Tunable filters provide flexibility for agile networks

Philip Duggan, Maurice Tremblay, Nisan Rowhani, and Manjeet S. Ner

Current optical networks are mostly "fixed" with a limited degree of flexibility. Service providers need agile networks that allow wavelengths or bands to be manipulated across nodes to create profit-generating services. Tunable filters are key enablers for agile networks, and are used primarily for tunable add/drop multiplexers, amplified spontaneous-emission suppression, and wavelength monitoring.

Tunable-filter technologies and their applications include Fabry-Perot (FP) etalons, temperature- or mechanically tuned fiber Bragg gratings (FBGs), diffraction gratings (DGs) (waveguide or bulk), tilt-interference filters (TIFs), and linear variable filters (LVFs).

Fabry-Perot interferometers or etalons are simple optical filters made up of two parallel reflective surfaces that produce a comb spectral filter in which the transmission peaks are separated equally in frequency. The parameters that define a Fabry-Perot interferometer are free spectral range (FSR) and finesse. FSR is the wavelength spacing between peaks. Finesse is defined as the ratio of the FSR to the 3-dB bandwidth (BW) of an individual peak. It is a measure of the resolving power of the Fabry-Perot interferometer

Increasing the reflectivity of the mirrors achieves greater contrast because more light is allowed to add up constructively (see Fig. 1).

The transmission spectrum of a Fabry-Perot filter is the Airy function, given by

Tuning can be achieved by adjusting the mirror spacing. To tune one FSR, the plate spacing must be adjusted by l/2 (0.8 nm at 1600 nm). The relative shift in wavelength is equal to the relative change in mirror spacing. For example, if the spacing is 30 µm (40-nm FSR at 1550 nm) then a 1-nm change in mirror spacing is equivalent to a 50-pm shift in wavelength. Accurate control over the relative position of the mirrors is essential for this technology.

Many of the tuning mechanisms for FP filters are nonlatching and lack extensive reliability characterization. For example, piezoelectric transducers, although suitable for this application, degrade over time when continuous voltage is applied. Another more promising tuning mechanism for FP filters is electrostatically driven microelectromechanical systems (MEMS), which offers low cost and a small footprint.

Fabry-Perot filters can also be tuned by changing the index of refraction of the cavity or spacer material. A suitable candidate, liquid crystals, can exhibit large index changes for small applied voltages. In practice, material loss and uniformity has limited the finesse of liquid-crystal FP filters to below 100.

To achieve finesse of greater than 1000, the two mirror surfaces must be parallel to within approximately one arcsecond. Deviations result in additional loss and filter broadening. Therefore, an extremely stable mechanical platform is required. Active parallelism adjustments have also been successfully demonstrated.

Thin-film technology can be used to produce low-loss narrowband filters called tilt interference filters. The filter structure can be modified to produce various band shapes from flat-topped to Gaussian.

Wavelength tuning is achieved by varying the angle of incidence, q, on the filter (see Fig. 2). The tuning equation is given by

where λ_θ is the center wavelength at normal incidence, and neff is the filter effective index of refraction. Note that the wavelength shift is quadratic for small angles, with the largest shifts occurring at highest angles of incidence. For example, to achieve a 40-nm wavelength shift, a 20° tilt is required.

One drawback of the TIF approach is that at higher tilt angles, polarization-dependent loss (PDL) can increase substantially, particularly near the passband edges. Additionally, an increased amount of loss ocurrs for high angles of incidence due to walk-off effects.

For example, a typical 50-GHz bandpass filter has approximately 20 ps of group delay at the center. This means the filter has an effective thickness of 4 mm (for neff = 1.5), which will cause the incident beam to walk off by approximately 1 mm at an angle of incidence of 20°. To minimize these effects, the input beam should be made significantly larger than the walk-off. In this instance an input beam diameter of at least 5 mm is required. The consequence is longer focal-length optics, a larger filter, and consequently, a larger footprint.

Linear variable filters (LVF) are thin-film filters that are grown with a slight taper in the film thickness along its length. Special masking techniques, which partially block the deposited film, are used when growing these filters to produce a linear slope.

The input signal is focused onto the filter surface for wavelength selection. The translation of the filter across the beam creates wavelength tuning. A motor (DC or stepper) and filter translation mechanism can achieve this mechanically. A stepper motor increases reliability as it is inherently latching. An encoder helps determine the filter position within 1 µm to calibrate the filter to within 1 GHz. Since the input beam is focused, the filter can be relatively short, leading to a compact device.

The LVF filter shape is broadened as a result of the finite extent of the beam along the linearly varying filter direction. This tends to produce spectral broadening and rounding of the intrinsic thin-film-filter design. For this reason, the beam must be properly focused onto the filter.

The focusing limitations are governed by the filter properties and Gaussian beam optics. The Rayleigh range of a Gaussian beam (the length over which a beam remains quasi-collimated) is given by

where n is the filter effective index and ω₀is the beam radius at the focus.

For the multiple reflections within an LVF film to interfere properly, the Rayleigh range of the focused input beam should be larger than the effective filter thickness. Although generally not preferred, the advantage of the "rounding" effect is reduced chromatic dispersion.

Thin-film technology such as the type used in tilt filters or LVFs are somewhat less amenable to three- and four-port operation. The reflected notch from these filters typically only provides 15-dB rejection of the selected channel. Multiple reflections of the filter can be used to increase this rejection but the free-space optics involved are complex.

Fiber Bragg gratings are made by exposing a fiber to an intense UV optical-interference pattern that causes the index of refraction to vary periodically along the length of the fiber core. This periodic variation in index acts as hundreds of weakly reflective mirrors. These reflections add in phase at the Bragg wavelength, λ_Β, which is determined by the fiber core index, ncore, and the Bragg period, Λ, as

Thus the FBG is a reflective notch filter.

The fiber-Bragg-grating approach, depending on the application, requires one or two circulators, increasing both losses and costs. FBGs allow for extremely narrow filters to be grown with low loss and a flat-top passband. As mentioned previously, the square passband tends to produce relatively high dispersion and may not be suitable for 40-Gbit/s systems. Tuning is achieved by mechanically stressing the grating, and a greater range is available in compression than in tension.

Stressing the fiber in any way has reliability concerns, particularly at the end attachment points. Another distinct disadvantage is the limit in tuning range possible, which in practice is 25 nm. Consequently, the associated costs and size can double in applications where tunability is required for the entire C- or L-Band.

Planar waveguides are an attractive technology for tunable-filter applications because of their compact size and robustness. Typical materials may include silicon/silicon dioxide (Si/SiO₂) and indium phosphide (InP), with tuning achieved thermally by changing the material indices.

In practice, the mode coupling used between waveguides and fibers have shown high losses, particularly for high-index waveguides, which have the smallest mode fields. As TE and TM modes propagate separately with differing effective indices, PDL is also a major issue. In theory, control of the waveguide aspect ratio would reduce PDL. However, PDL in excess of 0.5 dB is common. Waveguide Mach-Zehnder interferometers use a 3-dB splitter and an optical-path-length difference, or delay, to create interference. To provide a narrow-bandwidth filter, several MZ elements must be cascaded together, adding complexity, as all elements must be tuned in tandem.

A conventional bulk-ruled diffraction grating can also be used for tunable filter applications. A diffraction grating will take a collimated input beam and angularly resolve it into its spectral components. To achieve high spectral resolution, a well-collimated input to the DG is required. Additionally, the line density of the grating should be high, leading to longer focal-length optics and larger physical size.

Tuning the DG can be achieved by either tilting the grating or tilting a secondary mirror.

Diffraction gratings offer excellent spectral resolution and a Gaussian filter response. Disadvantages are high insertion loss, high PDL, and large footprint. Gratings are best suited for selecting multiple arbitrary wavelengths simultaneously, where the additional loss, cost, and complexity are more easily justified.

These various filter technologies offer different transmission shapes (see Fig. 3). The Gaussian filter shape corresponds approximately to the response of a DG or multistage MZ. This shape corresponds to the lowest chromatic dispersion.

Because FP filters have narrow passbands, they are less suitable for channel-selection applications. Cascading two Fabry-Perot interferometers can result in higher contrast and a slightly flatter top, but it adds to loss and design complexity.

No single technology fulfills all the performance and cost criteria required for the new architecture-agile networks. The linear variable filter, however, has a marginal edge over other technologies. The most appropriate technology for a specific application can be chosen by carefully considering the pros and cons of each type (see table).

Philip Duggan is senior technical staff member, Maurice Tremblay is product line manager, Nisan Rowhani is application engineer, and Manjeet S. Ner is product line manager at JDS Uniphase, Attenuators and Tunables Optical Layer Group, 3000 Merivale Rd., Nepean, Ontario, K2G 6N7 Canada. Philip Duggan can be reached at [email protected].