Optical amplifier measurements must meet stringent criteria

April 1, 2001

Jack Dupre

Erbium-doped fiber, Raman, and semiconductor optical amplifiers face rigid performance requirements. Each type of amplifier measurement has unique issues that must be resolved.

Higher channel density and wider optical bandwidth per fiber has placed stringent performance requirements on optical amplifiers used in booster, in-line, and preamplifier applications. Additional requirements are imposed by all-optical networking, in which channels are dynamically added and dropped. While the preponderance of optical amplifiers today is of the erbium-doped fiber (EDFA) type, Raman and semiconductor optical (SOA) amplifiers are also commonly used. Each amplifier type has unique measurement issues.

Gain and noise figure characterizations for amplifiers with multichannel stimuli are complex because the parameters depend on power levels in all input channels. To simplify test apparatus, signal-spontaneous noise figure, which is derived from amplified-spontaneous-emission (ASE) measurements on an optical spectrum analyzer (OSA), is commonly used. To measure ASE in the presence of the spontaneous emission from the laser sources, interpolated source subtraction (ISS) and time-domain extinction (TDE) methods are used.

To precisely characterize the gain and noise figure for a particular multichannel plan using these methods, the multichannel test source must duplicate the plan in wavelength and power level. This layout is problematic for testing amplifiers that are used with a number of channel plans because the source is very complex and must be reconfigured for alternative channel plans. By recognizing the largely homogeneous saturation characteristics of the EDFA, however, a reduced set of laser sources may be used with a minor compromise in measurement accuracy (see Fig. 1).1
A limitation on the reduction in the number of lasers is imposed by spectral hole burning (SHB). Spectral hole burning is a wavelength-localized depression in gain that is dependent on signal power (see Fig. 2). It is important that in regions where SHB is significant, at least one test laser must be placed within a spectral hole width. Careful measurements of SHB show that hole depth is deepest and width the narrowest in the 1535-nm region and is small above 1550 nm. Simulations suggest that 16 to 32 equally spaced lasers are required to eliminate errors due to SHB. Judicious wavelength placement (closer spacing in the 1535-nm region) could further reduce this number. For L-band designs, it has been reported that the localized gain depression caused by SHB is not a factor.2

A further refinement of the reduced source approach is to combine a small-signal probe with the saturating multichannel source. The probe may be either a tunable laser or broadband noise source. The broadband-noise probe method, often called noise gain profile (NGP), is generally preferred because of its measurement speed. It uses a noise source such as an edge-emitting LED to simultaneously probe the entire EDFA band. It is often implemented with modulated saturating laser sources and in conjunction with TDE (see Fig. 3). Compared to the direct TDE approach, NGP has some advantages. Resolution of the gain and noise figure data is limited only by the OSA's display resolution—not by the number of lasers in the multichannel source. By keeping the total input power constant, the gain and noise figure is largely immune to power-level variation of the individual lasers. Furthermore, measurement speed is independent of the number of lasers and so can be faster for large channel count.

While TDE and NGP are inherently different measurement methods, it is important that they measure the same gain and noise-figure parameters. To confirm this, measurements made with three methods were compared and results for gain and noise figure compiled (see Fig. 4). In addition to results for TDE and NGP, measurements were also made using the ISS method. For TDE and ISS, data are only available at the wavelengths of the multichannel source, while for NGP the data are continuous from 1528 to 1563 nm. The measurements for all three methods agree to within ±0.125 dB.

POLARIZATION EFFECTSAnother important issue in testing EDFAs is the effect of polarization. The test setups described above both incorporate polarization controllers to either scramble the state of polarization (SOP) in order to measure amplifier parameters for the average SOP, or vary the SOP and make measurements to collect the polarization dependency of the amplifier parameters.

When testing with a multichannel source, the polarization-hole-burning (PHB) phenomenon is of particular concern. To understand the effect of PHB, consider two signals in the same wavelength vicinity—one large and one small. The large signal sets the inversion level. An additional laser acts as a small signal probe. Because of PHB, the gain in the polarization state of the large signal is depressed. This gain depression will also impact the small signal if it is in the same polarization state. If the small signal is in the polarization state orthogonal to the large signal, however, it does not experience the gain depression. The difference between the gain in the polarization state aligned with the large signal and that for the orthogonal state is usually referred to as polarization-dependent gain.

With a multichannel source, it is usually desirable to measure gain with randomly aligned source polarization. This can be accomplished in three ways (see Fig. 5). The first method is to individually scramble each source. The second is to slowly scramble the combined sources followed by a birefringent element to decorrelate the SOP of adjacent channels. The third method is to utilize a fast polarization scrambler after the combiner. In this context, slow and fast scrambling is done with respect to the erbium-ion relaxation time.

RAMAN AMPLIFIER TESTINGRaman amplification increasingly is considered to extend the wavelength range for a transmission system or to extend the distance between repeaters. The active element that is pumped to achieve gain may be the actual link transmission fiber itself or, like a conventional EDFA, the active fiber may be packaged with the pump laser. In the case of Raman-pumped transmission fiber, significant performance advantages can be realized over EDFA-only systems.3 In addition, WDM pumping can provide a 100-nm operating band with very-flat gain characteristics.4 Test issues result primarily from a unique noise-generation mechanism and much faster time dynamics when compared to the EDFA.

The important noise mechanisms in a Raman amplifier are ASE and double-Rayleigh scatter noise. As with an EDFA, ASE can be measured with an OSA and the signal-spontaneous noise figure, Fsig-spon, can be calculated. Time-domain extinction, as discussed previously for the EDFA, cannot be used because of the significantly faster gain-relaxation time for Raman gain; interpolation must be used. Another significant issue is that double-Rayleigh scattering cannot be observed as an increase in ASE because it results in heterodyne noise in an optical receiver. The total noise figure—which includes the double-Rayleigh backscatter-noise component— must be measured with an electrical-spectrum-analyzer method.

The electrical-spectrum-analyzer test method for measuring Ftotal was once commonly used for EDFA measurements. Total noise figure is defined in terms of the electrical signal-to-noise ratios at the input and output of ideal e/o and o/e converters. Ftotal includes the effect of double-Rayleigh. But this method was abandoned because of the difficulty to obtain repeatable and accurate results and because, for the EDFA, Fsig-spon measured with an OSA was sufficient.

One of the major sources of measurement uncertainty was the calibration of the optical receiver. By using a relative-instrument-noise (RIN) standard,5 this major source of uncertainty can be greatly reduced. Consisting of an ASE source and a stable bandpass filter, the optical power is measured on a well-calibrated optical power meter (see Fig. 6). The intensity-noise spectral density is then known precisely and used as the calibration standard for the optical receiver.

While the EDFA has become the prevalent amplifier type and Raman is emerging to provide better network performance, the SOA continues to progress as a potentially more compact and less expensive alternative. While not usually considered as a commercially viable option for 1550-nm transmission systems because of the refinement of the EDFA, 32 channel x 10 Gbit/s over 160 km using SOAs has been attained in a laboratory experiment.

The semiconductor optical amplifier also has applications in optical switching and as a nonlinear device in applications such as wavelength conversion. The SOA may also be applied in the 1310-nm window where doped-fiber amplifiers have not proven to be viable.

Test methods for gain and noise figures of the SOA can benefit greatly from the EDFA experience. Signal-spontaneous noise figures using the OSA interpolation method may be used. Like the Raman amplifier, TDE cannot be used because of the short carrier lifetime.

Additional test issues relate to properties that are unique to the SOA. These include polarization dependence, gain ripple, and gain-clamping laser artifacts. In an EDFA, polarization dependencies are so small that they may be ignored in many applications. This is not true with the SOA in that it is inherently polarization-dependent and excellent design is required to minimize this effect. Gain ripple with subnanometer period is also inherent due to reflections between the gain-medium facets. Active gain clamping with a control laser is commonly employed to achieve automatic gain control and the resulting gain-clamping laser leakage needs to be measured. While many practical problems need to be overcome, the SOA promises to be a major competitor as the prevalent gain element in metro and access networks.

The author wishes to acknowledge Douglas Baney of Agilent Technologies Laboratories and Jim Stimple of Agilent Technologies, Lightwave Division, for their many contributions to this paper.


  1. D. Baney and J. Stimple, IEEE Phot. Tech. Lett. 8, 12 (1996).
  2. F. A. Flood, IEEE Phot. Tech. Lett. 12, 8 (2000).
  3. C. Fludger, A. Maroney, and N. Jolley, OFC 2000, paper FF2-2.
  4. Y. Emori and S. Namiki, OFC 1999, paper PD-19.
  5. D. M. Baney and R. L. Jungerman, OAA 1997, paper MB3.

Jack Dupre is a strategic business consultant in the Lightwave Division of Agilent Technologies, 1400 Fountaingrove Parkway, Santa Rosa, CA 95403. He can be reached at 707-577-2331 or [email protected].

Sponsored Recommendations

Scaling Moore’s Law and The Role of Integrated Photonics

April 8, 2024
Intel presents its perspective on how photonic integration can enable similar performance scaling as Moore’s Law for package I/O with higher data throughput and lower energy consumption...

Coherent Routing and Optical Transport – Getting Under the Covers

April 11, 2024
Join us as we delve into the symbiotic relationship between IPoDWDM and cutting-edge optical transport innovations, revolutionizing the landscape of data transmission.

Supporting 5G with Fiber

April 12, 2023
Network operators continue their 5G coverage expansion – which means they also continue to roll out fiber to support such initiatives. The articles in this Lightwave On ...

Moving to 800G & Beyond

Jan. 27, 2023
Service provider and hyperscale data center network operators are beginning to deploy 800G transmission capabilities – but are using different technologies to do so. The higher...