摘要
对一组线性瞬时混合信号,采用高斯混合模型拟合各个独立源的概率密度分布进行分离,其复杂度随信号源数量、高斯混合模型阶数的增加急剧上升。提出用统一计算设备架构(compute unified device architecture,CUDA)对该分离方法进行设计,实现该方法的并行加速处理。实验结果表明,此加速方案可以有效降低该盲分离方法的时间复杂度。
For a group of instantaneous linear mixing signals, the complexity of the blind separation method using Gaussian mixture model to fit their probability density functions becomes higher as either of the number of independent signal sources or the order number of Gaussian mixture model increases respectively. In this paper parallel implementation is proposed for this method on NVIDIA＇s CUDA platform to get accelerated processing. Experiment results show that our accelerating scheme can reduce time complexity of it.
出处
《自动化与信息工程》
2013年第1期21-25,共5页
Automation & Information Engineering
基金
广东省级大学生创新训练项目(1057312043)
广东药学院大学生创新实验项目(85)
