January 1, 2008

A Delicate Balance: Alternatives for Multi-Wavelength Systems

Co-propagating multiple wavelengths over a common fiber provides an option for logically segmenting an existing physical node.

By Dr. Tim Brophy, Fernando Villarruel and Dr. Kuang-Yi Wu &

Node splits and node segmentation are among the tools available to cable operators to deal with seemingly ever-increasing bandwidth demands and plant extensions. But how can a node be segmented when there is not enough fiber in the outside plant? Overlashing new fiber is certainly one possibility, albeit capital intensive.

A cost-effective alternative, transmitting multiple wavelengths through existing glass, is promising but poses certain challenges. Initial attempts using coarse wavelength division multiplexing (CWDM) have given way to dense WDM (DWDM) in the 1,310 nm band in order to overcome the impairments of widely spaced wavelengths. The requirements to provide full-band transmitters under the restrictions of fiber nonlinearities that go beyond the usual transmitter design constraints lead to tradeoffs among the link distance, optical power, and the number and spacing of the wavelengths in the system. While the fundamental issues arise because of fiber physical considerations, they are mitigated by considering the system as a whole and modifying or constraining the transmitters, the RF loading, and the passives in the system.
   
Here follow some possible solutions to choosing wavelengths that can successfully carry multiple forward path payloads and yield near single-transmitter performance. Preliminary data indicate substantial improvement relative to CWDM systems and a commercially viable product.

Applications

Networks today include an assortment of plant styles, each with its own character. There are established neighborhoods where physical growth is constrained by available real estate and which is only gradually changing the services required. This is a source of relatively stable and continuing revenue, and the introduction of new services such as voice over Internet protocol (VoIP) is slower than the uptake of video on demand (VOD) or digital tiers. Some new subdivisions adjacent to established neighborhoods need plant extensions. These “brown-fields” sometimes drive existing plant upgrades or require fiber builds all the way to the new homes, driven by builder competition and requirements. Although total green-field builds are becoming ever rarer, they give the opportunity to re-think physical plant logistics and have an impact on upgrades elsewhere to maintain commonality and implement best practices.

The driving application of co-propagating multiple wavelengths over a common fiber is to logically segment an existing physical node. (See Figure 1.) Segmentation is simple in nodes with available slots by adding the additional receivers and reconfiguring the RF output. For a node that has either a restricted number of RF ports or available receiver slots, an additional physical node, sometimes referred to as an “extender” node, can take advantage of the wavelengths by mounting the demultiplexer inline with the original node and installing fibers between the two nodes. As an added benefit, it may also be possible to balance the subscriber load on each of the ports (of both nodes).

In the near term, it is expected that segmentation will need fewer than four wavelengths, and buildouts would follow demand. In this situation, wavelength and path redundancy can be implemented to take advantage of common passive configurations. (See Figure 2.)

Further, if a service group is at a minimum useful size, unused capacity can by employed for commercial services overlay, or to provide a dedicated point-to-point link for a power user purchasing large capacity, which would otherwise come from the shared pool.

One possible network plan is to keep it looking like one network and at the same time be able to segment it into different groups as needed. This allows using a common set of network management tools for troubleshooting, maintaining and provisioning the physical plant, but also gives service divisions access to a group of customers in a single neighborhood, or the group that has kids in different age ranges. It allows targeting particular VOD offerings or sports packages; the physical partitions are designed to be easy to maintain, even during evolution toward all digital and reclamation of some analog channels.

Design

To meet system level requirements means ensuring point-to-point link specifications in the multi-wavelength environment. Overcoming system limiting impairments stemming from fiber and components involves working with existing plant and looking at the entirety of the system to remove the constraints without burdening the plant with costly wreck-outs or difficult integration. Early attempts at multi-wavelength solutions were hampered by composite second order (CSO) penalties from high levels of Raman crosstalk, which limited the number of wavelengths and the fiber link length.

Raman crosstalk levels are approximately proportional to the spacing between the generating wavelengths. Therefore, one obvious solution is to move the wavelengths closer together, to reduce the level of the crosstalk. In order to be compatible with a system CSO specification of -65dBc (cumulative from Raman, passives, and transmitter performance), two wavelengths in 25 km of fiber cannot be separated by more than about 5 nm, and for four wavelengths, this value drops to about 3 nm. In terms of balancing tradeoffs, Raman crosstalk leads to limitations determined by the mix of fiber length, optical launch power, and the number and spacing of wavelengths. In conjunction with these, crosstalk levels (of various origins) may also be improved by grooming of the RF signal driving the transmitter. Details of the grooming are proprietary and in the patent application process, but the net result is lower Raman Crosstalk and therefore an improvement in one or more of the tradeoff variables (i.e., fiber span, optical launch power, wavelength range, or number of channels launched).

Having determined the Raman limits, we turn to the actual wavelengths to be used. Four wave mixing (FWM) is an example of a non-linear, energy-transferring impairment that can occur when multiple wavelengths travel in the same direction in the fiber, producing new wavelengths as a result of the fiber properties.

There are two primary concerns: the location of the new wavelengths and their power. On-channel FWM products may cause two detrimental effects. The first is to generate noise in the RF spectrum when the signal and the beat are close enough so their optical spectra overlap. The noise is broadband and it will be visible on an RF spectrum analyzer, degrading the CNR of the desired signal. A second, even more insidious effect occurs when operating near the zero dispersion point (ZDP) of fiber. Even if wavelengths are chosen so that beats fall between channels, high-amplitude FWM products are generated. These FWM products mix with the original channels to form secondary FWM signals which always fall on the original wavelengths, and this produces a CSO degradation in that original channel.

The FWM products give multiple impairments near the ZDP, and dispersion combined with laser chirp degrades CSO. To ensure a more universal (craft-friendly, plug-and-play) solution for the installed fiber base, the ZDP may range many nanometers. Both longer and shorter wavelengths (away from ZDP) solve the FWM issue, but short wavelengths suffer from increases in fiber loss (to 0.42 dB/km at 1,271 nm) and CSO degradation, especially at higher RF frequencies. Also, evenly spaced channels will have mixed products that coincide with the desired wavelengths.           

When confronting this issue in WDM systems carrying Ethernet, Forgieri developed an algorithm to choose wavelengths so that the FWM products fall (harmlessly) in the spaces between the desired channels. An illustrative example is shown in Figure 3, where the shortest wavelength is given as a reference frequency, and the x-axis is plotted in terms of frequency offset from the reference. With this system in place, the wavelength span of eight channels is ~8 nm when the reference is 1,310 nm. Unfortunately, the additional CSO from the secondary FWM products and statistical range of absolute ZDP itself limit solutions that will work in systems operating too close to the ZDP. Conversely, the ability of existing electronic pre-distortion circuitry to correct for the chirp and dispersion induced CSO at slightly longer wavelengths (see Figure 4), which still benefit from the similarity to existing 1,310 nm transmitters, make the longer wavelength much more desirable.

In validating the analytical predictions with experiments, it became apparent that ripple in passives was a dominant source of CSO in the system. Using available descriptions the passband slope of the filter is limited to 0.1 dB/nm for CSO < -70 dBc. (See Figure 5.) For demultiplexers, which are subject to field temperature ranges, the problem is exacerbated by the requirement that the local slope (i.e., the slope the optical channel occupies at any given instant) vary as the filter moves with temperature and must be extended to include the potential wavelength drift of the source over its lifetime. By leveraging optical coating and packaging technologies, filters with superior specifications that are consistent with manufacturing processes have been realized. Polarization dependent loss issues are also well-controlled.

It remains to be shown that the impairments from the cross phase modulation (XPM) - which is also manifest as a crosstalk - is compatible with system performance. It has been shown analytically that XPM has the most impact on higher RF frequencies when the spacing between two wavelengths (one of which modulates the index of the fiber to the detriment of the other) is small. It then suffices to show that the two closest wavelengths do not impair each other. A plot of the total system level crosstalk (including XPM and Raman) is shown in Figure 6. The crosstalk level is approximately the same at high and low RF frequencies, which shows good balance between the Raman and XPM limitations. The two quadrature amplitude modulation (QAM) test channels are not shown, but the crosstalk level is higher because of the reduced channel power (-6 dB relative to analog). Each measures better than 56 dB below the desired carrier, giving a CSO that will meet system spec.

Performance

Using the criteria for wavelength selection, spacing, launch power, and link length together with well-controlled specifications on both transmitters and passive elements, a laboratory system has been characterized and tested for performance. (See Figure 7.)

The system comprised four wavelengths in the 1,330 nm range, whose polarization was aligned to yield the worst case Raman crosstalk. After multiplexing, a temporary tap (5 percent) was inserted to measure the polarization alignment. After 25 km of standard SMF-28 fiber, the fiber was again temporarily tapped to measure the FWM product amplitude and wavelength. After demultiplexing, the individual wavelengths were fed to a standard forward path node receiver. The RF content on the transmitters included NTSC-78 analog carriers (50-550 MHz) and 73 channels of QAM loading to 1,002 MHz. Two additional QAM channels (center frequencies 559 and 859 MHz) were fed with pseudo random bit streams and used to measure bit error rate (BER) and modulation error ratio (MER).

Performance results are recorded in Table 1 and compare favorably to standard performance 1,310 nm transmitters with 52/65/70 CNR/CSO/CTB. The demultiplexer, whose system location implies a more severe environment, was subjected to temperature extremes from -20 degrees C to 85 degrees C to simulate the node environment and ensure CSO did not degrade because of the passband slope variation with temperature. All node components were unaltered to ensure compatibility with existing plant. A plot showing the original wavelengths and the relative locations of the FWM products is shown in Figure 8.
 
...and cost

In all essentials, the transmitters are standard 1,310 nm forward path designs with the exception of wavelength selection and whose tuning provides adequate single wavelength performance.

Therefore, the commercial availability of these transmitters follows that of laser chips with the appropriate wavelength only, which, because of the relative quantities, are anticipated to be in the range of 10 to 15 percent higher cost than the 1,310 nm equivalents. The multiplexers and demultiplexers are matched to the transmitters and are also expected to be more costly because of the lower quantities available. Overall link costs are expected to be within 25 percent of comparable 1,310 nm links, which compares very favorably relative to the option of adding new fibers in a system, estimated in the $20,000/mile to $30,000/mile range depending on the aerial/underground ratio of the plant.

Summary

Preliminary test results of a four-wavelength system over 25 km of SMF-28 are compatible with system level requirements and at a very competitive price level to alternative schemes. Leveraging prior investments into linear 1,310 nm transmitter technology with the comparatively modest extension in wavelength into the 1,330 nm window yields a moderate price premium that, because of processing familiarity, will soon be commercially available in quantity. Hand in hand with the available transmitters are passive multiplexers and demultiplexers with extended temperature range of operation for use in field splice enclosures in common use in the HFC plant today.

The key to success has been the balancing of the nonlinearities occurring in the fiber with system level performance criteria, to ensure that overcoming the limitations is robust, repeatable, and compatible with real-world deployments. Refinements of the techniques and extensions to the product offerings are continuing so that operators will be provided with the flexibility needed to meet the increasing demands placed on their infrastructure to supply adaptable and targeted content to adaptable service groups.


Dr. Tim Brophy is director of Network Architectures, Transmission Network Systems, for Scientific Atlanta. Reach him at tibrophy@cisco.com. Fernando Villarruel is systems engineer, Transmission Network Systems, Scientific Atlanta. Reach him at villarf@cisco.com. Dr. Kuang-Yi Wu is senior staff engineer, Transmission Network Systems, for Scientific Atlanta. Reach him at kuanwu@cisco.com. The full white paper from which this article was adapted is available at www.scientificatlanta.com/multiwavelength_paper.