A fractal signal x(t) in biomedical engineering may be characterized by 1/f noise, that is, the power spectrum density (PSD) divergences at f=0. According the Taqqu’s law, 1/f noise has the properties of long-range dependence and heavy-tailed probability density function (PDF). The contribution of this paper is to exhibit that the prediction error of a biomedical signal of 1/f noise type is long-range dependent (LRD). Thus, it is heavy-tailed and of 1/f noise. Consequently, the variance of the prediction error is usually large or may not exist, making predicting biomedical signals of 1/f noise type difficult.
1. Introduction
Signals of 1/f noise type are widely observed in biomedical engineering, ranging from heart rate to DNA and protein, see, for example, [1–37], just to cite a few. Predicting such a type of signals is desired in the field [38–43]. A fundamental issue in this regard is whether a biomedical signal of 1/f noise type to be predicted is predicable or not.
The predictability of signals of non-1/f noise type is well studied [44–48]. However, the predictability of 1/f noise is rarely reported, to our best knowledge. Since many phenomena in biomedical engineering are characterized by 1/f noise [1–37], the predictability issue of 1/f noise is worth investigating.
Note that minimizing the mean square error (MSE) of prediction is a commonly used criterion in both theory and practical techniques of prediction, see, for example, [49–68]. Therefore, a sufficient condition for a biomedical signal x(t) to be predictable is that the variance of its predication error exists. If the variance of the predication error does not exist, on the contrary, it may be difficult to be predicted if not unpredictable. In the case of a signal being bandlimited, the variance of its predication error is generally finite. Consequently, it may be minimized and it is predictable. However, that is not always the case for biomedical signals of 1/f noise type.
Let x(t) be a biomedical signal in the class of 1/f noise. Then, its PDF is heavy-tailed, and it is LRD, see, for example, Adler et al. [69], Samorodnitsky and Taqqu [70], Mandelbrot [71], Li and Zhao [72]. Due to that, here and below, the terms, 1/f noise, LRD random function, and heavy-tailed random function are interchangeable.
Let p(x) be the PDF of a biomedical signal x(t) of 1/f noise type. Then, its variance is expressed by
(1)Var[x(t)]=∫-∞∞(x-μx)2p(x)dx,
where μx is the mean of x(t) if it exists. The term of heavy tail in statistics implies that Var[x(t)] is large. Theoretically speaking, in general, we cannot assume that Var[x(t)] always exists [72]. In some cases, such as the Pareto distribution, the Cauchy distribution, α-stable distributions [72], Var[x(t)] may be infinite. That Var[x(t)] does not exist is particularly true for signals in biomedical engineering and physiology, see Bassingthwaighte et al. [33] for the interpretation of this point of view.
Recall that a prediction error is a random function as we shall soon mention below. Therefore, whether the prediction error is of 1/f noise, or equivalently, heavy-tailed, turns to be a crucial issue we need studying. We aim at, in this research, exhibiting that prediction error of 1/f noise is heavy-tailed and accordingly is of 1/f noise. Thus, generally speaking, the variance of a prediction error of a biomedical signal x(t) of 1/f noise type may not exist or large. That is a reason why predicting biomedical signals of 1/f noise type is difficult.
The rest of this paper is organized as follows. Heavy-tailed prediction errors occurring in the prediction of biomedical signals of 1/f noise type are explained in Section 2. Discussions are in Section 3, which is followed by conclusions.
2. Prediction Errors of 1/f Noise Type
We use x(n) to represent a biomedical signal in the discrete case for n∈N, where N is the set of natural numbers. Let xN(n) be a given sample of x(n) for n=0,1,…,N-1. Denote by xM(m) the predicted values of x(n) for m=N,N+1,N+M-1. Then, the prediction error denoted by e(m) is given by
(2)e(m)=∑m=NN+M-1x(m)-xM(m).
If one uses the given sample of x(n) for n=N,N+1,…,2N-1 to obtain the predictions denoted by xM(m) for m=2N,2N+1,2N+M-1, the error is usually different from (2), which implies that the error e(m) is a random variable. Denote by p(e) the PDF of e(m). Then, its variance is expressed by
(3)Var[e(m)]=∑m=NN+M-1(e-μe)2p(e),
where μe is the mean of e(m).
Let P be the operator of a predictor. Then,
(4)xM(m)=PxN(n).
A natural requirement in terms of P is that Var[e(m)] should be minimized. Thus, the premise that Var[e(m)] can be minimized is that it exists.
It is obviously seen that Var[e(m)] may be large if p(e) is heavy tailed. In a certain cases, Var[e(m)] may not exist. To explain the latter, we assume that e(m) follows a type of heavy-tailed distribution called the Pareto distribution.
Denote by pPareto(e) the PDF of the Pareto distribution. Then [73], it is in the form
(5)pPareto(e)=abaea+1,
where e≥b, a>0, and b>0. The mean and variance of e(m) are, respectively, expressed by
(6)μe=aba-1,Var(e)=ab2(a-1)2(a-2).
The above exhibits that Var[e(m)] does not exist if a=1 or a=2 and if e(m) follows the Pareto distribution.
Note that the situation that Var[e(m)] does not exist may not occur if e(m) is light-tailed. Therefore, the question in this regard is whether e(m) is heavy-tailed if a biomedical signal x(n) is of 1/f noise. The answer to that question is affirmative. We explain it below.
Theorem 1.
Let x(n) be a biomedical signal of 1/f noise type to be predicted. Then, its prediction error is heavy-tailed. Consequently, it is of 1/f noise.
Proof.
Let rxx(k) be the autocorrelation function (ACF) of x(n). Then,
(7)rxx(k)=E[x(n)x(n+k)],
where k is lag and E the mean operator. Let rMM(k) be the ACF of xM(m). Then,
(8)rMM(k)=E[xM(m)xM(m+k)].
Let ree(k) be the ACF of e(m). Then,
(9)ree(k)=E[e(m)e(m+k)].
Note that
(10)ree(k)=E[e(m)e(m+k)]=E{[x(m)-xM(m)][x(m+k)-xM(m+k)]}=E[x(m)x(m+k)+xM(m)xM(m+k)E-xM(m)x(m+k)-x(m)xM(m+k)]=rxx(k)+rMM(k)-rMx(k)-rxM(k).
In the above expression, rMx(k) is the cross-correlation between xM(m) and x(m). On the other side, rxM(k) is the cross-correlation between x(m) and xM(m). Since rMx(k)=rxM(k), we have
(11)ree(k)=rxx(k)+rMM(k)-2rxM(k).
Recall that x(m) is 1/f noise. Thus, it is heavy-tailed and hence LRD. Consequently, for a constant c1>0, we have
(12)rxx(k)~c1k-α(k→∞)for0<α<1.
On the other hand, the predicted series xM(m) is LRD. Thus, for a constant c2>0, the following holds:
(13)rMM(k)~c2k-β(k→∞)for0<β<1.
In (11), if rxM(k) is summable, that is, it decays faster than rx(k) or rM(k), it may be ignored for k→∞. In this case, ree(k) is still non-summable. In fact, one has
(14)ree(k)~{c1k-α,0<α<β<1,c2k-β,0<β<α<1,(c1+c2)k-β,α=β.(k→∞),
On the other side, when rxM(k) is non-summable, re(k) is non-summable too. In any case, we may write ree(k) by
(15)ree(k)~ck-γ(k→∞)for0<γ<1.
Therefore, the prediction error e(m) is LRD. Its PDF p(e) is heavy-tailed according to the Taqqu’s law. Following [72], therefore, e(m) is a 1/f noise. This completes the proof.
3. Discussions
The present result implies that cautions are needed for dealing with predication errors of biomedical signals of 1/f noise type. In fact, if specific biomedical signals are in the class of 1/f noise, the variances of their prediction errors may not exist or large [72]. Tucker and Garway-Heath used to state that their prediction errors with either prediction model they used are large [74]. The result in this paper may in a way provide their research with an explanation.
Due to the fact that a biomedical signal may be of 1/f noise, PDF estimation is suggested as a preparatory stage for prediction. As a matter of fact, if a PDF estimation of biomedical signal is light-tailed, its variance of prediction error exists. On the contrary, the variance of the prediction error may not exist. In the latter case, special techniques have to be considered [75–78]. For instance, weighting prediction error may be a technique necessarily to be taken into account, which is suggested in the domain of generalized functions over the Schwartz distributions [79].
4. Conclusions
We have explained that the prediction error e(m) in predicting biomedical signals of 1/f noise type is usually LRD. This implies that its PDF p(e) is heavy-tailed and 1/f noise. Consequently, Var[e(m)] may in general be large. In some cases [72], Var[e(m)] may not exist, making the prediction of biomedical signals of 1/f noise type difficult with the way of minimizing Var[e(m)].
Acknowledgments
This work was supported in part by the 973 plan under the Project Grant no. 2011CB302800 and by the National Natural Science Foundation of China under the Project Grant no. 61272402, 61070214, and 60873264.
AoyagiN.StruzikZ. R.KiyonoK.YamamotoY.Autonomic imbalance induced breakdown of long-range dependence in healthy heart rate20074621741782-s2.0-33947527644TongS.JiangD.WangZ.ZhuY.GeocadinR. G.ThakorN. V.Long range correlations in the heart rate variability following the injury of cardiac arrest20073801-22502582-s2.0-3424785105210.1016/j.physa.2007.02.078SarlisN. V.SkordasE. S.VarotsosP. A.Heart rate variability in natural time and 1/f ‘noise’20098712-s2.0-7905146899610.1209/0295-5075/87/1800318003StruzikZ. R.HayanoJ.SomaR.KwakS.YamamotoY.Aging of complex heart rate dynamics200653189942-s2.0-3044445223910.1109/TBME.2005.859801AcharyaU. R.JosephK. P.KannathalN.LimC. M.SuriJ. S.Heart rate variability: a review20064412103110512-s2.0-3444765027510.1007/s11517-006-0119-0BatesJ. H. T.MaksymG. N.NavajasD.SukiB.Lung tissue rheology and 1/f noise19942266746812-s2.0-002853342210.1007/BF02368292HalleyJ. M.KuninW. E.Extinction risk and the 1/f family of noise models19995632152302-s2.0-003337538910.1006/tpbi.1999.1424WichmannM. C.JohstK.SchwagerM.BlasiusB.JeltschF.Extinction risk, coloured noise and the scaling of variance200568129402-s2.0-2214443903710.1016/j.tpb.2005.03.001YangZ.HoangL.ZhaoQ.KeeferE.LiuW.1/f neural noise reduction and spike feature extraction using a subset of informative samples2011394126412772-s2.0-7995442647510.1007/s10439-010-0201-5RuseckasJ.KaulakysB.Tsallis distributions and 1/f noise from nonlinear stochastic differential equations20118457051125BeckersF.VerheydenB.AubertA. E.Aging and nonlinear heart rate control in a healthy population20062906H2560H25702-s2.0-3374493500810.1152/ajpheart.00903.2005PilgramB.KaplanD. T.Nonstationarity and 1/f noise characteristics in heart rate19992761R1R92-s2.0-0032916303SzendroP.VinczeG.SzaszA.Pink-noise behaviour of biosystems20013032272312-s2.0-003488105010.1007/s002490100143MassieraG.Van CittersK. M.BiancanielloP. L.CrockerJ. C.Mechanics of single cells: rheology, time dependence, and fluctuations20079310370337132-s2.0-3654900829110.1529/biophysj.107.111641MuraseY.ShimadaT.ItoN.RikvoldP. A.Effects of demographic stochasticity on biological community assembly on evolutionary time scales2010814142-s2.0-7795112240910.1103/PhysRevE.81.041908041908YokogawaT.HaradaT.Generality of a power-law long-term correlation in beat timings of single cardiac cells2009387119242-s2.0-6765104854510.1016/j.bbrc.2009.05.143HaradaT.YokogawaT.MiyaguchiT.KoriH.Singular behavior of slow dynamics of single excitable cells20099612552672-s2.0-5884911549010.1529/biophysj.108.139691EkeA.HermánP.BassingthwaighteJ. B.RaymondG. M.PercivalD. B.CannonM.BallaI.IkrényiC.Physiological time series: distinguishing fractal noises from motions200043944034152-s2.0-003395887810.1007/s004240050957WestB. J.Fractal physiology and the fractional calculus: a perspective20101, article 1210.3389/fphys.2010.00012GrigoliniP.AquinoG.BolognaM.LukovićM.WestB. J.A theory of 1/f noise in human cognition200938819419242042-s2.0-6765011772010.1016/j.physa.2009.06.024GrüneisF.NakaoM.MizutaniY.YamamotoM.MeesmannM.MushaT.Further study on 1/f fluctuations observed in central single neurons during REM sleep19936831931982-s2.0-002735470110.1007/BF00224851ShengH.ChenY.-Q.QiuT.-S.Heavy-tailed distribution and local long memory in time series of molecular motion on the cell membrane2011101931192-s2.0-7955158812410.1142/S0219477511000429WestB. J.DeeringW.Fractal physiology for physicists: Lévy statistics19942461-211002-s2.0-2942574810DeeringW.WestB. J.Fractal physiology199211240462-s2.0-002688393910.1109/51.139035WestB. J.Physiology in fractal dimensions: error tolerance19901821351492-s2.0-002508051610.1007/BF02368426JoyeuxM.BuyukdagliS.SanreyM.1/f Fluctuations of DNA temperature at thermal denaturation200775692-s2.0-3454737933110.1103/PhysRevE.75.061914061914CattaniC.Fractals and hidden symmetries in DNA20102010312-s2.0-7795452576210.1155/2010/507056507056CattaniC.LaserraE.BochicchioI.Simplicial approach to fractal structures201220122195810110.1155/2012/958101HermanP.EkeA.Nonlinear analysis of blood cell flux fluctuations in the rat brain cortex during stepwise hypotension challenge2006269118911972-s2.0-3374759095910.1038/sj.jcbfm.9600266BaumertM.BaierV.VossA.Long-term correlations and fractal dimension of beat-to-beat blood pressure dynamics200554L549L5552-s2.0-2914446011210.1142/S0219477505003002CattaniC.On the existence of wavelet symmetries in Archaea DNA201220122167393410.1155/2012/673934PonomarevS. Y.PutkaradzeV.BishopT. C.Relaxation dynamics of nucleosomal DNA2009114510633106432-s2.0-7045022539310.1039/b910937bBassingthwaighteJ. B.LiebovitchL. S.WestB. J.1994Oxford University PressCraciunD.IsvoranA.AvramN. M.Long range correlation of hydrophilicity and flexibility along the calcium binding protein chains200938821460946182-s2.0-6864909785610.1016/j.physa.2009.07.042SiódmiakJ.UherJ. J.Santamaría-HolekI.KruszewskaN.GadomskiA.On the protein crystal formation as an interface-controlled process with prototype ion-channeling effect20073343133292-s2.0-4644911136210.1007/s10867-008-9076-1KouS. C.XieX. S.Generalized langevin equation with fractional gaussian noise: subdiffusion within a single protein molecule2004931841806032-s2.0-1974438249710.1103/PhysRevLett.93.180603ShengH.ChenY.-Q.QiuT.-S.2012SpringerPanellaM.Advances in biological time series prediction by neural networks2011621121202-s2.0-7995270189510.1016/j.bspc.2010.09.006ChoY.-R.ZhangA.Predicting protein function by frequent functional association pattern mining in protein interaction networks201014130362-s2.0-7684908712310.1109/TITB.2009.2028234CastroA.MarquesM. A. L.VarsanoD.SottileF.RubioA.The challenge of predicting optical properties of biomolecules: what can we learn from time-dependent density-functional theory?20091064694902-s2.0-7034917671610.1016/j.crhy.2008.09.001LüQ.WuH. J.WuJ. Z.A parallel ant colonies approach to de novo prediction of protein backbone in CASP8/9Science China Information Sciences. In pressYangB. R.QuW.WangL. J.ZhouY.A new intelligent prediction system model-the compound pyramid model2012553723736SuoJ. L.JiX. Y.DaiQ. H.An overview of computational photography201255612291248PapoulisA.A note on the predictability of band-limited processes1985738133213332-s2.0-0022106651ChenS. Y.YaoC. Y.XiaoG.YingY. S.WangW. L.Fault detection and prediction of clocks and timers based on computer audition and probabilistic neural networks3512Proceedings of the 8th International Workshop on Artificial Neural Networks, IWANN 2005: Computational Intelligence and Bioinspired SystemsJune 2005952959Lecture Notes in Computer Science2-s2.0-25144472095LymanR. J.EdmonsonW. W.McCulloughS.RaoM.The predictability of continuous-time, bandlimited processes20004823113162-s2.0-0033893305LymanR. J.EdmonsonW. W.Linear prediction of bandlimited processes with flat spectral densities2001497156415692-s2.0-003539902710.1109/78.928709DokuchaevN.The predictability of band-limited, high-frequency and mixed processes in the presence of ideal low-pass filters2008413872-s2.0-5474914402810.1088/1751-8113/41/38/382002382002WienerN.1964John Wiley & SonsKolmogorovA. N.Interpolation and extrapolation of stationary random sequences19415314ZadehL. A.RagazziniJ. R.An extension of Wiener's theory of prediction19502176456552-s2.0-000155022110.1063/1.1699725BhansaliR. J.Asymptotic properties of the Wiener-Kolmogorov predictor. I19743616173LevinsonN.A heuristic exposition of Wiener’s mathematical theory of prediction and filtering194726110119LevinsonN.The Wiener RMS (root mean squares) error criterion in filter design and prediction194725261278BhansaliR. J.Asymptotic mean-square error of predicting more than one-step ahead using the regression method19742313542MakhoulJ.Linear prediction: a tutorial review19756345615802-s2.0-0016495091ZimmermanD. L.CressieN.Mean squared prediction error in the spatial linear model with estimated covariance parameters199244127432-s2.0-000183964610.1007/BF00048668HuangD.Levinson-type recursive algorithms for least-squares autoregression2008114295315DeoR. S.Improved forecasting of autoregressive series by weighted least squares approximate REML estimation201228139432-s2.0-7995365085710.1016/j.ijforecast.2011.02.014RodríguezA.RuizE.Bootstrap prediction mean squared errors of unobserved states based on the Kalman filter with estimated parameters201256162742-s2.0-7996111953310.1016/j.csda.2011.07.010AbtM.Estimating the prediction mean squared error in gaussian stochastic processes with exponential correlation structure19992645635782-s2.0-0033242689KohnR.AnsleyC. F.Prediction mean squared error for state space models with estimated parameters19867324674732-s2.0-000119008110.1093/biomet/73.2.467BaillieR. T.Asymptotic prediction mean squared error for vector autoregressive models19796636756782-s2.0-0018702937NeelamegamP.JamaludeenA.RajendranA.Prediction of calcium concentration in human blood serum using an artificial neural network20114423123192-s2.0-7865057486110.1016/j.measurement.2010.09.035CarottiE. S. G.De MartinJ. C.MerlettiR.FarinaD.Compression of multidimensional biomedical signals with spatial and temporal codebook-excited linear prediction20095611260426102-s2.0-7044936723310.1109/TBME.2009.2027691BachtaW.RenaudP.CuvillonL.LarocheE.ForgioneA.GangloffJ.Motion prediction for computer-assisted beating heart surgery20095611255125632-s2.0-7044937991110.1109/TBME.2009.2026054LinH.-H.BeckC. L.BloomM. J.On the use of multivariable piecewise-linear models for predicting human response to anesthesia20045111187618872-s2.0-634428021910.1109/TBME.2004.831541AtalB. S.The history of linear prediction20062321541612-s2.0-33644616041AdlerR. J.FeldmanR. E.TaqquM. S.1998Boston, Mass, USABirkhäuserSamorodnitskyG.TaqquM. S.1994New York, NY, USAChapman & HallMandelbrotB. B.1998SpringerLiM.ZhaoW.On 1/f noiseMathematical Problems in Engineering. In pressKornG. A.KornT. M.1961McGraw-HillTuckerA.Garway-HeathD.The pseudotemporal bootstrap for predicting glaucoma from cross-sectional visual field data201014179852-s2.0-7684910669010.1109/TITB.2009.2023319CarliniM.CastellucciS.Modelling the vertical heat exchanger in thermal basin6785Proceedings of the International Conference on Computational Science and Its Applications (ICCSA '11)Springer277286Lecture Notes in Computer ScienceCarliniM.CattaniC.TucciA.Optical modelling of square solar concentrator6785Proceedings of the International Conference on Computational Science and Its Applications (ICCSA '11)Springer287295Lecture Notes in Computer ScienceBhansaliR. J.KokoszkaP. S.Prediction of long-memory time series: a tutorial review2003621321BisagliaL.BordignonS.Mean square prediction error for long-memory processes20024321611752-s2.0-003655702310.1007/s00362-002-0095-xLiM.LiJ.-Y.On the predictability of long-range dependent series2010201092-s2.0-7795352254510.1155/2010/397454397454