基于动量项与多系统自适应耦合理论的盲源分离算法研究

项目名称： 基于动量项与多系统自适应耦合理论的盲源分离算法研究

项目编号： No.61201457

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 电子学与信息系统

项目作者： 高颖

作者单位： 烟台大学

项目金额： 20万元

中文摘要： 自适应盲源分离技术的进步与发展是当今盲源分离理论能否真正投入实际应用的关键环节，而收敛速度和稳态误差间的矛盾问题又是影响自适应盲分离算法性能的核心因素。本项目将基于动量项与多系统自适应耦合理论，研究时变环境下高效的自适应盲源分离问题解决方案。首先，通过对动量项盲源分离技术的性能分析，采用随机梯度下降法设计动量因子的自适应更新规则，以获取更快的收敛速度；然后，结合牛顿波动方程和海森矩阵研究动量因子与步长参数的内在关系，通过对动量因子的自适应调整实时地获取步长的优选规律，进而突破步长参数对于系统稳态误差的限定；最后，基于多系统自适应耦合理论，结合多个代价函数以获取算法收敛速度和稳态误差的最优平衡。项目研究目的是针对现有自适应盲源分离理论存在的缺陷和难题，采用理论分析和仿真验证相结合的研究策略，以期对缓和收敛速度和稳态误差间的固有矛盾、提高盲源分离技术的实际应用能力方面提供一新的思路和解决方法。

中文关键词： 盲源分离；动量项；系统耦合；步长；收敛性

英文摘要： The progress and development of adaptive blind source separation is the key point about whether the blind source separation theory can be put into actual applications, whereas its performance is mainly influenced by the contradiction between the convergence speed and misadjusment. Based on the momentum term and adaptive convex combination of multiple systems, this project proposes a solution for making the adaptive blind source separation algorithms more effective under time-varying environment. Firstly, by analyzing the performance of the blind source separation technology adding a momentum term, our project would design an adaptive updating rule of the momentum factor based on the stochastic gradient descent method for achieving a faster convergence rate. Then, we will research the inherence between the momentum factor and step size based on the Newton wave equation and Hessian matrix, and thus obtain the optimized rule of step size by adjusting the momentum factor adaptively, which would make a breakthrough at restriction on step size and misajustment. Finally, in order to get an optimized balance between the convergence speed and misadjusment, our project will have resource to the adaptive convex combination of multiple systems theory and several contrast functions. This project aims at solving the limitatio

英文关键词： blind source separation；momentum term；convex system；step size；convergence