Networks using 10 GbE rely on more sophisticated modulation and coding techniques than networks needed at lower data rates.
4D-PAM5 (four-dimensional five-level pulse amplitude modulation) are the encoding and modulation techniques used in Gigabit Ethernet (GbE). 4D-PAM5 replaced the multilevel transmit MLT-3 encoding and modulation used for 100BASE-TX. IEEE 1000BASE-T (GbE) uses a combination of 8B1Q4 (8-bit to 1 four quinary [five] symbol), Trellis, Viterbi, and 4D methods for coding and decoding. This FAQ reviews the use of each of those tools in GbE, compares 4D-PAM5 with MLT-3, and closes with a glance at 4D-PAM16.
In GbE, the signal is first encoded using 8B1Q4; then, the 4D-PAM5 encoding method is applied. 4D-PAM5 coding uses five signaling levels and transmits on all four wire pairs at the same time, supporting a rate of 125 megabaud (symbols/sec) on each channel for an overall rate of 500 megabaud. Each symbol corresponds to two bits of data, resulting in a gross bit rate of 1,000 Mb/sec or 1 Mb/sec. The use of PAM5 is part of the continuing evolution in Ethernet encoding from the MLT-3 encoding used with 100 MbE Ethernet (Figure 1).
Redundancy is one of 4D-PAM5’s strengths. The scheme encodes two data bits per symbol on each channel, transmitting eight data bits per clock cycle. The number of unique signaling patterns needed to represent every 8-bit word is 28 = 256. With four channels and five signaling levels, producing 54 = 625 unique signal patterns is possible, giving 4D-PAM5 a significant amount of redundancy.
While MLT-3 uses two pairs of wire to achieve full-duplex operation, 4D-PAM5 achieves full duplex with a single wire pair. The large volume of traffic in 4D-PAM5 produces very complex voltage patterns that need management.
That’s where trellis encoding comes in. Trellis is a convolutional encoder that uses pseudo-random values to scramble the data and spread the transmission power as evenly as possible over the frequency spectrum. A redundant bit is added to the data so the receiver can recover the data even if some errors are introduced. This forward-error-correction (FEC) scheme is possible due to the highly redundant nature of 4D-PAM5 encoding, and it still leaves over 100 signal patterns for Ethernet control codes. At the receiver, the trellis encoding is decoded using the Viterbi algorithm.
What’s PAM16 used for?
Up to 1000BASE-T standard Cat5 cabling can be used. Cat5 cannot support higher bandwidths, and 10GBASE-T requires more complex PAM16 modulation, Tomlinson-Harashima precoding (THP), and more costly Cat6a cabling (Table 1). Fortunately, 10GBASE-T includes auto-negotiating and backward compatibility with 100BASE-T and 1000BASE-T, which lets users mix and match network functionality for specific use cases. The cost-benefit equation, however, changes significantly at 10GBASE-T.
At 10G, fiberoptic interconnects enter the picture. SFP+ copper interconnects have a lower installed cost compared with 10G SFP+ fiber interconnects. However, the copper solution consumes much more power, up to 2.5 W. In contrast, the maximum power consumption of a 10G optical transceiver is 0.8 W. Most data center operations will opt for lower energy consumption and lower operating costs with fiber optics even though they have a higher installation cost. Cat6a solutions are preferred on mixed networks that combine 100BASE-T, 1000BASE-T, and 10GBASE-T connectivity.
Summary
4D-PAM5 in IEEE 1000BASE-T GbE uses a combination of 8B1Q4, Trellis, Viterbi, and 4D methods for coding and decoding to replace the multilevel transmit MLT-3 encoding and modulation used for 100BASE-TX. Both 100BASE-TX and GbE can use Cat5 cabling. When speeds increase further to 10GBASE-T, Cat6a cabling must be used along with PAM16 modulation and Tomlinson-Harashima precoding.
References
10GBASE-T Ultimate Guide – Introduction, Cable, Pros and Cons, QSFPTEK
Line Coding Techniques, TechnologyUK
What is Ethernet, Teledyne Lecroy