Fa-Long Luo, Lehrstuhl fur Allgemeine und Theor. Elektrotech., Erlangen-Nurnberg Univ., Germany
R. Unbehauen, Lehrstuhl fur Allgemeine und Theor. Elektrotech., Erlangen-Nurnberg Univ., Germany
Hongqin Xue, Lehrstuhl fur Allgemeine und Theor. Elektrotech., Erlangen-Nurnberg Univ., Germany
This paper proposes an analog circuit approach to computing the eigenvectors corresponding to all the eigenvalues of a class of positive definite matrices. The proposed analog computational model can be considered specialized analog computers relying on strongly simplified models of elements. The key features of the proposed analog computational model are asynchronous parallel processing, continuous-time dynamics and high-speed computational capability. We show analytically and by simulations that the proposed circuit can provide the desired eigenvectors with arbitrarily small error during an elapsed time of only a few characteristic time constants of the circuit. In addition, the parameters of the circuit can be obtained from the given matrix without any computations. For the wider use, we also generalize this proposed method for the case in which the matrix takes the complex values. As a result, this proposed approach is satisfactory for many real-time applications fields.
Index Terms:
neural nets; eigenvalues and eigenfunctions; matrix algebra; analogue computers; special purpose computers; parallel processing; neural chips; continuous-time computation; eigenvectors; positive definite matrices; analog circuit approach; specialized analog computers; asynchronous parallel processing; continuous-time dynamics; high-speed computational capability; real-time applications fields
Citation:
Fa-Long Luo, R. Unbehauen, Hongqin Xue, "Continuous-time computation of the eigenvectors of a class of positive definite matrices," pdp, pp.464, 3rd Euromicro Workshop on Parallel and Distributed Processing, 1995