Robust method for periodicity detection and characterization of irregular cyclical series in terms of embedded periodic components

Physical Review E - Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics

Volumn 59, Issue 4, 1999, Pages 4013-4025

  Kanjilal, P P ^a   Bhattacharya, J ^a   Saha, G ^a  

^a INDIAN INSTITUTE OF TECHNOLOGY   (India)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 0000068037     PISSN: 1063651X     EISSN: None     Source Type: Journal    
DOI: 10.1103/PhysRevE.59.4013     Document Type: Article

References (70)

1. T. A. Denton, Am. Heart J. 120, 1419 (1990)
2. A. L. Goldberger, D. R. Rigney, and B. J. West, Sci. Am. 262, 42 (1990)
3. A. Babloyantz and A. Destexhe, Biol. Cybern. 58, 203 (1988)
4. L. Glass, Proc. R. Soc. London, Ser. A 413, 9 (1987).
5. P. P. Kanjilal, S. Palit, and G. Saha, IEEE Trans. Biomed. Eng. 44, 51 (1997).
6. A. Schuster, Terr. Magn. Atmos. Electr. 3, 13 (1898)
7. H. O. A. Wold, A Study in the Analysis of Stationary Time Series (Almqvist and Wiksells, Uppsala, Sweden, 1954)
8. B. P. Lathi, Signals, Systems and Controls (Harper & Row, New York, 1974)
9. S. M. Kay, Modern Spectral Estimation (Prentice Hall, Englewood Cliffs, NJ, 1988)
10. P. S. Naidu, Modern Spectrum Analysis of Time Series (CRC, Boca Raton, FL, 1996).
11. W. S. Pritchard and D. W. Duke, Int. J. Neurosci. 67, 31 (1992)
12. Machinery of the Mind, edited by E. R. John (Birkhauser, Boston, 1990)
13. A. C. K. Soong and I. J. M. Stuart, Biol. Cybern. 62, 55 (1988).
14. A. L. Goldberger, K. Kobalter, and V. Bhargava, IEEE Trans. Biomed. Eng. 33, 874 (1986)
15. M. C. Mackey and L. Glass, Science 197, 287 (1977).
16. B. T. Grenfell, J. R. Statist. Soc. B 54, 383 (1992)
17. L. F. Olsen and W. Schaffer, Science 249, 499 (1990)
18. W. P. London and J. A. Yorke, Am. J. Epidemiol. 98, 453 (1973).
19. R. Vautard, P. Yiou, and M. Ghil, Physica D 58, 95 (1992)
20. M. Ghil and R. Vautard, Nature (London) 350, 324 (1991).
21. J. Kurths and A. A. Ruzmaikin, Sol. Phys. 126, 407 (1990)
22. G. U. Yule, Philos. Trans. R. Soc. London, Ser. A 226, 267 (1927).
23. D. M. Raup, Nature (London) 314, 341 (1985)
24. M. R. Rampino and R. B. Stothers, Science 226, 1427 (1984)
25. D. M. Raup and J. J. Sepkoski, Jr., Proc. Natl. Acad. Sci. USA 81, 801 (1984).
26. T. Serre, Z. Kollath, and J. R. Buchler, Astron. Astrophys. 311, 833 (1996).
27. C. O. Weiss and J. Brock, Phys. Rev. Lett. 57, 2804 (1986).
28. D. C. Park, IEEE Trans. Inf. Theory 6, 442 (1991).
29. N. O. Weiss, Philos. Trans. R. Soc. London, Ser. A 330, 617 (1990).
30. A. Berger, J. L. Mélice, and I. van der Mersch, Philos. Trans. R. Soc. London, Ser. A 330, 529 (1990).
31. C.-S. Poon and C. K. Merrill, Nature (London) 389, 492 (1997).
32. P. C. Ivanov, Nature (London) 383, 323 (1996).
33. P. So, Phys. Rev. Lett. 76, 4705 (1996)
34. R. Badii, Rev. Mod. Phys. 66, 1389 (1994)
35. O. Binham and W. Wenzel, Phys. Rev. Lett. 63, 819 (1989)
36. D. P. Lathrop and E. J. Kostelich, Phys. Rev. A 40, 4028 (1989)
37. O. Binham and W. Wenzel, Phys. Rev. Lett. 63, 819 (1989)
38. Phys. Rev. Lett.P. Cvitanovic, 61, 2729 (1988).
39. G. H. Golub and C. F. Van Loan, Matrix Computations (Johns Hopkins University Press, Baltimore, 1989).
40. R. A. Thisted, Elements of Statistical Computing: Numerical Computations (Chapman and Hall, New York, 1988)
41. C. L. Lawson and R. J. Hanson, Solving Least Squares Problems (Prentice-Hall, Englewood Cliffs, NJ, 1974).
42. P. P. Kanjilal, Adaptive Prediction and Predictive Control (Peter Peregrinus, Stevenage, UK, 1995).
43. SVD and Signal Processing: Algorithms, Applications, and Architectures, edited by F. Deprettere (North-Holland, Amsterdam, 1988).
44. V. C. Klema and A. J. Laub, IEEE Trans. Autom. Control. AC-25, 164 (1980)
45. R. V. Patel, A. J. Laub, and P. M. Van Dooren, Numerical Linear Algebra Techniques for Systems and Control (IEEE, New York, 1994), pp. 1–35. There are many efficient and numerically stable algorithms 20 available for the computation of SVD of any matrix.
46. D. S. Broomhead and G. P. King, Physica D 20, 217 (1986)
47. A. I. Mees, P. E. Rapp, and L. S. Jennings, Phys. Rev. A 36, 340 (1987)
48. Phys. Rev. AA. M. Albano, 38, 3017 (1988)
49. E. J. Kostelich and T. Schreiber, Phys. Rev. E 48, 1752 (1993).
50. The “period-length spectrum or p spectrum” can also be called “singular value ratio” spectrum of (Formula presented) [P. P. Kanjilal and S. Palit, IEEE Trans. Signal Process. 43, 1536 (1995), 20]. Since, in place of (Formula presented) any other measure of closeness to rank oneness can serve, we use the generic term “p spectrum.”
51. The p spectrum is detrended using bidirectional low-pass filtering, [R. L. Longini, IEEE Trans. Biomed. Eng. 22, 432 (1975), 20], and the information in the positive direction only is retained. For bidirectional filtering, the data set is first passed through a linear filter, the filtered output is time reversed and passed through the same filter again; the time reversed second filter output subtracted from the raw signal forms the desired signal. Here a linear filter with a pole at 0.2 is used. This is a general signal processing exercise, as otherwise the spectrum rides on a trend; since the prime information is in the amplitude of the peaks (with respect to the neighborhood), the spectrum is clearer with detrending and low-pass bidirectional filtering.p-spectrum may be used without detrending and filtering also [e.g., Figs. 22(b), 22(d), and 22(f)], although the peaks will be less pronounced.
52. Here the periodicity index is derived as follows (which is not the only possible approach). The algorithm moves a data of window of size 4 over the p spectrum and identifies a peak [say (Formula presented)] to be present if its magnitude is at least 20% more than the moving average. The mean of (Formula presented) occurring at integral multiples (j) of the period length n serves as the estimated periodicity index (Formula presented)
53. SVD is a case of joint orthogonal transformation between U and (Formula presented) rotates the points in scatterplot space (of (Formula presented), while U rotates the points in variable space. The decomposed vectors (Formula presented) and (Formula presented) are considered “conjugate” [I. J. Good, Technometrics 11, 823 (1969)], and are expected to have related physical interpretations. For the present rank-one approximation, the perturbation bounds on (Formula presented) (Formula presented) and (Formula presented) are discussed in Ref. 2.
54. P. P. Kanjilal, G. Saha, and T. J. Koickal, IEEE Trans. Syst. Man Cybern. Pt. B 29, 1 (1999).
55. C. L. Mallows, Technometrics 15, 661 (1973). Minimization of (Formula presented) ensures optimal modeling (where with m data sets when p is the optimal model order leading to (Formula presented) residual sum of square error and n is the total no. of candidate regressors) as in C. Daniel and F. S. Wood, Fitting Equations to Data (Wiley, New York, 1971).
56. In the present case only up to quadratic terms have been used (which is not a limitation) subject to the availability of data. In place of the present model for (Formula presented) alternative nonlinear modeling methods [as in S. Haykin, Neural Networks: A Comprehensive Foundation (MacMillan, New York, 1994), 20 etc.] may be used.
57. In case of a dynamic series, all three periodicity attributes of the component(s) may vary with time. The present prediction scheme assumes constant periodicity; no restrictions are imposed on the variations of the scaling factors.
58. R. M. May, Nature (London) 261, 459 (1976)
59. R. M. MayProc. R. Soc. London, Ser. A 413, 217 (1987)
60. S. H. Strogatz, Nonlinear Dynamics and Chaos (Addison-Wesley, Reading, MA, 1994).
61. L. Cohen, Time Frequency Analysis (Prentice-Hall, Englewood Cliffs, NJ, 1995).
62. M. Henon, Commun. Math. Phys. 50, 69 (1976).
63. A. Weigend, D. Rumelhart, and B. Huberman, Int. J. Neural Syst. 1, 193 (1990)
64. M. Casdagli, J. R. Statist. Soc. B 54, 303 (1992);Weigend et al. (1990) and Berger et al. 14 considered the sunspot series to be low-dimensional chaotic while Casdagli (1992) concluded it to possess high-dimensional nonlinearity.
65. H. Tong, Nonlinear Time Series: A Dynamical System Approach (Clarendon, Oxford, 1996). Yule 8 (1927) developed linear stochastic models whereas stochastic nonlinear models have been proposed by Tong (1996).
66. G. Sugihara and R. M. May, Nature (London) 344, 734 (1990). It is said that chaoticity reflects in drooping correlation coefficient (e.g., for sunspot series) for long prediction horizons.
67. J. Theiler, Physica D 58, 77 (1992). We used the amplitude adjusted Fourier transform method as surrogate generator; an ensemble of 16 implementations was used.
68. U. Huebner, N. B. Abraham, and C. O. Weiss, Phys. Rev. A 40, 6354 (1989). The laser sequence has an underlying attractor dimension of 2.0–2.3 and an entropy rate (Formula presented) where T is the average intensity pulsing period. Our analysis unearths new structural features of the laser series in terms of period, and pattern variations (Fig. 55).
69. J. D. Farmer, Physica D 4, 366 (1982). For (Formula presented) correlation (Formula presented)
70. On similar analysis [J. Bhattacharya and P. P. Kanjilal (unpublished)] on x variable of Rossler series [N. H. Packard, J. P. Crutchfield, J. D. Farmer, and R. S. Shaw, Phys. Rev. Lett. 45, 712 (1980)], two periodic components are obtained, which lead to good predictability (e.g., prediction correlation coefficient as high as 0.9 for (Formula presented). Fourier domain analysis shows four dominant Fourier components of frequency 0.25, 0.22, 0.2, 0.167 (where a periodicity of (Formula presented) frequency 1);the corresponding strengths are LS estimated. Multistep prediction with these components produce poor validation results, unlike the proposed method.