Stability of a Class of Nonreciprocal Cellular Neural Networks

Leon O. Chua, T. Roska

Research output: Contribution to journalArticle

165 Citations (Scopus)

Abstract

Cellular neural networks provide a new and powerful approach to neural computing. Each cellular neural network is uniquely defined by a template. Many useful templates for various applications, such as geometric pattern recognition, have been published. Not only local but even global pattern features can be recognized in real time. This is one generic and remarkable property of cellular neural networks. If these networks are symmetric, i.e., if the feedback values between the cells are reciprocal these networks and their circuit realizations are then completely stable. Practical circuit realizations, however, inevitably give rise to nonreciprocity. In this paper, we show that for a class of practically important templates (positive/negative and opposite-sign templates), the complete stability property is assured even if the symmetry (reciprocity) condition is not met. Moreover, the nonreciprocity allowed in our theorems is not restricted to small or local perturbations in an otherwise reciprocal circuit.

Original languageEnglish
Pages (from-to)1520-1527
Number of pages8
JournalIEEE Transactions on Circuits and Systems
Volume37
Issue number12
DOIs
Publication statusPublished - 1990

Fingerprint

Cellular neural networks
Networks (circuits)
Pattern recognition
Feedback

ASJC Scopus subject areas

  • Engineering(all)

Cite this

Stability of a Class of Nonreciprocal Cellular Neural Networks. / Chua, Leon O.; Roska, T.

In: IEEE Transactions on Circuits and Systems, Vol. 37, No. 12, 1990, p. 1520-1527.

Research output: Contribution to journalArticle

@article{0b006c297c714e6daa8c7d926755ad66,
title = "Stability of a Class of Nonreciprocal Cellular Neural Networks",
abstract = "Cellular neural networks provide a new and powerful approach to neural computing. Each cellular neural network is uniquely defined by a template. Many useful templates for various applications, such as geometric pattern recognition, have been published. Not only local but even global pattern features can be recognized in real time. This is one generic and remarkable property of cellular neural networks. If these networks are symmetric, i.e., if the feedback values between the cells are reciprocal these networks and their circuit realizations are then completely stable. Practical circuit realizations, however, inevitably give rise to nonreciprocity. In this paper, we show that for a class of practically important templates (positive/negative and opposite-sign templates), the complete stability property is assured even if the symmetry (reciprocity) condition is not met. Moreover, the nonreciprocity allowed in our theorems is not restricted to small or local perturbations in an otherwise reciprocal circuit.",
author = "Chua, {Leon O.} and T. Roska",
year = "1990",
doi = "10.1109/31.101272",
language = "English",
volume = "37",
pages = "1520--1527",
journal = "IEEE Transactions on Circuits and Systems",
issn = "0098-4094",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
number = "12",

}

TY - JOUR

T1 - Stability of a Class of Nonreciprocal Cellular Neural Networks

AU - Chua, Leon O.

AU - Roska, T.

PY - 1990

Y1 - 1990

N2 - Cellular neural networks provide a new and powerful approach to neural computing. Each cellular neural network is uniquely defined by a template. Many useful templates for various applications, such as geometric pattern recognition, have been published. Not only local but even global pattern features can be recognized in real time. This is one generic and remarkable property of cellular neural networks. If these networks are symmetric, i.e., if the feedback values between the cells are reciprocal these networks and their circuit realizations are then completely stable. Practical circuit realizations, however, inevitably give rise to nonreciprocity. In this paper, we show that for a class of practically important templates (positive/negative and opposite-sign templates), the complete stability property is assured even if the symmetry (reciprocity) condition is not met. Moreover, the nonreciprocity allowed in our theorems is not restricted to small or local perturbations in an otherwise reciprocal circuit.

AB - Cellular neural networks provide a new and powerful approach to neural computing. Each cellular neural network is uniquely defined by a template. Many useful templates for various applications, such as geometric pattern recognition, have been published. Not only local but even global pattern features can be recognized in real time. This is one generic and remarkable property of cellular neural networks. If these networks are symmetric, i.e., if the feedback values between the cells are reciprocal these networks and their circuit realizations are then completely stable. Practical circuit realizations, however, inevitably give rise to nonreciprocity. In this paper, we show that for a class of practically important templates (positive/negative and opposite-sign templates), the complete stability property is assured even if the symmetry (reciprocity) condition is not met. Moreover, the nonreciprocity allowed in our theorems is not restricted to small or local perturbations in an otherwise reciprocal circuit.

UR - http://www.scopus.com/inward/record.url?scp=0025632457&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0025632457&partnerID=8YFLogxK

U2 - 10.1109/31.101272

DO - 10.1109/31.101272

M3 - Article

AN - SCOPUS:0025632457

VL - 37

SP - 1520

EP - 1527

JO - IEEE Transactions on Circuits and Systems

JF - IEEE Transactions on Circuits and Systems

SN - 0098-4094

IS - 12

ER -