A Kohonen neural network controlled all-optical router system

The internet is composed of thousands of networks, linking millions of users at academic, industrial and governmental institutions all over the world, offering opportunities to collaborate and to share resources, such as documents, software and data. The ever-increasing number of users confronts the network with an expansion in traffic density, leading to congestion situations during transfer of the data. In addition, future network services which offer an increasing amount of voice and video applications, demand more bandwidth and a high throughput. A cost-effective way to deliver bandwidth is to send multiple wavelengths through a single fibre through an optical network at very high data rates, using a (dense) wavelength-division multiplexing technology that operates in the 1550 nm wavelength window. Considering this context, packet switched all-optical networks not only offer much higher bandwidths than electronic networks due to their ultrafast serial processing, but they also provide for protocol heterogeneity (i.e. transparency of transport layer). In realizing such a network, the following requirements must be met. Optical buffering and contention within the network must be avoided, trading in bandwidth must minimize processing and electronic bottlenecks must be overcome by pushing all electronics to the periphery of the network. An integral part of an optical communication node is a wavelength routing mechanism, sorting optical packets along three dimensions: space, time and wavelength. This paper outlines the basis ideas of a Kohonen self-organizing feature map controlled all-optical router system, providing optimal paths between users without suffering contention problems