A Study Of GPU Parallelized Reconstruction Algorithm And Application Of Electromagnetic Environment In Reverberation Chamber

Reverberation chambers (RC) are large overmoded cavities in which the electromagnetic field is statistically uniform and isotropic. After years of development, RC became an Electromagnetic Compatibility (EMC) measurement system which was widely accepted. A well-operating RC with the electromagnetic field could be a multiple scattering and multipath environment. Numerical methods have been widely used to analyze the chamber environment because experimental methods can only sample the environment on some discrete points; however, The numerical analysis of the RC requires great computer resources. If the device has complex details or the frequency range to be investigated is much higher than the RC cutoff frequency, the computation workload and the simulation time will increase greatly. The rapidly evolving GPU heterogeneous parallelized computing techniques become one of the most powerful tool in computational electromagnetics. With the help of GPU, some complex electromagnetic problems (such as RC electromagnetic environment) can be simulated and solved.This dissertation studies the GPU accelerated numerical simulation methods for RC electromagnetic environment. Based on D. A. Hill’s plane wave integral representation theory for a reverberation chamber, a GPU-accelerated finite difference time domain (FDTD) method is proposed to recreate the RC electromagnetic environment. The simulated numerial results are validated by the theoretical results and the IEC61000-4-21 standard. Further more, statistical analysis are performed to obtain the optimal parameters. Finally, this work is extended to be applied in the biological electromagnetic dosimetry fields. The relevant work is supported by National Natural Science Foundations of China (No.30470429 and No.30870577).The main contributions of this work are:Firstly, RC electromagnetic environment reconstruction theory and its numerical validation are improved. Secondly, GPU accelerated RC electromagnetic environment reconstruction algorithm is proposed and optimized for computation efficiency. And finally, new biological electromagnetic dosimetry quanties are investigated to evaluate RC electromagnetic environment.The first contribution is the RC electromagnetic environment reconstruction theory. Based on D. A. Hill’s plane wave integration theory, the RC electromagnetic environment can be reproduced by a superposition of a finite number of random incident plane waves. The probabilistic distribution of incident plane wave parameters, zenith angle θ and azimuth angle (p, are deduced and suitably arranged to form the statistical properties attributed to the fields in RC. A numerical method is proposed to generate the samples, and the distribution of the samples are validated with Chi-square test, which improved the theoretical basis.Further more, in our numerical method, the availability of the electric field values in the whole working area allows the analysis of the uniformity and other statistical parameters.The recreated RC electromagnetic environment is verified at four metrics,1) correlation coefficients of electric field in different numerical simulations,2) hypothesis testing of electric field probabilistic distribution,3) hypothesis testing of the mean value of electric field in RC, and 4) standard deviation of electric field uniformity. All the numerial results are checked through the well-established reverberation chamber statistical theory and/or the IEC61000-4-21 standard, and statistical analysis are performed to obtain the optimal parameters in simulating the RCs.The second contribution is the optimization of the GPU accelerated FDTD algorithm used in RC simulation. The CPML model of the GPU FDTD method is time consuming. Parallel acceleration of CPML algorithm suffers from massive inefficiency division operations and the duplicated memory access. To achieve higher efficiency, we review the CPML theory and propose a novel division-free minimum-access parallel CPML (DFMA CPML) structure. By optimally rearranging the CPML inner iteration process, all the division operators can be eliminated and more than 50% arithmetic instructions could be saved. In the original two-step CPML iteration scheme,2/3 memory access of the Ψ auxiliary term is redundant. We propose a domain-decomposition joint-updating strategy, which decomposes the computation domain logically and joint the computing procedure accordingly. Benefiting from this strategy, data loading/writing operations in the CPML updating process can be reduced to the lower bound of CPML theory. Experimental results show that the proposed DFMA CPML structure can save up to 70% execution time of the traditional GPU-CPML technique without any accuracy loss.Meanwhile, a novel GPU accelerated multiple incident plane wave source based on total field/scatter field formulation (TF/SF) is proposed. Incident fields of a fixed point on TF/SF boundary is isolated from other points, so that the computing tasks of all incident fields of a fixed point can be combined and then executed by one GPU thread. By optimally rearranging the computing tasks of GPU threads and the data structure, only a few amount of fastest registers will be used instead of massive global memory. Therefore, any number of incident plane waves can be introduced to total field area simultaneously, and the data storage requirement and memory access operation can be reduced to slightly more than only one plane wave is introduced, which achieves the lower bound of TF/SF method. Experimental results show that the proposed procedure can save up to 64% execution time at the same time.The last contribution is the application of the biological electromagnetic. Based on the RC statistical theory and our numerical method, biological electromagnetic dosimetry of RC electromagnetic environment/in-cavity electromagnetic environment is investigated. It can be derived that in the RC/in-cavity environment, the Specific Absorption Rate (SAR) should be replaced by its accumulate average value (SAR) as the basic restriction quantity, and the environment power density P should be replaced by scalar power density S as the derived restriction quantity. S and 〈SAR〉have proportional relationship in quantity, and are independent from other factors. Comparing the ICNIRP guidelines, our choice of restriction quantities can describe the biological electromagnetic effect in RC/in-cavity environment more appropriately.At last a new strategy to evaluate the biological model absorption characteristic is discussed. This strategy uses RC/in-cavity environment as the facility and uses, 〈SAR〉 to evaluate the electromagnetic power absorbing in the model. The numerical results show that this strategy can produce the entirety evaluation result directly.