NTP Clock Discipline Principles

Similar documents
Timekeeping in the Interplanetary Internet

ENG2410 Digital Design Sequential Circuits: Part A

Department of Electrical Engineering, University of Waterloo. Introduction

ECE 545 Project Deliverables

CS 477/677 Analysis of Algorithms Fall 2007 Dr. George Bebis Course Project Due Date: 11/29/2007

Least Squares Optimal Filtering with Multirate Observations

ENSC Discrete Time Systems. Project Outline. Semester

Kinetic Model Completeness

Determining the Accuracy of Modal Parameter Estimation Methods

ECEN620: Network Theory Broadband Circuit Design Fall 2012

Lyapunov Stability Stability of Equilibrium Points

Physics 2010 Motion with Constant Acceleration Experiment 1

ENG2410 Digital Design Sequential Circuits: Part B

A Scalable Recurrent Neural Network Framework for Model-free

Bootstrap Method > # Purpose: understand how bootstrap method works > obs=c(11.96, 5.03, 67.40, 16.07, 31.50, 7.73, 11.10, 22.38) > n=length(obs) >

ECEN620: Network Theory Broadband Circuit Design Fall 2014

Chapter 3: Cluster Analysis

Public Key Cryptography. Tim van der Horst & Kent Seamons

Enhancing Performance of MLP/RBF Neural Classifiers via an Multivariate Data Distribution Scheme

3.4 Shrinkage Methods Prostate Cancer Data Example (Continued) Ridge Regression

Turing Machines. Human-aware Robotics. 2017/10/17 & 19 Chapter 3.2 & 3.3 in Sipser Ø Announcement:

Fall 2013 Physics 172 Recitation 3 Momentum and Springs

Multiple Source Multiple. using Network Coding

Chapter 3 Digital Transmission Fundamentals

SAP Note Missing documentation on enhancement MDR10001

Building research leadership consortia for Quantum Technology Research Hubs. Call type: Expression of Interest

SPH3U1 Lesson 06 Kinematics

15-381/781 Bayesian Nets & Probabilistic Inference

Computational modeling techniques

Differentiation Applications 1: Related Rates

ALE 21. Gibbs Free Energy. At what temperature does the spontaneity of a reaction change?

ROUNDING ERRORS IN BEAM-TRACKING CALCULATIONS

INSTRUCTIONAL PLAN Day 2

NUROP CONGRESS PAPER CHINESE PINYIN TO CHINESE CHARACTER CONVERSION

Early detection of mining truck failure by modelling its operation with neural networks classification algorithms

Supporting information

GENESIS Structural Optimization for ANSYS Mechanical

Synchronous Motor V-Curves

Materials Engineering 272-C Fall 2001, Lecture 7 & 8 Fundamentals of Diffusion

User Guide: Operation of ActiveAhead Mobile Application

OpenPowerNet Release Notes Version 1.7.0

and the Doppler frequency rate f R , can be related to the coefficients of this polynomial. The relationships are:

A NEW QUALITY INDICATOR FOR WIDEBAND UN- TERMINATION BY USING REFLECTIVE STANDARDS

CHAPTER 24: INFERENCE IN REGRESSION. Chapter 24: Make inferences about the population from which the sample data came.

PSU GISPOPSCI June 2011 Ordinary Least Squares & Spatial Linear Regression in GeoDa

Dead-beat controller design

OTHER USES OF THE ICRH COUPL ING CO IL. November 1975

ENG2410 Digital Design Arithmetic Circuits

Ecology 302 Lecture III. Exponential Growth (Gotelli, Chapter 1; Ricklefs, Chapter 11, pp )

Image Processing Adam Finkelstein & Tim Weyrich Princeton University

Cambridge Assessment International Education Cambridge Ordinary Level. Published

ELT COMMUNICATION THEORY

David HORN and Irit OPHER. School of Physics and Astronomy. Raymond and Beverly Sackler Faculty of Exact Sciences

BASIC DIRECT-CURRENT MEASUREMENTS

Sections 15.1 to 15.12, 16.1 and 16.2 of the textbook (Robbins-Miller) cover the materials required for this topic.

Scintillation and Cherenkov light detection with a 3 mm 3mm Silicon PhotoMultiplier

Purchase Order Workflow Processing

BASD HIGH SCHOOL FORMAL LAB REPORT

Lecture 02 CSE 40547/60547 Computing at the Nanoscale

N C R S I L V E R Q U A N T U M F A Q

Modelling of Clock Behaviour. Don Percival. Applied Physics Laboratory University of Washington Seattle, Washington, USA

Tree Structured Classifier

A Novel Electro-thermal Simulation Approach to Power IGBT Modules for Automotive Traction Applications

PE77D Series 3.3 V PECL Clock Oscillators November 2018

Physical Layer: Outline

the results to larger systems due to prop'erties of the projection algorithm. First, the number of hidden nodes must

[COLLEGE ALGEBRA EXAM I REVIEW TOPICS] ( u s e t h i s t o m a k e s u r e y o u a r e r e a d y )

o o IMPORTANT REMINDERS Reports will be graded largely on their ability to clearly communicate results and important conclusions.

Molecular Tagging Techniques Part - 2

Standard Title: Frequency Response and Frequency Bias Setting. Andrew Dressel Holly Hawkins Maureen Long Scott Miller

initially lcated away frm the data set never win the cmpetitin, resulting in a nnptimal nal cdebk, [2] [3] [4] and [5]. Khnen's Self Organizing Featur

Chapter 3 Digital Transmission Fundamentals

Performance Bounds for Detect and Avoid Signal Sensing

NUMBERS, MATHEMATICS AND EQUATIONS

Thermodynamics and Equilibrium

Lecture 24: Flory-Huggins Theory

Section I5: Feedback in Operational Amplifiers

How do we solve it and what does the solution look like?

Revisiting the Socrates Example

COMP 551 Applied Machine Learning Lecture 11: Support Vector Machines

Lecture 6: Phase Space and Damped Oscillations

Subject description processes

Guaranteeing Reliability with Vibration Simulation and Testing. Dr. Nathan Blattau

Energy and Power. Historical Scaling. Technology Scaling. CPU_power_dissipation

IB Sports, Exercise and Health Science Summer Assignment. Mrs. Christina Doyle Seneca Valley High School

Applying Kirchoff s law on the primary circuit. V = - e1 V+ e1 = 0 V.D. e.m.f. From the secondary circuit e2 = v2. K e. Equivalent circuit :

Predictive Coding. U n " S n

Methods for Determination of Mean Speckle Size in Simulated Speckle Pattern

Technical Bulletin. Generation Interconnection Procedures. Revisions to Cluster 4, Phase 1 Study Methodology

A water level indicator or other measuring device to determine the current depth to the water.

Pattern Recognition 2014 Support Vector Machines

SUMMER REV: Half-Life DUE DATE: JULY 2 nd

COMP9444 Neural Networks and Deep Learning 3. Backpropagation

Checking the resolved resonance region in EXFOR database

Jurnal Teknologi. Design Verification of Heat Exchanger for Ballast Water Treatment. Full paper

Chapter 16. Capacitance. Capacitance, cont. Parallel-Plate Capacitor, Example 1/20/2011. Electric Energy and Capacitance

Flipping Physics Lecture Notes: Simple Harmonic Motion Introduction via a Horizontal Mass-Spring System

Churn Prediction using Dynamic RFM-Augmented node2vec

(2) Even if such a value of k was possible, the neutrons multiply

Transcription:

NTP Clck Discipline Principles David L. Mills University f Delaware http://www.eecis.udel.edu/~mills mailt:mills@udel.edu Sir Jhn Tenniel; Alice s Adventures in Wnderland,Lewis Carrll 24-Aug-04 1

Traditinal apprach using phase-lck lp (PLL) Respnse t 10-ms Phase Step Respnse t 2-PPM Frequency Step Left graph shws the impulse respnse fr a 10-ms time step and 64-s pll interval using a traditinal linear PLL. Right graph shws the impulse respnse fr a 5-PPM frequency step and 64-s pll interval. It takes t lng t cnverge the lp using linear systems. A hybrid linear/nnlinear apprach may d much better. 24-Aug-04 2

Clck discipline design principles The clck discipline algrithm functins as a nnlinear, hybrid phase/frequency-lck (NHPFL) feedback lp. Detailed cmputer clck analysis yields the ptimum averaging interval depending n prevailing netwrk jitter and scillatr wander. Optimum value is determined in real time by measuring the jitter and wander separately. Clck state machine quickly cnverges time and frequency and suppresses transients resulting frm leap events, etc. Huff&puff algrithm crrects fr large utlyers and asymmetric delays Ppcrn spike suppressr clips nise spikes. 24-Aug-04 3

Clck discipline design apprach Phase nise due t netwrk jitter prevails at the lwer pll intervals, s a secnd-rder phase-lck lp (PLL) is the best frequency predictr. Frequency nise due t randm-walk scillatr wander prevails at the higher pll intervals, s a first-rder frequency-lck lp (FLL) is the best frequency predictr. A crafted heuristic algrithm is necessary t cmbine bth predictins. The NHPFL algrithm cmbines the time and frequency predictins in a seamless way fr pll intervals frm 16 secnds t 36 hurs. The PLL frequency adjustment is cmputed as the integral f past frequency ffsets. The FLL frequency adjustment is cmputed as the expnential average f past frequency ffsets. An additinal phase adjustment is necessary fr lp stabiility. The pll interval, which determines the lp time cnstant, is determined in respnse t measured jitter and wander. 24-Aug-04 4

Clck discipline algrithm NTP θ r + θ c Phase Detectr V d Clck Filter V s VFO Lp Filter x V c Clck Adjust y Phase/Freq Predictin V d is a functin f the NTP and VFO phase differences. V s depends n the stage chsen f the clck filter shift register. x is the phase crrectin and y the frequency adjustment cmputed by the predictin functins. The clck adjust prcess runs nce each secnd t adjust the VFO phase by V c. The lp behavir is determined by the lp filter parameters. 24-Aug-04 5

FLL/PLL predictin functins x Phase Crrect y FLL FLL Predict V s y Σ y PLL PLL Predict V s is the phase ffset prduced by the clck filter algrithm. x is the phase crrectin cmputed as the value f V s. y FLL is the frequency predictin cmputed as the average f past values f V s. y PLL is the frequency predictin cmputed as the integral f past values f V s. y FLL and y PLL are cmbined accrding t weight factrs determined by pll interval, update interval and Allan intercept. 24-Aug-04 6

Detailed calculatins The phase crrectin x and frequency predictins y PLL and y FLL are recalculated at each clck update. min( µ,2 y PLL = (4K 2 PLL τ ) θ τ ) 2 y FLL = K FLL θ max( µ, A The VFO adjustment V C is updated by the clck adjust prcess at nesecnd intervals. x dx = V y y dx x = x dx ζτ C = PLL + FLL + K PLL 2 x ) x = θ ζ = 1.2 ζ = 1 2 τ + 4 > A X therwise Cnstants Variables K PLL = 16 PLL gain τ pll interval (lg 2 ) K FLL = 8 FLL gain µ update interval A x = 1024s Allan intercept θ clck ffset θ ffset change since last update ζ damping factr 24-Aug-04 7

Pll adjust strategy Nte that as τ increases the phase nise φ P decreases with slpe 1, while the frequency nise φ F (τ << 1) increases with slpe +0.5. Thus, the minimum errr is when φ P = φ F (τ << 1). (Remember that τ is lg 2 f the actual pll interval.) Thus, the strategy is: If φ P > φ F (τ << 1) and θ < K G φ P, increase the hysteresis cunter h by τ. If h > K H, set h = 0 and increase τ by ne. Else, decrease h by tw, in rder t adapt t rapid frequency changes. if h < -K H, set h = 0 and decrease τ by ne. Cnstants Variables K H = 30 hysteresis limit φ P average phase differences K G = 4 hysteresis threshld φ F average frequency differences h hysteresis cunter 24-Aug-04 8

State machine peratins There are three threshlds which affect the state machine. Panic threshld (1000 s): exit t the perating system if ffset exceeds. Step threshld (128 ms): ignre if ffset exceeds until steput. Steput threshld (900 s): interval within which step spikes are ignred. When the discipline is started fr the first time, set the time and calculate a pssibly large frequency crrectin. Subsequently when the discipline is started, set the time nly if the ffset exceeds the step threshld. When calculating the frequency crrectin, cntinue t the steput threshld in rder t prduce an accurate value, then set the time and frequency. Once the initial time and frequency have been set, run the HNPFL algrithm and the pll-adjust algrithm. Ignre transients greater than the step threshld, unless the steput threshld is exceeded. 24-Aug-04 9

Clck state machine transitin functin NSET FSET 0: n step 1: step 2: steput and n step 3: steput and step 0, 1: set time, sc 0: 1: set time 3: set time/freq 0, 1: FREQ 1: sc TSET 2: set freq, sc 0: PLL, sc 0: PLL, sc 2: 3: set time/freq 1: SYNC 0: PLL, sc SPIK 24-Aug-04 10

Frequency ffset and pll interval frm simulatr 24-Aug-04 11

Leap secnd insertin TAI UTC = 31 s B A 23:59:59 23:59:60 00:00:00 TAI UTC = 32 s 23:59:58 Hardware time is read frm the prcessr cycle cunter that increments in the lw nansecnd range. Sftware time may nt step backward; it must increment frward at least 1 ns fr every reading. The clck is stepped backward at leap secnd 23:59:59, but sftware time stays the same (A), unless the clck is read. At the end f the leap secnd 23:59:60 the clck is ahead (B) in nansecnds the number f times it was read. 24-Aug-04 12

Clck discipline algrithm perfrmance The algrithm cnverges time within 5 ms and frequency within 2 PPM in a very shrt time with pll intervals up t 10 (1024 s). Time t cnverge with n frequency file is less than 20 min. Time t cnverge with frequency file and n iburst is less than 4 min. Time t cnverge with frequency file and iburst is less than 10 s. Previus designs culd take days t achieve this perfrmance. Fllwing slides shw results frm a simulatr run fr typical LAN Initial scillatr frequency ffset -400 PPM with wander parameter 1 s/s. Initial time ffset 600 s with netwrk jitter parameter 1 ms. These are parameters typical fr 10 Mb Ethernets and cmputer scillatrs. The pll interval rapidly adapts t frequency changes. The frequency (blue) is in PPM. The pll interval (green) is in lg2(s) units. It increases slwly it jitter is greater than wander and decreases rapidly therwise. 24-Aug-04 13

Further infrmatin NTP hme page http://www.ntp.rg Current NTP Versin 3 and 4 sftware and dcumentatin FAQ and links t ther surces and interesting places David L. Mills hme page http://www.eecis.udel.edu/~mills Papers, reprts and memranda in PstScript and PDF frmats Briefings in HTML, PstScript, PwerPint and PDF frmats Cllabratin resurces hardware, sftware and dcumentatin Sngs, pht galleries and after-dinner speech scripts Udel FTP server: ftp://ftp.udel.edu/pub/ntp Current NTP Versin sftware, dcumentatin and supprt Cllabratin resurces and junkbx Related prjects http://www.eecis.udel.edu/~mills/status.htm Current research prject descriptins and briefings 24-Aug-04 14