**Authors:** Joe Wiart 1,2, Pierric Kersaudy 1,3, Amal Ghanmi 3 , Nadege Varsier1,2
,
Abdelhamid Hadhem 1,2 , Picon Odile 2 , Bruno Sudret 4 and Raj Mittra5

**Source:** FERMAT,VOLUME12,ARTICLE,NOV-DEC-2015

**Abstract:** In this paper we propose a novel approach
which combines computational electromagnetics with
statistics to statistically characterize the variations of the
Radio Frequency (RF) exposure induced by inputs and
affected by variability or uncertainty. Conventional
numerical techniques such as the Monte Carlo Method,
typically used to solve such a problem, are not useful in
this case from a practical point of view since the
computation time needed to assess the effect of the
exposure is inordinately long for this type of problem. This
novel approach consists of characterizing the statistical
distribution of the output using a surrogate model which
is employed in the numerical method. The bottleneck is
encountered in the process of building a surrogate model
by using a parsimonious approach, because an extensive
set of computations are required by the Finite Difference
in Time Domain (FDTD) method, despite the fact that the
FDTD is a proven computationally efficient technique for
modeling problems in bio-electromagnetism. The
proposed method employs a truncated Generalized
Polynomial Chaos Expansion scheme in conjunction with
regression and Least Angle Regression (LARS) algorithms
to identify the polynomial which has a significant influence
and then to calculate the polynomial coefficient. The
accuracy assessment of the surrogate model is carried out
with the Leave-One-Out Cross Validation (LOOCV). In
this paper this method is used to characterize the variation
of the Specific Absorption Rate (SAR) induced in the head
by a mobile phone having a variable position relative to
the head.

**Index Terms:** Human exposure, Specific Absorption
Rate (SAR), Dosimetry, radiofrequency (RF), Finite
Difference in Time Domain (FDTD), Generalized
Polynomial Chaos Expansion, Least Angle Regression,
Leave One Out Cross Validation.

**View PDF**