Proceedings of the International Multiconference on Computer Science and Information Technology pp. 767 774 ISBN 978-83-60810-27-9 ISSN 1896-7094 CFEEEFFFEFBFFAEAC AEC EEEDB DACDB DEEE EDBCD BACE FE DD EEEDB DACDB DEEE EDBCD BACE FEEDD ABCADAEEFEEFFEB EAEFBFFAFAEFBFFAF EAFFACFCCFEEAEE BBFECEAECCECFEF FCEFAEFFAEEBFEBFEF EAFAEBFFAACCFBFAF FFDFBCEACEFEAF CAFAEBEAFFFECAE CEDFBAEEFFEBFFAFE AFFAFCFEBCADBFEA CFEEFAEEBFEBEFFACFAEFB FFAFEAFBEFFACFAEF BFFAFEAFCFEEA EAECDFEEF ABFFEE ABCDBFFAFA CAFAECAECCA A BCDEFFDDEBDCECFDµD DBDCDEEDEDDBEDF EEDDEEDBBD BEDFDBCCBCEDDDEBD CDEEDCFDEBDCECBE DBDDEBDBBEDEBDB DBDEDEB ABCDEF DFBEDFBC EBBEEBEDFEFFB DBDEEDBEDFBEDFE ECDEEDEBEDFEEB EDBDBECB CDBCBAEEB ABFCDEDBDCDBCDEEDECBDB DBECDE DEDEFDEBBEBFCE DEBDCDEDECBE = DCEDDB = ( ) ( ) ( ) ( ) ( ) = ( ) + ( ) ( ) + DCEDDBCD µ DCECFD A 2µBACA A A A D DA DA D D Σ 2µBC 2µFE ECDEEDEDEB DDCADCBFDBµ DCADCBFDBµDABDE DCCDADCCDBCDEDEABCDEF BDFEDEBDBEDFFED EEDCCBBµ FDDD DEEDBBEDF E F µ F < D DBEFFBDEBFD B BCEDDBBDEBFDEDBFDBB ECDEEE BCEDDBBDEBFDEED DEB ( ) ( ) = Ε 2µ DCBAEEBDDBFDE DFDBDCFDBDEDEB EDDDEFDEBB 978-83-60810-27-9/09/$25.00 c 2010 IEEE 767
768 PROCEEDINGS OF THE IMCSIT. VOLUME 5, 2010 EDBDFEDµBFE DEDCECBEEDDCF µbdbdddeeddb BBEDF DFµEDBBEDB BBEDFBDEED EEEDEDDEBBDDEBEB DBCDEFBDCFDµ optedbbe DEB µ = ( + ) DDBEFFBDEBFD B CFDFEDFD FDFEDFDEEDEBBD EBFCDBFDEFECEBBF BCBCDEDDEFDBECEBEBD BDBDECDBCDEEDEDBEDF EEEBDCCBBµEECB EEDDEDEDBFDED ECDFCBµDDECBEEE EDBDDEEDDBEDFDEBD DCADCBFDBµDABDE ABADCCDBCDEDEABCDEF ABEBBBEDFEED EDBDECBDDBBDEBFDE DCBCDFBABBD BEDFEEDDDBDDEDCD BDDBDEECDCBB DBBµFEEDEDBBF ECDDEBBEDBDB λ F µ λ F FEFFEBDBBDEBFDEB DECDDB BBBDEDDEED BBEDFBBBEDF EBBDFBDB B BBDEBDFDBDBBF EFEDDB EDEBC DEBDCECD EEDDBCDEFF EBDEBBDEED DCAADCFADCBFDB µ optadccdbcdedeabcdef DFEDEBBDCECFDµ opteef CBDDDBBDBEDFDEC FDµ optdfbbfcbfedddbcd DBµEDBDBC DEFDDEDBEDFBCECB Table I. CALCULATED VALUES OF A STEP SIZE PARAMETERS µ opt, µ max, µ B OF LMS ADAPTIVE FILTER FOR INPUT SIGNAL x(n) WITH DIFFERENT SSNR VALUES. B µ F µ B µ opt µ opt µ opt B µ F µ B µ opt µ opt µ opt DDDBDFB ξ DBDFEEFFFBξ FE ξ = = µ ξ FE B µ F µ B µ opt µ opt µ opt B µ F FEDFDEEFEBCFDDD CBEFBBBDBEDFEDB BCDEFEDEDF FEBFCEDEDDFB DEDCDEEDEDEBECBFD BEDFBFEDFD D BFDDDCDEDF BCDDDξ FE µ B µ opt µ opt µ opt
JAN VAŇUŠ, VÍTĚZSLAV STÝSKALA: APPLICATION OF OPTIMAL SETTINGS OF THE LMS FILTER 769 EFBDDτ FEDE τ ABC = = µλ µ λ EBDBBDEBFDEBDECD DB DDBEFFBDEBFD B DED λ EEDEFEBD EDBCCBEFDEDDECB DEEDDEFBDDτ FFCCEDBDBE CCBEFDB BBCDEEDD BFECDEDBCDEED = λ A BDDFEDEBBDCDEEDD EEFCBDDDDBDCDEEDE BDCECBEDBFF BCFDBDCDEEDEEDE BDCDEEDDDBED CECBEEBDDEFDBE EDFDBEDEB EDEBBDFEEBDBD CDEED BEDEBEBFCDBB DBDBCCBDD DDDBBBBDD BDCDEEDBEDFE DDEBDBEEDDB DEBDFEEFFEDEFEEDD DDBEEED DDDBBEEB FBDCDDBDFD EDDDE EFEEFFEDBDDBCBECD CBEDCBEBDE DDEE E EDBACEBE DEE E EDFEFCEEBFDFD BEFCFDCEFD EEFFEDBFCDDEB = FE { } EEFCBFEDEBDB EBDEBB CDECBEED B FE A D B A BEFCFDDEBBCEFDEDE BEEDDFDEDBBE DEBBCECBE C= EEF= EFC C EDFBFDBCE DFBFDECDEED C EEBFDEBBDCDEED ECDFD CEBBDDF TABLE II. THE VALUES OF DISTANCE d ARE CALCULATED BY COMPARING OF THE CZECH ISOLATED WORDS "JEDEN" (ONE) WITH THE WORDS "DVA" (TWO) FOR UP TO "PĚ T" (FIVE) AND COMPARED THE WORD "DVA" (TWO) WITH THE WORDS "JEDEN"(ONE), "TŘ I" (THREE) FOR UP TO "PĚ T"(FIVE). DE DE CD DEE E E DE DE CD DEE E E Table III. THE VALUES OF ORDER M OF ADAPTIVE FILTER AND DISTANCE d BETWEEN DESIRED SPEECH SIGNAL d(n) TO ADAPTIVE FILTER AND ERROR SIGNAL e(n) FROM ADAPTIVE FILTER CALCULATED BY WAY OF WIDROW METHODS (SIMULATED IN MATLAB). EB EB EB B µ F DEE E EEE µ F DEE E EEE µ F DEE E EEE B B B µ F µ F µ F DEE E EEE µ F µ F µ F µ F µ F µ F
770 PROCEEDINGS OF THE IMCSIT. VOLUME 5, 2010 DCCBBEEDEBEB BFCBDDBDCBCBFDB EDEBE EEDDDEBBDE EEBEBDBDBBE B EDBE EDDDEBBDEE EBEBDBDBBE B EDBE DDEBBDCEBEBD BEDBBE B EDBE DDEBBDCEBEBD BEDBBE B DFDBBBCDEFDFDBDCECF Dµ optddbdcdeedc CEEDDC ADEBBDCECFDµ optbcdef BFECDEEDDBDCDE ED BDBEEE EDBDCDEED DDDBBBEE DDDEDD DEBDBCDEE DDEEDDB DBCDEFBDCDEEDB BBDCDEEDDBFEE FFEDEDDBBFCE EBFCDBFDEFECEBBF BCBCDEDDEFDBECEBEDE BE ED TABLE IV. SEGMENTAL SIGNAL TO NOISE RATIO VALUES CALCULATED FOR THE SPEECH SIGNAL WORD "JEDEN" (FIG.1) TO ADDITIVE NOISE AND TO ADDITIVE WHITE NOISE. BEEEEDEB EDEBE EDEEDBE EDEEDBE EDEEDBE ADBDEB EDBBE EAD BDCDEED µ D TABLE V. THE VALUES OF ORDER M OF ADAPTIVE FILTER AND DISTANCE d BETWEEN DESIRED SPEECH SIGNAL d(n) TO LMS ADAPTIVE FILTER AND ERROR SIGNAL e(n) FROM LMS ADAPTIVE FILTER CALCULATED BY WAY OF DRAFT METHOD WITH DTW CRITERION (SIMULATED IN MATLAB). B B B B µ F DEE E EEE µ F DEE E EEE µ F µ F µ F DEE E EEE µ F µ F µ F µ F DEE E EEE µ F µ F µ F
JAN VAŇUŠ, VÍTĚZSLAV STÝSKALA: APPLICATION OF OPTIMAL SETTINGS OF THE LMS FILTER 771 DCDEEDEDBE EDEDDDDDBFDCEFECD BFDCBEDBEBDCDEEDDCEFE CDDEDECBDDF BDCDDEDBBEBDCD EEBDEDDEDEDBE BFDCEFECDCDEBEE CBFDDDDECCBED BEEDDBDDDEDBD EBFCDBFDEFECEBBF BCBCDEDDEFDBBECDEED EDEDDEBµ= EFDE ECDEBBEBECDEBE ECCB DDDCEDD BDCDBDCDEEDACEBD BFEDEDDEB B µ µ µ B µ D EDEBE ECEE DCEBDBE A AAA DCFDCA DFDBEDEDEBBD CDEBEEBFCEEFD EEDBDDBDCDEED EDBEDFECCEDEBBDCCEBB EDEBEECEFECDBDEEE EDEBEBEECD EFDBEBDEEEEB DEDDBEEBCDEBEBE EBFCDBFDEFECEBBF BCBCDEDDEFDBBECDEBE EDEDDEBµ= EFDE DFDBBBCDEFDDE BDCDEEDDDCEDBµB Table VI. THE OPTIMAL VALUES OF ORDER M OF FILTER AND DISTANCE d BETWEEN DESIRED SPEECH SIGNAL d(n) AND ERROR SIGNAL e(n) FROM LMS ADAPTIVE NOISE CANCELLER CALCULATED BY WAY OF DRAFT METHOD WITH DTW CRITERION (SIMULATED IN MATLAB). B B B B µ F DEE E EEE µ F D µ F D µ F D µ F DEE E EEE µ F D µ F D µ F D µ F DEE E EEE µ F D µ F D µ F D
772 PROCEEDINGS OF THE IMCSIT. VOLUME 5, 2010 DCDEEDEDBEDFEDCCEDEBB DCCEBBEDEBEBFDCE ADBCDEFDBBDC DEBEEDDEC EEBEEBFD CDEBEDE CBEBFDBDCDB CDEBEACEBDBFED EDDEBBBD µ µ µ EDEBE ECE ECEBDBE FFDDCADCACBA DFDBEDEDEBBDFEEB DBBDCDEEDEDBEDF ECCEDEBDBCCEBEBFD CEEEFCFDDEBBDBD DBCDEBEBEEDE BBDDEDAE CDEEEBFCBBFEE EEDEBEFDEDB BEDEBBECDEEE EBFCDBFDEFECEBBF BCBCDEDDEFDBECDEEE EDEBEDBCDEEDECCE DEBDBCCBEBFDCEEDEDDEBEFCFD BA CCEDEBBDDFDBEDEDEB EBDEDC DCDEBBDCECFDµ EDD DDEEDBBDBEDFE BECDEEDBC DEBEEFDE DCDDEBBDCDEED BBDDCECFDµ EBECDE EDBDCDEBE EFDE DDDBDCD EEBFCDEEDEDBEDF EEEBDCFDD BDCDEEDB DBEDE BDDEBDBEBF CDEEDBDCDACEBDEE E ED BCDEFDCFDBCDEEDED BEDFµ B TABLE VII. THE CALCULATION VALUES OF DISTANCE d, LENGTH M AND A STEP SIZE PARAMETER µ OF THE LMS ADAPTIVE NOISE CANCELLER FOR ( =10%, =20%, =30%) SSNR=6,676(dB), (SIMULATED IN MATLAB). µ µ µ µ F DEE E E DCFCEEBDDDCFDµ B DCDEBEEFCFDBD ABDBEDµ=1.10 8 DB µ=1.10 12 DBDCDE BE DBEDDB CDEFDDEBCFDµ DB µ=1.10 8 BDCDEBE EFCFDBDA CBEBFDBDCDB CDE BE EFCFD B AACEBDBFEDEDDEB BD µ TABLE VIII. THE CALCULATION VALUES OF DISTANCE d FOR SETTINGS OF LENGTH M=21 AND PARAMETERS µ (x(n) (Fig.8) WITH SSNR=6,676(dB)) (LMS ADAPTIVE NOISE CANCELLER WAS IMPLEMENTED ON DSK TMS320C6713). Settings of parameterµ AD B µ µ µ DEE E E E EEE AAA AAA DFDBEDEDEBBBCDE FDDECFDBCDEBE EFCFDBACCEEBEBF FEDEBEDBDBDFE DFEFCFDEEFDEBBE EDEBBCDEBBDBBDDEDBD EEDBEEDEBBDBFBDE B CBEDEBEBEBFFEDEBEDBDB DFBDBEE
JAN VAŇUŠ, VÍTĚZSLAV STÝSKALA: APPLICATION OF OPTIMAL SETTINGS OF THE LMS FILTER 773 CBBFFBEEDDC DEBEEFCFDBA EDBDEDEBE CBEDEB FDBEDEDEED CBEDEB EBFCDBFDEFECEBBF BCBCDEDDEFDBBECDE BEAEBFDCEEDEDDEBµ= EFCFDBDA EEDBDBFBDEEBBD BEEBBEBDBBBCDEBDE DEBEDEE EFCFDDEBBDCDEBEBBEBF FEDEBEDBDBDFBFBD EFBDCEFDDBDFED BDDDEBBDBEBFFFEBCB BCDECBDDEDBBDF BFDFBDBEDBDFDED DEBBFEBCBEDDBDBB EDEBE BBEDEBEDBBE BEBBEBDDEBEFBEDE FECEEDBFBFD FEBCBEBCCCCBEFD FBFDFEBCB BBBBFFBEBE EBEDEBBDBE ABEDEBBDCEFDDBBEE CBBFFBEEDBDD CDEBE EDBDEDEBE CBEDEB EDEDEBEC BEDEB EDEBBBEDEBBEBDBDBFF BEBEBBEDEBBDBE AA DECCEDBEEDEBBD CBCBFDBBDDDBCDEEDED BEDFECCEDEBBCCEBEBF CEBEFDEBEBD BDBEDEBBDEDBBB DFEEDEDBCECBEEE BCDEFDDEDCECFDµBBD CDEEDBDCDEBE CBCBFDBEEBDCDE EDEBBDDDBDCDEBE EDCCEDEBBCCEEDEBEBF CEEFCFDDEBBD AEEEFCFDDEBB BEBFFEDEBEDBDBDF BEFDEBBDBEBBCDEDEDEBE EE ECEFDBCEDABDB CEDDF
774 PROCEEDINGS OF THE IMCSIT. VOLUME 5, 2010 A ECCCCBDDDB DBDBDCCBD A BBDCCDBEABCDCA BEBAEDCC BEFDCCDBCBCFBCDE BEBCCC EDCCDBDEABA CC BBFEAAC EAEAFCDD CC BBBFDACAEABAC DDCC B BAAEC DEBCB DDCEDDBB AAFCFDFF BE BFFEDEBEDBDBDFEDDEBDE D EB DCBADBAC BACBAEEBEB BBCC CACDBBDBC DDBCC FFDDCAADEDCCDBABAC AFFCDCACDEADBADF BE BBECC AECCDCBACCDCA CDEDEBEB CC