基于图形处理器的时域有限差分算法研究

如何提高时域有限差分算法(FDTD)的运算效率一直是FDTD数值运算研究的核心问题之一.针对近年来图形处理器(GPU)运算能力的高速增长及GPU通用运算概念提出的背景,对GPU加速FDTD运算的潜力与研究现状进行了总结,并对GPU加速FDTD运算的并行实现原理进行了阐述,通过将其与其他典型硬件加速方式进行比较,指出了G...     

Quantitative spectral/spatial analysis of phased array coil in magnetic resonance imaging based on method of moment

A new approach for analysis of RF coils in magnetic resonance (MR) experiments is reported. Instead of assuming current distribution in conventional quasi-static algorithm, this approach transforms the coil geometry into an equivalent circuit for complex current calculation. Self and mutual inductance are taken into consideration. Frequency responses of RF coils and transverse magnetic field (B1) maps can be simulated. This approach is especially efficient for phased array coil design for its small matrix size when implemented on computers. Experiments on both single surface coil and phased array coils are consistent with simulation results. Index Terms-Magnetic resonance, method of moment, phased array coil, RF coil.     

基于GPU加速的地震图像重建技术

针对目前地层层析成像算法中正演算法存在计算量大、计算速度慢的问题,以图像处理器(GPU)为核心,研究并实现了一种基于GPU平台的时域有限差分(FDTD)正演算法。CUDA是一种由NVIDIA推出的GPU通用并行计算架构,也是目前较为成熟的GPU并行运算架构。而FDTD正演算法本身在算法特性上满足并行的要求,二者的结合将极大地加速程序的计算速度。在基于标准Marmousi速度模型的正演模拟中,程序速度提升30倍,而GPU正演图像与CPU正演结果误差小于千分之一。算例表明CUDA可以大大加速目前的FDTD正演算法,并且随着GPU硬件自身的发展和计算架构的不断改进,加速效果还将进一步提升,这将有利于后续波形反演工作的进展。     

An inverted-microstrip resonator for human head proton MR imaging at 7 tesla

As an extension of the previously developed microstrip transmission line (MTL) RF coil design, a high-frequency RF volume coil using multiple inverted MTL (iMTL) resonators for human head imaging at high magnetic field strength of 7 tesla (T) is reported. Compared to conventional MTL resonators, iMTL resonators can operate at higher frequency with lower losses and, thus, are suitable for designs of high-frequency RF volume coils with large coil size for human MR imaging and spectroscopy at high fields. An approach using capacitive terminations was analyzed and applied to the design of the iMTL volume coil for improving RF field homogeneity and broadening frequency-tuning range. A performance-comparison study was conducted between the prototype iMTL volume coil and a custom-built TEM volume coil at 7 T. The iMTL volume coil presents a comparable SNR and intrinsic B1 homogeneity to the TEM volume coil. Phantom and the human head images acquired using the iMTL volume coil are also presented. The proposed iMTL volume coil provides an efficient and alternative solution to design high-frequency and large-size volume coils for human MR applications at very high fields.     

采用粒子群算法的频率选择表面优化设计

 粒子群优化算法作为优秀的群体智能算法之一,已经被广泛应用于电磁优化问题中.通过与时域有限差分(FDTD)算法相结合,粒子群优化算法被运用于频率选择表面的优化设计.在此过程中,通过使用图形处理器(GPU)加速技术将FDTD算法速度提高近100倍,配合FDTD的宽频特性显著加速了优化过程.在此基础上,针对给定单元结构以及未知单元结构两种频率选择表面设计情况,分别采用带惯性权重的粒子群优化算法以及二进制离散粒子群优化算法进行优化.提出的两种优化流程在算例中得到验证与分析,证明了其可行性及高效性.     

高阶FDTD法分析电-大尺寸光波导器件

高阶时域有限差分(FDTD)法用于电-大尺寸平面光波导器件的时域分析,实现了高阶FDTD法的理想匹配层(PML)吸收边界条件;研究了高阶FDTD法的数值色散特性,并对平行介质带定向耦合器进行了数值模拟,所得结果与解析解非常一致。     

Shielded Microstrip Array for 7T Human MR Imaging

The high-frequency transceiver array based on the microstrip transmission line design is a promising technique for ultrahigh field magnetic resonance imaging (MRI) signal excitation and reception. However, with the increase of radio-frequency (RF) channels, the size of the ground plane in each microstrip coil element is usually not sufficient to provide a perfect ground. Consequently, the transceiver array may suffer from cable resonance, lower Q-factors, and imaging quality degradations. In this paper, we present an approach to improving the performance of microstrip transceiver arrays by introducing RF shielding outside the microstrip array and the feeding coaxial cables. This improvement reduced interactions among cables, increased resonance stability, and Q-factors, and thus improved imaging quality. An experimental method was also introduced and utilized for quantitative measurement and evaluation of RF coil resonance stability or “cable resonance” behavior.      

Computation of aperture antenna mutual coupling using FDTD andKirchhoff field transformation

A new method for computing the mutual coupling between aperture antennas using the finite-difference time-domain (FDTD) method together with the Kirchhoff near-field to near-field transformation is described. The method offers a reduction in computer storage, particularly for widely spaced antenna elements in an array     

Ultrahigh-field MRI whole-slice and localized RF field excitations using the same RF transmit array

In this paper, a multiport driving mechanism is numerically implemented at ultra high-field (UHF) magnetic resonance imaging (MRI) to provide 1) homogenous whole-slice (axial, sagittal, or coronal) and 2) highly localized radio frequency (RF) field excitation within the same slices, all with the same RF transmit array (here chosen to be a standard transverse electromagnetic (TEM) resonator/coil). The method is numerically tested using a full-wave model of a TEM coil loaded with a high-resolution/18-tissue/anatomically detailed human head mesh. The proposed approach is solely based on electromagnetic and phased array antenna theories. The results demonstrate that both homogenous whole-slice as well as localized RF excitation can be achieved within any slice of the head at 7 T (298 MHz for proton imaging).     

遗传算法与FDTD优化设计智能天线阵

研究了遗传算法(GA)和时域有限差分法(FDTD)结合来优化智能天线阵列。采用FDTD精确分析考虑互耦影响的智能天线阵,然后结合遗传算法以信噪比为优化基准对加权向量和天线的结构进行了一体化的优化设计,使用Fortran和MATLAB混合编程,优化设计了均匀直线阵和圆形智能天线阵。最后,通过实验天线验证了其有效性。     

FDTD analysis of phased array antennas

This work presents a new application of the finite-difference time-domain (FDTD) method to the generalized analysis of phased array antennas. The generality of the FDTD method brings important advantages to the phased array antenna analysis problem, allowing the modeling of complex conductor and dielectric geometries with relative ease. Additionally, a new broad-band FDTD periodic boundary condition is developed which allows the array problem to be simplified to a periodic unit cell computational domain. This hybrid frequency/time-domain periodic boundary condition enables solution of the periodic phased array problem for arbitrary scan conditions in a broadband fashion. The new method is applied to waveguide and stacked microstrip antenna arrays and the numerical results are compared to experimental or analytic solutions, demonstrating the validity and utility of this method     

Towards accelerating irregular EDA applications with GPUs

Recently graphic processing units (GPUs) are rising as a new vehicle for high-performance, general purpose computing. It is attractive to unleash the power of GPU for Electronic Design Automation (EDA) computations to cut the design turn-around time of VLSI systems. EDA algorithms, however, generally depend on irregular data structures such as sparse matrix and graphs, which pose major challenges for efficient GPU implementations. In this paper, we propose high-performance GPU implementations for a set of important irregular EDA computing patterns including sparse matrix, graph algorithms and message-passing algorithms. In the sparse matrix domain, we solve a core problem, sparse-matrix vector product (SMVP). On a wide range of EDA problem instances, our SMVP implementation outperforms all prior work and achieves a speedup up to 50× over the CPU baseline implementation. The GPU based SMVP procedure is applied to successfully accelerate two core EDA computing engines, timing analysis and linear system solution. In the graph algorithm domain, we developed a SMVP based formulation to efficiently solve the breadth-first search (BFS) problem on GPUs. We also developed efficient solutions for two message-passing algorithms, survey propagation (SP) based SAT solution and a register-transfer level (RTL) simulation. Our results prove that GPUs have a strong potential to accelerate EDA computing through designing GPU-friendly algorithms and/or re-organizing computing structures of sequential algorithms.     

A finite difference method for the design of gradient coils in MRI--an initial framework

This paper proposes a finite-difference (FD)-based method for the design of gradient coils in MRI. The design method first uses the FD approximation to describe the continuous current density of the coil space and then employs the stream function method to extract the coil patterns. During the numerical implementation, a linear equation is constructed and solved using a regularization scheme. The algorithm details have been exemplified through biplanar and cylindrical gradient coil design examples. The design method can be applied to unusual coil designs such as ultrashort or dedicated gradient coils. The proposed gradient coil design scheme can be integrated into a FD-based electromagnetic framework, which can then provide a unified computational framework for gradient and RF design and patient-field interactions.     

基于GPU通用计算平台的乐谱自动识别系统设计

在GPU通用计算平台上实现了一个钢琴独奏乐曲的乐谱识别系统,它读取WAV格式音频文件,利用GPU通用计算技术加速自相关函数算法来实现音高的识别,并综合考虑短时能量和基音周期的变化进行节拍划分。通过实际测试,验证了该乐谱识别系统的准确性,并证明了GPU并行计算对系统计算效率提升的效果：将计算时间减少到传统CPU计算时间的16%左右。     

基于异构平台的细胞神经网络算法研究

随着GPU的发展,其计算能力和访存带宽都超过了CPU,在GPU上进行通用计算具有成本低、性能高的特点。细胞神经网络由于其特有的性质,非常适合利用GPU进行并行计算,因此,该文提出了利用CU-DA实现的基于GPU的细胞神经网络异构算法,并应用在图像边缘检测上。实验结果证明,与传统的利用CPU实现的边缘检测方法相比,在速度上,基于GPU实现的图像边缘检测方法提高了数十倍,为细胞神经网络在实时图像、视频处理上的应用提供了新的方法。     

A hybrid inverse approach applied to the design of lumped-element RF coils

A combination of inverse procedures is employed in the design of radio-frequency (RF) coils with specific examples in, but not restricted to, magnetic resonance imaging. The first inverse procedure is the use of functional methods for the optimization of coil characteristics subject to restrictions on the field behavior. Continuous current distributions are derived from analysis of the fields they are required to produce. To make use of these distributions at a desired frequency, the method of moments is applied as a second inverse procedure to a discretized version of the current distribution. The advantage of this hybrid technique is that it provides a computational algorithm for optimization of feeding, tuning, impedance matching and other aspects of RF coil design. A prototype RF coil has been built using the engineering values predicted by the theory. Experimental results including images acquired from the prototype coil are presented.     

Electromechanical design and construction of a rotating radio-frequency coil system for applications in magnetic resonance

While recent studies have shown that rotating a single radio-frequency (RF) coil during the acquisition of magnetic resonance (MR) images provides a number of hardware advantages (i.e., requires only one RF channel, avoids coil-coil coupling and facilitates large-scale multinuclear imaging), they did not describe in detail how to build a rotating RF coil system. This paper presents detailed engineering information on the electromechanical design and construction of a MR-compatible RRFC system for human head imaging at 2 T. A custom-made (bladeless) pneumatic Tesla turbine was used to rotate the RF coil at a constant velocity, while an infrared optical encoder measured the selected frequency of rotation. Once the rotating structure was mechanically balanced and the compressed air supply suitably regulated, the maximum frequency of rotation measured ~14.5 Hz with a 2.4% frequency variation over time. MR images of a water phantom and human head were obtained using the rotating RF head coil system.     

Variable Power Combiner for RF Mode Shimming in 7-T MR Imaging

This contribution discusses the utilization of RF power in an MRI system with RF mode shimming which enables the superposition of circularly polarized modes of a transmit RF coil array driven by a Butler matrix. Since the required power for the individual modes can vary widely, mode-shimming can result in a significant underutilization of the total available RF power. A variable power combiner (VPC) is proposed to improve the power utilization: it can be realized as a reconfiguration of the MRI transmit system by the inclusion of one additional matrix network which receives the power from all transmit amplifiers at its input ports and provides any desired (combined) power distribution at its output ports by controlling the phase and amplitude of the amplifiers' input signals. The power distribution at the output ports of the VPC is then fed into the "mode" ports of the coil array Butler matrix in order to superimpose the spatial modes at the highest achievable power utilization. The VPC configuration is compared to the standard configuration of the transmit chain of our MRI system with 8 transmit channels and 16 coils. In realistic scenarios, improved power utilization was achieved from 17% to 60% and from 14% to 55% for an elliptical phantom and a region of interest in the abdomen, respectively, and an increase of the power utilization of 1?dB for a region of interest in the upper leg. In general, it is found that the VPC allows significant improvement in power utilization when the shimming solution demands only a few modes to be energized, while the technique can yield loss in power utilization in cases with many modes required at high power level.     

A modified FDTD (2, 4) scheme for modeling electrically largestructures with high-phase accuracy

A new fourth-order finite-difference time-domain (FDTD) scheme has been developed that exhibits extremely low-phase errors at low-grid resolutions compared to the conventional FDTD scheme. Moreover, this new scheme is capable of combining with the standard Yee (1966) scheme to produce a stable hybrid algorithm. The problem of wave propagation through a building is simulated using this new hybrid algorithm to demonstrate the large savings in computing resources it could afford. With this new development, the FDTD method can now be used to successfully model structures that are thousands of wavelengths large, using the present day computer technology     

Improvement of the quality factor of RF integrated inductors bylayout optimization

A systematic method to improve the quality (Q) factor of RF integrated inductors is presented in this paper. The proposed method is based on the layout optimization to minimize the series resistance of the inductor coil, taking into account both ohmic losses, due to conduction currents, and magnetically induced losses, due to eddy currents. The technique is particularly useful when applied to inductors in which the fabrication process includes integration substrate removal. However, it is also applicable to inductors on low-loss substrates. The method optimizes the width of the metal strip for each turn of the inductor coil, leading to a variable strip-width layout. The optimization procedure has been successfully applied to the design of square spiral inductors in a silicon-based multichip-module technology, complemented with silicon micromachining postprocessing. The obtained experimental results corroborate the validity of the proposed method. A Q factor of about 17 have been obtained for a 35-nH inductor at 1.5 GHz, with Q values higher than 40 predicted for a 20-nH inductor working at 3.5 GHz. The latter is up to a 60% better than the best results for a single strip-width inductor working at the same frequency