The 60802 challenge of meeting time accuracy goals across ...

121
McCall, Stanton with Woods, Weingartner, and Schlechter IEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021 McCall, Stanton with Woods, Weingartner, and Schlechter IEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021 The 60802 challenge of meeting time accuracy goals across long daisy-chains using 802.1ASโ„ข-2020 An analysis and a proposed path forward David McCall, Kevin Stanton, Greg Schlechter (Intel) Jordan Woods, Tom Weingartner (Analog Devices) September 2021 Interim of 802.1 TSN / 60802 v3 1

Transcript of The 60802 challenge of meeting time accuracy goals across ...

Page 1: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

The 60802 challenge of meeting time accuracy goals across long daisy-chains using 802.1ASโ„ข-2020An analysis and a proposed path forward

David McCall, Kevin Stanton, Greg Schlechter (Intel)Jordan Woods, Tom Weingartner (Analog Devices)

September 2021 Interim of 802.1 TSN / 60802

v3

1

Page 2: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Updates

โ€ข v2 โ†’ v3โ€“ Fixed minor typos and consistency issues.

โ€“ Updated speaker notes to match revised slide content.

โ€“ Changed top level references from Path Delay, ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ and ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐ถ๐‘œ๐‘Ÿ๐‘Ÿ๐‘’๐‘๐‘ก๐‘–๐‘œ๐‘› to Link Delay, ๐‘š๐‘’๐‘Ž๐‘›๐ฟ๐‘–๐‘›๐‘˜๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ and ๐‘š๐‘’๐‘Ž๐‘›๐ฟ๐‘–๐‘›๐‘˜๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐ถ๐‘œ๐‘Ÿ๐‘Ÿ๐‘’๐‘๐‘ก๐‘–๐‘œ๐‘› .

โ€ข v1 โ†’ v2โ€“ In ๐‘๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ and ๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ sections, changed to reflect that pDelayResp, ๐‘ก3

& ๐‘ก4 are used to measure NRR (not pDelayReq, ๐‘ก1

& ๐‘ก2 )

โ€“ In ๐‘๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ section, fixed calculation of NRR to represent ๐‘๐‘…๐‘…๐ด๐ต (was previously, erroneously, ๐‘๐‘…๐‘…๐ต๐ด). Also changed graphic to

make it clear that ๐‘๐‘…๐‘…๐ด๐ต is the ratio measured from A of its clock rate relative to B.

โ€“ Fixed ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘๐‘…๐‘… equation (should be /2)

โ€“ Fixed ๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐ท๐‘Ÿ๐‘–๐‘“๐‘ก equations (should be ๐‘๐‘™๐‘œ๐‘๐‘˜๐ท๐‘Ÿ๐‘–๐‘“๐‘ก๐บ๐‘€ ๐‘๐‘™๐‘œ๐‘๐‘˜๐ท๐‘Ÿ๐‘–๐‘“๐‘ก ๐‘› not ๐‘๐‘™๐‘œ๐‘๐‘˜๐ท๐‘Ÿ๐‘–๐‘“๐‘ก ๐‘› ๐‘๐‘™๐‘œ๐‘๐‘˜๐ท๐‘Ÿ๐‘–๐‘“๐‘ก๐บ๐‘€)

โ€“ Fixed discussion of ๐‘๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ in most recent time series simulationsโ€ข NRR is measured across seven pDelay Intervals prior to taking the median of the past seven measurements. (Previously, only taking the median

was discussed.)

โ€“ Completed formulae in Backup and added slides for relevant formulae at the end of each sectionโ€ข Formulae in Backup include consideration of the units for each factor and secondary errors which will not be included in the initial model. Formulae

in the main presentation do not.

โ€“ Changed some โ€œpDelayโ€ references to โ€œPath Delayโ€ where the latter is more appropriate (i.e. the reference isnโ€™t to a pDelaymessage or used in a formula).

p. 2

2

Page 3: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Abstract

Industrial Automation Systems require microsecond-accurate time across long daisy-chains of devices using IEEE Std. 802.1ASโ„ข-2020 as specified by IEEE/IEC 60802.

Simulated protocol and system parameters have thus far either been judged impractical or have failed to meet the time-accuracy requirement.

In this contribution we:

1. Summarize the assumptions, requirements and results to date

2. Challenge certain assumptions

3. Propose an approach that we believe could lead to a solution

p. 3

3

Page 4: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Some Requirements & Assumptions to Date

โ€ข Requirements:โ€“ Topology: 64-100 network hops (i.e. 2 endstations + 63-99 devices in a chain)

โ€“ Time Accuracy: Not to exceed +/-1us with respect to the GM at any nodeโ€ข Measured at each Working Clock, e.g. used by 802.1Qbv (i.e. GCL) and the Application Clock

โ€ข Assumptions:โ€“ Local Clock reference frequency (in each system) is an XO, i.e. a quartz crystal oscillator

โ€“ The 802.1AS GM does not have better PPM stability than the other nodes

โ€“ No averaging of the pDelay measurement is performed

โ€“ Temperature gradients induce PPM gradients, resulting in time-duration-measurement errors

โ€“ Various Ambient Temperature profiles (TAmbient) have been usedโ€ข With various transfer functions of PPM = f(TCrystal)]

โ€“ Zero thermal resistance between TAmbient and TCrystal (0oK/W)

Note: Assumptions that we recommend revisiting are marked with โ€œโ€

p. 4

4

Page 5: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Other Principles

โ€ข Initial 60802 standard should not trigger new silicon design / spinsโ€“ Requirements should be met by commodity switch and endpoint silicon today that

were designed to support IEEE Std. 802.1ASโ„ข-2020

โ€ข The solution should not unnecessarily limit the deployment of IEEE Std. 1588โ„ข-2019 security mechanismsโ€“ With todayโ€™s 802.1AS-2020-capable silicon

โ€ข Once viable 802.1AS-2020 parameters are chosen, the worst-case PPM/sec allowed per time-aware system should be derived

โ€ข The function of PPM vs Tambient is a property of the electrical, mechanical and thermal design of the board / chassis / systemโ€“ Thus the PPM / TAmbient requirement should be specified, but not the method by

which the PPM / TAmbient requirement must be met by the system design

p. 5

5

Page 6: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Viable Parameter Selections

โ€ข Sync Interval

โ€ข Sync-to-Follow-up delay

โ€ข pDelay Interval

โ€ข Timestamp Granularity

โ€ข Rx/Tx Timestamp Asymmetry:

โ€ข Residence Time (in addition to Sync-to-Follow-up delay)

A simple analytical Model is needed to more quickly choose scenarios for more detailed time-series simulationQuickly iterate a Model over minutes, then invest the >1 week for the detailed simulation

p. 6

6

Page 7: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Variable Time Error (vTE): Timestamp Model

Local Clock(f Hz)

PHY/MACRx

0 < e < Vr

PHY/MAC Tx

0 < e < Vt

Rx Timestamp Tx Timestamp

Va

lue

PTP Relay Instance

e: The time errorVr: Variable Rx delayVt: Variable Tx delay1/f: Timestamp granularity

p. 7

7

Page 8: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 8McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Motivation & Results (so far)

โ€ข Develop a method of iterating fasterโ€ข Input: key 802.1AS parametersโ€ข Output: time sync accuracy over 64 and 100hops

โ€ข Approach by modelling the errors and how they accumulateโ€ข Requires an analysis of the errors first

โ€ข Analysis of the errors has already suggested some assumptions inherent in the current time series simulation that should be revisited

โ€ข This presentation walks through the error analysis and the most important insights it reveals

โ€ข Goal is to introduce the overall approach share the insightsโ€ข There are some areas that require further analysis; will present in a later session

โ€ข Next steps:โ€ข Build the modelโ€ข Complete the additional analysis

This is a very graphical presentation, but the uploaded version includes extensive speaker notes (which you are reading now!).

There is a lot of information in this presentation. The main section Iโ€™m about to go through is nearly over 100 slides, but many of them are builds, so we should get through them fairly quickly.

8

Page 9: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 9McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Summary of Main Insights

โ€ข meanLinkDelay should be an average of pDelay measurements.โ€ข Should nearly eliminate meanLinkDelayerror

โ€ข NRR should not be averaged.โ€ข Provided (Timestamp Errors / pDelay Interval) is negligible in terms of ppm

โ€ข There are significant benefits to the GM having greater ppm stability (lower ppm/s).

โ€ข It is hard (impossible?) to eliminate Timestamp granularity and Dynamic Timestamp error effects from Residence Time

โ€ข Need more work to see how far these elements can realistically be reduced

โ€ข Synchronising pDelayReq to arrive just before Sync would minimise the effect Clock Drift induced errors it NRR and RR.

โ€ข There may be opportunities to compensate for NRRerrorโ€ข And therefore RRerror

9

Page 10: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 10McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Link DelayResidence Time

Time Sync

GM 10

2 3 99End

Station100

97 98

Link DelayResidence Time

Link DelayResidence Time

Link DelayResidence Time

Link DelayResidence Time

Link DelayResidence Time

Link Delay

This is a simple view of how time sync works.

The basics are: measure Link Delays and Residence Times; add them to a Sync message as it is passed from the GM to the End Station.

10

Page 11: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 11McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

TIMECorrection Field

Rate Ratio

Link DelayResidence Time

NRR

Time Sync

GM 10

2 3 99End

Station100

97 98

Link DelayResidence Time

NRR

Link DelayResidence Time

NRR

Link DelayResidence Time

NRR

Link DelayResidence Time

NRR

Link DelayResidence Time

NRR

Link DelayNRR

TIMECorrection Field

Rate RatioTIME

TIMECorrection Field

Rate Ratio

TIMECorrection Field

Rate Ratio

TIMECorrection Field

Rate Ratio

TIMECorrection Field

Rate Ratio

TIMECorrection Field

Rate Ratio

Adding a bit more detail, this is how thatโ€™s done in practice.

The Sync message contains the TIME of transmission from the GM, a Correction Field of the delay that has accumulated since it was transmitted, and a measure of the Rate Ratio, i.e. the ratio between the current clock speed and the GMโ€™s clock speed.

Itโ€™s worth pointing out that each device is responsible for updating the Correction Field based on the incoming Link Delay plus itโ€™s own Residence Time. Also a device measures the Neighbor Rate Ratio between itself and the upstream device. It then adds that to the incoming Rate Ratio. And finally uses the new Rate Radio to modify the measured Link Delay and the Residence Time prior modifying the Correction Field, and passing on everything to the next device. Rate Ratio is calculated and used for this step because the Correction Field is in terms of the GM clock, not the local clock โ€“ which is used to measure Link Delay and Residence Time - and multiplying by the Rate Ratio accomplishes that translation.

11

Page 12: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 12McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Time Sync โ€“ Elements & Relationships

Time Sync

Link Delay Residence Time

Building up the logical elements and connection of this processโ€ฆthe simple model looks like thisโ€ฆ

12

Page 13: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 13McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Time Sync โ€“ Elements & Relationships

GM TIMEEnd Station 

Time

Link Delay Residence Time

Corrections

Rate Ratio

Corrections

Rate Ratio

Accumulated Corrections

Rate Ratio (Accumulated NRRs)

As mentioned, the Time Sync is actually accomplished by accumulating the delays (i.e. the corrections) and measuring the rate ratio between the GM and the End Station.

13

Page 14: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 14McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Time Sync โ€“ Elements & Relationships

GM TIMEEnd Station 

Time

Link Delay Residence Time

INCOMING OUTGOING

CorrectionAdjustment

Neighbor Rate Ratio

Correction

Rate Ratio

Correction

Rate Ratio

Rate Ratio

At each bridge there is a process for taking the incoming Correction Field and Rate Ratio and adding effects of the local pDelay (between the local and upstream device) and Residence Time.

Iโ€™ll be referring back to this diagram later, but when it comes to analysing the errors involved in this process, itโ€™s helpful to take a slightly different viewโ€ฆ

14

Page 15: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 15McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Link Delay

Time Sync โ€“ Elements & Relationships

Dynamic Time Sync Error

Timestamp Granularity Error

Residence Time

Dynamic Timestamp Error

Neighbor Rate Ratio Rate Ratio

Clock Drift

Link Delay Error Residence Time Error

Neighbor Rate Ratio Error

Rate RatioError

This is the start of the error analysis.

Timestamp Granularity Error, is related to the maximum resolution of Timestamps.

Dynamic Timestamp Error is in addition to Granularity and can be very implementation dependant.

Clock Drift is just the way that each device has itโ€™s own local clock which is running a slightly different speedโ€ฆand also changing that speed over time (mostly due to temperature changes; or at least those are the changes that have the greatest impact on our analysis).

All other errors are all rooted in these three.

Not all of the relationships need to be modelled. Figuring out which can be ignored is part of the analysis.

15

Page 16: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 16McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Link Delay Error โ€“Timestamp Granularity & Dynamic Time Stamp Error

16

Page 17: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 17McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Time Sync โ€“ Elements & Relationships

Dynamic Time Sync Error

Time Stamp Granularity

Link Delay Error Residence Time Error

Dynamic Time Stamp Error

Neighbor Rate Ratio Error

Rate RatioError

Clock Drift

Time Stamp Granularity

Link Delay Error

Dynamic Time Stamp Error

To be clear, weโ€™re only looking at these three elements in this section. We arenโ€™t looking at any of the others.

17

Page 18: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 18McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A (e.g. 99)

B (e.g. 98)

t1

t2 t3

t4

(t2 โ€ t1) + (t4 โ€ t3)

2pDelay

t2 โ€ t1 = 14 ns t4 โ€ t3 = 14 ns

pDelayReq pDelayResp

=

This is the idealised viewโ€ฆthe Platonic ideal of how pDelay is measured.

Note that โ€œAโ€ is device 99. It initiates a pDelayReq to the upstream device (98)โ€ฆwhich responds with a pDelayRespโ€ฆafter the pDelay Response Time.

The default maximum for pDelay Response is 10ms, which is much longer than the pDelay which in this case is 14nsโ€ฆindicating a cable length of 4.2mโ€ฆbut for the moment Iโ€™m going to model this shorter pDelay Response Time to make things legible and illustrate some principlesโ€ฆ

18

Page 19: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 19McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16 32 40 48 56 64 72 80

0 8 16 32 40 48 56 64 72 80

t1

t2 t3

t4

(t2 โ€ t1) + (t4 โ€ t3)

2= 14 ns

t2 โ€ t1 = 14 t4 โ€ t3 = 14

pDelayReq pDelayResp

24

24

Letโ€™s put some numbers on the time line. Weโ€™ll use a resolution of 8ns, which is what youโ€™d get with a clock speed of 125MHz.

Butโ€ฆclocks A & B arenโ€™t the sameโ€ฆthey arenโ€™t counting the same numbers at the same time.

19

Page 20: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 20McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16 24 32 40 48 56 64 72 80

8000 8008 8016 8024 8032 8040 8048 8056 8064 8072 8080

t1

t2 t3

t4

(t4 โ€ t1) โ€ (t3 โ€ t2)

2= 14 ns

t4 โ€ t1 = 46

t3 โ€ t2 = 18

pDelayReq pDelayResp

Weโ€™ll model this by just adding a nice, round 8,000 to the lower line.

This requires changing the equation to subtract Bโ€™s pDelay Response time (on the lower timeline) from the time between A sending the Request and receiving the Response. Itโ€™s mathematically equivalentโ€ฆand also allows compensation for differing clock speeds (via Neighbor Rate Ratioโ€ฆwhich we are not considering in this section).

20

Page 21: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 21McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16 24 32 40 48 56 64 72 80

t1 t4

= 16 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1 = 48

t2 t3

8000 8008 8016 8024 8032 8040 8048 8056 8064 8072 8080

t3 โ€ t2 = 16

8ns is our Timestamp Granularityโ€ฆwhich means the pDelay Req could arrive at any time between clock ticks and the Timestamp will actually reflect the next clock tick.

Iโ€™m keeping things simple here by assuming the TX time is exactly on a clock tick (and measured as such)โ€ฆ.but you can see the effect this has. The measurement of pDelay is 16ns, not 14ns.

21

Page 22: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 22McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16 24 32 40 48 56 64 72 80

t1 t4

= 16 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1 = 48

t2 t3

8000 8008 8016 8024 8032 8040 8048 8056 8064 8072 8080

t3 โ€ t2 = 16

Timelines A &B are independent of each other and move relative to each other due to different PPMs of Local Clocks.

Iโ€™ll deal with the movement relative to each other in a moment. For now, letโ€™s just look at this measurement and the fact that the timelines are independent. This means the clock ticks donโ€™t necessarily line up exactly. They mightโ€ฆbut itโ€™s more likely there will be an offset.

22

Page 23: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 23McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16

t1

t2 t3

= 16 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

8000 8008 8016 8024 8032 8040 8048 8056 8064 8072 8080

pDelayReq pDelayResp

t4 โ€ t1 = 48

t3 โ€ t2 = 16

24 32 40 48 56

t464 72 80

Offset: 0ns

23

Page 24: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 24McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16

t1

t2 t3

= 16 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

8000 8008 8016 8024 8032 8040 8048 8056 8064 8072 8080

pDelayReq pDelayResp

t4 โ€ t1 = 48

t3 โ€ t2 = 16

24 32 40 48 56

t464 72 80

Offset: 1ns

24

Page 25: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 25McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16

t1

t2

= 16 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

8008 8016 8024

pDelayReq pDelayResp

t4 โ€ t1 = 48

t3 โ€ t2 = 16

24 32 40 48 56

t464 72 80

t3

8032 8040 8048 8056 8064 8072 8080

Offset: 2ns

At an offset >2ns, the measurement is no longer 16nsโ€ฆ

25

Page 26: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 26McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16 24 32 40 48 56 64 72 80

t1 t4

= 20 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1 = 56

t3 โ€ t2 = 16

8008 8016 8024 8032 8040 8048 8056 8064 8072 8080

t2 t3

Offset: 3ns

โ€ฆitโ€™s worse. 20ns.

This, by the way, is to be expected. With a Timestamp Granularity of 8ns, and a divide by 2 in the equation, we can expect a measurement granularity of 4ns.

26

Page 27: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 27McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16

t1

= 20 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1 = 56

t3 โ€ t2=16

24 32 40 48 56 64 72 80

t4

8008 8016 8024 8032 8040 8048 8056 8064 8072 8080

t2 t3

Offset: 4ns

27

Page 28: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 28McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16

t1

= 20 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1 = 56

t3 โ€ t2 = 16

24 32 40 48 56 64 72 80

t4

8008 8016 8024 8032 8040 8048 8056 8064 8072 8080

t2 t3

Offset: 5ns

28

Page 29: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 29McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16

t1

= 16 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1 = 48

t3 โ€ t2 = 16

24 32 40 48 56

t464 72 80

8008 8016 8024 8032 8040 8048 8056 8064 8072 8080

t2 t3

Offset: 6ns

At 6ns the measurement gets back to 16ns.

29

Page 30: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 30McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16

t1

= 16 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

8096

pDelayReq pDelayResp

t4 โ€ t1 = 48

t3 โ€ t2 = 16

24 32 40 48 56

t464 72 80

8008 8016 8024 8032 8040 8048 8056 8064 8072 8080

t2 t3

Offset: 7ns

30

Page 31: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 31McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16

t1

= 16 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

8096

pDelayReq pDelayResp

t4 โ€ t1 = 48

t3 โ€ t2 = 16

24 32 40 48 56

t464 72 80

8008 8016 8024 8032 8040 8048 8056 8064 8072 8080

t2 t3

Offset: 0ns

And then weโ€™re back to where we started, with an offset of 0ns.

31

Page 32: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 32McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Offset / ns Measured pDelay / ns

0 โ€“ 2 16

>2 โ€“ 6 20

>6 โ€“ <8 16

Average = 18 ns โ‰  14 ns

OK to average as, over multiple measurements, the timelines will move relative to each other.But on itโ€™s own that doesnโ€™t help.

Soโ€ฆdoes this even work at all?

The good news is that yes, it does. There are two reasonsโ€ฆ

32

Page 33: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 33McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16 24 X+40 X+48 X+56 64X+32

t1

t2 t3

t4

= 16 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

8000 8008 8016 8024 X+8040 X+8048 X+8056 X+8064X+8032

pDelayReq pDelayResp

t4 โ€ t1 = X+48

t3 โ€ t2 = X+16

X

X8032

32

Firstly, there is a relatively large amount of time between the left and right ends of the timeline due to the fact that pDelay Response Time is much longer than pDelayโ€ฆa fact I previously ignored.

33

Page 34: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 34McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16 24 X+40 X+48 X+56 64X+32

t1

t2

t4

= 16 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

8000 8008 8016 8024

t3

X+8040 X+8048 X+8056 X+8064X+8032

pDelayReq pDelayResp

t4 โ€ t1 = X+48

t3 โ€ t2 = X+16

XA

XB8032

32

And secondly, that relatively long period of time is measured by two free running local clocks. Running at different speedsโ€ฆand drifting slightly relative to each other.

34

Page 35: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 35McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16 24 X+40 X+48 X+56 64X+32

t1 t4

= 16 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1 = X+48

t3 โ€ t2 = X+16

XA

XBt2

8000 8008 8016 8024

t3

X+8040 X+8048 X+8056 X+8064X+80328032

32

Already know that timelines A & B move relative to each other.

35

Page 36: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 36McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

0 8 16 32

t1

t3

X+40 X+48 X+56 64X+32

t4

= 16 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

X+8040 X+8048 X+8056 X+8064X+8032

pDelayReq pDelayResp

t4 โ€ t1 = X+48

t3 โ€ t2 = X+16

XA

XBt2

8000 8008 8016 8032 8040

40

Phase changes of Local Clocks betweenpDelayReq & pDelayResp mean the two ends of the timeline can be treated as independent.

Because the left and right ends of the timeline can effectively be decoupled from each otherโ€ฆbecause there are different numbers of clock ticks in XA and XB (which is eventually compensated for by NRRโ€ฆwhich we arenโ€™t not currently considering right now)โ€ฆyou can look at the measurement probabilisticallyโ€ฆwhich looks like thisโ€ฆ

36

Page 37: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 37McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

t1

t2 ยฑ4ns

t4 ยฑ4ns

= 14 ns ยฑ4 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1

t3 โ€ t2

XA

XB t3TimestampGranularity

This shows only error due to Timestamp Granularityโ€ฆ8nsโ€ฆcentred on the 14ns, i.e. ยฑ4 ns

37

Page 38: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 38McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

t1 ยฑ4ns

t2 ยฑ4ns

t4 ยฑ4ns

= 14 ns ยฑ8 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1

t3 โ€ t2

XA

XB t3 ยฑ4nsTimestampGranularity

Add TX Timestamp Granularity.

38

Page 39: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 39McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

t1 ยฑ4ns + ยฑ2ns

t2 ยฑ4ns + ยฑ2ns

t4 ยฑ4ns + ยฑ2ns

= 14 ns ยฑ12 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1

t3 โ€ t2

XA

XB t3 ยฑ4ns + ยฑ2nsTimestampGranularity

Dynamic Timestamp Error

Add Dynamic Time Stamp Error, ยฑ2 ns.

Indications are that this error should be less than Timestamp granularity. It varies with implementation and Iโ€™ll let Kevin talk about that a bit more laterโ€ฆbut for the moment letโ€™s just model it as this probability.

If you add together all 8 uniform probabilitiesโ€ฆ

39

Page 40: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 40McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

t1 ยฑ6ns

t2 ยฑ6ns

t4 ยฑ6ns

= 14 ns ยฑ12 ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1

t3 โ€ t2

XA

XB t3 ยฑ6ns

pDelay Error ยฑ12ns 

Combined Timestamp Error

You get this.

At each TX and RX point, two uniform probabilities combine to give a triangular probabilityโ€ฆand those combine to an overall bell curve probability.

40

Page 41: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 41McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

t1

t2 t3

t4

= 14 ns (on average)(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1 = X+28 (on average)

t3 โ€ t2 = X (on average)

XA

XB

Which can be averaged out.

41

Page 42: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 42McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Link Delay Error โ€“ Observations

โ€ข Averaging of pDelay measurements should be mandatoryโ€ข Delivers meanLinkDelay with much lower error

โ€ข Helpful if implementations exhibit error probability distributions that can be averaged out to zero

โ€ข For Ethernetโ€ฆโ€ข The goal is to estimate a relatively fixed quantityโ€ข Very low bandwidth lowโ€pass filter is appropriate due to stability of pDelay. (IIR?)โ€ข Some dedicated startup behaviour may be appropriate to arrive quickly at a useful average

โ€ข For wirelessโ€ฆโ€ข More complex due to multiโ€path and potential movement of devices, butโ€ฆโ€ข Unlikely to have long chains of devices, or at least long chains of mobile devices.โ€ข Out of scope for the current discussion.

42

Page 43: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 43McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Link Delay & Timestamp Errors โ€“Current Simulation

43

Page 44: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 44McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

t1 ยฑ4ns + ยฑ2ns

t2 ยฑ4ns + ยฑ2ns

t4 ยฑ4ns + ยฑ2ns

= 14ns ยฑ12ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1

t3 โ€ t2

XA

XB t3 ยฑ4ns + ยฑ2nsTimestampGranularity

Dynamic Timestamp Error

This is the model we just saw. The simulation does not do this. Instead it doesโ€ฆ

44

Page 45: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 45McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

t1 ยฑ4ns + ยฑ8ns

t2 ยฑ4ns + ยฑ8ns

t4 ยฑ4ns + ยฑ8ns

= 14ns ยฑ24ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1

t3 โ€ t2

XA

XB t3 ยฑ4ns + ยฑ8nsTimestampGranularity

Dynamic Timestamp Error

This.

Dynamic Timestamp Error is ยฑ8nsโ€ฆbut itโ€™s either +8 or -8 with a 50:50 change of either for each and every timestamp. Thatโ€™s represented by the two vertical linesโ€ฆone at +8, one at -8.

Working out the probabilitiesโ€ฆ

45

Page 46: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 46McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

t1 ยฑ12ns

t2 ยฑ12ns

t4 ยฑ12ns

= 14ns ยฑ24ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1

t3 โ€ t2

XA

XB t3 ยฑ12nsCombined Timestamp Error

pDelay Error ยฑ24ns 

1

6

1

4 4

This is Case 1 (and most other cases) in 60802-garner-new-simulation-results-new-freq-stab-model-0421-v02.pdf, i.e. the most recent set of time series simulations.

And there is no averaging of pDelay in the current simulation. The latest pDelaymeasurement is passed directly into the Correction Field.

Ohโ€ฆand if you are wondering if this could result in a negative pDelay, i.e. the simulation of time travelโ€ฆyes, it could. But not with the current simulationโ€™s pDelay value of 150ns.

There were also three other simulations of interest in that same run of simulations.

46

Page 47: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 47McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

t1

t2

t4

= 14ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1

t3 โ€ t2

XA

XB t3Combined 

Timestamp Error

pDelay Error 0ns 

Case 13 in 60802-garner-new-simulation-results-new-freq-stab-model-0421-v02.pdf

This turned both errors to zero. Which, for Link Delay is the equivalent of what averaging pDelay measurements should accomplish.

47

Page 48: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 48McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

= 14ns ยฑ16ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1

t3 โ€ t2

XA

XBCombined 

Timestamp Error

pDelay Error ยฑ16ns 

t1 ยฑ8ns

t2 ยฑ8ns

t4 ยฑ8ns

t3 ยฑ8ns

1 6 14 4

Case 14 in 60802-garner-new-simulation-results-new-freq-stab-model-0421-v02.pdf

This removed Timestamp Granularity Errorโ€ฆbut kept Dynamic Timestamp Error.

48

Page 49: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 49McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

t1 ยฑ4ns

t2 ยฑ4ns

t4 ยฑ4ns

= 14ns ยฑ8ns(t4 โ€ t1) โ€ (t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1

t3 โ€ t2

XA

XB t3 ยฑ4nsCombined Timestamp Error

pDelay Error ยฑ8ns 

Case 15 in 60802-garner-new-simulation-results-new-freq-stab-model-0421-v02.pdf

And this did the oppositeโ€ฆkept Timestamp Granularity Error but removed Dynamic Timestamp Error.

49

Page 50: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 50McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

CaseTimestamp

Granularity / nsDynamic

Timestamp Error

1 ยฑ4 ยฑ8

14 0 ยฑ8

15 ยฑ4 0

13 0 0

pDelay Error ยฑ24ns 

pDelay Error ยฑ16ns 

pDelay Error ยฑ8ns 

pDelay Error 0ns 

Comparison of Analysis & Latest Simulations

1855 2841

496 733

27 32

96%

26%

1.4%

105%

27%

1.2%

1924 2711

max|dTE| 64 Hops / ns max|dTE| 100 Hops / ns

Note that the simulation removed the effects of Timestamp Granularity and Dynamic Timestamp Error from both Link Delay and Residence Time.

While Link Delay could get close to this via averagingโ€ฆyou canโ€™t do that with Residence Time. So youโ€™ll get some big benefits from averagingโ€ฆbut not quite as much as shown here.

50

Page 51: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 51McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Modelling Link Delay Error โ€“ Timestamp Granularity  & Dynamic Timestamp Errorโ€ข Model linear error with uniform distribution for both factors.

โ€ข Allow separate RX & TX values for both, i.e. four factorsโ€ฆ

โ€ข Add โ€œError Correction Factorโ€ to account for averaging.โ€ข One value to cover all errors from all four factors.

โ€ข 25% Effec ve โ†’ Removes 25% of the error.

Timestamp Granularity Dynamic Time Stamp Error

RX

TX

51

Page 52: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 52McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Formulae โ€“ pDelayerrorTS

๐‘š๐‘’๐‘Ž๐‘›๐ฟ๐‘–๐‘›๐‘˜๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ 1 ๐’Ž๐’†๐’‚๐’๐‘ณ๐’Š๐’๐’Œ๐‘ซ๐’†๐’๐’‚๐’š๐’†๐’“๐’“๐’๐’“๐‘ช๐’๐’“๐’“๐’†๐’„๐’•๐’Š๐’๐’ ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘† ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘๐‘…๐‘…

๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘†

๐‘ก4๐‘‡๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ก1๐‘‡๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ก3๐‘‡๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ก2๐‘‡๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ

2

๐‘ก1๐‘‡๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ป๐‘บ๐‘ฎ๐‘ฌ๐‘ป๐‘ฟ ๐‘ซ๐‘ป๐‘บ๐‘ฌ๐‘ป๐‘ฟ ๐‘ก2๐‘‡๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ป๐‘บ๐‘ฎ๐‘ฌ๐‘น๐‘ฟ ๐‘ซ๐‘ป๐‘บ๐‘ฌ๐‘น๐‘ฟ

๐‘ก3๐‘‡๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ป๐‘บ๐‘ฎ๐‘ฌ๐‘ป๐‘ฟ ๐‘ซ๐‘ป๐‘บ๐‘ฌ๐‘ป๐‘ฟ ๐‘ก4๐‘‡๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ป๐‘บ๐‘ฎ๐‘ฌ๐‘น๐‘ฟ ๐‘ซ๐‘ป๐‘บ๐‘ฌ๐‘น๐‘ฟ

52

Page 53: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 53McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Neighbor Rate Ratio Error

53

Page 54: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 54McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Time Sync โ€“ Elements & Relationships

Dynamic Time Sync Error

Time Stamp Granularity

Link Delay Error Residence Time Error

Dynamic Time Stamp Error

Neighbor Rate Ratio Error

Rate RatioError

Clock Drift

Time Stamp Granularity

Dynamic Time Stamp Error

Neighbor Rate Ratio Error

Clock Drift

54

Page 55: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 55McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

B (e.g. 98)

A (e.g. 99)

t3

t4

pDelayResp

t3' โ€“ t3

t4' โ€“ t4

YA

YB

pDelayRespโ€™

t3'

t4'

=NRRAB @ t4' (t3' โ€“ t3)

(t4' โ€“ t4)

NRRAB

The Platonic idealโ€ฆ

Note that A uses pDelayResp messages from B to A measure the Rate Ratio between the two (NRRAB). These are the same pDelayResp messages that are uses to measure Link Delay between A and B. (Iโ€™m just not showing the pDelayReqmessages). Also note that the intent here is to measure NRR at t4.

55

Page 56: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 56McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

B

A

t3

t4

t3' โ€“ t3

t4' โ€“ t4

1s

1s

t3'

t4'

localClkB125MHz + 10ppm

localClkB125MHz + 10ppm

localClkA125MHz โ€ 10ppm

localClkA125MHz โ€ 10ppm

874,998,750

1,875,001,250

1,000,000,000

2,000,000,000

=NRRAB @ t4' (t3' โ€“ t3)

(t4' โ€“ t4)=

124,998,750

125,001,250โ†’ โ€0.002%   =   โ€20 ppm

pDelayResp pDelayRespโ€™

NRRAB

Put some realistic number on itโ€ฆ

56

Page 57: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 57McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

t4 ยฑ6ns

t4 ยฑ6ns

B

A

t3' โ€“ t3

t4' โ€“ t4

1s

1s

localClkB125MHz + 10ppm

localClkB125MHz + 10ppm

localClkA125MHz โ€ 10ppm

localClkA125MHz โ€ 10ppm

874,998,750

1,875,001,250

1,000,000,000

2,000,000,000

t3 ยฑ6ns

t4 ยฑ6ns

=NRRAB @ t4' (t3' โ€“ t3)

(t4' โ€“ t4)=

124,998,750

125,001,250โ†’ โ€0.002%   =   โ€20 ppm

pDelayResp pDelayRespโ€™

NRRAB

These are the errors that we could model.

Timestamp Granularity Error & Dynamic Timestamp Error (shown combined)

Clock Drift (not the +10ppm and -10ppmโ€ฆbut the fact that those ppm values are changing over time, i.e. at the left end of the timeline, the points they arenโ€™t+10ppm and -10ppm)

57

Page 58: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 58McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

What factors to model?Factor YES / NO

Time Stamp Granularity& Dynamic Time Stamp Error

NOยฑ20ns over 1s = ยฑ0.02ppmWill be swamped by other errors.

Drift Rate of Local Clocks A & B (ppm/s)(Difference between Local Clock A ppm/s andLocal Clock B ppm/s at start)

YES

For most recent temp profile, max possible difference between two clocksโ€™ rate of change of ppm is ยฑ1.2ppm/s.ยฑ1.2ppm/s over 1s = ยฑ1.2ppmMay be a significant contributor to overall dynamic time sync error.

Rate of change of Drift Rate of Local Clocks A&B(Difference between Local Clocksโ€™ ppm/s at start & finish.)

NO

For most recent temp profile, max rate of change of ppm/s (i.e.ppm/s per second) is ~ยฑ0.006ppm/s2.Rounding up to ยฑ0.01ppm/s2, over 1s (or 5s) the change is not significant.(Note: ignores discontinuities at start and end of temperature ramp, which are not realistic.)

It turns out that the Timestamp errors will be small relative to other errors and can be ignored.

The same applies to the rate of change of clock drift (in ppm/s2), which is low enough that, over the periods of time we are interested in, clock drift can be treated as constant, i.e. linear.

58

Page 59: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 59McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Sources of Dynamic Time Sync Error

Dynamic Time Sync Error

Time Stamp Granularity

Link Delay Error Residence Time Error

Dynamic Time Stamp Error

Neighbor Rate Ratio Error

Rate RatioError

Clock Drift

Neighbor Rate Ratio Error

Clock Drift

So, we only need to look at Clock Driftโ€ฆ

59

Page 60: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 60McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

B

A

1s

1s

localClkB125MHz + 10ppm

localClkB125MHz + 10ppm

localClkA125MHz โ€ 10ppm

localClkA125MHz โ€ 10ppm

874,998,750

1,875,001,250

1,000,000,000

2,000,000,000

=NRRAB @ t4' (t3' โ€“ t3)

(t4' โ€“ t4)=

124,998,750

125,001,250โ†’ โ€0.002%   =   โ€20 ppm

pDelayResp pDelayRespโ€™

NRRAB

t4 ยฑ6ns

t4 ยฑ6ns

t3' โ€“ t3

t4' โ€“ t4

t3 ยฑ6ns

t4 ยฑ6ns

Which means we can remove the Timestamp Errors and also modify the diagram to remove the 1s discontinuity in the middle so that weโ€™re looking at the appropriate scaleโ€ฆ

60

Page 61: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 61McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

B

A

t3' โ€“ t3

t4' โ€“ t4

1s

localClkB125MHz + 10ppm

localClkB125MHz + 10ppm

localClkA125MHz โ€ 10ppm

localClkA125MHz โ€ 10ppm

874,998,750

1,875,001,250

1,000,000,000

2,000,000,000

t3

t4

t3'

t4'

=NRRAB @ t4' (t3' โ€“ t3)

(t4' โ€“ t4)=

124,998,750

125,001,250โ†’ โ€0.002%   =   โ€20 ppm

pDelayResp pDelayRespโ€™

NRRAB

Then squeeze things together to make a bit of roomโ€ฆ

61

Page 62: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 62McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

B

A

t3' โ€“ t3

t4' โ€“ t4

1s

localClkB125MHz + 10ppm

localClkB125MHz + 10ppm

localClkA125MHz โ€ 10ppm

localClkA125MHz โ€ 10ppm

874,998,750

1,875,001,250

1,000,000,000

2,000,000,000

t3

t4

t3'

t4'

=NRRAB @ t4' (t3' โ€“ t3)

(t4' โ€“ t4)=

124,998,750

125,001,250โ†’ โ€0.002%   =   โ€20 ppm

pDelayResp pDelayRespโ€™

NRRAB

62

Page 63: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 63McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

B

A

1s

localClkB125MHz + 10.6ppm

localClkB125MHz + 10ppm

localClkA125MHz โ€ 10ppm

localClkA125MHz โ€ 10ppm

874,998,750

1,875,001,250

1,000,000,000

2,000,000,000

0.6

0

ppmDriftB

โ€0.6ppm/s

=NRRAB @ t4' (t3' โ€“ t3)

(t4' โ€“ t4)=

124,998,750

125,001,250โ†’ โ€0.002%   =   โ€20 ppm

pDelayResp pDelayRespโ€™

NRRAB

t3' โ€“ t3

t4' โ€“ t4

t3

t4

t3'

t4'

Firstโ€ฆadd a -0.6ppm drift on Bโ€™s local clock.

63

Page 64: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 64McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

B

A

1s

localClkB125MHz + 10.6ppm

localClkB125MHz + 10ppm

localClkA125MHz โ€ 10ppm

localClkA125MHz โ€ 10ppm

874,998,713

1,875,001,250

1,000,000,000

2,000,000,000

0.6

0

ppmDriftB

Effective Measurement Point

โ€0.6ppm/s

=NRRAB @ t4' (t3' โ€“ t3)

(t4' โ€“ t4)=

124,998,750

125,001,288โ†’ โ€0.00203%   =   โ€20.3 ppm

pDelayResp pDelayRespโ€™

NRRAB

t3' โ€“ t3

t4' โ€“ t4

t3

t4

t3'

t4'

This means that Bโ€™s local clock was running faster at the left of the timeline than at the rightโ€ฆspecifically 0.6ppm faster.

Which means that there were more than 125,001,250 clock ticks between t3 and t3โ€™โ€ฆwhich means the fraction is smallerโ€ฆby 0.3ppm.

Only half of the ppm drift shows up in the NRR because you need to integrate under the curveโ€ฆor rather line, because we can treat clock drift as linear. Another way of thinking about this is that weโ€™re effectively measuring NRR half way between the current and the previous pDelayResp.

64

Page 65: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 65McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

B

A

1s

localClkB125MHz + 10.6ppm

localClkB125MHz + 10ppm

localClkA125MHz โ€“ 10.6ppm

localClkA125MHz โ€ 10ppm

874,998,713

1,875,001,288

1,000,000,000

2,000,000,000

0.6

0

ppmDriftB

0

โ€0.6

ppmDriftA

Effective Measurement Point

Effective Measurement Point

โ€0.6ppm/s

+0.6ppm/s

=NRRAB @ t4' (t3' โ€“ t3)

(t4' โ€“ t4)=

124,998,712

125,001,288โ†’ โ€0.00206%   =   โ€20.6 ppm

pDelayResp pDelayRespโ€™

NRRAB

t3' โ€“ t3

t4' โ€“ t4

t3

t4

t3'

t4'

Adding in a positive clock drift to Aโ€™s local clock has a similar effect.

65

Page 66: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 66McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

B

A

1s

localClkBdriftRateB

localClkB

localClkAdriftRateA

localClkA

=

mNRRerror (ppm)

2

(driftRateB โ€ driftRateA)*pDelayInterval

=2

(driftRateB โ€ driftRateA)*(t2' โ€“ t2)

(Difference between measured and actual Neighbor Rate Ratio @ t1)

(For simulation)

pDelayResp pDelayRespโ€™

NRRAB

t3' โ€“ t3

t4' โ€“ t4

t3

t4

t3'

t4'

66

Page 67: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 67McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Neighbor Rate Ratio Error โ€ Observations

โ€ข There is a difference between actual NRR and measured NRR (mNRR)โ€ข When clock rates are stable mNRR is very accurate

โ€ข Almost no difference between real NRR and mNRRโ€ข No large impact from timestamp granularity or errorsโ€ข No benefit from averaging mNNR (or taking a median of previous values) if (Timestamp Errors / pDelay Interval) is negligible in terms of ppm (it actually makes things worse)

โ€ข When clock rates are drifting mNRR may have significant errorsโ€ข Significant difference between real NRR and mNRRโ€ข Minimised when clock rates of neighboring devices drift in the same directionโ€ข Maximised when clock rates of neighboring devices drift in different directions

โ€ข PPM vs. Temp curve for XO means the latter could be more common than expected

โ€ข There is the possibility of compensating for Neighbor Rate Ratio Error

67

Page 68: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 68McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Modelling Neighbor Rate Ratio โ€“ 1

โ€ข Use formula from earlier to model NRRerror at pDelayReqโ€ฆ

โ€ข Model driftRate as stable (0ppm) or varying ยฑ0.6ppm (linear uniform distribution)

โ€ข X% stable; 100โ€X% varying

โ€ข Add โ€œError Correction Factorโ€ to account for modelling of drift (NRRdrift), correction of NRR (cNRR), and correction of other calculations via use of aNRR

โ€ข One value to cover all errors related to NRRerror and drift.โ€ข 25% Effec ve โ†’ Removes 25% of the error.

2

(driftRateB โ€ driftRateA)*pDelayIntervalmNNRerror =

68

Page 69: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 69McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Modelling Neighbor Rate Ratio โ€“ 2

โ€ข Actual error when NRR is used (NRRerror) will be larger than mNRRerror

if time has passed since measurement, allowing clocks to drift further apart.  Relevant for Residence Time Error.

โ€ข Model time Since mNRR as uniform distribution with min=0, max=pDelay.

โ€ข Apply same NRRerror correction factor to second termโ€ข 25% Effec ve โ†’ Removes 25% of the error.

NNRerror =   mNRRerror +   (driftRateB โ€ driftRateA) * time Since mNRR

69

Page 70: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 70McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Current Simulation

โ€ข Mean pDelay Interval: 31.25ms

โ€ข Would mean mNRRerror = ยฑ0.009375ppm with no average / median calculations, butโ€ฆ

โ€ข NRR is measured based on t3 & t4 from most recent pDelayResp messageโ€ฆand t3 & t4 from the pDelayResp 7 pDelay Intervals prior.  The median of the most recent measurement and the 6 prior measurements is then used as mNRR.

โ€ข Using t3 & t4 from the pDelayResp 7 pDelay Intervals prior pushes back the effective measurement point to 7/2, i.e. 3.5 pDelay Intervals prior to mNRR (vs. 0.5 pDelay Intervals prior, i.e. 7x worse).

โ€ข As rate of change of clock drift is (ppm/s2) is low, clock drift can be treated as linear, so the median will be the NRR from 3 pDelay intervals prior, i.e. the effective measurement point is an additional 3 pDelay Intervals prior.

โ€ข Worst case mNRRerror due to drift is therefore 13x worse: ยฑ0.12ppm

โ€ข pDelay Interval is short enough that Timestamp errors have an appreciable effect

โ€ข ยฑ24ns over 31.25ms is ยฑ0.77ppm

โ€ข Worst Case mNRRerror: ยฑ0.89ppm

โ€ข Effect of Clock drift on calculations made between NRR measurements is low due to low pDelay Interval

โ€ข ยฑ1.2ppm over 31.25ms is ยฑ0.0375ns

70

Page 71: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 71McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Formulae โ€“NRRerror

๐‘š๐‘๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘› ๐’…๐’“๐’Š๐’‡๐’•๐‘น๐’‚๐’•๐’†๐‘ฌ๐’“๐’“๐’๐’“๐’„๐’๐’“๐’“๐’†๐’„๐’•๐’Š๐’๐’

๐’‘๐‘ซ๐’†๐’๐’‚๐’š๐‘ฐ๐’๐’•๐’†๐’“๐’—๐’‚๐’2

๐’„๐’๐’๐’„๐’Œ๐‘ซ๐’“๐’Š๐’‡๐’• ๐‘› 1 ๐’„๐’๐’๐’„๐’Œ๐‘ซ๐’“๐’Š๐’‡๐’• ๐‘›

71

Page 72: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 72McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Link Delay Error โ€“Neighbor Rate Ratio Error

72

Page 73: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 73McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Time Sync โ€“ Elements & Relationships

Dynamic Time Sync Error

Time Stamp Granularity

Residence Time Error

Dynamic Time Stamp Error

Neighbor Rate Ratio Error

Rate RatioError

Clock Drift

Link Delay ErrorLink Delay Error

Neighbor Rate Ratio Error

73

Page 74: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 74McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

A

B

t1

t2 t3

t4

(t4 โ€ t1) โ€ mNRR(t3 โ€ t2)

2

pDelayReq pDelayResp

t4 โ€ t1

t3 โ€ t2

XA

XB

pDelay =

localClkA localClkA

localClkB localClkB

mNRRerror(t3 โ€ t2)

2pDelayerrorNRR (ns)   =  

74

Page 75: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 75McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

WORST CASE:

=   3ns

Conclusion: Model pDelayErrorNNR

NRRerror(t3 โ€ t2)

2pDelayerrorNRR (ns)   =  

10ms * (0.6ppm)

2pDelayerrorNRR (ns)   =  

75

Page 76: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 76McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Formulae โ€“ pDelayerrorNNR

๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘๐‘…๐‘… ๐‘š๐‘๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ

๐’‘๐‘ซ๐’†๐’๐’‚๐’š๐‘น๐’†๐’”๐’‘๐’๐’๐’”๐’†2

๐‘š๐‘’๐‘Ž๐‘›๐ฟ๐‘–๐‘›๐‘˜๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ 1 ๐’Ž๐’†๐’‚๐’๐‘ณ๐’Š๐’๐’Œ๐‘ซ๐’†๐’๐’‚๐’š๐’†๐’“๐’“๐’๐’“๐‘ช๐’๐’“๐’“๐’†๐’„๐’•๐’Š๐’๐’ ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘† ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘๐‘…๐‘…

76

Page 77: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 77McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Rate Ratio Error

77

Page 78: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 78McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Time Sync โ€“ Elements & Relationships

Dynamic Time Sync Error

Time Stamp Granularity

Residence Time Error

Dynamic Time Stamp Error

Neighbor Rate Ratio Error

Rate RatioError

Clock Drift

Link Delay Error

Neighbor Rate Ratio Error

Rate RatioError

Clock Drift

78

Page 79: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 79McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

RateRation = RateRationโ€1 * NRR

79

Page 80: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 80McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

1 2 3GM

Start with a โ€œsimpleโ€ analysis.

80

Page 81: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 81McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

1 2 3

125MHz +5 ppmDrift = 0 ppm/sNRR32 = +15 ppmmNRR32 = +15 ppm NRRerror32 = โ€0.3 ppm

125MHz โ€10 ppmDrift = 0 ppm/sNRR21 = โ€20 ppm

mNRR21 = โ€20.3 ppmNRRerror21 = โ€0.3 ppm

RR = 10 ppmmRR1 = +10.3 ppmRRerror1 = +0.3 ppm

RR = โ€10 ppmmRR2 = โ€10 ppmRRerror2 = 0 ppm

RR = +5 ppmmRR3 = +5 ppmRRerror3 = 0 ppm

GM

Drift = 0 ppm/s 125MHz +10 ppmDrift = โ€0.6 ppm/sNRR1GM = +10 ppm

mNRR1GM = +10.3 ppmNRRerror1GM = +0.3 ppm

RRerrorRange ppm

+0.3

0

Simple in thatโ€ฆif you feed NRR directly into RR, and donโ€™t account for time between measuring mNRR and using mNRR, i.e. applying it to Rate Ratio.

For pDelay Interval = 1s

โ€ฆwhich is the 802.1AS-2020 default value, and the value used in this presentation when looking at NRR error.

See earlier in the presentation for full analysis.

The RRerror at 1 is fully removed at 2. This makes sense as RR is trying to calculate the ratio of Local Clock to GM clock. If everything is working correctly, the error will depend on the drift between the twoโ€ฆnot the drift of clocks in the transmission path.

81

Page 82: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 82McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

1 2 3GM RRerrorRange ppm

+0.3

โ€0.3125MHz +5 ppmDrift = 0 ppm/sNRR32 = +15 ppm

mNRR32 = +15.6 ppm NRRerror32 = +0.3 ppm

125MHz โ€10 ppmDrift = +0.6 ppm/sNRR21 = โ€20 ppm

mNRR21 = โ€20.6 ppmNRRerror21 = โ€0.6 ppm

RR = 10 ppmmRR1 = +10.3 ppmRRerror1 = +0.3 ppm

RR = โ€10 ppmmRR2 = โ€10.3 ppmRRerror2 = โ€0.3 ppm

RR = +5 ppmmRR3 = +5 ppmRRerror3 = 0 ppm

Drift = 0 ppm/s 125MHz +10 ppmDrift = โ€0.6 ppm/sNRR1GM = +10 ppm

mNRR1GM = +10.3 ppmNRRerror1GM = +0.3 ppm

This continues to line up as we add Clock Drift to 2โ€ฆ

82

Page 83: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 83McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

1 2 3GM RRerrorRange ppm

+0.3

โ€0.3125MHz +5 ppmDrift = +0.6 ppm/sNRR32 = +15 ppmmNRR32 = +15 ppm NRRerror32 = +0 ppm

125MHz โ€10 ppmDrift = +0.6 ppm/sNRR21 = โ€20 ppm

mNRR21 = โ€20.6 ppmNRRerror21 = โ€0.6 ppm

RR = 10 ppmmRR1 = +10.3 ppmRRerror1 = +0.3 ppm

RR = โ€10 ppmmRR2 = โ€10.3 ppmRRerror2 = โ€0.3 ppm

RR = +5 ppmmRR3 = +5 ppm

RRerror3 = โ€0.3 ppm

Drift = 0 ppm/s 125MHz +10 ppmDrift = โ€0.6 ppm/sNRR1GM = +10 ppm

mNRR1GM = +10.3 ppmNRRerror1GM = +0.3 ppm

โ€ฆand 3.

83

Page 84: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 84McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

1 2 3GM RRerrorRange ppm

+0.6

0

125MHz +5 ppmDrift = +0.6 ppm/sNRR32 = +15 ppmmNRR32 = +15 ppm NRRerror32 = 0 ppm

125MHz โ€10 ppmDrift = +0.6 ppm/sNRR21 = โ€20 ppm

mNRR21 = โ€20.6 ppmNRRerror21 = โ€0.6 ppm

RR = 10 ppmmRR1 = +10.6 ppmRRerror1 = +0.6 ppm

RR = โ€10 ppmmRR2 = โ€10.3 ppmRRerror2 = 0 ppm

RR = +5 ppmmRR3 = +5 ppmRRerror3 = 0 ppm

Drift = +0.6 ppm/s 125MHz +10 ppmDrift = โ€0.6 ppm/sNRR1GM = +10 ppm

mNRR1GM = +10.6 ppmNRRerror1GM = +0.6 ppm

Adding Clock Drift to the GM is very different because it affects the Rate Ratio Error at every single downstream device.

Worse, it pushes them all in one direction, which could have further implicationsโ€ฆwhen you start adding in the next layer of Rate Ratio Errorโ€ฆ

84

Page 85: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 85McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

However, there is an additional source of error.  

(driftRateGM โ€ driftRatelocal) * pDelayIntervalRRerror (ppm)   =

2

Note: Effect of GM clock drift skews RRerror for all downstream instances in the same direction.This will have implications for how errors accumulate.

Therefore, model GM clock separately, i.e. separate ppm/s range vs. other clocks in the sysem.(A more accurate GM clock could address up to 50% of basic Rate Ratio related errors in the system.

Note that the latest time series simulations do not include any GM clock error.)

RateRatio is used during Time Sync processing.

RateRatio is effectively measured at the same time as NRR.  There is a time delay between measurement and useโ€ฆwhich is different at each node.

(NRR is measured as part of pDelay exchange.)

?

Which looks a lot like the NRRerror equation, but with driftRateGM instead of driftRatenโ€1

The โ€œsimpleโ€ analysis suggest that the equation at the top of the screen is sufficient. But the time delay means we need to dig deeperโ€ฆ

85

Page 86: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 86McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Time Sync โ€“ Elements & Relationships

GM TIMEEnd Station 

Time

pDelay Residence Time

INCOMING OUTGOING

CorrectionAdjustment

Neighbor Rate Ratio

Correction

Rate Ratio

Correction

Rate Ratio

Rate Ratio

pDelay

CorrectionAdjustment

Correction

Rate Ratio

Neighbor Rate Ratio

Correction

Rate Ratio

Rate Ratio

CorrectionAdjustment

Link Delay

86

Page 87: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 87McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

1 2 3GM

87

Page 88: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 88McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

1

2

3

GM

pDelayResp Sync

pDelayResp Sync

SyncpDelayResp

pDelayRsp Sync

t3โ€pDelayGM1

t4โ€pDelayGM1

t1โ€syncGM1

t2โ€syncGM1

t1โ€sync12

t2โ€sync12

t1โ€sync23

t2โ€sync23

t3โ€pDelay12

t4โ€pDelay12

t3โ€pDelay23

t4โ€pDelay23

t1โ€sync23

A more complex analysis includes different periods between pDelay and Sync messages. The minimum is close to zero (at scales of tens of hundreds of milliseconds) while the maximum is the pDelay Interval.

Note that in this graphic Iโ€™ve used an unrealistically long Residence Time to make it visible at this scale. The error analysis is unaffected.

88

Page 89: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 89McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Sync

t3โ€pDelayGM1

t4โ€pDelayGM1

mNRRGM1

Sync

1

2

3

GM

Sync

t1โ€syncGM1

t2โ€syncGM1

t1โ€sync12

t2โ€sync12

t1โ€sync23

t2โ€sync23

t3โ€pDelay12

t4โ€pDelay12mNRR12

t3โ€pDelay23

t4โ€pDelay23mNRR23

Sync

t1โ€sync23

mNRRGM1 addedto RR & used for Correction Field Adjustment

mNRR12 addedto RR & used for Correction Field Adjustment

mNRR23 addedto RR & used for Correction Field Adjustment

pDelayResp

pDelayResp

pDelayResp

pDelayRsp

NRR is measured at last pDelayResp message (mNRR)โ€ฆbut added to Rate Ratio at end of Residence Time (Sync message TX). This means that the amount of adjustment to Rate Ratio (i.e. NRR) is effectively measured at mNRR.

89

Page 90: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 90McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

t3โ€pDelayGM1

t4โ€pDelayGM1

mNRR1GM = +10.6 ppmmRR1 = +10.6 ppm

1

2

3

0.5s

GMt1โ€syncGM1

t2โ€syncGM1

t1โ€sync12

t2โ€sync12

t1โ€sync23

t2โ€sync23

t3โ€pDelay12

t4โ€pDelay12

t3โ€pDelay23

t4โ€pDelay23

t1โ€sync23

125MHz 0 ppmDrift = 0 ppm/s

125MHz +10 ppmDrift = โ€0.6 ppm/s

125MHz โ€10 ppmDrift = 0 ppm/s

125MHz +5 ppmDrift = 0 ppm/s

RR1

+10 ppm

RRerror1 = +0.6 ppm

RRerrorRange ppm

+0.6

0

ppmDrift1GMEffective Measurement 

Point

pDelayResp

pDelayResp

pDelayResp

pDelayRsp

Sync

Sync

Sync

Sync

Looking at the first stepโ€ฆ

mNRR is effectively measured pDelay Interval / 2 prior to the incoming pDelaymessage. When pDelay Interval is 1 second, thatโ€™s 0.5s prior to the mNRRmeasurement, i.e. it has an error of +0.3ppm.

But that is 0.5s prior to when the measurement is actually usedโ€ฆby which time there is an additional 0.5s of drift leading to an additional +0.3ppm of error, for a total of +0.6ppm.

90

Page 91: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 91McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

0

timeSync

t3โ€pDelayGM1

t4โ€pDelayGM1

mNRR1GM = +10.6 ppmmRR1 = +10.6 ppm

timeSync

1

2

3

0.5s

GM

timeSync

t1โ€syncGM1

t2โ€syncGM1

t1โ€sync12

t2โ€sync12

t1โ€sync23

t2โ€sync23

t3โ€pDelay12

t4โ€pDelay12mNRR21 = โ€20.9 ppmmRR2 = โ€10.3 ppm

t3โ€pDelay23

timeSync

t1โ€sync34

1s

125MHz 0 ppmDrift = 0 ppm/s

125MHz +10 ppmDrift = โ€0.6 ppm/s

125MHz โ€10 ppmDrift = 0 ppm/s

125MHz +5 ppmDrift = 0 ppm/s

RR1

+10 ppm

RR2

โ€10 ppm

RRerror1 = +0.6 ppm

RRerror2 = โ€0.3 ppm

RRerrorRange ppm

+0.6

ppmDrift21 Effective Measurement Point

โ€0.3

t4โ€pDelay23

pDelayResp

pDelayResp

pDelayResp

pDelayRsp

In the second step, the delay between measuring NRR and using the measurement is at itโ€™s maximum, i.e. a pDelay Interval of 1 secondโ€ฆwhich means the effective measurement point is 1.5 seconds prior to when the value is used.

The NRR error at the point when itโ€™s added to the Rate Ratio is therefore off by -0.9ppmโ€ฆwhich doesnโ€™t just cancel out the +0.6ppm error from the previous step, it overshoots by 0.3ppm.

Note that the calculations shown above do not explicitly include the effect of clock drift during Residence Time. However, this does not affect the error analysis.

โ€ข Clock1 will drift slightly (slower, since its drift rate is -0.6ppm) between t2-sync12

(where itโ€™s marked as +10ppm) and t1-sync23 where itโ€™s implicitly used in the calculation of RR2 butโ€ฆ

โ€ข Clock drift over a typical Residence Time (default 802.1AS maximum is 10ms) is not significant, e.g. ยฑ1.2ppm over 10ms is 0.012ppm. And even this can be ignored becauseโ€ฆ

โ€ข NRRerror and RRerror are driven by a clockโ€™s drift rate, not itโ€™s absolute value, and the rate of change of clock drift (ppm/s2) is low enough that, over the time periods of interest, clock drift can be considered linear. (See meanLinkDelayerror

for details.)

91

Page 92: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 92McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

timeSync

t3โ€pDelayGM1

t4โ€pDelayGM1

mNRR1GM = +10.6 ppmmRR1 = +10.6 ppm

timeSync

1

2

3

0.5s

GM

timeSync

t1โ€syncGM1

t2โ€syncGM1

t1โ€sync12

t2โ€sync12

t1โ€sync23

t2โ€sync23

t3โ€pDelay12

t4โ€pDelay12mNRR21 = โ€20.9 ppmmRR2 = โ€10.3 ppm

t3โ€pDelay23

t4โ€pDelay23mNRR32 = +15 ppmmRR3 = +4.7 ppmtimeSync

t1โ€sync34

1s

0.25s

125MHz 0 ppmDrift = 0 ppm/s

125MHz +10 ppmDrift = โ€0.6 ppm/s

125MHz โ€10 ppmDrift = 0 ppm/s

125MHz +5 ppmDrift = 0 ppm/s

RR1

+10 ppm

RR2

โ€10 ppm

RR3

+5 ppm

RRerror1 = +0.6 ppm

RRerror3 = โ€0.3 ppm

RRerror2 = โ€0.3 ppm

RRerrorRange ppm

+0.6

โ€0.3

pDelayResp

pDelayResp

pDelayResp

pDelayRsp

There is no relative drift between 2 and 3, so there is no error in the NRR measurementโ€ฆwhich means there is still nothing to cancel out the 0.3ppm error in Rate Ratio.

The basic mNRR errors from the previous section (the simple analysis) cancel out, because pDelay interval is constant (or close enough; if you want to be really accurate, youโ€™d need to model variation in pDelay interval, but for the purpose of this error model, weโ€™ll assume there is no variation).

But the errors due to drift can remain in the system. This is the same inputs as for the simple analysisโ€ฆbut with added delay. In the simple analysis the RRerrorRange

was ยฑ0.3ppm. In this case, itโ€™s already higher. And it can get worse.

92

Page 93: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 93McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

timeSync

t3โ€pDelayGM1

t4โ€pDelayGM1

mNRR1GM = +10.6 ppmmRR1 = +10.6 ppm

timeSync

1

2

3

0.5s

GM

timeSync

t1โ€syncGM1

t2โ€syncGM1

t1โ€sync12

t2โ€sync12

t1โ€sync23

t2โ€sync23

t3โ€pDelay12

t4โ€pDelay12mNRR21 = โ€21.8 ppmmRR2 = โ€11.2 ppm

t3โ€pDelay23

t4โ€pDelay23mNRR32 = +15.45 ppmmRR3 = +0.45 ppmtimeSync

t1โ€sync23

1s

0.25s

125MHz 0 ppmDrift = 0 ppm/s

125MHz +10 ppmDrift = โ€0.6 ppm/s

125MHz โ€10 ppmDrift = +0.6 ppm/s

125MHz +5 ppmDrift = 0 ppm/s

RR1

+10 ppm

RR2

โ€10 ppm

RR3

+5 ppm

RRerror1 = +0.6 ppm

RRerror3 = โ€0.75 ppm

RRerror2 = โ€1.2 ppm

RRerrorRange ppm

+0.6

โ€1.2

pDelayResp

pDelayResp

pDelayResp

pDelayRsp

Adding a +0.6ppm/s drift to 2โ€™s clock means the mNRR is off by -1.2pmm (the difference between 2 and 1โ€™s clocks) multiplied by 1.5s, which is the time from the effective measurement point (pDelay Interval / 2, i.e. 0.5s) plus the delay between mNRR and when itโ€™s used (a full pDelay Interval in this case, i.e. 1s).

-1.2ppm x 1.5s = -1.8ppm, so mNRR21 isnโ€™t -20ppm, or -20.6ppm, itโ€™s -21.8ppm.

On the previous slide mRR2 corrected mRR1 (of +0.3ppm) and overshot by 0.3ppm to end up at -0.3ppm. In this case, with 2โ€™s clock also driftingโ€ฆin the opposite direction to 1โ€ฆand such a long delay between mNRR21 and the Sync messageโ€ฆthe overshoot is now an additional 0.9ppm to end up at -1.2ppm.

Some of that is offset over the next hop by the difference in drift rate between 3 and 2. In the simple analysis of this scenario, RRerror2 was -0.3ppm and all of the error (i.e. 0.3ppm) was offset when moving on to RRerror3. In this scenario 0.45ppm is offsetโ€ฆwhich makes sense since the effective measurement point is 0.75s prior to when the measurement is used. But that still leaves -0.75ppm remaining, which will be passed down the chain of devices and, if every subsequent device has zero drift, will remain at the End Station.

93

Page 94: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 94McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

timeSync

t3โ€pDelayGM1

t4โ€pDelayGM1

mNRR1GM = +10.6 ppmmRR1 = +10.6 ppm

timeSync

1

2

3

0.5s

GM

timeSync

t1โ€syncGM1

t2โ€syncGM1

t1โ€sync12

t2โ€sync12

t1โ€sync23

t2โ€sync23

t3โ€pDelay12

t4โ€pDelay12mNRR21 = โ€21.8 ppmmRR2 = โ€11.2 ppm

t3โ€pDelay23

t4โ€pDelay23mNRR32 = +15 ppmmRR3 = +3.8 ppmtimeSync

t1โ€sync34

1s

0.25s

125MHz 0 ppmDrift = 0 ppm/s

125MHz +10 ppmDrift = โ€0.6 ppm/s

125MHz โ€10 ppmDrift = +0.6 ppm/s

125MHz +5 ppmDrift = +0.6 ppm/s

RR1

+10 ppm

RR2

โ€10 ppm

RR3

+5 ppm

RRerror1 = +0.6 ppm

RRerror3 = โ€1.2 ppm

RRerror2 = โ€1.2 ppm

RRerrorRange ppm

โ€1.2

+0.6

pDelayResp

pDelayResp

pDelayResp

pDelayRsp

Adding the +0.6pp drift to device 3 (as was done in the simple analysis) has the expected effect. There is no relative drift between 3 and 2, so NRRerror is zero and RRerror3 remains the same as RRerror2.

94

Page 95: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 95McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

timeSync

t3โ€pDelayGM1

t4โ€pDelayGM1

mNRR1GM = +11.2 ppmmRR1 = +11.2 ppm

timeSync

1

2

3

0.5s

GM

timeSync

t1โ€syncGM1

t2โ€syncGM1

t1โ€sync12

t2โ€sync12

t1โ€sync23

t2โ€sync23

t3โ€pDelay12

t4โ€pDelay12mNRR21 = โ€21.8 ppmmRR2 = โ€10.6 ppm

t3โ€pDelay23

t4โ€pDelay23mNRR32 = +15 ppmmRR3 = +4.4 ppmtimeSync

t1โ€sync34

1s

0.25s

125MHz 0 ppmDrift = +0.6 ppm/s

125MHz +10 ppmDrift = โ€0.6 ppm/s

125MHz โ€10 ppmDrift = +0.6 ppm/s

125MHz +5 ppmDrift = +0.6 ppm/s

RR1

+10 ppm

RR2

โ€10 ppm

RR3

+5 ppm

RRerror1 = +1.2 ppm

RRerror3 = โ€0.6 ppm

RRerror2 = โ€0.6 ppm

RRerrorRange ppm

โ€0.6

+1.2

pDelayResp

pDelayResp

pDelayResp

pDelayRsp

Note that a GM Clock Drift has a similar effect as in the simple analysis: it pushes every subsequent Rate Ratio measurement off in one direction, but in this case the effect is amplified by the initial delay between mNRR1GM.

95

Page 96: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 96McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Observations

โ€ข Effect of RRerror can be significant

โ€ข See section on effect of RRerror Residence Time Error for details

โ€ข RRerror due to mNRRerror cancels out at the next node

โ€ข RRerror due to additional Clock Drift between when NRR is measured (mNRR) and then used as part of Time Sync does not necessarily cancel out.

โ€ข Can accumulate to some degree. Should average to zero, unlessโ€ฆ

โ€ข GM clock drift affects every device in the chain, and in the same direction, i.e. affects the average

โ€ข Less time between pDelay message (used to measure NRR) and TimeSync is better

โ€ข Lower pDelay Interval is one method

โ€ข If pDelay Interval is < Sync Interval many pDelay messages are โ€œwastedโ€

โ€ข Setting pDelay to a multiple of Sync Interval and synchronising to send pDelay just before Sync would be more efficient

โ€ข Anything that can be done to compensate for NRRerror will flow through to RRerror

96

Page 97: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 97McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Modelling Rate Ratio Error

โ€ข Model two aspectsโ€ฆโ€ข Error from mNNRerror

โ€ข Errors out along the chain, nodeโ€toโ€nodeโ€toโ€node.

โ€ข RRerrorNRR never deviates by more than mNRRerrorRangeโ€ข See earlier for equation

โ€ข Error from drift between time of mNNR measurement and Time Syncโ€ข Should average out eventuallyโ€ฆbut with potentially large variances as the errors wonโ€™t cancel out nodeโ€toโ€nodeโ€toโ€node.

โ€ข RRerrorDrift may vary by a lot more than mNRRerrorRangeโ€ข Need more work to model and quantify this.

97

Page 98: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 98McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Formulae โ€“RRerror

๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘๐‘…๐‘… ๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐ท๐‘Ÿ๐‘–๐‘“๐‘ก

๐‘‘๐‘’๐‘™๐‘Ž๐‘ฆ๐‘š๐‘๐‘…๐‘…_๐‘†๐‘ฆ๐‘›๐‘~๐‘ˆ ๐‘š๐‘–๐‘› 0, ๐‘š๐‘Ž๐‘ฅ ๐’‘๐‘ซ๐’†๐’๐’‚๐’š๐‘ฐ๐’๐’•๐’†๐’“๐’—๐’‚๐’ 1 ๐’‘๐‘ซ๐’†๐’๐’‚๐’š๐‘น๐’†๐’”๐’‘๐‘บ๐’š๐’๐’„๐’„๐’๐’“๐’“๐’†๐’„๐’•๐’Š๐’๐’

๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘๐‘…๐‘… ๐’…๐’“๐’Š๐’‡๐’•๐‘น๐’‚๐’•๐’†๐‘ฌ๐’“๐’“๐’๐’“๐’„๐’๐’“๐’“๐’†๐’„๐’•๐’Š๐’๐’

๐’‘๐‘ซ๐’†๐’๐’‚๐’š๐‘ฐ๐’๐’•๐’†๐’“๐’—๐’‚๐’2

๐’„๐’๐’๐’„๐’Œ๐‘ซ๐’“๐’Š๐’‡๐’•๐‘ฎ๐‘ด ๐’„๐’๐’๐’„๐’Œ๐‘ซ๐’“๐’Š๐’‡๐’•

๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐ท๐‘Ÿ๐‘–๐‘“๐‘ก ๐‘› ๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐ท๐‘Ÿ๐‘–๐‘“๐‘ก ๐‘› 1 ๐’…๐’“๐’Š๐’‡๐’•๐‘น๐’‚๐’•๐’†๐‘ฌ๐’“๐’“๐’๐’“๐’„๐’๐’“๐’“๐’†๐’„๐’•๐’Š๐’๐’ ๐‘‘๐‘’๐‘™๐‘Ž๐‘ฆ๐‘š๐‘๐‘…๐‘…_๐‘†๐‘ฆ๐‘›๐‘ ๐’„๐’๐’๐’„๐’Œ๐‘ซ๐’“๐’Š๐’‡๐’•๐‘ฎ๐‘ด ๐’„๐’๐’๐’„๐’Œ๐‘ซ๐’“๐’Š๐’‡๐’• ๐‘›

๐‘‘๐‘’๐‘™๐‘Ž๐‘ฆ๐‘š๐‘๐‘…๐‘…_๐‘†๐‘ฆ๐‘›๐‘ is modelled as a uniform continuous distribution withminimum = 0 and a maximum = ๐’‘๐‘ซ๐’†๐’๐’‚๐’š๐‘ฐ๐’๐’•๐’†๐’“๐’—๐’‚๐’ 1 ๐’‘๐‘ซ๐’†๐’๐’‚๐’š๐‘น๐’†๐’”๐’‘๐‘บ๐’š๐’๐’„๐’„๐’๐’“๐’“๐’†๐’„๐’•๐’Š๐’๐’ . 

98

Page 99: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 99McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

pDelay Error โ€“ Rate Ratio Error

99

Page 100: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 100McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Time Sync โ€“ Elements & Relationships

Dynamic Time Sync Error

Time Stamp Granularity

Link Delay Error Residence Time Error

Dynamic Time Stamp Error

Neighbor Rate Ratio Error

Rate RatioError

Clock Drift

Link Delay Error

Rate RatioError

100

Page 101: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 101McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

pDelayErrorRR (ns) = pDelay * RRError

pDelayErrorRR (ns) = 150ns * (ยฑ10 ppm)  

= 0.0015ns (i.e. 1.5ps)

Conclusion: Do Not Model pDelayErrorRR

WORST CASE: RRError (ppm) = ยฑ10

150ns pDelay โ†’ 45m 

101

Page 102: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 102McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Residence Time Error โ€“Timestamp Errors& Rate Ratio Error

102

Page 103: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 103McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Time Sync โ€“ Elements & Relationships

Dynamic Time Sync Error

Time Stamp Granularity

Link Delay Error Residence Time Error

Dynamic Time Stamp Error

Neighbor Rate Ratio Error

Rate RatioError

Clock Drift

Time Stamp Granularity

Residence Time Error

Dynamic Time Stamp Error

Rate RatioError

103

Page 104: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 104McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

t1out

t2inA

Sync

t1out โ€ t2in

XB

Sync

localClkA localClkA

residenceTime = RateRatio (t1out โ€ t2in)

104

Page 105: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 105McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

B

CU

t1out ยฑ6ns

t2in ยฑ6nsA

Sync

t1out โ€ t2in

XB

Sync

localClkA localClkA

residenceTime = RateRatio (t1out โ€ t2in)

residenceTimeerror = t2inerror + t1outerror + RRerror (t1out โ€ t2in)

localClkA localClkA

GM

Sync

105

Page 106: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 106McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

residenceTimeErrorRR (ns) = (t1out โ€ t2in) * RRError

WORST CASE:

Conclusion: Model residenceTimeErrorRR

Conclusion: Model residenceTimeerrort2in and residenceTimeerrort2out

residenceTimeErrorRR (ns) = (t1out โ€ t2in) * RRError

= (10ms) * 10ppm

= 100ns

RRError (ppm) = ยฑ 10

150ns pDelay โ†’ 45m 

106

Page 107: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 107McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Formulae โ€“ residenceTimeerror

๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘† ๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘…๐‘…

๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘† ๐‘ก1๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ก2๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ

๐‘ก2๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ป๐‘บ๐‘ฎ๐‘ฌ๐‘น๐‘ฟ ๐‘ซ๐‘ป๐‘บ๐‘ฌ๐‘น๐‘ฟ ๐‘ก1๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ป๐‘บ๐‘ฎ๐‘ฌ๐‘ป๐‘ฟ ๐‘ซ๐‘ป๐‘บ๐‘ฌ๐‘ป๐‘ฟ

๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘…๐‘… ๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐’“๐’†๐’”๐’Š๐’…๐’†๐’๐’„๐’†๐‘ป๐’Š๐’Ž๐’†

107

Page 108: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 108McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Time Sync โ€“ Errors to Model

Dynamic Time Sync Error

Time Stamp Granularity

pDelay Residence Time

Dynamic Time Stamp Error

Neighbor Rate Ratio Rate Ratio

Clock Drift(Separate GM Model)

Link Delay Error Residence Time Error

Neighbor Rate Ratio Error

Rate RatioError

See Backup for Summary of Probabilities & Formulae

108

Page 109: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 109McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Summary of Main Insights

โ€ข meanLinkDelay should be an average of pDelay measurements.โ€ข Should nearly eliminate meanLinkDelayerror

โ€ข NRR should not be averaged.โ€ข Provided (Timestamp Errors / pDelay Interval) is negligible in terms of ppm

โ€ข There are significant benefits to the GM having greater ppm stability (lower ppm/s).

โ€ข It is hard (impossible?) to eliminate Timestamp granularity and Dynamic Timestamp error effects from Residence Time

โ€ข Need more work to see how far these elements can realistically be reduced

โ€ข Synchronising pDelayReq to arrive just before Sync would minimise the effect of Clock Drift.

โ€ข There may be opportunities to compensate for NRRerrorโ€ข And therefore RRerror

109

Page 110: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Recommended Next Steps

โ€ข Complete the Model

โ€ข Validate the Model using existing simulation(s) from Geoff

โ€ข Report back with a specific proposal containing 802.1AS parameters (e.g. rates etc.) and other recommendations that would meet the time accuracy goals for JP 60802 review

p. 110

110

Page 111: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 111McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Backup

111

Page 112: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 112McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Input Errors

Error DistributionDefault Min

Default Max Equation Unit

clockDriftGMClockDriftGMmin

& ClockDriftGMmax

X% Stable (zero)(100โ€X)% Uniform

โ€0.6 +0.6 ๐‘๐‘™๐‘œ๐‘๐‘˜๐ท๐‘Ÿ๐‘–๐‘“๐‘ก๐บ๐‘€~๐‘ˆ ๐‘๐‘™๐‘œ๐‘๐‘˜๐ท๐‘Ÿ๐‘–๐‘“๐‘ก๐บ๐‘€๐‘š๐‘–๐‘›, ๐‘๐‘™๐‘œ๐‘๐‘˜๐ท๐‘Ÿ๐‘–๐‘“๐‘ก๐บ๐‘€๐‘š๐‘Ž๐‘ฅ ppm/s

clockDriftX% Stable (zero)(100โ€X)% Uniform

โ€0.6 +0.6 ๐‘๐‘™๐‘œ๐‘๐‘˜๐ท๐‘Ÿ๐‘–๐‘“๐‘ก~๐‘ˆ ๐‘๐‘™๐‘œ๐‘๐‘˜๐ท๐‘Ÿ๐‘–๐‘“๐‘ก, ๐‘๐‘™๐‘œ๐‘๐‘˜๐ท๐‘Ÿ๐‘–๐‘“๐‘ก ppm/s

TSGETX Uniform โ€4 +4 ๐‘‡๐‘†๐บ๐ธ๐‘‡๐‘‹~๐‘ˆ ๐‘‡๐‘†๐บ๐ธ๐‘‡๐‘‹, ๐‘‡๐‘†๐บ๐ธ๐‘‡๐‘‹ ns

TSGERX Uniform โ€4 +4 ๐‘‡๐‘†๐บ๐ธ๐‘…๐‘‹~๐‘ˆ ๐‘‡๐‘†๐บ๐ธ๐‘…๐‘‹, ๐‘‡๐‘†๐บ๐ธ๐‘…๐‘‹ ns

DTSETX Uniform โ€2 +2 ๐ท๐‘‡๐‘†๐ธ๐‘‡๐‘‹~๐‘ˆ ๐ท๐‘‡๐‘†๐ธ๐‘‡๐‘‹, ๐ท๐‘‡๐‘†๐ธ๐‘‡๐‘‹ ns

DTSERX Uniform โ€1 +1 ๐ท๐‘‡๐‘†๐ธ๐‘…๐‘‹~๐‘ˆ ๐ท๐‘‡๐‘†๐ธ๐‘…๐‘‹, ๐ท๐‘‡๐‘†๐ธ๐‘…๐‘‹ ns

Formulae below take into account the unit of the input value.Formulae in the main presentation do not.

112

Page 113: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 113McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Other Variable Inputs

Error Default Value Unit

pDelay 150 ns

pDelayInterval 1,000 ms

pDelayResponse 10 ms

residenceTime 10 ms

Formulae below take into account the unit of the input value.Formulae in the main presentation do not.

113

Page 114: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 114McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Correction Factor Inputs

Error Default Value Unit

meanLinkDelayerrorTScorrection 0 %

driftRateErrorcorrection 0 %

pDelayRespSynccorrection 0 %

114

Page 115: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 115McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Formulae โ€“ Top Level

๐ท๐‘‡๐ธ ๐‘ฅ ๐‘š๐‘’๐‘Ž๐‘›๐ฟ๐‘–๐‘›๐‘˜๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘› ๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘›

๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘† ๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘…๐‘…

The initial model will not include factors shown in grey.x is number of hops.

ns

ns

ns

๐‘š๐‘’๐‘Ž๐‘›๐ฟ๐‘–๐‘›๐‘˜๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ 1 ๐’Ž๐’†๐’‚๐’๐‘ณ๐’Š๐’๐’Œ๐‘ซ๐’†๐’๐’‚๐’š๐’†๐’“๐’“๐’๐’“๐‘ช๐’๐’“๐’“๐’†๐’„๐’•๐’Š๐’๐’ ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘† ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘๐‘…๐‘… ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘…๐‘…

115

Page 116: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 116McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Formulae โ€“ meanLinkDelayerror

๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘†

๐‘ก4๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ก1๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ก3๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ก2๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ

2

๐‘ก1๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ป๐‘บ๐‘ฎ๐‘ฌ๐‘ป๐‘ฟ ๐‘ซ๐‘ป๐‘บ๐‘ฌ๐‘ป๐‘ฟ ๐‘ก2๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ป๐‘บ๐‘ฎ๐‘ฌ๐‘น๐‘ฟ ๐‘ซ๐‘ป๐‘บ๐‘ฌ๐‘น๐‘ฟ

๐‘ก3๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ป๐‘บ๐‘ฎ๐‘ฌ๐‘ป๐‘ฟ ๐‘ซ๐‘ป๐‘บ๐‘ฌ๐‘ป๐‘ฟ ๐‘ก4๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ป๐‘บ๐‘ฎ๐‘ฌ๐‘น๐‘ฟ ๐‘ซ๐‘ป๐‘บ๐‘ฌ๐‘น๐‘ฟ

๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘๐‘…๐‘…

๐‘š๐‘๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ

10๐’‘๐‘ซ๐’†๐’๐’‚๐’š๐‘น๐’†๐’”๐’‘๐’๐’๐’”๐’† ๐Ÿ๐ŸŽ๐Ÿ”

2๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘†

ns

ns

ns

ns

ns

๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘…๐‘…

๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ

10๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ 1 ๐’‘๐‘ซ๐’†๐’๐’‚๐’š๐’†๐’“๐’“๐’๐’“๐‘ช๐’๐’“๐’“๐’†๐’„๐’•๐’Š๐’๐’ ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘† ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘๐‘…๐‘… ns

๐‘š๐‘๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ

๐’‘๐‘ซ๐’†๐’๐’‚๐’š๐‘น๐’†๐’”๐’‘๐’๐’๐’”๐’†2

๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘†

10 ns

๐‘š๐‘’๐‘Ž๐‘›๐ฟ๐‘–๐‘›๐‘˜๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ 1 ๐’Ž๐’†๐’‚๐’๐‘ณ๐’Š๐’๐’Œ๐‘ซ๐’†๐’๐’‚๐’š๐’†๐’“๐’“๐’๐’“๐‘ช๐’๐’“๐’“๐’†๐’„๐’•๐’Š๐’๐’ ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘† ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘๐‘…๐‘… ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘…๐‘…

116

Page 117: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 117McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Formulae โ€“ residenceTimeerror

ns๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘† ๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘…๐‘…

๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘† ๐‘ก1๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ก2๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ

๐‘ก2๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ป๐‘บ๐‘ฎ๐‘ฌ๐‘น๐‘ฟ ๐‘ซ๐‘ป๐‘บ๐‘ฌ๐‘น๐‘ฟ ๐‘ก1๐‘†๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ป๐‘บ๐‘ฎ๐‘ฌ๐‘ป๐‘ฟ ๐‘ซ๐‘ป๐‘บ๐‘ฌ๐‘ป๐‘ฟ

๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘…๐‘…

๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ

10๐’“๐’†๐’”๐’Š๐’…๐’†๐’๐’„๐’†๐‘ป๐’Š๐’Ž๐’† 10 ๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘†

ns

ns

๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐’“๐’†๐’”๐’Š๐’…๐’†๐’๐’„๐’†๐‘ป๐’Š๐’Ž๐’†๐‘Ÿ๐‘’๐‘ ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’๐‘‡๐‘–๐‘š๐‘’๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘‡๐‘†

10

ns

ns

117

Page 118: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 118McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Formulae โ€“NRRerror

๐‘š๐‘๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘› ๐’…๐’“๐’Š๐’‡๐’•๐‘น๐’‚๐’•๐’†๐‘ฌ๐’“๐’“๐’๐’“๐’„๐’๐’“๐’“๐’†๐’„๐’•๐’Š๐’๐’

๐’‘๐‘ซ๐’†๐’๐’‚๐’š๐‘ฐ๐’๐’•๐’†๐’“๐’—๐’‚๐’๐Ÿ ๐Ÿ๐ŸŽ๐Ÿ‘

๐‘ต๐‘น๐‘น๐‘ฐ๐’๐’•๐’†๐’“๐’—๐’‚๐’๐’†๐’“๐’“๐’๐’“

๐Ÿ๐ŸŽ๐Ÿ— ๐’„๐’๐’๐’„๐’Œ๐‘ซ๐’“๐’Š๐’‡๐’• ๐‘› 1 ๐’„๐’๐’๐’„๐’Œ๐‘ซ๐’“๐’Š๐’‡๐’• ๐‘› ppm

๐‘ต๐‘น๐‘น๐‘ฐ๐’๐’•๐’†๐’“๐’—๐’‚๐’๐’†๐’“๐’“๐’๐’“ ๐‘ก4๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ก4๐‘๐‘Ÿ๐‘’๐‘ฃ๐‘–๐‘œ๐‘ข๐‘ ๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ก3๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ก3๐‘๐‘Ÿ๐‘’๐‘ฃ๐‘–๐‘œ๐‘ข๐‘ ๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ

๐‘ก3๐‘๐‘Ÿ๐‘’๐‘ฃ๐‘–๐‘œ๐‘ข๐‘ ๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ป๐‘บ๐‘ฎ๐‘ฌ๐‘ป๐‘ฟ ๐‘ซ๐‘ป๐‘บ๐‘ฌ๐‘ป๐‘ฟ ๐‘ก4๐‘๐‘Ÿ๐‘’๐‘ฃ๐‘–๐‘œ๐‘ข๐‘ ๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘ป๐‘บ๐‘ฎ๐‘ฌ๐‘น๐‘ฟ ๐‘ซ๐‘ป๐‘บ๐‘ฌ๐‘น๐‘ฟ

The factors ๐‘ก3๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ and ๐‘ก4๐‘ƒ๐ท๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ are the same as in the ๐‘๐ท๐‘’๐‘™๐‘Ž๐‘ฆ๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ calculation.(Same value. No need for new random numbers, as the same pDelayResp message is used to measure NRR.)

ns

ns

118

Page 119: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 119McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Formulae โ€“ RRerror

๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ ๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘๐‘…๐‘… ๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐ท๐‘Ÿ๐‘–๐‘“๐‘ก ppm

๐‘‘๐‘’๐‘™๐‘Ž๐‘ฆ๐‘š๐‘๐‘…๐‘…_๐‘†๐‘ฆ๐‘›๐‘~๐‘ˆ 0, 1 ๐’‘๐‘ซ๐’†๐’๐’‚๐’š๐‘น๐’†๐’”๐’‘๐‘บ๐’š๐’๐’„๐’„๐’๐’“๐’“๐’†๐’„๐’•๐’Š๐’๐’ ๐’‘๐‘ซ๐’†๐’๐’‚๐’š๐‘ฐ๐’๐’•๐’†๐’“๐’—๐’‚๐’

๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐‘๐‘…๐‘… ๐’…๐’“๐’Š๐’‡๐’•๐‘น๐’‚๐’•๐’†๐‘ฌ๐’“๐’“๐’๐’“๐’„๐’๐’“๐’“๐’†๐’„๐’•๐’Š๐’๐’

๐’‘๐‘ซ๐’†๐’๐’‚๐’š๐‘ฐ๐’๐’•๐’†๐’“๐’—๐’‚๐’๐Ÿ ๐Ÿ๐ŸŽ๐Ÿ‘

๐‘ต๐‘น๐‘น๐’Š๐’๐’•๐’†๐’“๐’—๐’‚๐’๐’†๐’“๐’“๐’๐’“

๐Ÿ๐ŸŽ๐Ÿ— ๐’„๐’๐’๐’„๐’Œ๐‘ซ๐’“๐’Š๐’‡๐’•๐‘ฎ๐‘ด ๐’„๐’๐’๐’„๐’Œ๐‘ซ๐’“๐’Š๐’‡๐’•

๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐ท๐‘Ÿ๐‘–๐‘“๐‘ก ๐‘› ๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ๐ท๐‘Ÿ๐‘–๐‘“๐‘ก ๐‘› 1 ๐’…๐’“๐’Š๐’‡๐’•๐‘น๐’‚๐’•๐’†๐‘ฌ๐’“๐’“๐’๐’“๐’„๐’๐’“๐’“๐’†๐’„๐’•๐’Š๐’๐’

๐‘‘๐‘’๐‘™๐‘Ž๐‘ฆ๐‘š๐‘๐‘…๐‘…_๐‘†๐‘ฆ๐‘›๐‘

10๐’„๐’๐’๐’„๐’Œ๐‘ซ๐’“๐’Š๐’‡๐’•๐‘ฎ๐‘ด ๐’„๐’๐’๐’„๐’Œ๐‘ซ๐’“๐’Š๐’‡๐’• ๐‘›

ppm

ppm

ms

The factor ๐‘ต๐‘น๐‘น๐’Š๐’๐’•๐’†๐’“๐’—๐’‚๐’๐’†๐’“๐’“๐’๐’“ is the same as in the ๐‘ต๐‘น๐‘น๐’†๐’“๐’“๐’๐’“ calculation as it is the same error.

Does not include any effect of changing clock drift (ppm/s2) duringResidence Time.  See speaker notes in ๐‘…๐‘…๐‘’๐‘Ÿ๐‘Ÿ๐‘œ๐‘Ÿ section for details.

119

Page 120: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 120McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Special Cases

๐’„๐’๐’๐’„๐’Œ๐‘ซ๐’“๐’Š๐’‡๐’• 0 ๐’„๐’๐’๐’„๐’Œ๐‘ซ๐’“๐’Š๐’‡๐’•๐‘ฎ๐‘ด

๐’“๐’†๐’”๐’Š๐’…๐’†๐’๐’„๐’†๐‘ป๐’Š๐’Ž๐’†๐’†๐’“๐’“๐’๐’“ ๐‘ฅ 0

For the first hop (๐’ ๐Ÿ), ๐’ ๐Ÿ ๐ŸŽ, i.e. the first device in chain, which is the GM.

For the last hop (๐’ ๐’™), the Sync message is not passed on so there is no Residence Time Error.

120

Page 121: The 60802 challenge of meeting time accuracy goals across ...

McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN / 60802 Interim, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

p. 121McCall, Stanton with Woods, Weingartner, and SchlechterIEEE 802.1 TSN Interim, Virtual, September 2021 Available at http://www.ieee802.org/1/files/public/docs2021

Of Interestโ€ฆ

โ€ข What happens when there is GM Clock Drift?โ€ข The ability to control clockDriftGM separately allows this to be investigated.

โ€ข How much do the various factors influence DTE?โ€ข Also: how do they accumulate along the chain?

โ€ข The separation of various factors until late in the process allows detailed analysis

โ€ข How effective are the various correction fields?

Careful construction of the model will help when it comes to extracting the values of interest (minimums, maximums, means, etcโ€ฆ) as the 

model is run, without taking excessive time or using excessive memory.

121