Multi-tiered approach to monitoring measures network performance


Jim Sirkis and Alan Kersey

Next-generation optical networks add functionality with smart amplifiers, dynamic gain equalizers, and optical crossconnects, but increasing complexity makes it more important to monitor spectral details of each WDM channel. Three distinct classes of optical monitoring—optical channel monitors, optical channel analyzers, and DWDM bit-error-rate testers—provide the essential, fast measurements required.

Service providers continue to struggle with the effect Internet traffic has on their infrastructures. In contrast to the predictable growth and distribution of voice traffic, Internet traffic by its nature occurs in bursts and its impact on network capacity continues to be difficult to anticipate. In addition, service providers are seeking transport systems that offer lower overall installed costs without sacrificing reach, bandwidth, or signal quality.

The key to implementing this strategy is the continued reduction of expensive regeneration equipment through extended use of optical transparency in transport links and subnetworks, which, in general, requires cascading more amplifier sites between optical-electrical-optical regeneration and performing certain routing and management functions directly in the optical domain. While this strategy increases reach and enables more highly interconnected optical layers, it can compromise the quality of the optical transport signals, particularly in terms of channel power uniformity and noise.

Consequently, future optical networks will rely to a greater extent on monitoring techniques that can be performed directly in the optical domain. This requirement will not be met by one universal device but by a family of optical-monitoring elements that provide a three-class tiered capability. This tiered strategy to optical-network monitoring provides monitoring capabilities that are relevant for differing needs within the network (see Fig. 1).

The application drivers of these monitoring approaches are quite diverse. For example, consider the effect of cascaded amplifiers on the optical channels in a DWDM system. Nonuniformities in the gain profile of saturated, passively gain flattened optical amplifiers are magnified by a multiplicative effect when more than a few amplifiers are cascaded.1

Signal distortions are even worse when amplifiers are operated in less than ideal conditions, which can occur when aging or environmental effects impact the gain profile, or when the total power entering amplifiers changes due to power imbalances or adding (or dropping) channels. The net effect of these network dynamics is the emergence of optical channel monitors (OCMs) and dynamic-gain-equalization modules to locally measure and balance the channel powers. Because of their linkage with amplifiers, OCMs will be used frequently in network links, offering a rich commercial opportunity. While initially targeted for channel-power equalization applications, OCMs will find use in other applications such as fault isolation and channel-routing supervision. The OCM provides the first tier of optical-monitoring capability.

Network designers have long recognized that equalizing channel power brings stability to the network, but it does not necessarily improve the optical signal-to-noise ratio. In fact, an unwanted consequence of cascading amplifiers is the accumulation of amplified spontaneous emission noise in each DWDM channel, even when the channel powers are equalized.2 In addition, the need for increased bandwidth and spectral efficiency has led to ever tighter control over DWDM wavelengths, especially in 25-GHz systems.

These factors created the demand for the second tier of channel monitors, called optical channel analyzers, capable of accurately measuring channel power, wavelength, and optical signal-to-noise ratio. Optical channel analyzers (OCAs) play an invaluable role in commissioning new DWDM networks and provisioning new DWDM channels in existing networks. For example, virtually all of the new generation of long-haul and ultralong-haul optical networks have automated end-to-end power-balancing algorithms that involve measuring the power at the receivers with an OCA and adjusting the laser power at the transmitter—all communicated through an optical supervisory channel. Similar functions exist for regulating channel wavelength as well. Because OCAs are more costly, they will be used less frequently than OCMs and in different locations.

Even with their impressive measurement fidelity, OCAs will never satisfy all network monitoring requirements because optical signal-to-noise ratio (OSNR) measurements do not provide a direct measurement of data quality.3 This is why bit-error-rate testers (BERTs) have always been used for qualifying, troubleshooting, and turning up networks, although this has primarily been done using bench-top or portable instrumentation. However, the rise in multiprotocol data transmission has increased the need for a third tier of monitoring capability, a permanent bit-error-rate test functionality integrated into certain network elements. This approach guarantees compliance with service-level agreements, while at the same time avoiding the expense of translating native data formats to higher-level data protocols.

The preferred implementation of permanent bit-error-rate testing in DWDM networks is to use a high-isolation tunable receiver combined with a single bit-error-rate test chip set to allow the bit-error rate of individual channels to be examined on command. This combination strikes the right balance between measurement fidelity and hardware cost. Even so, bit-error-rate tests are relatively expensive, and therefore will be relegated to use in network terminal elements and intersections between different service providers' networks.

Optical channel monitors (OCMs) are designed to provide fast and cost-effective channel-power measurements in DWDM systems. So-called "smart amplifiers" are the primary application driving OCM development, although other applications are starting to emerge, particularly related to elements that provide for a dynamic optical layer, such as optical crossconnect switches and reconfigurable add/drop multiplexers.

Interestingly, OCAs (not OCMs) were originally envisioned as the measurement tool for smart amplifiers. Responding to the potentially lucrative market opportunity, a wide variety of OCA solutions were proposed, including bulk diffraction gratings or blazed fiber Bragg gratings with indium gallium arsenide (InGaAs) detector arrays, scanning Fabry-Perot filters, arrayed waveguides with potodetector arrays, and tunable Bragg gratings.4,5,6

It soon became apparent, however, that the high cost of these OCA solutions would ``limit their deployment to applications absolutely requiring OSNR and wavelength measurement, which often does not include smart amplifiers. This revelation led to the demand for a more limited, lower-cost optical monitor, thus leading to the emergence of OCMs. Unit cost being a prime driver has led to additional solutions to those mentioned above, including OCMs based on microsystems, scanning spectrometers, and micro-optical benches.

The functional differences between OCMs and OCAs are ultimately derived from their respective filter function characteristics. Because OSNR measurement is not required, the OCM filter function design is driven only by the need to uniquely identify individual channels and to measure their power. From a functional perspective, OCMs measure absolute channel power to within ±0.5 dB, identify channels without prior knowledge of the channel plan, and make full C-, L-, or S-band measurements in less than 0.5 seconds. Filter characteristics are factored into the OCM design (see Fig. 2).

This filter function, which was designed for an OCM operating with 50-GHz channel-plan networks, has a resolution bandwidth (RBW) of 180 pm and an adjacent channel isolation of -27 dB. These two filter specifications play different roles in the OCM design. The RBW dictates the channel resolving power, which in this case means that two equal power channels can be uniquely identified if their channel separation is greater than 180 pm. This RBW is much narrower than the 50-GHz channel spacing, and is required to account for unequal channel powers and for drift in the channel wavelength.

Alternatively, the adjacent channel isolation plays an important role in determining OCM power accuracy because the total measured power is given by the integrated power under the OCM filter function. The worst-case scenario in this regard is when the two adjacent channels contain much more power than the channel being measured. The example filter function was designed by assuming the adjacent channels will be no more than 10-dB stronger than the center channel, and as a result, the -27-dB adjacent-channel isolation of this filter yields a maximum crosstalk error of 0.17 dB. This leaves a 0.33- to -0.5-dB error band to accommodate the other error sources, such as polarization-dependent loss, calculation errors, and environmentally dependent loss mechanisms.

It is worth noting that RBW and adjacent channel isolation are not independent, their linkage arises through the type of filter used in the OCM design. For example, a Fabry-Perot filter will have worse adjacent-channel isolation characteristics than a bulk diffraction grating with an identical RBW. This interdependence between RBW, adjacent-channel isolation, and OCM performance has led to the use of peak-to-valley ratio (PVR) as an aggregate metric for OCM filter performance (see Fig. 3).

Using data with simple centroid and peak detection algorithms is sufficient to identify channels and calculate their respective powers to within ±0.5 dB (see Fig. 4).

Finally, it is worth noting that while the filter function discussed seems nearly ideal for OCM applications, deconvolution by computer simulation can always be used to minimize the effects of adjacent-channel crosstalk for OCMs with suboptimal filter functions. However, deconvolution always adds measurement uncertainties and increases computing power requirements.

The role played by optical channel analyzers (OCAs) in network systems is much different from that of OCMs. Optical channel analyzers offer a higher level of functionality because they accurately measure optical power, center wavelength, and OSNR. In contrast to OCMs, OCAs generally provide channel power accuracies better than ±0.5 dBm, center-wavelength accuracies better than ±20 pm, and OSNR measurements down to 35 dB with a ±0.1-dB accuracy. These requirements place very different constraints on the requisite filter function.

Accurate wavelength measurement implies highly stable and referenced wavelength measurement capability, whereas the power and OSNR requirements have competing implications. Accurate channel-power measurements require a filter RBW wide enough to encompass the laser signal sidebands to prevent sideband attenuation, yet a narrow enough RBW to have strong adjacent-channel isolation to avoid crosstalk error.

In contrast, accurate OSNR measurements can only be made with a narrow RBW filter. Assuming that the primary source of noise is amplified spontaneous emission, then signal noise is measured at the interchannel positions and interpolated to the channel's center wavelength. In this case, a narrow RBW (typically on the order of 100 pm) enables the interchannel noise to be measured without contributions from power in the adjacent channels.7 Note also that the defacto industry standard is to provide OSNR normalized to a 100-pm RBW.

The two obvious approaches to meeting the complete OCA measurement requirements are either to use two distinct filters, or to use a filter with variable RBW capability (see Fig. 5, top). This figure shows the spectra of a tunable Bragg grating filter with a center-wavelength tuning range over the entire C-band, and a RBW tunable from 100 pm to 550 pm. These RBWs correspond to an OCA designed for 50-GHz network systems. The wider filter has a RBW of 260 pm and an adjacent-channel isolation greater than -35 dB. For a 10-Gbit/s system, this 260-pm RBW limits the sideband error to less than 0.1 dB, whereas the -5-dB adjacent-channel isolation limits the crosstalk error to less than 0.03 dB.

On the other hand, the 135-pm RBW setting allows accurate noise measurement, provided the filter can be accurately positioned at the interchannel locations. Using a 260-pm RBW filter (see Fig. 5, center) and a 135-pm RBW filter (see Fig. 5, bottom) increase the ability to resolve channels and noise for a given composite DWDM optical signal. Of course, positioning the filter at interchannel locations and meeting the ±20-pm channel-wavelength accuracy requires precise wavelength control and referencing. For example, as the 260-pm RBW filter is tuned over the 4-THz L-band for temperatures ranging from 0°C to 65°C the center wavelength accuracy can be determined (see Fig. 6). The ±6-pm (±0.75-GHz) error band exhibited by this filter meets all envisioned OCA measurement applications.

In spite of the enhanced functionality offered by OCAs, they still have limitations because OSNR does not account for important signal impairments such as chromatic dispersion or polarization-mode dispersion. Therefore there will always be instances where data streams with good OSNR characteristics will have unacceptable bit-error rates.8 As a result, bit-error-rate tests will always play an important role in network monitoring.

Ultimately, the goal of any monitoring system is to provide a metric to indicate and ensure the quality of service the carrier can provide its customers. As we have shown, OCMs and OCAs provide measurements of the principal analog optical parameters of the optical signals, but provide no indication of the quality of the data encoded onto each optical carrier. Whereas amplifier tilt, accumulated gain imbalances, and OSNR can be tracked by an OCA, the effects of transmission impairment such as chromatic dispersion and polarization-mode dispersion are not directly observable using direct optical monitoring.

To provide a measure of the quality of the signal at the data level, a measure of the bit-error rate, or Q-factor, of the transmitted data is required.9 This is the highest tier of optical monitoring that can be implemented; it measures directly the quality of the optical transmission and does not depend on an inference between OSNR and bit-error rate in the optical layer.

Bit-error-rate testing is invaluable at the system level in certain network conditions. First, as a network link is commissioned, the equipment provider and operator have a need to understand the optical transport-signal quality by analysis of bit-error rates. An important situation where bit-error-rate testing, in some form, is absolutely required is at the interface of two operators' networks, where there are direct commercial or legal consequences of confirming quality of service. As a result, optical-electrical-optical conversions at these intersections are common and necessary. In essence, this type of measurement can help avoid disputes between network operators and ensure quality of service is maintained and propagated over such interfaces.

In next-generation networks, where increased transparency will prevail, performing quality-of-service measurements within "all-optical islands" (or domains) will be all the more difficult because client information (SONET/SDH, ATM, and so on) is only available at the edges of optical transport systems, at regeneration sites, or at optical-electrical-optical interfaces between different operators' networks. As a result, there will always be a requirement to have a flexible optical monitor located strategically within the optical transport network that can directly access the bits of any channel in a DWDM system.

Measuring bit-error rate, or Q-factor, for each channel in a DWDM system can be a laborious task, particularly with 50- and 25-GHz systems, which can have upwards of 100 and 200 channels, respectively. Typically, bit-error-rate test chip sets are used with optical drop units that are sequentially stepped over the ITU grid to evaluate all channels using semi-automated or automated processes.

Jim Sirkis is director of advanced technology and Alan Kersey is chief technology officer at CiDRA, 50 Barnes Park North, Wallingford, CT 06492. They can be contacted at 203-294-4211 or and


  1. P. C. Beck, N. A. Olson, and J. R. Simpson, Academic Press, San Diego, 1999.
  2. S. H. Yun et al., IEEE Phot. Tech. Lett., 11 (10), 1229 (1999).
  3. G. Jacobsen, L. Gillner, O. Franson, J. Opt. Comm., 21 (1), 20 (2000).
  4. J. L. Wagener et al., Proc. ECOC/IOOC'97, Edinburgh, U.K., 5, 65 (1997).
  5. W. L. Emkey, Proc. NFOEC, Tuesday (2000).
  6. C. Miller and L. Pelz, cmocarticle.pdf.
  7. J. Chappell and S. DeMange, WDM Solutions, (November 2000).
  8. C. P. Larson, and P. O. Andersson, Opt. Net. Mag., (October, 2000).
  9. G. Bendelli et al., Proc. Of ECOC, (2000).
More in Home