Random Processes Metamodeling Applied to Dosimetry

Advanced computational tools take advantage of progress in high-performance calculation (HPC). Nowadays the wireless network design is simulated using software for the purpose of predicting coverage, capacity and throughput of planned networks. Despite these progress in HPC, uncertainty quantification requests enormous calculations which are often unaffordable due to the computational burden. Surrogate models have been investigated to overcome such a limitation and are used to date in most of the engineering domains. Previous study explored a 4G-induced exposure to Electromagnetic Fields in terms of propagation environments using stochastic city models. These stochastic models are governed by a few Constant Parameters (CP), such as, built surface, height of building or length of streets. The influence of these CP on the EMF exposure should be investigated by a sensitivity analysis. To do so, an efficient surrogate method dedicated to random processes is needed. Relative study is important but scant. To overcome such a limit, this paper proposes an innovative approach to built surrogate models for stochastic simulators based on Karhunen-Loève expansion combined with the polynomial chaos expansion to surrogate the covariance.