Healthy skepticism greets WDM claims

Wavelength division multiplexing systems are becoming ever more popular as carriers seek to meet the insatiable demand for bandwidth that growing Internet and corporate data traffic creates.

As the number of channels in commercial systems continues to increase, industry experts question the ultimate capacity of dense WDM systems. A recent laboratory experiment in Japan set a WDM transmission record of 2.6 Tb/s, transmitting 132 channels with each carrying 20 Gb/s of data through 120 km of conventional fiber.

In line with these encouraging results, several WDM equipment vendors have announced future availability of increased numbers of wavelengths-and hence total capacity-on a single fiber. In reality, however, practical trade-offs exist among the number of WDM channels, the channel data rate and span characteristics that limit the effective use of fiber transmission bandwidth for both long-haul and metropolitan networks.

One fundamental restriction on the number of optical channels is the set of wavelengths supported by optical amplifiers, which carriers need to overcome fiber and optical multiplexing losses.

The most commonly available optical amplifiers for use in WDM networks are erbium doped silica fiber amplifiers that feature a 28 to 30 nm wide bandpass within the conventional or C wavelength band ranging from 1530 to 1560 nm.

Optical carriers can be packed into this spectrum with minimum spacing that is a function of the bit rate, fiber properties and practical stability of optical sources and filters. State-of-the-art WDM systems can achieve 50 GHz (0.4 nm) spacing for OC-48 channels on conventional non-dispersion shifted fiber or non-zero dispersion-shifted fiber.

With improvements in optical technologies, this spacing may drop to 25 GHz within the next two years. Current 50 GHz systems therefore can provide 64 to 80 channels within the C band. At OC-192 rates, fiber nonlinearities and optical spectral filtering requirements lead to a minimum spacing of 100 to 200 GHz, resulting in up to 32 to 40 channels within the same bandpass.

Although these considerations determine the set of possible wavelength channels, deployment requirements such as economics, operations policies and procedures, and fiber plant characteristics ultimately decide the total system capacity.

Trade-offs over the long haul A long-haul WDM transmission system has typical span losses of 25 dB, which are compensated using line amplifiers (Figure 1). In these systems, the maximum non-regenerated span depends on many factors, including:

* Number of WDM channels,

* Maximum output power of optical amplifiers,

* Amplifier noise figures and

* Linear-chromatic dispersion, polarization mode dispersion-and nonlinear-cross phase modulation, self-phase modulation, four wave mixing-characteristics of fiber.

Intermediate regenerators should not be deployed before demand for bandwidth or grooming proves to be economically beneficial.

In other words, trade-offs come into play between the number of WDM channels and the achievable span.

For long-haul systems, WDM transmission is limited by non-flat amplifier gain and by optical signal-to-noise ratio degradation due to accumulated amplifier noise. Fiber non-linearities such as four-wave mixing and cross phase modulation generally are not significant at 2.5 Gb/s for single mode fiber for distances of less than 1000 km. But they have a greater impact for non-zero dispersion shifted fiber and severely impact transmission using dispersion shifted fiber.

Self-phase modulation effects are minimized by the relatively low power per channel as the number of channels increases beyond 16. However, transmission at 10 Gb/s requires more than 6 dB higher transmitted signal power than is needed for 2.5 Gb/s to maintain adequate optical signal-to-noise ratio.

The higher signal power required for 10 Gb/s can result in severe degradations because of fiber non-linearities, especially cross phase modulation. Polarization mode dispersion of installed fiber plant also can place limit 10 Gb/s transmission to less than 400 km.

Table 1 indicates the trade-offs between the number of WDM channels, the bit rate and the number of spans because of accumulated amplifier noise and limited output power for the particular case of 25 dB span losses. These results were obtained with the following assumptions and constraints: optical signal-to-noise ratio of >20 dB for 2.5 Gb/s and >26 dB for 10 Gb/s, and gain-flattened erbium-doped fiber amplifiers with a noise figure of less than 6 dB.

Two erbium-doped fiber amplifier output powers are assumed: 17 dBm, which conforms to accepted eye safety practice, and 20 dBm to demonstrate the effect of limited output power.

The WDM multiplexer/demultiplexer losses are weakly dependent on the number of channels.

What about increasing the number of WDM channels on the number of fiber spans, each with 25 dB? The key limitation is optical signal-to-noise ratio degradation with increasing numbers of erbium-doped fiber amplifiers and channels (hence decreasing per-channel power).

These results are presented in Figure 2, which assumes a 75 km length for 25 dB span loss. Increasing the bit rate to 10 Gb/s reduces the number of spans because of the higher required optical signal-to-noise ratio, greater than 26 dB. However, fiber non-linearities and polarization mode dispersion may further limit the non-regenerated distance for 10 Gb/s WDM transmission.

Wavelength stability is key in metro A metro WDM system shows typical span losses below about 20 dB, which would require few or no fiber amplifiers (Figure 3). This figure estimates the per-channel power levels at various points in the system, assuming directly modulated DFB laser transmitters with average output power of 3 dBm.

The WDM multiplexer loss is nearly independent of the channel number for dispersive WDM multiplexers such as arrayed waveguide gratings. Therefore, for metro WDM systems, the loss budget does not vary significantly with the number of channels but is strongly affected by the bit rate because of the receiver sensitivity.

In this example, span losses up to 18 dB can be accommodated for 2.5 Gb/s using avalanche photodiode receivers with about -33 dBm sensitivity, allowing a margin of 3 to 4 dB. For 10 Gb/s, an optical pre-amplifier would be needed before the receiving WDM demultiplexer to meet the same span loss budget of 18 dB.

To reduce costs, directly modulated DFB lasers can be used for metro WDM systems, where the total fiber length will be less than 100 km. However, the transient chirp of these lasers may lead to pulse distortions if the WDM filter bandpass is too narrow or the laser wavelength drifts to the filter's edge.

Laser wavelength stability will be an important consideration for 64-channel systems with 50 GHz channel spacing; 10 Gb/s transmission would require external modulation to reduce transient chirp.

Such rapid advancements in optical technology have made it possible to build WDM systems today with 60 to 80 OC-48 channels on a single fiber. This number may increase to 250 or more wavelengths within a few years with the availability of new optical sources and amplifiers. There are significant trade-offs, however, between the maximum non-regenerated span and the number of channels for a given data rate.

To maintain compatibility with deployments of 600 km systems, WDM equipment transporting greater than 32 OC-48 channels (through upgrades or new installations) will require higher-power amplifiers exceeding industry safety limits. Revised operational procedures and policies to support these conditions may be necessary.

At OC-192 rates, the number of WDM channels and the non-regenerated span are further limited because of high optical signal-to-noise ratio requirements, as well as impairments caused by fiber nonlinearities and polarization mode dispersion.

In short-haul metropolitan WDM systems, the amplifier and nonlinearity constraints are reduced and the total achievable capacity outpaced by the loss of multiplexing technology, which may ultimately be greater for short-haul networks.