Optical signal-to-noise ratio characterization demands precision and flexibility
Physical layer testing of dense wavelength-division multiplexed (DWDM) systems in research and development, manufacturing, during installation, and for maintenance purposes includes the measurement of channel wavelength, channel power, and channel optical signal-to-noise ratio (OSNR). The two instruments typically used for making these measurements are the multiwavelength meter and the optical spectrum analyzer (OSA). While the multiwavelength meter has outstanding wavelength accuracy, it lacks the dynamic range required for making most OSNR measurements.
The wavelength accuracy of OSAs continues to improve, and they have a dynamic range more suited for measuring the system noise required to calculate OSNR. The dynamic range performance of many OSAs is demonstrated by measuring closely spaced continuous-wave (CW) carriers. But in most real-world applications, DWDM systems consist of modulated carriers.
Making good channel power measurements on modulated carriers requires a resolution bandwidth wide enough to encompass the modulation sidebands of the signal. A good noise measurement requires high close-in dynamic range for the rejection of adjacent channel power, and the ability to accurately measure the power density of the broadband noise.
Several methods can be used to optimize the various physical layer measurements of DWDM systems using an OSA. In addition, several accuracy considerations should be kept in mind in making these OSA measurements.
Channel wavelength measurements
Several automated functions are available on modern OSAs to identify the channel wavelength. The most popular is probably the peak-search function. Unfortunately, peak search does not return the most accurate channel wavelength value. There are several reasons for this lack of accuracy.
The first is that the peak of the signal is dependent on the shape of the filter used in detecting the signal. This filter shape is primarily due to the slit present in the optical path of the OSA and it determines the resolution bandwidth of the monochromator. Because the angle of the light incident on the slit varies over wavelength, the shape of the filter is not perfectly symmetric. As a result, the peak wavelength measured is not necessarily the channel center wavelength.
The other issue with peak search has to do with the digitizing of the detected signal. Assume a perfectly flat symmetrical filter is used to measure the signal. In this case, there are multiple points at the top of the filter with the same amplitude. Which of these is the peak?
A more accurate function to measure the center wavelength would be the 3-dB or center-of-mass methods. The center-of-mass method uses a weighted average of all of the points near the peak of the signal to determine the center wavelength of the signal and is probably the most accurate method. Unfortunately, most OSAs do not offer this functionality. The 3-dB method uses the wavelength values down 3 dB on either side of the peak to determine the center wavelength and is a function found on most OSAs. In evaluating the error between the center-of-mass method and the 3-dB method, we have found the differences to be negligible.
While the wavelength accuracy of the OSAs has been improving, there is still an important difference between the ways an OSA and multiwavelength meter make their measurements. With the multiwavelength meter, the incoming or unknown signal is beat against a reference light source to determine its wavelength. This difference is critical when measuring long-term signal drift. Because the signal is constantly compared to a reference signal in a multiwavelength meter, any changes in ambient temperature or pressure of the measuring instrument have little impact on the measured wavelength.
With an OSA, the measured signal travels in free space through the air in the monochromator. Any changes in the ambient temperature or pressure in this air affect the index of refraction of the air and thus will impact wavelength accuracy. The impact of these changes on the wavelength accuracy could be on the order of 3 to 5 pm per °C. When attempting to measure the wavelength drift of a channel or channels in a system, this error could be significant. For long-term drift measurements, the multiwavelength meter is still the measuring instrument of choice.
Making accurate signal power measurements on modulated carriers requires a resolution bandwidth wide enough to encompass the modulation sidebands of the signal. The narrow resolution bandwidth (RBW) advertised by OSA manufacturers provides little benefit in making channel power measurements if it is narrower than the modulation bandwidth of the signal. In this case, components of the signal to be measured are attenuated by the filter edges of the OSA. The goal is to have a resolution bandwidth great enough to encompass most of the modulation sidebands and a dynamic range good enough to reject adjacent signal content.
For a 2.4-Gbit/s modulated signal, a 10-pm RBW will introduce an error of approximately 1.4 dB in the channel power measurement. For a 10-Gbit/s modulated signal, a 15-pm RBW will introduce an error of approximately 5 dB in the measurement of the channel power. For these measurements, a 0.2-nm RBW is a far better choice. As long as the adjacent channels are rejected, the wider the better (see Fig. 1 and Fig. 2). For 100-GHz (~0.8 nm) spaced channels, a 0.5-nm RBW could be used.
Overall, the amplitude accuracy of an OSA is better for larger resolution bandwidths. As the resolution bandwidth increases, the edge shape becomes indicative of the overall filter shape. The edge shape can vary slightly over wavelength, and it is this variation that affects the amount of light incident on the photodetector, and thus the measured signal power. For a 60-pm RBW, the rising/falling edge makes up the entire filter shape. For a 0.2-nm RBW, these edges account for only 20% of the overall filter shape. A greater portion of the filter shape is the flat top. For this reason, the widest possible resolution bandwidth is chosen for accurate and repeatable signal amplitude measurements (see Fig. 3).
While wide resolution bandwidths are desired for accurate signal measurements, unless the filter is a "brick-wall" filter with extremely steep filter skirts, it will not have the dynamic range required to reject the adjacent channel signals when measuring the noise between these channels. However, as noted, choosing a narrow resolution bandwidth to make accurate noise measurements means sacrificing the channel power measurement. Both measurements need to be accurate to effectively calculate OSNR (see Fig. 4).
Furthermore, if truly measuring the noise between the channels, this noise is broadband so the amplitude measured is dependent upon the measurement bandwidth selected. For example, a noise level of -50 dBm measured with a 0.1-nm RBW would measure -40 dBm if a 1-nm RBW were used. For consistent noise measurements, noise power density should be used, and the reference bandwidth used should be stated-in this case, -50 dBm/0.1nm.
Additionally, for accurate noise measurements, knowing the resolution bandwidth of the filter is not enough. To determine the amount of broadband light incident on the photodetector, the "equivalent noise bandwidth" is required. Because the resolution bandwidth is only the nominal 3-dB width of the OSA optical filter, it does not define the total filter shape. Several filters with a 3-dB bandwidth could have various filter shapes and, therefore, various levels of light incident on the photodetector (see Fig. 5).
The goal is to determine the noise power density relative to a perfect 0.1-nm filter with a flat top and vertical edges-a brick-wall filter. While it is impossible to build such an optical filter, it is possible to calibrate the OSA and provide calibrated noise markers.
The result of calibrated noise markers is that the power is measured over the effective resolution of the filter, rather than providing a single-point power measurement as most OSAs measure. Because noise can vary from point to point (thus the name "noise"), this is a far more repeatable measurement.
The next step is to determine where to measure the noise. Choices include midway between two channels, at the minimum point between two channels, or at a fixed offset. For fixed channel spacing and the same modulation rate, midway between the channels may make sense. For channels with different modulation rates or unequally spaced channels, the minimum point between the channels may be a better choice. Finally, for the first and last channels and for systems where channels are missing, if the other two methods are not applicable, then the fixed offset method can be used.
Once a noise component is measured on either side of a system channel, the noise component at the center wavelength is interpolated. It is this value, along with the channel power, that is used to calculate the channel OSNR.
Tying it all together
The resolution bandwidth requirements for a good channel power measurement differ from those required for an accurate noise measurement. Making both of these measurements using the same resolution bandwidth compromises one or the other of the measurements. As a result, the accuracy of the calculated OSNR suffers.
One solution is a WDM channel analysis application that provides an automated dual sweep measurement of DWDM channels. In addition, the noise component can be measured using calibrated noise markers to enhance the accuracy of the resulting OSNR measurement.
For the most accurate measurement of channel wavelength in DWDM systems, the multiwavelength meter is still the instrument of choice. It also provides the most stable measurement for system drift measurements. Optical spectrum analyzers, with continued improvements in wavelength accuracy, provide excellent channel wavelength measurements and the ability to more accurately measure channel power and optical signal-to-noise ratio.
Jerry Chappell is a product marketing section manager and Scott DeMange is a product manager from Agilent Technologies, 1400 Fountaingrove Parkway, Santa Rosa, CA 95403. Jerry Chappell can be contacted at e-mail: email@example.com.
FIGURE 1. Amplitude error can be depicted as a function of the resolution bandwidth for 2.5 Gbit/s signals.
FIGURE 2. Amplitude error can be depicted as a function of the resolution bandwidth for 10 Gbit/s signals.
Figure 3. As resolution bandwidth increases, amplitude error due to filter edges is reduced.
Figure 4. Two sweeps of a DWDM signal: the power sweep (top) allows for more accurate measurement of the modulating signal, but cannot accurately measure the noise (bottom) between the channels; the noise sweep provides better resolution between the channels, but offers less accurate measurements of the modulated signal amplitude.
Figure 5. Two different filters, shown overlapped, with the -3-dB point. Different filter shapes could result in different powers being measured.