Crosstalk prediction in non-uniform cable bundles based on neural network

The statistical approaches for estimating crosstalk in random cable bundles require significant computational effort. The “worst-case” method can mitigate overmuch computation, but it gives a too conservative prediction. In order to account for these problems, a neural network approach to predict crosstalk in non-uniform cable bundles at low frequencies where circuits are electrically small is proposed. A BP neural network model is trained by Levenberg-Marquardt algorithm based on statistical simulation results calculated by RDSI algorithm. By comparing the predicted results and the simulation ones, an adequate match between them shows that the proposed neural network method has the ability to predict crosstalk in non-uniform cable bundles rapidly and accurately.