System challenges drive component choices

Arthur J. Lowery

A rainbow has a continuous spectrum of color. Present wavelength-division multiplexing (WDM) systems-like a young child`s perception of a rainbow-have several discrete colors but with large gaps between them. As happened to the radio spectrum, however, demand for capacity is forcing the use of the entire spectrum of the low-loss regions of optical fiber`s attenuation window: a true rainbow, without gaps.

So what impact is this having on the development of optical components? What advances will have to be made, and how can we design systems where electronics is partially replaced by all-optical technologies to form optical networks? This article explores three fundamental ways a system`s capacity can be improved and how they depend on the performance of all components along a typical WDM system (see Fig. 1).

Performance vs. Capacity

Ideally, WDM channels do not interact (there is no optical crosstalk) and as many channels can be added as required, each operating at a high data rate. In practice, however, interaction occurs, and the number of channels is limited by the performance of the components.1

The three requirements for providing a higher-bit-rate WDM link are broader optical bandwidth (using more of the fiber`s low-loss regions), higher bit rates per channel (say from 2.5 to 10 Gbit/s), and a higher density of individual channels (thus reduced channel spacing, say from 100 to 25 GHz). The total capacity of the system is the product of these three factors. For example, 30 nm of optical bandwidth (3750 GHz), with a channel density of 1/(100 GHz), and a bit rate per channel of 10 Gbit/s gives a total capacity of (3750 × 10/100) = 375 Gbit/s.

System capacity is also directly affected by the performance of the individual components in it. By identifying the limitations of all components in a system, it becomes easy to visualize which components are limiting total capacity (see Fig. 2).

Optical transmitter. Before WDM, a transmitter`s wavelength was unimportant: the absolute wavelength was only roughly specified to be close to the dispersion minimum of the fiber. However, WDM transmitters need wavelength stability to prevent drift into the wavelength allocation of an adjacent channel. This requirement increases with channel density.

In addition, a modulated transmitter`s spectrum should fit within its allocated optical channel. If the laser bandwidth is significantly broadened when modulated, its energy can fall outside the demultiplexer`s bandwidth. This produces patterning effects because extreme frequencies of the transmitted sequence are rejected, limiting the spectral efficiency of the system.

External modulators can have zero chirp and so give an optical bandwidth of just under two times the data rate due to the upper and lower modulation sidebands. Band-limiting the modulation waveform can lower the optical bandwidth, but special modulation formats such as duobinary, m-ary, or single-sideband are required to increase spectral efficiency to more than 1 bit/s/Hz.2

Wavelength multiplexer. The simplest form of multiplexer is a wavelength-independent tree of optical couplers. Because of power loss and inability to reject out-of-band noise from transmitters, a wavelength-dependent multiplexer is preferred. As spectral efficiency is increased, its characteristics become critical, similar to the demultiplexers (see below).

Optical amplifier. Doped-fiber amplifiers compensate for fiber and splitting losses. The performance measures of amplifiers are bandwidth, gain, gain flatness, output power, and noise figure. Multiple-dopant amplifiers giving large bandwidths are required to increase the optical bandwidth of the system. However, gain flatness suffers and is difficult to compensate for as the channel count is increased. Gain flatness also depends on the amount of loss compensated for, which is proportional to system length. To maintain adequate signal-to-noise ratios in the channels, the optical power per channel must be kept high, especially in long links with chains of amplifiers. Thus, the output power of an amplifier must increase with an increase in both optical bandwidth and channel density, requiring multiple high-power pump lasers.

Amplifiers generate amplified spontaneous emission (ASE), which is approximately proportional to their gain in excess of unity. The ASE beats with the signal at the photodiode to produce in-band electrical noise. This electrical noise is generally proportional to the bit rate per channel. For systems using incoherent sources, such as some code-division multiplexed schemes, noise can scale with increased bit rate and channel density.

Optical fiber. The optical fiber is often seen as a perfect transmission medium with almost limitless bandwidth. Relative to copper technology this is true, but in practice the optical fiber can be the most limiting component to a system`s capacity, especially as distance is increased to multispan amplified systems, because fiber-dominated limitations scale with length.

Fiber loss, fortunately, is independent of the modulation bandwidth of a single channel (this is not the case with copper, which has frequency-dependent loss). However, the low-loss windows of fiber are limited, and increasing the optical bandwidth will require higher-gain amplifiers.

Fiber dispersion is a critical limitation for bit rate per channel. The effect of dispersion on receiver sensitivity scales with the square of the bit rate per channel because the transmitted bandwidth scales with the data rate, which creates pulse spreading. It is the pulse spreading relative to the bit period that affects the receiver sensitivity.

Compensation for dispersion can be achieved using dispersion-compensating fiber, planning the fiber plant to have near-zero overall dispersion, or using filters such as chirped Bragg gratings with a compensating phase characteristic. The difficulty of dispersion compensation using filters increases with optical bandwidth because longer differential delays are required in the filter. This needs a long fiber Bragg grating, for example.

Fiber nonlinearity, although small compared to active materials, causes crosstalk in WDM systems by four-wave mixing (FWM), timing jitter due to cross-phase modulation (XPM), and power transfer by Raman scattering. All effects are proportional to channel power and scale with the number of channels.

Channel power is dependent on data rate to obtain a given electrical signal-to-noise ratio. The exact limitation is highly dependent on the local dispersion along the fiber (the dispersion map), the link`s length, the channel spacing, polarizations, and powers; it generally requires powerful numerical simulation to assess.

Polarization-mode dispersion (PDM) causes timing jitter due to the random birefringence of fibers. This generally accumulates with length. However, the effect of timing jitter depends on its value relative to a bit period, so that PDM must be reduced as data rate is increased.

Optical demultiplexer. The optical demultiplexer separates out WDM channels with a minimum of optical crosstalk and directs them to individual channel receivers. With a low channel density, this is relatively easy, especially for low data rates. A power splitter followed by a simple bandpass filter for each channel filter can be used if a loss is acceptable. However, if the spectral efficiency is increased to close to unity, the design of the filters has to be much more sophisticated, requiring near-flat passband amplitude characteristics and a sharp transition to their stopbands. The stopband attenuation must be sufficient for the combined crosstalk of all the unwanted channels. Filter design must also include near-linear phase characteristics, otherwise the filter itself will become a dispersive element.

A filter`s attenuation characteristic changes with denser and higher channel capacity systems (see Fig. 3). A higher optical bandwidth implies that the filter must have a high rejection over a wide stopband, which may be difficult to achieve with some periodic-response filter designs.

Receivers. Because the receivers are placed after the optical demultiplexer, they should be affected only by the individual channel data rate on which their sensitivity depends. Although the receiver itself has little dependence on total optical bandwidth (apart from the photodiode`s wavelength sensitivity), the bit error rate (BER) at the end of the receiver is of course the ultimate test of system performance.

All-optical networks. All-optical networks replace electronic switching and wavelength conversion with optical functionality, such as optical cross-connects (OXCs) and optical add-drop multiplexers (OADMs). Being `analog` in nature (unless full optical regeneration is developed further), these networks require careful consideration of optical crosstalk and multipath interference.3

Low levels of crosstalk can have a significant effect because of the coherent mixing of optical fields. Even if the fields are from different transmitters, or carry different data, or come from the same transmitter but over a ghost path longer by more than the coherence length of the laser, coherent mixing will cause large penalties. Also, adding/dropping channels during switching will cause millisecond power transients unless sophisticated dynamic gain control schemes are used.

Designing performance

Clearly the design of an optical component can directly and significantly affect the performance of an optical system. The system can cost hundreds of millions of dollars: the component can cost tens of dollars. New design methods must be found because of the pressures for increased performance, increasingly sophisticated systems, and reduced design cycles.

One possibility would be to tightly specify the performance of each component to ensure the successful operation of the system as a whole. However, this would lead to overly conservative design, which is not sustainable in a highly competitive industry. An attractive alternative is to employ computer-aided design and optimization to optical network systems and to replace the hardware prototype with software simulations (see Fig. 4).

Sophisticated and comprehensive simulation tools have been developed to assess all of the degradations described. The programs allow detailed simulations of the performance of components (using bidirectional models to identify the effects of reflections and unwanted resonant cavities). They also allow full signal simulation of the transmission (to identify the interaction of fiber dispersion and nonlinearities) and parameterized simulation of networks (for rapid identification of signal-to-noise ratios, amplifier operating conditions and crosstalk paths).

With simulation software, even the smallest of components can be designed and assessed within very large systems, enabling us to develop a WDM system with no gaps in the rainbow.

REFERENCES

1. IEEE J. Quantum Electron. 34(11), 2053 (Nov. 1998).

2. T. Ono, Technical Digest of OFC `99 (San Diego, CA, Feb. 1999), paper FE4.

3. Y. Shen, K. Lu, and W. Gu, J. Lightwave Technol. 17(5), 759 (May 1999).

FIGURE 4. Simulation of performance degradation in a ring network provides a schematic for assessing crosstalk due to filter performance. BER can be plotted against filter characteristics. (Courtesy Virtual Photonics` Photonic Transmission Design Suite)