About the robustness of CNN linear templates with bipolar images

This paper defines robustness measures for CNN linear templates. The measures are described for the normal unity gain CNN and also for a very high gain CNN. Inaccuracies in templates are next introduced and formulas for new modified measures are defined. This information can be very useful to CNN hardware designers when determining the acceptable inaccuracies in the coefficient realizations. If some template is found not to be good in presence of inaccuracies the robustness can be increased. One way to increase the defined template robustness is discussed. The increase in robustness factors are calculated and finally one example is given for the use of the described method.

[1]  T. Roska,et al.  Random variations in CNN templates: theoretical models and empirical studies , 1994, Proceedings of the Third IEEE International Workshop on Cellular Neural Networks and their Applications (CNNA-94).

[2]  Michiel Steyaert,et al.  Evaluation of CNN template robustness towards VLSI implementation , 1994, Proceedings of the Third IEEE International Workshop on Cellular Neural Networks and their Applications (CNNA-94).

[3]  Josef A. Nossek,et al.  Design of robust cellular neural networks , 1993 .

[4]  L. Chua,et al.  An analytic method for designing simple cellular neural networks , 1991 .

[5]  Luigi Fortuna,et al.  Cellular Neural Networks: A Survey , 1995 .

[6]  Leon O. Chua,et al.  CNN cloning template: hole-filler , 1990 .

[7]  Lin-Bao Yang,et al.  Cellular neural networks: theory , 1988 .