Preparing for the bandwidth explosion
By John Spencer, P.E.
This case study illustrates how to overcome economic and technical issues on a large fiber-optic ring network.
In 2009, a mid-size carrier was experiencing rapid growth in demand for Gigabit Ethernet (GbE) and 10GbE services. It operated a 26-node ring with a circumference of almost 1400 km. The ring had been cobbled together with a combination of eight-channel CWDM and eight-channel DWDM systems.
|Figure 1. Add/drop and pass-through ROADM implementations.|
The DWDM systems had been placed to accommodate 10-Gbps systems in the core of the ring, and the CWDM placed for 2.5-Gbps systems in the rural areas of the system. As such, it was difficult to grow the ring. The network had become high-maintenance from a network management and planning perspective.
The carrier considered a number of replacement DWDM system capacities (8-channel, 16-channel, 32-channel, or 40-channel). It wanted to start as small as possible, but have the ability to grow easily by simply placing transponders. The design needed to be cost-effective, simple to engineer and maintain, and easily scalable. The carrier had a timeline of turning up service within three months of awarding the contract. The network upgrade proposal was sent to multiple suppliers for competitive bid.
This article looks at some of the issues involved in upgrading a large network made up of different technologies and capacities, using the network just described as an example. These issues are economic and technical, and involve planning, cutover, and ongoing maintenance.
Initially, the carrier was unsure what DWDM system capacity to deploy. One alternative was an eight-channel system that would grow in bands of eight wavelengths as demand emerged. While it represented a lowest first-cost approach, an alternate proposal of starting with a 40-channel DWDM system turned out to be cost-competitive, even with an initial deployment of 8-10 wavelengths.
A significant disadvantage of the eight-channel DWDM system was its cost to scale. To add a ninth wavelength, visits would be needed to each site to install the second band of eight wavelengths. As a result, the 40-channel system was chosen and has proven to be the best choice in light of the actual growth around the ring.
The use of reconfigurable optical add/drop multiplexers (ROADMs) also was considered. As an alternative to the optical add/drop multiplexers (OADMs) that add or drop all 40 wavelengths, ROADMs could be placed at each node. As new circuits were added, visits still would need to be made at the end points (Nodes 12 and 20 in Figure 1) to add the new Ethernet transmission equipment, but connecting the circuit at all the intermediate nodes could be done remotely via the ROADMs.
An important advantage of a ROADM is that when coupled with an optical channel monitor, optical power measurements and adjustments can be made remotely or in fully automated fashion. When wavelengths are added, these adjustments are required to keep the power levels through the EDFAs in the ring at the levels needed to manage optical signal-to-noise ratio (OSNR) and receiver power levels, and prevent optical cross-talk.
Figure 2. Increased fiber and wavelength utilization via muxponders
Without ROADMs, as traffic demands around the ring increase, the time required by technicians to keep power levels balanced will increase proportionally; hence, ROADMs make a more compelling case from an operational expense (opex) perspective, particularly in large rings. In this specific case, however, ROADMs were not within the carrier's initial budget, but were considered as a future upgrade. As the number of wavelengths added increases, the cost of power balancing will increase–making ROADMs more attractive.
Two types of muxponders were considered. One type would multiplex four OC-48s onto a 10-Gbps line. The other would multiplex eight GbEs onto a 10GbE line. Both would efficiently use wavelengths to carry either four OC-48s or eight GbEs on a single wavelength (see Figure 2). The latter muxponder proved very attractive given the strong anticipated GbE circuit growth.
However, the muxponders were deferred until later in lieu of using the existing wavelength capacity. In the future, when the 40-channel DWDM system begins to reach capacity, muxponders will be added.
In the end, the carrier was very pleased to have a plan that addressed current needs with the 40-channel DWDM system and a migration path with ROADMs and muxponders to address future needs.
There are a number of technical issues to consider when providing capacity via a large DWDM ring.
Fiber loss: A carrier can be unaware of the actual loss in a fiber span for a number of reasons. The carrier may have been able to estimate the span loss for a previous bandwidth addition and still have sufficient system margin. However, estimated losses may not be adequate for a bit-rate upgrade. For example, a carrier might get by with estimated losses on short or medium-length spans but may encounter a redesign on a longer span due to underestimated loss. On the other hand, overestimated losses can cause overdesign and the unnecessary placement of EDFAs. The carrier may also be using leased fiber and be dependent on the records of the leasing company.
Actual field measurements are required to ensure that adequate margin is available and the placement of amplifiers and the number of amplifiers required meets the system design parameters.
Chromatic and polarization-mode dispersion: Chromatic dispersion is predictable and stable and can be compensated via dispersion compensation modules. However, it is essential to know the fiber type in the span(s) to accurately predict the chromatic dispersion characteristics.
In this case study, after the new DWDM system had been installed along with amplifiers and dispersion compensation modules, pre-turn-up testing showed a higher than expected bit error rate (BER) on a particular span. After measurements, it was determined that the leasing company's records were incorrect. The type of fiber according to the records was ITU G.652–but the type of fiber in the span was actually ITU G.655, which has different dispersion characteristics. The correct fiber type was entered into the engineering design tool and dispersion recalculated. Dispersion compensation modules were changed out accordingly and the span performed within BER limits.
Polarization mode dispersion (PMD) is less predictable and must be measured to know exactly its possible impact on system performance. PMD can especially be a problem on fiber placed prior to 1994.
Connectors and reflections: A dirty connector is one of the most common network problems but the simplest to correct. Dirty connectors cause higher span loss and contribute to greater reflections, and also affect chromatic dispersion and PMD measurements.
|Figure 3. Loss properties of Gaussian and wideband filters.|
Connector inspection and cleaning kits are readily available to check and clean a connector in a matter of minutes. The simple cleaning of a connector at both ends of a fiber span can eliminate many possible sources of system performance degradation and enable accurate measurements of other parameters. It is important to clean at the planning stage so that accurate measurements are fed into the engineering design tool.
DWDM AWG filter types: High-channel-count optical multiplexers/demultiplexers are based on array waveguide grating (AWG) technology. AWGs have typically two filter types associated with them: Gaussian or flat-top, also referred to as wideband. The spectral loss properties of these two filter types are demonstrated in Figure 3. Figure 3 is based on two back-to-back AWGs making a single optical add/drop multiplexer.
How a particular wavelength will be attenuated as it traverses a long chain of DWDM OADMs will depend, in part, on the frequency tolerance of that wavelength. Wavelengths generated by tunable transmitters are inherently stable, with a typical frequency tolerance of ± 2.5 GHz. Wavelengths generated by SFPs or XFPs are less stable, with a typical tolerance of ± 12.5 GHz.
As seen in Figure 3, a wavelength generated by an SFP/XFP that is offset by ± 12.5 GHz would have an additional 1.5-dB loss over a wavelength with no offset when passing through an OADM using Gaussian filters; the additional loss of the offset wavelength when using wideband filters would be negligible. If a ± 12.5-GHz offset wavelength generated by an SFP/XFP traversed 10 Gaussian OADMs, the power disparity with respect to a zero-offset wavelength could grow to 15 dB. By using wideband OADMs, however, the effect can be greatly mitigated.
The importance of basic testing
Several essential tests need to be made to ensure proper operation of a DWDM system. A simple cleaning of the connectors can insure that loss, reflection, chromatic dispersion, and PMD tests are all made accurately.
A fiber loss test will ensure that spans perform as engineered and designed. As mentioned previously, a chromatic dispersion test can ensure accurate data is used in design calculations. PMD testing is the most complex test and is necessary to ensure proper operation at 10 Gbps and higher.
Optical power balancing can be highly repetitive in large DWDM networks with many wavelengths being passed through EDFAs. When wavelengths are added to a DWDM ring, the power of all the wavelengths around the ring will potentially need to be adjusted. This has traditionally been done by making measurements with an optical power meter or spectrum analyzer and adjusting power levels accordingly by inserting attenuators. This is a very time-consuming process that involves making trips to multiple locations.
Going forward, installed optical channel monitors will provide measurements continuously without technician involvement. When used with ROADMs, optical power levels can be adjusted continuously, automatically, and remotely, again with no technician involvement.
What the carrier learned
The carrier learned a great deal from this exercise. They made the decision to install a higher-capacity 40-channel DWDM system to accommodate the anticipated growth. They decided against a lower-count eight-channel DWDM system that would have had a lower initial cost, but would have incurred greater cost to add wavelength bands as traffic grew. They developed a migration strategy to address the increased operational cost of optical power balancing through adding ROADMs and optical channel monitors in their DWDM ring network. As capacity is added and wavelengths are filled, muxponders will be deployed and wavelengths freed as necessary.
There are many technical issues in building a large DWDM ring. Most are straightforward, such as proactively cleaning dirty connectors and making appropriate tests to verify loss and chromatic dispersion, especially when the fiber is leased by the carrier.
The key lessons learned were the importance of planning with the big picture in mind and of appropriate testing and fiber characterization. While several of these lessons were learned along the way, this deployment stayed within the carrier's budget, was installed within its three-month time limit, and cut over without any service interruption.
John Spencer, P.E., is currently business and technology strategist for Optelian.