瞬时梯度去相关在线EASI算法  

A New Online EASI Algorithm with Instantaneous Gradient Decorrelation

作　　者：杨华[1] 张杭[1] 杨柳[1] 

机构地区：[1]解放军理工大学通信工程学院,江苏南京210007

出　　处：《信号处理》2016年第1期119-126,共8页Journal of Signal Processing

摘　　要：针对在线EASI算法的瞬时梯度与真实梯度存在偏差,并且偏差会随着分离矩阵的迭代更新传递和累积下去,造成分离性能下降的问题,提出了一种瞬时梯度去相关的在线EASI算法—IGDA-EASI算法。IGDA-EASI算法通过消除瞬时梯度间的相关性,减小偏差累积,从而提高算法的分离性能。经仿真实验验证,该算法相较传统在线EASI算法在收敛速度和分离精度方面都获得了较大提高。在信道时变的情况下,IGDA-EASI算法同样具有更好的分离性能。并且在动量项EASI算法中IGDA算法依然具有适用性。Aiming at the problem of performance degradation caused by the existing deviations between instantaneous gradi- ent and true gradient, and furthermore, the deviations would transfer and accumulate to next separation matrix renewal process, a new online EASI algorithm with instantaneous gradient deeon＇elation （IGDA-EASI） is proposed. By eliminating the correlation between the instantaneous gradients, this algorithm decreases the accumulation of the deviation and improves the separation performance effectively. Simulation results show that both the convergence speed and separation accuracy are well improved compare to the traditional online ESAI algorithm. When channel is time-varying the new algorithm also has better separation performance. And the IGDA also shows its applicability to the modified momentum term EASI algorithm.

关 键 词：盲源分离 等变自适应分离算法 瞬时梯度去相关 在线算法 动量项算法 

分 类 号：TN911.7[电子电信—通信与信息系统]