Characterize DWDM Optical Amplifiers
Christian Hentschel, Hewlett-Packard, Boeblingen, Germany; and Doug Baney, Hewlett-Packard Laboratories, Palo Alto, CA -- Test & Measurement World, 11/1/1998
In multichannel dense-WDM (DWDM) optical communication systems, per-channel gain and noise figures describe the performance of an erbium-doped fiber amplifier (EDFA). Optical-gain measurements are fairly simple to make. In contrast, though, measuring noise figures presents a challenge because you first need precise measurements of amplified spontaneous emission (ASE), which aren’t easy to obtain.The noise figure, F, of an optical amplifier is given by Equation 1
where rASE represents the spectral density (W/Hz) of the ASE in the same polarization and at the same frequency, n, as the optical signal. The variable G represents the optical gain.1
 |
| Figure 1. Three tunable lasers pump an EDFA undergoing testing. When you remove one wavelength, you compensate by adding power at the other two wavelengths. |
 |
| Figure 2. This power graph shows two sets of measurement results made with all three lasers and only two lasers driving the EDFA. Turning off the laser at the middle wavelength and adding more power at the two other wavelengths lets you make ASE measurements at the middle wavelength. |
 |
| Figure 3. Measurements centered at about 1540 mm show how the power substitution technique provides good results at the center wavelength. |
 |
| Figure 4. Changing the wavelength to about 1550mm shows that the signal substitution technique also works at this wavelength. |
The optical presence of the signal itself complicates measuring the ASE density at the signal wavelength because the large signal masks the lower-level ASE noise. Standard ASE interpolation techniques are usually limited to channel spacings of 1.6 nm or greater, because the limited spectral resolution of many optical spectrum analyzers (OSAs) does not permit accurate measurements of the ASE at the standardized channel spacings of 0.8 nm. Alternatively, time-domain extinction,2 sometimes in combination with the reduced-source technique,2–4 can yield the ASE spectral density. This type of test requires laser pulsing, which usually reduces the input power substantially.
We have developed a new measurement technique, the signal-substitution technique, which allows us to measure the EDFA ASE directly at the signal wavelength. Unlike the time-domain extinction technique, the signal-substitution method does not require pulsing the laser sources. And compared with ASE interpolation, this new technique significantly reduces the OSA spectral resolution requirement, permitting more accurate measurements.
Using signal substitution, you switch off, or otherwise remove, the signal for the channel that you want to characterize. Then, you increase the power in the adjacent channels by a certain amount (see Eq. 2 below). This increase ensures that the total number of photons generated by the amplifier remain constant, hence the amplifier saturation state, ASE generation, and spectral gain remain constant.
With the signal at the measurement wavelength turned off, you increase the power in the immediately adjacent lower and upper channel so that the EDFA remains in the constant-saturation state. The power substitution takes place within a spectral range that is smaller than the width of the EDFA’s spectral holeburning (SHB) effect. The SHB effect is typically 4 to 6 nm wide, whereas the substitution takes place within typically 1.6 or 3.2 nm. You can assume such a small spectral range provides locally homogeneous broadening and negligible SHB. Equation 2 expresses this relationship, which ensures that the photon flux caused by the increased power in the adjacent channels is equal to the photon flux in the disabled channel:4
Gn represents the channel gains measured prior to substitution. The subscript n refers to the nth WDM channel, and DP refers to the input power added to the adjacent channels. The increase in power requires approximately 2 dB of reserve power for each channel.
In EDFA measurements that involve many channels, possible gain and ASE changes caused by spectral hole burning are further reduced because the saturation effect is widely distributed over the EDFA’s spectral range. Because the channels are closely spaced compared to absolute wavelength, you can remove the wavelength dependence from the equation above. And by adding power equally to the adjacent channels you can simplify Equation 2 to
At the longest and shortest wavelengths in the DWDM channel’s spectrum, you should apply the additional power to the only remaining adjacent channel, assuming you have enough reserve power. Alternatively, in order to save input power, you can interpolate the ASE obtained for the interior-channel and the measured exterior ASE to calculate the ASE for the end channel.
Ideally, you would have to subtract the ASE power from the measured signal output power in order to calculate the gain values to use in Equation 3. You don’t have to perform the subtraction, because a) at high input powers, the ASE contribution is negligible, and b) at low powers, a small inaccuracy in gains has no influence on the saturation state. During the final calculations of gain and noise figure, you should make the ASE corrections, however, because you will know the ASE values then.
Our experiments show that we can maintain the local and global amplifier saturation as we turn off power to the middle channel and add the appropriate powers to the adjacent channels. The maintenance of the saturation becomes obvious when you compare the ASE values near the substitution wavelength (locally) and the ASE within the remainder of the spectrum (globally).
We combined three tunable lasers and directed the signal to a 1480-nm-pumped EDFA undergoing testing (Fig. 1). An OSA measures signal powers and ASE spectral density. We set the three lasers to wavelengths of 1528.2 nm, 1529 nm, and 1529.8 nm with optical powers of –9 dBm. This caused gain compression of about 15.5 dB, in comparison with the unsaturated gain.
By using just the three lasers and centering their wavelengths close to 1530 nm, we tested the most challenging conditions in light of the strong saturation effects of just three channels and the strong and spectrally narrow nature of the EDFA SHB near the 1530-nm wavelength. A similar experiment run with 16 or 80 channels will yield even better results than those we obtained for the three lasers. Using more channels reduces the effect of each individual channel on the EDFA saturation and reduces local SHB.
The graph in Figure 2 shows two measurements: one in which we measured the ASE with all three channel signals present, and one in which we turned off power to the middle channel and increased the power to the adjacent channel according to Equation 3. The graph shows an excellent match—to within 0.05 dB between the ASE spectras in the spectral region of significant gain slope.
We repeated our measurements at a center wavelength near 1540 nm (Fig. 3). Individual channels powers of –20 dBm result in about 5 dB of compression. At 1540 nm, the ASE results for the three-channel case are within 0.06 dB of those for the two-channel case with added powers to maintain the saturation condition.
We made a third series of measurements (Fig. 4) centered at about 1550 nm with channel powers set to –9 dBm each, which results in 12.5 dB of gain compression. At this wavelength, the two ASE spectra agree to within 0.03 dB. SHB is evident in the measurement as a “gentle” valley around 1550 nm.
Additional experiments run with powers as low as –40 dBm per channel also showed excellent agreement between the three-channel measurements and the two-channel power-substituted measurements.
Overall, we found that when the adjacent channels lie within the width of the spectrally burned hole, you can maintain the local and global saturation state by increasing the adjacent channel powers, provided that the total output power on these channels remains constant. For the 100-GHz ITU wavelength plan, you can maintain the saturation state because the resulting 300-GHz substitution wavelength span is narrower than the SHB hole widths, even at the worst-case region near 1530 nm. The 300 GHz represents the total spacing of the three adjacent wavelengths.
Even under the 200-GHz wavelength plan, the channel separations are still less than the narrowest SHB hole widths, so this method will still produce accurate results particularly if you have a large number of channels. We confirmed that this technique provides accurate results by making additional measurements for which we used four lasers with 1.6-nm channel spacings. T&MW
FOOTNOTES
1. Fiber Optic Test and Measurement, Chapter 13, D. Derickson, editor, Prentice Hall, PTR, Englewood Cliffs, NJ. 1998.
2. Baney, D. M., “Gain and noise characterization of EDFAs for WDM applications,” paper Wa1, Proceedings of the Optical Fiber Communications Conference (OFC), Dallas, TX, 1997. Optical Society of America, Washington, DC.
3. Chou, H., and J. Stimple, “Inhomogeneous gain saturation of erbium doped fiber amplifiers,” Optical Amplifiers and their Applications Conference (OAA), Technical Digest Series Vol. 8, 1995. Optical Society of America, Washington, DC.
4. Baney, D. M., and J. Stimple, “WDM EDFA gain characterization with a reduced set of saturating channels,” IEEE Photonics Technology. “Letters,” p. 1615–1617, December 1996.
FOR FURTHER READING
Vobis, Joachim, “Use a New Technique to Characterize FO Amplifiers,” Test & Measurement World, Newton, MA. December 1996. pp. 23–30.
Christian Hentschel is the manager of the lightwave standards laboratory at HP in Boeblingen, Germany. He has a Dr.Ing. degree and has written a number of papers on lightwave
measurements.
Doug Baney
is a project manager for telecommunication technologies at HP Laboratories in Palo Alto, CA. He received a Ph.D. from the Ecole Nationale Superieure des Telecommunications in Paris, France.
