1. Choosing the Right Optical PMD Solutions ------------------------------------------- Paul Kolesar Lucent Technologies for IEEE 802.3z, 9/96 This text further refines and clarifies the message I tried to convey at the meeting in Coeur d'Alene. In addition to providing supporting text for my slides, please note this document provides the following modifications to the handout material given in Coeur d'Alene: 1. In slide 3, the entry for Extended Horizontals under the ISO 11801 column has been corrected to indicate "300 proposed" to reflect the recent agreement by ISO/IEC JTC1/SC25/WG3 to consider including the contents of TIA TSB-72 "Centralized Optical Fiber Cabling Guidelines" in the next revision of ISO 11801. 2. In slide 4, the Modal Bandwidth entry under ISO 11801 has been appended with a note stating that a modification of the 200/500 specification to 160/500 is presently in ballot. 3. In slide 5, the standard cell from IEC 793-2 for the closest match of 50 um fiber to the requirements of the building cabling standards and the needs of 802.3z has been corrected. It is changed from 400/400 to 400/600 MHz-km to meet the minimum long wavelength specification of 500 MHz-km from the building cabling standards. 4. In slide 18, the recommended alternate fiber selection has been changed from 50 um with 400/400 bandwidth to 50 um with 400/500 bandwidth. This change includes the proviso that 802.3z request IEC to add a 400/500 cell to the 793-2 standard for 50 um fiber. Also, explanations have been added to the Distance Goals and Fiber Selections recommendations. 2. Outline ---------- - Review of Fiber Cabling Standards - Distance Needs and Capabilities - Short Distance PMD Choices - Equalization Techniques - Recommendations 3. Building Cabling Distance Specifications (meters) ---------------------------------------------------- Segment TIA 568-A ISO 11801 FD - BD 500 max 500 max BD - CD 1500 1500 FD - CD 2000 2000 (3000 SM) (3000 SM) Ext. Horiz. 300 300 proposed 4. Building Cabling Fiber Specifications ---------------------------------------- Fiber Spec. TIA 568-A ISO 11801 Preferred Type 62.5 62.5 Alternate Type not applic. 50 Modal Bandwidth 160/500 200/500 minimum (160/500 @ 850/1300 nm modification (MHz-Km) in ballot) 5. Standard-Cell Dual-Window Bandwidths from IEC 793-2 ------------------------------------------------------ Fiber Type 50um (A1a) 62.5um (A1b) Transmission wavelengths (nm) 850 1300 850 1300 Modal Bandwidth 200 400 160 200 categories in MHz 200 600(ISO) 160 500(TIA) referred to 1 km 400 400 200 200 (minimum) 400 600(802.3z) 200 400 400 800 200 600(ISO) 400 1000 250 1000 400 1200 - - 400 1500 300 800 600 1000(FC) - - The specification of fiber bandwidth performance should meet the minimum requirements of the building cabling standards, and be selected from the performance cells of IEC 793-2. The 160/500 cell for 62.5 um fiber exactly matches the TIA 568-A building cabling standard. The 200/600 cells for both 62.5 and 50 um fiber are the lowest performance cells to meet the present ISO 11801 cabling standard of 200/500. However, a proposed addendum to ISO 11801 may change the specification to match the 160/500 performance of TIA 568-A. This addendum in presently out for ballot and requires 75% approval to take effect. Even if it is approved, it will only change the selection for 62.5 um fiber, as the 200/600 cell for 50 um fiber is still the lowest performance selection to meet the 160/500 proposal. The 600/1000 cell for 50 um fiber is the lowest performance cell to meet the 500/500 specifications of Fibre Channel. This forces Fibre Channel to use one of the highest standard performance grades. Rather than follow the Fibre Channel example, I propose that 802.3z specify the 400/600 cell for 50 um fiber, or, preferably, request IEC to add a 400/500 cell. This will ensure adequate performance while not restricting the user to purchase the highest level of standard cell performance. 6. Fiber Backbone Distance Distribution (per Compaq Survey of 7/96) ------------------------------------------------------------------- This bar chart shows that of fiber runs within buildings 92% are shorter than 300 meters. Of fiber runs between buildings over 80% are longer than 500 meters. Less than 5% of all runs lie between 300 and 500 meters. This information is invaluable in determining the optimal alignment of the available optical solutions to the needs of the user, thus helping to ensure the greatest market acceptance of this standard. The dividing line between low-cost short-distance solutions and higher-cost longer-distance solutions is clear. If possible, the most cost-effective solutions should support distances up to at least 300 meters, leaving the longer distances for higher-cost solutions if necessary. This can facilitate a useful boundary between the deployment of the incompatible short and long wavelength solutions for MIS managers: short wavelength for in-building connections and long wavelength for between-building connections. Techniques that extend the range of the low-cost solution from 300 to 500 meters will help to meet the letter of the building cabling standards, but are of little practical value to customers. Therefore 802.3z should not preferentially recommend solutions to address distances longer than 300 meters, if those solutions result in more costly transceivers or operation on other than the installed base of 62.5 micron fiber. 7. Multimode Distance Limitations as a Function of BWmod, lc and lw ------------------------------------------------------------------- This graph plots the theoretical distance limitations of sped-up Fibre Channel transceivers (with center wavelengths (lc) of 850- and 780-nm and Full-Width Half-Max (FWHM) spectral widths (lw) between 5 and 9.4 nm) against the modal bandwidth (BWmod) of fiber ranging from 100 to 500 MHz-km. Interesting points are disclosed in the following two slides. Where certain combinations are inadequate to meet the distance objectives they are noted as needing a certain improvement factor or restriction on spectral width. 8. 300+ m Solutions for selected modal bandwidths ------------------------------------------------- lc(nm) Fiber Modal Bandwidth (MHz-km) 160 200 300 850 needs 1.2x improvement OK OK 780 needs 1.5x improvement needs 1.2x improvement OK 9. 500+ m Solutions for selected modal bandwidths ------------------------------------------------- lc(nm) Fiber Modal Bandwidth (MHz-km) 160 200 300 400 500 850 2.0x impr. 1.6x impr. lw < 5 nm OK OK 780 2.5x impr. 2.0x impr. 1.45x impr. lw < 5 nm lw < 7 nm 10. Short Distance Optical PMD Choices for Installed Base of 62.5um ------------------------------------------------------------------- For 300 m distances: - 850 nm on 200 MHz-km - 850 nm on 160 MHz-km w Res. Lch. or Eq. (1.2x) - 780 nm on 200 MHz-km w Res. Lch. or Eq. (1.2x) - 780 nm on 160 MHz-km w Res. Lch. and/or Eq. (1.5x) For 500 m distances: - 850 nm on 200 MHz-km w Res. Lch. and/or Eq. (1.6x) - 850 nm on 160 MHz-km w Res. Lch. and Eq. (2.0x) - 780 nm on 200 MHz-km w Res. Lch. and Eq. (2.0x) - 780 nm on 160 MHz-km w Res. Lch. and Eq. (2.5x) According the same Compaq survey shown previously, 84% of the premises network installed base in the U.S. is 62.5 um fiber, 7% is 50 um, and 9% is singlemode. The 62.5 um fiber is mostly of 160/500 MHz-km performance with some 200/500 MHz-km grade installed as a result of the improvement of fiber specifications by Lucent Technologies (AT&T) in 1994 in response to the ISO 11801 building cabling specifications. Using 850-nm transceivers provides an immediate solution to 300+ meters on the ISO-grade fiber. All other combinations require some amount of improvement. Attractive improvement techniques include restricted modal launch and receiver equalization. Restricted modal launch investigations have just begun within TIA FO2.2 with preliminary results varying widely from 10% to 700% improvement. Equalization techniques have been deployed for gigabit under sea lightwave links. Simple receiver equalization circuits can provide dramatic improvements shown later. The use of one or both of these techniques can potentially extend the reach of short wavelength solutions to fit the customers needs and the specifications of the cabling standards for in-building cabling distances. However, using only 850-nm devices has a few advantages as shown in the next slide. 11. Advantages of Using 850 nm Instead of 780 nm ------------------------------------------------ For 300+ m Solutions: - Immediate Solution for ISO 11801 Links (200 MHz-km) - Needs Only ~1.2x Improvement for TIA 568-A Links(160 MHz-km) achievable with Restricted Launch or simple Equalization For 500+ m Solutions: - Immediate Solution for 400 MHz-km Fiber - Workable on 160 MHz-km Fiber with Restricted Launch and simple Equalization 12. Power Penalties of Various Equalization Techniques ------------------------------------------------------ This slide is extracted from B.L.Kasper's work published in the September 1982 Bell System Technical Journal entitled "Equalization of Multimode Optical Fiber Systems". It plots the power penalties of four systems against the ratio of bit rate to channel bandwidth (R). Curve A represents a non-equalized system representative of today's Fibre Channel solutions. Curve B adds an 11-tap transversal (linear) equalizer in the receiver. Curve C adds a 5-tap decision feedback equalizer (DFE) to the transversal equalizer of curve B. Curve D uses the DFE alone. Also plotted are the equivalent penalties for multi-level encoding schemes. The curves show that a 5-tap DFE alone is nearly as effective as the same DFE in combination with an 11-tap transversal equalizer. In addition, using the DFE alone provides a 1.5 times improvement in R for no additional power penalty compared to the Fibre Channel techniques where R = 2. 13. Performance Comparison of 1- and 5-Tap DFEs ----------------------------------------------- This graph is also taken from Kasper's work cited above. It plots the performance of the 5-tap DFE against a simple 1-tap DFE. It shows that out to 1.75 times improvement (R = 3.5), the 1-tap circuit is within 0.4 dB of the 5-tap solution. The reason is that up to a bit rate of 3.5 times the channel bandwidth, almost all of the inter-symbol interference (ISI) is due to just the first preceding bit (provided sampling is done before the pulse peak). 14. How DFE Compensates for ISI ------------------------------- This slide graphically depicts the way DFEs work to eliminate ISI. It shows the data eye opening for systems with and without significant ISI and how the adjustment of the decision threshold based on the previous bit improves signal recovery in the receiver. 15. Block Diagram of 1-Tap DFE with Analog Feedback --------------------------------------------------- This slide shows a simple DFE consisting of a comparitor, a flip-flop and two resistors. The threshold of the comparitor is adjusted dynamically using a weighted amount of feedback from the flip-flop. It is extracted from work by J.H.Winters published in July 1992 in the Journal of Lightwave Technology entitled "Adaptive Nonlinear Cancellation for High-Speed Fiber-Optic Systems". 16. Block Diagram of 1-Tap DFE with Multiple Decision Circuits -------------------------------------------------------------- This slide shows another implementation where two decision circuits with different fixed thresholds can be used with a 2:1 mux to produce a 1-tap DFE. It is also extracted from Winter's work noted above. Both of these circuits can be built with off-the-shelf components to work at gigabit rates. 17. DFE Summary --------------- Fixed 1-Tap DFE is simple and effective - Works at gigabit rates - Adds few gates to receiver chip - Predetermined threshold levels (non-adaptive) - As effective as multi-tap out to 1.75x improvement Theoretical power budget impact - No penalty out to 1.5x distance improvement - 1.5 dB penalty at 2.0x distance improvement 18. Recommendations ------------------- Distance Goals - 300 m min. on MM for short wavelength solutions - 3000 m min. on SM (to align with building cabling standards) Fiber Selections - Preferred: 62.5um w 200/500 bandwidth (aligned w ISO 11801) - Alternate: 50um w 400/500 bandwidth (add cell to IEC 793-2) Short Wavelength Sources - 850 nm w restricted launch (1.2x for 160 MHz-km Fiber) - 780 nm w restricted launch and equalization (1.5x) Distance Characterizations on Multimode for Short Wavelengths - Relative to fiber bandwidth from 160 - 400 MHz-km These are my recommendations: 1. It is clear from the Compaq survey that low-cost short wavelength solutions can help ensure the success of this standard by extending their reach to cover the practical needs of in-building cabling out to at least 300 meters. Techniques that extend the range of the low-cost solution from 300 to 500 meters will help to meet the letter of the building cabling standards, but are of little practical value to customers. Therefore 802.3z should not preferentially recommend solutions to address distances longer than 300 meters, if those solutions result in more costly transceivers or operation on other than the installed base of 62.5 micron fiber. 2. The singlemode distance objective should be changed from the present 2000 meters to 3000 meters to harmonize with building cabling standards. This change will not impact the cost of the singlemode solutions proposed to date, as their power budgets are sufficient to support twice this length. 3. The multimode fiber choices should be aligned with ISO 11801 and IEC-793-2. The preferred fiber type should be 62.5 um with bandwidth ratings aligned with the minimum values specified by IS 11801. 62.5 um fiber should be the preferred fiber because: 1. it is the installed base media (85% in the U.S.), 2. it is the recommended fiber type by IEEE 802.8 for all of 802, 3. it is the preferred fiber type for 10BASE-F, 4. it is the preferred fiber type for 100BASE-FX, 5. it is the only multimode fiber specified by the TIA 568-A building cabling standard, 6. it is the preferred multimode fiber type specified by the ISO 11801 building cabling standard, 7. it causes no incompatibility or mixed-media management issues for the installed base. The alternate fiber type should be 50 um fiber with bandwidth of 400/500 MHz-km to meet the minimum specifications of IS 11801 and the needs of IEEE 802.3 for extending short wavelength solutions to meet the 500+ meter building cabling standard limitations for in-building backbones. IEEE 802.3 should request IEC to add the 400/500 MHz-km cell to IEC 793-2. If this is not accepted, then specify 50 um with 400/600 MHz-km performance. 4. For short wavelength sources the industry should converge on 850-nm technology for its performance advantages over 780-nm technology, and to avoid operating on the poorer performance fringe of the optical fiber short wavelength window. If 802.3z forces this convergence by narrowing the center wavelength specifications to exclude 780-nm devices, then I would be comfortable at this point to rely only on restricted launch techniques to provide the 1.2 times improvement required to boost the supportable distances on the installed base of fiber to the needed 300+ meters. The more likely scenario is that 802.3z will adopt the broad center wavelength specification of Fibre Channel, which allows both 780-nm and 850-nm devices. In this case, restricted launch techniques may be insufficient to provide the 1.5 times improvement required to reach the 300+ meter mark. If required, additional improvement can be obtained by using simple receiver equalization in combination with restricted! launch techniques. I am not suggesting that two different enhancement values should be applied to coexisting 850 and 780-nm devices, because that will cause interoperability problems. The needed enhancement will be determined by the lowest performance technology combination, and should be applied to all variants to avoid incompatibilities. 5. Distance specifications for short wavelength solutions should address the differences in the capability of the various performance grades in the installed base of multimode fibers. Distance specifications should not artificially limit the supportable distance to that of the lowest performance grade of fiber. To do so would rob the user of the performance they paid for in order to simplify the specification. Rather than divide the specification according to core size (assuming a certain bandwidth for each), distinguish distance capability according to modal bandwidth ratings. At a minimum, discrete distance specifications should be provided for 160, 200, and 400 MHz-km grades. In addition, a curve spanning the 160 to 600 MHz-km range would cover all of the standard cell grades of IEC-793-2, allowing the user to match the grade to its capability without over burdening the Gigabit Ethernet specification.