Abstract:

With the advances of dense wavelength division multiplexing (DWDM) technology, the number of wavelengths in a wavelength division multiplexed (WDM) network increases to hundreds or more per fiber, and each wavelength operates at 10Gbps or higher. While raw bandwidth has increased by more than four orders of magnitude over the last decade or so, capacity of switches has only been up by a factor of ten. Switching speed is the bottleneck at the core of the optical network infrastructure. Consequently, a challenge is to design cost-effective WDM switches that can scale in size beyond a hundred of inputs and outputs, and at the same time, switch fast.

Typically, there are two request models widely considered. In one model, a connection request asks to go from a wavelength on an input fiber of the WDM switch to a particular wavelength on an output fiber. In the other, a connection only needs to get to a particular output fiber, irrespective of what wavelength it will be on.

Complexity . Since it's difficult to compare the cost between different constructions, we need to somehow quantify the networks. There are several ways to do that. First one will be counting the number of components in the networks, especially the wavelength converters. As wavelength conversion is unavoidable when the source and destination involve different wavelengths, also full wavelength converters are still not practical, limited wavelength converters (LWC) are preferred when performing wavelength conversion. Hence our objective is to minimize the total number of LWCs without affecting the nonblockingness of the networks. Secondly we model the network as a graph and study the complexity of these graphs, which is defined as the number of edges in the graph. This gives a good estimate of the cost of the networks. Here we extend some known results from traditional circuit switching networks to the WDM networks.

Construction . We will give both theoretical and practical constructions. Theoretical constructions will realize some upper bounds on the result from our analysis of the complexity and many time can lead to good practical constructions. We will also give novel constructions of strictly nonblocking and rearrangeably nonblocking WDM switches for both request models in unicast using limited range wavelength converters and Arrayed Waveguide Grating Routers. We will also give several cost-effective constructions for multicast under both request models. Our designs are all relatively simple and easy to be laid out, consume little power, do no accumulate much noise, and are useful for both optical circuit-switching and optical packet/burst switching.