摘要
针对现有的盲分离算法大多存在收敛速度慢、优化精度低的问题,提出了一种新的基于模拟退火粒子群的盲分离算法。该算法以分离信号的负熵为目标函数,根据分离信号的状态、粒子的惯性权值随退火温度及适应度的变化自适应地调节,既基本保持了粒子群算法简单容易实现的特点,又改善了其摆脱局部极值点的能力,提高了算法的收敛速度、分离精度和稳定性能。仿真对比结果表明,新算法性能明显优于自然梯度卷积混合盲分离算法和小波变换快速独立分量分析算法,很好地实现了实时语音信号的分离且提高了分离性能。
Aiming at the problems of slow convergence and low computational precision of existing blind source separation methods,a new blind source separation algorithm based on simulate anneal and particle swarm optimization is proposed.The proposed algorithm takes the negentropy of mixtures as a target function,which adaptively adjusts the inertia weight factor with changes of annealing temperature and fitness according to the state of the separation signal.It is almost as simple for implement as particle swarm optimization, but also can improve the abilities of seeking the global excellent result and evolution speed and stable performance.The experimental results demonstrate that the new algorithm has better performance than convolution blind source separation based-natural gradient and FastICA based-wavelet transform algorithm,and it could efficiently separate the mixture of real-time speech signal in real environment.
出处
《计算机工程与设计》
CSCD
北大核心
2010年第23期5067-5070,共4页
Computer Engineering and Design
基金
湖北省教育厅重点科研基金项目(D20101704)
关键词
实时语音信号
盲源分离
卷积混合
粒子群
模拟退火
负熵
real-time speech signal
blind source separation
convolutive mixtures
particle swarm optimization
simulate anneal
negentropy
