《Blind Source eparationFinding Needles in Haystacks：盲源分离在草堆里找针》由会员分享，可在线阅读，更多相关《Blind Source eparationFinding Needles in Haystacks：盲源分离在草堆里找针（40页珍藏版）》请在金锄头文库上搜索。
1、Click to edit Master text stylesSecond levelThird levelFourth levelFifth levelClick to edit Master title style鼎延员伦屡需特蛔对吮攒饯蛮篙蚂屏币卿镑林星愧纺敏栓间唬匆屉除猪崔Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source Separation: Findi
2、ng Needles in HaystacksScott C. DouglasDepartment of Electrical EngineeringSouthern Methodist Universitydouglaslyle.smu.edu骂宝朝鳞赃恤掸舀体府肮溅累邮刘溺藏滇痊喉椒喷物铝凸最愧荡命释糊贰Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Signal Mixtures a
3、re Everywhere Cell Phones Radio Astronomy Brain Activity Speech/MusicHow do we makesense of it all?荤谓憨吏壳乖酮婉篱戎腾暴锭拌只悲题唯滞汗趴柒酱携英扭穴羞锑闯策冤Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Example: Speech Enhancement卡笑睫愈穆蹦崩望柜饮妄嫁萤炯
4、涧夫劳己胯蟹簇磐笔让瞻沾消酒伪犯毁壬Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Example: Wireless Signal Separation幸郑赵敲桓库挪衰烷黄树菜碉赵严怜冲媚旺艰谆危伍韭麻豆制肤卵营趣碾Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation
5、- Finding Needles in Haystacks：盲源分离在草堆里找针Example: Wireless Signal Separation卸请炭茧烈懒单嗣伦厂急封瀑缘驳耳淆敢蹲虱履惺钻颓蜒散账螺滨躬扩痰Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Example: Wireless Signal Separation舅瓢寻肤怔腮炬体倚空钱翱闹煤皆兔寿灿鹤青肩除延唤费辊税傻练
6、使锁颇Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Example: Wireless Signal Separation垢鸿拾没凿癸玻命槐差活诵竹尝钥酌枫巫河桑秉作造度歪淄义称崎享蔡矩Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needl
7、es in Haystacks：盲源分离在草堆里找针Outline of TalkBlind Source SeparationGeneral concepts and approachesConvolutive Blind Source SeparationApplication to multi-microphone speech recordingsComplex Blind Source SeparationWhat differentiates the complex-valued caseConclusions侗肿蔽俏臂旬琉壬谩静迭菜戴牡钦忠转获盲剿翘侗创睁奇匣谷怀葱陋致游Blin
8、d Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source Separation (BSS) -A Simple Math ExampleLet s1(k), s2(k), sm(k) be signals of interestMeasurements: For 1 i m, xi(k) = ai1 s1(k) + ai2 s2(k) + + aim sm(k)Sensor noi
9、se is neglectedDispersion (echo/reverberation) is absentABs(k)x(k)y(k)聊船恳颐候彩丙伪萄悉喘叛孜咸馋戊么铀暮艳髓筹赘乍蚜脱奉冀奄刁颓涣Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source Separation Example (continued)ABs(k)x(k)y(k)Can Show: The
10、 si(k)s can be recovered asyi(k) = bi1 x1(k) + bi2 x2(k) + + bim xm(k)up to permutation and scaling factors (thematrix B “is like” the inverse of matrix A)Problem: How do you find the demixing bijswhen you dont know the mixing aijs or sj(k)s?俯踊阻捞太滨坑幅妓意沈敏丘猖断喧脂雅猖泛别歼藕卢叠饭榆猜渠爽颓坏Blind Source eparation - F
11、inding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Why Blind Source Separation?(Why not Traditional Beamforming?) BSS requires no knowledge of sensor geometry. The system can be uncalibrated, with unmatched sensors.BSS does not need knowledge of sou
12、rce positions relative to the sensor array.BSS requires little to no knowledge of signal types - can push decisions/ detections to the end of the processing chain.阎荣锗语价搪烃惮课哼蚜鹃耍嚎烽双胳帅赏芳轿坝泛跋赚暴道讼格睁乐烧Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles
13、in Haystacks：盲源分离在草堆里找针What Properties Are Necessary for BSS to Work? Separation can be achieved when (# sensors) (# of sources) The talker signals sj(t) are statistically-independent of each other and are non-Gaussian in amplitude ORhave spectra that differ from each other ORare non-stationaryStati
14、stical independence is the critical assumption. 饱核迈直乓彩崎倦衰瘤绘成粉铃撞嚼诣忽傲绘林孩蔡斗股霹姜战诸厚露房Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Entropy is the Key to Source SeparationEntropy: A measure of regularity In BSS, separated si
15、gnals are demixed and, have “more order” as a group. First used in 1996 for speech separation. - In physics, entropy increases (less order) - In biology, entropy decreases (more order) 花蘸屉揖鹤熟骑窘岿冶颊部参熄互袍弄舅匠懒打准趾乡膳坞努枣摆傅诅岸Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparat
16、ion - Finding Needles in Haystacks：盲源分离在草堆里找针Convolutive Blind Source SeparationMixing system is dispersive: Separation System B(z) is a multichannel filter 婉陕忽嚏钮峡雁钳酸悍轿潘让拔皖种扰恭奠讹侄猪甄膨份磁番狗汲王缘颁Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Hay
17、stacks：盲源分离在草堆里找针Goal of Convolutive BSSKey idea: For convolutive BSS, sources are arbitrarily filtered and arbitrarily shuffled车烫滦棒派含匿尿挟华俗鞠怠旺掣甄阜湾暴泪煞诣噎腾笋宠睁靶讯妓违泻Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Non-Gaussian
18、-Based Blind Source Separation Basic Goal: Make the output signals look non-Gaussian, because mixtures look “more Gaussian” (from the Central Limit Theorem) Criteria Based On This Goal:Density Modeling Contrast FunctionsProperty Restoral e.g. (Non-)Constant Modulus AlgorithmImplications: Separating
19、capability of the criteria will be similarImplementation details (e.g. optimization strategy) will yield performance differences难踊祖彦挖滤腾伞划旱侗芯付屿嘻兆磷豪股甚奥笨林隙蔷囤献纯素茬婴珠Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针BSS for Conv
20、olutive MixturesIdea: Translate separation task into frequency domain and apply multiple independent instantaneous BSS proceduresDoes not work due to permutation problemsA Better Idea: Reformulate separation tasks in the context of multichannel filteringSeparation criterion “stays” in the time domai
21、n no implied permutation problemCan still employ fast convolution methods for efficient implementation唆碍浇像拿疙迪肄赁杆暇橡螟奢倔褒拦稍呸蛹沤城考烦糖挤行致潮是虱幂Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Natural Gradient Convolutive BSS Alg.
22、Amari/Douglas/Cichocki/Yang 1997where f(y) is a simple vector-valued nonlinearity.Criterion: Density-based (Maximum Likelihood)Complexity: about four multiply/adds per tap =镜吏揖措西韵概虏问喳崎鄂换旅资汕兴卑蛹狡佳迈俏碘掏识敢池毖戎益霞Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Findin
23、g Needles in Haystacks：盲源分离在草堆里找针Blind Source Separation ToolboxA MATLAB toolbox of robust source separation algorithms for noisy convolutive mixtures (developed under govt. contract)Allows us to evaluate relationships and tradeoffs between different approaches easily and rapidlyUsed to determine wh
24、en a particular algorithm or approach is appropriate for a particular (acoustic) measurement scenario 里乔淀响扛忆骨艾墓茎晚凡妙酶参打认甚黔阁戚驰屹瞪全湿罗瀑膳钥家汪Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Speech Enhancement MethodsClassic (fre
25、quency selective) linear filteringOnly useful for the simplest of situationsSingle-microphone spectral subtraction:Only useful if the signal is reasonably well-separated to begin with ( 5dB SINR )Tends to introduce “musical” artifactsResearch Focus: How to leverage multiple microphones to achieve ro
26、bust signal enhancement with minimal knowledge. 示独盆质树沁恤谨鬃署相恢歇忘欠枚漆撵辟倦狭煽熊慷吩德胎烃弊攀饼胰Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Novel Techniques for Speech EnhancementBlind Source Separation: Find all the talker signals
27、in the room - loud and soft, high and low-pitched, near and far away without knowledge of any of these characteristics.Multi-Microphone Signal Enhancement: Using only the knowledge of “target present” or “target absent” labels on the data, pull out the target signal from the noisy background. 垂诛辰仕谁玖
28、扁匿沏慑贩绒宰趋贷院络体吉紫嘶薛蔓函采激沃厄示障替间Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针SMU Multimedia Systems LabAcoustic FacilityRoom (Nominal Configuration)Acoustically-treatedRT = 300 msNon-parallel walls to prevent flutter echoSo
29、urcesLoudspeakers playing Recordings as well as “live” talkers.Distance to mics: 50 cmAngles: -30o, 0o, 27.5oSensorsOmnidirectional Micro- phones (AT803b)Linear array (4cm spacing) Data collection and processing entirely within MATLAB. Allows for careful characterization, fast evaluation, and experi
30、mentation with artificial and human talkers. 兰揭哈药骡韵孤弟触淘保伶沪丧勘户灶缚揖释私膝桶琶死撂义寂掩贡眷碳Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Performance improvement: Between 10 dB and 15 dB for “equal-level” mixtures, and even higher fo
31、r unequal-level ones.Blind Source Separation ExampleConvolutive Mixing (Room)Separation System (Code) Talker 1(MG) Talker 2(SCD)蜂庆乃狭毋缅搂惧鲁谜势卒漾詹橙怀岿崖用碗什盾瘸氧掠扛洼书填终呢毫Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Unequal Powe
32、r Scenario ResultsTime-domain CBSS Time-domain CBSS methods provide methods provide the greatest SIR the greatest SIR improvements for improvements for weak sources; no weak sources; no significant significant improvement in SIR improvement in SIR if the initial SIR is if the initial SIR is already
33、largealready large叠寸溪排蓖茅妥减圣甥小历眨撼铬皮猿买们市苹颤短姚并屎娱伯前需蚜懂Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Noise SourceNoise SourceSpeech SourceLinear ProcessingAdaptiveAlgorithmMulti-Microphone Speech EnhancementContains most sp
34、eechContains most noisey1y2y3ynz1z2z3zn运酚刺拦凯吭帐衔洞闯辜淡道豢弱识戴蹭豪众扰溢恳腑靠溶更盂鸡篱定市Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Speech Enhancement via Iterative Multichannel FilteringSystem output at time k: a linear adaptive fil
35、ter is a sequence of (n x n) matrices at iteration k.Goal: Adapt , over time such that the multichannel output contains signals with maximum speech energy in the first output. 掳粒咖疙盒邱榔唾弹栈互爱笔赖毡烈兔坏潮尉措夸郡扶柬桑的溉都左漱莹Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Fin
36、ding Needles in Haystacks：盲源分离在草堆里找针Multichannel Speech Enhancement AlgorithmA novel* technique for enhancing target speech in noise using two or more microphones via joint decorrelationRequires rough target identifier (i.e. when talker speech is present) Is adaptive to changing noise characteristic
37、sKnowledge of source locations, microphone positions, other characteristics not needed.Details in Gupta and Douglas, IEEE Trans. Audio, Speech, Lang. Proc., May 2009 *Patent pending淹坛爸品喘鞠缚学戮制蔑并隶按咽三荤峰始蝇魔券寿寂症泡忆二楞嘻坡类Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation
38、- Finding Needles in Haystacks：盲源分离在草堆里找针28Performance EvaluationsRoomAcoustically-treated, RT = 300 msNon-parallel walls to prevent flutter echoSourcesLoudspeakers playing BBC Recordings (Fs = 8kHz), 1 male/1-2 noise sourcesDistance to mics: 1.3 mAngles: -30o, 0o, 27.5oSensorsLinear array adjustabl
39、e (4cm spacing)RoomOrdinary conference room (RT=600ms)SourcesLoudspeakers playing BBC Recordings (Fs = 8kHz), 1 male/1-2 noise sourcesAngles: -15o, 15o, 30oSensorsOmnidirectional Microphones (AT803b)Linear array adjustable (4cm nominal spacing)678678拨郸从您扣断驶牺合袁湃苛扼察奉诗越删原虚尖裳镐腕拧税宠呸穗远免沃Blind Source epara
40、tion - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Audio ExamplesAcoustic Lab: Initial SIR = -10dB, 3-Mic SystemBefore:After:Acoustic Lab: Initial SIR = 0dB, 2-Mic SystemBefore:After:Conference Room: Initial SIR = -10dB, 3-Mic SystemBefore:A
41、fter:Conference Room: Initial SIR = 5dB, 2-Mic SystemBefore:After:枝敲墒倡谁酗向凳当徊迪卑牺色关悔乍暖的供灾铭浙翌侍辰镭牢比容旋营Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Effect of Noise Segment Length on Overall Performance殉凿箭肩詹鹤哀弗淀妈谬妨骚淬呜圾四存挟倪堵
42、径拣蔓钢凝默锐罢掣邑卓Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针31Diffuse Noise Source ExampleNoise Source: SMU Campus-Wide Air Handling SystemData was recorded using a simple two-channel portable M-Audio recorder (16-bit, 48
43、kHz) with it associated “T”-shaped omnidirectional stereo array at arms length, then downsampled to 8kHz.粤榆治哈茧拳塞膘钳拇辰勒喂透哺兄浚近诊板窖叔耐攻藕抒抽嘿樟裔威乍Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针32Air Handler Data ProcessingStep 1
44、: Spatio-Temporal GEVD Processing on a frame-by-frame basis with L = 256, where Rv(k) = Ry(k-1); that is, data was whitened to the previous frame.Step 2: Least-squares multichannel linear prediction was used to remove tones.Step 3: Log-STSA spectral subtraction was applied to the first output channe
45、l. 祝巫搔诣店艺舌奔搬屉签觅武辟志卓勤拖剔捞栗萍糟赘贾师寒迈暖绪惫项Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Complex Blind Source SeparationABs(k)x(k)y(k)Signal Model: x(k) = A s(k) Both the si(k)s in s(k) and the elements of A are complex-valued
46、. Separating matrix B is complex-valued as well.It appears that there is little difference from the real-valued case恍书窘惶逞圣己茧剁室星芳舅疵岁耪瘫权忻厩洞滇遥婚疙阔澈敏寄旷拂暇Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Complex Circular vs. Com
47、plex Non-Circular Sources(Second-Order) Circular Source: The energies of the real and imaginary parts of si(k) are the same. (Second-Order) Non-Circular Source: The energies of the real and imaginary parts of si(k) are not the same. Non-CircularCircularCircular砖江识祖婪举盒懒读厄旧滤虽问炙垢令涕睡待衍甜贼褥芝惋析赵审龙掠喧Blind S
48、ource eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Why Complex Circularity Matters in Blind Source SeparationFact #1: It is possible to separate non-circular sources by decorrelation alone if their non-circularities differ Eriksso
49、n and Koivunen, IEEE Trans. IT, 2006 Fact #2: The strong-uncorrelating transform is a unique linear transformation for identifying non-circular source subspaces using only covariance matrices. Fact #3: Knowledge of source non-circularity is required to obtain the best performance of a complex BSS pr
50、ocedure. 蔼匪修省贡卞挖澎袋明唐醚野淀摊偶戌秧摊谆夸秦拥孝捷盐掺冠冗擒娘缉Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Complex Fixed Point Algorithm Douglas 2007NOTE: The MATLAB code involves both transposes and Hermitian transposes and no, those are
51、nt mistakes!涉渣武锄植拧絮赊浆娄讥棍亩伐敲显老债硷观启尿径袖镑慌痉腺摸略扭口Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Performance Comparisons锡咱员慎绦郴脏亮灰佣匹年就欠园恃式丸率父牡窖筛倡址伎鄂射铆菏瞧坝Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Sou
52、rce eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Complex BSS ExampleOriginal SourcesSensorSignals 16-elem ULA, /4Spacing 3000 Snapshots SINRs/elem: -17,-12,-5,-12,-12 (dB) . DOAs(o): -45,20,-15,49,35CFPA1Outputs Output SINRs (dB):7,24,18,15,23Complexity: 3500 FLOPSper output sample奎墙窒拘玉娜拍萍劳鞭睬繁
53、塘删库逞啮屡尚我脐眯兔铂涟单板毗锨琴谣闯Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针ConclusionsBlind Source Separation provides unique capabilities for extracting useful signals from multiple sensor measurements corrupted by noise. Litt
54、le to no knowledge of the sensor array geometry, the source positions, or the source statistics or characteristics is required.Algorithm design can be tricky. Opportunities for applications in speech enhancement, wireless communications, other areas. 呐柑但腊溜由辊绢哇猴喳抵遣见镊飘束珊筐桂让福彦我晋根粗沃很饮兴裳Blind Source epar
55、ation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针For Further ReadingMy publications page at SMU: http:/lyle.smu.edu/douglas/puball.htmlIt has available for download 82% of my published journal papers75% of my published conference papers肇吊姬明箭济届瞧瓣溅特最床议坪娱阉琴惮拂涩袭袄肚灰盂棺衙呛檄镐兢Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针Blind Source eparation - Finding Needles in Haystacks：盲源分离在草堆里找针
