可变脉冲激励的磁共振功放的非线性建模和预失真线性化研究

Fast magnetic resonance imaging (MRI) has been a key technology of rapid screening and accurate diagnosis of major diseases such as cancers in China. In the fast MRI system, magnetic resonance power amplifier （PA） is a key component; unfortunately it has bifurcation and dynamic variable nonlinearities due to injected fast variable pulses. Therefore, the imaging quality of the fast MRI system is significantly deteriorated by PA nonlinearities. In other words, the PA linearity or nonlinearity linearization ultimately determines the performance of the fast MRI imaging system. The traditional nonlinear models ignore the nonlinear problems of thermal rapid change of PA, and thus it is difficult to accurately model the bifurcation nonlinearity of magnetic resonance PA. Further, traditional/existing pre-distortion technologies are difficult to track and adapt to the fast change of the variable pulse signal of the magnetic resonance PA, and thus the fast switching cannot be realized between the different pre-distorters. In this project, nonlinear behavior models of magnetic resonance PA will be explored and investigated, in which the effect of thermal memory will be introduced, and the temperature and the variable input signal of PAs will be also considered. Finally, multi-dimensional nonlinear behavior models based on the above will be explored and investigated. Owing to the fast change of the pulse signal of magnetic resonance PA, dynamic deviation of the PA nonlinearity will be introduced, so the variable incremental pre-distortion circuits will be proposed and investigated to realize the fast switching of the pre-distorters. At last, the linearized magnetic resonance PA will be tested to meet the high quality imaging of MRI system.

快速磁共振成像是我国实现肿瘤等重大疾病快速筛查和精准诊断的重点发展技术。磁共振功率放大器作为快速磁共振成像系统的核心部件，由于脉冲快速可变，其非线性存在分叉且动态可变，造成磁共振成像系统成像质量显著恶化。传统非线性模型忽略功放热快速变化带来的非线性问题，难以对磁共振功放的分叉非线性进行精确建模；传统预失真架构受硬件制约，难以适应磁共振功放的不同脉冲信号的快速变化，无法实现不同预失真器间的快速切换。本项目将研究磁共振功放的非线性行为模型，引入热记忆效应，将温度和功放输入信号作为模型的输入，功放输出信号作为模型的输出，提出基于温度输入的多维度非线性行为模型；并针对磁共振功放脉冲快速变化的特点，引入功放非线性的动态偏差，提出可变增量预失真架构，以实现预失真器的快速切换，最终实现线性磁共振功放，满足磁共振系统快速扫描时高质量成像的需要。

本项目首先建立了磁共振功放非线性仿真和实验平台，该测试平台由矢量信号发生器（安捷伦N5182A）、带数字板卡的矢量信号分析仪（安捷伦N9020A）、任意波形发生器（安捷伦33250A）和高功率磁共振功放等组成。该测试平台既可用于捕获磁共振功放输入输出端基带IQ信号，又能够验证不同非线性模型和预失真器的性能。然后本项目联合鑫高益医疗设备股份有限公司开发了具有自主知识产权的应用于1.5T磁共振系统的18KW功率放大器样机。接着本项目基于该放大器分析了在功率大幅动态变化的射频脉冲作用下磁共振功放非线性的特点，提出了时间功率混合分段非线性模型，时间功率混合分段模型是针对磁共振功放的激励为辛格脉冲信号时的精确非线性模型，能够对磁共振功放的大动态分叉非线性进行精确建模，从而更好地跟踪高功率磁共振功放的非线性特征。并基于时间功率混合分段模型，本项目又提出了分段级联数字预失真器用于矫正高场磁共振高功率射频功放的大动态范围分叉非线性和射频发射链路的IQ缺陷，从而提高MRI系统图像质量，为实现高场MRI射频功放的国产化奠定基础。紧接着，本项目又提出了实复混合时延神经网络模型用于建模宽带射频功放的非线性，并基于该模型，构建了数字预失真器，通过对被测功放，进行线性化验证，证实了实复混合时延神经网络数字预失真器相比记忆多项式数字预失真器和实数时延神经网络数字预失真器，能够更好的抑制宽带射频功放的带外再生频谱。本项目还提出了改进型并行双非线性广义记忆效应模型，以准确建模被测非线性系统的强动态非线性。同时，为了进一步降低传统非线性模型的系数数量，进而减少训练时所耗的计算资源量，从而方便硬件实现，本项目提出了一种宽带射频功放的动态非线性建模系统，完成了基于系数分解的非线性建模装置设计。