQUENCY STABILITY ANALYSIS 9. keep track of the clock identities and transactions. identifying it only by the file name. can have any required resolution. hour. month. day. It may also be necessary to integrate the clock data with other information (e.4000. minute and second. Timetags Timetags are often associated with phase or frequency data. 4713 BC.

Freq. D. 2. S. Symp.. "Picosecond Time Difference Measurement System". Edited by E. Proc. (formerly Timing Solutions Corp. S.A. 2. 9. 1987. Tutorial at 1993 IEEE Freq. S. Howe and L Erb. pp. Howe.R.Their Measurement and Characterization''. Feb. CO 80301 USA.A. 48. 7. June 1984. Proc.). Barnes.. 35th Annu. "Frequency and Time . J. Symp. Contrl. Allan. May 1981.A. Symmetricom. Symp. "Oscillator-Stability Analyzer Based on a Time-Tag Counter". pp.. 1985. A7 Frequency & Phase Comparator Measurment System. Chapter 12. on Freq. Telecommunications and Data Acquisition Progress Report 42-88. Levine. Precision Frequency Control. CA. 314-320. Proc. D. J. 4. 11. Data Sheet.A. pp. 38th Annu. Symp. Inc. Stein and G. 200-212. Boulder.. June 1. Totnes. Proc. D. Symmetricom. Model 5120A Phase Noise Test Set. Gray. "Performance of an Automated High Accuracy Phase Measurement System".W. Stein. 8.A. C. D. D. p. 3.191-416. Hilliard. NPO-20749. pp. CO 80301 USA. Sessions 3D and 4D.A. "Introduction to Robust Statistics and Data Filtering". Academic Press. Plymouth Road. Stein. Data Sheet. "Software for Two Automated Time Measurement Systems". Symp. Contrl.). Contrl.R. 29th Annu. Oct-Dec 1986. (formerly Timing Solutions Corp. 10. "Properties of Signal Sources and Measurement Methods''. Gernot M. Vol. Quartzlock (UK) Ltd. Data Sheet.W. ISBN 0-12-280602-6. NASA Tech Briefs. May 1975. Ballato.Glaze. 101 . Winkler. Greenhall. 1993. D. Jet Propulsion Laboratory. Gothic. June 1982.R. Greenhall. Model 5110A Time Interval Analyzer. May 2001. 5. 404-411. Contrl.. pp. pp. 36th Annu. 483-486. Freq. 147. Gifford. on Freq. "Frequency Stability Review". C.SECTION 9 MEASURING SYSTEMS References for Measuring Systems 1. Contrl. New York. Inc. TQ9 5LH England. 6. Pasadena. Gerber and A. Boulder. Devon. Allan and J.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS 102 .

large frequency offset and significant drift.099785679257264e-06 8. Visual inspection of data is an important preprocessing step! Analyst judgment may be needed for less obvious outliers.899732698242008e-06 Etc.199836723454638e-06 8. Plot 2 Convert the phase data to frequency data and examine it. the purpose of the analysis is to characterize the noise in the presence of an outlier. Step 1 Description Open and examine a phase data file.999873025741449e-07 1. Using simulated data for a high-stability rubidium frequency standard. the following example shows a typical analysis flow.399836191440859e-06 6.000000000000000e+00 8. Nevertheless. 103 .599873851209537e-06 4.699869215098003e-06 3.299833612216789e-06 7.499887627663997e-06 5.999774896024524e-06 9.799890526185009e-06 2. The phase data is just a ramp with slope corresponding to frequency offset. An obvious outlier exists that must be removed to continue the analysis.SECTION 10 ANALYSIS PROCEDURE 10 Analysis Procedure A frequency stability analysis can proceed along several paths. 0. as the circumstances dictate.

one should try to determine the cause of the outlier. A line shows a linear fit to the frequency data. The frequency spike of 1x10-9 corresponds to a phase step of 900 nsec over a single 900-second measurement interval. Remove the outlier. nine 10 MHz carrier cycles. Data taken at a higher rate would help to determine whether the anomaly happened instanttaneously or over some finite period. The resulting quadratic shape is due to the frequency drift. Timetags can help to relate the outlier to other external events.HANDBOOK OF FREQUENCY STABILITY ANALYSIS 3 In an actual analysis. One can just begin to see phase fluctuations around the quadratic fit to the phase data. 104 . The noise and drift are now visible. 3 4 Remove the frequency offset from the phase data. which appears to be quite appropriate.

The raw phase data will be essentially a straight 105 . Remember that frequency corresponds to the slope of the phase. remove the frequency drift from the frequency data of Step #3. 6 7 Convert the phase residuals to frequency residuals. and flicker FM noise at longer tau (t0 slope). however. both at the simulated levels shown in the annotations of the first plot.SECTION 10 ANALYSIS PROCEDURE 5 Remove the frequency drift.1. which is now clearly visible. Analyst judgment is needed to make the best choices. where the phase slope is the most negative. One exception. Other drift models may be more appropriate. Some experience is needed to interpret phase data like these. The results show white FM noise at short averaging times (τ-0. Alternatively. Data Precision There are relatively few numerical precision issues relating to the analysis of frequency stability data.5 slope) . There are subtle differences in removing the linear frequency drift as a quadratic fit to the phase data compared with removing it as a linear fit to the frequency data (different noise models apply). leaving the phase residuals for noise analysis. so the frequency is lowest near the end of the record. is phase data for a highly stable frequency source having a relatively large frequency offset. Perform a stability analysis using the overlapping Allan deviation. 10.

Their occurrence indicates that the statistics are not stationary. Phase data may be converted to frequency data. Gaps. while their numerical precision obviously depends upon the variable digits of the data set. Frequency drift may be removed from frequency data by subtracting a least-squares linear fit. Similar considerations apply to the quadratic phase term (linear frequency drift). In general. or available to fill gaps and to correct otherwise avoid such bad data points. called the bisection drift. the objective is to remove as much of the deterministic behavior as possible. jumps and outliers can occur frequency stability data. Usually the for outliers in a consistent manner. and it may be necessary to divide the data into portions and analyze them separately. which may have to be removed before doing an analysis to obtain meaningful results. or removal of outliers. Frequency jumps can also be a problem for stability analysis. The latter. The latter are particularly useful to describe the stabilization of a frequency source. Jumps and Outliers It is common to have gaps and outliers in a set of raw Gaps. Outliers can be replaced by gaps or filled with interpolated values. Frequency outliers are found by comparing each data point with the median value of the data set plus or minus some multiple of the absolute median deviation. A large number of digits must be used unless the frequency offset is removed by subtracting a linear term from the raw phase data. or by a least squares linear fit. An offset may be removed from frequency data by normalizing it to have an average value of zero. and vice versa. Missing or erroneous data in frequency measurements and they may occur due to power outages. there is little error propagation in forming the summary statistics. equipment must be handled before performing malfunctions and interference. which may require data averaging. Preprocessing Preprocessing of the measurement data is often necessary before the actual analysis is performed. Methods are may not be possible or practical to repeat the run. For long-term tests. 10. and it is particularly important to explain all phase discontinuities. Frequency offset may be removed from phase data by subtracting a line determined by the average of the first differences.HANDBOOK OF FREQUENCY STABILITY ANALYSIS line (representing the frequency offset). it a stability analysis. Frequency drift may be removed from phase data by a least squares or 3-point quadratic fit. by subtracting a line determined by the first differences or by calculating the drift from the difference between the two halves of the data. reason for the gap or outlier is known. obtaining random residuals for subsequent noise analysis. 106 . is equivalent to the three-point fit for phase data. Many frequency stability measures involve averages of first or second differences. Thus. These median statistics are more robust because they are insensitive to the size of the outliers.2.3. or by subtracting the average of the second differences. Plotting the data will often show the bad points. 10. and the instability information is contained in the small deviations from the line. Other more specialized log and diffusion models may also be used. Phase and frequency data can be combined for a longer averaging time. frequency offset and drift.

and gaps can then be inserted as placeholders to maintain equally spaced data. A zero value in fractional frequency data can also occur as the result of the conversion of two equal adjacent phase data points (perhaps because of limited measurement resolution). Uneven Spacing Unevenly spaced phase data can be handled if they have associated timetags by using the individual timetag spacing when converting it to frequency data. Similarly. For phase data. zero should be treated as valid data if it is the first or last point.g. and then using the two values immediately before and after any interior gaps to determine linearly interpolated values within the gap. 107 . in effect placing the frequency data on an average uniform grid. gaps and outliers can be avoided but not eliminated. 10. and the value should be adjusted to. and a value of zero (0) is often used to denote a gap. the data may be analyzed by using the average timetag spacing as the analysis tau. judgment is needed to assure a credible result. Some analysis processes can be performed with data having gaps by skipping over them. For example. 10. Missing points can be found by examining the timetags associated with the data. but other calculations cannot be. valid data having a value of zero can be replaced by some very small value (e. perhaps at some speed penalty.4. Analysis of Data with Gaps Care must be taken when analyzing the stability of data with missing points and/or gaps. and. if either of the two points involved in the first difference is a gap. by first removing any leading and trailing gaps. if the tau differences are reasonably small.6. outliers can be replaced with gaps for the same reason. if there are many of them. Then. in the calculation of the Allan variance for frequency data. tau variations of ±10 % will yield reasonable results as long as the exact intervals are used for the phase to frequency conversion. In these cases. however. that Allan variance pair is skipped in the summation. Many analysis functions can produce meaningful results for data having gaps by simply skip those points that involve a gap. For fractional frequency data. Gaps may be filled in phase or frequency data by replacing them with interpolated values. the results will be suspect. say.SECTION 10 ANALYSIS PROCEDURE Gaps and outliers can occur in clock data due to problems with the measuring system or the frequency source itself.5. 1e-99 to distinguish it from a gap. Like death and taxes. 10. 1e-99). Gap Handling Gaps should be included to maintain the proper implied time interval between measurements. It may be more prudent to simply analyze a gap-free portion of the data. It is therefore often necessary to replace the gaps with interpolated values. While completely random data spacing is not amenable to this process. Those points are not real data. These gaps can span multiple points.

while others have speed limitations. is compared with the median value of the data set. plus or minus the desired multiple of the MAD. m. Because two phase points are needed to determine each frequency point (as the difference between the phase values divided by their tau). and the factor 0. Analysis of phase data resulting from the conversion of frequency data having a large gap is not recommended. based on their deviation from the median of the data.7. Variance Analysis Variance analysis functions can include provisions for handling gaps. and (2) check the results against the normal Allan deviation (which has the simplest. using a deviation limit in terms of the median absolute deviation (a 5sigma limit is common). 10. Conversion from frequency to phase is more problematic because of the need to integrate the frequency data. Some of these functions yield satisfactory results in all cases.10. Filling gaps can often help for the total variances. the results will depend on coding details included in addition to the basic variance algorithm. which can cause a discontinuity in the phase record. The speed limitations are caused by more complex gap checking and frequency averaging algorithms. In all cases. Phase-Frequency Conversions Phase to frequency conversion is straightforward for data having gaps. and a gap of N phase points causes N+1 frequency gaps.6745 makes the MAD equal to the standard deviation for normally distributed data. which is then replaced 108 . while the poor results are associated with the total variances for which conversion to phase data is required. y(i).11. Two general rules apply for the variance analysis of data having large gaps: (1) use unconverted phase data. The average frequency value is used to calculate the phase during the gap. The latter is most apparent at longer averaging times where the averaging factor is comparable to the size of the gap.6745 }. 10. Drift Analysis Drift analysis functions generally perform well for data having gaps.m | / 0. While the definition of an outlier is somewhat a matter of judgment. fastest gap handling ability. Spectral Analysis Gap filling in spectral analysis functions can affect the low frequency portion of the spectrum. Outlier Recognition The median absolute deviation (MAD) is recommended as its means of outlier recognition. or provide unsatisfactory results for data having large gaps.8. defined as MAD = Median { | y(i) . a single phase gap will cause two frequency gaps. It is the median of the scaled absolute deviations of the data points from their median value. 10. Each frequency data point.9. it is important to find and remove such points in order to use the rest of the data. The MAD is a robust statistic based on the median of the data. where m = Median { y(i) }. This is a robust way to determine an outlier. 10.HANDBOOK OF FREQUENCY STABILITY ANALYSIS 10. provided that missing data are represented by gaps to maintain a regular time sequence.

Visual inspection can provide vital insight into the results. 3. which should be an adjunct to. An automatic outlier removal algorithm can iteratively apply this method to remove all outliers. there is essentially no difference between the two plotting methods. V. Chichester. It is important to explain all outliers. frequency reference. Tutorial at 1993 IEEE Freq. This presentation properly conveys the integral nature of the phase data. Symp.12.S. etc. 10. and the overlapping form provides better confidence than the original 109 . Phase data are plotted as line segments connecting the data points. and. 3rd Edition. As the density of the data points increases. Discrete phase jumps are a particular concern. Obviously. if they are related to the RF carrier frequency. Outliers in Statistical Data.R. may indicate a missing cycle or a problem with a digital divider. U. 2.. Winkler. Gernot M. Barnett and T. Missing data points are shown as gaps without lines connecting the adjacent points. ISBN 0-471-93094-6. Auxiliary information such as monitor data can be a big help in determining the cause of outliers. D. Jumps. or environmental control are often easier to identify if multiple devices are under test. "Use of Robust Statistical Techniques in Time Scale Formation". and is an important preprocessor before numerical analysis. Frequency data are plotted as horizontal lines between the frequency data points. References for Gaps. A plot also shows much about the validity of a curve fit. The overlapping Allan variance is recommended in most cases. Preliminary Report. thereby determining whether they are due to the measurement process or the device under test. while a common change in all measurement readings points to a reference problem.13. John Wiley & Sons. A log of all measurement system events should be kept to facilitate outlier identification. visual inspection of the data.. Sessions 3D and 4D. and mimics the analog record from a frequency counter.) that could account for the problem. The Allan variance is the most widely used time-domain stability measure. which can be a source of confusion.SECTION 10 ANALYSIS PROCEDURE by a gap. Naval Observatory Contract No. A phase jump will correspond to a frequency spike with a magnitude equal to the phase change divided by the measurement interval. and Outliers 1. Contrl. This shows the averaging time associated with the frequency measurement. equipment failures. a common gap in all measurement channels points to a failure of the measurement system. Variance Selection It is the user's responsibility to select an appropriate variance for the stability analysis. 1994. Percival. 10. June 1.B.g. "Introduction to Robust Statistics and Data Filtering". Failures of the measurement system. especially where the frequency drift is small or has been removed. 1982. Data Plotting Data plotting is often the most important step in the analysis of frequency stability. An important first step is to correlate the bad point with any external events (e. power outages. Such a frequency spike will produce a stability record that appears to have a (large magnitude) white FM noise characteristic. and not a substitute for. 1993. N70092-82-M-0579. Lewis.

and. 10. Three-Cornered Hat Any frequency stability measurement includes noise It is sometimes necessary to contributions from both the device under test and the determine the individual noise reference. The total and Thêo1 variance can be used for even better confidence at large averaging factors (but at the expense of longer computation time). again. Ideally. Normally. but TVAR is generally preferred. Or. The modified Allan variance is recommended to distinguish between white and flicker PM noise. while the latter provides tau multiples of ten). while MTIE measures the peak time deviations. The all tau option can be useful as a form of spectral analysis to detect cyclic disturbances (such as temperature cycling). In all cases. The individual variances may be determined by the expressions σ²a= ½ [ σ²ab + σ²ac . A doing so by using the results of special case is that of two identical units where half multiple measurements to estimate of the measured variance comes from each. the overlapping Hadamard variance is recommended over its normal form for analyzing stability in the presence of divergent noise or frequency drift.HANDBOOK OF FREQUENCY STABILITY ANALYSIS "normal" version.σ²ac ] σ²c= ½ [ σ²ac + σ²bc . Otherwise. a total form of it is available for better confidence at long tau. octave or decade spacing is used (the former has even spacing on a log-log plot. and the the variance of each source. it may be useful to employ the so-called "three-cornered hat" method for determining the variance of an individual source. The time variance provides a good measure of the time dispersion of a clock due to noise. measured deviation can be corrected for one unit by dividing it by √2. Methods exist for it can be removed by subtracting its variance. The choice of tau interval depends mainly on whether interference mechanisms are suspected that cause the stability to vary periodically. b and c whose variances add σ²ab= σ²a + σ²b σ²ac= σ²a + σ²c σ²bc= σ²b + σ²c . Given a set of three pairs of measurements for three independent frequency sources a. measurement. the results are reported in terms of the deviations.σ²bc ] σ²b= ½ [ σ²ab + σ²bc .14. TIE rms can also be used to assess clock performance. if the noise of the reference is known. the reference noise would be low contributions of the two sources that enough that its contribution to the measurement is contribute to a variance negligible.σ²ab ] . 110 . Finally.

441e-12 1.SECTION 10 ANALYSIS PROCEDURE Although useful for determining the individual stabilities of units having similar performance.147e-12 1.830e-12 2. An example of the use of the three-cornered hat function to correct stability data for reference noise is shown below.280e+02 257 5.275e-12 3. which contains the a priori reference stability values multiplied by √2. Negative variances are a sign that the method is failing (because it was based on insufficient measurement data. denoted as source A.203e-12 6.039e-13 111 .854e-12 1. denoted by B and C.663e-12 3. as shown in the following diagram: The Unit Under Test (UUT).923e-12 7.826e-12 5.600e+01 481 2. The three-cornered hat method should be used with discretion.. by identical stability data files A-B and A-C.000e+00 505 5. is measured against the reference.108e-12 9. The reference is measured against itself by stability data file B-C.367e-12 8.546e-12 1. to determine which unit is best).058e-12 3.400e+01 385 1. or for which there are insufficient data. The three sets of stability data should be measured simultaneously. the method may fail by producing negative variances for units that have widely differing stabilities. The three-cornered hat function may be used to correct a stability measurement for the noise contribution of the reference.000e+00 497 3.000e+00 509 7.157e-12 2.200e+01 449 1. It is best used for units having similar stability (e. and it is not a substitute for a low noise reference. This problem is most likely to arise at long tau.625e-13 4. The resulting stability data are shown in the following table.720e-12 2.448e-12 9.075e-12 4.g. Simulated overlapping Allan deviation stability data for the unit under test versus the reference were created by generating and analyzing 512 points of frequency data with tau = 1 second and σy(1 sec) = 1e-11.370e-12 2.520e-12 4. or because the units under test have disparate or correlated stability).000e+00 511 9. Stability Data for Unit Under Test versus Reference Tau # Sigma Min Sigma Max Sigma 1. if the units are correlated.025e-12 6.820e-13 7.269e-12 1.

Stability Data for Unit Under Test versus Reference Tau # Sigma Min Sigma Max Sigma 1.458e-13 6.HANDBOOK OF FREQUENCY STABILITY ANALYSIS A similar stability file is used for the reference.360e-13 1.135e-13 3.173e-13 1.368e-13 2.000e+00 511 1.200e+01 449 2.000e+00 509 1.854e-13 1.490e-12 1.257e-13 8.349e-13 4. Simulated overlapping Allan deviation stability data for the reference versus the reference was created by generating 512 points of frequency data with tau = 1 second and σy(1) = 1.681e-13 3.070e-12 4.058e-13 2.000e+00 505 7. the Allan deviations of the reference source are multiplied by √2.000e+00 497 5.637e-13 1.436e-12 1.070e-13 8.285e-13 The corrected UUT and reference stabilities are plotted below: 112 .550e-12 2.280e+02 257 1.631e-13 7.924e-13 6. Since it represents a measurement of the reference against itself.451e-13 2.025e-12 9.400e+01 385 1.846e-13 5.329e-13 1.152e-13 2.414e-12.600e+01 481 3.

one can easily write closed-form expressions for the separated variances from measurements of M clocks.SECTION 10 ANALYSIS PROCEDURE Here the reference stability is about 1x10-12τ-1/2. and the σ²ij are the measured Allan variances for clock i versus j at averaging time τ. Appropriate use of the three-cornered hat method to correct stability measurements for reference noise applies where the reference stability is between three to ten times better than that of the unit under test. Using σ2ii = 0 and σ2ij = σ2ji. The fraction of remaining degrees of freedom for unit i as a result of performing a threecornered hat instead of measuring against a perfect reference is given by: Γ = (2·σ4i) / (2·σ4i + σ²a·σ²b + σ²a·σ²c+ σ²b·σ²c) . and has questionable confidence for less than the former (and a better reference should be used). independent of the averaging time and noise type. 113 . Note that the B and C columns of corrected stability values both represent the reference source.4 for three units having the same stability. and the corrected UUT instability is slightly less than the uncorrected values. 12]. The three-cornered hat technique can be extended to M clocks (subject to the same restriction against negative variances) by using the expression σ 2i = where 1 M −2 LM σ NM∑ M j =1 M 2 ij −B OP QP . The ratio of the number of degrees of freedom is 0. B= 1 2 M −1 b LM σ OP g ∑ NM∑ QP M 2 kj k =1 j =1 . The error bars of the individual variances may be set using χ² statistics by first determining the reduced number of degrees of freedom associated with the three-cornered hat process [11. The correction is negligible for more the latter (see above).

HANDBOOK OF FREQUENCY STABILITY ANALYSIS

References for Three-Cornered Hat
1. 2. 3. The term "Three-cornered hat" was coined by J.E. Gray of NIST. J.E. Gray and D.W. Allan, "A Method for Estimating the Frequency Stability of an Individual Oscillator", Proceedings of the 28th Annual Symposium on Frequency Control, May 1974, pp. 243-246. S.R. Stein, "Frequency and Time - Their Measurement and Characterization'', Chapter 12, Section 12.1.9, Separating the Variances of the Oscillator and the Reference, pp. 216217, Precision Frequency Control , Vol. 2, Edited by E.A. Gerber and A. Ballato, Academic Press, New York, 1985, ISBN 0-12-280602-6. J. Groslambert, D. Fest, M. Oliver and J.J. Gagnepain, "Characterization of Frequency Fluctuations by Crosscorrelations and by Using Three or More Oscillators", Proceedings of the 35th Annual Frequency Control Symposium, May 1981, pp. 458-462. P. Tavella and A. Premoli, "A Revisited Tree-Cornered Hat Method for Estimating Frequency Standard Instability", IEEE Transactions on Instrumentation and Measurement, IM-42, February 1993, pp. 7-13. P. Tavella and A. Premoli, "Characterization of Frequency Standard Instability by Estimation of their Covariance Matrix", Proceedings of the 23rd Annual Precise Time and Time Interval (PTTI) Applications and Planning Meeting, December 1991, pp. 265276. P. Tavella and A. Premoli, "Estimation of Instabilities of N Clocks by Measuring Differences of their Readings", Metrologia, Vol. 30, No. 5, 1993, pp. 479-486. F. Torcaso, C.R. Ekstrom, E.A. Burt and D.N. Matsakis, "Estimating Frequency Stability and Cross-Correlations", Proceedings of the 30th Annual Precise Time and Time Interval (PTTI) Systems and Applications Meeting, December 1998, pp. 69-82. F. Torcaso, C. Ekstrom, E. Burt and D. Matsakis, “Estimating the stability of N clocks with correlations,” IEEE Trans. Instrum. Meas. 47(5), 1183, 2000. C. Audoin and B. Guinot, The Measurement of Time, Cambridge University Press, ISBN 0-521-00397-0, 2001, Section 5.2.8. C. Ekstrom and P. Koppang, "Three-Cornered Hats and Degrees of Freedom", Proceedings of the 33rd Annual Precise Time and Time Interval (PTTI) Systems and Applications Meeting, November 2001, pp. 425-430. C.R. Ekstrom and P.A. Koppang, "Error Bars for Three-Cornered Hats", IEEE Trans. UFFC, Vol. 53, No. 5, pp. 876-879, May 2006. J.A. Barnes, The Analysis of Frequency and Time Data, Austron, Inc. May 1992.

4. 5. 6.

7. 8. 9. 10. 11. 12. 13.

10.15. Reporting
The results of a stability analysis are usually presented as a combination of textual, tabular and graphic forms. The text describes the device under test, the test setup, and the methodology of the data preprocessing and analysis, and summarizes the results. The assumptions and analysis choices should be documented so that the results could be reproduced. The report often includes a table of the stability statistics. Graphical presentation of the data at each stage of the analysis is generally the most important aspect of presenting the results. For example, these are often a series of plots showing the phase and frequency data with an aging fit, phase and frequency residuals with the aging removed, and stability plots with noise fits and error bars. Plot titles, subtitles, annotations and inserts can be used to clarify and emphasize the data presentation. The results of several stability runs can be combined, possibly along with specification limits, into a single composite plot. The various elements can be combined into a single electronic document for easy printing and transmittal. 114

SECTION 11

CASE STUDIES

11 Case Studies
This section contains several case studies to further illustrate methodologies of frequency stability analysis.

11.1. Flicker Floor of a Cesium Frequency Standard
The purpose of this analysis is to examine the flicker floor of a commercial cesium beam tube frequency standard. An instrument of this kind can be expected to display a white FM noise characteristic out to long averaging times. At some point, however, the unit will typically “flicker out”, with its stability plot flattening out to a flicker FM noise characteristic, beyond which its stability will no longer improve. Determining this point requires a lengthy and expensive run, and it is worthwhile to use an analytical method that provides the best information at long averaging factors. The effort of a more elaborate analysis is far easier than extending the measurement time by weeks or months. This is the five-month frequency record for the unit under test. It has already been “cleaned up” for any missing points, putting in gaps as required to provide a timecontinuous data set. The data look very “white” (see Section 6.1), the frequency offset is very small (+7.2x10-14), and there is no apparent drift (+1.4x10-16/day). Overall, this appears to be a very good record. An overlapping Allan deviation plot shows a white FM noise level of about 8.2x10-12τ-1/2 out to about 5 days, where the stability levels off at about 1.2x10-14. While this is very respectable behavior, one wonders what the stability actually is at the longer averaging times. But to gain meaningful confidence using ADEV there, the run would have to be extended by several months.

115

HANDBOOK OF FREQUENCY STABILITY ANALYSIS

The total deviation can provide better confidence at the longer averaging times. It seems to indicate that the stability continues to improve past 10 days, where is drops below 1x10-14. But the results are not conclusive.

Thêo1 can provide even better long-term information, at the expense of a longer calculation time. This Thêo1 plot seems to show even more clearly that the stability continues to improve at longer averaging times, well into the pp1015 range. It assumes white FM noise (no Thêo1 bias removal).

A ThêoH analysis with automatic bias removal, which combines the AVAR and ThêoBR statistics, requires an unacceptably long time for the original data set. By averaging the data by a factor of 5 (τ = 500 sec), the analysis can be performed in several hours. The results are essentially the same as for Thêo1 above. The Thêo1 bias is 1.676, intermediate between white and flicker FM noises for medium to large averaging factors. One can conclude that this cesium beam frequency standard reaches a stability slightly better than 1x10-14 at an averaging time on the order of 1 month.

116

about 1x10-12 at 5 days). plus significant frequency drift. 117 . If the linear frequency drift is removed before performing the AVAR stability analysis.SECTION 11 CASE STUDIES 11.. the stability plot shows a white FM noise (τ-1/2) characteristic changing to flicker FM noise (t0) at longer averaging times. If an Allan variance analysis is performed directly on these data without drift removal.e. In particular. It is usually best to use a stability plot only to show the noise. and analyzing and removing the drift separately.2. the stability at the longer averaging times is degraded. the stability plot has a τ+1 slope seconds that beyond 105 corresponds to the drift (i. Hadamard Variance of a Source with Drift These frequency data simulate a typical rubidium frequency standard (RFS) with a combination of white and flicker FM noise.

The HDEV results are essentially identical to those of the drift-removed ADEV. White PM Flicker PM Examination of the phase data is a good first step. This can be more convenient. An experienced analyst will immediately notice the difference between the white and flicker noise plots. especially when analyzing a combination of sources having differing amounts of drift (e.HANDBOOK OF FREQUENCY STABILITY ANALYSIS The Hadamard variance is insensitive to linear frequency drift. the frequency data shows little noise type discrimination.3. Phase Noise Identification Consider the problem of identifying the dominant type of phase noise in a frequency distribution amplifier. Examination of the frequency data would be appropriate to distinguish between white and flicker FM noise. 118 . In contrast. It can therefore be used to perform a stability analysis on the original data without first having to remove the drift. but will find it harder to quantify them. because the differencing process whitens both. 11. cesium and rubidium units). How should these data be analyzed to best determine the noise type? Simulated white and flicker PM noise at equal levels are analyzed below in several ways to demonstrate their ability to perform this noise identification.. Assume that time domain stability measurements have been made comparing the amplifier’s input and output signals. however.g.

93. respectively.0. That ID method is quite effective even for mixed noise types.5 and –1. for which the slopes are –1. is able to distinguish between white and flicker PM noise. Even better discrimination is possible with the autocorrelation function. and is able to quantitatively estimate the power law noise exponents as +1.99 and +0. respectively. 119 . The modified Allan deviation.006 and 0.SECTION 11 CASE STUDIES The Allan deviation is not able to distinguish between white and flicker PM noise. Both have slopes of about 1.780 for these white and flicker PM noise data. as shown by the superimposed noise fit lines. The lag 1 ACF is 0. by virtue of its additional phase averaging.

Detection of Periodic Components Frequency stability data can contain periodic fluctuations due to external interference. The interference level is too low to be visible in the full frequency data plot. and it can therefore be necessary to detect discrete spectral components when performing a stability analysis. Several analytical methods are used to detect this periodic component.000 points of τ= 1 second simulated white FM noise at a level of 1x10-11τ-1/2 that contains sinusoidal interference with a period of 500 seconds at a peak level of 1x10-12.4. This example uses a set of 50. environmental effects. or the source itself.HANDBOOK OF FREQUENCY STABILITY ANALYSIS 11. The latter simulates interference that might occur as the result of air conditioner cycling. 120 .

The interference is clearly visible in the power spectral density (PSD). Here the stability is equal to the underlying white FM noise level. The interference is quite visible in an “all tau” stability plot as a null that occurs first at an averaging time of 500 seconds (the period of the interference). there is just a hint of the interference (10 cycles over 5000 data points). which has a bright component at 2 mHz. 121 .SECTION 11 CASE STUDIES By zooming in. corresponding to the 500 sec period of the interference.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS The interference is less visible in an autocorrelation plot. and have a τ-1 envelope set by the vibrational phase modulation. It is a good analysis policy to examine the power spectral density when periodic fluctuations are visible on a stability plot and periodic interference is suspected. Nulls in the Allan deviation occur at averaging times equal to the multiples of the 20 Hz sinusoidal modulation period. 122 . The PSD has equivalent information.5. where the stability is determined by the white PM noise. Peaks in the Allan deviation occur at the modulation half cycles. Stability Under Vibrational Modulation This plot shows the stability of an oscillator with a combination of white PM noise and a sinusoidal component simulating vibrational modulation. 11. but its log scale makes low-level components more apparent. where cyclic behavior is barely noticeable with a 500-lag period.

but some of the records are quieter than others in the longer term. Reference: C. In particular. Greenhall. Thus adding a single large (say 106) central outlier to the 1000point test suite of section 12. CA. A diagonal slope downward across the plot corresponds to an aging of about –2. providing a quick comparison of their behaviors. all plotted with the same scales. still contain enough detail to allow subtle comparisons very quickly.6. The eye can easily see the similarities and differences. A figure like this immediately shows that (a) all the units have negative frequency aging of about the same magnitude that stabilizes in about the same way. could be used to provide additional information. All the data have τ = 900 seconds (15 minutes). (b) there are occasional gaps in some of the records. Pasadena.A. Plots for 40 units are shown with their serial numbers. far better than a set of numbers or bar graphs would do. which would be harder to do using a set of larger plots on separate pages. "Frequency Stability Review". although small. Jet Propulsion Laboratory. Closer inspection of even these small plots can reveal a lot of quantitative information if one knows the scale factors.5x10-11 (1x10-12/div). The following figure shows the stabilization period of a production lot of rubidium clocks. The plots. as shown in this stability plot. 1987. (d) some of the units take longer than others to reach a certain aging slope. Color coding. and can immediately select units based on some criterion. although not used here. 123 .SECTION 11 CASE STUDIES 11. 11. White FM Noise of a Frequency Spike The Allan deviation of a frequency record having a large spike (a phase step) has a τ-1/2 characteristic [1].16386e4. Telecommunications and Data Acquisition Progress Report 42-88. Feb.7.4x10-13/day. Composite Aging Plots A composite plot showing the aging of a population of frequency sources can be an effective way to present this information visually. (c) all of the units have about the same short-term noise. These plots are inspired by Edward Tufte’s book The Visual Display of Quantitative Information. these plots all have a full x-scale time range of nine weeks (1 week/div) and a full y-scale frequency range of 1.4 will give a data set with σy(τ) = [(106)⋅2/(1000-1)]1/2 = 3.

ISBN 9780961392109. 124 .R. Tufte. The Visual Display of Quantitative Information. Graphic Press. 1983.HANDBOOK OF FREQUENCY STABILITY ANALYSIS 001 002 003 004 005 006 007 008 009 010 011 012 013 014 015 016 017 018 019 020 021 022 023 024 025 026 027 028 029 030 031 032 033 034 035 036 037 038 039 040 Reference: E.

There is a continuing need to validate the custom software used to analyze time domain frequency stability. general purpose mathematical and spreadsheet programs such as MathCAD®. The standard and normal Allan variances should be approximately equal for white FM noise. 3. Software Validation Considerable effort is needed to ensure that the results obtained from frequency stability analysis software are correct. Consistency Checks: The new program should be verified for internal consistency. plotted. The overlapping method provides better confidence of the stability estimates. Copies of those test data are available on-line [3]. 2]. Published Results: The results of a published analysis can be compared with the new program output. Other Programs: The results obtained from other specialized stability analysis programs (such as that from a previous generation computer or operating system) can be compared with the new program output. 125 . General Purpose Programs: The results obtained from industry standard. analyzed. 2. such as producing the same stability results from phase and frequency data. For other averaging factors. Several methods are available to validate frequency stability analysis software: 1. analyzed. Manual Analysis: The results obtained by manual analysis of small data sets (such as in NBS Monograph 140Annex 8. 6. Mature commercially available software should be used whenever possible instead of developing custom software. Known values of frequency offset and drift can be inserted. operates on current computer hardware and operating systems.SECTION 12 SOFTWARE 12 Software Software is necessary to perform a frequency Specialized software is needed to stability analysis. Several suggested validation methods are shown. and large data sets needed to interactively perform and document a complete stability analysis. includes the latest analytical techniques.1. 5. 4. The normal and modified Allan variances should be identical for an averaging factor of 1. and the normal and overlapping Allan variances should be approximately equal. because those calculations perform a frequency stability generally involve complex specialized algorithms analysis. The various methods of drift removal should yield similar results. One important validation method is comparison of the program results against a test suite such as the one in References [1. Matlab® and Excel® can be compared with the new program output. 12. User feedback and peer review is important.E) can be compared with the new program output. and has been validated to produce the correct results. the modified Allan variance should be approximately one-half the normal Allan variance for white FM noise. Simulated Data: Simulated clock data can also serve as a useful cross check. and the methods listed below can help ensure that correct results are obtained. This is always good to do to get a “feel” for the process. and removed. It is convenient to use an integrated software package that combines all of the required analytical tools. and modeled. Known values of power-law noise can be generated.

9770 91. NBS Monograph 140.8082 85. This frequency data is also normalized to zero mean by subtracting the average value.0000 Table II. A small data set is also an advantage for finding "off-by-one'' errors where an array index or some other integer variable is slightly wrong and hard to detect in a larger data set.00000 103.8889 100.22222 157.0 657.80607 91. Annex 8.78849 86.25 102.11111 9. Annex 8.E of NBS Monograph 140 shown in Table I.50203 43.11111 20.33333 -2. and then integrated to obtain phase values.22945 91.22945 91. a very useful starting point to verify frequency stability calculations. since a small data set can easily be entered and analyzed manually. Test Suites The following tables summarize the values for several common frequency stability measures for both the classic NBS data set and a 1000-point portable test suite.33333 166.88889 -144.35831 116. Table I. nevertheless.95287 74.HANDBOOK OF FREQUENCY STABILITY ANALYSIS 12.11111 34. NBS Monograph 140.56794 .11111 114.5 -2. While nine data points are not sufficient to calculate large frequency averages.2.7980 85.22945 52.83606 87.3.20000 839.E Test Data # 1 2 3 4 5 6 7 8 9 10 Frequency 892 809 823 798 671 644 883 903 677 Normalized Frequency 103.61487 91. Those numbers were used as an early example of an Allan variance calculation.8889 809 -10. 12.59112 2 4 893. A listing of the properties of this data set is shown in Table II.22222 111.11111 123.90379 75.5 802.67135 70.875 830.6039 115.55555 -96.E Test Data Statistics Averaging Factor # Data Points Maximum Minimum Average Median Linear Slope Intercept Standard Deviation [1] Normal Allan Deviation Overlapping Allan Dev Modified Allan Dev Time Deviation Hadamard Deviation Overlap Hadamard Dev Hadamard Total Dev Total Deviation Modified Total Dev Time Total Deviation 126 1 9 903 644 788.11111 -117.80607 70.88889 0.16396 93.44444 48.80607 70.22945 75. they are.55 809. NBS Data Set A "classic'' suite of frequency stability test data is the set of nine 3-digit numbers from Annex 8.88889 Phase (τ=1) 0.11111 -111.88889 94.

052754e-01 9.577220e-03 9.897804e-02 3.4.133666e-04 3. n3 = 633705974.965736e-02 9.910861e-02 3. 231-1.943883e-01 2.687202e-01 2.897745e-01 4.237638e-02 3.172376e-02 3.563623e-01 1.570469e-01 5. or by the Box-Muller method [5].056376e-03 4.371760e-03 4. c= Slope at end Standard Dev [5] Normal Allan Dev [6] Overlap Allan Dev [8] Mod Allan Dev [7.489368e-01 4. 1000-Point Test Suite The larger frequency data test suite used here consists of 1000 pseudo-random frequency data points. Thus the normalized value of n0 is 0.5748904732.418528e-01 1.104214e-04 1.648320e-04 5.571498e-06 2.957453e-01 1.442759e-01 7.631172e-01 5.248477e-03 4. a= y(t)=a·ln (bt+1)+c.798849e-01 6. These numbers may be divided by 2147483647 to obtain a set of normalized floating-point test data ranging from 0 to 1. avoids a collapse to zero).011791e-03 7. A spreadsheet program is a convenient and reasonably universal way to generate these data. frequency offset and/or drift terms may be added to the data.4].170921e-02 1.SECTION 12 SOFTWARE Table II Note: [1] Sample (not population) standard deviation.594973e+00 4.159953e-02 6. n2 = 1209410747. 1000-Point Frequency Data Set Averaging Factor # Data Points Maximum Minimum Average [1] Median Linear Slope [2. this simpler data set is adequate for software validation.253382e-00 3.138988e-03 1.614787e-02 9. Table III.922319e-01 2. It is produced by the following prime modulus linear congruential random number generator: ni+1 = 16807 ni Mod 2147483647 This expression produces a series of pseudo-random integers ranging in value from 0 to 2147483646 (the prime modulus.581083e-02 9. Similarly. These conversions can also be done by a spreadsheet program.104214e-05 9. 12. etc.737500e-01 4.206656e-02 3.134743e-02 6.897745e-01 5.922319e-01 2.003371e-01 2.499161e-02 3.420429e+02 4.241343e-02 2.3] Intercept [3] Bisection Slope [2] 1st Diff Slope [2] Log Fit [2. yielding a set of 1001 phase data points.287774e-02 1. and could be produced by summing a number of independent uniformly distributed data sets.047888e-01 5.884664e-01 2.943883e-01 2.396338e-01 10 100 7.533354e-01 4. While a data set with a normal (Gaussian) distribution would be more realistic.490910e-06 4.922319e-01 1.8] Time Deviation [8] Hadamard Deviation Overlap Had Dev Hadamard Total Dev Total Deviation Modified Total Dev Time Total Deviation 1 1000 9.839644e-01 -6. b= y'(t)=ab/(bt+1).058103e-02 3.545924e-01 4. it produces the sequence n1 = 395529916. When started with the seed n0 = 1234567890.865258e-01 -6.897745e-01 4. The values of this data set will be uniformly distributed between 0 and 1.517561e-04 5.752293e-01 100 10 5.867547e-01 -6.237080e-05 9.406530e-02 2. The frequency data may be converted to phase data by assuming an averaging time of 1.296352e-02 9.807261e-01 1.104214e-06 1.979804e-05 4.320847e-00 127 .943883e-01 2.922319e-01 2.

Applies only to F PM noise. [5] Theoretical R(n) for W FM noise = 0.06 Min σy(τ)= 5.224944e-02 [8] 69.842. [4] Exact results will depend on iterative algorithm used.035201e-01[8] 114. noise-independent CI estimate =σy(τ)/√N=1. [6] Expected value equal to standard deviation for white FM noise.45 Min σy(τ)= 8. [8] Double-sided 95 % confidence interval: p = 0.177 6. 12.158 and 0.34 Max σy(τ)=7. B1=0. R(n)=0.001594e-02. Annex C.77 RW FM -2 0.172376e-02 W FM[5].99 W FM 0 0. [8] Calculated with listed averaging factors from the basic τ = 1 data set. [2] Simple. Annex C. Table IV.2886751.044412e-02 [7] 72.000 for W FM noise and 0. Data not suited to log fit.870 9. [3] This CI includes κ(α) factor that depends on noise type: Noise α κ(α) W PM 2 0. Annex C.667 for F and W PM noise.4 has an example of a total deviation calculation.950.025 and 0.159953e-02 W FM # χ² df=146. and Annex D contains examples of confidence interval calculations.87 F FM -1 0.99 F PM 1 0.3] ±1σ CI.620 Conf Interval Remarks CI= 8. Expected value = 1/√12 = 0.07 Max σy(τ)= 1.2 contains an example of an Allan deviation calculation.419961e-02 [8] 122. Error Bars for n=1000 Point =1 Data Set with Avg Factor=10 Allan Dev Type #Pts Normal 99 Overlapping 981 Modified [4] 972 Sigma Value Noise Type & Ratio 9. 128 .500 and 0. IEEE Standard 1139-1999 IEEE Standard 1139-1999. [2] All slopes are per interval. [6] χ² for 95% CF Max σy(τ)=1.5.975. [5] Sample (not population) standard deviation.HANDBOOK OF FREQUENCY STABILITY ANALYSIS Table III Notes: [1] Expected value = 0.223942e-02 [8] 181. IEEE Standard Definitions of Physical Quantities for Fundamental Frequency and Time Metrology – Random Instabilities contains several example of stability calculations.014923e-01 [7] 119.262 for F PM noise.75 [4] BW factor 2πfhτ0 = 10.384 # χ² df=94. [7] Equal to normal Allan deviation for averaging factor = 1.965736e-02 W FM [1].64 Max σy(τ)= 7.713870e-03 [2. [6] Double-sided 68 % confidence interval: p = 0.71 Table IV Notes: [1] Theoretical B1=1. [3] Least squares linear fit.3 has an example of a modified Allan deviation calculation. [7] Single-sided 95 % confidence interval: p = 0.5.

B.com. Flannery. Hamilton Technical Services. 3. 2. Press. 1995 IEEE Freq. www. Symp. "IEEE Standard Definitions of Physical Quantities for Fundamental Frequency and Time Metrology . "Addendum to a Test Suite for the Calculation of Time Domain Frequency Stability". July 1999. 880-882. 1988. Riley. Vetterling. 1996 IEEE Freq.wriley... SC 29907 USA. Press.H.T. 5.A. W. June 1996. Teukolsky and W. S. Stable32 Software for Frequency Stability Analysis. 4. U. IEEE Std 1139-1999.K. 129 . Proc.J. W. pp. Symp. Riley. W. Beaufort.J.Random Instabilities". "A Test Suite for the Calculation of Time Domain Frequency Stability".. ISBN 0-521-35465-X. Numerical Recipes in C. Cambridge. Cambridge Univ.SECTION 12 SOFTWARE References for Software 1. June 1995. Contrl. 360-366. Proc.P. Contrl. pp.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS 130 .

It uses the second differences of the fractional frequencies. The uncertainty associated with a measurement. Valid when noise power is much smaller than the carrier power. See Tau. To remove the average value from phase or frequency data . the number of days since noon on January 1. A set of fractional frequency values. Bandwidth. £(f) = 10·log[½ · Sφ(f)]. 4713 BC. The maximum time interval error of a clock.4000. 131 £(f) MJD Modified Sigma MTIE Normalize Phase Data . HVAR. A modified version of the Allan or total variance that uses phase averaging to distinguish between white and flicker PM noise processes. A set of time deviates. hertz. where i denotes equally-spaced time samples.SECTION 13 GLOSSARY 13 Glossary The following terms are used in the field of frequency stability analysis Aging Allan Variance Averaging Averaging Time BW Confidence Limit Drift Frequency Data Hadamard Variance The change in frequency with time due to internal effects within the device. The MJD is the Julian Date . x[i] with units of seconds.000. dBc/Hz. the ratio of the SSB phase noise power in a 1 Hz BW to the total carrier power. and is unaffected by linear frequency drift. A three-sample variance. Often a 68 % confidence level or error bar. y[i]. where i denotes equally-spaced time samples. The Modified Julian Date is based on the astronomical Julian Date. The two-sample variance σ²y(τ) commonly used to measure frequency stability. that is similar to the twosample Allan variance. The process of combining phase or frequency data into samples at a longer averaging time.2. Called “phase” data to distinguish them from the independent time variable.5. The change in frequency with time due to all effects (including aging and environmental sensitivity).

The one-sided spectral density of the time deviations. often the twosample or Allan deviation. sec²/Hz. y(t) = [ν(t)−ν0]/ν0] = x'(t). A two-sample variance similar to the Allan variance with improved confidence at large averaging factors. The instantaneous time deviation from a nominal time. hertz. This dependent time variable is often called "phase" to distinguish it from the independent time variable t. The one-sided spectral density of the phase deviations. See Tau. where ν0 is the nominal frequency. Theoretical variance #1. The one-sided spectral density of the fractional frequency deviations. Single sideband. A biased removed version of Thêo1. σy(τ). where ν0 is the nominal frequency Total Variance x(t) y(t) 132 . The time interval error of a clock. rad²/Hz. x(t) = φ(t)/2πν0. The change in frequency per tau interval.HANDBOOK OF FREQUENCY STABILITY ANALYSIS Phase Noise Sampling Time Sigma Slope SSB Sf(f) Sx(f) Sy(f) Tau Total Thêo1 ThêoBR ThêoH TIE The spectral density of the phase deviations. A variance using an extended data set that provides better confidence at long averaging times. Can be expressed as the rms time interval error TIE rms or the maximum time interval error MTIE. seconds. A hybrid combination of ThêoBR and the Allan variance. a variance providing stability data out to 75 % of the record length. The square root or deviation of a variance. The instantaneous fractional frequency deviation from a nominal frequency. The interval between phase measurements or the averaging time used for a frequency measurement. 1/Hz.

July 1999.SECTION 13 GLOSSARY References for Glossary 1. "IEEE Standard Definitions of Physical Quantities for Fundamental Frequency and Time Metrology .Random Instabilities". IEEE Std 1139-1999. Glossary of Time and Frequency Terms issued by Comite Consultatif International de Radio Communication – International Telecommunications (CCITT) Union. Geneva. 133 . 2. Switzerland.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS 134 .

Blair (Editor). Contrl. 12. "Characterization of Clocks and Oscillators". IEEE. Academic Press.Their Measurement and Characterization''. 5. 647-654. G. 35th Annu.A.. IEEE Press. Allan and J. 2.F. These references are arranged by topic and sorted by date. 247-291. "Properties of Signal Sources and Measurement Methods''. "A Brief Review of Frequency Stability Measures". Dec. Jet Propulsion Laboratory. Proc. 1985.A. Nov. Only the 1990 and later PTTI proceedings are now available on-line. 5. “Characterization of Frequency Stability”. 42nd Annu..pl. Vol. UFFC-34. Department of Commerce. "Time and Frequency: Theory and Fundamentals". 31th Annu. 16. May 1974. D. 1978. Telecommunications and Data Acquisition Progress Report 42-88. “Atomic Timekeeping and the Statistics of Precision Signal Generators. 105-120.A. ISBN 0-12-280602-6. Jan.gov/cgi-bin/showpubs.W. Barnes. Hamilton Technical Services papers available on-line at http://www. Chapter 12. J. June. Symp.W.Lesage and C. Edited by E. 11. Proc. June 1977. Vol. 6. June 1967. Proc. pp. 135 • . "Oscillator Specifications: A Review of Classical and New Ideas".” IEEE Proceedings. 13. Freq. and Prediction of Precision Clocks and Oscillators”.A. All of the FCS and UFFC references are available on-line for IEEE UFFC members. Vol. IEEE Trans. Allan. 4. March 1990. Symp. 521-539. 6. 4. 207-219. Vol. Control Symp. Kroupa (Editor). May 1981.com. Ballato. 54. "Standard Terminology for Fundamental Frequency and Time Metrology". Also available on line at NIST web site. 1990.W. Proc. D.A. Oct-Dec 1986.BIBLUOGRAPHY 14 Bibliography • Notes 1. Ultrasonics. Barnes.191-416. pp.E. 489-527. Audoin.L.R. 7. C. vol. 1-47. Allan. 1987. pp. pp. pp. 1966. D. 18. Stein. 2. Institute of Electrical and Electronic Engineers. IEEE. 10th PTTI Meeting. 15. NBS Monograph 140.nist. Proc. P. U. National Bureau of Standards. pp. None of the EFTF or older IEEE I&M proceedings are available on-line. 2. J. Rutman. Class Notes. Estimation. et.W Allan. D.Walls (Editors). pp. Nov. 1986. pp. General 1. al.A. 1976. 10. 54. "Frequency and Time . IEEE Trans. V. S. Meas. pp. Frequency Stability: Fundamentals and Measurement. Department of Commerce. Howe and F. 17.S. 74. 1966. J. “Time and Frequency (Time-Domain) Characterization.. et al.wriley. 4. CA. Barnes. 1983. Vol. 9. J. "Frequency Stability Review". Special Issue on Frequency Stability. Proc. Tetu. 1979. Vanier and M. Vol. 8th PTTI Meeting. on Freq. Gerber and A. Radio Science. “Characterization and Measurement of Time and Frequency Stability”. Allan. 8. IM-20. Pasadena.291-310. Winkler.S.. Feb. NIST Technical Note 1337.. 55.. D. 14. Proc. Howe.B Sullivan. Proc. on Freq. National Institute of Standards and Technology. B. Feb. 6. NIST papers are available on-line at http://tf. May 1971. “Characterization of Precision Clocks and Oscillators”. No. Greenhall. Contrl. Nov.A. ISBN 087942-171-1. D.W. pp. Ferroelectrics and Freq. Contrl. D.. 2. 14. 419-425. Precision Frequency Control. D. NRL Workshop. 1987. Vol. Instrum. 3. U. IEEE. New York. New York. 3. No. No. 1988. "Time Domain Measurement of Frequency Stability".

IM-47. et al. 7. • Standards and Specifications 1. International Radio Consultative Committee (C. IEEE Std 1139-1999. Guide to the Expression of Uncertainty in Measurement.A. pp.. J. Freq. Crystal. IEEE. Sydnor (Editor). "Characterization of Frequency Stability". MIL-PRF-55310. 1991. 21. 22. D. 1990. Proc. W. 2. June 1997. NBS Technical Note 394. Proc. D. Hamilton Technical Services. “Spectral Ambiguity of Allan Variance”. Tutorial at the 2003 Intl. IEEE Trans.I. Hamilton Technical Services.W. G.W. Symp.HANDBOOK OF FREQUENCY STABILITY ANALYSIS 19. W. 142-150. Oct. Freq. Instrum. NASA SP-80.W. 20. Allan. Report 580. “Variances Based on Data with Dead Time Between the Measurements”. Vol. "Allan Variance". Austron. June 1998. Vol. Barnes. Barnes. Dec. July 1991. 27. 5.R. Jet Propulsion Laboratory. W. "The Generalized Autocovariance: A Tool for Clock Statistics". San Francisco: Holden-Day. Classic (Pre-Allan variance) 1. IEEE-NASA Symp. IEEE. NIST Technical Note 1318. Cont. D. on the Definition and Measurement of Short-Term Frequency Stability. J. R. Vol. J. 28.J. Greenhall. 5. Allan. Allan. Pasadena.S Department of Commerce. Greenhall. • • 136 . "The Calculation of Time Domain Frequency Stability”. "IEEE Standard Definitions of Physical Quantities for Fundamental Frequency and Time Metrology . Vol. "The Basics of Frequency Stability Analysis". Audoin and B. U. 2001 25. Greenhall. C. 24. C. 2. "The Statistics of Atomic Frequency Standards''. Jenkins.J. IM-20. 79. 4. Allan Variance 1. July 1999.). Oscillators.P. Riley. Instrum. March 1991. "Time and Frequency Metrology: Current Status and Future Considerations''. C. C. 1995. pp. and Time Forum. 1997 Intl. "Does Allan Variance Determine the Spectrum?".J. Nov. pp. Guinot. 79. "Characterization of Frequency and Phase Noise". pp. Riley. pp. 3. 1964. Inc. 623-627. National Bureau of Standards. 2. IEEE. 1-9. May 2003. Cont. "Characterization of Frequency Stability in Precision Frequency Sources". 4.S. "The Analysis of Frequency and Time Data".L. Proc. International Standards Organization. 23.A. Barnes.C. May 1999. 5th European Freq.A. 221-230. ITUR Handbook. Allan's TIME. 1986. Symp. 54.A. Box and G. Proc. L.A. “The Selection and Use of Precise Frequency Systems”. Meas. IEEE Trans. Telecommunications and Mission Operations Progress Report. 1995. 105-120. 6. pp. “Techniques for Frequency Stability Analysis”.. No. U. The Measurement of Time. 3. CA. J. Rutman and F. 42-137. General Specification For. 2.. Feb. 1970. Special Issue on Time and Frequency.Random Instabilities". Vol. Proc.M. 3. Walls. National Institute of Standards and Technology. Department of Commerce. Riley.A. Proc. "Characterization of Frequency Stability". Cambridge University Press. 26.. ISBN 92-67-10188-9.. Time Series Analysis: Forecasting and Control. No. 358-365. 1970. Meas. May 1971. July 1991. No. 1966. Vol.E.

Instrum.J.A.A. Proc. “Characterization of Frequency Stability: Analysis of the Modified Allan Variance and Properties of Its Estimate”. 201-213. pp. C. Meas. Davis. D. Contrl. 225-236. 10. "Hadamard Transform Image Coding". 4. Allan. Lesage and T. D. M. Kane and H. 2000. pp. "A Modified Allan Variance with Increased Oscillator Characterization Ability".C.K. 3. 3. des telecom. Andrews.. C. M. Hutsell. Proc. IEEE Trans. June 1989. pp. Dec. 346-353. "Processus a Accroissements Stationnaires". Contrl. "A Frequency-Domain View of TimeDomain Characterization of Clocks and Time and Frequency Distribution Systems''. P. Barnes. Contrl. Proc. "Estimating the Modified Allan Variance". 1995 IEEE Freq.38-67. N. French mathematician. Greenhall. D. No. IM-33. 3. S. No. Contrl. 7-8. Howe. E. March 1981. May 1991. Tome 30. et al. pp.. Symp. Proc. Vol. July-Aug. Weiss.. D. “The Hadamard Variance”. Symp. Baugh. Allan and H. pp. "Frequency Modulation Analysis with the Hadamard Variance". Allan.W. 25th Annu. Meas. April 1971.A. 4. "New Inexpensive Frequency Calibration Service From NIST''. 696-703. 8. "Extended Variances and Autoregressive Moving Average Algorithm for the Measurement and Synthesis of Oscillator Phase Noise".A. Proc. Greenhall.. Vol. on Freq. pp.. "A Shortcut for Computing the Modified Allan Variance". pp. pp. "Pulsar-Appropriate Clock Statistics". Hamilton Technical Services. pp. Instrum. IEEE. Hellwig. K. Symp.. Dec.A. pp. 1995. 27th PTTI Meeting. 44th Annu. Proc. 43rd Annu. No. Proc. Symp. Josties. Symp.BIBLUOGRAPHY 8.J. Picinbono. Weiss and J. 1. 2. Dec. 3. • • 137 . 45th Annu. 1996. 2.A.T. "Relating the Hadamard Variance to MCS Kalman Filter Clock Estimation". 4. Hironaka.T. R. J. 1984. Okayama.L. 262-264. 1975. 291-302. 32nd PTTI Meeting. pp. 667-678. on Freq. D. Characterization. J. Proc. Roberts. Ann.A. 470-474. Allan and J. and D. Riley. Jespersen. Jan. Hadamard Variance 1. 332-336. IM-46. Ayi. "Interpreting Oscillatory Frequency Stability Plots".331-335. June 1990. W. 1996. S. May 1981. pp. D. 725-732. IEEE Trans. Visr and J. Contrl. 1969.W. Proc. “Time Deviation and Time Prediction Error for Clock Specification. 5. C. Greenhall. Symp. June 1997. Contrl. Time Variance and Time Error Prediction 1. Howe. No. 5.. Pratt. 211-212.W.N. pp... D. Vol. and Application”. Wan. 9. 28th PTTI Meeting. D. Hutsell. 2006.A. 28th PTTI Meeting.. 1992 IEEE Freq.. Symp. Matsakis and F. Proc. Proc. Jacques Saloman Hadamard (1865-1963). Symp. W. “A Total Estimator of the Hadamard Function Used For GPS Operations”. 2. "Operational Use of the Hadamard Variance in GPS". Contrl. Levine. 35th Annu. Contrl. B. 222-225.D. May 1995. 2000 IEEE Freq. May 1992. 107-116. May 2002 • Modified Allan Variance 1. 7. on Freq. Nov. Dec. "The Third-Difference Approach to Modified Allan Variance". 6. on Freq. pp.W. on Freq. Proc. 57. pp.

D. Symp. "An Extension of the Allan Variance with Increased Confidence at Long Term. 2000. Weiss and D. Greenhall. Vol.A. Sept. 5. Bregni and L. Cont. pp." Proc. Peppler. April 1999. IEEE Trans.A.K. 2. May 2003. June 2001. 8. 10. Ferroelectrics and Freq. 6.K. Cont. Symp. 1183-1191. "Generalization of the Total Variance Approach to the Modified Allan Variance". D.A.A. 233-238. pp." Proc. Greenhall and W. . Howe. 9. Proceedings of the IEEE ICC. D. Symp. C. Contrl. C. . Freq..J. an Estimator of Long-Term Frequency Stability”. 1997 PTTI Meeting. "Total Variance as an Exact Analysis of the Sample Variance". Ferroelectrics and Freq. "Methods of Improving the Estimation of Long-term Frequency Variance. 1997.A. D. 279-290. pp. Lainson. Howe.A. March 1997. 883-889. S. Note: An old definition of TDEV..B Percival. 2. Modified Total Variance 1. D.A. Howe and F. Proceedings of the 35th Annual Precise Time and Time Interval (PTTI) Systems and Applications Meeting. Total Variance 1. IEEE Trans. No. 127-132. 1102-1110. "Definitions of Total Estimators of Common TimeDomain Variances". Freq. "Uncertainty of Stability Variances Based on Finite Differences". Howe. D. 321-329.. 1997. Ultrasonics. Symp. pp. 1999.A. 192-198. 2001 Intl. 5. "Simulation Study Using a New Type of Sample Variance. Proc. June 1995. Proc. Percival and D. Proc. 39-48.A. Dec. Jmoda. 1996 Intl. Howe. Proc. Proceedings of the 2003 IEEE International Frequency Control Symposium .. Howe. Cont. 1998 IEEE Int. Howe and T. 1997 PTTI Meeting. Symp. June 2006. D. and ThêoH 1.J. Freq. D. ThêoBR.A. 127-132. Cont. 7.A. June 1996 4. “Total TDEV”.HANDBOOK OF FREQUENCY STABILITY ANALYSIS • Modified Hadamard Variance 1. “The Total Deviation Approach to Long-Term Characterization of Frequency Stability”.K. Howe and T.A. Thêo1.J. D. December 2003. Dec.A. 31st PTTI Meeting. Contrl. Howe and K. " Very Long-Term Frequency Stability: Estimation Using a Special-Purpose Statistic". D. Time Total Variance 1.B. pp. Riley. M. 1093-1099. 2001 Intl.A. . pp. 1995 IEEE Int.A. Howe and D. 1997 European Frequency and Time Forum. 91-99..A. Freq. pp. Howe and T. 1999 Joint Meeting of the European Freq. 1995. Howe & C. June 1998. "Definitions of Total Estimators of Common TimeDomain Variances". Freq.A. Proc. Proc. Dec. "Effect of Drift on TOTALDEV". 1999. “Total Variance. Peppler. Vernotte. Lainson. pp. • • • • 138 . 1995 PTTI Meeting. Symp. Vol. Proc.267-276. Dec. UFFC-46. "Total Variance: A Progress Report on a New Frequency Stability Characterization". UFFC-47. pp.A. pp. Howe. Ultrasonics. June 2001. Howe and K. pp. and Time Forum and the IEEE Freq. Cont. "Total Variance Explained". Contrl. Greenhall. pp. 5. 2. D. “Improved Estimation of the Hurst Parameter of LongRange Dependent Traffic Using the Modified Hadamard Variance”. D. Sept. Peppler. D. pp." Proc. Proc. No. 3.

2. IEEE Trans. 1992. PTTI Meeting. 28th Annu. pp. Oct. Tasset. Meas. May 1992. • • 139 . Ivens. Vol. pp. Dec. 5. Tavella. 4. Galleani and P.J. E. 1240-1244. Proc. 24th PTTI Meeting. "The Characterization of Clock Behavior with the Dynamic AllanVariance". J. 1997 European Frequency and Time Forum. "ThêoH Bias-Removal Method". E. S. Meo. 6. 413-424. November 1997. 1996. March 1992. Howe. 2000. Instrum. Dodone and S. T. 2005 Joint FCS/PTTI Meeting. Symp. Gagnepain.. Master's Thesis. Dec. 5. I&M. 2003 Joint FCS/EFTF Meeting. pp. Dec. 91-99. pp. 49. Howe. "Measurement of Maximum Time Interval Error for Telecommunications Clock Stability Characterization”. Bregni and S.A. L. “The Range Covered by a Clock Error in the Case of White FM”. 788-792. 2004 EFTF . A. Meyer and F. 1998. S. Maccabruni. Groslambert and J. pp. London. PTTI Meeting. Meas. 2. M. 4. IM-46. Proc. 46th Annu.. 7. 5. Instrum. Vernotte. April 1993. P. 342-350. 2006 IEEE Freq. Travella. Tasset.. D. 900-906. No. Vernotte. 3. 3. pp. Contrl. “Synopsis of Timing Measurement Techniques Used in Telecommunucations”. "Fast Computation of Maximum Time Interval Error by Binary Decomposition". Bregni. Proc. June 2006. "Simultaneneous Measurement of Drifts and Noise Coefficients of Oscillators: Application to the Analysis of the Time Stability of the Millisecond Pulsars" Proc. No. IEEE Trans. "Tracking Nonstationarities in Clock Noises Using the Dynamic Allan Variance". Zampetti. Multi-Variance Analysis 1. "Clock Stability Characterization and Measurement in Telecommunications”. 24th PTTI Meeting. Tavella.J. Meas. 30th Annu. F. IEEE Trans. “The Range Covered by a Random Process and the New Definition of MTIE”. Bregni. Proc. Dec. 1997. "Simulating the Wander Accumulation in a SDH Synchronisation Network". pp. • MTIE 1. pp. Instrum. Contrl. University College. Dec. "Time Stability: An Improvement of the Multi-Variance Method for the Oscillator Noise Analysis". IM-42..A. Proc. 343-347. 4..BIBLUOGRAPHY 2. Vernotte. Lantz. Proc. Metrologia 43 (2006). pp. 313-326. D. Dynamic Stability 1.J. E. F. S322-S331. pp. F. 6. "Oscillator Noise Analysis: Multivariance Measurement". High-Confidence Statistic that Improves on the Allan Deviation”. Proc. J. Lantz. 6. F. "A New Multi-Variance Method for the Oscillator Noise Analysis". March 1997. "A Multi-Variance Analysis in the Time Domain". Naraghi. pp. Vernotte. IEEE Trans. Leschiutta. “ThêoH: A Hybrid. pp. Lantz.A. 1284-1294. 119123. P. 1992. 3. UK. Vol. 2. Galleani and P. Symp. IM-45. No. pp. J. Proc. 2. E. Lantz. Gagnepain. No. McGee-Taylor and T. Vol. L. 6th European Frequency and Time Forum. Proc. Dec. “Thêo1: Characterization of Very Long-Term Frequency Stability. on Freq. F. Vol. S. 1996. 49-60. G. 284-289. Walter.N. Howe and T. D. 239-244. Travella and D. Proc. Groslambert and J.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS

Confidence Intervals 1. J.A. Barnes and D.W. Allan, Variances Based on Data with Dead Time Between the Measurements", NIST Technical Note 1318, U.S. Department of Commerce, National Institute of Standards and Technology, 1990. 2. C.A. Greenhall, "Recipes for Degrees of Freedom of Frequency Stability Estimators'', IEEE Trans. Instrum. Meas., Vol. 40, No. 6, pp. 994-999, Dec. 1991. 3. M.A. Weiss and C. Hackman, "Confidence on the Three-Point Estimator of Frequency Drift'', Proc. 24th Annu. PTTI Meeting, pp. 451-460, Dec. 1992. 4. M.A.Weiss and C.A. Greenhall, "A Simple Algorithm for Approximating Confidence on the Modified Allan Variance and the Time Variance", Proc. 28th Annu. PTTI Meeting, pp. 215-224, Dec. 1996. 5. D.A. Howe, "Methods of Improving the Estimation of Long-Term Frequency Variance", Proc. 11th European Freq. and Time Forum, pp. 91-99, March 1997. 6. W.J. Riley, "Confidence Intervals and Bias Corrections for the Stable32 Variance Functions", Hamilton Technical Services, 2000. 7. F. Vernotte and M. Vincent, "Estimation of the Uncertainty of a Mean Frequency Measurement", 8. P. Lesage and C. Audoin, "Estimation of the Two-Sample Variance with a Limited Number of Data", pp. 311-318. 9. P. Lesage and C. Audoin, "Characterization of Frequency Stability: Uncertainty due to the Finite Number of Measurements", IEEE Trans. Instrum. Meas., Vol. 22, No. 2, pp.157-161, June 1973. 10. K. Yoshimura, "Degrees of Freedom of the Estimate of the Two-Sample Variance in the Continuous Sampling Method", IEEE Trans. Instrum. Meas., Vol. 38, No. 6, pp. 1044-1049, Dec. 1989. 11. C.R. Ekstrom and P.A. Koppang, “Error Bars for Three-Cornered Hats”, IEEE Trans. Ultrasonics, Ferroelectrics and Freq. Contrl., Vol. 53, No. 5, pp. 876-879, May 2006. 12. C. Greenhall and W. Riley, "Uncertainty of Stability Variances Based on Finite Differences", Proc. 35th PTTI Meeting, Dec. 2003. 13. T.N. Tasset, D.A. Howe and D.B. Percival, "Thêo1 Confidence Intervals", Proc. 2004 Joint FCS/UFFC Meeting, pp. 725-728, Aug. 2004. Drift Estimation and Removal 1. C.A. Greenhall, "Removal of Drift from Frequency Stability Measurements", Telecommunications and Data Acquisition Progress Report 42-65, July-Aug 1981, Jet Propulsion Laboratory, Pasadena, CA, pp. 127-132, 1981. 2. J.A. Barnes, "The Measurement of Linear Frequency Drift in Oscillators'', Proc. 15th Annu. PTTI Meeting, pp. 551-582, Dec. 1983. 3. M.A. Weiss and C. Hackman, "Confidence on the Three-Point Estimator of Frequency Drift", Proc. 24th Annu. PTTI Meeting, pp. 451-460, Dec. 1992. 4. M.A. Weiss, D.W. Allan and D.A. Howe, "Confidence on the Second Difference Estimation of Frequency Drift'', Proc. 1992 IEEE Freq. Contrl. Symp., pp. 300-305, June 1992. 5. “Long Term Quartz Oscillator Aging - A New Approach”, The Standard, HewlettPackard Co., pp. 4-5, Jan. 1994. 6. L.A. Breakiron, “A Comparative Study of Clock Rate and Drift Estimation”, 7. G.Wei, “Estimations of Frequency and its Drift Rate”, IEEE Trans. Instrum. Meas., Vol. 46, No. 1, pp. 79-82, Feb. 1997. 8. C.A. Greenhall, "A Frequency-Drift Estimator and Its Removal from Modified Allan Variance", Proc. 1997 IEEE Freq. Contrl. Symp., pp. 428-432, June 1997.

140

BIBLUOGRAPHY

9.

F. Vernotte and M. Vincent, “Estimation of the Measurement Uncertainty of Drift Coefficients Versus the Noise Levels", Proc. 12th European Freq. and Time Forum, pp. 222-227, March 1998. Bias Functions J.A. Barnes, “Tables of Bias Functions, B1 and B2, for Variances Based on Finite Samples of Processes with Power Law Spectral Densities”, NBS Technical Note 375, Jan. 1969. J.A. Barnes, "The Analysis of Frequency and Time Data", Austron, Inc., Dec. 1991. W.J. Riley, "Confidence Intervals and Bias Corrections for the Stable32 Variance Functions", Hamilton Technical Services

• 1. 2. 3.

Noise Identification 1. J.A. Barnes, "The Generation and Recognition of Flicker Noise", NBS Report 9284, U.S Department of Commerce, National Bureau of Standards, June 1967. 2. J.A. Barnes, “Effective Stationarity and Power Law Spectral Densities”, NBS Frequency-Time Seminar Preprint, Feb. 1968. 3. J.A. Barnes and D.W. Allan, “Recognition and Classification of LF Divergent Noise Processes”, NBS Division 253 Class Notes, circa 1970. 4. J.A. Barnes, "Models for the Interpretation of Frequency Stability Measurements", NBS Technical Note 683, U.S Department of Commerce, National Bureau of Standards, Aug. 1976. 5. C.A. Greenhall and J.A. Barnes, "Large Sample Simulation of Flicker Noise", Proc. 19th Annu. PTTI Meeting, pp. 203-217, Dec. 1987 and Proc. 24th Annu. PTTI Meeting, p. 461, Dec. 1992. 6. D. Howe, R. Beard, C. Greenhall, F. Vernotte and W. Riley, "A Total Estimator of the Hadamard Function Used for GPS Operations", Proc. 32nd PTTI Meeting, pp. 255-268, Nov. 2000. 7. W.J. Riley and C.A. Greenhall, "Power Law Noise Identification Using the Lag 1 Autocorrelation", Proc. 18th European Freq. and Time Forum, April 2004. 8. W.J. Riley, "Confidence Intervals and Bias Corrections for the Stable32 Variance Functions", Hamilton Technical Services. Simulation 1. J.A Barnes and D.W. Allan, "A Statistical Model of Flicker Noise", Proc. IEEE, Vol. 54, No. 2, pp. 176-178, Feb. 1966. 2. J.A. Barnes, "The Generation and Recognition of Flicker Noise", NBS Report 9284, U.S Department of Commerce, National Bureau of Standards, June 1967. 3. N.J. Kasdin and T. Walter, "Discrete Simulation of Power Law Noise", Proc. 1992 IEEE Freq. Contrl. Symp., pp. 274-283, May 1992. 4. T. Walter, “Characterizing Frequency Stability: A Continuous Power-Law Model with Discrete Sampling”, IEEE Trans. Instrum. Meas., Vol. 43, No. 1, pp. 69-79, Feb. 1994. 5. S.K. Park and K.W. Miller, "Random Number Generators: Good Ones are Hard to Find'', Comm. ACM, Vol. 31, No. 10, pp. 1192-1201. 6. S. Bregni, L. Valtriani, F. Veghini, "Simulation of Clock Noise and AU-4 Pointer Action in SDH Equipment", Proc. of IEEE GLOBECOM '95, pp. 56-60, Nov. 1995. 7. S. Bregni , "Generation of Pseudo-Random Power-Law Noise Sequences by Spectral Shaping" in Communications World. Editor: N. Mastorakis. WSES Press, 2001. 141

HANDBOOK OF FREQUENCY STABILITY ANALYSIS

Dead Time 1. J.A. Barnes and D.W. Allan, “Variances Based on Data with Dead Time Between the Measurements", NIST Technical Note 1318, U.S. Department of Commerce, National Institute of Standards and Technology, 1990. 2. D.B Sullivan, D.W Allan, D.A. Howe and F.L. Walls (Editors), "Characterization of Clocks and Oscillators", NIST Technical Note 1337, U.S. Department of Commerce, National Institute of Standards and Technology, TN-296-335, March 1990. 3. J.A. Barnes, "The Analysis of Frequency and Time Data", Austron, Inc., Dec. 1991. 4. D.A. Howe and E.E. Hagn, "Limited Live-Time Measurements of Frequency Stability", Proc. Joint FCS/EFTF, pp. 1113-1116, April 1999. 5. W.J. Riley, "Gaps, Outliers, Dead Time, and Uneven Spacing in Frequency Stability Data”, Hamilton Technical Services. 3-Cornered Hat 1. J.E. Gray and D.W. Allan, "A Method for Estimating the Frequency Stability of an Individual Oscillator", Proc. 28th Freq. Contrl. Symp., pp. 243-246, May 1974. 2. J. Groslambert, D. Fest, M. Oliver and J.J. Gagnepain, "Characterization of Frequency Fluctuations by Crosscorrelations and by Using Three or More Oscillators", Proc..35th Freq. Contrl. Symp., pp. 458-462, May 1981. 3. S.R. Stein, "Frequency and Time - Their Measurement and Characterization'', Chapter 12, Section 12.1.9, Separating the Variances of the Oscillator and the Reference, pp. 216-217, Precision Frequency Control , Vol. 2, Edited by E.A. Gerber and A. Ballato, Academic Press, New York, 1985, ISBN 0-12-280602-6. 4. P. Tavella and A. Premoli, "Characterization of Frequency Standard Instability by Estimation of their Covariance Matrix", Proc. 23rd PTTI Meeting, pp. 265-276. Dec. 1991. 5. P. Tavella and A. Premoli, "A Revisited Tree-Cornered Hat Method for Estimating Frequency Standard Instability", IEEE Trans. Instrum. Meas., IM-42, pp. 7-13, Feb. 1993. 6. C.R. Ekstrom and P.A. Koppang, "Error Bars for Three-Cornered Hats", IEEE Trans. UFFC, Vol. 53, No. 5, pp. 876-879, May 2006. 7. W.J. Riley, "Application of the 3-Cornered Hat Method to the Analysis of Frequency Stability", Hamilton Technical Services. Domain Conversions 1. A.R. Chi, “The Mechanics of Translation of Frequency Stability Measures Between Frequency and Time Domain Measurements”, Proc. 9th Annu. PTTI Meeting, pp. 523-548, Dec.1977. 2. J. Rutman, "Relations Between Spectral Purity and Frequency Stability”, pp. 160165. 3. R. Burgoon and M.C. Fisher, “Conversion Between Time and Frequency Domain of Intersection Points of Slopes of Various Noise Processes”, 32th Annu. Symp. on Freq. Contrl., pp.514-519, June 1978. 4. W.F. Egan, “An Efficient Algorithm to Compute Allan Variance from Spectral Density”, IEEE Trans. Instrum. Meas., Vol. 37, No. 2, pp. 240-244, June 1988. 5. F. Vernotte, J. Groslambert and J.J. Gagnepain, “Practical Calculation Methods of Spectral Density of Instantaneous Normalized Frequency Deviation from Instantaneous Time Error Samples”, Proc. 5th European Freq. and Time Forum, pp. 449-455, March 1991.

142

Gerber and A. Howe and L Erb. Sept. 3. Chapter 12. Symp. May 1992. Riley. U. Telecommunications and Data Acquisition Progress Report 42-88. on Freq. D. 1987. 483-486.T. Proc. 880-882. Levine. Numerical Recipes in C. Devon.Their Measurement and Characterization''..A. pp. Stein. Plymouth Road.A. Gray. W. "Software for Two Automated Time Measurement Systems". 1985. p. N70092-82-M-0579. Edited by E. Measurements 1. pp. Freq. "Addendum to a Test Suite for the Calculation of Time Domain Frequency Stability". "Performance of an Automated High Accuracy Phase Measurement System". Contrl. 3. 8. and Time Forum. John Wiley & Sons. • F. 1994. Proc. Ultrasonics. W. pp.K. Academic Press.). 3.B. Asmar and K. CO 80301 USA. C. V. June 1984. S. June 1982.P.. pp.A. 1982. Vol. Oct-Dec 1986. Barnes. Contrl. Symp.. Robust Statistics 1. Symp. Glaze. CA. "Properties of Signal Sources and Measurement Methods''.W. Symp. pp.R. Data Sheet. Percival. Pasadena. Teukolsky and W. TQ9 5LH England. J.. Model 5110A Time Interval Analyzer. Proc. Greenhall. Oudrhiri. Contrl. 4. Gifford. "Stable32 Frequency Domain Functions". Boulder.BIBLUOGRAPHY 6. June 1996. Contrl. Hilliard. Data Sheet. Riley... 1468-1472. S. A7 Frequency & Phase Comparator Measurement System. Vetterling. M. Symp. 2. 2. 1995 IEEE Freq. 200-212. Stein. May 1975. May 2001. 9. Cambridge. Greenhall. IEEE Trans. Proc. 1992 IEEE Freq. Contrl. Press. pp. 29th Annu. pp. Totnes. "A Test Suite for the Calculation of Time Domain Frequency Stability". 4. Naval Observatory Contract No. 1996 IEEE Freq. 544-548.R. Flannery.J.R. on Freq.S. Tutorial at 1993 IEEE Freq. 48. Contrl. June 1995. No. 11th European Freq. Proc. Howe. 3rd Edition. Riley. Ferroelectrics and Freq. Contrl. Symp. Greenhall. Allan and J. 404-411. Feb. S. Contrl. Contrl. Gernot M. S.J. 5. "Use of Robust Statistical Techniques in Time Scale Formation". 5. D.A. "Introduction to Robust Statistics and Data Filtering". 1988. "A Shortcut for Computing the Modified Allan Variance''. C. March 1997. NPO-20749. Stein and G. Press. 36th Annu.A. Freq. 360-366. pp.. Jet Propulsion Laboratory.A. D. C. 7. “Limitations on the Use of the Power-Law Form of Sy(f) to Compute Allan Variance”. Symp. Outliers in Statistical Data. S.W. Barnett and T. J. Quartzlock (UK) Ltd. Thomson. pp. 143 • • . Preliminary Report. U.A. 35th Annu. Cambridge Univ.. "Frequency and Time . Chichester. 2. 314-320.A. 147. NASA Tech Briefs. Symp.H. D. Winkler. 52. Precision Frequency Control. 2005. Hamilton Technical Services. TDEV and MTIE”. D.. 6. ISBN 0-12-280602-6. 1993. 9. 2. “Some Approach to Computation of ADEV. pp.216-217. W. pp. "Picosecond Time Difference Measurement System". Proc. 7.191-416. Kasznia. Gothic. May 1981. Lewis. "Frequency Stability Review". (formerly Timing Solutions Corp. Symmetricom. Sessions 3D and 4D. D. Ballato. Proc. D. Proc. New York. pp. "Oscillator-Stability Analyzer Based on a Time-Tag Counter". June 1. Inc. Allan.J. Vol. B.. ISBN 0-471-93094-6. 38th Annu. 262-264. W. Computation and Algorithms 1.

HANDBOOK OF FREQUENCY STABILITY ANALYSIS 144 .

58.... 3.. 86.... 37... 135 Bisection...... 54... 56....... 100..... 45........ 24.. 22. 60...... 47.... 132 Allan Variance (AVAR) ..... 15.............. 57. 22.......... 58..... 67.... 19.. 91. 95..... 37........... 80....... 105... 15......ii Heterodyne...........INDEX Index 3 3-Cornered Hat .... 53 Confidence Interval 19.. 36.......... 96...... 106.. 110.... 107 Frequency Drift... 56.. 109. 30............ii Feedback. 132 Measuring System ... 46.. 9.. 73... 50. 70 Dual Mixer Time Interval (DTMF) .. 26......... 19...... 31.... 74...... 61............ 36. 97. 131.......... 46....... 103 Autocorrelation. 54..................... 26..... 131 All Tau...53.. 107 Median.. 106.... 50........ 96............. 108..4. 108 Median Absolute Deviation (MAD) .. 30.. 4. 24...... 46.. 131.. 26. 19. 91.. 21.............. 106. 53. 26. 53....... 63.... 37........ 66... 22. 26...... 46........ 57 Autocorrelation Function (ACF)..... 29................ 109 L Lag 1 Autocorrelation.. 14.. 1. 70............. 67..... 3.. 20.. 30............ 33....... 128.... 100 Dead Time ........ 25....... 43...... 68........... 64..... 72. 99.. 34.............................. 73.. 33........... 85............... 131................... 40..... 70........... 67......... 57 Log Model ..................... 15. 25...... 29........ 70 145 . 55....................... 99 I IEEE-Standard-1139.. 132 B Bias Function.......... 109....... 47........... 113........ 127........ 114 A Aging . 18... 52........ 106. 110.. 3........ 18...... 19... 125 First Difference9............. 16.... 108.............. 47. 49........ 1. 18.... 131 Hamilton Technical Services .. 128 D Data Array ... 29.. 99 Data Arrays.3... 55. 49.................. 99 Diffusion........ 16.... 125......... 109....... 70 M Many Tau..... 97 Dynamic Stability .... 111. 46...... 107................. 40.. 21.. 53......... 113. 59... 68. 14. 50. 21.... 41.......... 38........... 15. 108 MIL-O-55310B............. 96. 109 Glossary . 34 Hadamard Variance ...... 61. 74..... 67........ 49.. 1...... 53.... 9.... 95.... 50....... 36 Hadamard Total Variance . 4... 73... 21.........ii Equivalent Degrees of Freedom (edf) ..... 110. 99 Histogram ......... 113.......... 58........ 41....... 97.. 10....... 106. 51.. 53 Error Bars ... 70 Maximum Time Interval Error (MTIE)41....... 54. 57.. 14.... 91. 61 Bibliography . 69.... 126... 35. 24.. 111... 33.......... 33.. 39.. 106......... 131 Frequency Offset ... 69... 13..... 15...... 31..... 23. 125..... 16.. 27.......... 99...... 106 C Chi Square (Χ2) ............. 42... 112.... 107..... 131 F Fax .. 55 Averaging Time (tau) .... 45... 47.... 25........ 37.... 132 Analysis Procedure ...... 44 E E-Mail. 69........ 131 H Hadamard Deviation . 85 Drift .... 123. 106. 24. 110 Allan Deviation (ADEV) . 11 J Jumps.... 106 Domain Conversion........ 25. 53................. 108............... 7 Database .. 65.... 22... 10... 95...... 59.. 127 G Gaps............... 74... 39..... 31.... 53.... 113...... 53..... 45............ 53. 43..... 38.. 46..... 46.. 58..

. 108.. 31... 123 Overlapping Allan Deviation........... 79.. 64 V Validation ................. 59... 49......... 23....... 33.. 129... 110... 27... 53..... 40.. 24............. 24 Syntonization...... 23. 24...... 43...... 58 Stride . 22......... 41... 131.. 132 Three-Point Fit.. 128 Total Variance (TOTVAR) 25..... 64...........HANDBOOK OF FREQUENCY STABILITY ANALYSIS Modified Allan Deviation (MDEV)15...... 25 T Test Suite ....... 58.... 53..... 33.... 41 Noise Identification . 23..... 125... 21.. 15.................. 59....... 42... 80.... 91.. 61 Noise Spectra........... 41..41............... 52..... 75... 54.... 18..... 51. 59.. 50 Time Variance (TVAR) .............. 127 Stable32 .. 110 Timetags ...... 81 Power Law Noise .... 53......... 40.... 92 Power Spectral Density (PSD).. 50.....41.......... 37...... 9.. 110.. 110. 65... 31.......... 14... 75 Two-Way Satellite Time Frequency Transfer (TWSTFT). 80. 57. 97 Time Interval Error (TIE) .. 47. 52.... 106...... 99.......... 53..... 23........... 46....... 57........ 58..... 38. 95.... 15 Synchronization .... 26...... 68 R Regression ...... 16. 125 P Phase Noise Diagram....... 107 Total Deviation (TOTDEV)....... 7 Y y (Frequency Data) . 29......... 43......... 21........ 17..... 39... 54.. 42.......... 131 Software . 53.............. 69...... 132 Time Total Variance (TTOT) ...... 15........ 125 Modified Total Variance (MTOT). 108.. 46. 31......... 86.................. 25... iii.......9............ 69...... 23... 14....... 32.. 50. 109. 111. 132 N NewThêo1 ............. 34...... 47..34... 23... 125.............. 85............ 56. 41.... 47... 33 Time Interval Counter (TIC)........ 123.... 35. 33... 45.. 109........ 79 O Outlier.. 110. 125. 7 146 .... 109... 37............. 128 Modified Allan Variance (MVAR) 15....... 56. 50...8......... 15.. 81 Q Quadratic ..... 96.. 131 Standard Variance..... 46.......... 99... 50... 132 ThêoBR ........... 132 ThêoH. 46............. 47.. 39..... 32..... 127 X x (Phase Data).. 18........... 100........... 69 S Second Difference .. 22...... 39....... 74.. 131 Transfer Function ... 41.. 44. 65.... 30.. 127 Thêo1 4.. 42... 126... 1.... 37.. 132 Time Deviation (TVAR).. 81. 32... 63... 112 Overlapping Allan Variance ........ 106..... 59 MTIE ......13.. 106 Time Deviation (TDEV).. 51... 34... 61...... 30.. 64..... 79........ 39....

NOTES 147 .

HANDBOOK OF FREQUENCY STABILITY ANALYSIS 148 .

Sign up to vote on this title
UsefulNot useful

Master Your Semester with Scribd & The New York Times

Special offer for students: Only $4.99/month.

Master Your Semester with a Special Offer from Scribd & The New York Times

Cancel anytime.