Complex modulation comes to optical fiber

Fiber is running out of bandwidth just as dial-up lines did years ago. Complex modulation again solves the problem.

By Martin Rowe, Senior Technical Editor, Test & Measurement World -- EDN, April 8, 2010

The demand for greater data throughput seems endless, and it is accelerating faster than many people expected, creating bottlenecks in fiber-optics networks. Digital transmissions of 100 Gbps, which companies are just now introducing, should alleviate some of these bottlenecks. A year ago, most of the work surrounding 100-Gbps links started with 10 10-Gbps lanes over short distances. Since then, Verizon has deployed the first long-haul 100-Gbps link using four 25-Gbps lanes (Reference 1). With it comes complex modulation that optical communications have never before used.

The new modulation schemes are necessary for handling long-distance transmissions. Short-haul communications, the so-called client side within campuses and local metropolitan areas, don’t need complex modulation because their distances are short enough to accommodate the higher speeds (Figure 1).

This is not the case for long-haul transmissions—the “line side” of networks for which service providers need transmissions of hundreds of kilometers. Adding fiber to compensate for additional lanes is just too expensive. “Carriers need to squeeze 100-Gbps throughput rates into their existing fiber plants, many of which were designed for 10 Gbps and some of which were designed for 2.5-Gbps fiber links,” says Pavel Zivny, a product engineer at Tektronix.

Simply squeezing a 100-Gbps NRZ (nonreturn-to-zero) stream into existing fiber is impractical. Current DWDM (dense-wavelength-division-multiplexing) fibers use 50-GHz spacing between channels. Although that channel spacing is sufficient for 10-Gbps data streams using NRZ modulation, it is too narrow for 100-Gbps NRZ streams. “You can’t put 100-Gbps streams right on the carrier,” says Mike Schnecker, business-development manager at LeCroy, because, for a 100-Gbps NRZ signal, each bit is just 10 psec wide.

“Because of crosstalk between adjacent channels, 100-Gbps data streams can’t be used in DWDM systems,” says Hiroshi Goto, an optical-product specialist at Anritsu. “PMD [polarization-mode dispersion] and CD [chromatic dispersion] prevent that [scenario]. There’s too much distortion. The pulses distort and overlap.”

To work around the problem, the OIF (Optical Internetworking Forum) has recommended using complex modulation to squeeze more bits per second per hertz from existing fiber. The OIF-proposed modulation uses QPSK (quadrature-phase-shift keying) and two polarizations to achieve 100-Gbps throughput on a single wavelength. QPSK is common in digital RF communications, but it’s new to fiber-optics communications.

A 100-Gbps link consists of two 50-Gbps streams in two polarizations—TE (transverse electric) and TM (transverse magnetic)—that propagate in two orthogonal polarization planes. Each 50-Gbps stream consists of 25G symbols/sec. QPSK modulation packs 2 bits into one symbol. Because the QPSK signal travels in two polarizations, it is called either DP-QPSK (dual-polarization QPSK) or PM-QPSK (polarization-mode QPSK); the terms are interchangeable, and both are commonly used. This article uses DP-QPSK when referring to the two polarizations and QPSK when referring to one polarization.

Complex modulation

Figure 2 illustrates the modulation process. A single 100-Gbps bit stream splits into TE and TM polarizations. That step produces two carriers at the same frequency. Each carrier then undergoes I/Q (in-phase/quadrature) modulation, resulting in two 25G-symbol/sec streams. The total is 100 Gbps, but the actual data rate is somewhat higher (see sidebar "What's in a G?"). The polarization splitter in figure 2 appears before the QPSK modulators. Some transceiver designs may place the I/Q modulators first and then split the modulated signals into two polarizations.

QPSK modulation places 2 bits per symbol by phase-shifting a carrier of light in response to incoming bit pairs (00, 01, 10, 11). Each symbol represents 2 bits. A receiver demodulates each symbol into its 2 bits and produces a 50-Gbps digital data stream. In addition, bits undergo precoding before modulation and decoding after modulation (Reference 3). A receiver then produces four 25-Gbps electrical signals after it demodulates and decodes the incoming DP-QPSK signal.

QPSK signals carry twice the number of bits per symbol that NRZ signals carry. Thus, the two modulations produce signals that degrade differently as they pass through fiber. Peter Andrekson, director of EXFO Sweden, explains that QPSK signals are more susceptible to noise and nonlinear phase distortion than NRZ signals. “Because of the higher noise susceptibility, QPSK-modulated signals will require higher power than NRZ signals,” he says.

QPSK signals have an important advantage over NRZ signals, though. They’re less susceptible to bit errors from chromatic dispersion and group delay at the same bit rate. That’s because one UI (unit interval) of a 100-Gbps data stream is 10 psec wide. Because line-side transmissions use four 25-Gbps lanes, each symbol is 40 psec wide, which results in a lower bandwidth.

The 40-psec-wide symbol of a 25G-symbol/sec stream is shorter and requires more bandwidth than a 10-Gbps, 100-psec-wide NRZ signal. Thus, the 25G-symbol/sec signal is more susceptible to errors from dispersion than a 10-Gbps NRZ signal, but it’s less susceptible to degradation than a 100-Gbps NRZ signal. “There is a trade-off between complexity and SNR [signal-to-noise ratio] versus dispersion tolerance and hardware bandwidth at a given bit rate,” explains Andrekson.

The DP-QPSK technology is so new that no transceiver modules exist for the line side. Chris Cole, senior member of the technical staff at Finisar, explains that line-side transceiver modules are larger than client-side modules (Figure 3), which a multisource agreement currently defines (Reference 4). Cole notes that designers can even implement line-side transceivers as line cards rather than as modules.

Test will change, too

The shift from NRZ to DP-QPSK modulation brings the constellation diagram to the forefront of fiber-optics test. Although constellation diagrams are common in RF wireless transmissions, they’re new to optical communications. Constellation diagrams are the first measurement you make on a QPSK transmission. Constellation diagrams provide information about the transmitted signal’s integrity. Dispersion and nonlinearities can cause signal degradation, resulting in distortion. shows constellation diagrams for both polarizations in a DP-QPSK signal. The constellation’s points are clearly visible in Figure 4, but they can become indistinguishable in the presence of too much distortion.

The two lower-right traces in show the QPSK-modulated signal’s magnitude (upper trace) and phase (lower trace). Note the apparent discontinuities on the phase-angle diagram. They result from phase shifts due to the encoding of bit pairs in the QPSK modulation.

For testing the optical DP-QPSK signal, you can use an optical-modulation or an optical-signal analyzer. These instruments produce constellation diagrams, decode them into electrical data streams, and display them as eye diagrams. Agilent Technologies, Anritsu, EXFO, and Optametra serve this market, and Optametra’s product employs a Tektronix oscilloscope.

“There’s no test specification for the 100-Gbps long-haul optical waveform, so test-equipment makers must talk to the optical-module makers to find out what they need to measure,” says Finisar’s Cole. “Each company will have different needs.” Cole also notes that test equipment must support 28G- and 32G-symbol/sec signals. “There are DP-QPSK test systems that run at 22G symbols/sec for 40-Gbps links, but new equipment will need to run at 28G and 32G symbols/sec to support 100-Gbps links.”

Testing the receiver side of optical transceivers is even more up in the air because specifications do not yet exist for stressed-receiver testing. Cole says that test equipment must be able to generate DP-QPSK signals, and that requirement can introduce controlled impairments, such as chromatic dispersion and polarization-mode dispersion. These impairments cause the TE and TM carriers to rotate as they pass through fiber. The impairments must produce stressed eye patterns after demodulating and decoding so that engineers can measure the signals once they’re in electrical form.

Figure 4 also shows the two eye diagrams (upper right) representing two 25-Gbps lanes from one polarization. “You’ll have to look at eye-mask margins, jitter, and extinction ratio; that’s the same as for 10-Gbps links,” says Cole.

Engineers now use oscilloscopes and BER (bit-error-rate) testers to analyze eye diagrams. Some engineers use high-bandwidth oscilloscopes to capture DP-QPSK signals. “Because of the modulation, signals at the receiver look like noise,” says LeCroy’s Schnecker. “Signals are no longer repetitive, and thus you need a real-time oscilloscope.” Zivny of Tektronix has also worked with engineers using real-time oscilloscopes on DP-QPSK signals. A four-channel oscilloscope lets you see all four decoded, demodulated data streams with high timebase correlation.

Engineers developing DP-QPSK transceivers use BER testers to produce the 25-Gbps data streams for each I and Q phase of a QPSK signal. They also use BER testers to measure BER on the demodulated, decoded signals. BER testers from Agilent Technologies and SyntheSys Research can measure BER at data rates as high as 28 Gbps.

Over the next few years, the industry will continue to develop 100-Gbps line-side transmissions. Test specifications will also emerge as optical-module manufacturers work with test-equipment makers and standards bodies to identify test issues and to develop test procedures and equipment.

A version of this article appeared in the March 2010 issue of EDN’s sister publication Test & Measurement World.

References

“Verizon Deploys Commercial 100G Ultra-Long-Haul Optical System on Portion of Its Core European Network,” Verizon, Dec 14, 2009. Di Minico, Chris, “802.3ba Cu specifications,” Jan 2008, IEEE. “Trends and Issues in Ultra-High-Speed Transmission Technologies: For the implementation of 100 GbE/40 GbE and long haul transmission by optical modulation,” Technical Note, Anritsu, 2009. “CFP Multi-Source Agreement Draft 1.0,” March 23, 2009.

What's in a G?

The terms “100G,” “40G,” and “25G” refer to the data throughput of an optical link. Because of formatting and FEC (forward-error correction), actual data rates are higher than the numbers indicate. For example, the data rate for a 100-GbE (Gigabit Ethernet) transmission is actually 103.125 Gbps, but the data throughput is 25 Gbps. Therefore, each 25-Gbps lane actually carries 25.78125 Gbps (26 Gbps) for the client side. So, if a test-equipment manufacturer claims that its products have 26-Gbps speed, it means that the product covers the 25.78125-Gbps data rate. A 27.739-Gbps (28 Gbps) data rate is also under consideration for Ethernet client-side networks. For line-side networks, long-haul transmissions, links need additional FEC. The line rate for 100-Gbps links with 7% FEC is about 112 Gbps, which translates to about 28 Gbps on each lane. According to the Optical Internetworking Forum’s “100G Ultra Long Haul DWDM Framework Document,” the exact rate has not yet been specified (Reference A). These transmissions can also use a higher FEC overhead of 20%, which increases the bit rate to about 32 Gbps. Reference A. “100G Ultra Long Haul DWDM Framework Document,” Optical Internetworking Forum, www.oiforum.com/public/documents/OIF-FD-100G-DWDM-01.0.pdf.