Dynamically reconfigurable wdm networks remain a challenging goal
Dynamically reconfigurable wdm networks remain a challenging goal
Achieving all-optical wdm network operation may require dynamic channel-power equalization of multiple wavelengths and data-shifting among wavelengths for efficient and robust operation
alan e. willner, bogdan hoanca
Univ. of southern California
timothy day
new focus inc.
Wavelength-division multiplexing (wdm) technology, in which multiple channels are located on different wavelengths and are simultaneously transmitted down the same optical fiber, has greatly expanded transmission capacity and enabled highly flexible wavelength-routed optical networks.
wdm has been critically enabled by the development of erbium-doped fiber amplifiers, which can amplify multiple wavelength channels simultaneously. These all-optical, fiber-based amplifiers possess high gain and low additive noise, qualities that ideally suit communications systems. wdm technology also relies heavily on recently developed wavelength-dependent devices and techniques, such as integrated wavelength routers, fiber Bragg gratings, and multiple-wavelength lasers. As a practical matter, a commercially available wavelength-tunable laser enables the evaluation of wavelength-dependent limitations in wdm systems experiments.
Research in other areas could further enhance the operation of wdm networks. The dynamic channel-power equalization of many multiple-wavelength signals for robust operation, and all-optical shifting of data from one wavelength to another for dynamically reconfigurable, high-throughput networks represent two such pursuits.
Dynamic equalization
Although passive gain equalization has been used to flatten the spectrally nonuniform gain of erbium-doped fiber amplifiers, such techniques are suitable only for static point-to-point links. Dynamic channel power equalization is needed to accommodate the time variations of several system variables, including input channel power, amplifier gain, gain nonuniformity, and link losses.
In one possible dynamic channel power equalization method, a dense wavelength-division multiplexer (dwdm), such as an integrated wavelength router, and an array of polarization-independent acousto-optic modulators are used (see Fig. 1). The multiplexer first separates the incoming wdm stream into parallel paths of single-wavelength channels. Then, each acousto-optic modulator acts as an independent loss element for an individual wdm channel. Combining the modulators with another dwdm and a feedback control circuit allows the independent equalization of all channels, with the channel spacing dictated only by the multiplexer.
The dynamic equalization module reduces the power differential among, say, three wdm channels from 10 dB to 1 dB and responds to modulation changes in less than 0.1 microsecond. To simulate channel power fluctuations, two channels are intensity modulated at a 10-MHz rate (see Fig. 2a). The third channel is left unmodulated and serves as a reference channel.
The dynamic equalization module effectively smoothes out the optical power variations of the modulated channels from 10 dB to within 1dB of the reference channel (see Fig. 2b). As an additional benefit of this method, most of the interchannel amplified-spontaneous-emission (ase) noise from the optical amplifiers is attenuated because the channels are passing through the dwdm optical filter; any ase appearing between channels is not coupled back into the output. Based on this method, a fully equalized 1000-km long-distance link has been demonstrated by circulating data repeatedly through a 100-km fiber loop that incorporates the erbium-doped fiber amplifiers and the dynamic equalization module.
Another approach to dynamic equalization involves the use of an acousto-optic tunable filter, in which each wdm channel is placed within one of the multiple independent passbands whose transmission value can be tuned. This flexible approach accommodates any changing set of wavelengths for the incoming channels but does not allow channel spacings closer than 2 nm.
Wavelength-shifting
Wavelength management makes the best use of available wavelength resources and maintains high throughput in a dynamically reconfigurable wdm network. Specifically, the number of wavelength channels available for routing is usually limited by the bandwidth of optical components, the wavelength generation and stability of wavelength-dependent devices, and nonlinear transmission effects. These limitations, along with possible changing traffic patterns and link outages, may necessitate the re-use of a relatively small number of wavelengths across a large network. In robust and flexible wdm networks, wavelength-shifting may be essential for wavelength management, and would be performed all-optically to ensure minimum speed bottlenecks and maximum transparency in both data rate and data format.
In all-optical wavelength-shifting experiments, wavelength-shifting can be produced in a semiconductor optical amplifier based on cross-gain compression. To allow higher data rates and better signal shaping, other wavelength-shifting techniques using semiconductor optical amplifiers include integrated interferometric configurations and four-wave mixing.
Despite some shortcomings, cross-gain compression is considered the easiest and most robust wavelength-shifting method. For wavelength-shifting using cross-gain compression, a strong intensity-modulated pump laser is used to compress the amplifier gain only when the laser pump output is "high," or "1" (see Fig. 3). Consequently, the gain takes on the inverse time function of the modulated input pump signal. A weak continuous wave "probe" laser, operating at a second wavelength and coupled into the amplifier simultaneously with the intensity modulated pump, experiences this same gain modulation. Because the output of the probe wavelength equals the input multiplied by the gain, the output probe becomes an inverted replica of the input pump signal. In this manner, wavelength-shifting technology holds the potential for forming several key building blocks for a future packet-switched wdm network.
Subcarrier header routing
Wavelength-shifting can also be used to perform the all-optical routing of optical packets (see Fig. 4). In this method, incoming packets enter the wavelength routing node on wavelength A (lA) and are routed out of the node on a different wavelength (l1 through lN), depending on the header information. In our method, the header for each packet is subcarrier-multiplexed on a 1.5-GHz radio-frequency wave and sent along with the baseband data (operating at a 1-Gbit/sec rate).
At each wavelength routing node, the wavelength path is set according to the packet destination encoded in the header. Once the wavelength path is set, the baseband data are wavelength-shifted onto the destination wavelength, and the packet is routed to the output. Throughout the routing node, the baseband data remain in optical form, allowing transparency in both data rate and data format.
Functioning in a test wdm routing node, four input packets are cyclically wavelength-routed error-free from lA to l1, l4, l2, and l3 (see Fig. 5). Subcarrier-multiplexed headers are useful because they
are lower-speed headers that can be recovered using relatively inexpensive electronics,
alleviate the problem of resolving many wavelengths since only one optical detector is required to recover all the subcarrier headers located on different incoming wavelengths.
Wavelength-shifting can also be used to resolve output-port data-packet contention. Contention occurs when two packets arrive simultaneously in a switch at different input ports and on the same wavelength, and need to be routed to the same output port. Under normal circumstances, one packet would lose contention and either be buffered locally or be routed out of the undesired port. To avoid delays and loss of throughput, one packet can be shifted onto a free available wavelength, and then both packets can be routed to their desired output port without affecting each other.
Contention among 1-Gbit/sec input data packets arriving at port A and port B of a 2 ¥ 2 switch can be resolved in real time by comparing subcarrier-multiplexed packet control headers (see Fig. 6). The contention detection mechanism is based on encoding the packet wavelength and destination using subcarrier multiplexing. Therefore, a subcarrier frequency corresponding to a packet wavelength allows a single detector to fully recognize any one out of multiple incoming wavelengths. If contention is detected, one packet is all-optically wavelength-shifted to a free wavelength. For the packet that has been wavelength-shifted, the header itself is replaced and updated with new routing information.
To demonstrate successful contention resolution using wavelength-shifting in a wdm network (see Fig. 7), consider a time slot (0 to 640 nsec) in which two data packets are present at input port A, one on wavelength l1 and another on wavelength l2. At the same time, assume a packet enters input port B on wavelength l1. If all three packets request to be routed to output port C, there is contention between ports A and B on wavelength l1. To overcome this contention problem at output port C, the packet from port B is wavelength-shifted onto a third available wavelength, l3, thereby resolving contention without loss of throughput.
In the second time slot, 640 to 1280 nsec, because no l1 packet is present at port A, no contention occurs, and no wavelength-shifting onto l3 needs to be performed at output port C. In the third time slot, 1280 to 1920 nsec, contention is again present. In this situation, the packet from input B is again wavelength-shifted to avoid contention.
Many research groups in the wdm community have demonstrated basic building blocks and even networks, but a dynamically reconfigurable wavelength-routed packet-switched network is not yet available commercially. Dynamic power equalization, dynamic routing, and network reconfiguration are still challenging research issues for achieving future high-throughput, cost-effective wdm networks. u
Alan E. Willner is associate professor, Department of Electrical Engineering--Systems, at the University of Southern California (usc), Los Angeles, CA. Bogdan Hoanca is a graduate research assistant, Department of Electrical Engineering--Systems, at usc. Timothy Day is vice president, engineering, at New Focus Inc., Santa Clara, CA.
The authors acknowledge the contribution of the following researchers to this paper: Jin-Xing Cai and Kai-Ming Feng, graduate research assistants, Department of Electrical Engineering--Systems, at usc; William Shieh, member of technical staff, Time and Frequency Systems Research Group, Jet Propulsion Laboratory, Pasadena, CA; Eugene Park, principal member of technical staff, Telesis Technologies, San Ramon, CA; and David Norte, senior member of technical staff, Lucent Technologies, Denver, CO.