基于原子库优化的测井声波信号分离研究
|
摘要
随着盲信号分离技术的发展,该技术已经广泛应用在图像处理,地震探测,语音识别,生物医学等方面。大多数盲信号分离算法对多通道接收的混合信号有有效地分离效果,但是对单通道频谱重叠严重的混和信号采用现有的一些盲信号分离方法,有一定的局限性。国内外已有学者对单通道混合的AM-FM信号进行分离,并且在知道先验信息的前提下对比分离出的各个混和信号与源信号的波形图,验证了算法的有效性。本论文在研究了已有算法的基础上,采用提取的波至时刻优化原子库的MP稀疏分解算法对测井声波信号纵波及横波中混合信号进行分离,得到混合信号中各个分量信号的频率、幅度和相位等信息。此信息有助于分析计算岩石的岩性、孔隙、油层等特征。
测井声波信号是时频重叠的信号,单独在时域、频域和空域滤波都比较难分离,现有的研究方法也不能完全达到预期效果,如何分离测井声波信号,得到更多相关信息是目前研究的难题。针对测井声波信号的特点及单通道分离混和信号的原理,本论文选取利用原子库优化的MP稀疏分解方法,将该方法应用于测井声波信号分离中。主要工作如下：
1、首先简单介绍了盲信号的模型、分类和分离算法。接着研究了测井声波信号的特点,用三个信号分量叠加得到测井声波信号的检测信号,为接下来的研究做好准备。
2、详细介绍了原子库优化的方法。由于提取的后续波波至时刻存在误差,在应用到原子参数设置时需要增加波至时刻的左右采样点数。通过对模型信号的仿真实验,得出结论：提取的波至时刻需要左右增加四个采样点数,才能减小提取的波至时刻误差。并且当信号起始点时刻重叠程度达到10时将无法正确分离。模型信号的仿真实验中,对比分离出的信号与原信号的均方误差,说明分离算法的有效性。
3、将提取的波至时刻应用于稀疏分解原子参数离散化过程中,优化测井信号分离所用的原子库,分别分离模型信号,实际测井声波信号。实验结果表明,基于原子库优化的MP稀疏分解方法可将测井声波信号中的首波及后续波各混合信号分离出来,通过实验仿真可以看出运行时间较快,降低了搜索最优原子计算复杂度。对比基于提取波至时刻来优化原子库的MP稀疏分解算法和MP分离信号的算法。得出结论,利用本文提出的方法可有效判断混和信号个数并且减小原子库的大小,减小程序运行时间,得到分离结果。
With the development of blind signal separation technology.the technology has been widely used in image processing,seismic exploration.speech recognition, biomedical and so on. But most of the blind signal separation algorithm for multi-channel receiving mixed signals effective separation of overlapping single-channel mixed signal is seriously some of the existing BSS methods have some limitations. AM-FM mixed signal have been separated by domestic and foreign scholars in single-channel, and in the context of prior information that will separate the signal from the signal source verified by comparing the effectiveness of the algorithm. Existing methods has been studied the based on the use the wave to extract moment optimization atom library MP sparse decomposition algorithm longitudinal spread of logging in mixed-signal wave to be separated. Mixed signals are the various component signals of the frequency, amplitude and phase information. This information helps to analyze the calculation of the lithology, porosity, reservoir and other information.
Logging acoustic signal is time-frequency overlapping signals, alone in the time domain, frequency domain, spatial domain filtering are more difficult to separate, existing methods can not achieve the desired results, how to separate logging acoustic signals, extracting more useful information is our research goal. Acoustic signals for logging in the single-channel characteristics and the principle of separation of mixed signals, the atom library optimized MP sparse decomposition method is selected, the method is applied in logging acoustic signal. Main job is as follows:
1, First the blind source model is introduced, classification and separation algorithms. Then study the characteristics of logging acoustic signals, with three signal acoustic logging signal detection signal. to prepare for the next study.
2, Introduce atom library optimization method. Extraction of subsequent wave moment in the application to the atom parameter settings prior to reduce errors influence. Based on the simulation experiment of model signal, the conclusion:the extraction of wave moment need to add four sampling points, can reduce the error extract wave moment. And when a signal starting point when up to 10 times overlap degree will not be correct separated. The simulation experiment of model signal, the contrast of signal and isolated the mean square error of the original signal, mean-square explain the effectiveness of separation algorithm.
3, Put forward by optimizing the wave to extract combining with the MP algorithm, will extract wave moment applied to sparse decomposition of atomic parameter settings, then optimized atomic library. Respectively, the actual separation model acoustic signal wave logging. Experimental results show that the MP based on atomic library optimization method sparse decomposition of logging acoustic signals first affected each subsequent wave of mixed signal isolated, through the simulation can see running time quickly. Reduced the search optimal atomic computational complexity. Based on wavelet to extract contrast moments to optimize atoms library sparse decomposition algorithm and MP algorithm of separated signal. Conclusion, the proposed method can effectively and reduce the number of mixed signal judgment, reduce the size of the atomic library programs running time, get separation results.
测井声波信号是时频重叠的信号,单独在时域、频域和空域滤波都比较难分离,现有的研究方法也不能完全达到预期效果,如何分离测井声波信号,得到更多相关信息是目前研究的难题。针对测井声波信号的特点及单通道分离混和信号的原理,本论文选取利用原子库优化的MP稀疏分解方法,将该方法应用于测井声波信号分离中。主要工作如下：
1、首先简单介绍了盲信号的模型、分类和分离算法。接着研究了测井声波信号的特点,用三个信号分量叠加得到测井声波信号的检测信号,为接下来的研究做好准备。
2、详细介绍了原子库优化的方法。由于提取的后续波波至时刻存在误差,在应用到原子参数设置时需要增加波至时刻的左右采样点数。通过对模型信号的仿真实验,得出结论：提取的波至时刻需要左右增加四个采样点数,才能减小提取的波至时刻误差。并且当信号起始点时刻重叠程度达到10时将无法正确分离。模型信号的仿真实验中,对比分离出的信号与原信号的均方误差,说明分离算法的有效性。
3、将提取的波至时刻应用于稀疏分解原子参数离散化过程中,优化测井信号分离所用的原子库,分别分离模型信号,实际测井声波信号。实验结果表明,基于原子库优化的MP稀疏分解方法可将测井声波信号中的首波及后续波各混合信号分离出来,通过实验仿真可以看出运行时间较快,降低了搜索最优原子计算复杂度。对比基于提取波至时刻来优化原子库的MP稀疏分解算法和MP分离信号的算法。得出结论,利用本文提出的方法可有效判断混和信号个数并且减小原子库的大小,减小程序运行时间,得到分离结果。
With the development of blind signal separation technology.the technology has been widely used in image processing,seismic exploration.speech recognition, biomedical and so on. But most of the blind signal separation algorithm for multi-channel receiving mixed signals effective separation of overlapping single-channel mixed signal is seriously some of the existing BSS methods have some limitations. AM-FM mixed signal have been separated by domestic and foreign scholars in single-channel, and in the context of prior information that will separate the signal from the signal source verified by comparing the effectiveness of the algorithm. Existing methods has been studied the based on the use the wave to extract moment optimization atom library MP sparse decomposition algorithm longitudinal spread of logging in mixed-signal wave to be separated. Mixed signals are the various component signals of the frequency, amplitude and phase information. This information helps to analyze the calculation of the lithology, porosity, reservoir and other information.
Logging acoustic signal is time-frequency overlapping signals, alone in the time domain, frequency domain, spatial domain filtering are more difficult to separate, existing methods can not achieve the desired results, how to separate logging acoustic signals, extracting more useful information is our research goal. Acoustic signals for logging in the single-channel characteristics and the principle of separation of mixed signals, the atom library optimized MP sparse decomposition method is selected, the method is applied in logging acoustic signal. Main job is as follows:
1, First the blind source model is introduced, classification and separation algorithms. Then study the characteristics of logging acoustic signals, with three signal acoustic logging signal detection signal. to prepare for the next study.
2, Introduce atom library optimization method. Extraction of subsequent wave moment in the application to the atom parameter settings prior to reduce errors influence. Based on the simulation experiment of model signal, the conclusion:the extraction of wave moment need to add four sampling points, can reduce the error extract wave moment. And when a signal starting point when up to 10 times overlap degree will not be correct separated. The simulation experiment of model signal, the contrast of signal and isolated the mean square error of the original signal, mean-square explain the effectiveness of separation algorithm.
3, Put forward by optimizing the wave to extract combining with the MP algorithm, will extract wave moment applied to sparse decomposition of atomic parameter settings, then optimized atomic library. Respectively, the actual separation model acoustic signal wave logging. Experimental results show that the MP based on atomic library optimization method sparse decomposition of logging acoustic signals first affected each subsequent wave of mixed signal isolated, through the simulation can see running time quickly. Reduced the search optimal atomic computational complexity. Based on wavelet to extract contrast moments to optimize atoms library sparse decomposition algorithm and MP algorithm of separated signal. Conclusion, the proposed method can effectively and reduce the number of mixed signal judgment, reduce the size of the atomic library programs running time, get separation results.
引文
[1]彭耿,黄知涛,姜文利等单通道盲信号分离研究进展与展望[J].中国电子科学研究院学报,2009,4(3)：268-277.
[2]王祝文,刘菁华,聂春燕基于Hilbert-Huang变化的阵列声波测井信号时频分析[J].中国地质大学学报,2008,33(3)：387-392.
[3]胡婧,张更新等盲信号分离技术综述.数字通信世界,2010,4.
[4]马建仓,牛奕龙,陈海洋盲信号处理[M].国防工业出版社,2006.
[5]J. Herault, C. Jutten. Detection de grandeus primitives dans un message composites par une architecture de calcul neuromimetique en apprentissage non supervise Nice, Freace 1985.
[6]张发启盲信号处理及应用[M].西安电子科技大学出版社,2006.
[7]楼红伟,胡光锐基于teager能量算子和小波变换的语音识别特征参数[J]上海交通大学学报,2003,37(增刊)：83-85.
[8]Les Atlas, Christiaan Janssen Coherent Modulation Spectral Filtering for Single-channel Music Source Separation[C].ICASSP 2005,Philadelphia, PA. USA,2005,4:461-464.
[9]Alexey Ozerov,Pierrick PH Lippe.et al One Microphone Singing Voice Separation Using Source-adapted Models[C].2005,90-93.
[10]James Hopgood,Peter J W Reyner Single Channel Nonstationary Stochastic Signal Separation Using Linearr Time-vary Filters[J].2003,51(7):1739-1752.
[11]蔡权伟, 魏平, 肖先赐 单信道多信号分量分离.通信学报,2006,27(6)：49-56.
[12]沈慧芳,赖宏慧瞬时混合盲信号分离问题的自适应算法比较[A].科技广场,2009,(9)：10-12.
[13]蔡权伟,魏平,肖先赐基于能量算子的单信道重叠信号盲分离方法.中国科学杂志社,2008,38(4)：607-619.
[14]Christopher J James,D Lowe Single Channel Analysis of Electromagnetic Brain Signals Through ICA in a Dynamical Systems Framework[C]. BMBS2001,1974-1977.
[15]Ping Gao,Ee-Chien Chang,Lonce Wyse Blind Separation of Fetal ECG from Single Mixture Using SVD and ICA[C].ICICS-PCM 2003,3:1418-1422.
[16]C.J.James and D.Lowe Single Channel Analysis of Electromagnetic Brain Signals Through ICA In A Dynamical Systems Framework.2001,1974-1977.
[17]Warner E S.Proudler I K.Single-channel Blind Signal Separation of Filtered MPSK Signal[J].IEE Proceedings of Radar,Sonar&Navigation,2003, 150(6):396-402.
[18]J.Lee,K.J.Lee,S.K.Yoo Development of a new Signal Processing Algorithm based on Independent Component Analysis for Single Channel ECG Data.2004,224-226.
[19]Manuel J.Reyes-Gomez,Daniel P.W.Ellis,Nebojsa Jojic Multiband Audio Modeling for Single-channel Acoustic Source Separation[C].ICASSP 2004,Montreal,Quebec,Canada,2004,5:641-644.
[20]Bhilksha Raj,Madhusudana V.S.Shashanka,Paris Smaragdis Latent Dirichlet Decomposition for Single Channel Speaker Seeparation[C]. ICASSP 2006. Toulouse, France,2006,5:821-824.
[21]Gil-Jin Jang,Te-Won Lee Single-Channel Signal Separation Using Time-Domain Basis Functions[J].IEEE 2003,10:168-171.
[22]强琳,刘贵忠阵列声波全波列测井信号波至提取的小波变换方法[J].石油物探,1997,36(2)：55-61.
[23]江万哲,章成广时频分析方法在声波测井信息提取中的应用[J].江汉石油学院学报,2005,27(6)：736-738.
[24]楚泽涵,赵培华长源距声波测井资料分析处理[J]地球物理学报,1992.35(6)：771-779.
[25]李长文,余春昊,王文先,等阵列声波测井资料的分波处理及应用[J].测井技术,1997,21(1)：1-8.
[26]陈强,张超谟 长短时窗能量比法提取全波列测井纵横波信息[J].江汉石油学院学报,2006,28(3)：72-75.
[27]宋袆,谌海云,等 基于小波变换的偶极声波测井横波首波的提取[J].石油天然气学报,2008,30(5)：73-76.
[28]王祝文,刘菁华,聂春燕时频分析的重排方法及其在声波测井信号处理中的应用[J].吉林大学学报,2007,37(5)：1042-1046.
[29]Leon Cohn.Time-frequency analysis:theory and applications[M].Englewood Cliffs:Prentice Hall,1994.
[30]Shie Qian.Introduction to time-frequency and wavelet transforme[M]. Beijing:China Machine Press,2005.
[31]陶钧基于时频信号分析的阵列声波测井信号处理研究[D].天津大学,2009.
[32]林盛,刘业新,李衍达基于时-频分解技术的全波列声波测井信号处理[J]. 清华大学学报,1997,37(3)：63-66.
[33]M.E.Willis and M.N.Toksoz.Automatic P and S Velocity Determination from Full Waveform form Digital Acoustics Logs.Geophysics.Vol.48,No.12,1983.
[34]J. D. Ingram, C. F. Morris, E. E. MacKnight. andT. W. Parks, Directpase determination of S-waveve locities from acoustic waveform log, Geophysics, Vol.50, Nov.,1985, P.1746-1755.
[35]段竹文,马英卓,黄庆程提取全波列声波测井信号波前的一种新方法[J].哈尔滨商业大学学报,2002,18(6)：635-643.
[36]马鸿飞,李横晖,樊昌信语音分析的能量算子方法[J].西安电子科技大学学报,1997,24(1)：8-14.
[37]Teager H M.Some Observitions on Oral Air Flow During Phonation.IEEE Trans on ASSP.1980,28(5):P599-P601.
[38]Petros Maragos, Thomas Quatieri and Jame F. Kaser. Speech Nonlinearitis. Modulations and Energy Operatiors. ICASSP 91.1991:P421-P424.
[39]Kaiser J F. On a simple algorithm to calculate the energy of a signal. In: Proc ICASSP'90. Piscataway:IEEE Operation Center,1990.381—384.
[40]Petros Maragos,James F.Kaiser and Thomas Quatieri.Energy Separation in Signal Modulations with Application to Speech Analysis.IEEE Transaction on Signal Processing.1993(10):P3024-P3051.
[41]Balasubramanian Santhanam and Petros Maragos.Harmonic Analysis and Restoration of Separation Methods for Periodic Signal Mixtures.Santhanam & Maragos:submitted to signal Processing October 27.1997.
[42]王青松基于能量算子解调法的滚动轴承故障诊断技术研究[D].重庆大学,2004.
[43]Belouchrani A, Cardoso J F. Maximum likelihood source separation for discrete sources[J]. In:Proc EUSIPCO.1994.768-771.
[44]Zibulevsky M, Pearlmutter B A. Blind source separation by sparse decomposition in a signal dictionary[J]. Neural Computation.2001,13(4): 863-882.
[45]Lewicki M S, Sejnowski T J. Learning overcomplete representations [J].Neural Computation.2000.12(2):337-365.
[46]Zhang L Q, Amari S. Cichocki A. Natural gradient approach to blind separation of over-and undercomplete mixtures[J]. In:Proc ICA1999, Aussois,1999.455-460.
[47]徐尚志.苏勇.叶中付.欠定条件下的盲分离算法[J].数据采集与处理.2006,21(2)：128-132.
[48]Zhang Chunmei.Yin ZhongKe, Chen Xiangdong. Xiao Mingxia. Signal Overcomplete representation and sparse decomposition based on redundant dictionaries[J]. Chinese Science Bulletin,2005,50(23):2672-2677.
[49]李雨听,尹忠科,王建英MP稀疏分解算法及其在语音识别中的应用[J].计算机工程与应用,2010,46(1)：122-124.
[50]尹忠科,王建英,邵君基于原子结构特性的信号稀疏分解[J].西南交通大学学报,2005,40(2)：173-178.
[51]王建英,尹忠科,张春梅信号与图像的稀疏分解及初步应用[M].成都：西南交通大学出版社,2006.
[52]Mallat S, Zhang Z. Matching pursuit with time-frequency dictionaries [J]. IEEE Trans, on Signal Processing.1993,41(12):3397-3415.
[53]朱正亚,周致强全波列声波测井数据的时域处理方法[J].石油物探,1991,30(2)：76-82.
[54]Cwdric Fevotte and Simon J. Godsill.A Bayesian Approach for Blind Separation of Sparse Sources [J]. IEEE Trans. on Audio Speech.2006,14(6): 2174-2188.
[2]王祝文,刘菁华,聂春燕基于Hilbert-Huang变化的阵列声波测井信号时频分析[J].中国地质大学学报,2008,33(3)：387-392.
[3]胡婧,张更新等盲信号分离技术综述.数字通信世界,2010,4.
[4]马建仓,牛奕龙,陈海洋盲信号处理[M].国防工业出版社,2006.
[5]J. Herault, C. Jutten. Detection de grandeus primitives dans un message composites par une architecture de calcul neuromimetique en apprentissage non supervise Nice, Freace 1985.
[6]张发启盲信号处理及应用[M].西安电子科技大学出版社,2006.
[7]楼红伟,胡光锐基于teager能量算子和小波变换的语音识别特征参数[J]上海交通大学学报,2003,37(增刊)：83-85.
[8]Les Atlas, Christiaan Janssen Coherent Modulation Spectral Filtering for Single-channel Music Source Separation[C].ICASSP 2005,Philadelphia, PA. USA,2005,4:461-464.
[9]Alexey Ozerov,Pierrick PH Lippe.et al One Microphone Singing Voice Separation Using Source-adapted Models[C].2005,90-93.
[10]James Hopgood,Peter J W Reyner Single Channel Nonstationary Stochastic Signal Separation Using Linearr Time-vary Filters[J].2003,51(7):1739-1752.
[11]蔡权伟, 魏平, 肖先赐 单信道多信号分量分离.通信学报,2006,27(6)：49-56.
[12]沈慧芳,赖宏慧瞬时混合盲信号分离问题的自适应算法比较[A].科技广场,2009,(9)：10-12.
[13]蔡权伟,魏平,肖先赐基于能量算子的单信道重叠信号盲分离方法.中国科学杂志社,2008,38(4)：607-619.
[14]Christopher J James,D Lowe Single Channel Analysis of Electromagnetic Brain Signals Through ICA in a Dynamical Systems Framework[C]. BMBS2001,1974-1977.
[15]Ping Gao,Ee-Chien Chang,Lonce Wyse Blind Separation of Fetal ECG from Single Mixture Using SVD and ICA[C].ICICS-PCM 2003,3:1418-1422.
[16]C.J.James and D.Lowe Single Channel Analysis of Electromagnetic Brain Signals Through ICA In A Dynamical Systems Framework.2001,1974-1977.
[17]Warner E S.Proudler I K.Single-channel Blind Signal Separation of Filtered MPSK Signal[J].IEE Proceedings of Radar,Sonar&Navigation,2003, 150(6):396-402.
[18]J.Lee,K.J.Lee,S.K.Yoo Development of a new Signal Processing Algorithm based on Independent Component Analysis for Single Channel ECG Data.2004,224-226.
[19]Manuel J.Reyes-Gomez,Daniel P.W.Ellis,Nebojsa Jojic Multiband Audio Modeling for Single-channel Acoustic Source Separation[C].ICASSP 2004,Montreal,Quebec,Canada,2004,5:641-644.
[20]Bhilksha Raj,Madhusudana V.S.Shashanka,Paris Smaragdis Latent Dirichlet Decomposition for Single Channel Speaker Seeparation[C]. ICASSP 2006. Toulouse, France,2006,5:821-824.
[21]Gil-Jin Jang,Te-Won Lee Single-Channel Signal Separation Using Time-Domain Basis Functions[J].IEEE 2003,10:168-171.
[22]强琳,刘贵忠阵列声波全波列测井信号波至提取的小波变换方法[J].石油物探,1997,36(2)：55-61.
[23]江万哲,章成广时频分析方法在声波测井信息提取中的应用[J].江汉石油学院学报,2005,27(6)：736-738.
[24]楚泽涵,赵培华长源距声波测井资料分析处理[J]地球物理学报,1992.35(6)：771-779.
[25]李长文,余春昊,王文先,等阵列声波测井资料的分波处理及应用[J].测井技术,1997,21(1)：1-8.
[26]陈强,张超谟 长短时窗能量比法提取全波列测井纵横波信息[J].江汉石油学院学报,2006,28(3)：72-75.
[27]宋袆,谌海云,等 基于小波变换的偶极声波测井横波首波的提取[J].石油天然气学报,2008,30(5)：73-76.
[28]王祝文,刘菁华,聂春燕时频分析的重排方法及其在声波测井信号处理中的应用[J].吉林大学学报,2007,37(5)：1042-1046.
[29]Leon Cohn.Time-frequency analysis:theory and applications[M].Englewood Cliffs:Prentice Hall,1994.
[30]Shie Qian.Introduction to time-frequency and wavelet transforme[M]. Beijing:China Machine Press,2005.
[31]陶钧基于时频信号分析的阵列声波测井信号处理研究[D].天津大学,2009.
[32]林盛,刘业新,李衍达基于时-频分解技术的全波列声波测井信号处理[J]. 清华大学学报,1997,37(3)：63-66.
[33]M.E.Willis and M.N.Toksoz.Automatic P and S Velocity Determination from Full Waveform form Digital Acoustics Logs.Geophysics.Vol.48,No.12,1983.
[34]J. D. Ingram, C. F. Morris, E. E. MacKnight. andT. W. Parks, Directpase determination of S-waveve locities from acoustic waveform log, Geophysics, Vol.50, Nov.,1985, P.1746-1755.
[35]段竹文,马英卓,黄庆程提取全波列声波测井信号波前的一种新方法[J].哈尔滨商业大学学报,2002,18(6)：635-643.
[36]马鸿飞,李横晖,樊昌信语音分析的能量算子方法[J].西安电子科技大学学报,1997,24(1)：8-14.
[37]Teager H M.Some Observitions on Oral Air Flow During Phonation.IEEE Trans on ASSP.1980,28(5):P599-P601.
[38]Petros Maragos, Thomas Quatieri and Jame F. Kaser. Speech Nonlinearitis. Modulations and Energy Operatiors. ICASSP 91.1991:P421-P424.
[39]Kaiser J F. On a simple algorithm to calculate the energy of a signal. In: Proc ICASSP'90. Piscataway:IEEE Operation Center,1990.381—384.
[40]Petros Maragos,James F.Kaiser and Thomas Quatieri.Energy Separation in Signal Modulations with Application to Speech Analysis.IEEE Transaction on Signal Processing.1993(10):P3024-P3051.
[41]Balasubramanian Santhanam and Petros Maragos.Harmonic Analysis and Restoration of Separation Methods for Periodic Signal Mixtures.Santhanam & Maragos:submitted to signal Processing October 27.1997.
[42]王青松基于能量算子解调法的滚动轴承故障诊断技术研究[D].重庆大学,2004.
[43]Belouchrani A, Cardoso J F. Maximum likelihood source separation for discrete sources[J]. In:Proc EUSIPCO.1994.768-771.
[44]Zibulevsky M, Pearlmutter B A. Blind source separation by sparse decomposition in a signal dictionary[J]. Neural Computation.2001,13(4): 863-882.
[45]Lewicki M S, Sejnowski T J. Learning overcomplete representations [J].Neural Computation.2000.12(2):337-365.
[46]Zhang L Q, Amari S. Cichocki A. Natural gradient approach to blind separation of over-and undercomplete mixtures[J]. In:Proc ICA1999, Aussois,1999.455-460.
[47]徐尚志.苏勇.叶中付.欠定条件下的盲分离算法[J].数据采集与处理.2006,21(2)：128-132.
[48]Zhang Chunmei.Yin ZhongKe, Chen Xiangdong. Xiao Mingxia. Signal Overcomplete representation and sparse decomposition based on redundant dictionaries[J]. Chinese Science Bulletin,2005,50(23):2672-2677.
[49]李雨听,尹忠科,王建英MP稀疏分解算法及其在语音识别中的应用[J].计算机工程与应用,2010,46(1)：122-124.
[50]尹忠科,王建英,邵君基于原子结构特性的信号稀疏分解[J].西南交通大学学报,2005,40(2)：173-178.
[51]王建英,尹忠科,张春梅信号与图像的稀疏分解及初步应用[M].成都：西南交通大学出版社,2006.
[52]Mallat S, Zhang Z. Matching pursuit with time-frequency dictionaries [J]. IEEE Trans, on Signal Processing.1993,41(12):3397-3415.
[53]朱正亚,周致强全波列声波测井数据的时域处理方法[J].石油物探,1991,30(2)：76-82.
[54]Cwdric Fevotte and Simon J. Godsill.A Bayesian Approach for Blind Separation of Sparse Sources [J]. IEEE Trans. on Audio Speech.2006,14(6): 2174-2188.