Until recently, the development of dense wavelength division multiplexing (DWDM) technology has focused solely on scaling to higher channel counts, following pundits’ predictions that future high-bandwidth applications would create a bandwidth shortage. This shortage has proved largely illusory so far, and DWDM development has had to adapt to meet additional demands. One example is reconfigurable networks, which provide flexibility and fast bandwidth, enabling carriers to realize a return on their investment in infrastructure. The new and significant demands that reconfigurable networks place on DWDM components have led to a variety of new multiplexing and demultiplexing technologies. Current DWDM options include thin-film filters, fiber Bragg gratings, arrayed waveguide gratings and hybrid devices based on free-space optics and diffraction gratings. The application of these technologies is governed by a number of factors, including price (both initial cost and price per channel), performance, footprint, channel counts, scalability, power consumption and manufacturability. Like arrayed waveguide gratings, free-space diffraction gratings multiplex light through induced phase shift and interference. However, their passive operation frees them from the need for heaters and enables much smaller components. High-channel-count applications Major network hubs multiplex and demultiplex many channels through single optical fibers — typically 40 channels for metropolitan and 80 or more for long-haul hubs. This requires technology with high channel counts and good uniformity. Traditionally, thin-film filters coupled by circulators (or more recently, interleavers) provide one approach to achieving high channel counts. The efficiency of these devices does not rely on thermal compensation, and their bandpass filter function is flat, which enables the layering of many filters in a system. However, they also present several potential problems. First, although single thin-film filters have good optical properties, when cascaded in series with other thin-film filters, circulators and interleavers, their cumulative losses can impair overall DWDM system performance at high channel counts. And although prices have dropped substantially, the economics of cascaded thin-film filters are poor because interleavers, circulators and other coupling technologies must be included in the overall system cost. DWDM systems based on cascaded thin-film filters are also bulky, often requiring 19-in. racks. Alternative high-channel-count DWDM filters use parallel processing, provided either by free-space diffraction gratings or by arrayed waveguide gratings. These technologies separate all channels in a single step rather than through a sequential series of filtering steps. Parallel processing of light, provided either by free-space diffraction gratings or by arrayed waveguide gratings, separates all channels in a single step rather than through a sequential series of filtering steps. The advantages are numerous. First, these technologies allow high-channel counts without requiring additional coupling components, which results in a much smaller footprint, lower cost, higher reliability and increased optical performance. Arrayed waveguide gratings, a relatively new technology, multiplex light sources by causing the differential phase shifts of “beamlets” to interfere. These devices have made great advances in optical performance over the past few years, particularly with regard to insertion loss. But in general, they are outperformed by free-space diffraction grating technology. Free-space diffraction gratings also separate light in a parallel fashion and operate using the same physical principles as arrayed waveguide gratings: induced phase shift and interference. But free-space diffraction gratings have the added advantage of being passive. That is, unlike arrayed waveguide gratings, they do not require heaters to perform under all environmental conditions. New athermal designs for arrayed waveguide gratings have been tested, but their optical performance remains poor compared with heated versions. Another advantage of free-space diffraction gratings is their improved channel isolation, which results from a more accurate phase shift imparted by the grating. The technology also offers better channel accuracy, lower polarization-dependent loss and lower chromatic and polarization mode dispersion. This last benefit stems from the fact that the free space within the device incurs no birefringence. Finally, free-space diffraction grating designs are often “single-ended,” in that fibers enter and exit on the same side of the device. This can enable efficient board layouts because it must accommodate only one fiber bend radius. One issue common to all dispersive technologies — including both arrayed waveguide gratings and free-space diffraction gratings — is that they naturally produce a Gaussian passband shape. When coupled with other devices having a Gaussian passband, the effective filter function narrows to the point where the system is incapable of high-data-rate transmission. Flattop versions of both technologies have appeared that achieve a desirable filter function with an approximately 2- to 3-dB penalty in insertion loss. Such devices outperform the corresponding optical performance parameters for high-channel-count subsystems based on thin-film-filters. Low-channel-count applications Another DWDM application involves add/drop sites or nodes. Current network architectures use fixed add/drop functionality, in which a particular node drops the same channels with no possibility of reconfiguration without physical intervention on site. Nominally, the number of channels dropped at a node varies from one to eight, but typical sites drop either four or eight channels. Whatever the channel count, the dropped wavelengths require demultiplexing from the network and a multiplexer to add channels on the main fiber. Thus, there is a significant need for low-channel-count devices. Here the advantage of free-space diffraction gratings over thin-film filters and fiber Bragg gratings is less clear-cut. Low-channel-count applications dilute the ability of free-space diffraction gratings (and arrayed waveguide gratings) to amortize the fixed cost of the optics and packaging over a large number of fibers. Meanwhile, thin-film filters can be made relatively inexpensively and can perform well in applications with, say, fewer than 16 channels. However, at least two emerging grating technologies have the potential to address this issue. Dual-input free-space diffraction gratings — as demonstrated by Zolo Technologies — share fixed costs of packaging and optics over two and potentially more multiplexer/demultiplexers housed in the same package. This reduces the cost per channel up to a factor of two compared with separately housed multiplex assemblies. It also reduces the footprint by a factor of two. Thus, a dual-input eight-channel device has approximately the same performance and cost per channel as a single 16-channel device. Another free-space diffraction grating technology that could improve per-channel costs for low-channel count applications was developed at the National Research Council in Ottawa. It is a hybrid technology using a curved, etched echelle grating that couples light into and out of the device via planar waveguides. It has two potential advantages for low-channel-count applications. First, the device has only two optical components — the grating and the waveguide coupler — making it likely that the technology will be inexpensive even at low channel counts. Second, because the device is very small, its packaging costs may be reduced. However, issues remain. Although the yield on the etched gratings has reportedly improved over the last few years, very high yields will be critical to maintaining low product costs. Also, these devices are not passive; they require temperature control for proper operation, which increases cost and the potential for failure. Functional integration Future networks will utilize a different, more flexible architecture, particularly at the add/drop sites. Network architects covet the ability to provision bandwidth according to demand, and this could fuel development of reconfigurable add/drop multi- and demultiplexing at the network nodes. Such technology would enable the network operator, with a software command, to increase or decrease the number of channels dropped and added at any site at any time. Although some companies now offer reconfigurable add/drop multiplexers, the devices typically comprise individual components. Consequently, they are expensive, bulky and prone to failure. A great debate brews. What is the best means of achieving higher-level, functionally integrated devices such as reconfigurable add/drops and dynamic channel equalizers? A few arrayed waveguide grating companies offer package-level-integrated channel equalizers, but such devices have not yet integrated the waveguide grating and attenuation function monolithically onto the same silicon substrate. In fact, it could be argued that true integration will not be possible or even desirable for many years. There are two problems with true monolithic integration using planar waveguide circuits. First, except for one or two companies, arrayed waveguide grating yields are very low. A reconfigurable add/drop multiplexer requires two (perhaps 40-channel arrayed waveguide gratings) on the same substrate with a 2 x 2 switching technology displaced between the two waveguides for every channel. The yield for such devices is not likely to hit double digits anytime soon. Second, these devices would include active components on a monolithic substrate (which also must be heated) with two low-yield passive devices. So what happens when one of the active components fails? The entire device must be scrapped. When these items cost a few hundred dollars apiece (i.e., not per channel), scrapping failed devices may be an option. But it is clearly not an option now. Alternatively, functionally integrated free-space diffraction grating devices rely on a paradigm that Zolo has already demonstrated: reusing the same grating and optics to perform two separate multiplex and demultiplex operations. Unlike arrayed-waveguide-grating-based reconfigurable add/drops, free-space diffraction devices would require only a single grating. The switching technology can be a simple one-dimensional array of microelectromechanical systems mirrors. The array need have only as many mirrors as there are channels on the network. This type of array can be made relatively inexpensively and with good yields. Perhaps the greatest benefit is that free-space integrated devices enable the ability to switch out the failed active component during the manufacturing process, if not in the field. Meet the author Andrew Sappey is chief technology officer at Zolo Technologies in Louisville, Colo.