摘要
频域分离算法将时域计算复杂的解卷积问题变换为频域简单的瞬时分离问题。采用基于自然梯度的迭代函数实现瞬时盲分离[2],在频域用瞬时盲分离解决时域卷积盲分离问题时,遇到的两个难点就是频点间幅度和排列次序不确定性,为此我们分别采用归一化方法和分阶段分离方法[5]来解决上述两个问题,仿真实验证明了算法的有效性。
A frequency-domain processing can transform computationally a complex convolute BSS problem in time domain to easi- er-to-olve instantaneous RSS problem in frequency domain. In this paper the iterative function of instantaneous RSS is realized based on natural-gradient. To solve time-domain convolute BSS problem with instantaneous BSS in frequency domain, two challenges is encountered： amplitude inconsistency and permutation inconsistency. In this paper, it adapts normalization and multistage separation respectively to solve these two problems. Simulations have shown the effectiveness of this algorithm.
出处
《山西电子技术》
2008年第3期63-65,共3页
Shanxi Electronic Technology
关键词
自然梯度
频域盲分离
排列次序不一致性
natural-gradient
frequency-domain blind separation
permutation inconsistency
