Gigabit Ethernet--Fiber-Optic Physical Layer 12 March 1996: Ethernet Plenary Stan Swirhun - Vixel Corp. (sswirhun@vixel.com) ============================================================== Goal: feasibility of PAR & assumptions (1250Mb/s) discuss open issues w.r.t. to physical layer implementation (i.e. balancing cost vs. performance) Physical Layer Issues -- Open Fiber Control (OFC) vs No - OFC -- Operation at 1250 Mb/s - distance limitations 62.5 micron fiber - MM fiber modal bandwidth: what is it? - effect of fiber dispersion on link length -- Start at Optical Specifications - technical issues affecting price - future developments over Gb/s Ethernet lifetime Technology vs. Physical Contraints -- Technology, primarily--- - SERDES & CDR function implementation: 10-bit interfaces work at 100Mb/s; 125Mb/s achievable - Analog electronics implementation -- Physical, primarily--- - Fiber media properties - especially modal bandwidth Specification Assumptions -- Speed: 8B/10B at 1000Mb/s = 1250Mb/s -- Media/cable plant: 300m-500m goal, 62.5 multimode base (50 MM future?) N typical connectors @ 0.2dB each -- Transmitter: short wavelength--CD laser; Vertical Cavity Surface Emitting Laser -- Safety: Eye safety compliance; OFC or no-OFC -- Optrical interface connector: existing, standardized OFC vs. No-OFC -- Non-OFC is preferred - fewer components, introduced as a "feature" - simplicity, eliminates cause of interoperability problems - eliminates autonegotiation (advantage?); allows loops -- OFC may provide one advantage: additional link budget - OFC launch = +1.3dBm to -7dBm receive = >-13dBm => 6dB budget - No-OFC launch = -5.0 dBm to -10dBm receive = >-16dBm => 6dB budget Fiber "Reach" at 1250Mb/s -- Limited by both fiber and source -- Laser-baseed multimode optical fiber bandwidth limited by: - fiber modal dispersion: 62.5 core fiber exhibits higher dispersion than 50micron core - rise and fall time of optical (laser) source: faster sources = longer links - margin and specified level of robustness -- BUT... - spec'd multimode fiber modal bandwidth UNDERESTIMATES true modal bandwidth achieved by low-mode laser launch (TIA working group to describe this) - fast optical sources (e.g. VCSELs) available--faster edge rates obtained with more power dissipation - ANSI Fibre Channel is "closest cousin" to GB/s Ethernet whereby Fibre Channel has taken a fairly conservative approach Feasibility: Experimental -- 850nm VCSEL-based Tx; Commercial integrated PD-preamp - <350ps rise/fall time - 3-6 fiber sections and connectors - measurements @ OC-24 (1.244Gb/s) and full speed FC (1.063Gb/s) - results same (not statistically significant) -- summary experimental BER penalties (eye center) - 62.5micron MM fiber: @ 300m fiber, <0.2dB delta (1.06 to 1.24Gb/s) @ 1.24Gb/s, <1.0dB delta (10m to 500m) - 50micron MM fiber: @ 1.24 Gb/s, <0.4dB delta (500m to 900m) @ 1.24Gb/s, <0.6dB delta (10m to 500m) -- Conclusions: Less than 1dB of dispersion penalty @ 62.5 MMF suggests that modal bandwidth is greater than "worst case spec"; simulation predicts >1.8dB dispersion penalty with 500m @ 400MHz-km Modeling Reach @ 1250Mb/s -- Assumptions - fiber: minimum specified 62.5micron MMF: 160-200MHz-km mimimum specified 50micron MMF: 400-500MHz-km - sources: VCSEL = 850nm, <300ps edge rates CD Laser = 780nm, <400ps edge rates - budgets: Link budget = 6dB=>Cable plant budget = 4dB, Penalty budget = 2dB -- Results - Maintaining 2dB penalty budget with VCSEL is feasible (just) - Maintaining 2dB penalty budget with slower source is "iffy" Optical Budget Issues -- Candidate Approach - Short-wave FC spec: 4dB cable plant + 2dB penalties - FC-like eye mask template? -- Cable Plant Loss - Maximum of 4dB cable plant loss 62.5mic MMF @ 4.5dB/km x 0.5km = 2.3 dB 8-11 MM connectors @ 0.15dB +0.30/-0.15 = 1.7dB -- Receiver Sensitivity - Must be specified w.r.t. a BER (10**-12)? - "penalties" reduce AC signal or increase noise: a) eye closure penalty due to fiber BW, laser edge rates (rise/fall) and finite receiver bandwidth == ~0.8dB b) extinction ratio (assumed in test, negligible) c) noise sources due to modal noise, RIN == ~0.8dB d) off center eye closure and jitter == ~0.4dB Price Penalties Occur... -- If receiver sensitivity required < -16dBm, price may suffer... - few % increment for each additional dB in sensitivity - low-cost integrated detector/pre-amp may reach fundamental performance floor near -20dBm -- If OFC is required price may suffer... - 5% increment for inclusion of OFC circuit - few % decrement - for laser and optical sub-assembly specs - package size, reliability, power dissipation, negatively impacted Conclusions -- Feasibility of 1250Mb/s - 300-500m - 62.5MMF ? - Yes, with work; several issues - likely require > 6dB link budget - demand better specification on fiber bandwidth