基于数据分块的自适应变步长盲源分离算法
|
摘要
本文基于经典的自然梯度盲源分离算法提出了一种新型数据分块处理的自适应变步长改进思路,在平稳和非平稳环境中进行正定盲信号分离。其中数据分块处理结合了批处理和自适应在线处理的优点,文中给出了其更新公式的详细的推导过程;变步长则在原有自适应算法的基础上,通过引入性能指数来构造目标函数,反馈到更新公式上,通过选取合适的经验参数自适应的调节步长,在一定程度上寻求稳态误差和收敛速度这对固有矛盾的平衡点,弥补固定步长存在的缺陷。仿真结果表明,所提方法具有在线算法实时跟踪快变非平稳环境的优点,并且对步长有较强的自适应调节能力,收敛速度快,稳态误差小,能以更小的运算量,更短的数据处理时间有效分离混合信号。
A novel adaptive variable-step blind source separation method based on the natural gradient with data-block processing is proposed,which could cope with the determined blind source separation in both stationary and non-stationary environments. Data-block processing combines the benefits of batch processing and adaptive online processing,and its detailed derivation of online updating formula is given in the article;The performance index(PI)is utilized to construct the objective function,which could feed back to the updated formula,so that the step size can be adjusted adaptively by selecting appropriate empirical parameters to seek the balance-point of the inherent contradiction between convergence speed and steady-state error and compensate for the shortcomings of fixed-step size. The simulation results show that the proposed method has the advantages of on-line algorithm,adjusts the step size flexibly,and gets faster convergence speed and better steady-state error. The mixed signals can be separated effectively in shorter time and smaller computation cost.
A novel adaptive variable-step blind source separation method based on the natural gradient with data-block processing is proposed,which could cope with the determined blind source separation in both stationary and non-stationary environments. Data-block processing combines the benefits of batch processing and adaptive online processing,and its detailed derivation of online updating formula is given in the article;The performance index(PI)is utilized to construct the objective function,which could feed back to the updated formula,so that the step size can be adjusted adaptively by selecting appropriate empirical parameters to seek the balance-point of the inherent contradiction between convergence speed and steady-state error and compensate for the shortcomings of fixed-step size. The simulation results show that the proposed method has the advantages of on-line algorithm,adjusts the step size flexibly,and gets faster convergence speed and better steady-state error. The mixed signals can be separated effectively in shorter time and smaller computation cost.
引文
[1] Yoshioka T,Nakatani T,Miyoshi M,et al. Blindseparation and dereverberation of speech mixturesby joint optimization[J]. IEEE Transactions on Au-dio Speech&Language Processing,2011,19(1):69-84.
[2] Liu Q,Wang W,Jackson P J B,et al. Sourceseparation of convolutive and noisy mixtures usingaudio-visual dictionary learning and probabilistictime-frequency masking[J]. IEEE Transactions onSignal Processing,2013,61(22):5520-5535.
[3] Abolghasemi V,Ferdowsi S,Sanei S. Fast and in-coherent dictionary learning algorithms with appli-cation to fMRI[J]. Signal Image&Video Process-ing,2015,9(1):147-158.
[4] Sardouie S H,Shamsollahi M B,Albera L,et al.Interictal EEG noise cancellation:GEVD and DSSbased approaches versus ICA and DCCA basedmethods[J]. Irbm,2015,36(1):20-32.
[5]陈豪,白琳,周治宇.盲源分离技术的发展及应用浅谈[J].空间电子技术,2013,10(1):1-6.
[6]周治宇,陈豪.基于分块自适应的在线Infomax及其扩展算法[J].信号处理,2010,26(11):1632-1637.
[7]邓灵,张天骐,金静,等.基于改进EASI算法的多跳频信号盲源分离[J].科学技术与工程,2014,14(2):45-49.
[8]李振璧,王康,姜媛媛.基于串音误差与分离度的变步长EASI盲源分离算法[J].计算机工程,2017,43(8):108-113.
[9]季策,杨坤,陶奕名,等.一种非平稳环境下的自适应变步长盲源分离算法[J].控制与决策,2016,31(4):735-739.
[10]张天骐,马宝泽,强幸子,等.带自适应动量因子的变步长盲源分离方法[J].通信学报,2017,38(3):16-24.
[11]张延良,师晨旭,张伟涛,等.基于分级迭代变步长的自然梯度盲源分离算法[J].测控技术,2017,36(1):1-4.
[12]Ou S,Wang X,Gao Y. Adaptive improved RLS al-gorithm for blind source separation[J]. Wydawnict-wo SIGMA-NOT,2013,89(3b):81-83.
[13]梅铁民.盲源分离理论与算法[M].西安:西安电子科技大学出版社,2012.
[14]余先川,胡丹.盲源分离理论与应用[M].北京:科学出版社,2011.
[15]杨杰,关丽坤,任学平.不同步长下自然梯度算法分离状态的比较[J].机械设计与制造,2012(7):202-204.
[16]刘俊良,于凤芹.基于分离度变步长的自然梯度算法的语音分离[J].计算机工程与应用,2012:216-219.
[2] Liu Q,Wang W,Jackson P J B,et al. Sourceseparation of convolutive and noisy mixtures usingaudio-visual dictionary learning and probabilistictime-frequency masking[J]. IEEE Transactions onSignal Processing,2013,61(22):5520-5535.
[3] Abolghasemi V,Ferdowsi S,Sanei S. Fast and in-coherent dictionary learning algorithms with appli-cation to fMRI[J]. Signal Image&Video Process-ing,2015,9(1):147-158.
[4] Sardouie S H,Shamsollahi M B,Albera L,et al.Interictal EEG noise cancellation:GEVD and DSSbased approaches versus ICA and DCCA basedmethods[J]. Irbm,2015,36(1):20-32.
[5]陈豪,白琳,周治宇.盲源分离技术的发展及应用浅谈[J].空间电子技术,2013,10(1):1-6.
[6]周治宇,陈豪.基于分块自适应的在线Infomax及其扩展算法[J].信号处理,2010,26(11):1632-1637.
[7]邓灵,张天骐,金静,等.基于改进EASI算法的多跳频信号盲源分离[J].科学技术与工程,2014,14(2):45-49.
[8]李振璧,王康,姜媛媛.基于串音误差与分离度的变步长EASI盲源分离算法[J].计算机工程,2017,43(8):108-113.
[9]季策,杨坤,陶奕名,等.一种非平稳环境下的自适应变步长盲源分离算法[J].控制与决策,2016,31(4):735-739.
[10]张天骐,马宝泽,强幸子,等.带自适应动量因子的变步长盲源分离方法[J].通信学报,2017,38(3):16-24.
[11]张延良,师晨旭,张伟涛,等.基于分级迭代变步长的自然梯度盲源分离算法[J].测控技术,2017,36(1):1-4.
[12]Ou S,Wang X,Gao Y. Adaptive improved RLS al-gorithm for blind source separation[J]. Wydawnict-wo SIGMA-NOT,2013,89(3b):81-83.
[13]梅铁民.盲源分离理论与算法[M].西安:西安电子科技大学出版社,2012.
[14]余先川,胡丹.盲源分离理论与应用[M].北京:科学出版社,2011.
[15]杨杰,关丽坤,任学平.不同步长下自然梯度算法分离状态的比较[J].机械设计与制造,2012(7):202-204.
[16]刘俊良,于凤芹.基于分离度变步长的自然梯度算法的语音分离[J].计算机工程与应用,2012:216-219.