By Rob Howald
The communication community can't seem to get enough on fiber optics. Optics promises to be one of the hottest, if not the hottest, arena in communications in 2001.
Welcome to another year of Communication Systems Design and, hopefully, another worthwhile year of reading "Building Blocks." Finally, we can safely say New Millennium without some numerical nerd reminding us we are really off by a year.
Four years of feedback
I want to send out thanks to the readership for taking the time to provide feedback. I do appreciate everyone who has made an effort to e-mail me with comments, suggestions, and even questions. I've received enough feedback now, after the four years of columns so far, to characterize some common themes.
One main group of e-mails comes from folks who read a column that touches on an area they are involved in at a deeper level. I like these e-mails, because two of the three versions of these e-mails are positive. One such subgroup says, roughly, "very nice, you hit XYZ right on the head, nice job, now I see a bigger picture."
Another subgroup adds something that I did not mention or cover or know of, which often ends up in me learning something. This was difficult at first, since I had to let go of the premise that I knew everything, but has proven to be rewarding.
Finally, I get an occasional e-mail asking for technical advice on topics unrelated to column discussions. Based on this type of technical feedback, I do intend to spend much of the year discussing topics that remain tutorial in nature. The popular e-mail vote seems to favor this tutorial type of material, as opposed to more advanced topics.
Last month, we introduced some basics of fiber-optic communications. Included in that introduction was the idea of multi-mode and single-mode fiber as a waveguide, wavelength windows of operation and associated optical attenuation, pulse transmission, multi-mode and chromatic dispersion, and linewidth. This month, we continue to extend our fiber-optic vocabulary. We will extend our discussion of wavelengths by touching on wave division multiplexing, then talk about optical sources - including some of their performance characteristics.
Wavelengths and WDM
The focus of last month's discussion pointed out the various wavelength "windows" through which light could be effectively propagated down a fiber. The discussion worked its way around to focusing on some of today's common wavelength regions - 1310 nm and 1550 nm.
Both the 1310- and 1550-nm regions were conveniently low loss regions, with the 1550-nm region being slightly lower per kilometer. The 1310-nm region lines up nicely with the zero dispersion point on typical single-mode fiber, meaning that light pulses will not be distorted (ideally) as they traverse the fiber.
In contrast, fiber that is not dispersion-shifted (designed to move the zero dispersion point elsewhere, such as near 1550 nm) will have dispersive characteristics for 1550-nm sources. Despite the lower loss, this dispersion can limit the distance that 1550-nm sources can travel, as the detection of the pulses becomes more difficult.
Now, consider the wavelengths of light (and this is for all of you RF people) as a couple of information carriers, and change nanometers (nm) to Megahertz (MHz). This conversion gives us signals at 1.31 and 1.55 GHz. Ignoring for a moment any possible bandwidth occupancy issues, these two carriers could share the same RF spectrum, on a coaxial cable, for example, without interference. This is simple frequency division multiplexing (FDM), a concept used everywhere.
The same concept can be applied to optics, and the technique is called, wave division multiplexing (WDM). A single fiber can carry both 1310 nm and 1550 nm by coupling the two sources together onto a fiber.
Of course, there are numbers and names of parameters for describing how the wavelengths need to be kept away from one another, just as the RF domain has the adjacent channel interference (ACI) concept.
The receiver end gets trickier, as to operate off a single detector of light, the wavelengths would have to avoid sharing spectrum. This scenario is unlikely in digital optics. Thus, the wavelengths need to be separated at the receive end before detecting. This 1310-/1550-nm technique goes by various names that describe the wavelength difference, such as broad WDM, wide WDM, or coarse WDM.
The DWDM case
A more interesting and more difficult case is dense WDM, or DWDM. This technology takes further advantage of the idea described with traditional WDM. DWDM fits many more wavelengths onto a fiber by packing in wavelengths that are very close to one another. However, the wavelengths are carefully controlled in terms of source quality and wavelength accuracy and movement. As such, the higher performance lasers used are more costly.
The additional cost associated with sources and the passive multiplexers needed to support wavelength separation, which once were prohibitive, have come down quite a bit as optics has become more commercially available. DWDM wavelengths are standardized on the ITU grid so that equipment can be interchangeable.
Obviously, each color (the cute way to say wavelength) that can be added linearly increases the carrying capacity of the system. As an example of wavelength spacing, eight wavelengths spaced by 10 nm can be placed in the 1550-nm window beginning at 1510 nm and ending at 1580 nm.
Systems with higher density and more colors are possible. Thus, a single fiber, OC-192 link (10 Gbps) can be built to a carrying capacity of 160 Gbps over the same fiber strand by deploying DWDM with 16 wavelengths. The fact that no more fiber needs to be laid to do this is extremely valuable, because of the cost involved.
Light sources for optical transmission are not mysterious, although the variations and manipulations of the traditional transmit devices can be intricate. There are two basic types: light-emitting diodes (LEDs) and lasers. As an interesting bit of techno-trivia, laser is actually an acronym for Light Amplification by Stimulated Emission of Radiation.
Learning about LEDs
First, let's consider the basic LED. LEDs actually have a technical purpose beyond good (green) and bad (red). While LED engineering has definite trade show implications, the LEDs used for networking purposes have some significant differences.
Because of the loss versus wavelength concepts presented earlier, networking LEDs operate in these lower loss regions when used for transport, as opposed to indicator LEDs, which obviously have to work in wavelength regions where you can see them. While transport LEDs operate in the windows previously described, with a low end of perhaps 780 nm or 850 nm, eyeball wavelengths are below 700 nm.
Another key difference for networking LEDs is that, because efficient coupling of optical sources into fibers is critical, these LEDs emit their light over a smaller area in order to get as much as possible into the fiber.
Fortunately LEDs are inexpensive, rugged, and easily fabricated and obtained. Unfortunately, LEDs have broad linewidths and emitting areas, affecting dispersion performance and coupling efficiency. Figure 1 shows a sample linewidth of a 1310 LED.
In contrast, the linewidth on lasers in this wavelength window may be on the order of a half nm or less, to about 5 nm. Additionally, LEDs can be slower (lower maximum data rates), based on how quickly they can be turned off. This is unfortunate, in that it gets in the way of one of the attractive benefits of the fiber - its inherently high bandwidth. Numbers on the order of hundreds of Mbps are typical.
Laser Lore
Since I did not burden you with LED physics, I will not spend valuable words on laser physics, either. However, this type of physics is interesting to re-read in a nerdy sort of way, as we are manipulating effects that can be traced right down to our keen understanding of things at the atomic level (both LEDs and lasers).
In any event, commercially used lasers come in two main types: Fabry-Perot (FP), and distributed feedback (DFB). The former is the economy car, while the latter is the luxury sedan.
The FP provides the necessary optical feedback to support lasing action (see, you should go ahead and read about the physics) by implementing a cavity for light via reflectors (mirrors) at the ends. The DFB, as the name implies, provides a distributed mechanism to implement such feedback, by etching a periodic structure into the active area.
The result for both is very narrow linewidths, as given by the numbers above for the FP (wider) and DFB (narrower). Thus, the coherence of the light emissions from DFBs is better than FPs, and FPs are better than LEDs in this regard.
While lasers can be more expensive, less rugged physically, and may exhibit poor environmental performance, there are many situations where only a laser is up to the task, such as high data-rate links or when distance and/or linearity are required.
Microwave interlude
Now, back to you microwave buffs for a moment. As you recall, I likened the waveguide effect of RF plumbing to how optics works. For optical transmission, it is the same principle played out in a wildly different portion of the spectrum, and it dramatically changes the size of the waveguide. For you playing along at home, just for fun, this waveguide frequency dependence can also be recognized when driving down the road and entering a long tunnel with your favorite AM radio station (is there such a thing?) on, as the physical size of the tunnel and its material structure results in waveguide propagation cutoff for low frequency AM stations.
Now, recall that those waveguides became resonant cavities when you put ends on them. It is fascinating to know that, even today, some of our best filter performances can be achieved this way - basically using a soup can as a resonant cavity.
Recall also that those cavities have more than one mode - any integer wavelength of the dimensions was resonant. The same effect occurs with FP lasers - multiple discrete modes can exist. These widened, discrete lines of spectrum for the FP have the same dispersion issues associated with them that we spoke of before for LEDs. Yet there are actually some beneficial aspects about the spread. The benefits are associated with modal noise in multi-mode fiber. This has to do with the potential for wonderfully coherent sources, such as DFBs, to powerfully add both constructively and destructively when the situation is not ideal. Some spectrum spread tends to reduce these peaks and valleys.
Entering DFB grating
The DFB grating approach can be designed to have a single dominant mode of very narrow linewidth. What's so great about that? Well, the DWDM idea is a no-brainer. Narrower width sources can be packed together more tightly without fear of overlap. Thus, DFBs in the 1550-nm window create the opportunity for substantial capacity increases from a single fiber.
Another repeatedly mentioned situation is dispersion. linewidth spreading is more likely to result in dispersive distortion and to interfere with the successful pulse detection (assuming digital optics). Monochromatic sources (literally meaning single-color or very narrow linewidth) associated with DFBs, when emitting at the zero dispersion point of the fiber can go a long way without fear of dispersion, in principle. Another way to approach it is that you can operate at very high data rates without pulse dispersion problems.
We have tossed out the word linewidth many times in this column, now let's quantify it to illustrate this principle. A typical way to describe the linewidth is by using full-width half-maximum (FWHM) linewidth. FWHM is the linewidth at the half amplitude levels of the light spectrum. Figure 1 shows the bandwidth-distance product, in GHz-km, for various source linewidths. The GHz-BW obviously spikes up, on a logarithmic scale no less, sharply as the linewidth decreases. Furthermore, the DFB linewidth used above (a half nanometer or less) isn't even on Figure 2.
For example, the FWHM = 2 nm case peaks at, say, 30,000 GHz-km. This means that a 10-Gbps link would, even using a very conservative 10-GHz bandwidth (60% is more appropriate), be safely undistorted through 3000 km! Of course, optical attenuation says you wouldn't see it at the other end anyway, but you get the idea.
The other important thing to remember is that, linewidth aside, accuracy of the precise wavelength matters a lot, as moving off the zero dispersion point even a little changes things drastically, and this is often what must be designed for. You can see that the DFB offers both nice distance options for 1310-nm sources, and nice WDM options for 1550-nm sources. The luxury sedan tends to have more options than the econo car - but I bet you can at least get that AM radio in the economy car.
So when should you use an FP and when a DFB? These are basically pay-as-you need situations. FP's are less expensive and less performing, but have a boatload of applications that they are good enough for. Long distance, high performance, high-speed applications begin the thought process for DFBs.
Next time...
Next month, I'll continue my survey of fiber-optic technology and systems. I'll move into optical detection and push to the surface some of the issues, trade-offs, and link performance parameters related to this month's discussion.
I hope to venture into what kinds of things are happening beyond this fundamental overview of point-to-point optical links presented here.
There is so much going on in this arena, I could fill a year of columns with it. It is amazing to think about how drastically the landscape has changed in a relatively short time. When I received my undergraduate and master's degrees, there was no mandatory coursework in optics at all, or even in PCs, broadband anything, or Internet-related anything (IP, networking, Web basics).
Perhaps you've heard that term for an on ramp that merges into the far left lane of a major highway? An exciting traffic gimmick in Philadelphia (but no doubt born in Boston), it is known as the popular Merge or Die Lane. In engineering today, it has basically become Learn or Die. It can be very invigorating and educational - or it can inspire pursuit of an MBA.
Puzzled by a network processing design issue?
