In October 1995, two new 100-Mbit/s local area network standards were published by the Institute of Electrical and Electronics Engineers: IEEE 802.12, based on a demand priority (DP) access method, and IEEE 802.3u, based on a collision sense multiple access/collision detection (CSMA/CD) access method. Subsequently, there has been much interest in increasing the operating speed of these standards beyond 100 Mbits/s.[1] This imposes design challenges for the two media involved: optical fiber and copper cabling. Optical-fiber-based approaches are discussed elsewhere in this issue. Here we will examine schemes that are designed to use existing" copper cable installations, specifically data-grade cable, Category 5. This cable is already installed in locations that have foflowed building wiring standards.

A critical objective for the copper solution was to cost less than the fiber solution and this meant low complexity was required. We show that transmission at 424.8 Mbits/s using Category 5 cable can meet both industrial and the more stringent domestic emissions regulations. Furthermore, the design is robust in operation and of a complexity not greatly in excess of that used for the 100-Mbit/s rate.[2] The data rate of 424.8 Mbits/s is equivalent to the Fibre Channel rate of 531 Mbits/s before 8B10B coding (mapping 8 bits to 10 bits) and was chosen in anticipation of other IEEE 802 physical layers (PHYS) also fohowing the route of compatibility with Fibre Channel speeds to leverage existing components.

Compared with fiber, a copper system has a number of problems peculiar to it. A long metallic conductor is prone to act as a radio antenna, and this could lead to interference with other equipment (emissions) and unwanted pickup from other equipment (susceptibility). In addition, the copper cables are isolated by transformer coupling to avoid ground loops and other undesirable dc effects. These properties together with the transfer characteristic of the cable determine what can be successfully transmitted over the cable in a real-world environment.

Architectural Requirements: Speed, Bidirectionality

In a shared-medium access method such as DP or CSMA/ CD, full-duplex data transmission is not possible, since only one station (or none) has access to the shared channel at any point in time. However, half-duplex data transmission is suitable and an important refinement is possible. Network control traffic, including, for example, requests for access to the shared media, can be allowed to travel upstream when data flow is downstream, or vice versa. This helps the efficiency of the network, making it into a hybrid duplex scheme in which data and control can flow simultaneously in opposite directions, but neither is full-duplex.[3]

Having four pairs in one Category 5 cable means that there are alternative duplexing schemes to the traditional singlechannel frequency-division multiplexing (FDM) or hybridplus-echo-canceller approach. The bandwidth of one pair can be dedicated to a reverse control-signaling channel with the remaining three pairs for the forward data channel. This 3 + 1 scheme creates an asymmetric duplex scheme in terms of the bandwidth available in each direction. Having the still relatively high bandwidth of a single pair for control signaling is useful not because control traffic is high-bandwidth but because prompt detection of control codes is advantageous in terms of network performance. Asymmetric duplexing using selected pairs is notably less complex than FDM or hybrid plus echo canceller because of the lower component count. As shown in Figure 1, only two of the four pairs need be half-duplex; the other two can be simplex. Near-end crosstalk (NEXT) is no longer the dominant noise source in this Category 5 system as it was in the Category 3 100-Mbit/s systems.[5] Extenally induced noise is doniinant. Having a solution without FDM, echo cancellation, or NEXT cancellation is the pivotal step in forming a low-complexity system design.

Speeds greater than 100 Mbits/s are of interest for extending the existing standards, and in particular, speeds matching the Fibre Channel rates offer the possibility of leveraging existing components including drivers and clock recovery circuits. The first two Fibre Channel rates offering a marked speed increase over 100 Mbits/s are 531 Mbits/s and 1062.5 Mbits/s. However, these include the overhead of an 8B10B code designed assuming a single serial transmission medium. Since the copper solution divides the data among three pairs, the 8B10B code could be replaced with something more appropriate for this application. The raw data rates for the two Fibre Channel rates then become 424.8 Mbits/s and 849.6 Mbits/s. Extending these rates to the 3 + 1 asymmetric duplex scheme gives per-pair rates of 141.6 Mbits/s and 283.3 Mbits/s.

Signaling: Multilevel Signaling, Coding, Control

Earlier work[6] had shown that transmitting basic NRZ data at 155 Mbits/s was unlikely to satisfy domestic emissions regulations (e.g., FCC B) and might even prove problematical in meeting the less stringent industrial regulations (e.g., FCC A). Regulations begin at 30 Mhz, so reducing transmitter spectral energy above this frequency is an obvious approach to reducing emissions. Thus, a per-pair rate of 283.3 Mbits/s immediately seems far less suitable for a copper implementation than 141.6 Mbits/s. Several bandwidth compressing modulation schemes were studied, including quadrature amplitude modumon (QAM), partial response (PR) changes I and IV, and pulse amplitude modulation (PAM).[7,8] QAM is a two-dimensional scheme requiring complex in-phase and quadrature filters, and while PR has good bandwidth compression, this comes at the expense of clock recovery. PAM requires dc balancing but is a relatively straightforward scheme to implement and when m levels are used reduces bandwidth requirements by 1092(m). If excess bandwidth a is also reduced below 100% then an extra factor (1 + [alpha])/2 is gained to give the overall relationship: