Dynamic gain equalizers balance power in long-haul systems


The amplification of multiple optical channels has increased the distance between electrical regenerators almost an order of magnitude and has reduced long-haul transmission costs. Further increases in optical transmission distances are limited by the accumulated power differences among multiple channels, which reduce the optical signal-to-noise ratio and create the need for electrical regeneration. These power differences come from many sources, such as the nonuniform loss spectrum of transmission fibers, power transfer due to stimulated-Raman scattering, and the gain ripple in optical amplifiers.

Dynamic gain equalizers (DGEs), also known as dynamic gain-flattening filters, can overcome these power differences and enable extended-reach long-haul systems. Because DGEs have such an effect on improving system efficiencies and reducing cost, a key metric in evaluating DWDM systems has become the link distance allowed without regeneration.

Standard erbium-doped optical amplifiers have a gain that varies across the wavelength spectrum due to the atomic energy levels of the erbium ion. Static gain-flattening filters, based on thin-film coatings or long-period Bragg gratings, are designed with a loss function inverse to the erbium gain curve and are added to reduce the amplifier gain variation to within ±0.5 dB. After 20 amplifier spans, the target distance for extended-reach application, this static gain error sums to 10 dB. Also, these static filters cannot compensate for the power transfer caused by stimulated-Raman scattering, changes in the amplifier that occur with changes in channel usage, changing traffic patterns, or amplifier aging effects. These varying power differences limit the signal-to-noise ratio achieved at the receiver.

Several DWDM system manufacturers have introduced extended-reach and ultralong-haul DWDM systems that reduce cost by reducing the number of electrical regenerators along a link. However, these designs place stringent demands on the allowable gain errors or nonuniformity along the link. To meet this need, several component manufacturers have introduced DGEs.

Because the market for DGEs is relatively new, several technologies are vying for attention and no clear favorite has emerged. Available products are based on planar lightwave circuits, microelectromechanical systems (MEMS), liquid crystal, and acousto-optic technology. While much of the development work for DGEs focuses on the filtering function of the device, the dynamic function of the DGE must include a sensor and control logic (see Fig. 1). Often these functions are as critical as the filtering response of the DGE, and as complex to implement.

When evaluating the competing technologies, some key performance criteria must be met, including the spectral-response function. Individual technologies can provide gain equalization by modifying the gain envelope across the entire spectrum, or by grooming on a channel-by-channel basis. A continuous response function is preferred, however, because this avoids passband narrowing through multiple devices in a link, and makes the device insensitive to bit-rate and channel spacing.

Residual gain ripple or error function should also be considered. This parameter is critical to enable ultralong-reach applications because the accumulated error function, along with the polarization-dependent loss (PDL), will ultimately limit the transmission distance. Residual gain ripple below ±0.25 dB is required, and performance beyond this is welcomed.

Another important parameter to consider is PDL. While the PDL is an element of the overall loss budget, it affects the system margin differently. Increasing the amplifier gain can address a higher insertion loss, but the unpredictable nature of the PDL on the signal creates a varying power at the receiver. The receiver operates in a linear regime over a limited range, and if the PDL plus other varying effects exceeds this range, bit errors can occur. Values below 0.3 dB are needed.

Of course, low insertion loss is critical to avoid degrading the amplifier noise figure. However, adding the DGE in the mid-stage of a dual-stage amplifier can minimize impact on the noise figure and accommodate insertion losses of up to 6 dB. Finally, a dynamic range of 10 dB is adequate to compensate for the amplifier gain variation.

In addition to the optical parameters, size, power consumption, and the cost-reduction potential of the various technologies are critical. Currently, DGEs are comparable in size to amplifiers, but integrated components offer hope to reduce this in the future. Regarding cost, system designers may trade off technical specification in components if they believe one technology offers a potential for significant cost savings.

System designers have several DGE technologies to evaluate. The first class of technologies creates a dynamic spectral response by controlling the amplitude of multiple sinusoidal filters with differing periods. The ideal DGE response is the Fourier transform of the inverse gain spectrum. The number of Fourier elements or stages used in the device determines the resolution of the DGE and the residual gain ripple. The acousto-optic effect, either in a bulk crystal or directly in the fiber, is one method to create the controllable response filter. These approaches have been inefficient and require high RF power.

Liquid crystals can also be used to control dynamic loss with multiple stages with different phase retardations. Varying the polarization rotation in the liquid crystal can control the loss of the individual stages. Finally, a lattice filter can be made by cascading multiple Mach- Zehnder interferometer stages in a planar lightwave circuit (PLC) and controlling the interference of the stages using the thermo-optic effect.

A general challenge to all these approaches is the control algorithm used in the DGE. Often the calculation uses a convergence method, iterating the response multiple times to arrive at the minimal error function. This iterative approach can create instability in the control loop.

An alternative approach with a much more predictive control algorithm is based on slicing the spectrum and controlling individual elements. These approaches employ a demultiplexing element to provide the spectral slices. In combination with a diffraction grating, MEMS are one example of this technique. Each spectral slice falls on a MEMS element, which can be adjusted to control the loss. Liquid-crystal technology can also be employed after a diffractive element to create a DGE.

To maintain a continuous response, interference of the spectral slices can be used. An interesting example using this approach is to include an arrayed waveguide grating (AWG) and thermo-optic phase-shifters within one arm of a Mach-Zehnder interferometer.1 The AWG provides the spectral slicing, allowing control of individual wavelength regions. Rather than controlling the amplitude of these spectral regions with a variable optical attenuator, a phase-shifter is added to create a controlled phase delay. A second AWG is used as a multiplexer and the output then combines with the unaltered second path of the Mach-Zehnder interferometer. Interference at the Mach-Zehnder recombination point creates a dynamic range in excess of 10 dB, and the output spectrum is continuous due to oversampling within the AWG spectrum. The spectral resolution and maximum local slope is determined by the AWG design and is increased with the number of channels in the AWG.

In this planar waveguide approach, no moving parts are required (see Fig. 2), which offers improved reliability compared to some of the other technology approaches. Also, if hybrid integration is used, the device can have low power consumption and low PDL. Finally, the planar approach should offer significant cost reduction potential with increases in wafer size and volume production.

While the amplifier application of DGEs has been our main focus here, the technologies discussed also apply to modified versions of the DGE. A simpler version of a DGE exists in which a dynamic gain tilt compensator corrects only for slope errors across the gain curve rather than compensating for an arbitrary variation in the gain curve. These slope errors could be caused by amplifier gain tilt, or by the linearly varying stimulated-Raman-scattering loss. Both liquid-crystal and PLC technologies apply to this problem.

A second version of a DGE includes the capability to block individually selected channels. Clearly the approaches based on spectral slices rather than Fourier elements are more suited for this need. These products can equalize the express channels and block the dropped channels in add/drop nodes.2 The effective device must have an extinction ratio of greater than 40 dB for the blocked channels, and must be installed with the full channel count and channel density of the system.

In all the applications described above, the relatively high cost of the DGE at $10,000 to $12,000 limits their use to every four or five amplifiers. While significant design activity exists at the major system manufacturers, DGEs are not yet deployed in volumes that drive these prices down, and volumes will depend on the recovery of the long-haul DWDM market. To be the technology of choice for DGEs when the market recovers, the product must meet not only the optical performance specification but also the reliability and cost required by the system integrators.

Bob Shine is director of marketing at WaveSplitter Technologies, 46430 Fremont Blvd., Fremont CA 94538. He can be reached at bob_shine@wavesplitter.com.

  1. C. R. Doerr et al., IEEE Photon.Tech. Lett. 12, 1195 (2000).
  2. J. Bayne and M. Sharma, Lightwave, 69 (December 2001).
More in Home