So Good, it works on Barbed Wire

In 1995, I had the privilege of serving as the chief technical editor of the Fast Ethernet specification. In that capacity, I got to know many of the design teams working on chips to support the standard. The group from Broadcom was working particularly hard on an all-digital implementation of a subset of Fast Ethernet called 100BASE-T4 (different from the 100BASE-TX version that ultimately prevailed in the market).

Broadcom's T4 chip contained one of the first all-digital adaptive equalizers built for use on a fast serial link. Adaptive equalization was necessary in this case because the severely limited bandwidth of 100m of fairly low-grade Category 3 data wiring filters out all the fast edges in a signal, turning a crisp transmitted signal into slush at the far end of the cable. A good adaptive equalizer reverses the filtering effect of the cable, restoring the received data to its normal appearance.

By 1995, the use of digital adaptive equalization at lower speeds was well-established, as in telephone and satellite modems. These products used programmable-DSP cores to operate at speeds of approximately 100 kbps. In contrast to the programmable approach, Broadcom used dedicated digital state machines to speed those same algorithms by a factor of 1000.

The value of the Broadcom T4 design was its ability to work at high speeds on horrible cables. In many cases, you could use it on the cables you already had without upgrading. To demonstrate the power of what it felt was the world's best chip, Broadcom demonstrated its operation using the world's worst cable.

Paige demonstrates Fast Ethernet on barbed-wire cabling. In 1998, Wide-Band Systems demonstrates Gigabit Ethernet running on four pairs of old, rusty barbed wire.

At Interop that year, Broadcom set up a 2×4-ft glass case containing eight parallel strands of barbed wire configured as four differential pairs, each running straight from side to side, suspended in air. The wires were ugly and rusty and had nasty little barbs all over them. A transmitter and a reel of Category 3 data cabling were on one side of the case. The data cabling led to the glass case where it coupled onto the four barbed-wire pairs. The other side of the case coupled through more Category 3 cabling to a receiver.

During the show, lo and behold, Broadcom's demonstration flawlessly conveyed 100 Mbps of data through the barbed wire. "Buy our parts" was the message the Broadcom marketing folks wanted to impress on their audience.

I'd like you to receive a different message: Only four properties really affect the performance of most digital transmission structures. The "big four" transmission-line properties are impedance, delay, high-frequency loss, and crosstalk.

Crosstalk in a barbed-wire configuration is controlled by enforcing a large spacing between the pairs, as compared with the much smaller spacing between the individual wires of each pair. The glass case in Broadcom's demonstration was easily large enough to accommodate such spacing, so crosstalk wasn't a problem.

What about the high-frequency loss? It wasn't a problem either. The T4 system divides its data among the four pairs, so that each pair operates at only 25 Mbps. At that low frequency the skin-effect resistance of 4 ft of barbed wire is insignificant, and the overall high-frequency loss in the glass case at 25 Mbps was practically nil.

The signal delay is less on barbed wire than on an equivalent length of PVC-insulated Category 3 wiring, due to the use of an air dielectric between the barbed strands. This difference in delay was insignificant, however, because of the serial nature of the communications architecture.

Finally, if you consider the characteristic impedance, you find that this quantity is just a fixed number, such as 75 or 150 Ω. For most cables, it varies little with frequency of 1 to 100 MHz, but it varies significantly with the spacing between the wires. You can intentionally set the spacing to create almost any impedance you want. Inside the glass case, the spacing between barbed strands was set to create an impedance of 100 Ω, the same impedance as in the Category 3 UTP cabling on either side of the glass case. Thus, the case introduced no impedance discontinuity.

In summary, the barbed wire had zero impact on signal quality. The signals went through perfectly undistorted. The only thing the barbed wire did was impress the heck out of Broadcom's customers.

Next time you look at a transmission line, I hope you'll focus on the big four properties: characteristic impedance, high-frequency loss, delay, and crosstalk. These properties determine how well a transmission structure functions, regardless of the physical appearance or configuration of its conductors.


The author thanks Paul Sherer at 3Com for sponsoring his work on Fast Ethernet.