|M.Sc Student||Lahat Eitan|
|Subject||Cellular Oscillator Networks|
|Department||Department of Electrical Engineering||Supervisors||Professor Meir Orenstein|
|Professor Emeritus Yehoshua Zeevi|
In this thesis we study a new computational model, denoted cellular oscillator network (CON). This model represents a collection of coupled non-linear elements, and may be useful for problem solution as well as image processing related operations. This model is based on semiconductor laser physics: arrays of lasers put close together exhibit complex non-linear phenomena. The CON model resembles in its general structure the cellular neural network (CNN) model which was developed as a computational method, inspired by the behavior of coupled electronic oscillator arrays. The CON model should be regarded as an extension of the CNN. By the use of different non-linearity type and different coupling between elements, the CON provides different computational capabilities.
The structural dissimilarities between CON and CNN lead to different dynamics, an issue which we studied both analytically and by means of simulations. We analyzed autonomous arrays (with no external input) of one, two and three elements. We found the steady state and periodic solutions, and their existence and stability domains. Differences in corresponding properties were tracked down to the specific structural dissimilarities causing them. As the CNN is naturally real-valued and the CON is naturally complex-valued, a complex-valued CNN model and real-valued CON model were proposed and used where suitable. The differences we found between CON and CNN systems indicate that CON is a richer model than CNN. Furthermore, the differences indicate that the CON has computational advantages. In this work we established the foundations for finding them.