Scheduling unicast and multicast traffic in WDM local optical networks and backpropagation training for an optical neural network

There are three parts in this dissertation: scheduling in WDM local single-hop optical networks, scheduling traffic in multicast supported WDM local single-hop optical networks and backpropagation training for an optical neural network using self-lensing effect.; A single-hop network is attractive in a local area environment where all the nodes can be connected to a single passive star coupler (PSV). The major obstacle to the construction of a good single-hop network is the large latency (tuning time) of tunable devices (transmitter or receiver) when they are tuned from one wavelength to another. To overcome this problem, some scheduling algorithms are needed. In this dissertation, several scheduling algorithms for variable-length messages are proposed. The tuning time of the transmitter and receiver is a main parameter in the algorithms. Some simulations are performed to different loads of the networks.; With the advent of telecommunication services and computer applications requiring multi-destination communication, it is now likely that a significant portion of the overall traffic in future communication environments will be multicast traffic. In this dissertation, I developed several scheduling mechanisms about scheduling traffic in a multicast supported single-hop optical network. In addition, I also presented two protocols about how the nodes get group member information.; Artificial neural networks (ANNs) are becoming a more accepted means for doing non-standard computations. Many tasks that were once difficult, or even impossible, for a traditional numerical processor can now be conducted using ANNs. The main characteristic of the neural network is the parallel processing and the high capacity connections between neurons. Optical neural networks offer an inherent parallelism that allows high capacity connections between neurons. That makes the optical neural network a competent candidate for implementing artificial neural networks. Among all the supervised learning algorithms, backpropagation is probably the most widely used. An experiment is set up to demonstrate that an optical neural network can be trained by backpropagation training algorithm. Several logic gates are successfully trained by using the back propagation training method.