摘要
定义了一种新的映射关系,有效地简化了应用径向基函数神经网络(RBFNN)实现波束形成时训练数据的产生.采用两个网络并行处理的方法提高了网络收敛速度,通过后续处理逼近维纳解.仿真实验表明,该算法的信号跟踪能力与最小方差无畸变响应(MVDR)算法的跟踪能力十分接近.但由于该算法结合了神经网络容错能力强的特点和并行计算的结构优势,比MVDR算法更有效地提高了运行速度,并且对系统误差具有更强的鲁棒性.
Defining a new mapping relationship, the generation process of training data is effectively simplified when using the radial basis function neural network (RBFNN) to realize beamforming. The parallel processing with two networks are used to improve the network convergence rate and approximate Wiener solution via post-processing. Computer simulations show that the signal trackability of this algorithm approximates very well that of the minimum variance distortionless response (MVDR) algorithm. But because this algorithm combines the strong fault tolerability of neural network with the architectural advantage of parallel operation, it can make the computation more rapid and more robust to systematic error in comparison to MVDR algorithm.
出处
《东北大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2006年第2期161-164,共4页
Journal of Northeastern University(Natural Science)
基金
河北省教育厅科学研究指导计划项目(Z2004103)