Both local SAR values for FDTD and RBFNN at 50 MHz and at 450 MHz are shown in Figures 4(a) and 4(b), respectively. These figures were drawn for the 300th cell that shows the center of x axis.
Results obtained for these two methods are good in track. Based on this point, RBFNN model can provide automatic estimation of local SAR values in the working frequency values and requires coordinate location for each layer. This simulation was realized for three-layer cylindrical human model; however, this method can be easily used for different structures.
Variation of maximum error levels which occur in the output of the network is shown in Figure 5. We have reached to these values by choosing maximum values between RBFNN results and FDTD results for each frequency in test step. Accuracy levels of the network changed with using different frequency values. In the designed model, the error level can be reduced by using an optimum sigma (width of the Gauss Kernel function) value. The variations are given in Figures 4 and5 for Ο 50.
The test accuracy of RBFNN changes depending on the sigma parameter. The variation of maximum error values which is calculated for all frequency values in the test data with respect to sigma parameter is shown in Figure 6. The sigma is an effective parameter for the performance of network. The determination of optimum sigma values is important for the accuracy of the system. In this work, Gaussian Kernel width was determined experimentally for maximum performance.
5. CONCLUSION
In this study, a new approach based on RBFNN was presented for the estimation of local SAR values for the three-layer cylindrical human model. The target data were determined with FDTD solution, which was used to determine the local SAR, when a plane wave hits the PML-surrounded three-layer cylinder. The results show that RBFNN is an effective model for this problem space, and RBFNN design solutions are well matched with the FDTD solution results. This method was proposed for the three-layer cylindrical human model; however, this topology can be applied for different structures.