A New Vlsi Architecture for Adaptive Filter

VLSI Architecture
of 6
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  National Conference on “Smart Instrumentation, Measurement and Applications” A NEW VLSI ARCHIEC!RE #R A$A%IVE ILER   R.Aarthi 11 1  PG Scholars, Department of Applied Electronics Engineering,SNS College of Engineering, Coimbatore– 641 !  Abstract -  In this paper the architecture is  proposed for Sign Sign LMS algorithms which is obtained from the conventional LMS recursion by replacing the tap-input vector x(n) and error vector e(n) with the vector sgnx(n)! and sgne(n)! in the weight updation e uation#  Implementation of this function may be cheaper than the conventional LMS recursion$ especially in high speed applications such as biotelemetry these types of recursions may be necessary#  %eywords &   Adoptive filters$ LMS$A'  I. I  NTRODUCTION  Signal denoising is a field of engineering that studies methods used to recoer an srcinal signal from nois! signals corru ted #! different t! es of noises. Noise ma! #e in the form of $hite noise%   ink noise% #a##le noise and man! other t! es of noise resent in the enironment. Oer the last decades% noise remoal from signals is an area of interest of researchers during signal rocessing. Ada tie algorithms are mostl! used for signal denoising. The a lication areas of are 1D or &D  #iomedical signal anal!sis% roducing ' anal!(ing irregular signals. Ada tie algorithms can decom ose a signal into seeral scales that re resent different fre)uenc! #ands and at each scale% the osition of signal*s instantaneous structures can #e determined a ro+imatel!. Such a ro ert! can #e used for denoising. Although $ithout information on the signal to #e anal!sed% filters select information #! strongl! reducing its )uantit!. The method is #ased on error alue in the signal that each alue of the desired signal is com ared to the filter out ut. Using ada tie filters to remoe noise from a signal re)uires identif!ing $hich com onents contain the noise% and then reconstructing the signal. The signal used here is s eech or ,C-. S eech signal is the #asic tool for diagnosis. During ac)uisition seeral noises contaminate the signal arameters. One of the dominant noise that contaminate the s eech signal is due to atmos here distur#ances. The noises that affect the s eech signal are $hite noise% ink noise%  #a##le noise and man! other t! es of noise   resent in the enironment. east mean s)uare /0S algorithm is one of the algorithm used in ada tie noise canceller to denoise the s eech signals. The algorithm used to u date the filter coefficient is the east 0ean S)uare /0S algorithm $hich is kno$n for its sim lification% lo$ com utational com le+it!% and #etter   erformance in different running enironments. II.  IT,RATUR, R,2I,3  A #PGA based ANC   0ohammed 4ahoura 5 6assan ,((aidi /&711 im lemented an efficient 89-A:#ased ra id   rotot! ing of an ada tie noise canceller /ANC using east 0ean S)uare /0S in ;U9 2irte+:II 9ro deelo ment #oard and ;ilin+ S!stem -enerator . Ne$ arallel and se)uential architectures of the ANC are ro osed and successfull! a lied to remoe noise from electrocardiogram and s eech signals. It is o#sered that #! roiding com ara#le filtering   erformances that of the arallel architectures% the   ro osed se)uential s!stem re)uired fe$er material resources.  $ %&S based adapti'e filters  0d. <ia Ur Rahman% Rafi Ahamed Shaik% D 2 Rama =oti Redd!% /&711 discuss im lementation seeral signed 0S #ased ada tie filters% $hich are com utationall! su erior haing multi lier free $eight u date loo s are ro osed for noise cancellation in the ,C- signal. Different filter structures are resented to eliminate the dierse forms of noise like >76( o$er line interference%  #aseline $ander% muscle noise and the motion artifact. 8inall!% these algorithms $ere a lied on real ,C- signals o#tained from the 0IT:4I6 data #ase and com ared their erformance $ith the conentional 0S algorithm. The results sho$ that the erformance of the signed regressor 0S algorithm is su erior than conentional 0S algorithm% the erformance of signed 0S and sign:sign 0S #ased reali(ations are com ara#le to that of the 0S #ased filtering techni)ues in terms of signal to noise ratio and com utational com le+it!. C (ptim)m Step Si*e %east &ean S+)are (SS%&S- algorithm So$m!a.I% Ch Sathi Ra?u% 0d. <ia Ur Rahman and D 2 Rama =oti Redd! /&711 resented a noel ada tie filtering techni)ue to remoe artifacts from cardiac signal #ased on o timum ste si(e strateg!.This resultant ada tie algorithm is called as o timum ste si(e least mean s)uare /OSS0S algorithm. The ada tie filters essentiall! minimi(es the mean:s)uared error #et$een a rimar! in ut% $hich is the nois! &'SRC(EIE )  SIMA*&+  National Conference on “Smart Instrumentation, Measurement and Applications” ,C-% and a reference in ut% $hich is either noise that is correlated in some $a! $ith the noise in the rimar! in ut or a signal that is correlated onl! $ith ,C- in the rimar! in ut. 8inall!% this algorithm $as a lied on real ,C- signal o#tained from the 0IT:4I6 data #ase and com ared the erformance $ith the conentional 0S algorithm. The results clears that the   erformance of the OSS 0S algorithm is su erior to that of the 0S #ased algorithm in noise reduction III.  SCO9, O8 T6, 9RO,CT The 8IR filter com utations la! an im ortant role in the DS9 s!stems as the ada tie t! e. The 8IR filtering is a conolution o eration $hich can  #e ie$ed as the $eighted summation of the in ut se)uences. The large amount of multi l!:and accumulate o erations attracts man! researchers to stud! this issue. The 8IR 8ilter consists of shifters% adders% and multi liers. These constitutes can #e chosen on the demand of the designer. One of the most o ular ada tie algorithms aaila#le in the literature is the stochastic gradient algorithm also called least:mean:s)uare /0S. Its o ularit! comes from the fact that it is er! sim le to #e im lemented. As a conse)uence% the 0S algorithm is $idel! used in man! a lications. east 0ean : S)uare /0S ada tie filter is the main com onent of man! communication s!stemsB traditionall!% such ada tie filters are im lemented in Digital Signal 9rocessors /DS9s. Sometimes% the! are im lemented in ASICs% $here erformance is the ke! re)uirement. 6o$eer% man! high:  erformance DS9 s!stems% including 0S ada tie filters% ma! #e im lemented using 8ield 9rogramma#le -ate Arra!s /89-As due to some of their attractie adantages. Such adantages include fle+i#ilit! and   rogramma#ilit!% #ut most of all% aaila#ilit! of tens to hundreds of hard$are multi liers aaila#le on a chi . The 0S ada tie filter en?o!s a num#er of adantages oer other ada tie algorithms% such as ro#ust #ehaior $hen im lemented in finite: recision hard$are% $ell understood conergence #ehaiour and com utational sim licit! for most situations as com ared to least s)uare methods. The #elo$ figures e+ lains the t! es of 8IR  I2. T9,S O8 8IR 8IT,RSThe figure 1 sho$s that direct form 8IR it limited #! the critical ath corres onding to the longest com utation time among all aths that contain (ero dela!s.   8ig.1 Direct form 8IR The figure & sho$s the #inar! tree direct form sho$s the critical ath is reduced $hereas figure 1. sho$s that Trans osed form sho$s that it oercomes the critical ath limitation #! retiming dela!s into the adder chain.   8ig.& 4inar! tree direct form 8IR 8ig. Trans osed form 8IR  'SRC(EIE )  SIMA*&+  National Conference on “Smart Instrumentation, Measurement and Applications” 2.  ,AST 0,AN SEUAR, A-ORIT60 The e+traction of high:resolution signals from recordings contaminated $ith #ack ground noise is an im ortant issue to inestigate. The goal for signals enhancement is to se arate the alid signal com onents from the undesired artefacts% so as to   resent an signal that facilitates. ,as! and accurate inter retation. 0an! a roaches hae #een re orted in the literature to address signal enhancement. In recent !ears% ada tie filtering has #ecome one of the effectie and o ular a roaches for the rocessing and anal!sis of the signal and other #iomedical signals. Ada tie filters ermit to detect time ar!ing otentials and to track the d!namic ariations of the signals. 4esides% the! modif! their #ehaiour according to the in ut signal. Therefore% the! can detect sha e ariations in the ensem#le and thus the! can o#tain a #etter signal estimation. Ada tie solution #ased on the 0S algorithm is suggested for noise cancellation in signal. The reference in uts to the 0S algorithm are deterministic functions and are defined #! a eriodicall! e+tended% truncated set of orthonormal #asis functions. The 0S algorithm o erates on an FinstantaneousF #asis such that the estimate. In a recent stud!% ho$eer% a stead! state conergence anal!sis for the 0S algorithm $ith deterministic reference in uts sho$ed that the stead!:state $eight ector is #iased% and thus% the ada tie estimate does not a roach the 3iener solution. To handle this dra$#ack another strateg! $as considered for estimating the coefficients of the linear e+ ansion% namel!% the #lock 0S /40S algorithm% in $hich the coefficient ector is u dated onl! once eer! occurrence #ased on a  #lock gradient  estimation. A ma?or adantage of the #lock% or the transform domain 0S algorithm is that the in ut signals are a ro+imatel! uncorrelated. 2I. S/GN S/GN A%G(0/2&  The sign sign algorithm is o#tained from the conentional 0S recursion #! re lacing the ta :in ut ector +/n and error ector e/n $ith the ector sgn G+/nH and sgn Ge/nH. Com le+it! reduction of the noise cancellation s!stem%   articularl! in a lications such as $ireless  #iotelemetr! s!stem has remained a to ic of intense research. This is #ecause of the fact that $ith increase in the s eech signal data transmission rate% the channel im ulse res onse length increases and thus the order of the filter increases. These algorithms reduce the com utational com le+it! of the ada tie algorithm $ithout affecting the signal )ualit!. In order to achiee this% the sign #ased ada tie algorithm is considered. This algorithm en?o!s less com utational com le+it! #ecause of the sign   resent in the algorithm. This algorithm re)uires no multi lications as in the 0S algorithm% thus making them attractie from ractical im lementation oint of ie$. 2II. ,;A09, 8OR NOIS, CANC,ATIONAda tie filters are re)uired $hen it is necessar! for a digital filters characteristics to #e aria#le% ada ting to changing signal characteristics. Such filters can #e classed as self designing in that the! rel! u on a recursie mathematical algorithm% that allo$ them to   erform satisfactoril! in an enironment $here com lete kno$ledge of the signal of interest /desired signal and noise is not kno$n. As a conse)uence% ada tie filters% such as the 0S /least mean s)uared algorithm hae #een used in man! real $orld a lications such as #iomedical signal enhancement% s!stem identification and noise cancellation. The follo$ing e+am le demonstrates the enhancement of a 1776( signal  #uried in #and limited $hite noise% #! irtue of a 7th order 8IR 0S filter. 2III. ADA9TI2, IN, ,N6ANC,0,NT /A,Ada tie line enhancement /A, refers to the case $here a nois! signal% +/n consisting of a sinusoidal com onent% s/n is aaila#le and the re)uirement is to remoe the noise art of the signal% n/n. This ma! #e achieed #! the arrangement sho$n #elo$. It consists of a de:correlation stage% s!m#oli(ed #! and an ada tie   redictor. The de:correlation stage attem ts to remoe an! correlation that ma! e+ist #et$een the sam les of noise% #! shifting those sam les a art. As a result% the redictor can onl! make a   rediction a#out the sinusoidal com onent of u/n% and $hen ada ted to minimi(e the instantaneous s)uared error out ut% e/n% the line enhancer $ill #e a filter o timi(ed /the 3iener solution or tuned to the sinusoidal com onent. -'SRC(EIE )  SIMA*&+  National Conference on “Smart Instrumentation, Measurement and Applications” 8ig.J Ada tie noise cancellationThe 8igure J sho$s a little further% notice ho$ the in ut to the ada tie filter is actuall! sam les  #ehind the rimar! in ut signal% +/n% since the de:correlation stage has introduced a hase shift. Therefore% in order to time align the enhanced signal% K/n $ith the in ut signal% +/n the ada tie filter must #e a#le to  redict ahead in time /formall! referred as for$ard linear rediction%  #! o timi(ing its filter coefficients in a least s)uares sense% such that the instantaneous s)uared error is minimi(ed. I;. ADA9TI2, NOIS, CANC,,R /ANCANC sho$n in figure L is an interesting a lication of ada tie filter that has #een used in a $ide range of signal rocessing s!stems. The! include o$er line interference /L7M>76( elimination from electrocardiogram /,C- signals % fetal ,C- e+traction % echo cancelation in long:distance tele hone and satellite communications % ada tie noise cancelation for hearing aids %am#ient noise reduction from #reath sound measurements % etc. The ada tie noise canceller is #ased on a finite im ulse res onse /8IR filter $hose coefficients are automaticall! ad?usted using a gradient descendant algorithm. The least mean s)uare /0S algorithm is commonl! used to minimi(e the coast function /error.8ig.L 4lock diagram of Ada tie noise canceller  ;. ADA9TI2, NOIS, CANC,ATIONThe #lock diagram of an ada tie noise canceller that consists of t$o in ut signals and an ada tie filter. The rimar! in ut contains the $anted signal sk lus the uncorrelated noise n7k . The reference in ut is the noise n1k $hich is uncorrelated $ith the signal sk #ut correlated in some unkno$n $a! $ith the corru ting noise n7k. The reference in ut is rocessed $ith the ada tie filter that iteratiel! ad?usts its im ulse res onse to minimi(e the mean s)uare error /0S, #et$een its out ut and the rimar! in ut. The ada tie filter is o timi(ed to roduce an out ut !k that must #e as close as ossi#le to n7k . This out ut is su#tracted from the rimar! in ut to roduce the s!stem out ut ek  sk n7k  !k . Assume that signal and noise are uncorrelated% the mean s)uare error /0S, is gien #!  E[e k 2   ] =E[ s k   ]  2  +E[(n ok   –y  k   ) 2   ]  ;I. T9,S O8 0S 4AS,D ADA9TI2, 8IR 8IT,RSP.Ne$ i elined arallel of the 0S:#ased ada tie 8IR filter. PSe)uential architectures of the 0S: #ased ada tie 8IR filter.Com ared to e+isting high:s eed s!stems the   ro osed i elined architectures ensure the same filtering erformances #ut the! resents adantage to re)uire less hard$are resources  #ecause the! need less dela! and shifter elements. In addition% the se)uential i elined structure needs onl! t$o multi liers regardless of the filter si(e. The arallel architectures is $ell suited for small:si(e filters% $hile the se)uential one is more a ro riate for large:si(e filters. The arallel architecture is $ell suited for a high sam ling rate re)uirement and a small num#er of coefficients. +'SRC(EIE )  SIMA*&+


Aug 3, 2017
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks