A noise resilient distributed inference framework for body sensor networks
Common Phase Error due to Phase Noise in OFDM - Estimation and Suppression
COMMON PHASE ERROR DUE TO PHASE NOISE IN OFDM-ESTIMATION AND SUPPRESSIONDenis Petrovic,Wolfgang Rave and Gerhard FettweisV odafone Chair for Mobile Communications,Dresden University of Technology,Helmholtzstrasse18,Dresden,Germany{petrovic,rave,fettweis}@ifn.et.tu-dresden.deAbstract-Orthogonal frequency division multiplexing (OFDM)has already become a very attractive modulation scheme for many applications.Unfortunately OFDM is very sensitive to synchronization errors,one of them being phase noise,which is of great importance in modern WLAN systems which target high data rates and tend to use higher frequency bands because of the spectrum availability.In this paper we propose a linear Kalmanfilter as a means for tracking phase noise and its suppression.The algorithm is pilot based.The performance of the proposed method is investigated and compared with the performance of other known algorithms.Keywords-OFDM,Synchronization,Phase noise,WLANI.I NTRODUCTIONOFDM has been applied in a variety of digital commu-nications applications.It has been deployed in both wired systems(xDSL)and wireless LANs(IEEE802.11a).This is mainly due to the robustness to frequency selective fading. The basic principle of OFDM is to split a high data rate data stream into a number of lower rate streams which are transmitted simultaneously over a number of orthogonal subcarriers.However this most valuable feature,namely orthogonality between the carriers,is threatened by the presence of phase noise in oscillators.This is especially the case,if bandwidth efficient higher order modulations need to be employed or if the spacing between the carriers is to be reduced.To compensate for phase noise several methods have been proposed.These can be divided into time domain[1][2]and frequency domain approaches[3][4][5].In this paper we propose an algorithm for tracking the average phase noise offset also known as the common phase error(CPE)[6]in the frequency domain using a linear Kalmanfilter.Note that CPE estimation should be considered as afirst step within more sophisticated algorithms for phase noise suppression[5] which attempt to suppress also the intercarrier interference (ICI)due to phase noise.CPE compensation only,can however suffice for some system design scenarios to suppress phase noise to a satisfactory level.For these two reasons we consider CPE estimation as an important step for phase noise suppression.II.S YSTEM M ODELAn OFDM transmission system in the presence of phase noise is shown in Fig. 1.Since all phase noise sources can be mapped to the receiver side[7]we assume,without loss of generality that phase noise is present only at the front end of the receiver.Assuming perfect frequency and timing synchronization the received OFDM signal samples, sampled at frequency f s,in the presence of phase noise can be expressed as r(n)=(x(n) h(n))e jφ(n)+ξ(n).Each OFDM symbol is assumed to consist of a cyclic prefix of length N CP samples and N samples corresponding to the useful signal.The variables x(n),h(n)andφ(n)denote the samples of the transmitted signal,the channel impulse response and the phase noise process at the output of the mixer,respectively.The symbol stands for convolution. The termξ(n)represents AWGN noise with varianceσ2n. The phase noise processφ(t)is modelled as a Wiener process[8],the details of which are given below,with a certain3dB bandwidth∆f3dB.,0,1,2...m lX l=,0,1,2...m lR l=Fig.1Block diagram of an OFDM transmission chain.At the receiver after removing the N CP samples cor-responding to the cyclic prefix and taking the discrete Fourier transform(DFT)on the remaining N samples,the demodulated carrier amplitude R m,lkat subcarrier l k(l k= 0,1,...N−1)of the m th OFDM symbol is given as[4]:R m,lk=X m,lkH m,lkI m(0)+ζm,lk+ηm,lk(1)where X m,lk,H m,lkandηm,lkrepresent the transmitted symbol on subcarrier l k,the channel transfer function andlinearly transformed AWGN with unchanged variance σ2n at subcarrier l k ,respectively.The term ζm,l k represents intercarrier interference (ICI)due to phase noise and was shown to be a gaussian distributed,zero mean,randomvariable with variance σ2ICI =πN ∆f 3dB s[7].The term I m (0)also stems from phase noise.It does not depend on the subcarrier index and modifies all subcarriers of one OFDM symbol in the same manner.As its modulus is in addition very close to one [9],it can be seen as a symbol rotation in the complex plane.Thus it is referred to in the literature as the common phase error (CPE)[6].The constellation rotation due to CPE causes unaccept-able system performance [7].Acceptable performance can be achieved if one estimates I m (0)or its argument and compensates the effect of the CPE by derotating the received subcarrier symbols in the frequency domain (see Eq.(1)),which significantly reduces the error rate as compared to the case where no compensation is used.The problem of esti-mating the CPE was addressed by several authors [3][4][10].In [3]the authors concentrated on estimating the argument of I m (0)using a simple averaging over pilots.In [10]the argument of I m (0)was estimated using an extended Kalman filter,while in [4]the coefficient I m (0)itself was estimated using the LS algorithm.Here we introduce an alternative way for minimum mean square estimation (MMSE)[11]of I m (0)using a linear scalar Kalman filter.The algorithm is as [4]pilot based.III.P HASE N OISE M ODELFor our purposes we need to consider a discretized phase noise model φ(n )=φ(nT s )where n ∈N 0and T s =1/f s is the sampling period at the front end of the receiver.We adopt a Brownian motion model of the phase noise [8].The samples of the phase noise process are given as φ(n )=2πf c √cB (n )where f c is the carrier frequency,c =∆f 3dB /πf 2c [8]and B (n )represents the discretizied Brownian motion process,Using properties of the Brownian motion [12]the fol-lowing holds:B (0)=0and B (n +1)=B (n )+dB n ,n ∈N 0where each increment dB n is an independent random variable and dB n ∼√T s N (0,1).Noting that φ(n )=2πf c √cB (n )we can write the discrete time phase noise process equation asφ(n +1)=φ(n )+w (n )(2)where w (n )∼N (0,4π2f 2c cT s )is a gaussian randomvariable with zero mean and variance σ2w =4π2f 2c cT s .IV.CPE E STIMATION U SING A K ALMAN F ILTER Since all received subcarriers within one OFDM symbolare affected by the same factor,namely I m (0),the problem at hand can be seen as an example of estimating a constant from several noisy measurements given by Eq.(1)for which purpose a Kalman filter is well suited [11].For a Kalmanfilter to be used we need to define the state space model of the system.Define first the set L ={l 1,l 2,l 3,...l P }as a subset of the subcarrier set {0,1,...N −1}.Using Eq.(1)one can writeR m,l k =A m,l k I m,l k (0)+εm,l k(3)where A m,l k =X m,l k H m,l k and I m,l k (0)=I m (0)for all k =1,2...,P .Additional indexing of the CPE terms is done here only for convenience of notation.On the other hand one can writeI m,l k +1(0)=I m,l k (0).(4)Equations (3)and (4)are the measurement and processequation of the system state space model,where A m,l k represents the measurement matrix,while the process matrix is equal to 1and I m,l k (0)corresponds to the state of the system.The measuring noise is given by εm,l k which combines the ICI and AWGN terms in Eq.(1),the varianceof which for all l k equals σ2ε=(σ2ICI +σ2n ).The process noise equals zero.Note that the defined state space model is valid only for one OFDM symbol.For the state space model to be fully defined,knowledge of the A m,l k =X m,l k H m,l k is needed.Here we assume to have ideal knowledge of the channel.On the other hand we define the subset L to correspond to the pilot subcarrier locations within one OFDM symbol so that X m,q ,q ∈L are also known.We assume that at the beginning of each burst perfect timing and frequency synchronization is achieved,so that the phase error at the beginning of the burst equals zero.After the burst reception and demodulation,the demodulated symbols are one by one passed to the Kalman filter.For a Kalman filter initialization one needs for eachOFDM symbol an a priori value for ˆI m,l 1(0)and an a priori error variance K −m,1.At the beginning of the burst,when m =1,it is reasonable to adopt ˆI −1,l 1(0)=1.Within each OFDM symbol,say m th,the filter uses P received pilot subcarriers to recursively update the a priori value ˆI −1m,l 1(0).After all P pilot subcarriers are taken into account ˆI m,l P (0)is obtained,which is adopted as an estimate ofthe CPE within one OFDM symbol,denoted as ˆIm (0).The Kalman filter also provides an error variance of the estimateof I m,l P (0)as K m,P .ˆI m,l P(0)and K m,P are then used as a priori measures for the next OFDM symbol.The detailed structure of the algorithm is as follows.Step 1:InitializationˆI −m,l 1(0)=E {I −m,l 1(0)}=ˆI m −1(0)K −m,1=E {|I m (0)−ˆIm −1(0)|2}∼=E {|φm −ˆφm −1|2}=σ2CP E +K m −1,Pwhere σ2CP E =4π2N 2+13N +N CP ∆f 3dBf s(see [10]),K 0,P =0and φm =arg {I m (0)}.Repeat Step2and Step3for k=1,2,...,P Step2:a-posteriori estimation(update)G m,k=K−m,kH H m,lkH m,lkK−m,kH Hm,l k+(σ2ICI+σ2n)ˆIm,l k (0)=ˆI−m,l k(0)+G m,k[R m,lk−H m,l kˆI−m,l k(0)]K m,k=(1−G m,k H m,lk )K−m,kStep3:State and error variance propagationK−m,k+1=K m,k(5)ˆI−m,l k+1(0)=ˆI m,lk(0)Note that no matrix inversions are required,since the state space model is purely scalar.V.CPE C ORRECTIONThe easiest approach for CPE correction is to derotate all subcarriers l k of the received m th symbol R m,lkby φm=−arg{ˆI m(0)}.Unambiguity of the arg{·}function plays here no role since any unambiguity which is a multiple of2πrotates the constellation to its equivalent position in terms of its argument.The presented Kalmanfilter estimation algorithm is read-ily applicable for the decision feedback(DF)type of algo-rithm presented in[4].The idea there was to use the data symbols demodulated after thefirst CPE correction in a DFE manner to improve the quality of the estimate since that is increasing the number of observations of the quantity we want to estimate.In our case that would mean that after thefirst CPE correction the set L={l1,l2,l3,...l P}of the subcarriers used for CPE estimation,which previously corresponded to pilot subcarriers,is now extended to a larger set corresponding to all or some of the demodulated symbols. In this paper we have extended the set to all demodulated symbols.The Kalmanfilter estimation is then applied in an unchanged form for a larger set L.VI.N UMERICAL R ESULTSThe performance of the proposed algorithm is investigated and compared with the proposal of[4]which is shown to outperform other known approaches.The system model is according to the IEEE802.11a standard,where64-QAM modulation is used.We investigate the performance in AWGN channels and frequency selective channels using as an example the ETSI HiperLAN A-Channel(ETSI A). Transmission of10OFDM symbols per burst is assumed.A.Properties of an EstimatorThe quality of an estimation is investigated in terms of the mean square error(MSE)of the estimator for a range of phase noise bandwidths∆f3dB∈[10÷800]Hz.Table1 can be used to relate the phase noise bandwidth with other quantities.Figures2and3compare the MSE of the LS estimator from[4]and our approach for two channel types and both standard correction and using decision feedback. Note that SNRs are chosen such that the BER of a coded system after the Viterbi algorithm in case of phase noise free transmission is around1·10−4.Kalmanfilter shows better performance in all cases and seems to be more effective for small phase noise bandwidths. As expected when DF is used the MSE of an estimator is smaller because we are taking more measurements into account.Fig.2MSE of an estimator for AWGN channel.Fig.3MSE of an estimator for ETSI A channel.Table 1Useful relationsQuantitySymbolRelationTypical values for IEEE802.11aOscillator constant c [1radHz]8.2·10−19÷4.7·10−18Oscillator 3dB bandwidth ∆f 3dB [Hz]∆f 3dB =πf 2cc 70÷400Relative 3dB bandwidth ∆f 3dB ∆f car∆f 3dBfsN 2·10−4÷13·10−4Phase noise energy E PN [rad]E PN =4π∆f 3dB∆fcar0.0028÷0.016Subcarrier spacing∆f car∆f car =f s N312500HzB.Symbol Error Rate DegradationSymbol error rate (SER)degradation due to phase noise is investigated also for a range of phase noise bandwidths ∆f 3dB ∈[10÷800]Hz and compared for different correc-tion algorithms.Ideal CPE correction corresponds to the case when genie CPE values are available.In all cases simpleconstellation derotation with φ=−arg {ˆIm (0)}is used.Fig.4SER degradation for AWGN channel.In Figs.4and 5SER degradation for AWGN and ETSI A channels is plotted,respectively.It is interesting to note that as opposed to the ETSI A channel case in AWGN channel there is a gap between the ideal CPE and both correction approaches.This can be explained if we go back to Eq.(1)where we have seen that phase noise affects the constellation as additive noise.Estimation error of phase noise affects the constellation also in an additive manner.On the other hand the SER curve without phase noise in the AWGN case is much steeper than the corresponding one for the ETSI A channel.A small SNR degradation due to estimation errors will cause therefore large SER variations.This explains why the performance differs much less in the ETSI A channel case.Generally from this discussion a conclusion can be drawn that systems with large order of diversity are more sensitive to CPE estimation errors.Note that this ismeantFig.5SER degradation for ETSI A channel.not in terms of frequency diversity but the SER vs.SNR having closely exponential dependence.It can be seen that our approach shows slightly better performance than [4]especially for small phase noise bandwidths.What is also interesting to note is,that DF is not necessary in the case of ETSI A types of channels (small slope of SER vs.SNR)while in case of AWGN (large slope)it brings performance improvement.VII.C ONCLUSIONSWe investigated the application of a linear Kalman filter as a means for tracking phase noise and its suppression.The proposed algorithm is of low complexity and its performance was studied in terms of the mean square error (MSE)of an estimator and SER degradation.The performance of an algorithm is compared with other algorithms showing equivalent and in some cases better performance.R EFERENCES[1]R.A.Casas,S.Biracree,and A.Youtz,“Time DomainPhase Noise Correction for OFDM Signals,”IEEE Trans.on Broadcasting ,vol.48,no.3,2002.[2]M.S.El-Tanany,Y.Wu,and L.Hazy,“Analytical Mod-eling and Simulation of Phase Noise Interference in OFDM-based Digital Television Terrestial Broadcast-ing Systems,”IEEE Trans.on Broadcasting,vol.47, no.3,2001.[3]P.Robertson and S.Kaiser,“Analysis of the effects ofphase noise in OFDM systems,”in Proc.ICC,1995.[4]S.Wu and Y.Bar-Ness,“A Phase Noise SuppressionAlgorithm for OFDM-Based WLANs,”IEEE Commu-nications Letters,vol.44,May1998.[5]D.Petrovic,W.Rave,and G.Fettweis,“Phase NoiseSuppression in OFDM including Intercarrier Interfer-ence,”in Proc.Intl.OFDM Workshop(InOWo)03, pp.219–224,2003.[6]A.Armada,“Understanding the Effects of PhaseNoise in Orthogonal Frequency Division Multiplexing (OFDM),”IEEE Trans.on Broadcasting,vol.47,no.2, 2001.[7]E.Costa and S.Pupolin,“M-QAM-OFDM SystemPerformance in the Presence of a Nonlinear Amplifier and Phase Noise,”IEEE mun.,vol.50, no.3,2002.[8]A.Demir,A.Mehrotra,and J.Roychowdhury,“PhaseNoise in Oscillators:A Unifying Theory and Numerical Methods for Characterisation,”IEEE Trans.Circuits Syst.I,vol.47,May2000.[9]S.Wu and Y.Bar-ness,“Performance Analysis of theEffect of Phase Noise in OFDM Systems,”in IEEE 7th ISSSTA,2002.[10]D.Petrovic,W.Rave,and G.Fettweis,“Phase NoiseSuppression in OFDM using a Kalman Filter,”in Proc.WPMC,2003.[11]S.M.Kay,Fundamentals of Statistical Signal Process-ing vol.1.Prentice-Hall,1998.[12]D.J.Higham,“An Algorithmic Introduction to Numer-ical Simulation of Stochastic Differential Equations,”SIAM Review,vol.43,no.3,pp.525–546,2001.。
Interplay of friction and noise and enhancement of disoriented chiral condensate
a r X i v :h e p -p h /0011003v 1 1 N o v 2000Interplay of friction and noise and enhancement of disoriented chiral condensateA.K.Chaudhuri ∗Variable Energy Cyclotron Centre 1-AF,Bidhan Nagar,Calcutta-700064Using the Langevin equation for the linear σmodel,we have investigated the effect of friction and noise on the possible disoriented chiral condensate formation.Friction and noise are supposed to suppress longwavelength oscillations and growth of disoriented chiral condensate domains.Details simulation shows that for heavy ion collisions,interplay of friction and noise occur in such a manner that formation of disoriented chiral condensate domains are enhanced.25.75.+r,12.38.Mh,11.30.RdIn recent years there is much excitement about the possibility of formation of disoriented chiral condensate (DCC).The idea was suggested by Rajagopal and Wilczek [1]and also by Bjorken and others [2].In hadron-hadron or in heavy ion collisions,a macroscopic region of space-time may be created within which the chiral order parameter is not oriented in the same direction in the internal O (4)=SU (2)×SU (2)space as the ordinary vacuum.The misaligned condensate has the same quark content and quantum numbers as do pions and essentially constitute a classical pion field.The system will finally relaxes to the true vacuum and in the process can emit coherent pions.Possibility of producing classical pion fields in heavy ion collisions had been discussed earlier by Anslem [3].DCC formation in hadronic or in heavy ion collisions can lead to the spectacular events that some portion of the detector will be dominated by charged pions or by neutral pions only.In contrast,in a general event all the three pions (π+,π−and π0)will be equally well produced.This may then be the natural explanation of the so called Centauro events [4].Microscopic physics governing DCC phenomena is not well known.It is in the regime of non-perturbative QCD as well as nonlinear phenomena,theoretical understanding of both of which are limited.One thus uses some effective field theory like linear σmodel with various approximations to simulate the chiral phase transition [5–7].In the linear sigma model chiral degrees of freedom are described by the the real O(4)field Φ=(σ,π→).Because of the isomorphism between the groups O (4)and SU (2)×SU (2),the later being the appropriate group for two flavour QCD,linear sigma model can effectively model the low energy dynamic of QCD [8].Explicit simulation with linear sigma model,indicate that DCC depends critically on the initial field configurations.With quench like initial condition DCC domains of 4-5fm in size can form [7].Initial conditions other than quench lead to much smaller domain size.Quench scenario assume that the effective potential governing the evolution of long wave length modes immediately after the phase transition at T c turns to classical one at zero temperature.It can happen only in case of very rapid cooling and expansion of the fireball.In heavy ion collisions quench like initial conditions are unlikely.Very recently,effect of external media on possible DCC is being investigated [9–16].Indeed,in heavy ion collisions,even if some region is created where chiral symmetry is restored,that region will be continually interacting with surrounding medium (mostly pions).The surrounding medium can be conveniently modelled by a white noise source and one can use Langevin equation for linear σmodel to simulate the DCC formation under the influence of external media.Recently it has been shown that in φ4model,hard modes can be integrated out to obtain a Langevin type of equations for the soft modes [9].Biro and Greiner [10]using a Langevin equation for the linear σmodel,investigated the interplay of friction and white noise on the evolution and stability of zero mode pion fields.In general friction and noise reduces the amplification of zero modes.But in some trajectories,large amplification may occur [10].We also obtain similar results [13].It is popular expectation that friction and noise will reduce the DCC formation probability.The expectation was found to be true in one dimensional calculations with zero modes only [10,13].To see how far this expectation is valid when higher modes are included,in the present letter we have investigated DCC formation in 3+1dimension.Two scenarios were considered.In scenario I,we use the equation of motion of linear σmodel with quenched initial condition to simulate DCC formation without friction and noise.In scenario II,we solve the Langevin equation for linear σmodel,with the same quenched initial condition to simulate DCC formation in presence of friction and parative study of these two scenarios reveal that certainly the noise and friction affect DCC formation probability.However contrary to one dimensional calculations,large DCC formation is seen in scenario II rather than in scenario I.Indeed,it will be shown that large DCC domains are formed in scenario II only,not in scenario I.This effect is particular to heavy ion collisions as explained below.The Langevin equation for linear σmodel,at temperature T can be written as,[∂2τ+ηi )∂∂x 2−∂2τ2∂2ion collisions.ησ,πandζσ,πare the friction coefficients and the white noise for theσandπfields.We note that if the friction and the noise terms are dropped from eq.1the resulting equation is for the scenario I,i.e.just the equation of motion of linearσmodel.The noise sourceζand the frictionηare not independent.They are related by thefluctuations-dissipation relation. We use white noise source with zero average and correlation as demanded from thefluctuation-dissipation relation,<ζa(τ,x,y,Y)>=0(2a) <ζa(τ,x,y,Y)ζb(τ′,x′,y′,Y′)>=2Tη1N )2m2σ1−4m2π1−exp(−m2σ/2mπT)1N )2N−11−4m2π4T(3b)Solution of eq.1require initialfields configurations.They were distributed according to a random Gaussian with,<σ>=(1−f(r))fπ(4a)<πi>=0(4b)<σ2>−<σ>2=<π2i>−<πi>2=f2π/6f(r)(4c)<˙σ>=<˙πi>=0(4d)<˙σ2>=<˙π2>=4f2π/6f(r)(4e)The interpolation functionf(r)=[1+exp(r−r0)/Γ)]−1(5) separates the central region from the rest of the system.We have taken r0=6.4fm andΓ=.7fm.The initialfield configurations corresponds to quench like condition[1]and as told in the beginning are unlikely to be obtained in a heavy ion collisions.We still use it as they are the most favourable initial conditions to produce DCC like phenomena. In the simulation results presented below,the initial time and temperature were assumed to beτi=1fm/c and T i=200 MeV.Effect of expansion was included through the cooling law,T(τ)=T iτii,j|π(i)||π(j)|(7)where the sum is taken over those grid points i and j such that the distance between i and j is r.Infig.1,we have compared the correlation function in scenario I and II.Initially there is no correlation length beyond the lattice spacing of1fm.Correlations starts to develop at later time.It increases for about7fm/c,then decreases again. Interestingly,larger correlation length is obtained in the scenario II,than in scenario I.Thus at7fm/c,correlation length in scenario I is only∼2fm,while that in scenario II is∼4fm.Increased correlation in scenario II is contrary to popular expectation that friction and noise will reduce correlation.Enhancement of correlation length in scenario II,with friction and noise is corroborated in the pionfield distribution also.Infig.2,we have shown the xy contour plot of theπ2component,at rapidity Y=0.Field distribution atτi=1 fm/c and after7fm/c of evolution are shown.Initially there is no correlation.Domain like structure is seen both in scenario I and II,after7fm/c evolution.The positive and negative components of theπ2separates out.Here again, much larger domains are formed in scenario II than in scenario I.It may be noted that domain like structure seen in one of the component ofπfield do not necessarily convert into physical domains.They are indication of larger correlation length only.Physical domain should contain either charged or neutral pion only.Thus in physical domain neutral to total pion ratio should differ considerably from the isospin symmetric value of1/3.Assuming that the pion density is proportional to the amplitude’s square,infig.3,we have shown the contour plot of the neutral to total pion ratio,at rapidity Y=0.π23R3(x,y,Y,τ)== π23dxdyNπtotalin one component of theπfield,there is no(large)physical domain formation in scenario I.On the contrary,large physical domains are seen to be formed in scenario II after7fm/c of evolution.We have also studied the rapidity distribution of the neutral to pion ratio in both the scenarios.Throughout the rapidity range,the ratiofluctuate about the isospin symmetric value of1/3,fluctuations being more in scenario II than in scenario I.However,it was also noted thatfluctuations will be considerably less when integrated over time.Rapidity distribution of neutral to total pion ratio may not be able to single out the DCC events.for7fm/c.shown.。
Teleportation and spin squeezing utilizing multimode entanglement of light with atoms
a r X i v :q u a n t -p h /0511174v 1 17 N o v 2005Teleportation and spin squeezing utilizing multimode entanglement of light withatomsK.Hammerer 1,E.S.Polzik 2,3,J.I.Cirac 11Max-Planck–Institut f¨u r Quantenoptik,Hans-Kopfermann-Strasse,D-85748Garching,Germany2QUANTOP,Danish Research Foundation Center for Quantum Optics,DK 2100Copenhagen,Denmark3Niels Bohr Institute,DK 2100Copenhagen,Denmark We present a protocol for the teleportation of the quantum state of a pulse of light onto the collective spin state of an atomic ensemble.The entangled state of light and atoms employed as a resource in this protocol is created by probing the collective atomic spin,Larmor precessing in an external magnetic field,offresonantly with a coherent pulse of light.We take here for the first time full account of the effects of Larmor precession and show that it gives rise to a qualitatively new type of multimode entangled state of light and atoms.The protocol is shown to be robust against the dominating sources of noise and can be implemented with an atomic ensemble at room temperature interacting with free space light.We also provide a scheme to perform the readout of the Larmor precessing spin state enabling the verification of successful teleportation as well as the creation of spin squeezing.PACS numbers:03.67.Mn,32.80.QkI.INTRODUCTIONQuantum teleportation -the disembodied transport of quantum states -has been demonstrated so far in sev-eral seminal experiments dealing with purely photonic [1]or atomic [2]systems.Here we propose a protocol for the teleportation of a coherent state carried initially by a pulse of light onto the collective spin state of ∼1011atoms.This protocol -just as the recently demonstrated direct transfer of a quantum state of light onto atoms [3]-is particularly relevant for long distance entanglement distribution,a key resource in quantum communication networks [4].Our scheme can be implemented with just coherent light and room-temperature atoms in a single vapor cell placed in a homogeneous magnetic field.Existing proto-cols in Quantum Information (QI)with continuous vari-ables of atomic ensembles and light [4]are commonly de-signed for setups where no external magnetic field is ap-plied such that the interaction of light with atoms meets the Quantum non-demolition (QND)criteria [5,6].In contrast,in all experiments dealing with vapor cells at room-temperature [3,7]it is,for technical reasons,ab-solutely essential to employ magnetic fields.In experi-ments [3,7]two cells with counter-rotating atomic spins were used to comply with both,the need for an exter-nal magnetic field and the one for an interaction of QND character.So far it was believed to be impossible to use a single cell in a magnetic field to implement QI protocols,since in this case -due to the Larmor precession -scat-tered light simultaneously reads out two non-commuting spin components such that the interaction is not of QND type.In this paper we do not only show that it is well possible to make use of the quantum state of light and atoms created in this setup but we demonstrate that -for the purpose of teleportation [8,9]-it is in factbetter to do so.As compared to the state resulting from the common QND interaction the application of an external magnetic field enhances the creation of correlations between atoms and light,generating more and qualitatively new,multimode type of entanglement.The results of the paper can be summarized as follows:(i )Larmor precession in an external magnetic field enhances the creation of entanglement when a collective atomic spin is probed with off-resonant light.The resulting entanglement involves multiple modes and is stronger as compared to what can be achieved in a comparable QND interaction.(ii )This type of entangled state can be used as a resource in a teleportation protocol,which is a simple generalization of the standard protocol [8,9]based on Einstein-Podolsky-Rosen (EPR)type of entanglement.For the experimentally accessible parameter regime the teleportation fidelity is close to optimal.The protocol is robust against imperfections and can be implemented with state of the art technique.(iii )Homodyne detection of appropriate scattering modes of light leaves the atomic state in a spin squeezed state.The squeezing can be the same as attained from a comparable QND measurement of the atomic spin [10,11].The same scheme can be used for atomic state read-out of the Larmor precessing spin,necessary to verify successful teleportation.We would like to note that it was shown recently in [12]that the effect of a magnetic field can enhance the capacity of a quantum memory in the setup of two cells.Teleportation in the setup of a single cell without mag-netic field was addressed in [13].The paper is organized as follows:The three points above are presented in sections II,III and IV,in this order.Some of the details in the calculations of sections III and IV are moved to appendices B and C.2II.INTERACTIONWe consider an ensemble of N at Alkali atoms with total ground state angular momentum F ,placed in a constant magnetic field causing a Zeeman splitting of Ωand ini-tially prepared in a fully polarized state along x .The collective spin of the ensemble is then probed by an offresonant pulse which propagates along z and is linearly polarized along x .Thorough descriptions of this inter-action and the final state of light and atoms after the scattering can befoundin [14,15,16,17,18]and espe-cially in [19,20,21,22]for the specific system we have in mind.We derive the final state here with a special focus on the effects of Larmor precession and light prop-agation in order to identify the light modes which are actually populated in the scattering process.In appendix A we show that the interaction is ade-quately described by a HamiltonianH =H at +H li +V,H at =Ω√N ph N at F a 1σΓ/2A ∆where N ph is the over-all number of photons in the pulse,a 1is a constant characterizing the ground state’s vector polarizability,σis the scattering cross section,Γthe decay rate,∆the detuning and A the effective beam cross section.Changing to a rotating frame with respect to H at by defining X I (t )=exp(−iH at t )X exp(iH at t )and evaluat-ing the Heisenberg equations for these operators yields the following Maxwell-Bloch equations∂t X I (t )=κT cos(Ωt )p (0,t ),(2a)∂t P I (t )=κT sin(Ωt )p (0,t ),(2b)(∂t +c∂z )x (z,t )=κcT[cos(Ωt )P I (t )−sin(Ωt )X I (t )]δ(z ),(∂t +c∂z )p (z,t )=0,where ∂t (z )denotes the partial derivative with respect to t (z ).These equations have a clear interpretation.Light noise coming from the field in quadrature with the classi-cal probe piles up in both,the X and P spin quadrature,but it alternately affects only one or the other,changing with a period of 1/Ω.Conversely atomic noise adds to the in phase field quadrature only and the signal comes alternately from the X and P spin quadrature.The out of phase field quadrature is conserved in the interaction.To solve this set of coupled equations it is convenient to introduce a new position variable,ξ=ct −z ,to elim-inate the z dependence.New light quadratures defined by ¯x (ξ,t )=x (ct −ξ,t ),¯p (ξ,t )=p (ct −ξ,t )also have a simple interpretation:ξlabels the slices of the pulse moving in and out of the ensemble one after the other,starting with ξ=0and terminating at ξ=cT .The Maxwell equations now read∂t ¯p (ξ,t )=0,(2c)∂t ¯x (ξ,t )=κcT[cos(Ωt )P I (t )−sin(Ωt )X I (t )]δ(ct −ξ).(2d)The solutions to equations (2a,2b,2c)areX I (t )=X I (0)+κT t 0d τcos(Ωτ)¯p (cτ,0),(3a)P I (t )=P I (0)+κT t 0d τsin(Ωτ)¯p (cτ,0),(3b)¯p (ξ,t )=¯p (ξ,0)(3c)and the formal solution to (2d)is¯x (ξ,t )=¯x (ξ,0)+(3d)+κT [cos(Ωξ/c )P I (ξ/c )−sin(Ωξ/c )X I (ξ/c )].As mentioned before,both atomic spin quadratures are affected by light but,as is evident from the solutions for X (t ),P (t ),they receive contributions from different and,in fact,orthogonal projections of the out-of-phase field.As we will show in the following,the corresponding projections of the in-phase field carry in turn the signal of atomic quadratures after the interaction.It is there-fore convenient to explicitly introduce operators for these modes [21].We define a cosine component before the in-teractionp in c = T Td τcos(Ωτ)¯p (cτ,0),(4a)x in c =TTd τcos(Ωτ)¯x (cτ,0)(4b)and a sine component p in s ,x ins with cos(Ωτ)replaced by sin(Ωτ).In frequency space these modes consist of spec-tral components at sidebands ωc ±Ωand are closely re-lated to the sideband modulation modes introduced in3[26]for the description of two photon processes.It is eas-ily checked that these modes are asymptotically canon-ical,[x in c ,p in c ]=[x in s ,p ins ]=i [1+O (n −10)]≃i ,and inde-pendent,[x in c ,p in s ]=O (n −10)≃0,if we assume n 0≫1for n 0=ΩT ,the pulse length measured in periods of Larmor precession.In terms of these modes the atomic state after the in-teraction X out =X I (T ),P out =P I (T )is given by X out =X in +κ2p inc ,P out =P in +κ2p ins .(5a)The final state of cosine (sine)modes is de-scribed by x out c(s),p outc(s),defined by equations (4)with ¯x (cτ,0),¯p (cτ,0)replaced by ¯x (cτ,T ),¯p (cτ,T )respec-tively.Since the out-of-phase field is conserved we have triviallyp out c =p in c ,p out s =p in s .(5b)Deriving the corresponding expressions for the cosine andsine components of the field in phase,x out c ,x outs ,raises some difficulties connected to the back action of light onto itself.This effect can be understood by noting that a slice ξof the pulse receives a signal of atoms at a time ξ/c [see equation (3d)]which,regarding equations (3a,3b),in turn carry already the integrated signal of all slices up to ξ.Thus,mediated by the atoms,light acts back on itself.The technicalities in the treatment of this effect are given in appendix B where we identify relevant ”back action modes”,x c ,1,p c ,1,x s ,1,p s ,1,in terms of which one can express x out c =x in c +κ2P in +κ√2 2p in s ,1,(5c)x out s=x in s −κ2X in−κ√2 2p in c ,1.(5d)The last two terms in both lines represent the effect of back action,part of which involves the already defined cosine and sine components of the field in quadrature.The remaining part is subsumed in the back action modes which are again canonical and independent from all other modes.Equations (5)describe the final state of atoms and the relevant part of scattered light after the pulse has passed the atomic ensemble and are the central result of this section.Treating the last terms in equations (5c,5d)as noise terms,it is readily checked by means of the separa-bility criteria in [27]that this state is fully inseparable,i.e.it is inseparable with respect to all splittings be-tween the three modes.For the following teleportation protocol the relevant entanglement is the one between atoms and the two light modes.Figure 1shows the von Neumann entropy E vN of the reduced state of atoms in its dependence on the coupling strength κand in com-parison with the entanglement created without magnetic field in a pure QND interaction of atoms and light.The amount of entanglement is significantly enhanced.FIG.1:Von Neumann Entropy of the reduced state of atoms versus coupling strength kappa for the state of equation 5(full line)and for the state generated without magnetic field in a pure QND interaction (dashed line)with the same cou-pling strength.Application of a magnetic field significantly enhances the amount of light-atom entanglement.III.TELEPORTATION OF LIGHT ONTOATOMSIn this section we will show how the multimode entan-glement between light and atoms generated in the scat-tering process can be employed in a teleportation proto-col which is a simple generalization of the standard proto-col for continuous variable teleportation using EPR-type entangled states [8,9].We first present the protocol and evaluate its fidelity and then analyze its performance un-der realistic experimental conditions.A.Basic protocolFigure 2depicts the basic scheme which,as usually,consists of a Bell measurement and a feedback operation.Input The coherent state to be teleported is encoded in a pulse which is linearly polarized orthogonal to the classical driving pulse and whose carrier frequency lies at the upper sideband,i.e.at ωc +Ω.The pulse envelope has to match the one of the classical pulse.As is shown in appendix B,canonical operators y,q with [y,q ]=i de-scribing this mode can conveniently be expressed in terms of cosine and sine modulation modes,analogous to equa-tions (4),defined with respect to the carrier frequency.One findsy =12(y s +q c ),q =−12(y c −q s ).(6)A coherent input amounts to having initially ∆y 2=∆q 2=1/2and an amplitude y , q with mean photon number n ph =( y 2+ q 2)/2.Bell measurement This input is combined at a beam splitter with the classical pulse and the scattered light.At the ports of the beam splitter Stokes vector compo-nents S y and S z are measured by means of standard po-larization measurements.Given the classical pulse in xFIG.2:Scheme for teleportation of light onto atoms:As de-scribed in section II,a classical pulse(linearly polarized along x)propagating along the positive z direction is scattered offan atomic ensemble contained in a glass cell and placed in a constant magneticfield B along x.Classical pulse and scat-tered light(linearly polarized along y)are overlapped with a with a coherent pulse(linearly polarized along z)at beam splitter B S.By means of standard polarization measurements Stokes vector components S y and S z are measured at one and the other port respectively,realizing the Bell measurement. The Fourier components at Larmor frequencyΩof the corre-sponding photocurrents determine the amount of conditional displacement of the atomic spin which can be achieved by ap-plying a properly timed transverse magneticfield b(t).See section III A for details.polarization this amounts to a homodyne detection of in-and out-of-phasefields of the orthogonal polarization component.The resulting photocurrents are numerically demodulated to extract the relevant sine and cosine com-ponents at the Larmor frequency[20].Thus one effec-tively measures the commuting observables˜x c=12 x out c+y c,˜x s=12 x out s+y s ,(7)˜q c=12 p out c−q c ,˜q s=12 p out s−q s .Let the respective measurement results be given by ˜Xc,˜X s,˜Q c and˜Q s.Feedback Conditioned on these results the atomic state is then displaced by an amount˜X s−˜Q c in X and −˜X c−˜Q s in P.This can be achieved by means of two fast radio-frequency magnetic pulses separated by a quar-ter of a Larmor period.In the ensemble average thefinal state of atoms is simply given byXfin=X out+˜x s−˜q c,Pfin=P out−˜x c−˜q s.(8)This description of feedback is justified rigorously in ap-pendix C.Relating these expressions to input operators,wefind by means of equations(5),(6)and(7)Xfin= 1−κ√2 2p in c+12x in s−16 κ2 P in−121−κ√√2 2p in s,1+q.(9b)This is the main result of this section.Teleportationfidelity Taking the mean of equations (9)with respect to the initial state all contributions due to input operators and back action modes vanish such that Xfin = y and Pfin = q .Thus,the am-plitude of the coherent input light pulse is mapped on atomic spin quadratures as desired.In order to prove faithful teleportation also the variances have to be con-served.It is evident from(9)that thefinal atomic spin variances will be increased as compared to the coherent input.These additional terms describe unwanted excess noise and have to be minimized by a proper choice of the couplingκ.As afigure of merit for the telepor-tation protocol we use thefidelity,i.e.squared over-lap,of input andfinal state.Given that the means are transmitted correctly thefidelity is found to be F= 2 (1+2(∆Xfin)2)(1+2(∆Pfin)2) −1/2.The variances of thefinal spin quadratures are readily calculated tak-ing into account that all modes involved are independent and have initially a normalized variance of1/2.In this way a theoretical limit on the achievablefidelity can be derived depending solely on the coupling strengthκ.In figure3we take advantage of the fact that the amount of entanglement between light and atoms is a monotonously increasing function ofκsuch that we can plot thefi-delity versus the entanglement.This has the advantage that we can compare the performance of our teleporta-tion protocol with the canonical one[8,9]which uses a two-mode squeezed state of the same entanglement as a resource and therefore maximizes the teleportationfi-delity for the given amount of entanglement.No physical state can achieve a higherfidelity with the same entan-glement.This follows from the results of[30]where it was shown that two-mode squeezed states minimize the EPR variance(and therefore maximize the teleportation fidelity)for given entanglement.The theoreticalfidelity achievable in our protocol is maximized forκ≃1.64cor-responding to F≃.77.But also for experimentally more feasible values ofκ≃1can thefidelity well exceed the classical limit[28,29]of1/2and,moreover,compari-son with the values achievable with a two-mode squeezed state shows that our protocol is close to optimal.FIG.3:(a)Theoretical limit on the achievablefidelity F ver-sus entanglement between atoms and light measured by the von Neumann entropy E vN of the reduced state of atoms.The grey area is unphysical.For moderate amounts of entangle-ment our protocol is close to optimal.(b)Coupling strengthκversus entanglement.The dashed lines indicate the maximal fidelity of F=.77which is achieved forκ=1.64.B.Noise effects and Gaussian distributed input Under realistic conditions the teleportationfidelity will be degraded by noise effects like decoherence of the atomic spin state,light absorption and reflection losses and also because the coupling constantκis experimen-tally limited to valuesκ≃1.On the other hand the classicalfidelity bound to be beaten will be somewhat higher than1/2since the coherent input states will nec-essarily be drawn according to a distribution with afinite width in the mean photon number¯n.In this section we analyze the efficiency of the teleportation protocol un-der these conditions and show that it is still possible to surpass any classical strategy for the transmission and storage of coherent states of light[28,29].During the interaction atomic polarization decays due to spontaneous emission and collisional relaxation.In-cluding a transverse decay thefinal state of atoms is given byX out= √βf X,(10a) P out= √βf P.(10b)as follows from the discussion in appendix A.βis the atomic decay parameter and f X,f P are Langevin noise operators with zero mean.Their variance is experimen-tally found to be close to the value corresponding to a coherent state such that f2X = f2P =1/2.Light absorption and reflection losses can be taken into account in the same way asfinite detection efficiency.For example the statistics of measurement outcome˜X s will not stem from the signal mode˜x s alone but rather from the noisy mode√ǫfx,swhereǫis the photon loss parameter and f x,s is a Langevin noise operator of zero mean and variance f2x,s =1/2.Analogous expres-sions have to be used for the measurements of˜x c,˜q s and ˜q c which will be adulterated by Langevin terms f x,c,f q,s and f q,c respectively.In principle each of the measure-ment outcomes can be fed back with an independently chosen gain but for symmetry reasons it is enough to distinguish gain coefficients g x,g q for the measurement outcomes of sine and cosine components of x and q re-spectively.Including photon loss,finite gain and atomic decay,as given in(10),equations(8),describing thefinal state of atoms after the feed back operation,generalize toXfin= βf X+g x √ǫf x,s (11a)−g q √ǫf q,c ,Pfin= βf P−g x √ǫf x,c (11b)−g q √ǫf q,s .For non unit gains a given coherent amplitude( y , q ) will not be perfectly teleported onto atoms and the cor-respondingfidelity will be degraded by this mismatch according toF( y , q )=2[1+2(∆Xfin)2][1+2(∆Pfin)2]·exp−( y − Xfin )21+2(∆Pfin)2 .If the input amplitudes are drawn ac-cording to a Gaussian distribution p( y , q )=exp[−( y 2+ q 2)/2¯n]/2π¯n with mean photon number¯n the averagefidelity[with respect to ( y , q )]is readily calculated.The exact expression in terms of initial operators can then be derived by means of equations(5),(6),(7)and(11)but is not particularly enlightening.Infigure4we plot the averagefidelity, optimized with respect to gains g x,g q,in its dependence on the atomic decayβfor various values of photon loss ǫ.We assume a realistic valueκ=0.96for the coupling constant and a mean number of photons¯n=4for the distribution of the coherent input.For feasible values ofβ,ǫ 0.2the averagefidelity is still well above the classical bound on thefidelity[28,29].This proves that the proposed protocol is robust against the dominating noise effects in this system.The experimental feasibility of the proposal is illus-trated with the following example.Consider a sam-ple of N at=1012Cesium atoms in a glass cell placed in a constant magneticfield along the x-direction caus-ing a Zeeman splitting ofΩ=350kHz in the F=4 ground state multiplet.The atoms are pumped into m F=4and probed on the D2(F=4→F′=3,4,5)FIG.4:(a)Averagefidelity achievable in the presenceof atomic decayβ,reflection and light absorption lossesǫ=8%,12%,16%,couplingκ=0.96and Gaussian dis-tributed input states with mean photon number¯n=4.Thefi-delity benchmark is in this case5/9(dashed line).(b)Respec-tive optimal values for gains g x(solid lines)and g q(dashedlines).transition.The classical pulse contains an overall num-ber of N ph=2.51013photons,is detuned to the blueby∆=1GHz,has a duration T=1ms and can have an effective cross section of A≃6cm2due to thermalmotion of atoms.Under these conditions the tensor polarizability can be neglected(∆/ωhfs≃10−1).Also n0=ΩT=350justifies the use of independent scatter-ing modes.The couplingκ≃1and the depumping ofground state populationη≃10−1as desired.IV.SPIN SQUEEZING AND STATE READ-OUTIn this section we present a scheme for reading out either of the atomic spin components X,P by means of a probe pulse interacting with the atoms in the one way as described in section II.The proposed scheme allows one,on the one hand,to verify successful receipt of the coherent input subsequent to the teleportation protocol of section III and,on the other hand,enables to generate spin squeezing if it is performed on a coherent spin state. It is well known[15,31]and was demonstrated exper-imentally[10,11]that the pure interaction V,as given in equation(1),can be used to perform a QND measure-ment of either of the transverse spin components.At first sight this seems not to be an option in the scenario under consideration since the local term H at,account-ing for Larmor precession,commutes with neither of the spin quadratures such that the total Hamiltonian does not satisfy the QND criteria[5,6].As we have shown in section II Larmor precession has two effects:Scattered light is correlated with both transverse components andsuffers from back action mediated by the atoms.Thus, in order to read out a single spin component one has to overcome both disturbing effects.Our claim is that this can be achieved by a simultane-ous measurement of x outc,p outs,p outs,1or x outs,p outc,p outc,1 if,respectively,X or P is to be measured.In the follow-ing we consider in particular the former case but every-thing will hold with appropriate replacements also for a measurement of P.As shown infigure5the set of observablesx outc,p outs,p outs,1can be measured simultaneously by a measurement of Stokes component S y after aπ/2rota-tion is performed selectively on the sine component of the scattered light.The cosine component of the corre-sponding photocurrent will give an estimate of x outcandthe sine component of p outs.Multiplying the photocur-rent’s sine component by the linear function defining the back action mode,equation(B1),will give in additionan estimate of p outs,1.Note that thefield out of phase is conserved in the interaction such thatp outs,1=p in s,1,p outc,1=p in c,1,(12) i.e.the results will have shot noise limited variance.It is then evident from equation(5c)that the respective photocurrents together with an a priori knowledge ofκare sufficient to estimate the mean X .The conditional variances after the indicated measure-ments are∆X2|{x out c,p out s,p out s,1}=(∆X in)222,(13b)corresponding to a pure state.Obviously the variance in X is squeezed by a factor(1+κ2/2)−1.Note that the squeezing achieved in a QND measurement without magneticfield but otherwise identical parameters is given by(1+κ2)−1.From this we conclude that the quality of the estimate for X ,as measured f.e.by input-output coefficients known from the theory of QND measurements [5,6],can be the same as in the case without Larmor precession albeit only for a higher couplingκ. Equations(13)are conveniently derived by means of the formalism of correlation matrices [32].For the operator valued vector R=(X,P, x c,p c,x s,p s,x c,1,p c,1,x s,1,p s,1)equations(5),(12) and(B2)define via R out=S(κ) R in a symplectic linear transformation S(κ).The contributions of p in c,2andp in s,2to x outs,1and x outc,1as given in(B2)are treated as noise and do not contribute to the symplectic trans-formation S but enter the input-output relation for the correlation matrix as an additional noise term as follows.The correlation matrix is as usually defined byγi,j=tr{ρ(R i R j+R j R i)}.The initial state is then an10×10identity matrix and thefinal state is γout=S(κ)S(κ)T+γnoise where the diagonal matrixFIG.5:Scheme for spin measurement:After the scattering a π/2rotation is performed on the scattered light modulated at the Larmor frequency such as to affect only the sine(cosine) component.Standard polarization measurement of S y and appropriate postprocessing allows to read out the mean of X(P),leaving the atoms eventually in a spin squeezed state.γnoise=diag[0,0,0,0,0,0,1,0,1,0](κ/2)4/15accountsfor noise contributions due correlations to second order back action modes c.f.equations(B2).In order to evaluate the atomic variances after a measurement ofx out c ,p outs,p outs,1the correlation matrixγout is split upinto blocks,γout= A C C T Bwhere A is the2×2subblock describing atomic variances. Now,the state A′after the measurement can be found by evaluating[32]A′=A−limx,n→∞C1(∆−iΓ/2) a01+ia1 F×(A3)8 with real dimensionless coefficients a j of order unity andΓthe excited states’decay rate.The non-hermitian partof the resulting Hamilton operator describes the effectof light absorption and loss of ground state populationdue to depumping in the course of interaction.In thefollowing we will focus on the coherent interaction and,for the time being,take into account only the hermitiancomponent.The effects of light absorption and atomicdepumping are treated below.Coherent interaction Since scattering of light occurspredominantly in the forward direction[22]it is legiti-mate to adopt a one dimensional model such that the(negative frequency component of the)electricfield prop-agating along z is given byE(−)(z,t)=E(−)(z) ey+E(−)(z,t) e xE(−)(z)=ρ(ωc) b dωa†(ω)e−ikzE(−)(z,t)=ρ(ωc)ωc/4πǫ0Ac and A denotes the pulse’scross sectional area,N ph the overall number of photonsin the pulse and T its duration.We restrict thefieldin x polarization to the classical probe pulse,since onlythe coupling of atoms to the y polarization is enhancedby the coherent probe.Furthermore we implicitly assumefor the classical pulse a slowly varying envelope such thatit arrives at z=0at t=0and is then constant fora time bining this expression for thefield withexpressions(A2)and(A3)for the atomic polarizability inequation(A1)yieldsV=−i κ4πJTbdω d zj(z) a(ω)e−i[(k c−k)z−ωc t]−h.c.where we defined a dimensionless coupling constant κ=N ph Ja1σΓ/2A∆withσthe scattering cross sec-tion on resonance.We now definefield quadratures for spatially localized modes[24,25]asx(z)=14πbdω a(ω)e−i(k c−k)z+h.c. ,(A4a)p(z)=−i4πbdω a(ω)e−i(k c−k)z−h.c. (A4b)with commutation relations[x(z),p(z′)]=icδ(z−z′) where the delta function has to be understood to have a width on the order of c/b.Since we assumed thatΩ≪b, the time it takes for such a fraction of the pulse to cross the ensemble is much smaller than the Larmor period 1/Ω.During the interaction with one of these spatiallylocalized modes the atomic state does not change appre-ciable and we can simplify the interaction operator to V= κ(JT)−1/2J z p(0)where J z= i F(i)z and we as-sumed that the ensemble is located at z=0and changedto a frame rotating at the carrier frequencyωc.A last approximation concerns the description of theatomic spin state.Initially the sample is prepared in acoherent spin state with maximal polarization along x, i.e.in the eigenstate of J x with maximal eigenvalue J. We can thus make use of the Holstein-Primakoffapprox-imation[23]which allows to describe the spin state as a Gaussian state of a single harmonic oscillator.The first step is to express collective step up/down operators (along x),J±=J y±iJ z,in terms of bosonic creation and annihilation operators,[b,b†]=11,asJ+=√11−b†b/2J b,J−=√11−b†b/2J. It is easily checked that these operators satisfy the correct commutation relations[J+,J−]=2J x if one identifies J x=J−b†b.The fully polarized initial state thus corresponds to the ground state of the harmonic oscillator.Note that this map-ping is exact.Under the condition that b†b ≪J one can approximate J+≃√2Jb†and therefore J z≃−i 2and P=−i(b−b†)/√。
Bosch 蒸汽干衣机商品说明书
1234Display panel / ButtonsCongratulations – you have chosen to buy a modern, high-quality Bosch domestic appliance.A distinctive feature of the condensation dryer with heat pump and automatic cleaning of the heat exchanger is its low energy consumption.Every dryer which leaves our factory is carefully checked to ensure that it functions correctly and is in perfect condition.Should you have any questions, our after-sales service will be pleased to help.Environmentally-responsible disposalDispose of packaging in an environmentally-responsible manner. This appliance is designated according to European directive 2002/96/EC which governs waste electrical and electronic equipment - WEEE).The directive provides an EU-wide framework for the return and recycling of used appliances.For further information about our products, accessories, spare parts and services, please visit: DryingÏ÷Spin speed Ï÷Spin speed exclusively to display the anticipated drying time (Ð÷Adjust drying level ContentsPageʋPreparation . . . . . . . . . . . . . . . . . . . . . . . . . . .2ʋControl panel . . . . . . . . . . . . . . . . . . . . . . . . . .2ʋDrying . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3/4ʋDisplay panel and buttons . . . . . . . . . . . . . . . .5ʋNotes on laundry . . . . . . . . . . . . . . . . . . . . . . .6ʋOverview of programmes . . . . . . . . . . . . . . . .7ʋImportant Information . . . . . . . . . . . . . . . . . . .8ʋConsumption rates. . . . . . . . . . . . . . . . . . . . . .9ʋNormal noises . . . . . . . . . . . . . . . . . . . . . . . .10ʋWhat to do if ... / After-sales service. . . . .11/12ʋSafety instructions . . . . . . . . . . . . . . . . . . . . .13Sort the laundryThe fluff filter consists of two parts.Clean the inner and outer fluff filter every 1.Open the door. Remove fluff from thedoor/door area.back together and reinsert into the appliance.6Labelling of fabricsFollow the manufacturer's care information.hc Drying tips––Remove the laundry and switch off Start/Pause buttonTwo-part fluff filterAir inletClean fluff filters reduce energy consumption.Depending on the selected programme, it may not be possible to select individual options button in the centre of the programme selector.Select the programme by turning the outer ring on the programme selector (the selector can be turned in both directions).Change the functions in the display using the buttons C/board dry extra,Dryeren Instruction manualDisplay Congratulations – you have chosen to buy a modern, high-qualityBosch domestic appliance.A distinctive feature of the condensation dryer with heat pump andautomatic cleaning of the heat exchanger is its low energyconsumption.Every dryer which leaves our factory is carefully checked to ensurethat it functions correctly and is in perfect condition.Should you have any questions, our after-sales service will bepleased to help.Environmentally-responsible disposalDispose of packaging in an environmentally-responsible manner.This appliance is designated according to European directive2002/96/EC which governs waste electrical and electronicequipment - WEEE).The directive provides an EU-wide framework for the return andrecycling of used appliances.For further information about our products, accessories, spare partsand services, please visit: Contents PageʋPreparation . . . . . . . . . . . . . . . . . . . . . . . . . . .2ʋControl panel . . . . . . . . . . . . . . . . . . . . . . . . . .2ʋDrying . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3/4ʋDisplay panel and buttons . . . . . . . . . . . . . . . .5ʋNotes on laundry . . . . . . . . . . . . . . . . . . . . . . .6ʋOverview of programmes . . . . . . . . . . . . . . . .7ʋImportant Information . . . . . . . . . . . . . . . . . . .8ʋConsumption rates. . . . . . . . . . . . . . . . . . . . . .9ʋNormal noises . . . . . . . . . . . . . . . . . . . . . . . .10ʋWhat to do if ... / After-sales service. . . . .11/12ʋSafety instructions . . . . . . . . . . . . . . . . . . . . .13button in the centre of the programme selector.Select the programme by turning the outer ring on the programmselector (the selector can be turned in both directions).1234DryingSort the laundryThe fluff filter consists of two parts.Clean the inner and outer fluff filter every 1.Open the door. Remove fluff from thedoor/door area.back together and reinsert into the appliance.6Remove the laundry and switch off Start/Pause buttonTwo-part fluff filterAir inletClean fluff filters reduce energy consumption.Change the functions in the display using the buttons C/board dry extra,Ï÷Spin speedÏ÷Spin speedexclusively to display the anticipated drying time (Ð÷Adjust drying level Labelling of fabricsFollow the manufacturer's care information. hc–Drying tips––Depending on the selected programme, it may not be possible to select individual optionsDryeren Instruction manual。
iProx无接近感触感应传感器安装说明说明书
iProx inductive proximity sensors—mounting close togetherDescriptionThe iProxீ is a powerful family of inductive proximity sensors featuring high sensing performance right out of the box. What makes the iProx unique from other inductive sensors is the ability to extensively customize the operating characteristics to suit a particular application.Mounting sensors close togetherWhen mounting iProx sensors close together, it is necessary to take into consideration problems that can be causedby two or more sensors communicating with each other (also known as “cross-talk”). This problem can arise when two or more sensors are mounted side by side (as shown in Figure 1) or facing each other (as shown in Figure 2).Figure 1. Proximity Sensors Mounted Side by SideFigure 2. Proximity Sensors Mounted Facing Each Other Standard inductive proximity sensors have a similar frequency and will interfere with each other when operated close together. Until iProx, your best solution may have been to buy special sensors designed to operate on different frequencies. The disadvantage of this solution is that your choice of sensing range and body style is usually very limited.The DC versions of iProx have three different noise immunity settings, while the AC versions have two settings. See Figure 3, Figure 4, Figure 5, and Figure 6 for optimum noise immunity settings based upon center-to-center spacing between mounted sensors. Note that in some cases, more than one noise immunity setting is available. In this case,the operator can choose the combinations of noise immunity settings most desirable for the application. The iProx breaks many of the traditional rules of inductive proximity sensors. For instance, it is possible to mount the sensors so that the sensing fields overlap, so long as the proper noise immunity settings are chosen.Figure 3. DC iProx Side-by-Side Configuration (S1)T able 1. DC iProx Side-by-Side Configuration (S1)Diameter (D)Sideby SideHigh NoiseImmunityFactoryDefault12 mm Shielded0–19 mmat 50 Hz0 mm–Infinityat 10 Hz30 mm–Infinityat 580 HzUnshielded0–40 mmat 50 Hz0 mm–Infinityat 10 Hz100 mm–Infinityat 300 Hz18 mm Shielded0–35 mmat 50 Hz0 mm–Infinityat 10 Hz80 mm–Infinityat 390 HzUnshielded0–75 mmat 50 Hz0 mm–Infinityat 10 Hz130 mm–Infinityat 150 Hz30 mm Shielded0–65 mmat 50 Hz0 mm–Infinityat 10 Hz110 mm–Infinityat 240 HzUnshielded0–75 mmat 50 Hz0 mm–Infinityat 10 Hz130 mm–Infinityat 90 Hz2Technical Data TD05301002EEffective June 2013iProx inductive proximity sensors—mounting close togetherEATON Figure 4. DC iProx Facing Configuration (S 2)T able 2. DC iProx Facing Configuration (S 2)Diameter (D)Side by SideHigh Noise ImmunityFactory Default12 mm Shielded 0–25 mm at 50 Hz 0 mm–Infinity at 10 Hz50 mm–Infinity at 580 HzUnshielded0–55 mm at 50 Hz 0 mm–Infinity at 10 Hz120 mm–Infinity at 300 Hz18 mm Shielded 0–45 mm at 50 Hz 0 mm–Infinity at 10 Hz100 mm–Infinity at 390 HzUnshielded0–90 mm at 50 Hz 0 mm–Infinity at 10 Hz160 mm–Infinity at 150 Hz30 mm Shielded 0–80 mm at 50 Hz 0 mm–Infinity at 10 Hz130 mm–Infinity at 240 HzUnshielded0–90 mm at 50 Hz 0 mm–Infinity at 10 Hz 160 mm–Infinity at 90 HzFigure 5. AC iProx Side-by-Side Configuration (S 1)T able 3. AC iProx Side-by-Side Configuration (S 1) ቢDiameter (D)Factory DefaultHigh Noise Immunity12 mm Shielded 0–19 mm/30 mm–Infinity at 30 Hz 0 mm–Infinity at 10 HzUnshielded0–40 mm/50 mm–Infinity at 30 Hz 0 mm–Infinity at 10 Hz18 mm Shielded 0–35 mm/80 mm–Infinity at 30 Hz 0 mm–Infinity at 10 HzUnshielded0–75 mm/130 mm–Infinity at 30 Hz 0 mm–Infinity at 10 Hz30 mm Shielded 0–65 mm/110 mm–Infinity at 30 Hz 0 mm–Infinity at 10 HzUnshielded0–75 mm/130 mm–Infinity at 30 Hz0 mm–Infinity at 10 HzቢThese specifications may not meet final product specifications.Figure 6. AC iProx Facing Configuration (S 2)T able 4. AC iProx Facing Configuration (S 2) ቢDiameter (D)Factory DefaultHigh Noise Immunity12 mm Shielded 0–25 mm/50 mm–Infinity at 30 Hz 0 mm–Infinity at 10 HzUnshielded0–55 mm/120 mm–Infinity at 30 Hz 0 mm–Infinity at 10 Hz18 mm Shielded 0–45 mm/100 mm–Infinity at 30 Hz 0 mm–Infinity at 10 HzUnshielded0–90 mm/160 mm–Infinity at 30 Hz 0 mm–Infinity at 10 Hz30 mm Shielded 0–80 mm/130 mm–Infinity at 30 Hz 0 mm–Infinity at 10 HzUnshielded0–90 mm/160 mm–Infinity at 30 Hz0 mm–Infinity at 10 Hzቢ These specifications may not meet final product specifications.ote:N There is a correlation between noise immunity and operatingfrequency. When setting the sensor for side-by-side sensing or high noise immunity, the operating frequency of the sensor will be reduced. Refer to the iProx Programming Software User Guide (P50228) for details.3Technical Data TD05301002EEffective June 2013iProx inductive proximity sensors—mounting close togetherEATON Setting iProx for side-by-side or facing sensingTo set up iProx sensors for side-by-side or face-to-face operation, just follow the simple procedure below. This process requires that the iProx programming software be installed on your computer. Consult the iProx Programming Software User Guide (P50228) for detailed installation and operating instructions.1. Connect the programming device to your computer. Connectthe remote programmer (E59RP1) to your computer’s serial or USB port. 2. Launch that the iProx programming software from the Start menu.3. Affix the magnetic puck to the face of the iProx sensor. To ensureproper mounting, it may be necessary to remove the mounting nuts from the sensor.4. Ensure that the sensor(s) you intend to program are powered on.5. The iProx programming software should automatically detect thesensor(s). See the “Connection Status” information displayed at the bottom of the “Getting Started” window to confirm that the software has detected the sensor(s). If the software does not detect the connected sensor, you may need to select a wider COM port range in the software settings. For more troubleshooting information, consult the iProx Programming Software User Guide (P50228) by clicking on the “Help” window and selecting “Contents/FAQs.”6. From the “Getting Started” window, click “Configure iProx” to modify the parameters of your iProx sensor.7. Under “Step 1: Select Sensor to Modify,” click the drop-downmenu and select the sensor you want to modify. It may take a few seconds to communicate with the sensor. Once this is complete, the rest of the screen should become enabled, allowing you to modify the parameters of the selected sensor.8. Under the “Response Time/Noise Immunity” section at the topright of the screen, drag the slider until you see the side-by-side value enabled.9. Click the “Program” button at the bottom of the screen. Repeatthis process for each sensor.Side-by-side vs. array sensingAlthough this procedure allows sensors to be operated close to other sensors in a side-by-side configuration (see Figure 7), Eaton does not recommend that iProx sensors be used in an array (Figure 8). It is possible that cross-ommunication can occur between sensors in an array. If the requirements of your application call for this arrangement, please contact Eaton’s Sensor ApplicationEngineering Department at 1-800-426-9184. They will work with youto find a solution.Figure 7. Sensors Oriented Side by SideFigure 8. Sensors Oriented in an ArrayTechnical Data TD05301002E Effective June 2013iProx inductive proximity sensors—mounting close togetherEaton is a registered trademark.All other trademarks are propertyof their respective owners.Eaton1000 Eaton BoulevardCleveland, OH 44122United States© 2013 EatonAll Rights ReservedPrinted in USAPublication No. TD05301002E / Z13430 June 2013。
振动噪声专业词汇
1 振动信号的时域、频域描述振动过程(Vibration Process)简谐振动(Harmonic Vibration)周期振动(Periodic Vibration)准周期振动(Ouasi-periodic Vibration)瞬态过程(Transient Process)随机振动过程(Random Vibration Process)各态历经过程(Ergodic Process)确定*过程(Deterministic Process)振幅(Amplitude)相位(Phase)初相位(Initial Phase)频率(Frequency)角频率(Angular Frequency)周期(Period)复数振动(Complex Vibration)复数振幅(Complex Amplitude)峰值(Peak-value)平均绝对值(Average Absolute Value)有效值(Effective Value,RMS Value)均值(Mean Value,Average Value)傅里叶级数(FS,Fourier Series)傅里叶变换(FT,Fourier Transform)傅里叶逆变换(IFT,Inverse Fourier Transform) 离散谱(Discrete Spectrum)连续谱(Continuous Spectrum)傅里叶谱(Fourier Spectrum)线*谱(Linear Spectrum)幅值谱(Amplitude Spectrum)相位谱(Phase Spectrum)均方值(Mean Square Value)方差(Variance)协方差(Covariance)自协方差函数(Auto-covariance Function)互协方差函数(Cross-covariance Function)自相关函数(Auto-correlation Function)互相关函数(Cross-correlation Function)标准偏差(Standard Deviation)相对标准偏差(Relative Standard Deviation)概率(Probability)概率分布(Probability Distribution)高斯概率分布(Gaussian Probability Distribution) 概率密度(Probability Density)集合平均(Ensemble Average)时间平均(Time Average)功率谱密度(PSD,Power Spectrum Density)自功率谱密度(Auto-spectral Density)互功率谱密度(Cross-spectral Density)均方根谱密度(RMS Spectral Density)能量谱密度(ESD,Energy Spectrum Density)相干函数(Coherence Function)帕斯瓦尔定理(Parseval''s Theorem)维纳,辛钦公式(Wiener-Khinchin Formula)2 振动系统的固有特*、激励与响应振动系统(Vibration System)激励(Excitation)响应(Response)单自由度系统(Single Degree-Of-Freedom System) 多自由度系统(Multi-Degree-Of- Freedom System) 离散化系统(Discrete System)连续体系统(Continuous System)刚度系数(Stiffness Coefficient)自由振动(Free Vibration)自由响应(Free Response)强迫振动(Forced Vibration)强迫响应(Forced Response)初始条件(Initial Condition)固有频率(Natural Frequency)阻尼比(Damping Ratio)衰减指数(Damping Exponent)阻尼固有频率(Damped Natural Frequency)对数减幅系数(Logarithmic Decrement)主频率(Principal Frequency)无阻尼模态频率(Undamped Modal Frequency)模态(Mode)主振动(Principal Vibration)振型(Mode Shape)振型矢量(Vector Of Mode Shape)模态矢量(Modal Vector)正交* (Orthogonality)展开定理(Expansion Theorem)主质量(Principal Mass)模态质量(Modal Mass)主刚度(Principal Stiffness)模态刚度(Modal Stiffness)正则化(Normalization)振型矩阵(Matrix Of Modal Shape)模态矩阵(Modal Matrix)主坐标(Principal Coordinates)模态坐标(Modal Coordinates)模态分析(Modal Analysis)模态阻尼比(Modal Damping Ratio)频响函数(Frequency Response Function)幅频特* (Amplitude-frequency Characteristics)相频特* (Phase frequency Characteristics)共振(Resonance)半功率点(Half power Points)波德图(BodéPlot)动力放大系数(Dynamical Magnification Factor)单位脉冲(Unit Impulse)冲激响应函数(Impulse Response Function)杜哈美积分(Duhamel’s Integral)卷积积分(Convolution Integral)卷积定理(Convolution Theorem)特征矩阵(Characteristic Matrix)阻抗矩阵(Impedance Matrix)频响函数矩阵(Matrix Of Frequency Response Function) 导纳矩阵(Mobility Matrix)冲击响应谱(Shock Response Spectrum)冲击激励(Shock Excitation)冲击响应(Shock Response)冲击初始响应谱(Initial Shock Response Spectrum)冲击剩余响应谱(Residual Shock Response Spectrum) 冲击最大响应谱(Maximum Shock Response Spectrum) 冲击响应谱分析(Shock Response Spectrum Analysis)3 模态试验分析模态试验(Modal Testing)机械阻抗(Mechanical Impedance)位移阻抗(Displacement Impedance)速度阻抗(Velocity Impedance)加速度阻抗(Acceleration Impedance)机械导纳(Mechanical Mobility)位移导纳(Displacement Mobility)速度导纳(Velocity Mobility)加速度导纳(Acceleration Mobility)驱动点导纳(Driving Point Mobility)跨点导纳(Cross Mobility)传递函数(Transfer Function)拉普拉斯变换(Laplace Transform)传递函数矩阵(Matrix Of Transfer Function)频响函数(FRF,Frequency Response Function)频响函数矩阵(Matrix Of FRF)实模态(Normal Mode)复模态(Complex Mode)模态参数(Modal Parameter)模态频率(Modal Frequency)模态阻尼比(Modal Damping Ratio)模态振型(Modal Shape)模态质量(Modal Mass)模态刚度(Modal Stiffness)模态阻力系数(Modal Damping Coefficient)模态阻抗(Modal Impedance)模态导纳(Modal Mobility)模态损耗因子(Modal Loss Factor)比例粘*阻尼(Proportional Viscous Damping)非比例粘*阻尼(Non-proportional Viscous Damping)结构阻尼(Structural Damping,Hysteretic Damping) 复频率(Complex Frequency)复振型(Complex Modal Shape)留数(Residue)极点(Pole)零点(Zero)复留数(Complex Residue)随机激励(Random Excitation)伪随机激励(Pseudo Random Excitation)猝发随机激励(Burst Random Excitation)稳态正弦激励(Steady State Sine Excitation)正弦扫描激励(Sweeping Sine Excitation)锤击激励(Impact Excitation)频响函数的H1 估计(FRF Estimate by H1)频响函数的H2 估计(FRF Estimate by H2)频响函数的H3 估计(FRF Estimate by H3)单模态曲线拟合法(Single-mode Curve Fitting Method)多模态曲线拟合法(Multi-mode Curve Fitting Method)模态圆(Mode Circle)剩余模态(Residual Mode)幅频峰值法(Peak Value Method)实频-虚频峰值法(Peak Real/Imaginary Method)圆拟合法(Circle Fitting Method)加权最小二乘拟合法(Weighting Least Squares Fitting method) 复指数拟合法(Complex Exponential Fitting method)1.2 振动测试的名词术语1 传感器测量系统传感器测量系统(Transducer Measuring System)传感器(Transducer)振动传感器(Vibration Transducer)机械接收(Mechanical Reception)机电变换(Electro-mechanical Conversion)测量电路(Measuring Circuit)惯*式传感器(Inertial Transducer,Seismic Transducer)相对式传感器(Relative Transducer)电感式传感器(Inductive Transducer)应变式传感器(Strain Gauge Transducer)电动力传感器(Electro-dynamic Transducer)压电式传感器(Piezoelectric Transducer)压阻式传感器(Piezoresistive Transducer)电涡流式传感器(Eddy Current Transducer)伺服式传感器(Servo Transducer)灵敏度(Sensitivity)复数灵敏度(Complex Sensitivity)分辨率(Resolution)频率范围(Frequency Range)线*范围(Linear Range)频率上限(Upper Limit Frequency)频率下限(Lower Limit Frequency)静态响应(Static Response)零频率响应(Zero Frequency Response)动态范围(Dynamic Range)幅值上限Upper Limit Amplitude)幅值下限(Lower Limit Amplitude)最大可测振级(Max.Detectable Vibration Level)最小可测振级(Min.Detectable Vibration Level)信噪比(S/N Ratio)振动诺模图(Vibration Nomogram)相移(Phase Shift)波形畸变(Wave-shape Distortion)比例相移(Proportional Phase Shift)惯*传感器的稳态响应(Steady Response Of Inertial Transducer)惯*传感器的稳击响应(Shock Response Of Inertial Transducer)位移计型的频响特* (Frequency Response Characteristics Vibrometer)加速度计型的频响特* (Frequency Response Characteristics Accelerometer) 幅频特*曲线(Amplitude-frequency Curve)相频特*曲线(Phase-frequency Curve)固定安装共振频率(Mounted Resonance Frequency)安装刚度(Mounted Stiffness)有限高频效应(Effect Of Limited High Frequency)有限低频效应(Effect Of Limited Low Frequency)电动式变换(Electro-dynamic Conversion)磁感应强度(Magnetic Induction,Magnetic Flux Density)磁通(Magnetic Flux)磁隙(Magnetic Gap)电磁力(Electro-magnetic Force)相对式速度传(Relative Velocity Transducer)惯*式速度传感器(Inertial Velocity Transducer)速度灵敏度(Velocity Sensitivity)电涡流阻尼(Eddy-current Damping)无源微(积)分电路(Passive Differential (Integrate) Circuit)有源微(积)分电路(Active Differential (Integrate) Circuit)运算放大器(Operational Amplifier)时间常数(Time Constant)比例运算(Scaling)积分运算(Integration)微分运算(Differentiation)高通滤波电路(High-pass Filter Circuit)低通滤波电路(Low-pass Filter Circuit)截止频率(Cut-off Frequency)压电效应(Piezoelectric Effect)压电陶瓷(Piezoelectric Ceramic)压电常数(Piezoelectric Constant)极化(Polarization)压电式加速度传感器(Piezoelectric Acceleration Transducer)中心压缩式(Center Compression Accelerometer)三角剪切式(Delta Shear Accelerometer)压电方程(Piezoelectric Equation)压电石英(Piezoelectric Quartz)电荷等效电路(Charge Equivalent Circuit)电压等效电路(Voltage Equivalent Circuit)电荷灵敏度(Charge Sensitivity)电压灵敏度(Voltage Sensitivity)电荷放大器(Charge Amplifier)适调放大环节(Conditional Amplifier Section)归一化(Uniformization)电荷放大器增益(Gain Of Charge Amplifier)测量系统灵敏度(Sensitivity Of Measuring System)底部应变灵敏度(Base Strain Sensitivity)横向灵敏度(Transverse Sensitivity)地回路(Ground Loop)力传感器(Force Transducer)力传感器灵敏度(Sensitivity Of Force Transducer)电涡流(Eddy Current)前置器(Proximitor)间隙-电压曲线(Voltage vs Gap Curve)间隙-电压灵敏度(V oltage vs Gap Sensitivity)压阻效应(Piezoresistive Effect)轴向压阻系数(Axial Piezoresistive Coefficient)横向压阻系数(Transverse Piezoresistive Coefficient)压阻常数(Piezoresistive Constant)单晶硅(Monocrystalline Silicon)应变灵敏度(Strain Sensitivity)固态压阻式加速度传感器(Solid State Piezoresistive Accelerometer) 体型压阻式加速度传感器(Bulk Type Piezoresistive Accelerometer) 力平衡式传感器(Force Balance Transducer)电动力常数(Electro-dynamic Constant)机电耦合系统(Electro-mechanical Coupling System)2 检测仪表、激励设备及校准装置时间基准信号(Time Base Signal)李萨茹图(Lissojous Curve)数字频率计(Digital Frequency Meter)便携式测振表(Portable Vibrometer)有效值电压表(RMS V alue V oltmeter)峰值电压表(Peak-value V oltmeter)平均绝对值检波电路(Average Absolute Value Detector)峰值检波电路(Peak-value Detector)准有效值检波电路(Quasi RMS V alue Detector)真有效值检波电路(True RMS Value Detector)直流数字电压表(DVM,DC Digital V oltmeter)数字式测振表(Digital Vibrometer)A/D 转换器(A/D Converter)D/A 转换器(D/A Converter)相位计(Phase Meter)电子记录仪(Lever Recorder)光线示波器(Oscillograph)振子(Galvonometer)磁带记录仪(Magnetic Tape Recorder)DR 方式(直接记录式) (Direct Recorder)FM 方式(频率调制式) (Frequency Modulation)失真度(Distortion)机械式激振器(Mechanical Exciter)机械式振动台(Mechanical Shaker)离心式激振器(Centrifugal Exciter)电动力式振动台(Electro-dynamic Shaker)电动力式激振器(Electro-dynamic Exciter)液压式振动台(Hydraulic Shaker)液压式激振器(Hydraulic Exciter)电液放大器(Electro-hydraulic Amplifier)磁吸式激振器(Magnetic Pulling Exciter)涡流式激振器(Eddy Current Exciter)压电激振片(Piezoelectric Exciting Elements)冲击力锤(Impact Hammer)冲击试验台(Shock Testing Machine)激振控制技术(Excitation Control Technique)波形再现(Wave Reproduction)压缩技术(Compression Technique)均衡技术(Equalization Technique)交越频率(Crossover Frequency)综合技术(Synthesis Technique)校准(Calibration)分部校准(Calibration for Components in system)系统校准(Calibration for Over-all System)模拟传感器(Simulated Transducer)静态校准(Static Calibration)简谐激励校准(Harmonic Excitation Calibration)绝对校准(Absolute Calibration)相对校准(Relative Calibration)比较校准(Comparison Calibration)标准振动台(Standard Vibration Exciter)读数显微镜法(Microscope-streak Method)光栅板法(Ronchi Ruling Method)光学干涉条纹计数法(Optical Interferometer Fringe Counting Method)光学干涉条纹消失法(Optical Interferometer Fringe Disappearance Method)背靠背安装(Back-to-back Mounting)互易校准法(Reciprocity Calibration)共振梁(Resonant Bar)冲击校准(Impact Exciting Calibration)摆锤冲击校准(Ballistic Pendulum Calibration)落锤冲击校准(Drop Test Calibration)振动和冲击标准(Vibration and Shock Standard)迈克尔逊干涉仪(Michelson Interferometer)摩尔干涉图象(Moire Fringe)参考传感器(Reference Transducer)3 频率分析及数字信号处理带通滤波器(Band-pass Filter)半功率带宽(Half-power Bandwidth)3 dB 带宽(3 dB Bandwidth)等效噪声带宽(Effective Noise Bandwidth)恒带宽(Constant Bandwidth)恒百分比带宽(Constant Percentage Bandwidth)1/N 倍频程滤波器(1/N Octave Filter)形状因子(Shape Factor)截止频率(Cut-off Frequency)中心频率(Centre Frequency)模拟滤波器(Analog Filter)数字滤波器(Digital Filter)跟踪滤波器(Tracking Filter)外差式频率分析仪(Heterodyne Frequency Analyzer) 逐级式频率分析仪(Stepped Frequency Analyzer)扫描式频率分析仪(Sweeping Filter Analyzer)混频器(Mixer)RC 平均(RC Averaging)平均时间(Averaging Time)扫描速度(Sweeping Speed)滤波器响应时间(Filter Response Time)离散傅里叶变换(DFT,Discrete Fourier Transform) 快速傅里叶变换(FFT,Fast Fourier Transform)抽样频率(Sampling Frequency)抽样间隔(Sampling Interval)抽样定理(Sampling Theorem)抗混滤波(Anti-aliasing Filter)泄漏(Leakage)加窗(Windowing)窗函数(Window Function)截断(Truncation)频率混淆(Frequency Aliasing)乃奎斯特频率(Nyquist Frequency)矩形窗(Rectangular Window)汉宁窗(Hanning Window)凯塞-贝塞尔窗(Kaiser-Bessel Window)平顶窗(Flat-top Window)平均(Averaging)线*平均(Linear Averaging)指数平均(Exponential Averaging)峰值保持平均(Peak-hold Averaging)时域平均(Time-domain Averaging)谱平均(Spectrum Averaging)重叠平均(Overlap Averaging)栅栏效应(Picket Fence Effect)吉卜斯效应(Gibbs Effect)基带频谱分析(Base-band Spectral Analysis)选带频谱分析(Band Selectable Sp4ctralAnalysis)细化(Zoom)数字移频(Digital Frequency Shift)抽样率缩减(Sampling Rate Reduction)功率谱估计(Power Spectrum Estimate)相关函数估计(Correlation Estimate)频响函数估计(Frequency Response Function Estimate) 相干函数估计(Coherence Function Estimate)冲激响应函数估计(Impulse Response Function Estimate) 倒频谱(Cepstrum)功率倒频谱(Power Cepstrum)幅值倒频谱(Amplitude Cepstrum)倒频率(Quefrency)4 旋转机械的振动测试及状态监测状态监测(Condition Monitoring)故障诊断(Fault Diagnosis)转子(Rotor)转手支承系统(Rotor-Support System)振动故障(Vibration Fault)轴振动(Shaft Vibration)径向振动(Radial Vibration)基频振动(Fundamental Frequency Vibration)基频检测(Fundamental Frequency Component Detecting) 键相信号(Key-phase Signal)正峰相位(+Peak Phase)高点(High Spot)光电传感器(Optical Transducer)同相分量(In-phase Component)正交分量(Quadrature Component)跟踪滤波(Tracking Filter)波德图(Bode Plot)极坐标图(Polar Plot)临界转速(Critical Speed)不平衡响应(Unbalance Response)残余振幅(Residual Amplitude)方位角(Attitude Angle)轴心轨迹(Shaft Centerline Orbit)正进动(Forward Precession)同步正进动(Synchronous Forward Precession)反进动(Backward Precession)正向涡动(Forward Whirl)反向涡动(Backward Whirl)油膜涡动(Oil Whirl)油膜振荡(Oil Whip)轴心平均位置(Average Shaft Centerline Position)复合探头(Dual Probe)振摆信号(Runout Signal)电学振摆(Electrical Runout)机械振摆(Mechanical Runout)慢滚动向量(Slow Roll Vector)振摆补偿(Runout Compensation)故障频率特征(Frequency Characteristics Of Fault)重力临界(Gravity Critical)对中(Alignment)双刚度转子(Dual Stiffness Rotor)啮合频率(Gear-mesh Frequency)间入简谐分量(Interharmonic Component)边带振动(Side-band Vibration)三维频谱图(Three Dimensional Spectral Plot)瀑布图(Waterfall Plot)级联图(Cascade Plot)阶次跟踪(Order Tracking)阶次跟踪倍乘器(Order Tracking Multiplier)监测系统(Monitoring System)适调放大器(Conditional Amplifier) 趋势分析(Trend Analysis)倒频谱分析(Cepstrum Analysis)直方图(Histogram)确认矩阵(Confirmation Matrix)通频幅值(Over-all Amplitude)幅值谱(Amplitude Spectrum)相位谱(Phase Spectrum)报警限(Alarm Level)。
001 (ISSCC tutorial)Noise Analysis in Switched-Capacitor Circuits
© 2011 IEEE
IEEE International Solid-State Circuits Conference
© 2011 IEEE
Thermal Noise Power
• Nyquist showed that
PSD ( f ) = 4kT
• The total average noise power of a resistor in a certain frequency band is therefore
– Examples: Audio systems, wireless transceivers, sensor interfaces
• Electronic noise directly trades with power dissipation and speed • Electronic noise is a major concern in modern technologies with reduced VDD
• The noise of a MOSFET operating in the triode region is approximately equal to that of a resistor • In the saturation region, the thermal noise can be modeled using a drain current source with power spectral density
• We can model the noise using an equivalent voltage or current generator
2 vn
= Pn ⋅ R = 4kT ⋅ R ⋅ Δf
Introduction to HF Radio Propagation
Introduction to HF Radio Propagation1. The Ionosphere1.1 The Regions of the IonosphereIn a region extending from a height of about 50 km to over 500 km, some of the molecules of the atmosphere are ionised by radiation from the Sun to produce an ionised gas. This region is called the ionosphere, figure 1.1.Ionisation is the process in which electrons, which are negatively charged, are removed from (or attached to) neutral atoms or molecules to form positively (or negatively) charged ions and free electrons. It is the ions that give their name to the ionosphere, but it is the much lighter and more freely moving electrons which are important in terms of high frequency (HF: 3 to 30 MHz) radio propagation. Generally, the greater the number of electrons, the higher the frequencies that can be used.During the day there may be four regions present called the D, E, F1 and F2 regions. Their approximate height ranges are:• D region 50 to 90 km;• E region 90 to 140 km;•F1 region 140 to 210 km;•F2 region over 210 km.During the daytime, sporadic E (section 1.6) is sometimes observed in the E region, and at certain times during the solar cycle the F1 region may not be distinct from the F2 region but merge to form an F region. At night the D, E and F1 regions become very much depleted of free electrons, leaving only the F2 region available for communications; however it is not uncommon for sporadic E to occur at night.Only the E, F1, sporadic E when present, and F2 regions refract HF waves. The D region is important though, because while it does not refract HF radio waves, it does absorb or attenuate them (section 1.5).The F2 region is the most important region for high frequency radio propagation as:•it is present 24 hours of the day;•its high altitude allows the longest communication paths;•it usually refracts the highest frequencies in the HF range.The lifetime of electrons is greatest in the F2 region which is one reason why it is present at night. Typical lifetimes of electrons in the E, F1 and F2 regions are 20 seconds, 1 minute and 20 minutes, respectively.Because the F1 region is not always present and often merges with the F2 region, it is not normally considered when examining possible modes of propagation. Throughout this report, discussion of the F region refers to the F2 region.Figure 1.1 Day and night structure of the ionosphere.1.2 Production and Loss of ElectronsRadiation from the Sun causes ionisation in the ionosphere. Electrons are produced when this radiation collides with uncharged atoms and molecules, figure 1.2. Since this process requires solar radiation, production of electrons only occurs in the daylight hemisphere of the ionosphere.When a free electron combines with a charged ion a neutral particle is usually formed,figure 1.3. Essentially, loss is the opposite process to production. Loss of electrons occurs continually, both day and night.500400300200100Figure 1.2 Production.Figure 1.3 Loss.1.3 Observing the IonosphereThe most important feature of the ionosphere in terms of radio communications is its ability to refract radio waves. However, only those waves within a certain frequency range will be refracted. The range of frequencies refracted depends on a number of factors (section 1.4). Various methods have been used to investigate the ionosphere,and the most widely used instrument for this purpose is the ionosonde, figure 1.4. Note that many references to ionospheric communications speak of reflection of the wave. It is, however, a refraction process.An ionosonde is a high frequency radar which sends very short pulses of radio energy vertically into the ionosphere. If the radio frequency is not too high, the pulses are refracted back towards the ground. The ionosonde records the time delay between transmission and reception of the pulses. By varying the oscillation frequency of the pulses, a record is obtained of the time delay at different frequencies.Frequencies less than about 1.6 MHz are interfered with by AM broadcast stations. As the frequency is increased, echoes appear first from the lower E region and subsequently, with greater time delay, from the F1 and F2 regions. Of course, at night echoes are returned only from the F2 region and possibly sporadic E since the other regions have lost most of their free electrons.Positively charged ionFree electronPositively charged ionFigure 1.4 Ionosonde operation.Today, the ionosphere is "sounded" not only by signals sent up at vertical incidence. Oblique sounders send pulses of radio energy obliquely into the ionosphere (the transmitter and receiver are separated by some distance). This type of sounder can monitor propagation on a particular circuit and observations of the various modes being supported by the ionosphere can be made. Backscatter ionosondes rely on echoes reflected from the ground and returned to the receiver, which may or may not be at the same site as the transmitter. This type of sounder is used for over-the-horizon radar. 1.4 Ionospheric VariationsThe ionosphere is not a stable medium that allows the use of one frequency over the year, or even over 24 hours. The ionosphere varies with the solar cycle, the seasons, the circuit and during any given day. So, a frequency which may provide successful propagation now, may not do so an hour later.1.4.1 Variations due to the Solar CycleThe Sun goes through a periodic rise and fall in activity which affects HF communications; solar cycles vary in length from 9 to 14 years. At solar minimum, only the lower frequencies of the HF band will be supported by the ionosphere, while at solar maximum the higher frequencies will successfully propagate, figure 1.5. This is because there is more radiation being emitted from the Sun at solar maximum, producing more electrons in the ionosphere which allows the use of higher frequencies.Figure 1.5 The relationship between solar cycles and E and F region frequencies at Townsville. Dotted vertical lines indicate start of each year. Note also theseasonal variationsThere are other consequences of the solar cycle. Around solar maximum there is a greater likelihood of large solar flares occurring. Flares are huge explosions on the Sun which emit radiation that ionises the D region causing increased absorption of HF waves. Since the D region is present only during the day, only those communication paths which pass through daylight will be affected. The absorption of HF waves travelling via the ionosphere after a flare has occurred is called a short wave fade-out (section 3.1). Fade-outs occur instantaneously and affect lower frequencies the most. Lower frequencies are also the last to recover. If it is suspected or confirmed that a fade-out has occurred, it is advisable to try using a higher frequency, if possible. The duration of fade-outs can vary between about 10 minutes to over an hour depending on the duration and intensity of the flare.1.4.2 Seasonal VariationsE region frequencies are greater in summer than winter, figure 1.5. However, the variation inF region frequencies is more complicated. In both hemispheres, F region noon frequencies generally peak around the equinoxes (March and September). Around solar minimum the summer noon frequencies are, as expected, generally greater than those in winter, but around solar maximum winter frequencies tend to be higher than those in summer. In addition, frequencies around the equinoxes (March and September) are higher than those in summer or winter for both solar maximum and1.4.3 ° latitude at H i g h e s t u s a b l e f r e q u e n c y (M H z )1.4.4 Daily VariationsFrequencies are normally higher during the day and lower at night, Figure 1.7. With dawn, solar radiation causes electrons to be produced in the ionosphere and frequencies increase reaching their maximum around noon. During the afternoon,frequencies begin falling due to electron loss and with darkness the D, E and F1regions disappear. So, communication during the night is by the F2 region and absorption of radio waves is lower. Through the night, frequencies gradually decrease,reaching their minimum just before dawn.Figure 1.7 E and F layer frequencies for a Singapore to Ho Chi Minh circuit sometime in a solar cycle.1.5 Variations in AbsorptionAbsorption was discussed in section 1.4.1 when describing how solar flares can cause disruptions or degradations to communication paths which pass through daylight.Absorption in the D region also varies with the solar cycle, being greatest around solar maximum. Signal absorption is also greater in summer and during the middle of the day, figure 1.8. There is a variation in absorption with latitude, with more absorption occurring near the equator and decreasing towards the poles. Lower frequencies are absorbed to a larger extent, so it is advisable to use as high a frequency as possible,particularly during the day when absorption is greatest.1012024********1618202224Time (UT)Frequency(MHz)Local time (hours)V e r t i c a l a b s o r p t i o n (d B )Figure 1.8 Daily and seasonal variations in absorption at Sydney, 2.2 MHz.Around the polar regions absorption can affect communications quite dramatically at times. Sometimes high energy protons ejected from the Sun during large solar flares will move down the Earth's magnetic field lines and into the polar regions. These protons can cause increased absorption of HF radio waves as they pass through the D region. This increased absorption may last for as long as 10 days and is called a Polar Cap Absorption event (PCA), section 3.2.1.6 Sporadic ESporadic E may form at any time during the day or night. It occurs at altitudes between 90 to 140 km (the E region), and may be spread over a large area or be confined to a small region. It is difficult to know where and when it will occur and how long it will persist. Sporadic E can have a comparable electron density to the F region. This implies that it can refract comparable frequencies to the F region. Sometimes a sporadic E layer is transparent and allows most of the radio wave to pass through it to the F region, however, at other times the sporadic E layer obscures the F region totally and the signal does not reach the receiver (sporadic E blanketing). If the sporadic E layer is partially transparent, the radio wave is likely to be refracted at times from the F region and at other times from the sporadic E. This may lead to partial transmission of the signal or fading, figure 1.9.Sporadic E in the low and mid-latitudes occurs mostly during the daytime and early evening, and is more prevalent during the summer months. At high latitudes, sporadic E tends to form at night.Signal partially transmitted by sporadic E and partially by the Flayer. Any received signal may be weak or fade.Wave passes through sporadic E and is refracted by the F layer. Figure 1.9 Some possible paths when sporadic E is present. The ground reflection will depend on the strength of the signal and other factors such as groundtype.1.7 Spread FSpread F occurs when the F region becomes diffuse due to irregularities in that region, which scatter the radio wave. The received signal is the superposition of a number of waves refracted from different heights and locations in the ionosphere at slightly different times. At low latitudes, spread F occurs mostly during the night hours and around the equinoxes. At mid-latitudes, spread F is less likely to occur than at low and high latitudes. Here it is more likely to occur at night and in winter. At latitudes greater than about 40°, spread F tends to be a night time phenomenon, appearing mostly around the equinoxes, while around the magnetic poles, spread F is often observed both day and night. At all latitudes there is a tendency for spread F to occur when there is a decrease in F region frequencies. That is, spread F is often associated with ionospheric storms (section 3.3).2. HF Communications2.1 Types of HF PropagationHigh Frequency (3 to 30 MHz) radio signals can propagate to a distant receiver, Figure 2.1, via the:•ground wave: near the ground for short distances, up to 100 km over land and 300 km over sea. Attenuation of the wave depends on antenna height, polarisation, frequency, ground types, terrain and/or sea state;•direct or line-of-sight wave: this wave may interact with the earth-reflected wave depending on terminal separation, frequency and polarisation;•sky wave: refracted by the ionosphere, all distances.2.2 Frequency Limits of Sky WavesNot all HF waves are refracted by the ionosphere, there are upper and lower frequency bounds for communications between two terminals. If the frequency is too high, the wave will penetrate the ionosphere, if it is too low, the strength of the signal will be lowered due to absorption in the D region. The range of usable frequencies will vary:•throughout the day;•with the seasons;•with the solar cycle;•from place to place;depending on the ionospheric region used for communications. While the upper limit of frequencies varies mostly with these factors, the lower limit is also dependent on receiver site noise, antenna efficiency, transmitter power, E layer screening (section 2.6) and absorption by the ionosphere.Figure 2.1 Types of HF propagation.2.3 The Usable Frequency RangeFor any circuit there is a Maximum Usable Frequency (MUF) which is determined by the state of the ionosphere in the vicinity of the refraction area(s) and the length of the circuit. The MUF is refracted from the area of maximum electron density of a region. Therefore, frequencies higher than the MUF for a particular region will penetrate thatregion. During the day it is possible to communicate via both the E and F layers usingdifferent frequencies. The highest frequency supported by the E layer is the EMUF, while that supported by the F layer is the FMUF.The F region MUF in particular varies during the day, seasonally and with the solar cycle. The data collected over the years displays a range of frequencies observed and the IPS predictions mirror this. A range of F region MUFs is provided in the predictions and this range extends from the lower decile MUF (called the Optimum Working Frequency, OWF), through the median MUF to the upper decile MUF. These MUFs have a 90%, 50% and 10% chance of being supported by the ionosphere, respectively. IPS predictions usually cover a period of one month, so the OWF should provide successful propagation 90% of the time or 27 days of the month. The median MUF should provide communications 50% or 15 days of the month and the upper decile MUF 10% or 3 days of the month. The upper decile MUF is the highest frequency of the range of MUFs and is most likely to penetrate the ionosphere, figure 2.2.Figure 2.2 Range of usable frequencies. If the frequency, f, is close to the ALF then the wave may suffer absorption in the D region. If the frequency is abovethe EMUF then propagation is via the F region. Above the FMUF the wave is likely to penetrate the ionosphere.The chances of successful propagation discussed above rely on the monthly prediction of solar activity being correct. Sometimes unforeseen events occur on the Sun resulting in the monthly predictions being inaccurate. The role of the Australian Space Forecast Centre (ASFC) at IPS is to provide corrections to the monthly predictions, warning customers of changes in communication conditions.The D region does not allow all frequencies to be used since the lower the frequency the more likely it is to be absorbed. The Absorption Limiting Frequency (ALF) is provided as a guide to the lower limit of the usable frequency band. The ALF is significant only for circuits with refraction points in the sunlit hemisphere. At night, the ALF falls to zero, allowing frequencies which are not usable during the day to successfully propagate.2.4 Hop LengthsThe hop length is the ground distance covered by a radio signal after it has been refracted once from the ionosphere and returned to Earth, figure 2.3. The upper limit of the hop length is set by the height of the ionosphere and the curvature of the Earth. For E and F region heights of 100 km and 300 km, the maximum hop lengths with an elevation angle of 4°, are 1800 km and 3200 km, respectively. Distances greater than these will require more than one hop. For example, a distance of 6100 km will require a minimum of 4 hops by the E region and 2 hops via the F region. More hops would be required with larger antenna elevation angles.Figure 2.3 Hop lengths based upon an antenna elevation angle of 4° and heights forthe E and F layers of 100 km and 300 km, respectively.If the elevation angle of the antenna is able to be altered easily or there is a choice of antennas, it may be possible to use a certain number of hops such that an opponent's position does not coincide with a reflection point on the Earth. This will however, be affected also by the time of day and what frequencies are available. However, in most cases the directional properties of the antenna are such that this is not possible.2.5 Propagation ModesThere are many paths by which a sky wave may travel from a transmitter to a receiver.The mode by a particular layer which requires the least number of hops between the transmitter and receiver is called the first order mode. The mode that requires one extra hop is called the second order mode. For a circuit with a path length of 5000 km, the first order F mode has two hops (2F), while the second order F mode has three hops (3F). The first order E mode has the same number of hops as the first order F mode. If this results in a hop length of greater than 2050 km, which corresponds to an elevation angle of 0°, then the E mode is not possible. This also applies to the second order E mode. Of course, the E region modes will only be available on daylight circuits.Simple modes are those propagated by one region, say the F region. IPS predictions are made only for these simple modes, Figure 2.4. More complicated modes consisting of combinations of refractions from the E and F regions, ducting and chordal modes are also possible, figure 2.5.E region(100 km)F region(300 km)Figure 2.4 Examples of simple propagation modes.Chordal modes and ducting involve a number of refractions from the ionosphere without intermediate reflections from the Earth. There is a tendency to think of the regions of the ionosphere as being smooth, however, the ionosphere undulates and moves, with waves passing through it which may affect the refraction of the signal. The ionospheric regions may tilt and when this happens chordal and ducted modes may occur. Ionospheric tilting is more likely near the equatorial anomaly, the mid-latitude trough and in the sunrise and sunset sectors. When these types of modes do occur,signals can be strong since the wave spends less time traversing the D region and being attenuated during ground reflections.Figure 2.5 Other propagation modes.Because of the high electron density of the daytime ionosphere in the vicinity of 15° of the magnetic equator (near the equatorial anomaly), transequatorial paths can use these enhancements to propagate on higher frequencies. Any tilting of the ionosphere may result in chordal modes, producing good signal strength over long distances.Ducting may result if tilting occurs and the wave becomes trapped between refracting regions of the ionosphere. This is most likely to occur in the equatorial ionosphere, near the auroral zone and mid-latitude trough. Disturbances to the ionosphere, such as travelling ionospheric disturbances (section 2.9), may also account for ducting and chordal mode propagation.E regionF region2.6 E Layer ScreeningFor daytime communications via the F region, the lowest usable frequency via the one hop F mode is dependent upon the presence of the E region. If the operating frequency for the 1F mode is below the two hop EMUF, then the signal is unlikely to propagate via the F region due to screening by the E region, figure 2.6. This is also because the antenna elevation angles of the 1F and 2E modes are similar.Figure 2.6 E layer screening occurs if communications are required by the 1F modeand the operating frequency is close to or below the EMUF for the 2Emode. Note the paths through the D region for each wave.A sporadic E layer may also screen a wave from the F region. Sometimes sporadic E can be quite transparent, allowing most of the wave to pass through it. At other times it will partially screen the F region leading to a weak or fading signal, while at other times sporadic E can totally obscure the F region with the possible result that the signal does not arrive at the receiver, figure 1.9 (section 1.6).2.7 Frequency, Range and Elevation AngleFor oblique propagation, there are three dependent variables:• frequency;• range or path length;• antenna elevation angle.The diagrams below illustrate the changes to the ray paths when each of these is fixed in turn.E regionF regionD regionFigure 2.7: Elevation angle fixed:• As the frequency is increased toward the MUF, the wave is refracted higher in the ionosphere and the range increases, paths 1 and 2;• At the MUF for that elevation angle, the maximum range is reached, path 3;• Above the MUF, the wave penetrates the ionosphere, path 4.Figure 2.7 Elevation angle fixed.Figure 2.8: Path length fixed (point-to-point circuit):• As the frequency is increased towards the MUF, the wave is refracted from higher in the ionosphere. To maintain a circuit of fixed length, the elevation angle must therefore be increased, paths 1 and 2;• At the MUF, the critical elevation angle is reached, path 3. The critical elevation angle is the elevation angle for a particular frequency, which if increased, would cause penetration of the ionosphere;• Above the MUF, the ray penetrates the ionosphere, path 4.Figure 2.8 Path length fixed.Figure 2.9: Frequency fixed:•At low elevation angles the path length is greatest, path 1;•As the elevation angle is increased, the path length decreases and the ray is refracted from higher in the ionosphere, paths 2 and 3;•If the frequency will return when sent vertically up into the ionosphere, then the skip distance is zero. However, if this is not the case, then as the elevation angle is increased, the range decreases. If the elevation angle is increased beyond the critical elevation angle for that frequency then the wave penetrates the ionosphere and there is an area around the transmitter within which no sky wave communications can be received, path 4. To communicate within the skip zone, the frequency must be lowered.Figure 2.9 Frequency fixed.2.8 Skip ZonesA propagation path will consist of high and low angle rays corresponding to the wave propagating from the transmit antenna at a range of angles. The high ray is transmitted at a high angle from the antenna. These rays may travel by different paths through the ionosphere. The edge of the skip zone corresponds to the high and low rays taking the same path through the ionosphere, with the resulting signal often being stronger. Within the skip zone the signal fades due to the waves penetrating the ionosphere.In figure 2.9, path 3 corresponds to the high and low rays taking the same path through the ionosphere, this corresponds to the MUF. As the frequency is increased toward the MUF, the height of refraction of the low angle ray increases and the height of refraction of the high angle ray decreases until they are both refracted at the same point in the ionosphere.Skip zones can often be used to advantage if it is desired that communications are not heard by a particular receiver. Selecting a different frequency will alter the size of the skip zone and if the receiver is within the skip zone and out of reach of the ground wave, then it is unlikely that it will receive the communications. However, factors such as sidescatter, where reflection from the Earth outside the skip zone results in the wavetransmitting into the skip zone may affect the reliability of this.Skip zones vary in size during the day, with the seasons, and with solar activity. During the day, solar maximum and around the equinoxes, skip zones generally are smaller in area. The ionosphere during these times has increased electron density and so is able to support higher frequencies.2.9 FadingMultipath fading results from dispersion of the signal by the transmitting antenna. A number of modes propagate which have variations in phase and amplitude. These waves may interfere with each other if they reach the receiver, figure 2.10.Figure 2.10 Multipath fading. The signal may travel by a number of paths which, if they arrive at the receiver and are of similar amplitude, may interfere and cause fading.Disturbances known as Travelling Ionospheric Disturbances, TIDs, may cause a region to be tilted, resulting in the signal being focussed or defocussed, figure 2.11. Fading periods of the order of 10 minutes or more can be associated with these structures. TIDs travel horizontally at 5 to 10 km/minute with a well defined direction of travel affecting higher frequencies first. Some originate in auroral zones following an event on the Sun and these may travel large distances. Others originate in weather disturbances. TIDs may cause variations in phase, amplitude, polarisation and angle of arrival of a wave.Polarisation fading results from changes to the polarisation of the wave along the propagation path. The receiving antenna is unable to receive parts of the signal; this type of fading can last for a fraction of a second to a few seconds.Skip fading can be observed around sunrise and sunset particularly, when the operating frequency is close to the MUF, or when the receiving antenna is positioned close to the boundary of the skip zone. At these times of the day, the ionosphere is unstable and the frequency may oscillate above and below the MUF causing the signal to fade in and out. If the receiver site is close to the skip zone boundary, as the ionosphere fluctuates, the skip zone boundary also fluctuates.Figure 2.11 Focussing and defocussing effects caused by tilting and travelling ionospheric disturbances (TIDs).2.10 NoiseRadio noise arises from internal and external origins. Internal or thermal noise is generated in the receiving system and is usually negligent when compared to external sources of noise. External radio noise originates from natural (atmospheric and galactic) and man-made (environmental) sources.Atmospheric noise, which is caused by thunderstorms, is normally the major contributor to radio noise in the HF band and will especially degrade circuits passing through the day-night terminator. Atmospheric noise is greatest in the equatorial regions of the world and decreases with increasing latitude. Its effect is also greater on lower frequencies, hence it is usually more of a problem around solar minimum and at night when lower frequencies are needed.Galactic noise arises from our galaxy. Receive antennas with higher frequencies more likely to be affected by this type of noise.Man-made noise includes ignition noise, neon signs, electrical cables, power transmission lines and welding machines. This type of noise depends on the technology used by the society and its population. Interference may be intentional, such as jamming, due to propagation conditions or the result of others working on the same frequency.Man-made noise tends to be vertically polarised, so selecting a horizontally polarised antenna may help in reducing noise. Using a narrower bandwidth, or a directional receiving antenna (with a lobe in the direction of the transmitting source and a null in the direction of the unwanted noise source), will also aid in reducing the effects of noise. Selecting a site with a low noise level and determining the major noise sources are important factors in establishing a successful communications system.2.11 VHF and 27 MHz PropagationVHF and 27 MHz are used for line-of-sight or direct wave communication, for example, ship-to-ship or ship-to-shore. The frequency bands are divided into channels and one channel is usually as good as the next. This is in contrast to medium frequency (MF: 300 kHz to 3 MHz) and HF where the choice of a frequency channel may be crucial for good communications.Because VHF and 27 MHz operate mainly by line-of-sight, it is important to mount the antenna as high as possible and free from obstructions. Shore stations are usually on the tops of hills to provide maximum range, but even the highest hills do not provide。
基于CEEMDAN
第 39 卷第 1 期电力科学与技术学报Vol. 39 No. 1 2024 年 1 月JOURNAL OF ELECTRIC POWER SCIENCE AND TECHNOLOGY Jan. 2024引用格式:尚海昆,张冉喆,黄涛,等.基于CEEMDAN-TQWT方法的变压器局部放电信号降噪[J].电力科学与技术学报,2024,39(1):272‑284. Citation:SHANG Haikun,ZHANG Ranzhe,HUANG Tao,et al.Partial discharge signal denoising based on CEEMDAN‑TQWT method for power transformers[J]. Journal of Electric Power Science and Technology,2024,39(1):272‑284.基于CEEMDAN‑TQWT方法的变压器局部放电信号降噪尚海昆,张冉喆,黄涛,林伟,赵子璇(东北电力大学现代电力系统仿真控制与绿色电能新技术教育部重点实验室,吉林吉林 132012)摘要:针对传统方法处理局部放电信号时存在振荡明显、消噪不彻底等问题,采用基于自适应白噪声完备集成经验模态分解(complete ensemble empirical model decomposition with adaptive noise,CEEMDAN)与可调品质因子小波变换(tunable Q⁃factor wavelet transform,TQWT)相结合的方法对局部放电信号进行消噪处理。
采用CEEMDAN将含噪变压器局部放电信号分解成多个固有模态函数(intrinsic mode function,IMF)分量,并利用相关系数判断IMF分量与原始信号的相关度。
将弱相关者视为劣质IMF,对其进行TQWT分解,利用能量占比与峭度指标来筛选小波子带,提取IMF的有效细节信息,进行TQWT逆变换,从而得到新的IMF分量;将强相关者视为优质IMF,与变换后的新IMF分量共同进行信号重构,得到消噪结果。
Acoustic Vector-Sensor Correlations in Ambient Noise
FORMANT FREQUENCY ESTIMATION IN NOISE
FORMANT FREQUENCY ESTIMATION IN NOISEBin Chen and Philipos C.Loizou*University of Texas at Dallas,Dept.of Electrical EngineeringRichardson,TX75083*loizou@ABSTRACTThis paper addresses the problem of formant frequency es-timation of speech signals corrupted by colored noise.The spectrum is sequentially segmented inot K segments so that each segment contains a single formant.A segmentation metric based on Wienerfilter theory is proposed for deter-mining the segment boundaries.A peak-picking algorithm is used for estimating the formant frequencies in each seg-ment.Results obtained using vowels embedded in+5dB S/N speech-shaped noise,indicated that the proposed algo-rithm produced formant frequencies which were compara-ble to those estimated in quiet.1.INTRODUCTIONApart from a variety of formant tracking approaches[1][2], considerable attention has been paid to methods based on linear prediction analysis(LPC)[3][4].However,captur-ing and tracking formants accurately from noisy speech is not easy,largely because the accuracy of root-finding algo-rithms based on LPC is sensitive to the noise level.In[5][6],a set of parallel digital formant resonators has been proposed for speech synthesis or formant frequency estimation.In this paper,we propose the use of a sequen-tial digital resonator model for spectrum segmentation.The spectrum segmentation is implemented sequentially from low to high frequencies.For each spectral segment,a dig-ital resonator isfirst determined to represent the spectral segment.A metric based on Wienerfilter theory is pro-posed to determine the segment boundaries.After identify-ing the spectral segments containing the formants,we apply a peak-picking algorithm on each spectral segment tofind the formant frequency.This approach was taken since the LPC-based digital resonators are sensitive to the noise level.A major advantage of the proposed method is that it deter-mines the segment boundaries sequentially and avoids the need for dynamic programming as done in[5]and[7]. This paper is organized as follows.Section2describes the formant estimation model,Section3presents the pro-posed formant frequency estimation algorithm,Section4presents the experimental results,and Section5gives theconclusions.2.FORMANT ESTIMATION MODELIn this section,a model is described for formant estima-tion that is implemented using a set of digital resonators.Each resonator represents a formant in a segment in the fre-quency domain.The spectrum is divided into segments suchthat only one formant resides in each segment.For the con-venience of representing the digital resonator,the segmentboundaries are assumed to befixed.In the next section,weshow how to determine the segment boundaries sequentiallyusing a Wiener-based metric.Each formant in a spectral segment k is represented by asecond-order predictionfilter.The second-order predictionfilter for the formant in the spectral segment k is given bythe all-pole model1/A k(z)=1/(1+a k z−1+βk z−2).The formants can be considered as being generated by a second-order system driven by white noise.A k(z)is a whiteningfilter that whitens the formant spectrum,i.e.,itflattens thespectrum in segment k.If A k(z)is used as a notchfilter,itwill notch the corresponding formant out of the spectrum.In our application,we adopt the notchfilter definition in[8]H k(z)=γ(1+αz−1+βz−2)(1) whereα=e−2πB,β=−2e−πB cosωandγ= 1/(1+α+β)are specified by the notch frequencyωand the bandwidth B.Note that A k(z)is similar to H k(z)ex-cept for the scalarγ.Thus,we canfind the notchfilter by determining the segmental system transfer function H k(z). According to[5],the optimum prediction coefficients of the notchfilter are given by:αoptk=r k(0)r k(1)−r k(1)r k(2)r k(0)−r k(1)(2a)βoptk=r k(0)r k(2)−r k(1)2k k(2b) where r k(m)are the autocorrelation coefficients obtained for segment kr k(m)=r(wk−1,w k)(m)=1Zωkωk−1|S(e jω)|2cos(mω)dω(3)Substitutingαoptk andβoptkobtained above to Eq.(1)givesus the desired notchfilter H k(z)of the k th band in the spec-trum.The scalarγis independent of minimization of theprediction error and is determined after theαoptk andβoptkare found.As in[5],we use a discrete approximation of the in-tegral in Eq.(3).The frequency range[0π]is divided into I equally spaced intervals∆ω(=π/I)with grid πi/I,i=0,1,...,I.Therefore,the segment boundaries ω0=0,...,ωk,...,ωK=πare replaced by the indices i0=0,...,i k,...,i K=I,and r k(m)is given byr k(m)=1i k X i=i k−1|S(i)|2cosµ2πmi¶(4)with S(i)=S(ω)|ω=e j(2πi/2I).The above autocorrelation sequence is determined for a specific spectral segment[i k−1 i k],and is expected to vary accordingly with the spectral segment.Experiments showed that the autocorrelation se-quence does not change much when a strong formant dom-inates the spectral segment even after the spectral segment is expanded to include a second formant.3.PROPOSED FORMANT FREQUENCYESTIMATION ALGORITHM IN NOISESo far we described a formant frequency model for a sin-gle spectral segment k.That is,we assumed that the seg-ment boundaries were known.In this section,we propose a segmentation metric,motivated by Wienerfilter theory,that identifies the boundaries of the K segments of the spectrum containing the K formants.Suppose that the input to a Wienerfilter is a signal with an additive noise,i.e.,x(n)=s(n)+n(n),and the desired signal is the noise,i.e.,d(n)=n(n).From the orthogonal-ity principle,we know thatE[e(n)x(n−l)]=0(5) =r nn(l)−∞X k=0h w(k)r xx(l−k)where h w(n)is the Wienerfilter,e(n)is the estimation er-ror,and r nn(l)and r xx(l)are the autocorrelation sequences of the noise and noisy speech signal respectively.For a given notchfilter h(n),we can produce the prediction resid-ual w(n)of the clean signal asw(n)=M−1X k=0h(k)s(n−k)(6)where h(0)=1,and M=3.Now,if we replace the Wienerfilter h w(n)in Eq.(5)with the notchfilter h(n),we get:E[e(n)x(n−l)](7a)=r nn(l)−M−1X k=0h(k)r xx(l−k)Since x(n)=s(n)+n(n),we get from Eq.(7a):E[e(n)x(n−l)]=r nn(l)−E[w(n)x(n−l)]−M−1P k=0h(k)r nn(l−k)=0(7b)Note that Eq.(7b)is no longer equal to zero since the notchfilter h(n)in Eq.(7b)is not the optimum Wienerfilter.Sincethe prediction residual w(n)is independent of the noisy sig-nal x(n),the second term E[w(n)x(n−l)]in Eq.(7b)oughtto be zero.In practice,however,w(n)becomes white onlyif h(n)whitens s(n˙).As the upper boundary of a segmentexpands,the notchfilter h(n)will gradually become moreand more matched with the formant in the segment,andE[w(n)x(n−l)]will become smaller and smaller.WhenE[w(n)x(n−l)]reaches its minimum,or E[e(n)x(n−l)]attains its maximum,the whole formant will be matched andcontained in the segment.As mentioned earlier,the notchfilter h(n)will not change much even if the next formantis included.That is,E[e(n)x(n−l)]reaches a maximumand saturates thereafter.The point at which the maximumis reached is indicative of a segment boundary.We there-fore use the energy of E[e(n)x(n−l)]as the segmentationmetric.The third term P h(k)r nn(l−k)in Eq.(7b)may alsobecome small as h(n)changes.In order to offset the effectof this undesired term,we add the term P h(k)r nn(l−k)in Eq.(7a).Thefinal segmentation metric then becomes:E k[e(n)x(n−l)](8)=r k nn(l)−M−1X m=0h k(m)r k xx(l−m)+M−1X m=0h k(m)r k nn(l−m)where h k(m)and r k(m)represent the notchfilter andthe autocorrelation sequence calculated from the k th spec-tral segment[ωk−1ωk]respectively.The energy ofEk[e(n)x(n−l)]is used as the segmentation metric andis denoted byE ex(ωk−1,ωk)=M−1X l=0E k[e(n)x(n−l)](9)The metric saturation point,which is also the segmentboundary point,is defined to be the point at which the fol-lowing condition is satisfied:¯¯¯¯E ex(w k+m)−E ex(w k)E ex(w k)¯¯¯¯<ε(10) where E ex(w k)denotes E ex(ωk−1,ωk)for simplicity.The delay index m is used to ensure that there is a long enough saturation period before a true saturation point is detected. Empirically,m should be selected such that the saturation period is no less than300Hz.The constantεis empirically determined.Figure1shows an example of the segmentation of a noisy vowel spectrum.Once the segmentation of the formant region is deter-mined,we considered peak-picking the spectrum.The basic idea is to segment the noise spectrum to have only one for-mant in each segment,and then for each segment,peak-pick the spectrum to get an estimate for the formant frequency of the noisy speech spectrum.The above segmentation algorithm requires access to the autocorrelation sequence of the clean signal,which we do not have.To estimate the clean autocorrelation sequence, we considered pre-processing the signal by the spectral sub-traction algorithm[9]to get an estimate of the enhanced signal spectrum.The autocorrelation sequence is obtained using Eq.(4)but with S(i)being replaced with the enhanced speech spectrum.3.1.Proposed AlgorithmThe proposed algorithm is outlined below: Initialization:k=1;i k−1=0;i k=1;K=desired number of formantsStep1.Loop(for segment k):(1)Calculate r(i k)xx (l)and r(i k)nn(l)using Eq.(4)(2)Use Equations(2a),(2b)and(4)to calculate the notchfilter h(i k)(n)(3)Use Equations(8)and(9)to estimateE ex(ωk−1,ωk)(4)if E ex(ωk−1,ωk)reaches a saturation point(ac-cording to Eq.10),then:k th boundary=i kPeak-pick spectrum to estimate formant fre-quency.go to Step2end(5)i k=i k+1EndStep2.k=k+1i k−1=i kif k>K,stopelse,go to Step1In our implementation,the autocorrelation sequence of the noise,r nn(l),was estimated using thefirst few speech-absent frames of the noisy speech signal.The speech sig-Table1.Standard deviations(Hz)of formant frequencyerrors for synthetic vowels using the proposed algorithm(SEF)and the LPC algorithm.The formant frequencies ofthe LPC algorithm were obtained in quiet,while the fre-quencies of the SEF algorithm were based on vowels em-bedded in+5dB speech-shaped noise.nal was processed using10-ms duration Hamming windowswith50%overlap between adjacent frames,and the spec-trum in Eq.4was obtained using the FFT.4.EXPERIMENTAL RESULTSThe proposed formant frequency estimation algorithm wasevaluated using real and synthetic vowels.Four naturalvowels,/u/,/a/,/ei/and/i/,corrupted by speech-shapednoise at+5dB S/N were used for evaluation.The vowelswere contained in the words“hood”,“hod”,“hayed”and“heed”and were produced by a male speaker.The esti-mated formant tracks are shown in Figure2.For compar-ative purposes,we also estimated the formant frequenciesof these vowels in quiet using two other methods based onLPC(16th order)and dynamic programming[5].As can beseen,our estimated formant frequencies are comparable tothe estimated formant frequencies in quiet.The same vowels were also synthesized using the Klattsynthesizer[6],and corrupted by a+5dB speech-shapednoise.Each test consisted of200trials in which the F1wasvaried±200Hz and the F2and F3frequencies were varied±150Hz around the center of the corresponding formant frequencies.Standard derivations were measured of the dif-ferences between the true formant frequencies and the es-timated formant frequencies.The results are tabulated inTable1.For comparative purposes,we also list the standarddeviations of the formant frequencies of the same vowelsestimated in quiet using the LPC method.Results indicatedthat the estimation of the F1frequency was more accuratethan the estimation of the F2and F3frequencies.5.SUMMARY AND CONCLUSIONSA new method for estimating formant frequencies in noisewas proposed based on sequential determination of spec-tral segments and formant frequencies.The spectrum wasFig.1.The top panel shows values of the segmentation met-ric as a function of frequency.Saturation point was esti-mated to be 1100Hz.Bottom panel shows the noisy spec-trum of the vowel /ey/.In this example,the F1region was determined to be 0-1100Hz.Fig.2.Formant tracks for four vowels in +5dB S/N es-timated using the proposed formant frequency estimation algorithm (SEF).For comparison,we superimpose the for-mant tracks of the vowels estimated in quiet by the LPC and dynamic programming based algorithms (Dyn)[5].sequentially segmented into K segments using a new seg-mentation metric based on Wiener filter theory.No speci fic assumptions were required for the statistics of the noise.Experimental results showed that the estimated formant fre-quencies of vowels embedded in +5dB speech-shaped noise were comparable to the formant frequencies estimated in quiet.6.REFERENCES[1]A.Crowe and M.A.Jack,”Globally optimizing formanttracker using generalized centroids,”Electron.Lett.,vol.23,pp.1019-1020,Sept.1987.[2]G.E.Kopec,”Formant tracking using hidden Markovmodels and vector quantization,”IEEE Trans.Acoust.,Speech,Signal Processing,vol.ASSP-34,pp.709-729,Aug.1986.[3]S.McCandless,”An algorithm for automatic formantextraction using linear prediction spectra,”IEEE Trans.Acoust.,Speech,Signal Processing,vol.ASSP-22,pp.135-141,1974.[4]R.C.Snell and inazzo,”Formant location fromLPC analysis data,”IEEE Trans.Speech Audio Process-ing,vol.1,pp.129-134,Apr.1993.[5]L.Welling and Hermann Ney,”Formant Estimation forSpeech Recognition,”IEEE Trans.Speech Audio Pro-cessing,vol.6,pp.36-48,Jan.1998.[6]D.H.Klatt,”Software for a cascade/parallel formantsynthesizer,”J.Acoust.Soc.Amer.,vol.67,pp.970-995,Mar.1980.[7]H.S.Chhatwal and A.G.Constantinides,”Speechspectral segmentation for spectral estimation and for-mant modeling,”in IEEE Int.Conf.Acoustics,Speech,and Signal Processing,Dallas,TX,apr.1987,pp.316-319[8]A.Watanabe,”Formant Estimation Method UsingInverse-Filter Control,”IEEE Trans.Speech Audio Pro-cessing,vol.9,pp.317-326,May 2001.[9]M.Berouti,R.Schwartz and J.Makhoul,”Enhance-ment of speech corrupted by acoustic noise”,IEEE Int.Conf.Acoustics,Speech,and Signal Processing,vol.4,pp.208-211,Apr 1979。
附加噪声的建设性的行动在最佳检测
International Journal of Bifurcation and Chaos,Vol.15,No.9(2005)2985–2994c World Scientific Publishing CompanyCONSTRUCTIVE ACTION OF ADDITIVE NOISE INOPTIMAL DETECTIONFRANC¸OIS CHAPEAU-BLONDEAU and DAVID ROUSSEAULaboratoire d’Ing´e nierie des Syst`e mes Automatis´e s(LISA),Universit´e d’Angers,62avenue Notre Dame du Lac,49000Angers,FranceReceived September20,2004;Revised November9,2004The optimal detection of a signal of known form hidden in additive white noise is examinedin the framework of stochastic resonance and noise-aided information processing.Conditionsare exhibited where the performance in the optimal detection increases when the level of theadditive(non-Gaussian bimodal)noise is raised.On the additive signal–noise mixture,when athreshold quantization is performed prior to the optimal detection,another form of improvementby noise can be obtained,with subthreshold signals and Gaussian noise.Optimization of thequantization threshold shows that even in symmetric detection settings,the optimal thresholdcan be away from the center of symmetry and in subthreshold configuration of the signals.Theseproperties concerning non-Gaussian noise and nonlinear preprocessing in optimal detection,aremeaningful to the current exploration of the various modalities and potentialities of stochasticresonance.Keywords:Stochastic resonance;signal detection;quantizer;noise.1.IntroductionMore and more studies have shown that noise is not necessarily always a nuisance,but can sometimes have a beneficial constructive action.This possi-bility has now been concretized in many different settings and conditions.Stochastic resonance is a generic denomination that can be used to designate such constructive manifestations of the noise[Moss et al.,1994;Chapeau-Blondeau&Godivier,1996; Gammaitoni et al.,1998;And`o&Graziani,2000]. Instances of stochastic resonance have been regis-tered in electronic circuits[Anishchenko et al.,1992, 1994;Godivier et al.,1997;Harmer&Abbott,2000; Morfu et al.,2003],optical devices[McNamara et al.,1988;Dykman et al.,1995;Jost&Saleh,1996; Vaudelle et al.,1998],neural processes[Bulsara et al.,1991;Douglass et al.,1993;Pantazelou et al., 1995;Chapeau-Blondeau&Godivier,1996],nano-technologies[Lee et al.,2003].Many possible dis-tinct forms have appeared for stochastic resonance,depending on the types of processes coupling signal and noise,and the various measures of performance receiving improvement from the noise.Inventory and analysis of these various forms and modalities of stochastic resonance are still ongoing endeavors. The developments are driven both by the impor-tant conceptual significance of stochastic resonance concerning the status of noise,and by its potentiali-ties for applications,especially for information pro-cessing.In particular,stochastic resonance has been investigated within standard signal processing prob-lems,like detection[Zozor&Amblard,2002;Saha &Anand,2003]or estimation[Chapeau-Blondeau &Rojas Varela,2001;Rousseau et al.,2003]of sig-nals in noise.Most forms of stochastic resonance observed so far concern suboptimal processes,in which a pro-cessing system is not tuned at its best,and where the noise is used to alter the operating conditions of the system so as to bring them closer to the best performance.Very recently,the possibility of some29852986 F.Chapeau-Blondeau &D.Rousseauform of stochastic resonance has been extended to optimal processes.Constructive action of the noise was reported in optimal detection of signals cor-rupted by non-additive phase noise in [Rousseau &Chapeau-Blondeau,2002;Chapeau-Blondeau,2003].In the present paper,we shall show that a similar property can be obtained in the more com-mon case of an additive signal–noise mixture.We shall also study the possibility of another type of improvement by noise when nonlinear preprocess-ing under the form of threshold quantization is per-formed prior to the optimal detection.2.Optimal DetectionWe consider a standard detection situation,where a deterministic signal s (t )can assume one among two known expressions s 0(t )(with prior probability P 0)or s 1(t )(with prior probability P 1=1−P 0).This signal s (t )is mixed to a noise η(t ),the resulting mixture forming the observable signal x (t ).This sig-nal x (t )is measured at N distinct times t k ,for k =1to N ,so as to provide N data points x k =x (t k ).We wish to use the data x =(x 1,...,x N )to decide whether the signal s (t )is s 0(t )(hypothesis H 0)or is s 1(t )(hypothesis H 1).According to classical detection theory [Van Trees,2001;Kay,1998],a given detector will decide hypothesis H 0whenever the data x =(x 1,...,x N )falls in the region R 0of I R N ,and it will decide H 1when x falls in the complementary region R 1of I R N .In doing so,the detector achieves an overall probability of detection error P er expressable asP er =P 1R 0p (x |H 1)d x +P 0R 1p (x |H 0)d x ,(1)where p (x |H j )is the probability density for observ-ing x when H j holds,with j ∈{0,1},and the nota-tion ·d x stands for the N -dimensional integral ··· ·dx 1···dx N .Since R 0and R 1are complementary in I R N ,one hasR 0p (x |H 1)d x =1− R 1p (x |H 1)d x ,(2)which substituted in Eq.(1)yieldsP er =P 1+R 1[P 0p (x |H 0)−P 1p (x |H 1)]d x .(3)The detector that minimizes P er can be obtained by making as negative as possible the integral over R 1on the right-hand side of Eq.(3).This is realized by including into R 1all and only those points x for which the integrand P 0p (x |H 0)−P 1p (x |H 1)is negative.This yields the optimal detector,that uses the likelihood ratioL (x )=p (x |H 1)p (x |H 0),(4)to implement the testL (x )H 1><H 0P 0P 1.(5)The minimal P er reached by the optimal detec-tor of Eq.(5)is expressable asP er =I R Nmin[P 0p (x |H 0),P 1p (x |H 1)]d x .(6)Since min(a,b )=(a +b −|a −b |)/2,the minimalprobability of error of Eq.(6)reduces toP er =12−12 I R N |P 1p (x |H 1)−P 0p (x |H 0)|d x .(7)We consider here that the signal–noise mixture x (t )is the additive mixturex (t )=s (t )+η(t ),(8)with η(t )a stationary white noise of cumulativedistribution function F η(u )and probability den-sity function f η(u )=dF η/du .Additive signal–noise mixture is a case very often met in practice.A more complicated nonlinear mixture is con-sidered in [Rousseau &Chapeau-Blondeau,2002;Chapeau-Blondeau,2003].It follows then,that the conditional densities factorize as p (x |H j )= Nk =1p (x k |H j ),withp (x k |H j )=f η[u −s j (t k )],(9)for j ∈{0,1}.Now this last Eq.(9)makes pos-sible the explicit evaluation of the optimal detec-tor of Eqs.(4)and (5),and of its performance [Eq.(7)].3.Constructive Role of NoiseThe level of noise η(t )is quantified by its rmsamplitude ση.A situation often met in practice is the case where η(t )in the mixture of Eq.(8)is a Gaussian noise.In this case,it is well-known that the performance P er in Eq.(7)of the opti-mal detector experiences a monotonic degradationConstructive Action of Additive Noise in Optimal Detection 2987as the noise level σηincreases.However,it is impor-tant to realize that the expectation of a monotonic degradation of the performance P er of an opti-mal detector when the noise level is raised,is not true in generality.This was illustrated in [Rousseau &Chapeau-Blondeau,2002;Chapeau-Blondeau,2003]with a nonlinear signal–noise mixture.We shall show here that the same can occur with the more common linear signal–noise mixture of Eq.(8)when it operates with certain non-Gaussian noises.For our demonstration,we consider in the sequel the basic situation where the signals to be detected are the constant signals s 0(t )=s 0and s 1(t )=s 1,for all t ,with two constants s 0<s 1.In the standard case where the white noise η(t )in Eq.(8)is zero-mean Gaussian,it is well-known that the optimal detector of Eqs.(4)and (5)reduces tox H 1><H 0s 0+s 12+σ2ηN s 1−s 0ln P 0P 1=x T ,(10)with x =N −1N k =1x k .This optimal test of Eq.(10)achieves the probability of error in Eq.(7)which readsP er =12 1+P 1erf √N x T −s 1√2ση−P 0erf√N x T −s 0√2ση.(11)It is easy to verify that the performance P er of Eq.(11)experiences a monotonic degradation as the noise level σηincreases.For the white noise η(t )in Eq.(8)we now turn to a non-Gaussian case,by way of the class of zero-mean Gaussian mixture with standardized proba-bility density (0<m <1)f gm (u )=12√2π1−m2 exp−(u +m )22(1−m 2) +exp−(u −m )22(1−m 2) ,(12)and cumulative distribution functionF gm (u )=12+14 erf u +m √2 1−m 2+erfu −m√21−m2 .(13)As m →0,Eq.(12)approaches the zero-mean unit-variance Gaussian density;as m →1,Eq.(12)approaches the zero-mean unit-variance dichotomic density at ±1.With f η(u )=f gm (u/ση)/ση,Fig.1shows different evolutions of the performance P er in Eq.(7)of the optimal detector,as the noise rms amplitude σηincreases.Figure 1reveals the possibility of nonmono-tonic evolutions of performance measure P er of the optimal detector,as the level σηof the Gaussian-mixture noise is raised.With no noise,at ση=0,0.511.522.533.5400.050.10.150.20.250.30.350.4noise rms amplitudee r r o r p r o b a b i l i t ym=0.8m=0.85m=0.9m=0.95m=0.99(a)0.511.522.533.540.050.10.150.20.250.30.35noise rms amplitudee r r o r p r o b a b i l i t ym=0.8m=0.85m=0.9m=0.95m=0.99(b)Fig.1.Probability of error P er of Eq.(7)for the optimal detector of Eq.(5),as a function of the rms amplitude σηof the Gaussian-mixture noise η(t )from Eq.(12)at different m .Also,s 0(t )≡s 0=−1,s 1(t )≡s 1=1and P 0=1/2;(a)N =1or (b)N =2.2988 F.Chapeau-Blondeau &D.Rousseauthe probability of detection error P er is always at its best value P er =0in Fig.1,as expected for an optimal detector operating in noise-free condition.When the noise level σηrises above zero in Fig.1,the probability of error P er starts gradually to degrade (to increase).However,this degradation of P er does not always proceed monotonically as σηis further increased.Conditions exist in Fig.1,where the performance P er can improve (decrease)when the noise level σηis further raised,over some ranges.This constructive action of the additive noise η(t )on the performance P er of the optimal detector,can be interpreted as a novel aspect of stochastic resonance.At even larger levels,the detrimental action of the noise resumes,and P er degrades again by increasing towards the least favorable value of 1/2.As seen in Fig.1,the constructive action occurs when the noise η(t )departs sufficiently from a Gaussian noise,i.e.when m in Eq.(12)is suffi-ciently close to 1.On the contrary,Gaussian noise or values of m approaching zero in Fig.1,lead to a monotonic increase of P er as σηis raised.A similar behavior with a constructive action of the noise,can be obtained with other non-Gaussian densities for η(t ).Let us consider passing a noise uniform over [−1,1]through the nonlinearityg (u )=1a βu1+(βu )2(14)parameterized by β>0,with a =1−arctan(β)/β.This produces a standardized noise whose probability density f sq (u )is zero for u outside [−g (1),g (1)],and otherwisef sq (u )=12βa [1−(au )2]3/2,(15)and its cumulative distribution function isF sq (u )=12+12βau 1−(au )2(16)over the support u ∈[−g (1),g (1)],and F sq (u )=0for u <−g (1)and F sq (u )=1for u >g (1).As β→0,one recovers the uniform noise over [−√3,√3].For increasing β,the density f sq (u )develops peaks at its two modes in −g (1)and g (1),up to β→+∞which yields a dichotomic noise at ±1.With f η(u )=f sq (u/ση)/ση,Fig.2shows dif-ferent evolutions of the performance P er in Eq.(7)of the optimal detector,as the noise rms amplitude σηincreases.0.511.522.533.5400.020.040.060.080.10.120.140.160.180.20.22noise rms amplitudee r r o r p r o b a b i l i t yβ=2β=3β=4β=5β=10Fig. 2.Probability of error P er of Eq.(7)for the opti-mal detector of Eq.(5),as a function of the rms ampli-tude σηof the noise η(t )from Eq.(15)at different β.Also,s 0(t )≡s 0=−1,s 1(t )≡s 1=1,P 0=1/2and N =1.Figure 2reveals that,as the noise level σηincreases,the possibility of a nonmonotonic evo-lution of the performance P er ,rather than a monotonic degradation,is preserved with the noise density of Eq.(15).This occurs in Fig.2for suf-ficiently large values of β,associated to the noise η(t )with a density from Eq.(15)with a sufficiently pronounced bimodal structure.The bimodal structure of the noise η(t )seems here,both in Figs.1and 2,to be an essential ingre-dient for observing the improvement by noise of P er .This can be understood qualitatively,because a zero-mean bimodal noise with rms amplitude ση,as used in Figs.1and 2,tends to concentrate its ampli-tudes around ±ση,as opposed to a unimodal noise which concentrates its amplitudes around zero.In a binary detection task,noise fluctuations around ±ση,especially for well-chosen ranges of σηin rela-tion to the levels s 0and s 1to be detected,can be less damageable than noise fluctuations around zero to the capacity of distinguishing between s 0and s 1.At the extreme,a purely dichotomic noise at ±σηwould let intact the capacity of distinguishing between s 0and s 1,as long as s 0±σηcannot be con-fused with s 1,which is the rule,except in the very special configuration where ση=s 1−s 0.Especially,when ση>s 1−s 0,the capacity of distinguishing between s 0and s 1is unaffected by the dichotomic noise,however large σηmay be.It is a reminiscence of this property of dichotomic noise,which is at work in Figs.1and 2,to allow the nonmonotonic evolution of the performance P er with continuousConstructive Action of Additive Noise in Optimal Detection2989bimodal noises interpolating between dichotomic and unimodal noises.Beyond the necessity of bimodal noises for the above mechanism to apply,what we wish to empha-size here is the conceptual significance in princi-ple of the results in Figs.1and2.These results establish that an optimal detector operating on an addditive signal–noise mixture can experience an improvement of its performance P er,when the noise level increases,over some ranges of the noise, instead of a monotonic degradation of P er.The same property was shown possible with nonlin-ear signal–noise mixture in[Rousseau&Chapeau-Blondeau,2002;Chapeau-Blondeau,2003],and it is extended here to the more common linear (additive)signal–noise mixture.It is clear in Figs.1 and2that the improvement of P er does not appear as soon as the noise levelσηis raised above zero.A nonzero amount of noiseη(t)has to pre-exist before improvement of P er by a further increase inσηis obtained;but a pre-existing amount of noise is usu-ally the rule in a signal-processing task.Also,for a non-Gaussianη(t),the increase of the rms ampli-tudeσηcannot be achieved by a simple addition of more noise if one wants to keep the same shape for the probability density fη(u),so as to match the conditions of Figs.1and2which increaseσηat fη(u)constant in shape.A more internal adjustable parameter,analog to a physical temperature,has to be assumed to increaseσηat fη(u)constant in shape.Alternatively,the change of the density fη(u)can be explicitly modeled as more noise is added(what is not done here).Improvement by noise can still be expected in these more elaborate conditions,since we show here that such improve-ment is robustly preserved over various shapes for fη(u),this point remaining to be explicitly explored. But again,beyond such issues which are oriented towards practical implementation of the proposed framework,what we want to emphasize here is its conceptual significance:the feasibility in principle of a form of improvement by noise in optimal detec-tion with additive signal–noise mixture.This pos-sibility is an important feature which complements all the aspects and properties known to stochastic resonance.4.Nonlinear Transformation BeforeDetectionIt sometimes happens that a nonlinear transfor-mation is performed on the signal–noise mixture x(t)=s(t)+η(t)prior to the detection process. Such a nonlinear transformation may be imposed by the physics of the sensing or measuring device. Let us consider here the nonlinear transformation, very often considered in the context of stochas-tic resonance,which produces the output signal y(t)asy(t)=sign[s(t)+η(t)−θ]=±1.(17) The transformation of Eq.(17)realizes a one-bit quantization of the input signal–noise mixture x(t)=s(t)+η(t),with quantization thresholdθ. It offers a parsimonious signal representation which can be useful for fast real-time processing;it also bears some similarity with neuronal coding.When the detection is based on y(t),the same considerations as in Sec.2yield the optimal detec-tor asL(y)=Pr{y|H1}Pr{y|H0}H1><H0P0P1,(18)with y=(y1,...,y N)and y k=y(t k)for k=1 to N.Equation(18)is the minimal-P er detector, achieving the minimum P er among all detection schemes based on y,this minimum beingP er=12−12y∈{−1,1}N|P1Pr{y|H1}−P0Pr{y|H0}|(19) with the sum performed over the2N distinct states (y1=±1,...,y N=±1)accessible to the data y.Forη(t)a white noise,we have Pr{y|H j}= Nk=1Pr{y k|H j},for j∈{0,1}.At any time t,we have according to Eq.(17),the conditional prob-ability Pr{y(t)=−1|H j}which is also Pr{s j(t)+η(t)≤θ},this amounting toPr{y(t)=−1|H j}=Fη[θ−s j(t)].(20)In the same way,we have Pr{y(t)=1|H j}=1−Fη[θ−s j(t)].This allows the explicit evaluation of the optimal detector of Eq.(18)and of its perfor-mance of Eq.(19).We again consider in the following,the basic situation where the signals to be detected are the constant signals s0(t)=s0and s1(t)=s1>s0,for all t.In this case,the optimal test of Eq.(18)can2990 F.Chapeau-Blondeau &D.Rousseaube reduced to a simpler expression taking the formN 1H 1><H 0ln P 0P 1−N lnF η(θ−s 1)F η(θ−s 0) ln1−F η(θ−s 1)1−F η(θ−s 0) −ln F η(θ−s 1)F η(θ−s 0)=N T ,(21)where N 1is the number,between 0and N ,of com-ponents y k at +1in the data y .This optimal test of Eq.(21)achieves the probability of errorP er =P 1N 1<N TC N N 1[1−F η(θ−s 1)]N 1×F η(θ−s 1)N −N 1+P 0N 1≥N TC NN 1[1−F η(θ−s 0)]N 1×F η(θ−s 0)N −N 1,(22)with C N N 1as the binomial coefficients.When the noise η(t )has a Gaussian-mixture density as in Eq.(12),then Fig.3represents the evolution of the probability of error P er of Eq.(22)for the optimal detector of Eq.(21),as the rms amplitude σηof the noise η(t )is raised.It is interest-ing to compare Fig.3characterizing the detection from the quantized data y ,to Fig.1characterizing the detection from the analog (unquantized)datax in otherwise similar conditions.Two important observations can be made in this respect.The first observation is that P er in the detection with the quantized data y is always larger (in the same noise condition)than with the analog unquan-tized data x .This observation has a natural expla-nation:one-bit quantization by y of the analog data x entails a loss of information,whence the reduced performance in detection.The reduced performance is even more pronounced at N =2in Fig.3(b)than at N =1in Fig.3(a),because the loss of information increases with the number N of data points.The trade-offis that y represents a much more parcimonious representation with only one bit per data point which can be useful for fast real-time processing,compared to x which in principle requires a infinite number (12to 16in practice)of bits per data point.The second observation is that the construc-tive action of noise present in Fig.1is absent in Fig.3.When the thresholding is done prior to the detection,the detector cannot benefit from a soft,smooth,analog representation of the data,which is somehow richer than the hard-thresholded rep-resentation imposed to base the decision in Fig.3.The constructive action of the noise can operate in the soft analog representation of Fig.1,but is sup-pressed in the thresholded representation of Fig.3.This has been revealed by the present analysis.This is somehow a novel aspect which enriches the0.511.522.533.5400.050.10.150.20.250.30.350.40.450.5noise rms amplitudee r r o r p r o b a b i l i t ym=0.8m=0.85m=0.9m=0.95m=0.99(a)0.511.522.533.5400.050.10.150.20.250.30.350.40.450.5noise rms amplitudee r r o r p r o b a b i l i t ym=0.8m=0.85m=0.9m=0.95m=0.99(b)Fig.3.Solid lines:probability of error P er of Eq.(22)for the optimal detector of Eq.(21)from the quantized data y ,as a function of the rms amplitude σηof the Gaussian-mixture noise η(t )from Eq.(12)at different m .The quantization threshold in Eq.(17)is θ=0.Also as in Fig.1:s 0(t )≡s 0=−1,s 1(t )≡s 1=1and P 0=1/2;(a)N =1or (b)N =2.The dotted lines are P er redrawn from Fig.1for the detection from the analog (unquantized)data x .Constructive Action of Additive Noise in Optimal Detection 2991properties known to stochastic resonance.Improve-ment by noise usually occurs,in known forms of stochastic resonance,for a nonlinear transformation of an input signal–noise mixture.Here,improve-ment by noise is possible on the linear input signal–noise mixture x (t ),and disappears after a nonlinear transformation on x (t ).However,if the nonlinear transformation of Fig.3performed with the quantization threshold θ=0,is modified by varying θ,then a construc-tive action of the noise can be recovered.Figure 4addresses a situation where the thresholding of Eq.(17)is performed in such a way that both sig-nals s 0(t )and s 1(t )to be detected are on the same side of the threshold θ.In Fig.4,when the noise η(t )is absent in Eq.(17),both signals s 0(t )and s 1(t )are always below the quantization threshold θ,and are there-fore always quantized exactly in the same way.In this case,at η(t )≡0,no discriminating detection is possible based on the quantized data y and the performance is at its worst,i.e.P er =1/2in Fig.4.Next,as the noise level σηis raised above zero,the presence of the noise η(t )progressively allows the signals s 0(t )and s 1(t )to be quantized differently by Eq.(17).This translates in Fig.4,into an improve-ment of the detection performance P er as the noise level σηincreases,up to an optimal nonzero noise level where the probability of detection error P er is minimized.The constructive action of the noise,or stochastic resonance,is recovered,under the form of a noise-assisted detection of subthreshold signals.For subthreshold signals,the constructive action of the noise is possible with the Gaussian mixture noise of Fig.4,but it is also possible with Gaussian noise,as shown in Fig.5.Figures 4and 5also confirm a remark made above for Fig.3,that the detection from the quan-tized data y never improves over the detection from the analog (unquantized)data x .This is observed whatever the position of the quantization threshold θ,either in a suprathreshold (Fig.3)or a subthresh-old (Fig.4)configuration of the signals s 0(t )and s 1(t ),as also confirmed in Fig.5.Another important observation in Figs.4and 5,is that a quantization threshold θin a subthreshold configuration,can lead to a better detection perfor-mance P er compared to θin a suprathreshold config-uration of the signals s 0(t )and s 1(t ).For instance,in Figs.4and 5,for the detection of s 0(t )≡−1and s 1(t )≡1,at large noise levels ση,the performance P er in the subthreshold configuration θ=1.1is gen-erally better than that in the suprathreshold config-uration θ=0.This is true for any N in Fig.4with Gaussian-mixture noise,and for N >1in Fig.5with Gaussian noise.This is an important property for the use of quantization devices,as often con-sidered for stochastic resonance:even in completely symmetric conditions of the signals and noise and process,the optimal configuration for the quanti-zation threshold is not necessarily at the center of symmetry θ=0.The above observation naturally leads to raise the issue of optimizing the quantization threshold θ0.511.522.533.5400.050.10.150.20.250.30.350.40.450.5noise rms amplitudee r r o r p r o b a b i l i t ym=0.8m=0.99(a)0.511.522.533.5400.050.10.150.20.250.30.350.40.450.5noise rms amplitudee r r o r p r o b a b i l i t ym=0.8m=0.99(b)Fig.4.Same as Fig.3,except that θ=1.1.The dashed lines are redrawn from Fig.3.2992 F.Chapeau-Blondeau &D.Rousseau0.511.522.533.5400.050.10.150.20.250.30.350.40.450.5θ=2noise rms amplitudee r r o r p r o b a b i l i t yθ=0θ=1.21.10.90.5(a)0.511.522.533.5400.050.10.150.20.250.30.350.40.450.5noise rms amplitudee r r o r p r o b a b i l i t yθ=00.5θ=0.91.1θ=1.2θ=2(b)Fig.5.Solid and dashed lines:probability of error P er of Eq.(22)for the optimal detector of Eq.(21)from the quantized data y ,as a function of the rms amplitude σηof the Gaussian noise η(t ),and different quantization thresholds θ.Also as in Fig.1:s 0(t )≡s 0=−1,s 1(t )≡s 1=1and P 0=1/2;(a)N =1or (b)N =2.In (b),the dotted line is P er of Eq.(11)for detection from the analog (unquantized)data x ;in (a)this line is superimposed to the curve at θ=0.0.511.522.533.5400.511.52noise rms amplitudeo p t i m a l t h r e s h o l dN=4N=300.511.522.533.5400.10.20.30.4noise rms amplitudee r r o r p r o b a b i l i t yN=23N=4N=2Fig.6.Performance of the optimal detector of Eq.(21)from the quantized data y ,as a function of the rms amplitude σηof the Gaussian noise η(t ),and different number N of data points.Also as in Fig.1:s 0(t )≡s 0=−1,s 1(t )≡s 1=1and P 0=1/2.Upper panel:optimal value θopt of the quan-tization threshold θin Eq.(17)minimizing P er of Eq.(22).Due to the symmetry of the process,−θopt is also an optimal threshold.Lower panel:Minimum P er at θopt .so as to maximize the performance (minimize P er )in given conditions of the noise η(t )and signals to be detected.For illustration,this issue is solved in Fig.6for conditions with Gaussian noise η(t ).With Gaussian noise,the results of Fig.6reveal that the optimal threshold θopt is never zero for N even,although it is always zero for N odd.The same trend that takes θopt away from zero is even more manifest with non-Gaussian noise η(t );for instance,with the Gaussian-mixture noise of Eq.(12),θopt never remains at zero,even for N odd.Moreover,Fig.6shows that the optimal threshold θopt can lie in a subthreshold configuration of the signals s 0(t )and s 1(t )to be detected,i.e.θopt is above 1for detection between s 0(t )≡−1and s 1(t )≡1at large values of the noise rms amplitude σηand N even.However,as also shown in Fig.6,when the process is tuned at θopt ,the resulting probability of detec-tion error P er is generally an increasing function of the noise level ση.Improvement of P er by increasing σηis feasible when θis not at its optimal position,as in Fig.4,but disappears when θis at its optimum θopt .Noise improvement occurs here in nonoptimal processes,while it occurred in optimal processes in Sec.3.5.Summary and OutlookWe have examined the optimal detection of a sig-nal of known form hidden in additive white noise,in the framework of stochastic resonance or noise-aided information processing.Several conclusions,meaningful in this framework,can be emphasized as follows.。
麦克风阵列基本原理英文教材
where r = |r| is the radial distance from the source, and k is the scalar wavenumber, given by 2π /λ. The spherical wave solution shows that the signal amplitude decays at a rate proportional to the distance from the source. This dependence of the amplitude on the distance has important implications for array processing algorithms when the source is in the near-field, as will be discussed in later sections. While sound waves are typically spherical in nature, they may be considered as plane waves at a sufficient distance from the source, and this approximation is often used to simplify mathematical analysis. The plane wave solution in Equation 3 is expressed in terms of two variables, time and space. Due to the well defined propagation of the signal, these two variables are linked by a simple relation, and thus the solution can be expressed as function of a single variable. If we formulate the plane wave solution as x(t, r) = Aej ω(t−β·r) where β = as
EEMD的开创性论文(2009年刊载于Advances in Adaptive Data Analysis)
Advances in Adaptive Data AnalysisVol.1,No.1(2009)1–41c World Scientific Publishing CompanyENSEMBLE EMPIRICAL MODE DECOMPOSITION:A NOISE-ASSISTED DATA ANALYSIS METHODZHAOHUA WU∗and NORDEN E.HUANG†∗Center for Ocean–Land–Atmosphere Studies4041Powder Mill Road,Suite302Calverton,MD20705,USA†Research Center for Adaptive Data AnalysisNational Central University300Jhongda Road,Chungli,Taiwan32001A new Ensemble Empirical Mode Decomposition(EEMD)is presented.This newapproach consists of sifting an ensemble of white noise-added signal(data)and treatsthe mean as thefinal true result.Finite,not infinitesimal,amplitude white noise isnecessary to force the ensemble to exhaust all possible solutions in the sifting process,thus making the different scale signals to collate in the proper intrinsic mode functions(IMF)dictated by the dyadicfilter banks.As EEMD is a time–space analysis method,the added white noise is averaged out with sufficient number of trials;the only persistentpart that survives the averaging process is the component of the signal(original data),which is then treated as the true and more physical meaningful answer.The effect ofthe added white noise is to provide a uniform reference frame in the time–frequencyspace;therefore,the added noise collates the portion of the signal of comparable scalein one IMF.With this ensemble mean,one can separate scales naturally without anya priori subjective criterion selection as in the intermittence test for the original EMDalgorithm.This new approach utilizes the full advantage of the statistical characteristicsof white noise to perturb the signal in its true solution neighborhood,and to cancel itselfout after serving its purpose;therefore,it represents a substantial improvement over theoriginal EMD and is a truly noise-assisted data analysis(NADA)method.Keywords:Empirical Mode Decomposition(EMD);ensemble empirical mode decompo-sitions;noise-assisted data analysis(NADA);Intrinsic Mode Function(IMF);shiftingstoppage criteria;end effect reduction.1.IntroductionThe Empirical Mode Decomposition(EMD)has been proposed recently1,2as an adaptive time–frequency data analysis method.It has been proved quite versatile in a broad range of applications for extracting signals from data generated in noisy nonlinear and nonstationary processes(see,e.g.,Refs.3and4).As useful as EMD proved to be,it still leaves some annoying difficulties unresolved.One of the major drawbacks of the original EMD is the frequent appearance of mode mixing,which is defined as a single Intrinsic Mode Function(IMF)either12Z.Wu&N.E.Huangconsisting of signals of widely disparate scales,or a signal of a similar scale resid-ing in different IMF components.Mode mixing is often a consequence of signal intermittency.As discussed by Huang et al.,1,2the intermittence could not only cause serious aliasing in the time–frequency distribution,but also make the phys-ical meaning of individual IMF unclear.To alleviate this drawback,Huang et al.2 proposed the intermittence test,which can indeed ameliorate some of the difficul-ties.However,the approach has its own problems:first,the intermittence test is based on a subjectively selected scale.With this subjective intervention,the EMD ceases to be totally adaptive.Second,the subjective selection of scales works if there are clearly separable and definable timescales in the data.In case the scales are not clearly separable but mixed over a range continuously,as in the case of the majority of natural or man-made signals,the intermittence test algorithm with subjectively defined timescales often does not work very well.To overcome the scale separation problem without introducing a subjective intermittence test,a new noise-assisted data analysis(NADA)method is proposed, the Ensemble EMD(EEMD),which defines the true IMF components as the mean of an ensemble of trials,each consisting of the signal plus a white noise offinite amplitude.It should be noted here that we use word‘single’instead of word‘data’in this paper(except in some part of Sec.2)because the purpose of this paper is to decompose the whole targeted data but not to identify the particular part that is known a priori as containing interesting information.Since there is added noise in the decomposition method,we refer the original data as‘signal’in most occasions. With this ensemble approach,we can clearly separate the scale naturally without any a priori subjective criterion selection.This new approach is based on the insight gleaned from recent studies of the statistical properties of white noise,5,6which showed that the EMD is effectively an adaptive dyadicfilter bank a when applied to white noise.More critically,the new approach is inspired by the noise-added anal-yses initiated by Flandrin et al.7and Gledhill.8Their results demonstrated that noise could help data analysis in the EMD.The principle of the EEMD is simple:the added white noise would populate the whole time–frequency space uniformly with the constituting components of different scales.When signal is added to this uniformly distributed white back-ground,the bits of signal of different scales are automatically projected onto proper scales of reference established by the white noise in the background.Of course, each individual trial may produce very noisy results,for each of the noise-added decompositions consists of the signal and the added white noise.Since the noise in each trial is different in separate trials,it is canceled out in the ensemble mean of a A dyadicfilter bank is a collection of band-passfilters that have a constant band-pass shape(e.g., a Gaussian distribution)but with neighboringfilters covering half or double of the frequency range of any singlefilter in the bank.The frequency ranges of thefilters can be overlapped.For example, a simple dyadicfilter bank can includefilters covering frequency windows such as50to120Hz, 100to240Hz,200to480Hz,etc.Ensemble Empirical Mode Decomposition3 enough trials.The ensemble mean is treated as the true answer,for,in the end, the only persistent part is the signal as more and more trials are added in the ensemble.The critical concept advanced here is based on the following observations:1.A collection of white noise cancels each other out in a time–space ensemble mean;therefore,only the signal can survive and persist in thefinal noise-added signal ensemble mean.2.Finite,not infinitesimal,amplitude white noise is necessary to force the ensembleto exhaust all possible solutions;thefinite magnitude noise makes the different scale signals reside in the corresponding IMF,dictated by the dyadicfilter banks, and render the resulting ensemble mean more meaningful.3.The true and physically meaningful answer to the EMD is not the one withoutnoise;it is designated to be the ensemble mean of a large number of trials consisting of the noise-added signal.This EEMD proposed here has utilized many important statistical characteris-tics of noise.We will show that the EEMD utilizes the scale separation capability of the EMD,and enables the EMD method to be a truly dyadicfilter bank for any data. By addingfinite noise,the EEMD eliminated largely the mode mixing problem and preserve physical uniqueness of decomposition.Therefore,the EEMD represents a major improvement of the EMD method.In the following sections,a systematic exploration of the relation between noise and signal in data will be presented.Studies of Flandrin et al.5and Wu and Huang6 have revealed that the EMD serves as a dyadicfilter for various types of noise.This implies that a signal of a similar scale in a noisy data set could possibly be contained in one IMF component.It will be shown that adding noise withfinite rather than infinitesimal amplitude to data indeed creates such a noisy data set;therefore, the added noise,havingfilled all the scale space uniformly,can help to eliminate the annoying mode mixing problemfirst noticed by Huang et al.2Based on these results,we will propose formally the concepts of NADA and noise-assisted signal extraction(NASE),and will develop a method called the EEMD,which is based on the original EMD method,to make NADA and NASE possible.The paper is arranged as follows.Section2will summarize previous attempts of using noise as a tool in data analysis.Section3will introduce the EEMD method, illustrate more details of the drawbacks associated with mode mixing,present concepts of NADA and of NASE,and introduce the EEMD in detail.Section4 will display the usefulness and capability of the EEMD through examples.Sec-tion5will further discuss the related issues to the EEMD,its drawbacks,and their corresponding solutions.A summary and discussion will be presented in the final section of the main text.Two appendices will discuss some related issues of EMD algorithm and a Matlab EMD/EEMD software for research community to use.4Z.Wu&N.E.Huang2.A Brief Survey of Noise-Assisted Data AnalysisThe word“noise”can be traced etymologically back to its Latin root of“nausea,”meaning“seasickness.”Only in Middle English and Old French does it start to gain the meaning of“noisy strife and quarrel,”indicating something not at all desirable. Today,the definition of noise varies in different circumstances.In science and engi-neering,noise is defined as disturbance,especially a random and persistent kind that obscures or reduces the clarity of a signal.In natural phenomena,noise could be induced by the process itself,such as local and intermittent instabilities,irresolv-able subgrid phenomena,or some concurrent processes in the environment in which the investigations are conducted.It could also be generated by the sensors and recording systems when observations are made.When efforts are made to under-stand data,important differences must be considered between the clean signals that are the direct results of the underlying fundamental physical processes of our inter-est(“the truth”)and the noise induced by various other processes that somehow must be removed.In general,all data are amalgamations of signal and noise,i.e.,x(t)=s(t)+n(t),(1) in which x(t)is the recorded data,and s(t)and n(t)are the true signal and noise,respectively.Because noise is ubiquitous and represents a highly undesirable and dreaded part of any data,many data analysis methods were designed specifi-cally to remove the noise and extract the true signals in data,although often not successful.Since separating the signal and the noise in data is necessary,three important issues should be addressed:(1)The dependence of the results on the analysis meth-ods used and assumptions made on the data.(For example,a linear regression of data implicitly assumes the underlying physics of the data to be linear,while a spectrum analysis of data implies the process is stationary.)(2)The noise level to be tolerated in the extracted“signals,”for no analysis method is perfect,and in almost all cases the extracted“signals”still contain some noise.(3)The portion of real signal obliterated or deformed through the analysis processing as part of the noise.(For example,Fourierfiltering can remove harmonics through low-pass filtering and thus deform the waveform of the fundamentals.)All these problems cause misinterpretation of data,and the latter two issues are specifically related to the existence and removal of noise.As noise is ubiquitous, steps must be taken to insure that any meaningful result from the analysis should not be contaminated by noise.To avoid possible illusion,the null hypothesis test against noise is often used with the known noise characteristics associated with the analysis method.6,9,7Although most data analysis techniques are designed specifi-cally to remove noise,there are,however,cases when noise is added in order to help data analysis,to assist the detection of weak signals,and to delineate the under-lying processes.The intention here is to provide a brief survey of the beneficial utilization of noise in data analysis.Ensemble Empirical Mode Decomposition5 The earliest known utilization of noise in aiding data analysis was due to Press and Tukey10known as pre-whitening,where white noise was added toflatten the narrow spectral peaks in order to get a better spectral estimation.Since then, pre-whitening has become a very common technique in data analysis.For exam-ple,Fuenzalida and Rosenbluth11added noise to process climate data;Link and Buckley,12and Zala et al.13used noise to improve acoustic signal;Strickland and Il Hahn14used wavelet and added noise to detect objects in general;and Trucco15 used noise to help design specialfilters for detecting embedded objects on the ocean floor experimentally.Some general problems associated with this approach can be found in the works by Priestley,16Kao et al.,17Politis,18and Douglas et al.19 Another category of popular use of noise in data analysis is more related to the analysis method than to help extracting the signal from the data.Adding noise to data helps to understand the sensitivity of an analysis method to noise and the robustness of the results obtained.This approach is used widely;for example, Cichocki and Amari20added noise to various data to test the robustness of the independent component analysis(ICA)algorithm,and De Lathauwer et al.21used noise to identify error in ICA.Adding noise to the input to specifically designed nonlinear detectors could also be beneficial to detecting weak periodic or quasi-periodic signals based on a physical process called stochastic resonance.The study of stochastic resonance was pioneered by Benzi and his colleagues in the early1980s.The details of the development of the theory of stochastic resonance and its applications can be found in a lengthy review paper by Gammaitoni et al.22It should be noted here that most of the past applications(including those mentioned earlier)have not used the cancellation effects associated with an ensemble of noise-added cases to improve their results.Specific to analysis using EMD,Huang et al.23added infinitesimal magnitude noise to earthquake data in an attempt to prevent the low frequency mode from expanding into the quiescent region.But they failed to realize fully the implications of the added noise in the EMD method.The true advances related to the EMD method had to wait until the two pioneering works by Gledhill8and Flandrin et al.7 Flandrin et al.7used added noise to overcome one of the difficulties of the original EMD method.As the EMD is solely based on the existence of extrema (either in amplitude or in curvature),the method ceases to work if the data lacks the necessary extrema.An extreme example is in the decomposition of a Dirac pulse (delta function),where there is only one extrema in the whole data set.To overcome the difficulty,Flandrin et al.7suggested adding noise with infinitesimal amplitude to the Dirac pulse so as to make the EMD algorithm operable.Since the decomposition results are sensitive to the added noise,Flandrin et al.7ran an ensemble of5000 decompositions,with different realizations of noise,all of infinitesimal amplitude. Though they used the mean as thefinal decomposition of the Dirac pulse,they defined the true answer asE{d[n]+εr k[n]},(2)d[n]=lime→0+6Z.Wu&N.E.Huangin which,[n]represents n th data point,d[n]is the Dirac function,r k[n]is a random number,εis the infinitesimal parameter,and E{}is the expected value.Flandrin’s novel use of the added noise has made the EMD algorithm operable for a data set that could not be previously analyzed.Another novel use of noise in data analysis is by Gledhill,8who used noise to test the robustness of the EMD algorithm.Although an ensemble of noise was used, he never used the cancellation principle to define the ensemble mean as the true answer.Based on his discovery(that noise could cause the EMD to produce slightly different outcomes),he assumed that the result from the clean data without noise was the true answer and thus designated it as the reference.He then defined the discrepancy,∆,as∆=mj=1t(cr j(t)−cn j(t))21/2,(3)where cr j and cn j are the j th component of the IMF without and with noise added, and m is the total number of IMFs generated from the data.In his extensive study of the detailed distribution of the noise-caused“discrepancy,”he concluded that the EMD algorithm is reasonably stable for small perturbations.This conclusion is in slight conflict with his observations that the perturbed answer with infinitesimal noise showed a bimodal distribution of the discrepancy.Gledhill had also pushed the noise-added analysis in another direction:he had proposed to use an ensemble mean of noise-added analysis to form a“Composite Hilbert spectrum.”As the spectrum is non-negative,the added noise could not cancel out.He then proposed to keep a noise-only spectrum and subtract it from the full noise-added spectrum at the end.This non-cancellation of noise in the spectrum,however,forced Gledhill8to limit the noise used to be of small magnitude, so that he could be sure that there would not be too much interaction between the noise-added and the original clean signal,and that the contribution of the noise to thefinal energy density in the spectrum would be negligible.Although noise of infinitesimal amplitude used by Gledhill8has improved the confidence limit of thefinal spectrum,Gledhill explored neither fully the cancella-tion property of the noise nor the power offinite perturbation to explore all possible solutions.Furthermore,it is well known that whenever there is intermittence,the signal without noise can produce IMFs with mode mixing.There is no justification to assume that the result without added noise is the truth or the reference sig-nal.These reservations notwithstanding,all these studies by Flandrin et al.7and Gledhill8had still greatly advanced the understanding of the effects of noise in the EMD method,though the crucial effects of noise had yet to be clearly articulated and fully explored.In the following,the new noise-added EMD approach will be explained,in which the cancellation principle will be fully utilized,even withfinite amplitude noise.Also emphasized is thefinding that the true solution of the EMD method should be theEnsemble Empirical Mode Decomposition7 ensemble mean rather than the clean data.This full presentation of the new method will be the subject of the next section.3.Ensemble Empirical Mode Decomposition3.1.The empirical mode decompositionThis section starts with a brief review of the original EMD method.The detailed method can be found in the works of Huang et al.1and Huang et al.2Different to almost all previous methods of data analysis,the EMD method is adaptive,with the basis of the decomposition based on and derived from the data.In the EMD approach,the data X(t)is decomposed in terms of IMFs,c j,i.e.,x(t)=nj=1c j+r n,(4)where r n is the residue of data x(t),after n number of IMFs are extracted.IMFs are simple oscillatory functions with varying amplitude and frequency,and hence have the following properties:1.Throughout the whole length of a single IMF,the number of extrema and thenumber of zero-crossings must either be equal or differ at most by one(although these numbers could differ significantly for the original data set).2.At any data location,the mean value of the envelope defined by the local maximaand the envelope defined by the local minima is zero.In practice,the EMD is implemented through a sifting process that uses only local extrema.From any data r j−1,say,the procedure is as follows:(1)identify all the local extrema(the combination of both maxima and minima)and connect all these local maxima(minima)with a cubic spline as the upper(lower)envelope;(2)obtain thefirst component h by taking the difference between the data and the local mean of the two envelopes;and(3)Treat h as the data and repeat steps1and 2as many times as is required until the envelopes are symmetric with respect to zero mean under certain criteria.Thefinal h is designated as c j.A complete sifting process stops when the residue,r n,becomes a monotonic function from which no more IMFs can be extracted.Based on this simple description of EMD,Flandrin et al.5and Wu and Huang6 have shown that,if the data consisted of white noise which has scales populated uniformly through the whole timescale or time–frequency space,the EMD behaves as a dyadicfilter bank:the Fourier spectra of various IMFs collapse to a single shape along the axis of logarithm of period or frequency.Then the total number of IMFs of a data set is close to log2N with N the number of total data points. When the data is not pure noise,some scales could be missing;therefore,the total number of the IMFs might be fewer than log2N.Additionally,the intermittency of signals in certain scale would also cause mode mixing.8Z.Wu&N.E.Huang3.2.Mode mixing problem“Mode mixing”is defined as any IMF consisting of oscillations of dramatically disparate scales,often caused by intermittency of the driving mechanisms.When mode mixing occurs,an IMF can cease to have physical meaning by itself,suggest-ing falsely that there may be different physical processes represented in a mode. Even though thefinal time–frequency projection could rectify the mixed mode to some degree,the alias at each transition from one scale to another would irrecov-erably damage the clean separation of scales.Such a drawback wasfirst illustrated by Huang et al.2in which the modeled data was a mixture of intermittent high-frequency oscillations riding on a continuous low-frequency sinusoidal signal.An almost identical example used by Huang et al.2is presented here in detail as an illustration.The data and its sifting process are illustrated in Fig.1.The data has its funda-mental part as a low-frequency sinusoidal wave with unit amplitude.At the three(a)(b)(c)(d)Fig.1.The veryfirst step of the sifting process.Panel(a)is the input;panel(b)identifies local maxima(gray dots);panel(c)plots the upper envelope(upper gray dashed line)and low envelope (lower gray dashed line)and their mean(bold gray line);and panel(d)is the difference between the input and the mean of the envelopes.Ensemble Empirical Mode Decomposition9 middle crests of the low-frequency wave,high-frequency intermittent oscillations with an amplitude of0.1are riding on the fundamental,as panel(a)of Fig.1 shows.The sifting process starts with identifying the maxima(minima)in the data.In this case,15local maxima are identified,with thefirst and the last coming from the fundamental,and the other13caused mainly by intermittent oscillations (panel(b)).As a result,the upper envelope resembles neither the upper envelope of the fundamental(which is aflat line at one)nor the upper one of the intermittent oscillations(which is supposed to be the fundamental outside intermittent areas). Rather,the envelope is a mixture of the envelopes of the fundamental and of the intermittent signals that lead to a severely distorted envelope mean(the thick gray line in panel(c)).Consequently,the initial guess of thefirst IMF(panel(d))is the mixture of both the low-frequency fundamental and the high-frequency intermittent waves,as shown in Fig.2.An annoying implication of such scale mixing is related to unstableness and lack of the uniqueness of decomposition using the EMD.With stoppage criterion given and end-point approach prescribed in the EMD,the application of the EMD to any real data results in a unique set of IMFs,just as when the data is processed by other data decomposition methods.This uniqueness is here referred to as“the mathematical uniqueness,”and satisfaction to the mathematical uniqueness is the minimal requirement for any decomposition method.The issue that is emphasizedFig.2.The intrinsic mode functions of the input displayed in Fig.1(a).10Z.Wu&N.E.Huanghere is what we refer to as“the physical uniqueness.”Since real data almost always contains a certain amount of random noise or intermittences that are not known to us,an important issue,therefore,is whether the decomposition is sensitive to noise.If the decomposition is insensitive to added noise of small butfinite ampli-tude and bears little quantitative and no qualitative change,the decomposition is generally considered stable and satisfies the physical uniqueness;and otherwise, the decomposition is unstable and does not satisfy the physical uniqueness.The result from decomposition that does not satisfy the physical uniqueness may not be reliable and may not be suitable for physical interpretation.For many traditional data decomposition methods with prescribed base functions,the uniqueness of the second kind is automatically satisfied.Unfortunately,the EMD in general does not satisfy this requirement due to the fact that decomposition is solely based on the distribution of extrema.To alleviate this drawback,Huang et al.2proposed an intermittence test that subjectively extracts the oscillations with periods significantly smaller than a pre-selected value during the sifting process.The method works quite well for this example.However,for complicated data with scales variable and continuously dis-tributed,no single criterion of intermittence test can be selected.Furthermore,the most troublesome aspect of this subjectively pre-selected criterion is that it lacks physical justifications and renders the EMD nonadaptive.Additionally,mode mix-ing is also the main reason that renders the EMD algorithm unstable:any small perturbation may result in a new set of IMFs as reported by Gledhill.8Obviously, the intermittence prevents EMD from extracting any signal with similar scales. To solve these problems,the EEMD is proposed,which will be described in the following sections.3.3.Ensemble empirical mode decompositionAs given in Eq.(1),all data are amalgamations of signal and noise.To improve the accuracy of measurements,the ensemble mean is a powerful approach,where data are collected by separate observations,each of which contains different noise.To generalize this ensemble idea,noise is introduced to the single data set,x(t),as if separate observations were indeed being made as an analog to a physical experiment that could be repeated many times.The added white noise is treated as the possible random noise that would be encountered in the measurement process.Under such conditions,the i th“artificial”observation will bex i(t)=x(t)+w i(t).(5) In the case of only one observation,each multiple-observation ensembles is mim-icked by adding not arbitrary but different realizations of white noise,w i(t),to that single observation as given in Eq.(5).Although adding noise may result in smaller signal-to-noise ratio,the added white noise will provide a relatively uniform ref-erence scale distribution to facilitate EMD;therefore,the low signal–noise ratiodoes not affect the decomposition method but actually enhances it to avoid the mode mixing.Based on this argument,an additional step is taken by arguing that adding white noise may help to extract the true signals in the data,a method that is termed EEMD,a truly NADA method.Before looking at the details of the new EEMD,a review of a few properties of the original EMD is presented:1.the EMD is an adaptive data analysis method that is based on local characteris-tics of the data,and hence,it catches nonlinear,nonstationary oscillations more effectively;2.the EMD is a dyadicfilter bank for any white(or fractional Gaussian)noise-onlyseries;3.when the data is intermittent,the dyadic property is often compromised in theoriginal EMD as the example in Fig.2shows;4.adding noise to the data could provide a uniformly distributed reference scale,which enables EMD to repair the compromised dyadic property;and5.the corresponding IMFs of different series of noise have no correlation with eachother.Therefore,the means of the corresponding IMFs of different white noise series are likely to cancel each other.With these properties of the EMD in mind,the proposed EEMD is developed as follows:1.add a white noise series to the targeted data;2.decompose the data with added white noise into IMFs;3.repeat step1and step2again and again,but with different white noise serieseach time;and4.obtain the(ensemble)means of corresponding IMFs of the decompositions asthefinal result.The effects of the decomposition using the EEMD are that the added white noise series cancel each other in thefinal mean of the corresponding IMFs;the mean IMFs stay within the natural dyadicfilter windows and thus significantly reduce the chance of mode mixing and preserve the dyadic property.To illustrate the procedure,the data in Fig.1is used as an example.If the EEMD is implemented with the added noise having an amplitude of0.1standard deviation of the original data for just one trial,the result is given in Fig.3.Here,the low-frequency component is already extracted almost perfectly.The high-frequency components,however,are buried in noise.Note that high-frequency intermittent signal emerges when the number of ensemble members increases,as Fig.4dis-plays.Clearly,the fundamental signal C5is represented nearly perfect,as well as the intermittent signals,if C2and C3are added together.The fact that the intermittent signal actually resides in two EEMD components is due to the aver-age spectra of neighboring IMFs of white noise overlapping,as revealed by Wu。
Decoherence of the Superconducting Persistent Current Qubit
a r X i v :c o n d -m a t /9910062v 3 [c o n d -m a t .m e s -h a l l ] 27 A p r 2000Decoherence of the Superconducting Persistent Current Qubit Lin Tian 1,L.S.Levitov 1,Caspar H.van der Wal 4,J.E.Mooij 2,4,T.P.Orlando 2,S.Lloyd 3,C.J.P.M.Harmans 4,J.J.Mazo 2,51Dept.of Physics,Center for Material Science &Engineering,2Dept.of Electrical Engineering and Computer Science,3Dept.of Mechanical Engineering,Massachusetts Institute of Technology;4Dept.of Applied Physics and Delft Institute for Microelectronics and Submicron Technologies,Delft Univ.of Technology;5Dept.de F´ısica de la Mataeria Condensada,Universidad de Zaragoza (February 1,2008)Decoherence of a solid state based qubit can be caused by coupling to microscopic degrees of freedom in the solid.We lay out a simple theory and use it to estimate decoherence for a recently proposed superconducting persistent current design.All considered sources of decoherence are found to be quite weak,leading to a high quality factor for this qubit.I.INTRODUCTION The power of quantum logic [1]depends on the degree of coherence of the qubit dynamics [2,3].The so-called “quality factor”of the qubit,the number of quantum operations performed during the qubit coherence time,should be at least 104for the quantum computer to allow for quantum error correction [4].Decoherence is an especially vital issue in solid state qubit designs,due to many kinds of low energy excitations in the solid state environment that may couple to qubit states and cause dephasing.In this article we discuss and estimate some of the main sources of decoher-ence in the superconducting persistent current qubit proposed recently [3].The approach will be presented in a way making it easy to generalize it to other sys-tems.We emphasize those decoherence mechanisms that illustrate this approach,and briefly summarize the results of other mechanisms.The circuit [3]consists of three small Josephson junctions which are connected in series,forming a loop,as shown in Fig.1.The charging energy of the qubits E C =e 2/2C 1,2is ∼100times smaller than the Josephson energy E J =¯h I 0/2e ,where I 0is the qubit Josephson critical current.The junctions discussed in [3]are 200nm by 400nm,and E J ≈200GHz.1FIG. asbyε0≈is≃The theH0=−ε0/2t(q1,q2)t∗(q1,q2)ε0/2,(1)where t(q1,q2)is a periodic function of gate charges q1,2.In the tight binding approximation[3],t(q1,q2)=t1+t2e−iπq1/e+t2e iπq2/e,where t1is the amplitude of tunneling between the nearest energy minima and t2is the tunneling between the next nearest neighbor minima in the model[3].Both t1and t2depend on the energy barrier height and width exponentially.With the parameters of our qubit design,t2/t1<10−3,the effect offluctuations of q1,2should be small.Below we consider a number of decoherence effects which seem to be most rele-vant for the design[3],trying to keep the approach general enough,so that it can be applied to other designs.2II.BASIC APPROACHWe start with a Hamiltonian of a qubit coupled to environmental degrees of freedom in the solid:H total=H Q( σ)+H bath({ξα}),where H Q=H0+H coupling:¯hH Q=∆· σby going to the frame rotating around the z−axis with the Larmor 2frequency∆=| ∆|.In the rotating frame the Hamiltonian(2)becomes:3H Q=¯hη (−ω)η (ω) (6)2πω2|φ⊥(t)|2 = dω|1−e iωt|2In thermal equilibrium,by virtue of the Fluctuation–Dissipation theorem,the noise spectrum in the RHS of (6)and (7)can be expressed in terms of the out-of-phase part of an appropriate susceptibility.III.ESTIMATES FOR PARTICULAR MECHANISMSHere we discuss the above listed decoherence mechanisms and use the expressions(6)and (7)to estimate the corresponding decoherence times.We start with the effect of charge fluctuations on the gates due to electromagnetic coupling to the environment modeled by an external impedance Z ω(see Fig.1),taken below to be of order of 400Ω,the vacuum impedance.The dependence of the qubit Hamiltonian on the gate charges q 1,2is given by (1),where q 1,2vary in time in response to the fluctuations of gate voltages,δq 1,2≈C g δV g (1,2),where the gate capacitance is much smaller than the junction capacitance:C g ≪C 1,2.The gate voltage fluctuations are given by the Nyquist formula: δV g (−ω)δV g (ω) =2Z ω¯h ωcoth ¯h ω/kT .In our design,|t (q 1,q 2)|≪ε0,and therefore fluctuations of q 1,2generate primar-ily transverse noise η⊥in (3),η⊥(t )≃(2π/¯h e )t 2C g δV g (t ).In this case,according to(7),we are interested in the noise spectrum of δV g shifted by the Larmor frequency ∆.Our typical ∆≃10GHz is much larger than the temperature k B T/h =1GHz at T =50mK,and thus one has ω≃∆≫kT/¯h in the Nyquist formula.The Nyquist spectrum is very broad compared to Larmor frequency and other relevant frequency scales,and thus in (7)we can just use the ω=∆value of the noise power.Evaluating |(1−e iωt )/ω|2dω=2πt ,we obtainR (t )= |φ⊥(t )|2 =2te t 2C g 2∆Z ω=∆(8)Rewriting this expression as R (t )=t/τ,we estimate the decoherence time asτ=∆−1¯h 2πC g t 2 2(9)where ¯h /2e 2≃4kΩ.In the qubit design e 2/2C g ≃100GHz,and t 2≃1MHz when t 2/t 1≤10−3.With these numbers,one has τ=0.1s.The next effect we consider is dephasing due to quasiparticles on supercon-ducting islands .At finite temperature,quasiparticles are thermally activated above the superconducting gap ∆0,and their density is ∼exp(−∆0/kT ).The contribution of quasiparticles to the Josephson junction dynamics can be modeled as a shunt resistor,as shown in Fig.1.The corresponding subgap resistance is inversely proportional to the quasiparticle density,and thus increases exponen-tially at small temperatures:R qp ≈R n exp ∆0/kT ,where R n is the normal state5resistance of the junction.For Josephson current I 0=0.2µA,R n ≈1.3kΩ.At lowtemperaturesthe subgap resistance is quite high,and thus difficult to measure[5].For estimates below we take R qp =1011Ωwhich is much smaller than what follows from the exponential dependence for T =50mK.The main effect of the subgap resistance in the shunt resistor model is generat-ing normal current fluctuations which couple to the phase on the junction.The Hamiltonian describing this effect isH qp coupling =i¯h 2¯h ∆ ε02kT (12)Taking R qp =1011Ω,T =50mK,and ε0/t 1=100,the decoherence times are τ =1ms and τ⊥=10ms.The decoherence effect of nuclear spins on the qubit is due to their magnetic field flux coupling to the qubit inductance.Alternatively,this coupling can be viewed as Zeeman energy of nuclear spins in the magnetic field B(r )due to the qubit.The two states of the qubit have opposite currents,and produce magnetic field of opposite sign.The corresponding term in (2)isH coupling =−σzr =r iµ B (r )· s (r )(13)where r i are positions of nuclei,µis nuclear magnetic moment and s (r i )are spin operators.Nuclei are in thermal equilibrium,and their spin fluctuations can be related to the longitudinal relaxation time T 1by the Fluctuation-Dissipation theorem.Assuming that different spins are uncorrelated,one hass ω(r )s −ω(r ) =2k B T χ′′(ω)1+ω2T 21,(14)6whereχ0=1/k B T is static spin susceptibility.The spectrum(14)has a very narrow width set by the long relaxation time T1.This width is much less then k B T and∆.As a result,only longitudinal fluctuationsη survive in(6)and(7).One hasφ2 (t) = dω|1−e iωt|2τ20 |t|−T1+T1e−|t|/T1 ,(16)τ0= 2µ2A similar theory can be employed to estimate the effect due to magnetic im-purities.The main difference is that for impurity spins the relaxation time T1is typically much shorter than for nuclear spins.If T1becomes comparable to the qubit operation time,the ensemble averaged quantities will describe a real dephas-ing of an individual qubit,rather than effects of inhomogeneous broadening,like for nuclear spins.IV.OTHER MECHANISMSSome sources of decoherence are not amenable to the basic approach considered above,such as radiation losses which we estimate to haveτ≃103s.Another such source of decoherence is caused by the magnetic dipole interaction between the qubits.This interaction between qubits is described byH coupling= i,j¯hλijσ(i)z⊗σ(j)z,¯hλij≈µiµjcan be made at least1ms which for f Rabi=100MHz gives a quality factor of105, passing the criterion for quantum error correction.In addition to the effects we discussed,some other decoherence sources are worth attention,such as low frequency chargefluctuations resulting from electron hopping on impurities in the semiconductor and charge configuration switching near the gates[8].These effects cause1/f noise in electron transport,and may contribute to decoherence at low frequencies.Also,we left out the effect of the acfield coupling the two low energy states of the qubit to higher energy states. Results of our numerical simulations of the coupling matrix in the qubit[3]show that Rabi oscillations can be observed even in the presence of the ac excitation mixing the states(to be published elsewhere).ACKNOWLEDGMENTSThis work is supported by ARO grant DAAG55-98-1-0369,NSF Award 67436000IRG,Stichting voor Fundamenteel Onderzoek der Materie and the New Energy and Industrial Technology Development Organization.[8]T.Henning et al.,Eur.Phys.J.B8.627(1999);V.A.Krupenin et al.,J.Appl.Phys.84,3212(1998);N.Zimmerman et al.,Phys.Rev.B56,7675(1997);E.H.Visscher et al,Appl.Phys.Lett.66,305(1995).10。
降噪的作文例子英语
降噪的作文例子英语Title: Finding Serenity in the Silence: The Art of Noise Reduction。
In today's bustling world, the incessant cacophony of urban life often threatens to drown out the tranquility we all crave. From the clamor of traffic to the ceaseless hum of electronic devices, the relentless noise can overwhelm our senses and leave us yearning for moments of peace. Fortunately, amidst the chaos, there exists a remedy: noise reduction.Noise reduction techniques encompass a variety of methods aimed at minimizing or eliminating unwanted sounds from our environment. Whether through technological advancements or simple lifestyle adjustments, the goal remains the same: to create spaces where serenity can flourish.One effective strategy for noise reduction is the useof soundproofing materials. By installing insulation, acoustic panels, or double-paned windows, individuals can significantly reduce the amount of external noise that penetrates their living or working spaces. This not only fosters a quieter environment but also enhances concentration and productivity, as distractions are minimized.Furthermore, embracing the concept of mindful listening can serve as a powerful tool in the quest for tranquility. Instead of allowing ourselves to be engulfed by the noise around us, we can learn to selectively tune in to the sounds that bring us joy and solace. Whether it's the gentle rustle of leaves in the wind or the soothing melody of a favorite song, mindful listening enables us to find pockets of peace amidst the clamor.In addition to physical measures, cultivating a mindset of simplicity can also contribute to noise reduction. In a society inundated with constant stimulation and excess, embracing minimalism can be liberating. By decluttering our physical spaces and streamlining our possessions, we createroom for clarity and calmness to emerge. In essence, less becomes more, and the unnecessary noise of materialism fades into the background.Beyond the realm of tangible solutions, the practice of mindfulness meditation offers a profound means of quieting the mind amidst external chaos. Through focused breathing and present moment awareness, individuals can cultivate inner stillness even in the midst of external tumult. By nurturing this inner sanctuary, we become less reactive to external stimuli, finding refuge in the depths of our own consciousness.Moreover, immersing oneself in nature provides arespite from the din of modern life. Whether it's a leisurely stroll through a tranquil forest or the mesmerizing rhythm of ocean waves, the natural world offers a sanctuary of silence. In these moments of communion with the earth, we rediscover our innate connection to the universe, finding solace in the whispers of the wind and the gentle lullaby of the earth's embrace.In conclusion, the pursuit of noise reduction is not merely about creating silence but about cultivating a deeper sense of harmony within ourselves and with the world around us. By integrating soundproofing techniques, mindful listening practices, minimalist principles, meditation, and immersion in nature, we can forge a path towards greater tranquility and inner peace. In the symphony of life, let us strive to find our own melody, one that resonates with serenity and stillness amidst the noise.。
学位 作文 噪音 英语
学位作文噪音英语Here is an essay on the topic of "Degree, Essay, Noise, English" with over 1000 words, written in English without any additional punctuation marks in the body:Pursuing a degree in the modern world is a multifaceted endeavor fraught with various challenges one must navigate with diligence and resilience The primary hurdle many face is the cacophony of noise that permeates our daily lives Amidst the constant barrage of digital stimuli and societal pressures it can be immensely difficult to maintain focus and cultivate the discipline required to succeed in academic pursuits Yet it is precisely in the face of such distractions that one must hone the skills necessary to produce a cohesive and compelling essay that can serve as the capstone of their educational journeyFor many students the act of crafting a substantial piece of academic writing presents a daunting prospect The sheer volume of research required the necessity of formulating a cogent thesis and the arduous task of seamlessly weaving disparate ideas into a unified narrative can seem overwhelming However it is through this process that one truly begins to develop as a critical thinker and effectivecommunicator skills that are not only invaluable within the confines of the classroom but also essential for navigating the complexities of the professional worldPerhaps the most significant obstacle in composing a successful essay lies in the ability to tune out the relentless noise that bombards us from all sides In an era defined by smartphone notifications constant social media engagement and the myriad other distractions vying for our attention it can be exceedingly challenging to carve out the mental space necessary for deep contemplation and sustained intellectual effort Yet it is precisely this capacity for focused attention that separates those who produce work of genuine substance from those who merely skim the surfaceCultivating such focus in the face of ubiquitous noise requires a multifaceted approach One must first recognize the debilitating impact that constant digital stimulation can have on cognitive function and make a concerted effort to limit unnecessary exposure to such sources of distraction This may involve setting aside dedicated blocks of time for uninterrupted work scheduling regular breaks to refresh and recharge or even implementing digital detox protocols to facilitate more meaningful engagement with the task at handIn addition to managing external sources of noise it is equallyimportant to address the internal chatter that can so often impede our ability to think clearly and write effectively The anxieties self-doubts and feelings of inadequacy that so many students grapple with can create a maelstrom of mental noise that makes it exceedingly difficult to maintain the focus and clarity required to produce a compelling essay It is therefore essential to develop strategies for quieting this inner turmoil through practices such as meditation mindfulness and self-compassionOnce one has established the necessary conditions for focused work the process of crafting a substantive essay can truly begin This entails meticulously researching the topic at hand carefully considering multiple perspectives and viewpoints and then synthesizing this wealth of information into a cohesive and persuasive argument Effective essay writing is not merely a matter of regurgitating facts but rather a dynamic interplay between research analysis and original insightThroughout this process the student must remain steadfast in their commitment to producing work of genuine substance resisting the temptation to succumb to the siren song of superficial quick fixes or formulaic templates Instead they must be willing to grapple with complexity delve into nuance and wrestle with ambiguity all while maintaining a keen sense of purpose and direction This requires not only a mastery of the technical aspects of essay composition but alsoa deep well of intellectual curiosity and a willingness to continually challenge one s own assumptions and biasesUltimately the ability to write a compelling essay in the face of ubiquitous noise is a skill that extends far beyond the confines of academia It is a testament to one s capacity for critical thinking effective communication and intellectual resilience qualities that are increasingly valuable in a world that is rapidly evolving and ever more complex For those who are able to develop these skills the rewards can be immense not only in the form of academic accolades but also in the personal growth and sense of mastery that comes from producing work of genuine substance。
Noise in an SSET-resonator driven by an external field
Over the past few years experiments on superconducting circuits have produced some rather impressive results. Superconducting elements have been used to produce two-level systems of various kinds which can be considered as artificial atoms [1, 2, 3, 4], and superconducting stripline resonators can act analogously to optical cavities for microwave fields [5]. This field of study has been referred to as circuit QED in analogy with cavity QED. One advantage of superconducting circuits is that rather than simply investigating the behavior of the system through the field emitted or reflected by the cavity, other elements can provide additional information. Mesoscopic conductors coupled to the resonator can provide additional information and at the same time the back action may lead to some interesting and subtle dynamics. Such back-action dynamics have also received considerable attention in the context of a mesoscopic conductor used to investigate the behavior of a mechanical resonator [6, 7, 8, 9, 10, 11, 12, 13]. One system that is of particular interest is that of a superconducting single electron transistor (SSET) coupled to a resonator, either mechanical [14, 15, 16, 17, 18, 19, 20] or composed of a superconducting stripline [23]. The coherent transport through this device at the Josephson quasiparticle resonance [24, 25] (JQP) allows a very lownoise current, and at the same time the sensitivity of the SSET to charge means that the resonator-SSET coupling is significant. The SSET biased at the JQP resonance can be considered analogous to a three-level atom, and the coupled system therefore shows behavior related to that of a micromaser [21, 22]. The low noise and relatively strong coupling in this device allows the observation of the non-trivial coupled dynamics that arises from the interaction between the resonator and SSET. When the SSET is biased so as to absorb energy from the resonator, the SSET acts as an
GuidelinesforCommunityNoise-NationalWindWatch
Health Effects of Wind Turbine NoiseNina Pierpont, MD, PhD()February 4, 2006Industrial wind turbines produce significant amounts of audible and low-frequency noise. Dr. Oguz A. Soysal, Professor and Chairman of the Dept. of Physics and Engineering at Frostburg State University in Maryland, measured sound levels over half a mile away from the Meyersdale, PA, 20-turbine wind farm. Typical audible (A-weighted) dB (decibel) levels were in the 50-60 range, and audible plus low-frequency (C-weighted) dB were in the 65-70 range.1 65-70 dB is the loudness of a washing machine, vacuum cleaner, or hair dryer.2 A difference of 10 dB between A and C weighting represents a significant amount of low-frequency sound by World Health Organization standards.3The noise produced by wind turbines has a thumping, pulsing character, especially at night, when it is more audible. The noise is louder at night because of the contrast between the still, cool air at ground level and the steady stream of wind at the level of the turbine hubs.4 This nighttime noise travels a long distance. It has been documented to be disturbing to residents 1.2 miles away from wind turbines in regular rolling terrain,5 and 1.5 miles away in Appalachian valleys.6At night, the WHO recommends, the level of continuous noise at the outside a dwelling should be 45 dB or less, and inside, 30 dB or less. These thresholds should be even lower if there is a significant low-frequency component to the sound, they add – as there is for wind turbines. Higher levels of noise disturb sleep and produce a host of effects on health, well-being, and productivity.7The decibel is logarithmic. Increasing the dB level by 10 multiplies the noise level by 10. Increasing the dB level by 20 multiplies the noise level by 100 (and 30 dB multiplies by 1000, etc.). Thus the 65 dB measured day and night half a mile from the Meyersdale wind farm is 100 times louder than the loudest continuous outdoor nighttime noise (45 dB) recommended by the WHO.Typical ordinances proposed or passed for NY State communities considering industrial wind turbines allow A-weighted noise levels of 50 dB and construction of turbines only 1000 ft. from dwellings. These ordinances meet neither WHO nor NYS DEC standards, especially compared to the very low ambient noise levels (with dB levels typically in the 20’s) in rural NY.8The health effects of excessive community noise are carefully documented in the WHO report with reference to scientific and medical literature. Effects relevant to wind turbines, in terms of dB levels and noise type, are paraphrased and summarized from this report:For people to understand each other easily when talking, environmental noise levels sh ould be 35 dB or less. For vulnerable groups (hearing impaired, elderly, children in the process of reading and language acquisition, and foreign language speakers) even lower background levels are needed. When noiseinterferes with speech comprehension, problems with concentration, fatigue, uncertainty and lack of1 Soysal, OA. 2005. Acoustic Noise Generated by Wind Turbines. Presented to the Lycoming County, PA Zoning Board 12/14/************************2 /noise/decibel.htm3 World Health Organization, 1999. Guidelines for Community Noise. Ed. by Berglund B et al. Available atwww.who.int/docstore/peh/noise/guidelines24van den Berg, FGP, 2005. “The beat is getting stronger: The effect of atmosphe ric stability on low frequency modulated sound of wind turbines.” Journal of Low Frequency Noise, Vibration, and Active Control, 24(1):1-24.5van den Berg, FGP, 2003. “Effects of the wind profile at night on wind turbine sound.” Journal of Sound and Vibration 277:955-970.6Linda Cooper, Citizens for Responsible Windpower, “Activist Shares Wind Power Concerns,” The Pendleton Times, March 3, 2005, p. 4.7 WHO, 1999. Guidelines for Community Noise.8 NYS DEC, 2001. Assessing and Mitigating Noise Impacts.self-confidence, irritation, misunderstandings, decreased work capacity, problems in human relations,and a number of stress reactions arise.9∙Wind turbine noise, as described above and experienced by many turbine neighbors, is easily within the decibel levels to disturb sleep. Effects of noise-induced sleep disturbance include fatigue, depressedmood or well-being, decreased performance, and increased use of sedatives or sleeping pills. Meas ured physiologic effects of noise during sleep are increased blood pressure and heart rate, changes inbreathing pattern, and cardiac arrhythmias.10 Certain types of nighttime noise are especiallybothersome, the authors note, including those which combine noise with vibration, those with low-frequency components, and sources in environments with low background noise.11 All three of thesespecial considerations apply to industrial wind turbines in rural NY State. Children, the elderly, andpeople with preexisting illnesses, especially depression, are especially vulnerable to sleep disturbance.∙Noise has an adverse effect on performance over and above its effects on speech comprehension. The most strongly affected cognitive areas are reading, attention, problem solving, and memory. Childrenin school are adversely affected by noise, and it is the uncontrollability of noise, rather than its intensity, which is most critical. The effort to tune out the noise comes at the price of increased levels of stresshormones and elevation of resting blood pressure. The adverse effects are larger in children with lower school achievement.12∙What is commonly referred to as noise “annoyance” is in fa ct a range of negative emotions, documented in people exposed to community noise, including anger, disappointment, dissatisfaction,withdrawal, helplessness, depression, anxiety, distraction, agitation, and exhaustion.13 Numerousreports from neighbors of new industrial wind turbine installations document these symptoms. Thepercentage of highly annoyed people in a population starts to increase at 42 dB, and the percentage ofmoderately annoyed at 37 dB.14Low-frequency sound is also sensed as pressure in the ears. It modulates the loudness of regular audible frequencies, and is sensed as a feeling or vibration in the chest and throat.15 Neighbors of industrial wind turbines describe the distressing sensation of having to breathe in sync with the rhythmic thumps of the turbine blades, especially at night when trying to sleep.The participants in noise studies are selected from the general population and are usually adults. Vulnerable groups of people are underrepresented. Vulnerable groups include:∙People with decreased personal abilities (old, ill, or depressed people)∙People with particular diseases or medical problems∙People (such as children) dealing with complex cognitive tasks, such as reading acquisition∙People who are blind or who have hearing impairment∙Fetuses, babies and young children∙The elderlyThese people may be less able to cope with the impacts of noise exposure and at greater risk for harmful effects than is documented in studies. Attention needs to be paid to them when developing regulations and setback requirements for industrial wind turbines and other sources of annoying and debilitating noise.Wind turbines also create moving visual disturbances, especially early and late in the day when the long shadows of moving blades sweep rhythmically over the landscape. That portion of the population which is susceptible to vertigo, unsteadiness, or motion sickness (including many children and a large proportion of the elderly) will be vulnerable to unsteadiness and nausea when subjected to this visual disturbance. People with seizure disorders are susceptible to triggering of seizures by the strobe effect of seeing the sun through the moving blades.9 WHO, 1999. Guidelines for Community Noise, pp. 42-44.10 Ibid, p. 44.11 Ibid. p. 4612 Ibid. pp. 49-5013 Ibid. p. 5014 Ibid. p. 5115 Moller, H. and CS Pedersen. 2004. Hearing at low and infrasonic frequencies. Noise & Health 6 (23):37-57.。
- 1、下载文档前请自行甄别文档内容的完整性,平台不提供额外的编辑、内容补充、找答案等附加服务。
- 2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
- 3、如文档侵犯您的权益,请联系客服反馈,我们会尽快为您处理(人工客服工作时间:9:00-18:30)。
With recent advances in low power wireless sensing technologies, the concept of BSN has shown significant strength in continuous monitoring of patients under their natural physiological status [1]. Due to the diversity of the environment and physiological conditions they may experience, understanding of the context within which the signals are collected plays an important role for the accurate prediction of adverse events. Reliable detection of patient activity, however, requires the use of a large number of context sensors around the body. This can potentially introduce a significant burden to power consumption and bandwidth utilisation. Directly sending all the sensory data to the centralised processing unit requires extensive transmission power and a high bandwidth at the central processing unit. Decreasing the transmission range and required bandwidth will greatly reduce the power consumption and prolong the life span of the sensors. Clustering data transmission among neighbourhood can also alleviate the problem of data collision.
5) Arc list
Current network
The arc A-C is not
added since B-C are A B C D E
barely correlated
6) Adding hidden nodes
R
H
H
A BCDE
Fig. 1. A diagrammatic overview of the construction of a BN with hidden nodes with complete-linkage clustering, where only the top arc is considered in each step.
D
D – decision (root) node
H – Hidden node
S1-S8 – Sensor nodes
H
H
S1
S2
S3
S4
S5
S6
S7
S8
Hidden nodes indicate redundant sensors Fig. 2. The structure of the BN with hidden nodes learned from the training dataset.
CurC,D)
(A,C)
Current network A B CDE
4) Arc list (A,C)
Current network
The arc C-D is not
added since C-E are A B C D E
not correlated
1) Arc list (A,B)
(D,E) (C,D) (A,C) (A,E) (B,D) (B,E) (A,D) (B,C) (C,E)
Current network A B CDE
Filter out arcs with low correlations
2) Arc list (D,E)
(C,D) (A,C)
To assess the overall performance of the proposed method, an ETH reference dataset was used. This consisted of sensor data obtained from accelerometers in performing eight different activities [2]. After extracting the temporal features from the dataset, reference data is constructed by selecting six representative features and two highly correlated features. From the reference data set, a BN is obtained by learning the structure from the training dataset, and hidden nodes were inserted to represent the dependency among correlated children [4], as shown in Fig.2.
13
2. Model Description
The main idea of introducing noise resilience to the sensing architecture is the introduction of hidden nodes to a traditional BN to resolve the violation of the independence assumption when there is a considerable amount of dependency among child nodes. This allows the transformation of a multiply connected network into a singly connected network. Complete-link clustering based on a pairwise-dependency measure, such as Pearson’s correlation coefficient or L1 metric, is used to form clusters of correlated child nodes as shown in Fig. 1. In practice, a close-form representation for deriving link matrices is difficult for a network with hidden nodes. This is because the hidden nodes are unobserved variables and the node statistics cannot be directly calculated. We used instead a backward propagation method, proposed by Kwoh et. al, is used for parameter learning [3].
* Department of Computing, Imperial College London, South Kensington Campus 180 Queen’s Gate, London, SW7 2AZ, United Kingdom. {st01, rck, benlo, dfg, gzy}@ /
A NOISE RESILIENT DISTRIBUTED INFERENCE FRAMEWORK FOR BODY SENSOR NETWORKS
Surapa Thiemjarus, Rachel King, Benny Lo, Duncan Gilles, Guang-Zhong Yang*
Abstract
Existing research has shown that inferencing with message passing is potentially useful for distributed sensing systems. However, resilience to communication error and node failures is major obstacle to overcome. Paskin et. al proposed a robust message passing algorithm for reasoning in a junction tree model [5]. By converting a standard multiply connected Bayesian Network (BN) [6] into a cluster tree and combining nodes into a clique, the problem of non-convergence and incorrect update of the posterior probabilities due to the loopy feedback in a multiply connected model can be avoided. Furthermore, existing research in BSN has also highlighted need for built-in redundancies in the sensor network for dealing with motion artefact and node failures. The purpose of this paper is to propose a Bayesian framework that permits distributed inferencing with a high level of noise tolerance.