Appendix E: 8B/10B Line Coding
When you are transmitting digital data across a LAN, WAN, or other link, it is important to design the link with the following features:
1. No average DC term (DC-free) when data are viewed over several bytes. So, only AC components are present. DC cannot pass through optical fiber.
2. Capability to allow for clock and bit recovery from received data stream.
3. Minimum frequency spectrum shape for greater cable reach.
Transmitting raw, uncoded user data bit streams will not meet any of these requirements. Why? It is likely for long strings of ones or zeros to occur, so conditions 1 and 2 are not met. Also, some data sequences have very high frequency components. Following are two different ways to meet the three conditions just given:
• Use a block code. One example encodes 8-bit user data as 10-bit valued code words. In its most basic form, 256 8-bit user values are mapped into 256 × 10-bit code words. This coding type, used by Fibre Channel, is called 8B/10B, although there are other popular ones, such as 8B/14B (used by the common CD) and 64B/66B (10 gigabit Ethernet). Using 8B/10B requires a 25 percent higher line bit rate due to the overhead of the code words. The 1G Ethernet line rate is 1.250 Gbps and the payload rate is 1 Gbps using 8B/10B coding. Each of the 256 code words has an equal number of ones and zeros for all but four cases, but a DC-free balance is maintained over several code words. Sometimes payload rates are quoted (as with GigE), and sometimes line bit rates are quoted (as with Fibre Channel). So, buyer beware. IBM patented this coding concept in 1984.
• Use a scrambler to randomize and statistically balance data. This approach does not expand the message as 8B/10B does, so it is more efficient. However, it does not guarantee a perfect DC-free balance, although blocks of reasonable size will be “nearly” balanced to high probability. Condition 3 is not strictly met, but scrambling does generate a predictable spectrum shape.
Both methods are in wide use. The common SDI link (SMPTE 259M) uses the scrambling method. Some pathological data patterns have been known to cause a poorly designed receiver to lose clock recovery and generate bands of colors instead of the payload image. Scrambling does not always eliminate certain problem sequences that 8B/10B, for example, can deal with effectively.