Multimode Fiber Gets a New Lease on Life
Dr. Stuart Walker and his graduate student, Pandelis Kourtessis, showed that it's possible to send an aggregate capacity of 204 Gbit/s over a distance of 3 km over multimode fiber. They say the bandwidth distance product of 0.6 Tbit/s/km is a world record -- far outreaching the previous record of 0.016 Tbit/s/km set by Lucent Technologies Inc. (NYSE: LU) in October 1999.
The press release claiming a world record confuses things a little by claiming 404 Gbit/s (see Boffins Set a Record). In fact, only 204 Gbit/s was available to carry data. The other 200 Gbit/s was comprised of the baseband signal (more on that in a minute).
Walker's aim was to show that there's plenty of mileage left in legacy fiber -- it can carry a lot more data than it's specified for. People want to upgrade to gigabit Ethernet without having to rip out their old fiber, he notes.
A lot of old datacom networks were built from multimode fiber, because the overall system cost is cheaper -- the transmitters for multimode fiber cost less than those for singlemode. Originally, the fiber itself was cheaper too, because it was easier to make -- the core (the part where the light is concentrated) of multimode fiber is much larger than that of singlemode fiber (see Optical Fiber). It should be noted that economies of scale have changed this, as sales of singlemode fiber have skyrocketed.
The drawback to multimode is that it has much lower bandwidth than singlemode. When light is launched into the fiber, it excites several optical "modes" -- depending on whether the light travels close to the center of the core or nearer to the outside -- and these travel at different speeds down the fiber. This results in "modal dispersion," an effect that's pretty similar to polarization mode dispersion (PMD) (see Chromatic Dispersion and Polarization Mode Dispersion (PMD)).
Lucent made its breakthrough using a completely new (at the time) multimode fiber, called LazrSPEED, that was specially optimized for 850 nanometer transmission -- the wavelength used in most datacom networks. The experiment comprised a single channel at 10 Gbit/s, which travelled a distance of 1.6 km. On the plus side, Lucent's experiment used commodity 850nm Vertical Cavity Surface Emitting Lasers (VCSELs) as the light source -- the cheapest available. But Lucent's use of a special fiber would not permit a simple system upgrade.
In contrast, the U.K. researchers used a special multiplexing scheme, along with 40-channel DWDM, to boost their system capacity. That means they used more expensive 1550nm light sources and required a multiplexer. Each wavelength carried 5.1 Gbit/s -- actually less than in Lucent's experiment.
In fact, Walker admits that they didn't use all 40 channels at once, because they didn't have 40 lasers (they're expensive, remember). "We had a block of 10 lasers, and two others," he says. "So we did the next best thing, which is look at co-channel interference."
The clever part of the experiment was in the electronics, according to Walker. The 5.1-Gbit/s data signal was composed of two separate 2.55-Gbit/s signals that were 90 degrees out of phase with each other, so that they would not interfere. This is a common technique in radio transmission, called subcarrier multiplexing, he says.
On top of that, the researchers added a "pilot tone" -- a pure electronic frequency at 5.1 GHz. The job of the pilot tone is to transport the clock signal along with the data, so that the detection circuits can lock onto the data immediately -- a zero latency system.
As a bonus, extracting the clock signal this way considerably simplifies the design of the receiving electronics, claims Walker, noting that the groundwork on high-frequency signals in multimode fiber was carried out by colleagues at Bristol University (see CLEO Report: Pick of the Papers).
The data signals and pilot tone are combined in the electronics, and the resulting signal, which now looks more like an analog than a digital signal, is sent to a optical modulator. After detection, the electronic signals must be separated before the data eyes can be examined. In every case, there were no errors, Walker claims.
But having achieved this huge data rate, the researchers are at a loss to explain it. Walker says that he wanted his student's research project to go out with a bang rather than a whimper, so they thought they'd just give it a try! Mathematical modeling of the results will take longer.
But whatever the reason it works, it works robustly. "We had to loan the fiber back to the teaching lab, so it was up and down the stairs on a regular basis." Yet, the experiment worked perfectly every time they put it back together, he claims.
— Pauline Rigby, Senior Editor, Light Reading