History and technology of wavelength division multiplexing
Please define wavelength division multiplexing.
Telecommunications make wide use of optical techniques where the carrier wave belongs to the classical optical domain (Figure 1). The wave modulation allows transmission of analog or digital signals up to a few GHz or Gbits/s on a carrier's very high frequency, typically 192 to 196 THz (infrared). In fact, the bit rate can be increased further by using several carrier waves that are propagating without significant interaction on the same cable.
It is evident that each frequency corresponds to a different wavelength. This technique could be named either Frequency Division Multiplexing (FDM) or Wavelength Division Multiplexing (WDM). The latter term is more often used; the first term is generally reserved for very close frequency spacings (typically less than 50 GHz corresponding to 0.4 nm). The terminology is not set in concrete.
With wavelength division multiplexing it is possible to couple sources emitting at different wavelengths λ1, λ2 ... λn into the same optical fiber. After transmission on the fiber, the λ1, λ2 ... λn signals can be separated toward different detectors at the fiber extremity (Figure 2). The component at the entrance must inject the signals coming from the different sources into the fiber with minimum losses-this is the multiplexer. The component separating the wavelengths is the demultiplexer. The multiplexer may be replaced by a simple optical coupler, but losses will increase.
Obviously, when the light propagation is reversed, the multiplexer becomes the demultiplexer, and conversely. It is important to note, however, that the coupling efficiency is not necessarily preserved in reverse operation. For example, if the multiplexer uses single-mode entrance fibers and a multimode output fiber, the coupling losses would be excessive in the reversed use.
Multiplexers designed with identical input and output fibers are usually reversible. Simultaneous multiplexing of input channels and demultiplexing of output channels can be performed by the same component: the multi/demultiplexer.
What are these multiplexers?
For the multiplexing (or separation) of wavelengths, interference filters or gratings can be used. However, wavelength division multiplexers using interference filters cannot be used when the number of channels is too high or when the wavelengths are too close. The main advantage of the grating is the simultaneous diffraction of all wavelengths and so it is possible to construct simple devices with a very large number of channels (with the exception of fiber gratings).
There are three types of gratings: classical, arrayed waveguide, or fiber.
In the classical grating, we use gratings in the Stimax configuration (Figure 3). The dispersive element was a grating embedded in a monoblock of silica, and the optical fibers were directly fixed to the block. The number of grooves on the grating (several tens to several thousands per millimeter) are obtained by using a diamond tool or by holographic photoetching. The grating has the property of diffracting light in a direction related to its wavelength (Figure 4). Hence an incident beam with several wavelengths is angularly separated in different directions. Conversely, several wavelengths λ1, λ2 ... λn coming from different directions can be combined in the same direction. The diffraction angle depends on the groove spacing and on the incidence angle.
The arrayed waveguide grating was designed to increase the resolving power, i.e., the fine splitting of the wavelengths. It was proposed around 1990 by Takahashi and others in Japan, and Dragone and others in the U.S. They increased the optical path difference between the diffracting elements by using a waveguide structure equivalent to the well-known Michelson echelon gratings in classical optics. The advantage is a smaller channel spacing. The disadvantages are a much smaller free spectral range that will limit the total number of channels and near-end crosstalk that affects bidirectionality.
A fiber grating is made by recording a Bragg grating in the core of single-mode fiber made photosensitive by doping with, for example, germanium. This grating can be used as an narrowband filter. It is necessary to use one grating per wavelength. So there is some limitation to the number of channels that can be obtained with these devices.
Please tell me about the history of WDMs and your part in it.
The optical multiplexing concept is not new. To our knowledge, it dates back at least to 1958, to an IEEE paper by R. T. Denton and T. S. Kinsel. About 20 years later, the first practical components for multiplexing were proposed from different laboratories, mainly in the U.S., Japan, and Europe.
We started our research on gratings in 1965. We took part in the engineering of ruling engines and to the introduction and development of holographic gratings in 1967 at Jobin Yvon. In those days, the main application was spectroscopy, and we had no idea of the possible applications to optical telecommunications. We realized that possibility only after having participated in the 1973 Summer School in Electromagnetism in Centre National d'tudes des Telecommunications in Lannion (France) devoted to guided waves in optical communication. Soon after, we developed a new grating optics coupler in 1974, but this component was only used for optical spectroscopy research and did not find application in optical telecommunications. In those days, optical telecommunications had more urgent problems to be solved. In 1980, when the technology began to mature, we introduced the Stimax configuration.
Since that time, we've refined the configuration and grating to achieve lower losses, lower polarization effect, smaller near-end crosstalk allowing full bidirectionality with many channels, routing capability, and more channels (the feasibility of devices with several hundred channels is now being demonstrated), smaller spacing (down 0.4 nm now).
In order to get more channels at smaller wavelength spacing, you need higher dispersion gratings, which means more grooves per millimeter. Unfortunately, this generally means gratings with more crosstalk and a larger polarization rate.
Is the crosstalk due to a diffraction or polarization effect?
No. Crosstalk is due to poorer quality in the positioning of the grooves, which is related to groove spacing. These are defects in the multiplexers. The crosstalk, or parasitic light, also comes from such problems as dispersion, channel spacing, wavelength conversion along the transmission fiber by four wave mixing, Brillouin, Raman, or other nonlinear effects. The theoretical minimum channel spacing, in the end, is related to "uncertainty" relationships (derived from the Heisenberg Uncertainty Relationship).
However, most of these problems can be avoided or reduced on practical networks. For example, on the so-called IBM Muxmaster, which transmits data on long links with 20 wavelengths (10 in each direction) at up to 2 Gbits/s each, the crosstalk of the WDM components (Stimax from Jobin Yvon) is almost negligible.
As for polarization, there is polarization from gratings. It can be shown in electromagnetic theory that the diffraction efficiency depends on the polarization of the incident beam. For a given wavelength, this effect increases when the groove spacing decreases. Typically this effect is small when the groove spacing is 10 times larger than the wavelength, but it becomes important when the groove spacing is reduced to a few wavelengths for higher angular dispersions. In general, for groove spacings less than one wavelength (the absolute limit being half a wavelength in the material), the grating almost totally polarizes the light.
Moreover, on integrated optic WDMs not only does the diffraction efficiency depend on the polarization, but also the position of the channel wavelengths may depend on polarization. Fortunately, such a problem does not exist on 3D devices.
Does polarization adversely affect WDM?
The polarization of the light transmitted along the fiber is not stable. If the transmission through the WDM component is sensitive to the polarization, we get unwanted signal variations.
In silica glass fibers, the best transmission wavelengths are 1.55 m (minimum attenuation) and 1.312 µm (minimum dispersion). How wide is the wavelength window (δλ about 1.55 m and 1.312 µm) to send multiple signals down such fibers? How many channels can you send with what channel width (signal full-width-half-maximum) and channel spacing through the 1.55-µm and 1.312-µm windows?
There are indeed about 15,000 GHz of optical frequency bandwidth in each 1300-nm and 1550-nm window (Figure 5). With 5-Gbits/s bit rate, the uncertainty relationship gives about 5-GHz limit for optical frequency spacing (or about 100-nm optical wavelength bandwidth). This would mean 3000 channels! Each channel can be separated by 0.03 nm. In reality, the separation between channels is now 1 nm or a few nanometers on most of the installed networks using WDM, with each wavelength signal having a full-width-half maximum smaller than 0.03 nm.
An International Telecommunication Union standardization is being discussed that would propose a frequency grid with separations of 100 GHz (about 0.8 nm) with multiples and submultiples. Now, the published minimum channel spacing is about 0.1 nm. However, nothing lower than 0.4 nm (50 GHz) spacing is commercially available. These spacings are obtained on WDM using "classical" gratings (Stimax from Jobin Yvon, for example), arrayed waveguide gratings (Lucent, NEC, for example), or Bragg grating filters (3M).
We made a WDM router with 25 input ports to 26 output ports at 0.8-nm spacing (similar to Figure 6) and demonstrated feasibility of routers with at least 157 input ports to 157 output ports at 0.05-nm spacing. (However, these 0.05-nm spacing components will remain useless until sources can be made for them.)
Is WDM easier for single mode rather than multimode fibers?
No, not necessarily, although single-mode fibers will give smaller devices for high density grating WDM. Is it easier to multiplex the signal in the electronic domain-in time (TDM), or in the optical domain-FDM or WDM? The answer is not easy and the optimum solution is generally found in the association of different techniques. For low bit rate services (<2 Mbits/s), it is generally better to use TDM techniques only. For uncompressed, high-definition television broadcasting (about 2 Gbits/s for studio quality), wavelength division multiplexing is highly recommended.
Applications such as video networks linking workstations, television studio center signal routing systems, videoconfer-ence networks, interactive video training systems, bank information service networks and data transfer networks between computers, integrated service digital networks (ISDN), tele-distribution, and generally all broadband networks increasingly use both time and wavelength multiplexed optical lines.
What about the future?
We already manufacture 131-channel WDMs. We think that many more channels-and why not 3000-channel WDM components?-are feasible. Three-thousand-channel classical grating spectrometers are selling readily, so why not WDMs? Now the main problem is to have enough stable fixed or tunable sources. Nowadays, the practical limit is a few tens of sources spaced at 50 GHz. The number of channels will depend on the progress with the sources.
It is now possible to find sources stable enough with up to 32 wavelengths or more at 100- or 50-GHz spacing. We need better definition of wavelengths, smaller spacings, and better stability for larger numbers of channels.
J.P. Laude is currently scientific director of ISA/Jobin Yvon. He also teaches short courses at the University of Paris (UP), the Ecole Superieure d'Optique (ESO) and the Institut National des Telecommunications in optics and spectroscopy. He received his Engineering Degree from ESO in 1963 and his Doctorate from UP in 1966. He received the 1997 SPIE Technology Award for his work on gratings and WDMs.