Measuring Time Error

www.calnexsol.com

Tommy Cook, CEO

• What is Time Error?

• Network devices.

• PRTC & Grand Master Clock Evaluation.

• Transparent Clock Evaluation.

• Boundary Clock Evaluation.

• Sources of uncertainty.

• Evaluation to G.8273.2 performance specification.

• Slave Clock Evaluation.

• Summary

Presentation overview

2

What is Time Error?

3

• Time Error measures the time difference between two clocks

• Time Interval Error measures change of time error• Starts at zero by convention, then tracks the dynamic time error

Time Error and Time Interval Error (TIE)

4

dTE

Time Error

0

cTE

Max|TE|

Time

Time Interval Error

0

Time Interval

• Time error generation of equipment clocks specified using four parameters:

• Constant Time Error (cTE) - specified in nanoseconds.

• Dynamic Time Error (dTEL) - specified with MTIE & TDEV masks,data pre-filtered by LP 0.1Hz.

• Dynamic Time Error (dTEH) - specified in nanoseconds,data pre-filtered by HP 0.1Hz.

• Max absolute time error (max|TE|) - specified in nanoseconds, data unfiltered.

• Network limits on time error at reference points specified using two parameters:

• Dynamic Time Error (dTEL) - specified with MTIE & TDEV masks,data pre-filtered by LP 0.1Hz.

• Max absolute time error (max|TE|) - specified in nanoseconds, data pre-filtered by LP 0.1Hz.

Time Error Specifications

5

Network Devices involved in Time delivery

6

Full Timing Support, (G.8275.1):

G.8275 Networks - Phase and Time Recovery

7

every switch/router on the path between T-GM and the T-TSC contains a T-BC

switch/routers containing a PTP T-BC

T-GM

T-GM

not all switch/routers on the path between T-GM and T-TSC contain a T-BC or T-TC

ordinary switch/routers

TC

Partial Timing Support, (G.8275.2):

PRTC and T-GM Testing

8

G.8272 defines the specification for a PRTC:• Max|TE| < 100ns, relative to the “applicable primary time standard” (e.g. UTC)

• dTE (or wander) limited by the MTIE mask:

Limits apply at both the 1pps phase output, and PTP time output if T-GM is integrated with the PRTC, (G.8272 Amendment 1).

PRTC Specification

9

How do you test a PRTC+T-GM?

10

PTP TimingMonitor

Device under Test:GNSS-Based PRTC+T-GM

PTP

PRC Ref.

Freq.

G.8272 Appendix I suggests three main ways of testing:

1. Comparison to a PRC frequency standard.

• Only addresses phase wander (i.e. dTE).

• Cannot verify max|TE| value.

2. Comparison to a reference receiver

• Addresses both phase wander and max|TE|.

• Uncertainty of the reference receiver may be too high.

3. Comparison to a national time standard

• Addresses both phase wander and max|TE|.

• Only available from a certified national time laboratory (e.g. NIST, NPL, PTB, etc).

PTP TimingMonitorDevice under Test:

GNSS-Based PRTC+T-GM

PTP 1pps+

ToD

PTP TimingMonitor

Device under Test:GNSS-Based PRTC+T-GM

PTP

NationalStandard

1pps+

ToD

• Produces an RF signal similar to what is seen by the receiver • Contains full orbital models of satellites

• Covers multiple constellations (e.g. GPS, Glonass, Beidou)

• Models atmospheric delays and disturbances

• Includes full position and navigation data for fixed or mobile receivers

• Produces a 1pps and ToD signal accurately synchronized to the RF signal• Enables timing receiver accuracy testing

GNSS simulators

11

Testing a PRTC with a GNSS Simulator

12

GNSS Simulator

Device under Test:GNSS-Based PRTC PTP Timing Monitor

Rb. Oscillator

PTP

RF

StableFrequency 1pps

Time-of-day

Reference Time InputTime

alignment to < 5ns

Transparent Clock evaluation

13

Packet Interfaces Siwtch/router

Switch/Router

Packet Interfaces

Transparent Clock

Residence Time BridgeMeasures time each PTP packet “resides” in the switch/router

What is a Transparent Clock?

14

SyncE SyncE

PTP Messages PTP Messages

Accuracy of the CorrectionField value:

Q1

Q2

Qn

Theoretical model:

•CorrectionField precisely reflects the delay through the equipment

• Ideal case – zero net PDV

Does it reflect the actual delay experienced by the Sync & Del_Req messages?

Types of CorrectionField inaccuracy;

1. Variable error• Caused by packet-to-packet variation in

CorrectionField accuracy.• Leads to residual PDV when PTP terminated.

2. Fixed Error

• Caused by CorrectionField value always being greater or less by a fixed value.

• Results in a fixed delay being measured when PTP terminated.

• Not a issue if Fixed Error is matched in forward and reverse direction.

• Differences between forward and reverse Fixed Error will produce asymmetry and hence create a Constant Time Error. 15

Evaluation Test BedRef. Time & Freq.

Slave

Transparent Clock Test Plan

16

Traffic Generator to load DUT

CaptureCapture

Master

Development Test Procedure to characterise actual performance

1. Measure the packet-by-packet latency across the TC.

2. Determine the change to the correctionField value for each message.

3. Accuracy is the difference in the actual latency compared to the change in Correctionfield value.

• Measure impact of CorrectionField on Sync PDV.

1. Vary traffic packet size.

2. Vary traffic priority.

3. Vary traffic utilisation.

• Repeat for Sync & Del_req PDV.Test in 1-Step and 2-Step modes.

G.8273.3 under development in ITU-T.

IEEE std C37.238-2011 PTP in Power Systems Applications.• Annex A:

TC TimeInaccuracy ≤50nsec.

T-TC

Boundary and Slave Clocks

17

G.8273.2 Telecom Boundary Clock (T-BC)

Slav

e

Mas

ter

EEC

SyncE SyncE

1pps

PTPPTP

Slave

Master

Time/Phase(1pps)

Frequency(T1/E1/SyncE)

Boundary Clocks reduce PDV accumulation by:

• Terminating the PTP flow and recovering the reference time

• Generating a new PTP flow using the recovered time

• No direct transfer of PDV

• Slave/Master combination

Telecom BCs (T-BCs) use SyncE to:

• Improve stability

• Improve holdover

T-BCs specified in ITU-T Recommendation G.8273.2

Boundary Clock, T-BC

18

G.8273.2 Telecom Boundary Clock (T-BC)

Slav

e

Mas

ter

EEC

SyncE SyncE

1pps

PTPPTP

Sources of uncertainty:• Timestamp noise in slave

• Phase noise and distortion in internal oscillator

• SyncE phase wander

• Path delay uncertainty in 1pps signal path

• Timestamp noise in master

Time Error is the combination of all these uncertainties

T-BC sources of uncertainty

19

e?

e?e?

e?

e?

PTP

G.8273.2 Telecom Boundary Clock (T-BC)

Slav

e

Mas

ter

EEC

SyncE Freq.T1/E1/SyncE

Phase1pps

PTP

Boundary Clock & Slave Clock comparison

20

G.8273.2 Telecom Time Slave Clock (T-TSC)

G.8273.2 Telecom Boundary Clock (T-BC)

Slav

e

Mas

ter

EEC

SyncE Freq.T1/E1/SyncE

Phase1pps

PTP

Slav

e

EEC

SyncE Freq.T1/E1/SyncE

Phase1pps

PTP

G.8273.2 Telecom Time Slave Clock (T-TSC)

• T-BC: Performance measured at the PTP, 1pps & Freq. outputs.• T-TSC: Performance measured at the 1pps & Freq. outputs.

• Important to verify performance on PTP output and 1pps output.

• When evaluating T-GM & T-BC:

• PTP used downstream therefore important as used to transfer time.

• 1pps Monitor point, but used once installed to verify behaviour and track down problems.

• Therefore, need to verify both!

Testing Time Error, 1pps or PTP?

21

PTP 2W cTE = 21nsec

1pps cTE = 207nsec

PTP T4 cTE = +162nsec

PTP T1 cTE = -119nsec

Boundary Clock evaluation

22

Five key elements to every ITU-T clock specification:

• Noise generation• The intrinsic noise generated by the clock itself with an “ideal” reference at

the input

• Noise tolerance• The maximum amount of noise the clock can tolerate at its input

• Noise transfer• The transfer function of the clock; usually defined as the bandwidth

• Transient response• The response of the clock to a transient at its input

• Holdover• How long a clock should maintain its output within specification after loss of

the input signal

For time clocks, “noise” means time error.

G.8273.2 T-BC Specification

23

Slav

e

Mas

ter

EEC

SyncE SyncE

1pps

PTPPTP

Maximum noise generation at both PTP and 1pps outputs:• Max|TE|: 100ns (class A), or 70ns (class B) (unfiltered)

• cTE: 50ns (class A), or 20ns (class B) (mean over 1000s)

• dTEL: 40ns MTIE, 4ns TDEV* (after filtering by 0.1Hz LP)

• dTEH: 20ns (after filtering by 0.1Hz HP) (proposed in draft revision)

G.8273.2 T-BC Noise Generation

24

Time

Tim

e Er

ror

Interval

Tim

e In

terv

al E

rro

r

“Ideal” Freq. Ref

Time

Tim

e Er

ror “Ideal” Time Ref

* - TDEV specified in Appendix hence not mandatory.

Slav

e

Mas

ter

EEC

SyncE SyncE

1pps

PTPPTP

With maximum “noise” on both PTP and SyncE inputs, check that:• No alarms are generated

• Clock does not switch references or go into holdover

• There are no performance limits to check

G.8273.2 T-BC Noise Tolerance

25

Time

Tim

e Er

ror

Interval

Tim

e In

terv

al E

rro

r

“Noisy” Freq. Ref

Time

Tim

e Er

ror “Noisy” Time Ref

PTP Noise tolerance:• T-BCs must tolerate the maximum dTE accumulation over the chain of T-BCs

• Specified using an MTIE mask in G.8271.1 Section 7.3:

SyncE Noise tolerance as defined in G.8262:

G.8273.2 Noise Tolerance Limits

26

G.8262-Y.1362(10)_F05

10

5.0

2.0

1.0

0.25

0.10.1 1.0 2.5 10 20 100 400 1000

MTIE [ s]m

Observation interval [s]t

Figure 5 from G.8262:Input wander tolerancefor EEC Option 1

MTIE

τ (s)

660ns

360ns

250ns

1µs

100ns0.01 0.0625 1 2.2 10 100 275 1000 100000.1

Figure 2 from G.8271.1:dTE network limit

Slav

e

Mas

ter

EEC

SyncE SyncE

1pps

PTPPTP

• Measures the transfer function of the clock, i.e. the bandwidth

• Measured by applying sinusoidal signals at different frequencies and measuring the relative attenuation

G.8273.2 T-BC Noise Transfer

27

Time

Tim

e Er

ror

Sinusoidal Time Error

Interval

Tim

e In

terv

al E

rro

r

Sinusoidal Wander

Time

Tim

e Er

ror

Slav

e

Mas

ter

EEC

PTP input is:• Noisy at high frequencies (e.g. packet jitter, timestamp quantization)

• Quiet at low frequencies (because traceable to a time reference)

Time Clock is a phase (time) locked loop:• Smooth out packet noise at high frequencies by following SyncE or local osc.

(i.e. PTP input is low-pass filtered)

• follows PTP input at low frequencies, correcting SyncE or oscillator wander (i.e. SyncE/oscillator noise is high-pass filtered)

SyncE input low-pass filtered by EEC, then high-pass filtered by Time Clock• Net result is a band-pass filter

G.8273.2: Noise Transfer

28

PTP

SyncE

T-BC

PTP to PTP flow

SyncE to SyncE flow

PTP

SyncE

1pps output

Noise Transfer• PTP to PTP:

Low pass filter, bandwidth between 0.05 to 0.1HzGain peaking ≤0.1dB

• SyncE to SyncE (G.8262): Low pass filter, bandwidth between 1 to 10Hz

• SyncE to PTP: Bandpass filter, lower cutoff 0.05 to 0.1Hz, upper 1 to 10HzGain peaking ≤0.1dB

G.8273.2: T-BC Clock Bandwidth

29

0.05 0.1 Frequency, Hz

Amplitude, dB

Amplitude, dB

0.05 0.1 Frequency, Hz1 10

Amplitude, dB

Frequency, Hz1 10

• PTP-to-PTP Transient Response ‘for further study’.

• Specifies that a T-BC should “reject” a SyncE transient on its input.• This means it should not follow the SyncE input until the signal is restored.

• Determined by monitoring the QL of the SyncE signal

• Defines an acceptable transient mask that the T-BC should remain within when a SyncE transient is applied at the input.

G.8273.2 Transient Response

30

T-BC output phase mask for first output phase error transient after start of SyncE rearrangement

Time after start of first output phase error transient (s)

0 5 10 15 20 25 30

Ma

xim

um

pha

se e

rror

(ns)

0

50

100

150

200

250

300

350

• G.8273.2 does not yet define holdover period for a T-BC

• Holdover provided at the end node rather than at each T-BC

• “Assisted” holdover may be provided using SyncE• Long term stability of the SyncE signal helps output to remain within

specification

• Even with assisted holdover, the clock will not remain in specification for more than a few seconds

G.8273.2 Holdover Specification

31

Slave clock evaluation

32

Five key elements to every ITU-T clock specification, (the same as T-BC but no PTP output to check, only 1PPS):

• Noise generation• The intrinsic noise generated by the clock itself with an “ideal” reference at

the input

• Noise tolerance• The maximum amount of noise the clock can tolerate at its input

• Noise transfer• The transfer function of the clock; usually defined as the bandwidth

• Transient response• The response of the clock to a transient at its input

• Holdover• How long a clock should maintain its output within specification after loss of

the input signal

For time clocks, “noise” means time error.

G.8273.2 T-TSC Specification

33

T-TSC Time Error Evaluation

34

Evaluation Test Bed

Impair

Capture

Ref. Time & Freq.

Capture

Master

TE: Difference between time output to ingress time;

• Max|TE| (unfiltered)• Constant TE• Dynamic TE.

Verify Performance specification in G.8273.2;

• Noise Generation

• Noise Tolerance

• Noise Transfer

• Transient Response

• Holdover

Slav

e

EEC Freq.T1/E1/SyncE

Slav

e

EEC

Phase

1pps, ToD

DUTT-TSC

• Time Error Metrics• Max|TE| (0.1Hz filtered for networks, unfiltered for equipment)• Constant TE, cTE• Dynamic TE; dTEL (MTIE & TDEV), dTEH (peak value)

• G.8272: PRTC and PRTC+T-GM performance parameters specified.• Time Error (max|TE|) & Wander Generation (dTE).

• G.8273.3 (draft): T-TC• Prove accuracy of inserted CorrectionField value.• {c37.238-2011 specifies accuracy today.}

• G.8273.2; Both T-BC and T-TSC performance parameters specified.• Noise Generation (max|TE|, cTE, dTEL, dTEH)• Noise Tolerance (dTE)• Noise Transfer (PTP-to-PTP, SyncE-to-PTP, SyncE-to-SyncE)• Transient Response (PTP-to-PTP, SyncE-to-PTP, SyncE-to-SyncE)• Holdover (tbd)

Summary

35

INTEGRITYPTP TIME MEASUREMENTS REQUIRE TRUE PRECISION

36

Tommy Cook

tommy.cook@calnexsol.com

+44 (0) 1506-671-416