Spectral equalization keeps optical signals in line

Allan Ashmead

Wavelength-division multiplexing (WDM) increases the transmission data rate through single-mode fibers by simultaneously propagating the light from spectrally different but equally powered laser-diode sources through the fiber. Each diode constitutes a channel, with 40 channels now common in many systems and channel spacing on the order of 50 GHz (0.4 nm). The diode bandwidth is typically an order of magnitude less than the channel spacing.

In WDM networks, channel powers must remain adequate and approximately the same as the light propagates from the transmitter to the receiver. In other words, the system ideally should have no spectral ripple. A system with no spectral ripple has a flat power-spectral density across its operating bandwidth. When channel powers are different, the system`s power-spectral density is not flat because the power in each channel, which constitutes a smaller bandwidth of light surrounding a central wavelength, is unequal. The wavelength management process of correcting these inequalities is called spectral equalization.

If the system`s spectrum is not managed, then inadequate power levels can decrease the system`s signal-to-noise ratio. Furthermore, other effects such as the amplification of amplified spontaneous emission (ASE) can increase the bit-error rate (BER) in a particular channel.

Spectral equalization is a significant wavelength-management task because a number of factors conspire to produce wavelength-dependent losses and gains throughout the system. Some are intentional, such as adding and dropping channels with an optical add/drop multiplexer (OADM). Others, such as absorption, scattering, and other nonlinear fiber effects, depend upon path length and the wavelength or dispersive material properties of the fiber.

Signal amplification

The losses in channel power necessitate periodic power amplification of the individual WDM signals. Amplifiers are placed at junction points to boost the attenuated optical signal. There are several electrical and optical methods for amplifying WDM signals. Regeneration is an electronic amplification technique in which WDM optical signal is converted to an electrical signal, amplified, and then converted back into an optical signal. Regenerators require an optical receiver, an electronic amplifier, and another optical transmitter. Furthermore, these regenerators are monochromatic devices requiring a different set of components for each wavelength or channel-making the system complex and expensive to maintain. Optical amplifiers are now used instead to amplify WDM signals and enable WDM technology from a cost and technical perspective, especially in the migration to all-optical networks (AONs).

Optical amplifiers are broadband-amplification devices that do not require converting light to an electrical signal for amplification. There are several types of optical amplifiers, including semiconductor optical amplifiers (SOAs), nonlinear amplifiers such as Raman amplifiers, and-the most common-erbium-doped fiber amplifiers (EDFAs). All of these amplifiers exploit some type of quantum mechanical effect in a material to create gain in the material and subsequent optical amplification of the input signal.

SOAs create a gain medium through population inversion by electrically pumping a semiconductor material such as indium gallium arsenide phosphide (InGaAsP). The weak-input WDM signals optically seed the gain medium and are amplified through stimulated emission. SOAs are small and can be easily integrated into fiber systems. Although SOAs have a large bandwidth, they are polarization-sensitive and, for the most part, require polarizing-preserving fibers, like regular semiconductor-laser diodes. SOAs also have a low signal-to-noise ratio and suffer from significant channel cross-talk.

Raman amplifiers use stimulated Raman scattering to amplify optical signals. Raman amplifiers use a lower-wavelength pump laser to excite the atoms in nondoped fibers to higher energy states. The weak-input WDM signals exploit the nonlinear behavior of the fiber medium by stimulating atoms in higher energy states to emit photons at longer wavelengths commensurate with the WDM signals. The stimulated light mixes with the WDM signal, resulting in optical amplification.

Similar to SOAs, Raman amplifiers operate over a wide bandwidth-wide enough to cover most of the useful spectrums of the S, C, and L bands. Unfortunately, Raman amplifiers require very long fibers and high-powered pump lasers. Although SOAs and Raman amplifier technologies are continually improving and show promise for use in certain applications, EDFAs are currently the most promising and generally used optical-amplifier technology.

Erbium-doped fiber amplifiers also work off of the principle of stimulated emission. The erbium ions in the doped fiber are excited to higher energy levels by optically pumping the medium. Although the erbium ions can be excited by many different wavelengths, the two most important wavelengths are 980 and 1480 nm because atomic relaxation from these states emits in the critical 1550-nm range. Light at 980 nm excites erbium ions to a higher energy state than light at 1480 nm.

These excited erbium ions relax to a lower metastable state through acoustical photon or phonon emission after several microseconds. Erbium ions in this state are then stimulated by WDM signals to emit photons, which mix with the signals, resulting in amplification. Light at 1480 nm excites erbium ions to the same metastable state as the 980-nm light. So both 980- and 1480-nm pumps are used in EDFAs and historically were chosen on the basis of economics, power, and noise, in that order (see Fig. 1).

Unfortunately, not all of the metastable states created by 980- and 1480-nm pumping are stimulated, and after a spontaneous lifetime on the order of milliseconds they eventually relax to a lower energy state, releasing photons in the process. This occurs even if there is no WDM signal. The spontaneously emitted photons can be further amplified, leading to ASE, which results in reduced signal-to-noise ratio and increased BER.

EDFA nonlinearity

In addition to the problem of ASE, EDFAs create spectral ripple primarily through the nonlinear behavior of their gain. As a result, the input channels or wavelengths are not uniformly amplified by the gain medium (see Fig. 2).

There are several phenomena contributing to the absence of flatness in the EDFA`s gain curve. First, the gain shape is a function of the pump power, which is a phenomenon called dynamic gain tilt. In other words, the gain profile changes with the total amount of pump power. Left unmanaged, dynamic gain tilt requires operating EDFAs at one gain level. This is an important issue, especially in OADMs where power is distributed per wavelength or channel. When an OADM drops a channel or a number of channels, this directly influences the gain distribution of subsequent EDFAs. As a result, the gain changes between OADMs.

An EDFA`s gain is also saturable. Gain saturation is a nonlinear behavior where the gain or amplification coefficient is not constant but is dependent on the power of the input field. If the gain coefficient-the amount of amplification per unit length-remains constant, then the amount of amplified power is proportional to the input power. But when the gain begins to saturate at increasingly higher-input powers, the gain coefficient nonlinearly decreases until stimulated emission depletes the excessive population inversion created by the pump laser. The amplifier produces its maximum output when the gain completely saturates; any further amplification becomes independent of the input WDM signal power.

Time-dependent relaxation oscillations and inhomogeneous broadening effects also contribute to nonlinear gain behavior. An inhomogeneously broadened gain medium is one created from different sets of ions, atoms, or molecules, each influencing the shape of the gain curve. These differing influences create an asymmetric gain curve. Temperature and spectral hole-burning also contribute to inhomogeneous broadening. Spectral hole-burning occurs when the lineshape function of a WDM signal saturates the gain around the center wavelength of a particular channel, creating a hole in the gain profile. This gain is now unavailable or reduced for amplification by other closely spaced channels.

Dynamic gain equalization

The migration of WDM systems to AONs will require an increase in the number of optical amplifiers and fiber lengths-exacerbating the above effects. The wavelength management solution is to implement spectral equalization through some form of dynamic gain-equalization. Currently there are two basic approaches: individual-channel equalization and Fourier filtering.

Individual-channel equalization separates or demultiplexes each channel. For example, individual channel powers are adjusted with liquid-crystal devices, and then recombined or multiplexed. However, there are a number of issues that must be addressed with this type of dynamic gain-equalization. Insertion loss, polarization-dependent loss (PDL), and dispersion are important technical issues, but cost remains the overriding issue with demultiplexing and multiplexing technologies. A cost of $500 to $1000 per channel makes the use of channel separation and recombination prohibitively expensive for large channel counts.

Dynamic gain-equalization through Fourier filtering requires splitting the gain curve into five to ten individual windows of 3 to 6 nm in width. The individual windows of the gain curve are then dynamically adjusted, independent of the number of channels. A number of groups from IBM, Lucent, and Nortel have demonstrated less than 0.5-dB gain flatness over the C-band, using as few as five windows.

There are a limited number of technologies that enable Fourier filtering, including Mach-Zehnder-based thermo-optic (TO) devices, acousto-optic tunable filters (AOTFs), and electronically switchable Bragg gratings (ESBGs). The Mach-Zehnder thermo-optic filter is essentially a temperature-controlled Mach-Zehnder waveguide interferometer. The amplitude from two different channels or wavelengths entering a directional coupler is equally split into two different paths, each having a separate path length. The optical path length is further controlled by changing the temperature of the refractive material in that path. The beams are then combined at a second direction coupler with two different outputs. Each output supports only one of the wavelengths under certain constructive phase conditions, and different wavelengths can be tuned through optical path differences by changing the temperature of the refractive material. Thermo-optic devices dissipate considerable amounts of heat into the substrate and are inherently slow.

AOTFs exploit the Bragg condition produced when acoustical waves are created in a refractive material in the direction of light propagation. The acoustical waves themselves are created through driving the material with a radio frequency (RF) signal. The acoustical waves set up areas of compression and expansion of the refractive index in the material, which emulates a periodic Bragg structure. The tunable filter passes only those wavelengths that fulfill the Bragg condition. AOTFs require high-power RF signals and carry a significant noise and cost overhead.

ESBGs also work by exploiting the Bragg condition (see Fig. 3). However, they can be created with holographic polymer-dispersed liquid-crystal technology (HPDLC), which embeds phase-volume holograms in polymer substrates through a process that provides direct control of the diffractive bandwidth and central wavelength. The grating is erased through the application of an electric field, which gives dynamic control of the coupling between the fiber core and cladding, while at the same time minimizing insertion and PDL losses. The technology is very fast, with switching speeds on the order of 100 µs, allowing a large number of gratings to be cascaded before there is any impact on the system response time.

The ESBG devices are fabricated by laying down a mixture of monomer and liquid crystal (LC) in a freestanding cell or waveguide. The material is then polymerized at the interference plane of two coherent laser beams, which causes phase separation of the LC into two regions of differing refractive index-one of pure polymer and one a mixture of LC and polymer-in essence, creating a phase-volume grating or hologram.

The electrodes are deposited onto a cover slip or the cell walls. Light incident on this periodic structure is diffracted out of the zero order when no voltage is applied to the grating. An applied AC voltage orients the optical axis of the LC molecules within the droplets to produce an effective refractive index that matches the polymer refractive index, creating a transparent cell.

ESBGs provide high-speed (~50 µs) response times, have low insertion (less than 1 to 2 dB) and PDL (less than 0.5 dB) losses, and reduce costs in the network by integrating the demux and mux functions within the device. None of the technologies have been deployed, and the challenge for developers of both ESBG and AOTF devices is to demonstrate clear technical and commercial advantages with respect to thermo-optics.

FIGURE 3. The evanescently coupled ESBG contains an HPDLC grating layer in which polymerization produces alternating planes of solid polymer and LC microdroplets with different refractive indexes. The index mismatch results in grating-selected wavelengths being diverted from the core. Applied voltage orients the optical axis of the LC molecules to produce a refractive index that matches the polymer refractive index, thereby erasing the grating.