A1.1 Optical Links

Communication systems involve the sending of information—typically electrical signals carrying digital data—between two points. Even for optical communication, which uses fiber to carry information in the form of light, electrical signals are first converted to optical ones before transmission, and converted back from optical to electrical at the receiver.

Communication is either unidirectional, known as simplex, or bidirectional (duplex). If both directions operate simultaneously, it is known as full-duplex [1].

For simplex communications, a transmitter is used at one end of the link and a receiver at the other—one to send data and one to receive. When bidirectional communication is involved, a receiver and transmitter pair—a transceiver—is required at each end.

The translation between electrical and optical signals for optical communication adds complexity and cost. So why do it? Because there are telling advantages when communicating optically. The bandwidth—the information-carrying capacity of the optical medium—is far greater in the optical domain than in the electrical domain, as is the reach.

Typically the receiver and transmitter are copackaged in an optical transceiver, also referred to as an optical module.

The main ways data is sent optically are shown in Fig. A1.1:

For duplex communications, the total fiber count is double the schemes above. One twist on these schemes is the bidirectional approach that scales capacity using wavelengths to avoid having to use multiple parallel fibers. The technique sends data in both directions on one fiber using a different wavelength for each.

These main three techniques help designers choose the degree of complexity needed to meet the optical performance requirements of a given application.

A single channel is the simplest and cheapest scheme. There is one laser at the transmitter and one photodetector at the receiver. But only so much data can be sent across the channel. The current highest-speed single-channel rate is 25 Gb/s, although a 40-Gb/s 2 km serial standard has been deployed commercially.

For short-reach applications, the transmitter’s laser is directly modulated. Here the laser is turned on and off to encode data so that a separate modulator is not required. But a modulator is used when higher optical performance and longer link distances are needed.

The modulator’s role, at its simplest, is like a card waved in front of the light source, to pass or block light as required to encode the data being sent. Naturally, adding a modulator adds cost to the transceiver.

The parallel, multiple-lane approaches—schemes 2 and 3 above—are used when the data to be sent exceeds the capacity of a single channel.

One approach is to use single channels in parallel in the form of a ribbon cable. The link is composed of a transceiver at each end connected with a ribbon cable. The ribbon fiber, with each strand carrying a wavelength, is attractive for shorter links. Each channel is relatively simple, but using multiple fibers adds cost especially as the link distance increases. Ribbon cable also adds wiring complexity that needs to be managed and routed within a data center.

One such example of a parallel link is the IEEE 100GBase-SR4 standard, as shown in Fig. A1.2. A 100-m, 100-Gb IEEE standard for the data center, the link uses four multimode fibers, each transmitting 25 Gb, and four fibers each receiving 25 Gb.

The Parallel Single Mode 4 - PSM4 - multisource agreement is another example using this approach.

Another example is the emerging short-reach 400-Gb Ethernet interface, dubbed 400GE-SR16, which uses sixteen 25-Gb channels in each direction. Accordingly, the 400-Gb short-reach interface will have 32 multimode fibers overall, an example of how channel count rises with data rate.

The second parallel approach is wavelength-division multiplexing. Here, channel parallelism is achieved using wavelengths across the fiber’s spectrum, commonly single-mode fiber. To send wavelengths across a fiber, multiple pairs of lasers and receivers are used as well as another optical function pair—a multiplexer and demultiplexer.

The multiplexer acts as an on-ramp, placing each wavelength onto the fiber, while the demultiplexer separates each wavelength at the destination. The difference from ribbon fiber is that a more complex transceiver is required at both ends, but for wavelength-division multiplexing only one fiber is used in each direction for full-duplex communication.

One example of wavelength-division multiplexing scheme is the 2-km CWDM4 multisource agreement, another 100-Gb interface for the data center. The CWDM4 transceiver uses four lasers and four photodetectors as well as a multiplexer and demultiplexer, with each wavelength carrying 25 Gb. Fig. A1.3 shows the optical functions of the CWDM4 transceiver.

Wavelength-division multiplexing is the bedrock of long-distance optical transmission. A complex transceiver design is needed to send and receive each wavelength hundreds or thousands of kilometers, each one typically carrying a 100-Gb signal. Up to 96 such wavelengths can be packed on a single-mode fiber.

A1.2 Optical Component Technologies

Optical designers must consider the following performance metrics when deciding which optical component technology to use:

A1.3 Attenuation Characteristics of Fiber

We mention different operating wavelengths in fiber: 850 nm for multimode fiber, and 1300 and 1550 nm for single-mode fiber. These are used with good reason: they are wavelength bands for the optical fiber where signal attenuation, measured in dB/km, is low and suits the transmission distances.

The attenuation curve in Fig. A1.4 makes clear why 1300 and 1550 nm are used for medium- and long-distance transmission since the loss is a fraction of a decibel per kilometer. Optical amplification and dispersion are other factors contributing to the transmission windows. Although the lowest loss and dispersion is at 1300 nm, the 1550-nm window emerged as a better wavelength for WDM transmission. One counterintuitive reason for this is that zero dispersion, which for a germanosilicate fiber is near 1300 nm, leads to nonlinear transmission impairments. Consequently, optical amplifiers and wavelength-division multiplexing were developed in the 1550-nm window.

Most long-distance transmission is between 1525 and 1565 nm in what is called the conventional or C band. The broad-based development and commercialization of erbium amplifiers is largely responsible for the C band’s popularity.

Erbium also amplifies between 1570 and 1610 nm, facilitating long-distance transmission in this long or L band region. As C and L band amplifier designs are different, two amplifiers are needed if transmission in both bands is desired. In Chapter 5, Metro and Long-Haul Network Growth Demands Exponential Progress, we discuss the C and L bands.

Operating at 850 nm, the attenuation is higher but the distances are substantially shorter. In fact, distance is limited by modal dispersion, the temporal spreading of the transmitted modes due to different propagation velocities.

Note that around 1400 nm, there is an absorption peak due to the effect of traces of water in the glass fiber. Fabrication techniques can be used to lower this absorption below 0.5 dB/km to expand the transmission window or fiber spectrum over which data can be sent.

A1.4 Optical Modules

Optical modules arose with the advent of 10-Gb/s transmission prior to the optical boom of 1999–2000.

At that time, equipment vendors such as Nortel and Lucent Technologies had their own optical component divisions to make and design optical interfaces for their systems. But around 2000, the then-leading telecom equipment makers divested their component divisions, ceding their component know-how to optical module and optical component companies. The equipment makers realized that their focus and cash could be better spent elsewhere, and that the influx of module-making companies during the optical boom would drive down the cost of interfaces.

This has been the business model for the past 15 years. Transceiver companies were able to aggregate volumes by selling to multiple equipment vendors and this, coupled with the fierce competition, has helped drive down costs.

For 10-Gb long-haul transmission, the first designs to market were line cards containing discrete state-of-the-art optical and chip components. Ten gigabit transmission was also the last speed to use simple on-off keying. This modulation scheme was also used for 40-Gb links that followed, but the move to 40 Gb also marked the introduction of more advanced modulation schemes based on phase that are now standard for 100-Gb and greater link speeds.

The first 10-Gb optical modules developed by third-party optical module companies were not alternatives for the highest performance 10-Gb links made by the equipment makers, but they achieved long-distance dense wavelength-division multiplexing–based transmission and offered a buying alternative for suitable links.

The first such modules were 5-by-7-in. 300-pin MSA modules. These were necessarily large to fit in all the components needed for long-haul performance. The 300-pin MSA supported 10-Gb transmission and was quickly followed by a 40-Gb MSA [4].

The market realized that Layer 3 applications provided the highest volumes and that small size was paramount for such applications. What followed was a string of announcements of ever-smaller form factors that ultimately led to the SFP+ (see Table A1.1).

The downside of such choice was that optical module makers had to choose which form factors to back at a time when it was unclear what end users would want.

A1.4.1 The Miniaturization of Modules

The SFP+ form factor has become the de facto module of choice for high-density designs—used for 48-port Ethernet switches inside the data center, e.g., this yields a total capacity of 480 Gb.

But it took more than a decade to move 10-Gb line-side technology from a discrete line card to fit into the SFP+ module. This may sound like a long time, but it is a remarkable achievement in terms of cramming ever-more-complex designs into the same form factor. There is now a suite of 10-Gb products in the SFP+ form factor, from 10-Gb short-reach interfaces to the tunable SFP+, where the tunable laser supports dense wavelength-division multiplexing transmission over several hundreds of kilometers.

The SFP+ is the most common pluggable optical module. Having an optical module that can be plugged into and detached from telecom and datacom equipment brings several benefits:

• Operators can replace the module should a failure occur without having to replace the complete line card, which would disrupt other links.

• The same form factor can support different interfaces, including future ones with enhanced performance. This enables easy upgrades once a platform is deployed.

• The operator of equipment can install interfaces as needed, saving up-front costs.

• Purchasers can choose among transceivers from multiple vendors based on cost and performance.

Today, the important pluggable form factors besides the SFP+ include the quad-channel version of the SFP+, known as the QSFP+, and the 28-Gb versions of the two—the SFP28 and the QSFP28, the latter of which is the 100-Gb form factor of choice in the data center.

Another important family of optical modules is the CFP MSA family [5], which includes the CFP, CFP2, CFP4, and CFP8 (see Fig. A1.5). There is a deliberate reduction in size from the CFP to the CFP4 to ensure ever-increasing interface densities on equipment. The CFP8 is approximately the size of the CFP2 and is being earmarked for 400-Gb Ethernet and, for line-side optics, the CFP8-ACO. A bigger size is needed to accommodate the greater number of channels 400-Gb requires and the greater heat such designs will generate.