New synchronization metrics for packet networks

1. New synchronization metrics for packet networks Teemu Laine – Supervisor: Prof. Samuli Aalto 6.8.2010

2. Presentation • Objectives • Research methods • What is synchronization? • Mobile network synchronization • Different methods for sync delivery • Timing over Packet (ToP) • Synchronization metrics • MAFE • MATIE • Conclusions

3. Objectives of the thesis • Introduction to synchronization, especially mobile network synchronization over packet based backhaul networks (Mobile backhaul is currently experiencing transition from TDM based technology to packet based IP/Ethernet). • Introduce metrics for packet sync • Verification of performance of the new metrics • Develop improved phase sync metric

4. Research methods • Background and theory • Literature: standards (& contributions), white papers, books • Performance verification of metrics • Implementing several metrics for a math software • Implementing slave clock simulations (i.e. PLL filters etc)

5. What is synchronization in this context? • Frequency sync • Phase sync • Time sync (providing time sync is one way to achieve phase sync) • Phase-locked loop (PLL) locks the local oscillator to the received reference signal.

6. Mobile network synchronization • All about delivering accurate reference signal • Frequency sync is always required (50 parts-per-billion) in the air interface at the base stations (in GSM, WCDMA, LTE) • Time/phase sync is additional requirement in some cases (e.g. in TDD networks)

7. Methods for delivering the reference timing signal • PDH/SDH • Inherently on physical layer, does not work without sync • Synchronous Ethernet • Freq reference delivery on PHY layer • Satellite timing (e.g. GPS) • High accuracy time & frequency • Adaptive Clock Recovery • Constant bit rate pseudo wire over packet network • Timing over Packet (ToP) • Timing delivered in packets

8. Timing over Packet (ToP) Packet delay variation (PDV) is the biggest problem!

9. High load Fraction of packets Low load Delay Minimum delay Average delay Maximum delay Synchronization metrics • Traditional clock metrics and mask: MTIE and TDEV/MDEV metrics and PDH mask for e.g. MTIE and TDEV • Traditional metrics are calculated from TIE data (i.e. local clock’s phase compared to ideal reference) • Packet metrics can be calculated from packet delay data (also kind of phase information) • Packet sync metrics need always packet selection process!

10. MATIE/MAFE • New packet sync metrics, MATIE for phase sycn and MAFE for frequency • Two adjacent sliding windows • Values are averaged within windows and maximum error between the windows is stored • Slope between averaged error = MAFE

11. MAFE as a fucntion of observation window • Red = MAFE calculated from PDV data • Blue = MAFE calculated from packet slave clock output (oscillator stability = ~1000s) • Black= MAFE calculated from another packet slave clock output (with better oscillator)

12. MATIE • Works well according to simulated slave clock • But not optimal metric for satisfying MTIE mask • Thus, full MTIE calculation from clock simulations • MATIE modification only better on test data

13. Conclusions • MAFE is very applicable for estimating ToP performance (frequency sync). • MAFE is very robust (performance compared against linear regression, slave clock simulations, modifications and other competing metrics). • MATIE applicable in some cases, MTIE calculated from clock simulation output might be more useful. • Time sync delivery needs always support from intermediate nodes and two-way messaging • All in all, ToP is the most cost-efficient sync delivery method over PSN  deployed rapidly at the moment • Most of the results are presented in ITU-T standardization body