VOICE & TIMING SOLUTIONS For a New Global Network A Closer Look at PDV and Oscillator for Packet Equipment Clocks WSTS 2010 Peter Meyer [email protected] Frequency, Phase & Time Synchronization over Packet Networks When deployed and inter-connected within the packet network the packet equipment clocks will allow frequency, phase and time to be transferred over the packet network Different types of packet equipment clocks (PEC) – PEC-M the input is physical timing and the output is packet timing signal – PEC-B the input is a packet timing signal and the output is a packet timing signal – PEC-S the input is a packet timing signal and the output is a physical timing signal PRS/PRC PRTC PEC-B PEC-B PEC-B PEC-B PEC-B PEC-B PEC-M C O M M U N I C A T I O N P R O D U C TS P.2 Feb 03, 2010 PEC-B PEC-S Network Limits Network limits specify the maximum noise present on the timing signal at different points in the network Network Limits PRS/PRC PRTC PEC-B PEC-B PEC-B PEC-B PEC-B PEC-B PEC-M C O M M U N I C A T I O N P R O D U C TS P.3 Feb 03, 2010 PEC-B PEC-S Equipment Clock Characteristics Equipment clock specifications define requirements for clock engines Jitter & Wander (Noise) Generation ideal signal – Output from PEC when synchronized to noise-free input Jitter & Wander Transfer measured generation PEC known noise measured noise transfer – Output from PEC given a known input PEC Jitter & Wander Tolerance network noise limit error free operation – Error free operation of the equipment with a known input of maximum noise output PEC C O M M U N I C A T I O N P R O D U C TS P.4 Feb 03, 2010 PEC-S Functional Model ITU-T G.8263 (draft) Annex includes a functional model of a PEC-S packet-based clock PEC differs from traditional EC with introduction of a packet selection block has been included C O M M U N I C A T I O N P R O D U C TS P.5 Feb 03, 2010 PEC-S Functional Model: Packet Selection & Low Pass Filter Goal of the packet selection block is to select from all the input packets to the packet equipment clock a certain subset that are the least affected by the packet switched network These packets would thus best reflect the timing signal at the transmitter Both the packet selection block and the low pass filter function to remove noise from the packet timing signal to faithfully re-create the timing source The ‘cleaned’ timing signal can then be used to discipline the local oscillator Eliminate Noise from Packet Timing Signal C O M M U N I C A T I O N P R O D U C TS P.6 Feb 03, 2010 Network Limits & Characteristics – Metrics Performance of PEC-S depends heavily on PDV, or the noise on the packet timing signal How to define the network limits & characteristics for PDV – New metrics – Under study at ITU-T Study Group 15 Question 13 – Existing metrics – Definitions exist for delay variation, loss, availability, etc. – Driven by non-synchronization applications (voice, video, data, wireless) – 3GPP: ‘all’ selection (e.g. 99% of packets less than 10 ms delay) – MEF 10.2: “Ethernet Services Attributes Phase 2” FD, IFDV, FLR – ITU Y.1540: “IP packet transfer and availability performance parameters” – IETF RFC3393: “IP Packet Delay Variation Metric for IP Performance Metrics (IPPM)” IPDV C O M M U N I C A T I O N P R O D U C TS P.7 Feb 03, 2010 Classic PDV Histogram PEC-S Functional Model: Packet Selection vs. ‘All’ Selection Several proposals to select packets experiencing the least delay, known as the floor delay These packets deemed to be least affected by PDV and therefore would best filter the noise from the PSN Seems reasonable ….. – Example where even an odd band of packets (10%-30%), a subset of the total range, can improve the result C O M M U N I C A T I O N P R O D U C TS P.8 Feb 03, 2010 PEC-S Functional Model: Packet Selection vs. ‘All’ Selection … Except when then ‘floor delay’ is unstable or such ‘lucky’ packets are few and far between The larger the time between ‘lucky’ packets the lower the sample rate and the less ability to see local oscillator movement The use of more packets would allow to better filter the noise caused by the PSN and better track local oscillator movements Example of network comparing ‘all’ packets with ‘floor delay’ packets C O M M U N I C A T I O N P R O D U C TS P.9 Feb 03, 2010 Some Networks Resist Packet Selection xDSL networks, even if unloaded, have characteristics that resist the concept of floor packet selection xDSL native asymmetry in upstream and downstream direction Access technologies such as GPON and EPON have their own mechanisms to transfer frequency & phase that may be used – PEC-S in the OLT and a PEC-M in the ONT Early interest in xDSL to also natively transfer frequency & phase between DSLAM and modem (e.g. NTR) C O M M U N I C A T I O N P R O D U C TS P.10 Feb 03, 2010 PDV Topics to Investigate Handling of transient detection & suppression, to address dynamic packet network bursts, re-routes and outage phenomena Dealing with a variety of PSN PDVs that have different characteristics from slow movements to fast movements Tackling a variety of physical layer transport technologies such as xDSL Ramp Fast Ramp C O M M U N I C A T I O N P R O D U C TS P.11 Feb 03, 2010 Ramp Square On/Off Square On/Off Congestion The Local Oscillator Oscillator datasheets generally specify performance in a few categories – Frequency: clock output (e.g. 20 MHz or 25 MHz) – Initial offset: absolute frequency offset at manufacture – Ageing: relative frequency change over different periods of time, such as 24 hours and 10 years – Temperature: relative frequency change over a specified temperature range – Supply voltage: relative frequency change over a specified power rail voltage range – Load: relative frequency change over a specified load change – Warm-up: the time required after power-up to meet the specification values – can range from minutes to days to a month! – Overall: worse case absolute frequency over ‘lifetime’ due to all causes C O M M U N I C A T I O N P R O D U C TS P.12 Feb 03, 2010 The Local Oscillator – Specified in the Frequency Domain for use in the Time Domain Generally oscillators do not specify performance parameters in the time domain … whereas most ITU-T requirements are specified in the time domain! Oscillator performance specifications cover frequency variation in the frequency domain – May see ADEV specified over 1 to 100 second tau ITU-T specification requirements generally provided in time domain – TIE, MTIE, TDEV MTIE w/ppb 1.00E-04 TDEV 1.00E-05 1.00E-05 1.00E-06 MTIE (s) 1.00E-06 TDEV (s) 1.00E-07 1.00E-07 1.00E-08 1.00E-09 1.00E-08 1.00E-10 1.00E-09 1.00E-11 1.00E-02 1.00E-02 1.00E-01 1.00E+00 1.00E+01 1.00E+02 1.00E+03 1.00E+04 1.00E+05 Observation time (s) E1, TDEV, G.823, SEC E1, TDEV, G.8261, EEC Option 1 E1, TDEV, G.823, Sync PDH T1, TDEV, T1.101, OC-N Sync Ref T1, TDEV, G.824, Sync Reference T1, TDEV, T1.101, DS1 Sync Ref P.13 Feb 03, 2010 1.00E+00 T1, TDEV, G.824, Option 2 SEC T1, TDEV, G.8261, EEC Option 2 T1, TDEV, T1.105.09, SONET Ref 1.00E+01 1.00E+02 1.00E+03 1.00E+04 1.00E+05 1.00E+06 Observation time (s) E1, MTIE, G.823, SEC E1, MTIE, G.8261, EEC Option 1 E1, MTIE, G.823, Sync PDH E1, MTIE, G.8261, Case 1 E1, MTIE, G.8261, Case 2A E1, MTIE, G.8261, Case 3 E1, MTIE, G.823, Traffic PDH T1, MTIE, T1.101, OC-N Sync Ref T1, MTIE, G.824, Sync Reference T1, MTIE, T1.101, DS1 Sync Ref T1, MTIE, G.8261, Case 1 T1, MTIE, G.8261, Case 2a T1, MTIE, G.8261, Case 3 50 ppb T1, MTIE, G.824, Traffic PDH T1, MTIE, T1.403, Traffic PDH 16 ppb C O M M U N I C A T I O N P R O D U C TS 1.00E-01 1.00E+06 A Frequency Domain Look at a Filtered Oscillator 10 What does the oscillator look like after high frequency components have been filtered? σ (τ) (pbb) Underlying movement is more clear once the short term variation is filtered Existing oscillator specifications may useable for a definition of a mask for frequency accuracy, such as mobile basestation 1 Hz Loop Filter -9 y 10 Allan Deviation after 1 Hz Filtering -8 10 -10 10 0 10 2 10 τ (sec) 4 10 mHz Loop Filter 15 1 mHz Loop Filter 8 8 6 6 10 4 4 0 2 FFO (pbb) FFO (pbb) FFO (pbb) 5 0 2 0 -2 -2 -5 -4 -4 -10 0 1 2 3 4 Seconds 5 6 7 8 x 10 C O M M U N I C A T I O N P R O D U C TS P.14 Feb 03, 2010 4 -6 0 1 2 3 4 S econds 5 6 7 8 x 10 4 -6 0 1 2 3 4 Seconds 5 6 7 8 x 10 4 A Time Domain Look at an Oscillator What does the oscillator look like in the time domain? – How to specify the oscillator for the packet selection and loop filter? Method used to view the oscillator in the time domain information presented in the next slides – Measure Oscillator TIE in lab over 24hrs – Extract FFO using a Simulink model – Filter FFO based on different loop filter settings – Subtract frequency offset for calculated FFO and estimate TIE – Estimate TIE for each loop filter Setting – Calculate TDEV based on estimated TIE to measure TDEV to measure FFO – Calculate ADEV based on calculated FFO TIE data from oscillator + Loop Filter - + + D C O M M U N I C A T I O N P R O D U C TS P.15 Feb 03, 2010 A Time Domain Look at an Oscillator With a 10 mHz loop filter this oscillator has a low TIE and TDEV noise 10 mHz Loop Filter TDEV for 10 mHz Loop Filter 4 10 60 Computed TDEV G.824 Envelop (T7/F5) G.824 Envelop (T6/F4) G.823 Envelop (T11/F9) G.823 Envelop (T13/F11) 3 10 40 2 10 TDEV (ns) TIE (ns) 20 0 -20 1 10 0 10 -1 10 -40 -60 -2 10 -3 10 0 0.5 1 1.5 2 Seconds 2.5 C O M M U N I C A T I O N P R O D U C TS P.16 Feb 03, 2010 3 3.5 4 x 10 5 -2 10 0 10 2 4 10 10 Observation interval τ (seconds) 6 10 A Time Domain Look at an Oscillator With a 1 mHz loop filter, equivalent of Stratum 3E, there is significantly MORE noise contributed by the oscillator A lower the loop filter will filter LESS oscillator noise Cannot keep lowering the loop filter to be more robust against PDV without increasing the cost of the equipment! 1 mHz Loop Filter 10 600 400 10 TDEV for 1 m Hz Loop Filter 4 Com puted TDEV G.824 Envelop (T7/F5) G.824 Envelop (T6/F4) G.823 Envelop (T11/F9) G.823 Envelop (T13/F11) 3 TDEV (ns) TIE (ns) 200 0 10 10 2 1 -200 10 0 -400 -600 0 1 2 3 4 Seconds 5 C O M M U N I C A T I O N P R O D U C TS P.17 Feb 03, 2010 6 7 8 x 10 4 10 -1 10 -2 10 0 2 10 10 Observation interval τ (seconds) 4 10 6 Effect of the Local Oscillator on the Packet Selection ‘Cleaned’ packet timing signal used to discipline local oscillator Will the oscillator movement impact on the packet selection to reduce estimated performance – If there was originally a stable floor delay, how does it appear to move based on a non-ideal local oscillator? – What is inter-packet gap between selected packets and how should this be adjusted to match the non-ideal local oscillator? + Packet Delay (Zoom) C O M M U N I C A T I O N P R O D U C TS P.18 Feb 03, 2010 = Oscillator Observed Packet Delay (Zoom) Oscillator Topics to Investigate Beneficial to have oscillator performance specified in the time domain Optimizations for packet-based clocks requirements, which may be different than traditional electrical-based clocks (e.g. Stratum clocks) Matching well the metrics and loop filter selection to an acceptable price point – Contributions for PEC-S without on path support at ITU-T generally have selection windows and/or loop filters around 1 mHz (GR-1244 Stratum 3E / G.812 Type III) and 3 mHz (G.812 Type I) C O M M U N I C A T I O N P R O D U C TS P.19 Feb 03, 2010 VOICE & TIMING SOLUTIONS For a New Global Network Thank-you for Your Time & Attention WSTS 2010