Transient thermal testing is an established technique for thermal analysis of electronic
components. While much attention has been paid to the experimental handling of transient
measurements, the subsequent data analysis with network identification by deconvolution
is also very challenging and should be performed with great care. One crucial step
is the transformation from a Foster to a Cauer network which is usually done by polynomial
long division. Here, polynomials of degree one hundred and more have to be handled.
To guarantee sufficient numerical accuracy for these computations, arbitrary-precision
arithmetic is applied. In turn, this increases computation times significantly, in
particular for polynomials of higher order, i. e. longer RC lines. In this work, two
alternative approaches for the Foster-to-Cauer transformation are investigated. The
established method of polynomial long division is compared to Khatwani's method and
Sobhy's method with respect to speed and accuracy. The results show a significant
speed advantage of Sobhy's method over polynomial long division for identical accuracies.
