Ethernet in the evolved fronthaul: synchronization and speed challenges

Transcription

1 intelligent Converged Network consolidating Radio and optical access arounduser equipment Ethernet in the evolved fronthaul: synchronization and speed challenges Nathan J. Gomes University of Kent New x-haul solutions for the 5G transport challenge A joint workshop of the icirrus, 5G-Crosshaul and 5G-Xhaul projects EUCNC 2017, Workshop 8, Monday 12 June Oulu, Finland This project has received funding from the European Union s Horizon 2020 research and innovation programme under grant agreement No (icirrus)

2 Acknowledgments This presentation describes work undertaken by several partners in the icirrus project. Particular thanks go to: Daniel Münch, Jörg-Peter Elbers ADVA Optical Networking Philippos Assimakopoulos, Mohamad Kenan Al-Hares University of Kent Volker Jungnickel, Kai Habel, Luz Fernandez, Malte Hinrichs Fraunhofer-HHI Philippos Assimakopoulos and Mohamad Kenan Al-Hares are also funded by the UK Engineering and Physical Sciences Research Council (EPSRC) NiRVANA project and DTA.

3 Outline Introduction: Evolved Fronthaul Architecture and the need for speed and synchronization Mechanisms for transport of high-priority/ptp messages Time-aware shaping (and comparison with Strict Priority) FUSION for deterministic latency High-speed, low-cost fronthaul DMT and PAM-4 Spectral upconversion technique Summary

4 Evolved Fronthaul Architecture Different RAN split points Will coexist with CPRI Fronthaul interface must transport different types of data and timing/synchronization Timing accuracy - order of 10ns (for MIMO schemes) Latency more relaxed (100µs or more) -> Playout buffers, as long as time is known accurately High-speed aggregated links are important

5 Strict priority and time-aware shaping Strict priority (e.g CM profile A) Queue with highest priority transmits first, then remaining queues in order of priority If HP frame arrives while LP queue is being served, HP frame must wait for current transmission to finish SP may be used only with extremely delay/jitter sensitive traffic such as PTP traffic, as can lead to starvation of LP queues Time-aware shaper (TAS) (e.g Qbv) TAS assigns specific window sections and allows only that traffic to pass through a bridge during the window (i.e. port gating). Prerequisite: overlaid time synchronization network Generic window section and subsection plan based on IEEE 802.1Qbv

6 Example OPNET results for SP and TAS Zoom-in and max PTP errors Low priority traffic (LP): constant frame-rate Frame size: Normal distribution, mean =1000 octets, variance =500 octets. Increasing guard period between best effort and priority sections can reduce and eventually eliminate FDV (and thereby PTP errors) tradeoff is link utilization No guard period: equivalent to Strict Priority Need to take into account windows for multiple hops (ranging protocols?) and timing inaccuracy in assigning windows and guard periods Monday 12 June, Oulu Finland

7 Switching / aggregation with Deterministic Latency using FUSION Alternative TSN approach investigated in icirrus Basic principle: Traffic is buffered, but for Guaranteed Service Transport (GST) gaps are preserved in aggregation (no FDV). Statistical Multiplexing (SM) traffic fills the gaps.

8 Theoretical analysis of FUSION approach (1)

9 Theoretical analysis of FUSION approach (2)

10 Initial experimental measurements of FUSION approach GST Experimental traffic shown measurements to experience of very end-to-end little latency variation (ony latency 136 ns with in this 10G case). inputs (2 SM, 1 GST) and on 100G trunk between 2 switches SM traffic experiences 2.4 µs latency variation.

11 Experimental Setup for High-Speed Evolved Fronthaul DSP Tx (offline) 25 GHz DAC 84 GS/s + - DA EML 27 GHz TEC & DFB-Current Control Offline DSP: Investigation of PAM-4 & DMT Data rates: up to 112 Gb/s Gray interface (1300 nm) Bias 16 GHz, 8bit resolution SSMF 0 km 20 km VOA & power meter 18 GHz, 8bit resolution ADC 84 GS/s + - PIN/ TIA 35 GHz DSP Rx (offline) Tx= transmitter DSP= digital signal processing DAC= digital analog converter GS/s= giga samples per second DA= driver amplifier EML= external modulated laser (distributed feedback laser + electroabsorbtion modulator) TEC= temperature control DFB= distributed feedback SSMF= standard single mode fiber VOA = variable optical attenuator PIN = positive intrinsic negative diode TIA = transimpedance amplifier ADC= analog digital converter Rx= receiver PAM= pulse amplitude modulation DMT= discrete multi-tone

12 Experimental Results DMT 100G PAM-4 BER= bit error rate ROP= received optical power Successful transmission over 10 km Higher EML output power (1 dbm) required to reach 20 km Tx-EQ: FFE21 Rx-EQ: FFE41

13 Investigation of spectral upconversion: Concept & Setup C. Kottke, et al. Performance of Single- and Multi-Carrier Modulation with Additional Spectral Up-conversion for Wideband IM/DD Transmission, Photonische Netze, Leipzig, 2017

14 Investigation of spectral upconversion: Results Performance of SC and MC modulation formats investigated SC-based modulation formats have better performance for flat channel MC-based modulations formats suited better to bandwidth limited channels Record high SC and MC transmission of 200 Gb/s and 224 Gb/s was reached

15 Summary TSN investigation Design of window sections in TAS taking into account variability in packet generation instances can reduce/eliminate latency variations Latency / Latency variation results from first measurements of within expected limits for FUSION deterministic latency approach Investigations of 100G high-speed data transmissions No performance difference between Nyquist PAM-4 and DMT at 112 Gb/s over up to 10 km SSMF with BERs below the CI-BCH FEC-threshold Record high SC and MC transmission of 200 Gb/s and 224 Gb/s respectively reached for a dual band system using IM/DD/spectral upconversion

16 Refined KPI table for fronthaul Different kind of traffics supported by the evolved fronthaul Fronthaul Requirements Data rate Max. latency (round-tripdelay) Legacy traffic (CPRI) 150 µs (CoMP) 440 µs (no CoMP) upper-phy split in down and uplink (no CoMP) 100 to 400 Gbps upper-phy split in downlink lower-phy split in uplink (CoMP) PDCP-RLC split 440 µs 150 µs 60 ms Min. frequency accuracy +/- 2 ppb (per hop) +/- 2 ppb (per hop) +/- 2 ppb (per hop) +/- 2 ppb (per hop) Min. phase and timing accuracy +/- 10 ns (MIMO & TX diversity) +/-1.36 µs (LTE TDD) +/- 30 ns +/- 30 ns not yet defined (expected to be about +/-1.36 µs (LTE TDD)) Max. latency imbalance +/- 16 ns +/- 163 ns +/- 163 ns not yet defined Max. BER

17 4.3 Switching / aggregating: Measurement Set-Up 10G Tester port A1 10G Tester port A2 GST link (GST: Guaranteed Service Transport) SM link (SM: Statistically Multiplexed Service) 10G 10G 10G Tester port B1 10G 100G 100G 10G 10G Tester port B2 10G 10G 10G Tester port C1 Aggregator / Switch (time sensitive) Aggregator / Switch (time sensitive) 10G Tester port C2 Latency / latency variation / Inter Packet Gap

18 100G: DMT DSP-steps Tx - DSP d(k) S/P M-QAM TS IFFT CP P/S Clipping Quantization Rx - DSP Synchr. S/P CP -1 FFT TS -1 Equalizer M-QAM -1 P/S d(k) Modulation formats Frame length FFT-Length 512 BPSK to 512-QAM 128 (124 DMT symbols + 4 training symbols) Cyclic Prefix (CP) 1/32 Clipping-Quotient Equalizer 12 db 1 tap, decision directed

19 100G: Nyquist PAM-4 DSP-steps Tx - DSP After driver After EML d(k) Mapping Level Adjust Resampling & Shaping Pre-Emphasize Quantization Rx - DSP Data rate: 112 Gb/s PAM4 Resampling Clk. Recovery Downsmpl. Normalization Equalization d(k) Sequence: four-level debruijn order 8, Gray-coded Shaping filter: Raised-Cosine with r=0.1 Tx equalizer: FFE with 21 taps Rx equalizer: FFE FFE=feed forward equalizer

20 100G: Nyquist PAM-4 Results Optimization of input power Distance: 0 km Tx-EQ: FFE21 Optimization of Tx-EQ & Rx-EQ ROP: 0 dbm Distance: 0 km Transmission over distance Tx-EQ: FFE21 Rx-EQ: FFE41 The family of lines shows the values for different BER BER= bit error rate EQ= equalizer FFE= feed forward equalizer FEC= forward error correction ROP= received optical power Rx= receiver Tx= transmitter

21 Investigation of spectral upconversion: Results II Performance of SC and MC modulation formats investigated SC-based modulation formats have better performance for flat channel MC-based modulations formats suited better to bandwidth limited channels Record high SC and MC transmission of 200 Gb/s and 224 Gb/s was reached