Abstract

One-dimensional broadcast networks are common, e.g., in the space dimension we have the crossbar networks, and in the wavelength dimension we have the star-based networks (e.g., “broadcast and select”). These networks have limited connectivity: the space-division networks are limited by the switch count and complexity of control, and the wavelength-division networks have a relatively small number of wavelength-channels due to spectrum limitation. In the proposed k-dimensional networks,¹ a communication channel between input and output is consists of k coordinates, one per dimension. Each input in the network broadcasts its fixed k coordinates to all the outputs. A connection is established when an output selects the appropriate coordinates (one per dimension) of a specific input. These networks are strictly non-blocking and have very high connectivity (>>1000) using only several dimensions (say 3). The high connectivity is achieved by relaxing the constraints on the size of each dimension since several dimensions share the load of the network, e.g., a 3-dimensional network with connectivity 1000 would require only 10 channels in each dimension. The network is optimal in essence that its complexity achieves the information lower bound through hierarchical multiplexing. An example of a 3-idimensional optical-wavelength, microwave-frequency and space is described and compared with some traditional networks.

© 1993 Optical Society of America