Ethernet Gigabit Optical Experiments ==================================== Jonathan Thatcher IBM Corporation June 8, 1996 Overall Status/Summary ---------------------- Initial testing of 1.25 Gbaud link with 275 meters of 62.5 micron fiber had insufficient margins for reliable operation. >2 dB if modal BW penalty for > 275 meters of fiber lead to a larger link budget (8 dB). Receiver (O/E) did not have sufficient margin at E-10 BER to support increased link budget. Possible to recover 1 dB of modal BW penalty by controlling launch conditions. This was rejected as an alternative for "normal operation." Opportunity for "expert link configuraters" to come in to extend links. Technique to be documented in an informative appendix. More rigorous testing using 210 meters of 62.5 micron fiber later showed reasonable results. High probability of building an optical link at 1.25 Gbaud at the distances and budgets shown in the HP/Vixel/IBM combined presentation using existing components. Please remember: there is no SERDES in this analysis. Initial Test Configuration. --------------------------- "Best case" electronics; controlled cross-talk conditions (i.e. none); operation a nominal power and temperature. Optical attenuator used to normalize input to photodetector. Results ------- 1st Insert A 2 dB sensitivity penalty (or a 10E3 BER penalty) can be seen caused by the introduction of 275 m of 62.5 micron fiber as compared to 50 micron fiber jumpers. Remember that the photodetector was normalized to -18 dBm using the optical attenuator. The majority of this penalty can be gained back by putting a short, 3 meter, 50 micron jumper at the transmit end of the 275 m of 62.5 micron fiber. 2nd Insert The respective eye diagrams can be seen. The post O/E amplifier included filters to closely approximate a fourth order BT filter. Improvements To The Test ------------------------ Built up two test cards using "standard" components: Laser Driver 780 nm CD Laser PreAmp with Integrated GaAs Photodetector (PAID) Post Amplifier Multiple pattern generators were used to operate the two cards asynchronously to induce crosstalk. Lasers driven to maximum modulation. Frequencies maximally skewed Random data ran through one side and `1010' data ran through the other. Optical attenuator used to reduce input to photodetector. Power stressed from 4.0 to 5.7 volts (nominal for these parts is 5V +/- 10%). Four pieces of 62.5 micron fiber (70 m, 139 m, and two 3 m jumpers) used to increase number of connectors. In some experiments, a 300 ps rise time converter was placed on the pattern generator to slow down the rise time input to the laser driver. The input voltage level of the high speed signals to the laser driver were stressed beyond specification in both directions. Test set up shown in Insert 3 Results ------- 4th & 5th Inserts Before and After the 300 ps rise time converter. Rise and fall times still much faster than desired; some increase in jitter. 6th, 7th, and 8th Inserts Transmit Eye Diagram for Card 1 without filter. Transmit Eye Diagram for Card 1 using an 800 MHz filter. Transmit Eye Diagram for Card 1 using a 1000 MHz filter. 9th, 10th, and 11th Inserts Ditto, for Card 2. 12th, 13th and 14th Inserts "Eye" diagram of post amp output without and with 210+ meters of 62.5 fiber. Optical input to PAID attenuated to -20 dBm (3 dB below spec; really harsh!) Power/BER penalty of fiber and crosstalk. Jitter Budgets ala FC --------------------- Derived random jitter numbers are scaled from the percentage of UI specified in Fibre Channel Appendix J. These are only approximate, since they do not obey the sum of squares law This is for reference only. We may not use FC's budgets. ======================================================================== Jitter Budget Measured Limits UI @ 1.25 Gb/s ps Derived from FC ps (UI * 800) E/O Converter ("b" to "S") -------------------------- Random Jitter 71.7 160 0.20 Deterministic Jitter 0 96 0.12 Total Jitter 71.7 256 0.32 O/E Converter ("S" to "c") -------------------------- Random Jitter 213.4 248 0.31 Deterministic Jitter 13.9 64 0.08 Total Jitter 227.3 312 0.39 Jitter Budget Source Measurements --------------------------------- Jitter Budget Measured Limits UI @ 1.25 Gb/s ps Derived from FC ps (UI * 800) Serial Electrical Input (Point "b") ----------------------------------- Random Jitter 27.3 96 0.12 Deterministic Jitter 26.5 64 0.08 Total Jitter 53.8 160 0.20 Optical Output (Point "S") -------------------------- Random Jitter 76.7 184 0.23 Deterministic Jitter 26.2 160 0.20 Total Jitter 102.9 344 0.43 Serial Electrical Output (Point "c") ------------------------------------ Random Jitter 226.8 312 0.39 Deterministic Jitter 40.1 248 0.31 Total Jitter 266.9 560 0.70 =======================================================================