A Machine Learning Framework for Enhancement and Recognition of Microphone Array Speech
|
|
- Nickolas Moody
- 5 years ago
- Views:
Transcription
1 A Machie Learig Framework for Ehacemet ad Recogitio of Microphoe Array Speech Chi-Hui Lee School of ECE, Georgia Tech I collaboratio with GT ad USTC teams 1 Outlie ad Talk Ageda DSP based o learig oliear spectral regressio Paradigm shift: spectral mappig with deep learig & big data Three classical sigle-chael DSP problems (Part 1) DNN-based speech ehacemet (SE) DNN-based source or speech separatio (SS) DNN-based speech dereverberatio Extesio to far-field microphoe array speech (Part 2) Two-stage architecture for SE/SS ad robust speech recogitio Multiple sources of iterfereces i reverberat coditios Speaker-depedet ehacemet (oly five-miute traiig) Black-box LVCSR (already clea- or multi-coditio traied) Comparig multi-chael DNN architectures & performaces Summary, supplemets, refereces ad recet efforts 2 1
2 Speech i Noisy Eviromet 1. Additive oise (mathematical mixig): y t = x t + t STFT Y l, k = X l, k + N(l, k) Focused i first part 2. Covolutioal oise: y t = x t h t 3. Mixed oise: y t = x t h t + t y t = [x t + v t ] h t + t Ofte solved with mixig sigal assumptios by mathematical optimizatio i covetioal approaches!! Focused i secod part 3 Part 1: Sigle-Chael Speech Ehacemet, Separatio ad Dereverberatio -- Speactral Mappig with DNN Regressio 4 2
3 Topic 1: Speech Ehacemet (SE) Speech ehacemet: improvig the itelligibility ad/or overall perceptual quality of degraded speech sigals usig digital sigal processig (DSP) techiques Oe of the most addressed classical SP problems Issues: musical oise ad o-statioary backgrouds Ehacig Noisy speech, Exhibitio, SNR=5dB Clea speech 5 Covetioal Speech Ehacemet Classified by the umber of sigal chaels 1. Sigle chael speech ehacemet Time ad frequecy iformatio Focused i Part 1 2. Array based speech ehacemet Time, frequecy ad spatial iformatio Focused i Part 2 Covetioal Techiques: math ad physics Spectral subtractio, Wieer filterig, maskig ˆ MMSE log spectral amplitude (MMSE-LSA) Optimally modified LSA (OM-LSA) May others for sigle- ad multi-chaels SE X ( l, k) Y ( l, k) Nˆ( l, k) 6 3
4 Learig-Based Speech Ehacemet Early: HMM-based speech estimatio (Erphraim & Malah, 1984) Deep deoisig autoecoder (Lu, Tsao, Matsuda, Hori, 13) Classificatio-based separatio (Wag & Wag, 13) Noliear regressio fuctio F(.) for spectral mappig i 12 most previous DNN efforts were for classificatio-based learig X(l, k) = F Y l, k + E(l, k) What is F(.)? What parameters? How may? How to obtai a lot of the traiig pairs, (Y, X)? Ay special assumptios? Geeralizatio issues? How to estimate the parameters? How to hadle mismatched spectral magitude & phase 7 DNN-Based SE System Overview Traiig Stage Clea/Noisy Samples Feature Extractio DNN Traiig SPL, Jauary 14 (#3 i SPS sice 14) Ehacemet Stage Noisy Samples Y t Feature Extractio l Y DNN Decodig T-ASLP, Jauary 15 (#1 i SPS sice 15) ˆX l Waveform Recostructio ˆX t f Y 1. Feature extractio: log-power spectra (LPS) 2. Waveform recostructio: overlap-add (OLA) algorithm 3. Traiig: RBM pre-traiig + back-propagatio fie-tuig 4. Phase (later) 8 4
5 DNN Based Spectral Mappig: A Paradigm Shift X 1*257=257 or More 48 h3 48 h2 48 h1 11*257=2827 Y (Output with a sigle frame of clea speech features) (Iput with multiple frames of oisy speech features) W 4 W 3 + ε 3 W 2 + ε 2 W 1 + ε 1 Deep Learig Big Data: Simulatio X= F Y + E High-dim Vector-to-vector oliear regressio: millio parameters 9 Noise-Uiversal SE-DNN A Hope DNN to lear the characteristics of may oise types Classificatios: Crowdig machie trasportatio aimal ature huma, etc. alarm cry G. Hu, 100 o-speech evirometal souds,
6 Ehaced Results: No-statioary Noise A utterace with machie gu oise at SNR= -5dB: with 104- oise ehaced (upper left, PESQ=2.78), MMSE ehaced (lower left, PESQ=1.86), 4-oise ehaced (upper right, PESQ=2.14), ad oisy speech (lower right, PESQ=1.85): 104NT-DNN ehaced PESQ=2.78 Log-MMSE ehaced PESQ=1.86 4NT-DNN ehaced PESQ=2.14 oisy,machie gu, SNR=-5dB PESQ=1.85 Eve the 4NT-DNN is much tha Log-MMSE, SE-DNN is capable of suppress highly ostatioary oise. Why? More ehacemet examples ca be foud at: home.ustc.edu.c/~xuyog62/demo/se_dnn.html 11 Ehaced Results: Real-World Speech Spectrograms of a utterace extracted from the movie Forrest Gump: DNN (left), Log-MMSE (right), ad oisy (middle) with usee oise Uiversal SE-DNN ehaced Log-MMSE ehaced Noisy Good geeralizatio capacity to real-world oisy speech Publicly available tool packages GPU C++ versio: Pytho versio:
7 Topic 2: Source & Speech Separatio (SS) Source separatio aims at separatig a target speaker speech from mixed speech with iterferig speakers (oe domiatig speaker) to improve the itelligibility ad overall perceptual quality of separated speech ad possibly for ASR/SID/LID usig acoustic sigal processig techiques Ideal DNN-based speech ehacemet / source separatio More tha oe-hour traiig speech from the target speaker Mixed speech SIR=0dB Separatio Separated Target Log-MMSE based speech ehacemet 13 Comparisos: M-F Mixture (5-mi Target, No-ideal) Mix (M+F -3dB) Supervised GMM (PESQ=1.79) Semi-supervised DNN (PESQ=2.68) Usupervised CASA (PESQ=1.26) Usupervised DNN (PESQ=2.62) Target(M) Sigle-Chael SS Supervised: both speakers kow Semi-supervised: oly target kow Usupervised: both speaker ukow Multi-chael BSS: ot compared T-ASLP, August 16 T-ASLP, July 17 Semi-supervised DNN: better tha supervised GMM Usupervised DNN: better tha state-of-the-art CASA 7
8 Topic 3: Speech Dereverberatio Reverb Ehaced Reverberat speech, No pre-processig, ASR errors Issues: A lot RIR RT60 Mismatched dereverberatio, less ASR errors, worst PESQ T-ASLP, Ja 17 J-STLP, Dec 17 Good DSP will Lead to Accurate ASR Matched dereverberatio ad SE, o ASR errors, best PESQ 15 Summary: So Far 1. Spectral mappig with deep learig & big data: a paradigm shift 2. Large traiig set: learig rich regressio structure Eve simulatio data ca be very useful if properly geerated 3. For DNN-based speech ehacemet, separatio, ad dereverberatio the results are amazigly good so far Multiple sources of iterfereces: ext target Array-based ehacemet (DNN too big?) ad ASR papers, SPL: #3, T-SALP: #1 top cited paper i IEEE SPS 5. Need to combie with covetioal techiques, e.g., IRM, IME 6. A New Hope: with proper pre-processig followed by itegrated post-processig leadig to robust ASR! But for black-box ASR? Lowest errors i CHiME-2, -4, & REVERB Challeges 16 Focused i Part 2 8
9 Part 2: Sigle- ad Multi-Chael Master-Voice Separatio ad Recogitio of Array Speech -- Prelimiaries with Two-Stage Ehacemet (Balacig Temporal ad Spatial Iformatio) 17 Issues: from Sigle-Chael to Multi-Chael Speech Ukow or imprecise array cofiguratios or microphoe types Robustess to room types, target, iterferece ad array positios Room-specific or geeral room coditios: RIR ad RT60 Whither temporal or spatial iformatio: iput vectors could be too log? Backed ASR is a black box, ofte multi-coditio traied. What to optimize? This talk: livig-room or i-vehicle applicatio scearios Coditio-specific but with more complex traiig data geeratio Multiple sources of iterfereces ad additive oises with room reverberatio Speaker-idepedet (SI) pre-processig could ot deliver satisfactory performaces SD master voice pre-processig performs better but how much traiig data? I this talk: less tha 5-miute master traiig voice, ad mostly clea-traied ASR 18 9
10 Acoustic Eviromet for Array Speech Simulatio Room Size: 6m*5.5m*3m Clea speech: WSJ Noise: OSU-100, NOISEX RIR: ISM, RT60: sec Traiig SD: 30 hours (from 40 to K utteraces) Testig SD: 1800 utteraces: usee speakers ad oises Talkig distace: 1-5 meters SINR: 5, 10 ad 15 db at the receivig microphoe Referece: Microphoe #1 A example 19 SINR > 5dB: or WER may exceed 100% DoA assumed kow by wake-up or cameras Sigle-Chael Speaker-Depedet Speech Separatio Baselie DNN cofiguratio ( sigle ) 3 hidde layer: 48 uits each Iput: 257-dim LPS features Temporal cotext size: 11-frame iput The dual outputs: estimated 257-dim LPS ad 257-dim IRM (ideal ratio mask or IRM for post-processig) features Multi-target learig N 1 2 E Xˆ ( Y, W, b ) X r N 1 1 N N , 0.7, 1.25 IRM ( Y, W, b) IRM. IRM 2 e X X e e N 2 2 Y ( d), IRM ( d) ˆ X ( ) ( ( ) ˆ d Y d X ( d)) / 2, IRM ( d) Xˆ ( d ), IRM 10
11 Overall Sigle-Chael ASR Result Summary Task: 230K-word WSJ speaker-idepedet recogitio with trigram LM perplex of 141 AM: CD-DNN-HMM traied with 70 hours of WSJ clea data, 6 hidde layers with 48 uits each, iput is 40-dim FMLLR, 11-frame expasio, output is 3455 shared states Test: Nov92, 8 speakers, 4 males ad 4 females, 300 utteraces, clea WER: ~3% SD Separator traiig: with 8-speaker clea adaptatio data, 40 utteraces each WER (i %) ad WERR (i paretheses i %) at 1m rage SINR = 5dB SINR = 10dB SINR = 15dB Noisy reverberat DNN processed (70.32) (64.23) (40.79) WER (i %) ad WERR (i paretheses i %) at 3m rage SINR = 5dB SINR = 10dB SINR = 15dB Noisy reverberat DNN processed (68.77) (63.45) (36.18) DNN architecture Ehacemet with 1 DNN Traiig: radomly selected 90 out of 240 hours 3 hidde layers, 48 uits i each 8-frame Iput: 8-chael LPS cocateated Temporal cotext size i each chael: 1 Dual outputs: 257-dim LPS dim IRM of chael 1 Proposed Multi-Chael DNN-Based Speech Ehacemet DNN Architecture 1 (DNN1) 0, 0.05, 0.7, 1.25 IRM-based post-processig ( pp ) Y ( d), IRM ( d) ˆ X ( ) ( ( ) ˆ d Y d X ( d)) / 2, IRM ( d) ˆ X ( d ), IRM N 1 2 E Xˆ ( Y, W, b ) X r N 1 1 N N 1 IRM ( Y, W, b) IRM
12 Proposed Two-stage Multi-Chael DNN-Based Speech Ehacemet DNN Architecture 2 (DNN2) DNN architecture it08 Stage 1: pre-ehacemet with 4 DNNs pre pre pre ad pre hidde layers, 48 uits i each Referece chael: chael 1 10-frame Iput : 2-chael LPS cocateated Temporal cotext size i each chael: 5 Outputs: ehaced 257-dim LPS of chael 1 Stage 2: itegratio with 1 DNN 3 hidde layers, 48 uits i each Iput with 4 or 12 frames: 4 ehaced LPS cocateated with 8-chael oisy LPS Dual outputs: 257-dim LPS dim IRM 23 Proposed Two-stage Multi-Chael DNN-Based Speech Ehacemet DNN Architecture 3 (DNN3) DNN architecture it0/8 Stage 1: pre-ehacemet with 2 DNNs pre ad pre hidde layers, 48 uits i each 12-frame Iput : 4-chael LPS cocateated Temporal cotext size i each chael: 3 Outputs: ehaced 257-dim LPS of chael 1 Stage 2: itegratio with 1 DNN 3 hidde layer with 48 uits i each Iput with 2 or 10 frames: 2 ehaced LPS cocateated with 8-chael oisy LPS Dual outputs: 257-dim LPS dim IRM 24 12
13 Prelimiary Results for Architecture Comparisos WER ad PESQ results at 3m for Speaker 447 SINR/dB RT60/s Average WER (%) Average PESQ Baselie DNN1 DNN2 DNN3 Post-processig oisy ch sigle pre it it pre it it it8+pp aechoic Overall PESQ ad ASR Result Summary Task: 230K-word WSJ cotiuous speech recogitio with trigram LM perplex of 141 AM: CD-DNN-HMM traied with 70 hours of WSJ1 clea data, 6 hidde layers with 48 uits each, iput is 40-dim FMLLR trasformed MFCC, 11-frame expasio, output is 3455 seoes Test: Nov92, 8 speakers, 4 males ad 4 females, about 1800 utteraces for each speaker Separator traiig: with 8-speaker clea adaptatio data, about five miutes each Average PESQ, WER ad WERR (i paretheses i %) over all 8 speakers for systems at SINR 5-15dB, RT s, ad 1-5m 1-5m Average PESQ Average WER % (WERR) oisy ch baselie: sigle (63.10%) proposed: it (62.04%) proposed: it8+pp (3.24%) Aechoic
14 Discussio o Clea- vs. Multi-Coditio Traiig Same sigle-chael ad multi-chael speech pre-processig AM: 41-dim fbak features, 11 frames expasio (per-utt cmv), 4 hidde layers, 1024 hidde odes WER(%) Spk 447 3m clea oisy MC RIR iterfere fbak MC RIR fbak 3.69 MC fbak 3.16 Clea fbak % 8-chael ehaced % % Clea ASR: 70 hours clea data MC fbak ASR: add 90 kids of oise, SNR=0dB 5dB 10dB, 280 hours oisy data MC RIR fbak ASR: covolve clea data with the RIR of 80 degree(rt60=0.2&0.3s), add 90 kids of oise, SNR= 0, 5, 10 db; 280 hours (cotai 70 hours aechoic data) MC RIR iterfere ASR: covolve clea data with RIR (RT60=0.2&0.3s), add 74 iterferers, add 90 kids of oise, SINR= 0, 5, 10 db, still 280 h (cotai 70 hours aechoic data) Ehaced speech works better for multi-coditio model. Better ehacemet put clea model o top? 33 A Example Test Utterace 3m 447 SINR=10dB RT60=0.2s post-processig Multi-chael with pp (PESQ:3.12) Multi-chael without post-processig (pp) (PESQ:3.10) without post-processig baselie sigle-chael clea oisy 0 Sigle chael baselie (PESQ:2.45) Clea Noisy ad reverberat (PESQ:2.24)
15 Summary: Two-Stage DNN Architecture (for SD Separatio ad SI Recogitio) Achieve sigificat PESQ improvemet ad WER reductio for multi-chael DNN, also effective i hadlig sigle-chael speech (5-mi SD traiig) Assume little o array cofiguratios; ot sesitive to array geometry Propose a ew two-stage ehacemet strategy: pre-ehacemet ad itegratio combiig both temporal ad spatial iformatio i spectra Need to replace kow with estimated power for real-world applicatios Power equalizatio caused about 10% degradatio: how to reduce it? Explore other techiques, e.g., SE for black-box clea- or multi-coditio ASR? Research further ito DNN architectures for array-based processig Ivestigate robustess issues i varyig rooms, positios ad array coditios 29 Ackowledgmet 15
16 Part 3: Supplemetary Slides Simulatio, Recet Efforts ad Refereces 31 Sigle-Chael SS: Defiitios Assumig mixed speech from oe target speaker ad oe iterferig speaker (more speakers later) Supervised separatio: both speakers kow GMM-based joit ad coditioal distributios Semi-supervised separatio: oly target kow Most reasoable sceario (more speakers later) This talk: DNN-based Usupervised separatio: both speakers ukow Computatioal auditory scee aalysis (CASA) This talk: DNN-based with geder mixture detectio Blid source separatio (BSS) Usually for multi-chael source separatio 32 16
17 Source Separatio: Real vs. Ideal A few miutes versus hours of target speech for traiig: Source Separatio Challege (SSC) target of GMM output target of DNN-2 output Iterferece of DNN-2 output Target: lay white by Y 6 please Iterferece : lay red with P 2 agai 3 3 Acoustic Eviromet for Array Speech Simulatio Room Size: 6m*5.5m*3m Speaker ID (geder) Target Directio(⁰) Iterferece Directio(⁰) 440 (M) (F) (F) (M) (F) (F) (M) (M) A example For sigle-chael, the sigle mic is Ref 1 Horizotal rage to the ceter is 1m, 3 m, 5m DoA assumed kow by wake-up or cameras 34 17
18 Data Simulatio Sigle Chael (1/2) Data Source: WSJ Corpus; OSU 100-type oise set Traiig Data Geeratio (for speaker-depedet master voice separatio) Target: 4 males, 4 females; ID umber from WSJ0 corpus; 40 clea utteraces for each Iterferig speakers: 72 speakers from WSJ0 corpus Noise: radom 90 types from OSU 100-type oise Corpus Oe simulated traiig utterace: 1 target utterace target Room Impulse Respose(RIR) + 1 radom iterferig utterace iterferig RIR + 1 radom oise Aliged utterace: target utterace Direct path of RIR; Room eviromets: rage 1m, 3m, 5m; each with 2 kids RT60s of 0.2s ad 0.3s; SINR cofiguratios (received at the microphoe): SINR = 5dB, SNR = 10, 15dB; SINR = 10dB, SNR = 15dB; SINR = 15dB, SNR = db (whe SNR was too low the ASR WER ofte exceeded 100%) Normalizatio: For each traiig ad testig target utterace, the reverberat utterace power = clea referece power = oisy reverberat power (ca be relaxed later with estimatio of TR60) Traiig data: SINR=5dB : SINR=10dB : SINR=15dB = 1 : 1 : 1; also iclude 40 SINR30dB SNR 31dB utteraces without reverberatio; about 000 simulated utteraces makig a 30-hour traiig set 35 Data Simulatio Sigle Chael (2/2) Testig Data Geeratio Target: the same 8 target with traiig, about 40 usee utteraces for each from WSJ0 Corpus Iterferig speakers: usee 10 speakers from WSJ0 Corpus Noise: usee 10 kids from OSU oise Corpus Oe simulated testig utterace: 1 target utterace target RIR + 1 radom iterferig utterace iterferig RIR + 1 radom oise Same SINR pairs: SINR = 5dB, SNR = 10, 15dB; SINR = 10dB, SNR = 15dB; SINR = 15dB, SNR = db 1m 3m 5m rage, both with RT60s of 0.2s ad 0.3s Testig data: about 1800 simulated utteraces (geerated Nov92 clea test utteraces) 300 utteraces for each situatio (too low SINR ofte caused over 100% WER) SINR/dB RT60/s
19 Data Simulatio Multi-Chael Traiig Data Geeratio (for speaker-depedet master voice separatio) Usig the same data geeratio strategy 8 chaels could have 8 times (240 hours) as much data as baselie (K utteraces) But oly 3 times (90 hours, utteraces) traiig data of all chaels were used Testig Data Geeratio About 1800 simulated utteraces are received by all chaels Data Geeratio Assumptios SINR was measured at the receivig microphoe With SINR at a lower level, the WER could exceed 100% DNN-Based Ehacemet: assumig kow power of desired aechoic outputs Estimated power (from estimated RT60) gave similar PESQ ad ASR results 37 Discussio o Utterace Power Equalizatio PESQ ad WER results w/o power equalizatio at 1-3 m Relaxig kow power assumptio usig data from Speaker 447 ( pest ): givig about 10% degradatio ID 447 1m 3m PESQ WER PESQ WER it it8+pest it8+pp it8+pp+pest
20 Selected Joural Publicatios 1. Y. Xu, J. Du, L.-R. Dai ad C.-H. Lee, A Experimetal Study o Speech Ehacemet Based o Deep Neural Networks, IEEE Sigal Processig Letters, Vol. 21, No. 1, pp , Jauary Y. Xu, J. Du, L.-R. Dai ad C.-H. Lee, A Regressio Approach to Speech Ehacemet Based o Deep Neural Networks, IEEE/ACM T-ASLP., Vol. 23, No. 1, pp. 7-19, Jauary J. Du, Y. Tu, L.-R. Dai, C.-H. Lee, A Regressio Approach to Sigle-Chael Speech Separatio via High- Resolutio Deep Neural Networks, IEEE/ACM T-ASLP., Vol. 24, No. 8, pp , B. Wu, K. Li, M. Yag, C.-H. Lee, A Reverberat-Time-Aware Approach to Speech Dereverberatio Based o Deep Neural Networks, IEEE/ACM T-ASLP., Vol. 25, No. 1, pp , Jauary Y. Wag, J. Du, L.-R. Dai, C.-H. Lee, A Geder Mixture Detectio Approach to Usupervised Sigle- Chael Speech Separatio Based o Deep Neural Networks, IEEE/ACM T-ASLP., Vol. 25, No. 7, pp , July Y.-H. Tu, J. Du, Q. Wag, X. Bao, L.-R. Dai ad C.-H. Lee, A Iformatio Fusio Framework with Multi- Chael Feature Cocateatio ad Multi-Perspective System Combiatio for Deep Learig Based Robust Recogitio of Microphoe Array Speech, Computer Speech & Laguage, Vol. 46, pp , B. Wu, K. Li, F. Ge, Z. Huag, M. Yag, S. M. Siiscalchi, ad C.-H. Lee, A Ed-to-Ed Deep Learig Approach to Simultaeous Dereverberatio ad Acoustic Modelig for Robust Speech Recogitio, IEEE J- STLP, Vol. 11, Issue 8, pp , December T. Gao, J. Du, L.-R. Dai ad C.-H. Lee, A uified DNN approach to speaker-depedet simultaeous speech ehacemet ad speech separatio i low SNR eviromets, Vol. 95, pp , Speech Com., Dec B. Wu, M. Yag, K. Li, Z. Huag, M. Siiscalchi, T. Wag ad C.-H. Lee, A Reverberatio-Time-Aware Approach Leveragig Spatial Ifo for Microphoe Array Dereverberatio, EURASIP J. o Advaces i Sigal Proc, Y.-H. Lai, Y. Tsao, X. Lu, F. Che, Y.-T. Su, J. K.-C. Che, M.-J. Lie, H.-Y. Che, L. P.-H. Li ad C.-H. Lee, A Noise Classificatio Based Deep Learig Noise Reductio Approach to Improvig Speech Itelligibility for Cochlear Implat Recipiets, to appear LISTEN i Ear Workshop, ad Hearig. 07/17/ Recet Joural ad Coferece Efforts Speaker-depedet ehacemet ad separatio (Speech Comm. 12/17) Array-based dereverberatio (EURASIP JASP, 12/17) Joit SS ad AM for multi-talker speech (SPS, 18) Multi-objective learig ad esemblig for Compact SE (T-ASLP, 07/18) Geeralized Gaussia desities for regressio error modelig (T-ASLP ad IS18) Multi-task learig of LPS ad IRM for SE (Iterspeech15) DNN-based VAD: SE followed by speech detectio (Iterspeech15) SNR-progressive learig for SE (Iterspeech16) ML approach to DNN parameter estimatio for SE (Iterspeech17) Geeratig mixig oises with oise basis fuctios for SE (Iterspeech17) Iterative mask estimatio ad post-processig for Array SE (for CHiME-4, ASPSIPA17) Combiig covetioal ad DNN techiques for SE ad ASR (ICASSP18) SE for speaker diarizatio (ICASSP18 ad Iterspeech18) Two-stage ehacemet of microphoe array speech for ASR (ISCSLP18) 40
Radar emitter recognition method based on AdaBoost and decision tree Tang Xiaojing1, a, Chen Weigao1 and Zhu Weigang1 1
Advaces i Egieerig Research, volume 8 d Iteratioal Coferece o Automatio, Mechaical Cotrol ad Computatioal Egieerig (AMCCE 7) Radar emitter recogitio method based o AdaBoost ad decisio tree Tag Xiaojig,
More informationMeasurement of Equivalent Input Distortion AN 20
Measuremet of Equivalet Iput Distortio AN 2 Applicatio Note to the R&D SYSTEM Traditioal measuremets of harmoic distortio performed o loudspeakers reveal ot oly the symptoms of the oliearities but also
More informationIntelligent location of two simultaneously active acoustic emission sources: Part II
1 Itelliget locatio of two simultaeously active acoustic emissio sources: Part II Tadej Kosel ad Igor Grabec Faculty of Mechaical Egieerig, Uiversity of Ljubljaa, Aškerčeva 6, POB 394, SI-11 Ljubljaa,
More informationA New Space-Repetition Code Based on One Bit Feedback Compared to Alamouti Space-Time Code
Proceedigs of the 4th WSEAS It. Coferece o Electromagetics, Wireless ad Optical Commuicatios, Veice, Italy, November 0-, 006 107 A New Space-Repetitio Code Based o Oe Bit Feedback Compared to Alamouti
More informationCHAPTER 5 A NEAR-LOSSLESS RUN-LENGTH CODER
95 CHAPTER 5 A NEAR-LOSSLESS RUN-LENGTH CODER 5.1 GENERAL Ru-legth codig is a lossless image compressio techique, which produces modest compressio ratios. Oe way of icreasig the compressio ratio of a ru-legth
More informationNonlinear System Identification Based on Reduced Complexity Volterra Models Guodong Jin1,a* and Libin Lu1,b
6th Iteratioal Coferece o Electroics, Mechaics, Culture ad Medicie (EMCM 205) Noliear System Idetificatio Based o Reduced Complexity Volterra Models Guodog Ji,a* ad Libi Lu,b Xi a research istitute of
More informationSPEECH ENHANCEMENT USING A MODULATION DOMAIN KALMAN FILTER POST-PROCESSOR WITH A GAUSSIAN MIXTURE NOISE MODEL. Yu Wang and Mike Brookes
14 IEEE Iteratioal Coferece o Acoustic, Speech ad Sigal Processig (ICASSP) SPEECH ENHANCEMENT USING A MODULATION DOMAIN KALMAN FILTER POST-PROCESSOR WITH A GAUSSIAN MIXTURE NOISE MODEL Yu Wag ad Mike Brookes
More informationA Novel Method of Artificial Bandwidth Extension Using Deep Architecture
INTERSPEECH 2015 A Novel ethod of Artificial Badwidth Extesio Usig Deep Architecture Bi Liu, Jiahua Tao, Zhegqi We, Ya Li, Daish Bukhari Natioal Laboratory of Patter Recogitio, Istitute of Automatio, Chiese
More informationAdaptive Multichannel Dereverberation for Automatic Speech Recognition
ITERSPEECH 07 August 0 4, 07, Stockholm, Swede Adaptive Multichael Dereverberatio for Automatic Speech Recogitio Joe Caroselli, Izhak Shafra, Aru arayaa, Richard Rose Google, Ic. {jcarosel, izhak, arut,
More informationOVERCOMPLETE BLIND SOURCE SEPARATION BY COMBINING ICA AND BINARY TIME-FREQUENCY MASKING
OVERCOMPLETE BLIND SOURCE SEPARATION BY COMBINING ICA AND BINARY TIME-FREQUENCY MASKING Michael Syskid Pederse 1,2, DeLiag Wag 3, Ja Larse 1 ad Ulrik Kjems 2 1 Iformatics ad Mathematical Modellig, Techical
More informationAkinwaJe, A.T., IbharaJu, F.T. and Arogundade, 0.1'. Department of Computer Sciences University of Agriculture, Abeokuta, Nigeria
COMPARATIVE ANALYSIS OF ARTIFICIAL NEURAL NETWORK'S BACK PROPAGATION ALGORITHM TO STATISTICAL LEAST SQURE METHOD IN SECURITY PREDICTION USING NIGERIAN STOCK EXCHANGE MARKET AkiwaJe, A.T., IbharaJu, F.T.
More informationPERFORMANCE OF MULTIBEAM MIMO FOR NLOS MILLIMETER WAVE INDOOR COMMUNICATION SYSTEMS CHAPTER 6
PERFORMANCE OF MULTIBEAM MIMO FOR NLOS MILLIMETER WAVE INDOOR COMMUNICATION SYSTEMS CAPTER 6 1 PERFORMANCE OF MULTIBEAM MIMO FOR NLOS MILLIMETER WAVE INDOOR COMMUNICATION SYSTEMS CAPTER 6 igh data rate
More informationSHORT-TERM TRAVEL TIME PREDICTION USING A NEURAL NETWORK
SHORT-TERM TRAVEL TIME PREDICTION USING A NEURAL NETWORK Giovai Huiske ad Eric va Berkum Dept. of Civil Egieerig - Uiversity of Twete - 7500 AE Eschede - The Netherlads E-mail: g.huiske@ctw.utwete.l ad
More informationMulti-Carrier Transmission over Mobile Radio Channels. Jean-Paul M.G. Linnartz Philips Research and TU/e
Multi-Carrier Trasmissio over Mobile Radio Chaels Jea-Paul M.G. Liartz Philips Research ad TU/e Outlie Itroductio to OFDM Discussio of receivers for OFDM ad MC-CDMA Itercarrier Iterferece, FFT Leakage
More informationJOINT NOISE AND MASK AWARE TRAINING FOR DNN-BASED SPEECH ENHANCEMENT WITH SUB-BAND FEATURES
JOINT NOISE AND MASK AWARE TRAINING FOR DNN-BASED SPEECH ENHANCEMENT WITH SUB-BAND FEATURES Qing Wang 1, Jun Du 1, Li-Rong Dai 1, Chin-Hui Lee 2 1 University of Science and Technology of China, P. R. China
More informationMid-level representations for audio content analysis *Slides for this lecture were created by Anssi Klapuri
Mid-level represetatios for audio cotet aalysis *Slides for this lecture were created by Assi Klapuri Represetatios 1 Sources: Ellis, Rosethal, Mid-level represetatios for Computatioal Auditory Scee Aalysis,
More informationEncode Decode Sample Quantize [ ] [ ]
Referece Audio Sigal Processig I Shyh-Kag Jeg Departmet of Electrical Egieerig/ Graduate Istitute of Commuicatio Egieerig M. Bosi ad R. E. Goldberg, Itroductio to Digital Audio Codig ad Stadards, Kluwer
More informationOutline. Motivation. Analog Functional Testing in Mixed-Signal Systems. Motivation and Background. Built-In Self-Test Architecture
Aalog Fuctioal Testig i Mixed-Sigal s Jie Qi Dept. of Electrical & Computer Egieerig Aubur Uiversity Co-Advisors: Charles Stroud ad Foster Dai Outlie Motivatio ad Backgroud Built-I Self-Test Architecture
More informationDIGITALLY TUNED SINUSOIDAL OSCILLATOR USING MULTIPLE- OUTPUT CURRENT OPERATIONAL AMPLIFIER FOR APPLICATIONS IN HIGH STABLE ACOUSTICAL GENERATORS
Molecular ad Quatum Acoustics vol. 7, (6) 95 DGTALL TUNED SNUSODAL OSCLLATOR USNG MULTPLE- OUTPUT CURRENT OPERATONAL AMPLFER FOR APPLCATONS N HGH STABLE ACOUSTCAL GENERATORS Lesław TOPÓR-KAMŃSK Faculty
More informationIntegrated Detection Method of Underwater Acoustic Fuze Based on IEMD, VIFD and ED
Advaces i Applied Acoustics Vol. Iss., November Itegrated Detectio Method of Uderwater Acoustic Fuze Based o IEMD, VIFD ad ED Qiao Hu Sciece ad Techology o Uderwater Iformatio ad Cotrol Laboratory, Xi
More informationFault Diagnosis in Rolling Element Bearing Using Filtered Vibration and Acoustic Signal
Volume 8 o. 8 208, 95-02 ISS: 3-8080 (prited versio); ISS: 34-3395 (o-lie versio) url: http://www.ijpam.eu ijpam.eu Fault Diagosis i Rollig Elemet Usig Filtered Vibratio ad Acoustic Sigal Sudarsa Sahoo,
More informationSummary of Random Variable Concepts April 19, 2000
Summary of Radom Variable Cocepts April 9, 2000 his is a list of importat cocepts we have covered, rather tha a review that derives or explais them. he first ad primary viewpoit: A radom process is a idexed
More informationAn Adaptive Image Denoising Method based on Thresholding
A Adaptive Image Deoisig Method based o Thresholdig HARI OM AND MANTOSH BISWAS Departmet of Computer Sciece & Egieerig Idia School of Mies, Dhabad Jharkad-86004 INDIA {hariom4idia, matoshb}@gmail.com Abstract
More informationA Novel Small Signal Power Line Quality Measurement System
IMTC 3 - Istrumetatio ad Measuremet Techology Coferece Vail, CO, USA, - May 3 A ovel Small Sigal Power Lie Quality Measuremet System Paul B. Crilly, Erik Leadro Boaldi, Levy Ely de Lacarda de Oliveira,
More informationVoice Command Recognition System Based on MFCC and VQ Algorithms
Voice Commad Recogitio System Based o MFCC ad VQ Algorithms Mahdi Shaeh, ad Azizollah Taheri Abstract The goal of this project is to desig a system to recogitio voice commads. Most of voice recogitio systems
More informationFingerprint Classification Based on Directional Image Constructed Using Wavelet Transform Domains
7 Figerprit Classificatio Based o Directioal Image Costructed Usig Wavelet Trasform Domais Musa Mohd Mokji, Syed Abd. Rahma Syed Abu Bakar, Zuwairie Ibrahim 3 Departmet of Microelectroic ad Computer Egieerig
More informationAPPLICATION NOTE UNDERSTANDING EFFECTIVE BITS
APPLICATION NOTE AN95091 INTRODUCTION UNDERSTANDING EFFECTIVE BITS Toy Girard, Sigatec, Desig ad Applicatios Egieer Oe criteria ofte used to evaluate a Aalog to Digital Coverter (ADC) or data acquisitio
More informationTehrani N Journal of Scientific and Engineering Research, 2018, 5(7):1-7
Available olie www.jsaer.com, 2018, 5(7):1-7 Research Article ISSN: 2394-2630 CODEN(USA): JSERBR 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38
More informationRecovering low frequencies for impedance inversion by frequency domain deconvolution
Recoverig low frequecies for impedace iversio by frequecy domai decovolutio Sia. Esmaeili*, CREWES, Uiversity of Calgary, sesmaeil@ucalgary.ca Gary. Frak. Margrave, CREWES, Uiversity of Calgary, margrave@ucalgary.ca
More informationA New Design of Log-Periodic Dipole Array (LPDA) Antenna
Joural of Commuicatio Egieerig, Vol., No., Ja.-Jue 0 67 A New Desig of Log-Periodic Dipole Array (LPDA) Atea Javad Ghalibafa, Seyed Mohammad Hashemi, ad Seyed Hassa Sedighy Departmet of Electrical Egieerig,
More informationSELEX Elsag. 5/18/2012 R. Pucci SDR 12 WinnComm 1
SELEX Elsag 5/18/01 R. Pucci SDR 1 WiComm 1 Military BU - SELEX Elsag Possible update of SDR Platforms to COGNITIVE architectures COGNITIVE MANAGER INTERFACE Geolocatio, Voice, Video, etc Applicatio Policy
More informationSpeech Enhancement Using Spectral Subtraction and Cascaded-Median Based Noise Estimation for Hearing Impaired Listeners
Speech Ehacemet Usig Spectral Subtractio ad Cascaded-Media Based Noise Estimatio for Hearig Impaired Listeers Satosh K. Waddi, Prem C. Padey, ad Nitya Tiwari Departmet of Electrical Egieerig Idia Istitute
More informationCFAR DETECTION IN MIMO RADARS USING FUZZY FUSION RULES IN HOMOGENEOUS BACKGROUND
CFAR DETECTION IN MIMO RADARS USING FUZZY FUSION RULES IN HOMOGENEOUS BACKGROUND Faycal Khaldi 1 ad Faouzi Soltai 2 1,2 Départemet d électroique, Uiversité des Frères Metouri Costatie Costatie 25, Algeria
More informationDesign of FPGA- Based SPWM Single Phase Full-Bridge Inverter
Desig of FPGA- Based SPWM Sigle Phase Full-Bridge Iverter Afarulrazi Abu Bakar 1, *,Md Zarafi Ahmad 1 ad Farrah Salwai Abdullah 1 1 Faculty of Electrical ad Electroic Egieerig, UTHM *Email:afarul@uthm.edu.my
More informationLETTER A Novel Adaptive Channel Estimation Scheme for DS-CDMA
1274 LETTER A Novel Adaptive Chael Estimatio Scheme for DS-CDMA Che HE a), Member ad Xiao-xiag LI, Nomember SUMMARY This paper proposes a adaptive chael estimatio scheme, which uses differet movig average
More informationA Data Field method for speech enhancement incorporating Binary Time-Frequency Masking
Jiaju HUANG Yafei ZHANG Xiogwei ZHANG Tao ZHU Istitute of Commad Automatio PLA Uiversity of Sciece ad Techology. A Data Field method for speech ehacemet icorporatig Biary Time-Frequecy Maskig Abstract.
More informationHigh-resolution speech signal reconstruction in Wireless Sensor Networks
High-resolutio speech sigal recostructio i Wireless Sesor Networks Adria Pazarloglou, Radu Stoleru, Ricardo Gutierrez-Osua Departmet of Computer Sciece, Texas A&M Uiversity {adria, stoleru, rgutier}@cs.tamu.edu
More informationChapter 12 Sound Waves. We study the properties and detection of a particular type of wave sound waves.
Chapter 2 Soud Waves We study the properties ad detectio o a particular type o wave soud waves. A speaker geerates soud. The desity o the air chages as the wave propagates. Notice that the displacemet
More informationApplication of Improved Genetic Algorithm to Two-side Assembly Line Balancing
206 3 rd Iteratioal Coferece o Mechaical, Idustrial, ad Maufacturig Egieerig (MIME 206) ISBN: 978--60595-33-7 Applicatio of Improved Geetic Algorithm to Two-side Assembly Lie Balacig Ximi Zhag, Qia Wag,
More informationA Simplified Method for Phase Noise Calculation
Poster: T-18 Simplified Method for Phase Noise Calculatio Massoud Tohidia, li Fotowat hmady* ad Mahmoud Kamarei Uiversity of Tehra, *Sharif Uiversity of Techology, Tehra, Ira Outlie Itroductio Prelimiary
More informationTotal Harmonics Distortion Reduction Using Adaptive, Weiner, and Kalman Filters
Wester Michiga Uiversity ScholarWorks at WMU Master's Theses Graduate College 6-2016 Total Harmoics Distortio Reductio Usig Adaptive, Weier, ad Kalma Filters Liqaa Alhafadhi Wester Michiga Uiversity, liquaa.alhafadhi@yahoo.com
More informationSPEECH ENHANCEMENT USING AN MMSE SPECTRAL AMPLITUDE ESTIMATOR BASED ON A MODULATION DOMAIN KALMAN FILTER WITH A GAMMA PRIOR
SPEECH ENHANCEMENT USING AN MMSE SPECTRAL AMPLITUDE ESTIMATOR BASED ON A MODULATION DOMAIN KALMAN FILTER WITH A GAMMA PRIOR Yu Wag? Mike Brookes? Egieerig Departmet Uiversity of Cambridge Uited Kigdom
More informationMeasurements of the Communications Environment in Medium Voltage Power Distribution Lines for Wide-Band Power Line Communications
Measuremets of the Commuicatios viromet i Medium Voltage Power Distributio Lies for Wide-Bad Power Lie Commuicatios Jae-Jo Lee *,Seug-Ji Choi *,Hui-Myoug Oh *, Wo-Tae Lee *, Kwa-Ho Kim * ad Dae-Youg Lee
More informationThe USTC System for Blizzard Challenge 2014
The USTC for Blizzard Challege 2014 Lig-Hui Che, Zhe-Hua Lig, Yi-Qig Zu, Ru-Qiag Ya, Yua Jiag, Xia-Ju Xia, Yig Wag, Natioal Egieerig Laboratory for Speech ad Laguage Iformatio Processig, Uiversity of Sciece
More informationObjectives. Some Basic Terms. Analog and Digital Signals. Analog-to-digital conversion. Parameters of ADC process: Related terms
Objectives. A brief review of some basic, related terms 2. Aalog to digital coversio 3. Amplitude resolutio 4. Temporal resolutio 5. Measuremet error Some Basic Terms Error differece betwee a computed
More information信號與系統 Signals and Systems
Sprig 24 信號與系統 Sigals ad Systems Chapter SS- Sigals ad Systems Feg-Li Lia NTU-EE Feb4 Ju4 Figures ad images used i these lecture otes are adopted from Sigals & Systems by Ala V. Oppeheim ad Ala S. Willsky,
More informationAdaptive MMSE Rake-Equalizer Receiver Design with Channel Estimation for DS-UWB System
Adaptive MMSE Rake-Equalizer Receiver Desig with Chael Estimatio for DS-UWB System Departmet of Electric Egieerig, Diwa Uiversity No.87-1, Nashi Li, Madou Tow, Taia Couty 7153, Taiwa, R.O.C. cyj@dwu.edu.tw
More informationLecture 14. Design of audio WM Model of digital audio CO ( WAV-format): - dependent samples with frequency 44,1кHz, - amplitude of samples is
Lecture 14. Desig of audio WM Model of digital audio CO WAV-format: - depedet samples wit frequecy 44,1кHz, - amplitude of samples is quatized o 2 16 =65536 levels 16 bits Remark 1. Tere exists a lossy
More informationGeneral Model :Algorithms in the Real World. Applications. Block Codes
Geeral Model 5-853:Algorithms i the Real World Error Correctig Codes I Overview Hammig Codes Liear Codes 5-853 Page message (m) coder codeword (c) oisy chael decoder codeword (c ) message or error Errors
More informationFinancial Time Series Forecasting Using Hybrid Wavelet-Neural Model
50 The Iteratioal Arab Joural of Iformatio Techology, Vol. 5, No., Jauary 208 Fiacial Time Series Forecastig Usig Hybrid Wavelet-Neural Model Jovaa Božić ad Djordje Babić School of Computig, Uiversity
More informationDesign of FPGA Based SPWM Single Phase Inverter
Proceedigs of MUCEET2009 Malaysia Techical Uiversities Coferece o Egieerig ad Techology Jue 20-22, 2009, MS Garde,Kuata, Pahag, Malaysia MUCEET2009 Desig of FPGA Based SPWM Sigle Phase Iverter Afarulrazi
More informationImage Contrast Enhancement Using Histogram Modification Technique
Image Cotrast Ehacemet Usig Histogram Modificatio Techique Shekhar R. Suralkar 1, Atul H. Karode 2, Maali S. Rathi 3 1 Asso. Prof. (E&TC Dept.), SSBT s COET, Jalgao, Idia 2 Asst. Prof. (E&TC Dept.), SSBT
More informationE X P E R I M E N T 13
E X P E R I M E N T 13 Stadig Waves o a Strig Produced by the Physics Staff at Colli College Copyright Colli College Physics Departmet. All Rights Reserved. Uiversity Physics, Exp 13: Stadig Waves o a
More informationBeam Pattern Scanning (BPS) versus Space-Time Block Coding (STBC) and Space-Time Trellis Coding (STTC)
It. J. Commuicatios, Network ad System Scieces, 2009, 6, 469-479 doi:10.4236/ijcs.2009.26051 Published Olie September 2009 (http://www.scirp.org/joural/ijcs/). 469 Beam Patter Scaig (BPS) versus Space-Time
More informationINCREASE OF STRAIN GAGE OUTPUT VOLTAGE SIGNALS ACCURACY USING VIRTUAL INSTRUMENT WITH HARMONIC EXCITATION
XIX IMEKO World Cogress Fudametal ad Applied Metrology September 6, 9, Lisbo, Portugal INCREASE OF STRAIN GAGE OUTPUT VOLTAGE SIGNALS ACCURACY USING VIRTUAL INSTRUMENT WITH HARMONIC EXCITATION Dalibor
More informationIntroduction to Markov Models
Itroductio to Markov Models But first: A few prelimiaries o text preprocessig Estimatig the probability of phrases of words, seteces, etc. What couts as a word? A tricky questio. How to fid Seteces?? CIS
More informationApplication of Image Fusion to Wireless Image Transmission
Applicatio of Image Fusio to Wireless Image Trasmissio Liae C. Ramac ad Pramod K. Varshey EECS Departmet, Syracuse Uiversity 121 Lik Hall, Syracuse, NY 13244 Email: lcramac@mailbox.syr.edu, varshey@cat.syr.edu
More informationPROJECT #2 GENERIC ROBOT SIMULATOR
Uiversity of Missouri-Columbia Departmet of Electrical ad Computer Egieerig ECE 7330 Itroductio to Mechatroics ad Robotic Visio Fall, 2010 PROJECT #2 GENERIC ROBOT SIMULATOR Luis Alberto Rivera Estrada
More informationCombined Scheme for Fast PN Code Acquisition
13 th Iteratioal Coferece o AEROSPACE SCIENCES & AVIATION TECHNOLOGY, ASAT- 13, May 6 8, 009, E-Mail: asat@mtc.edu.eg Military Techical College, Kobry Elkobbah, Cairo, Egypt Tel : +(0) 4059 4036138, Fax:
More informationCancellation of Multiuser Interference due to Carrier Frequency Offsets in Uplink OFDMA
Cacellatio of Multiuser Iterferece due to Carrier Frequecy Offsets i Upli OFDMA S. Maohar, V. Tiiya, D. Sreedhar, ad A. Chocaligam Departmet of ECE, Idia Istitute of Sciece, Bagalore 56001, INDIA Abstract
More informationAnalysis of SDR GNSS Using MATLAB
Iteratioal Joural of Computer Techology ad Electroics Egieerig (IJCTEE) Volume 5, Issue 3, Jue 2015 Aalysis of SDR GNSS Usig MATLAB Abstract This paper explais a software defied radio global avigatio satellite
More informationMEASUREMENT AND CONTORL OF TOTAL HARMONIC DISTORTION IN FREQUENCY RANGE 0,02-10KHZ.
ELECTRONICS 00 September, Sozopol, BLGARIA MEASREMENT AND CONTORL OF TOTAL HARMONIC DISTORTION IN FREQENCY RANGE 0,0-0KHZ. Plame Agelov Agelov Faculty for Computer Sciece, Egieerig ad Natural Studies,
More informationA Fuzzy Rule Base Assisted Adaptive Coding and Modulation Scheme for OFDM Systems
2012, TextRoad Publicatio ISSN 2090-4304 Joural of Basic ad Applied Scietific Research www.textroad.com A Fuzzy Rule Base Assisted Adaptive Codig ad Modulatio Scheme for OFDM Systems Atta-ur-Rahma a,c,*,
More information信號與系統 Signals and Systems
Sprig 2 信號與系統 Sigals ad Systems Chapter SS- Sigals ad Systems Feg-Li Lia NTU-EE Feb Ju Figures ad images used i these lecture otes are adopted from Sigals & Systems by Ala V. Oppeheim ad Ala S. Willsky,
More informationCalibrating Car-following Model with Trajectory Data by Cell Phone
Calibratig Car-followig Model with Trajectory Data by Cell Phoe Jiag Zeyu 1 jiagzy15@mails.tsighua.edu.c Wag Yog 3 13006373891@qq.com Wag Gagqiao 1 wag-gq15@mails.tsighua.edu.c Zhag Xiagpeg 3 364386058@qq.com
More informationGENERATE AND MEASURE STANDING SOUND WAVES IN KUNDT S TUBE.
Acoustics Wavelegth ad speed of soud Speed of Soud i Air GENERATE AND MEASURE STANDING SOUND WAVES IN KUNDT S TUBE. Geerate stadig waves i Kudt s tube with both eds closed off. Measure the fudametal frequecy
More informationCDS 270-2: Lecture 6-3 Optimum Receiver Design for Estimation over Wireless Links
CDS 70-: Lecture 6-3 Otimum Receiver Desig for stimatio over Wireless Lis Goals: Yasami Mostofi May 5, 006 To uderstad imact of wireless commuicatio imairmets o estimatio over wireless To lear o-traditioal
More informationSingle Bit DACs in a Nutshell. Part I DAC Basics
Sigle Bit DACs i a Nutshell Part I DAC Basics By Dave Va Ess, Pricipal Applicatio Egieer, Cypress Semicoductor May embedded applicatios require geeratig aalog outputs uder digital cotrol. It may be a DC
More informationOptimal Geolocation Updating for Location Aware Service Provisioning in Wireless Networks
Optimal Geolocatio Updatig for Locatio Aware Service Provisioig i Wireless Networks Siri Tekiay Amer Catovic tekiay@adm.jit.edu axc4466@jit.edu New Jersey Istitute of Techology Uiversity Heights, Newark,
More informationECE 333: Introduction to Communication Networks Fall Lecture 4: Physical layer II
ECE 333: Itroductio to Commuicatio Networks Fall 22 Lecture : Physical layer II Impairmets - distortio, oise Fudametal limits Examples Notes: his lecture cotiues the discussio of the physical layer. Recall,
More informationJoint Power Allocation and Beamforming for Cooperative Networks
It. J. Commuicatios, etwork ad System Scieces,, 4, 447-45 doi:.436/ijcs..4753 Published Olie July (http://www.scirp.org/joural/ijcs) Joit Power Allocatio ad Beamformig for Cooperative etworks Sodes Maadi,,
More informationAC : USING ELLIPTIC INTEGRALS AND FUNCTIONS TO STUDY LARGE-AMPLITUDE OSCILLATIONS OF A PENDULUM
AC 007-7: USING ELLIPTIC INTEGRALS AND FUNCTIONS TO STUDY LARGE-AMPLITUDE OSCILLATIONS OF A PENDULUM Josue Njock-Libii, Idiaa Uiversity-Purdue Uiversity-Fort Waye Josué Njock Libii is Associate Professor
More informationIllumination Compensation for 2-D Barcode Recognition Basing Morphologic
Research Joural of Applied Scieces, Egieerig ad Techology 5(12): 3273-3280, 2013 ISSN: 2040-7459; e-issn: 2040-7467 Maxwell Scietific Orgaizatio, 2013 Submitted: March, 09, 2012 Accepted: August 15, 2012
More informationThe Detection of Abrupt Changes in Fatigue Data by Using Cumulative Sum (CUSUM) Method
Proceedigs of the th WSEAS Iteratioal Coferece o APPLIED ad THEORETICAL MECHANICS (MECHANICS '8) The Detectio of Abrupt Chages i Fatigue Data by Usig Cumulative Sum (CUSUM) Method Z. M. NOPIAH, M.N.BAHARIN,
More informationImage Contrast Enhancement based Sub-histogram Equalization Technique without Over-equalization Noise
World Academy of Sciece, Egieerig ad Techology 5 9 Image Cotrast Ehacemet based Sub-histogram Equalizatio Techique without Over-equalizatio Noise Hyusup Yoo, Yougjoo Ha, ad Hersoo Hah Abstract I order
More informationIterative Frequency-Domain. Modulations. Rui Dinis
Iterative Frequecy-Domai Detectio for Sigle-Carrier Modulatios Rui Diis 1 Summary 1 Cotext ad motivatio 2 OFDM ad SC-FDE modulatios 3 IB-DFEtechiques 4 Applicatios MIMO systems Network diversity schemes
More informationPulse-echo Ultrasonic NDE of Adhesive Bonds in Automotive Assembly
ECNDT 6 - Poster 7 Pulse-echo Ultrasoic NDE of Adhesive Bods i Automotive Assembly Roma Gr. MAEV, Sergey TITOV, Uiversity of Widsor, Widsor, Caada Abstract. Recetly, adhesive bodig techology has begu to
More informationMarkov Modulated Punctured Autoregressive Processes for Traffic and Channel Modeling *
Markov Modulated Puctured Autoregressive Processes for Traffic ad Chael Modelig * Trista Pei-chu Che ad Tsuha Che + Departmet of Electrical ad Computer Egieerig Caregie Mello Uiversity, PA 1513 {peichu,
More informationProblem of calculating time delay between pulse arrivals
America Joural of Egieerig Research (AJER) 5 America Joural of Egieerig Research (AJER) e-issn: 3-847 p-issn : 3-936 Volume-4, Issue-4, pp-3-4 www.ajer.org Research Paper Problem of calculatig time delay
More informationA New Basic Unit for Cascaded Multilevel Inverters with the Capability of Reducing the Number of Switches
Joural of Power Electroics, ol, o, pp 67-677, July 67 JPE --6 http://dxdoiorg/6/jpe67 I(Prit: 98-9 / I(Olie: 9-78 A ew Basic Uit for Cascaded Multi Iverters with the Capability of Reducig the umber of
More informationECONOMIC LOT SCHEDULING
ECONOMIC LOT SCHEDULING JS, FFS ad ELS Job Shop (JS) - Each ob ca be differet from others - Make to order, low volume - Each ob has its ow sequece Fleible Flow Shop (FFS) - Limited umber of product types
More informationWhat Makes Audio Event Detection Harder than Classification?
What Makes Audio Evet Detectio Harder tha Classificatio? arxiv:1612.09089v4 [cs.sd] 17 May 2018 Huy Pha, Philipp Koch, Fabrice Katzberg, Marco Maass, Radoslaw Mazur, Ia McLoughli ad Alfred Mertis Istitute
More informationAn Improved SIC based Turbo Equalizer
Mauscript received July 5, 7; revised Oct. 6, 7 A Improved SIC based Turbo ualizer ARUNA TRIPATHY, SANT SARAN PATHAK, SASWAT CHAKRABARTI 3 Departmet of Electroics ad Electrical Commuicatio Egieerig GS
More informationUsing Color Histograms to Recognize People in Real Time Visual Surveillance
Usig Color Histograms to Recogize People i Real Time Visual Surveillace DANIEL WOJTASZEK, ROBERT LAGANIERE S.I.T.E. Uiversity of Ottawa, Ottawa, Otario CANADA daielw@site.uottawa.ca, lagaier@site.uottawa.ca
More informationImage Contrast Enhancement based Sub-histogram Equalization Technique without Over-equalization Noise
World Academy of Sciece, Egieerig ad Techology Iteratioal Joural of Electrical ad Computer Egieerig Image Cotrast Ehacemet based Sub-histogram Equalizatio Techique without Over-equalizatio Noise Hyusup
More informationNoncoherent Detection of Satellite-Based AIS Signal with Phase Offset Compensation
Sesors & Trasducers Vol. 166 Issue 3 March 2014 pp. 128-134 Sesors & Trasducers 2014 by IFSA Publishig S. L. http://www.sesorsportal.com Nocoheret Detectio of Satellite-Based AIS Sigal with Phase Offset
More informationSpectral and power efficiency trade-offs for multiple-antenna reception techniques under co-channel interference
EUROPEAN TRANSACTIONS ON TELECOMMUNICATIONS Eur. Tras. Telecomms. ; :5 7 Published olie i Wiley Olie Library (wileyolielibrary.com). DOI:./ett.535 RESEARCH ARTICLE Spectral ad power efficiecy trade-offs
More informationSymbol Error Rate Evaluation for OFDM Systems with MPSK Modulation
Symbol Error Rate Evaluatio for OFDM Systems with MPS Modulatio Yuhog Wag ad Xiao-Pig Zhag Departmet of Electrical ad Computer Egieerig, Ryerso Uiversity 35 Victoria Street, Toroto, Otario, Caada, M5B
More informationStudying for the Enhancement Captured Images at Different Lightness Directions and Levels Distribution Based on Using Histogram Equalization Method
Al-Saleh et.al. Iraqi Joural of Sciece, December, Vol. 53, No. 4, Pp. -7 Studyig for the Ehacemet Captured Images at Differet Lightess Directios ad Levels Distributio Based o Usig Histogram Equalizatio
More informationInternational Journal of Scientific & Engineering Research, Volume 4, Issue 10, October ISSN
Iteratioal Joural of Scietific & Egieerig Research, Volume 4, Issue, October-3 48 A ovel Noise estimatio by histogram equalizatio ad optimal filterig for robust speech ehacemet M.Shravai, P.Chadra Sekhar,
More informationIntroduction to Wireless Communication Systems ECE 476/ECE 501C/CS 513 Winter 2003
troductio to Wireless Commuicatio ystems ECE 476/ECE 501C/C 513 Witer 2003 eview for Exam #1 March 4, 2003 Exam Details Must follow seatig chart - Posted 30 miutes before exam. Cheatig will be treated
More informationADSP ADSP ADSP ADSP. Advanced Digital Signal Processing (18-792) Spring Fall Semester, Department of Electrical and Computer Engineering
ADSP ADSP ADSP ADSP Advaced Digital Sigal Processig (8-79) Sprig Fall Semester, 7 Departmet of Electrical ad Computer Egieerig OTES O RADOM PROCESSES I. Itroductio Radom processes are at the heart of most
More information(2) The MOSFET. Review of. Learning Outcome. (Metal-Oxide-Semiconductor Field Effect Transistor) 2.0) Field Effect Transistor (FET)
EEEB73 Electroics Aalysis & esig II () Review of The MOSFET (Metal-Oxide-Semicoductor Field Effect Trasistor) Referece: Neame, Chapter 3 ad Chapter 4 Learig Outcome Able to describe ad use the followig:
More informationDelta- Sigma Modulator with Signal Dependant Feedback Gain
Delta- Sigma Modulator with Sigal Depedat Feedback Gai K.Diwakar #1 ad V.Vioth Kumar *2 # Departmet of Electroics ad Commuicatio Egieerig * Departmet of Electroics ad Istrumetatio Egieerig Vel Tech Uiversity,Cheai,
More informationA study on the efficient compression algorithm of the voice/data integrated multiplexer
A study o the efficiet compressio algorithm of the voice/data itegrated multiplexer Gyou-Yo CHO' ad Dog-Ho CHO' * Dept. of Computer Egieerig. KyiigHee Uiv. Kiheugup Yogiku Kyuggido, KOREA 449-71 PHONE
More informationDecode-forward and Compute-forward Coding Schemes for the Two-Way Relay Channel
Decode-forward ad Compute-forward Codig Schemes for the Two-Way Relay Chael Peg Zhog ad Mai Vu Departmet of Electrical ad Computer Egieerig McGill Uiversity Motreal, QC, Caada H3A A7 Emails: peg.zhog@mail.mcgill.ca,
More informationNovel Steganography System using Lucas Sequence
Novel Stegaography System usig ucas Sequece Fahd Alharbi Faculty of Egieerig Kig Abdulaziz Uiversity Rabigh, KSA Abstract Stegaography is the process of embeddig data ito a media form such as image, voice,
More informationSIDELOBE SUPPRESSION IN OFDM SYSTEMS
SIDELOBE SUPPRESSION IN OFDM SYSTEMS Iva Cosovic Germa Aerospace Ceter (DLR), Ist. of Commuicatios ad Navigatio Oberpfaffehofe, 82234 Wesslig, Germay iva.cosovic@dlr.de Vijayasarathi Jaardhaam Muich Uiversity
More informationJournal of Computer Science, 9 (2): , 2013
Joural of Computer Sciece, 9 (2): 235-243, 203 ISSN 549-3636 203 Kaa ad Ravi, This ope access article is distributed uder a Creative Commos Attributio (CC-BY) 3.0 licese doi:0.3844/jcssp.203.235.243 Published
More informationTesting of methods for blind estimation of noise variance on large image database
Testig of methods for blid estimatio of oise variace o large image database Vladimir V. Luki a, Sergey K. Abramov a, Mikhail L. Uss a,b, Igor Marusiy a, Nikolay N. Poomareko a, Alexader A. Zelesky a, Beoit
More information