Andrew T. Walden

Spectral Analysis for Physical Applications: Multitaper and Conventional Univariate Techniques

Glossary of symbols 1. Introduction to spectral analysis 2. Stationary stochastic processes 3. Deterministic spectral analysis 4. Foundations for stochastic spectral theory 5. Linear time-invariant filters 6. Non-parametric spectral estimation 7. Multiple taper spectral estimation 8. Calculation of discrete prolate spheroidal sequences 9. Parametric spectral estimation 10. Harmonic analysis References Appendix: data and code via e-mail Index.

The Hilbert spectrum via wavelet projections

Non-stationary signals are increasingly analysed in the time-frequency domain to determine the variation of frequency components with time. It was recently proposed in this journal that such signals could be analysed by projections onto the time-frequency plane giving a set of monocomponent signals. These could then be converted to ‘analytic’ signals using the Hilbert transform and their instantaneous frequency calculated, which when weighted by the energy yields the ‘Hilbert energy spectrum’ for that projection. Agglomeration over projections yields the complete Hilbert spectrum. We show that superior results can be obtained using wavelet-based projections. The maximal-overlap (undecimated/stationary/translation-invariant) discrete wavelet transform and wavelet packet transforms are used, with the Fejér-Korovkin class of wavelet filters. These transforms produce decompositions which are conducive to statistical analysis, in particular enabling noise-reduction methodology to be developed and easily and successfully applied.

Statistical Properties and Uses of the Wavelet Variance Estimator for the Scale Analysis of Time Series

Abstract Many physical processes are an amalgam of components operating on different scales, and scientific questions about observed data are often inherently linked to understanding the behavior at different scales. We explore time-scale properties of time series through the variance at different scales derived using wavelet methods. The great advantage of wavelet methods over ad hoc modifications of existing techniques is that wavelets provide exact scale-based decomposition results. We consider processes that are stationary, nonstationary but with stationary dth order differences, and nonstationary but with local stationarity. We study an estimator of the wavelet variance based on the maximal-overlap (undecimated) discrete wavelet transform. The asymptotic distribution of this wavelet variance estimator is derived for a wide class of stochastic processes, not necessarily Gaussian or linear. The variance of this distribution is estimated using spectral methods. Simulations confirm the theoretical results. The utility of the methodology is demonstrated on two scientifically important series, the surface albedo of pack ice (a strongly non-Gaussian series) and ocean shear data (a nonstationary series).

A generalized demodulation approach to time-frequency projections for multicomponent signals

In this paper, we introduce a flexible approach for the time-frequency analysis of multicomponent signals involving the use of analytic vectors and demodulation. The demodulated analytic signal is projected onto the time-frequency plane so that, as closely as possible, each component contributes exclusively to a different ‘tile’ in a wavelet packet tiling of the time-frequency plane, and at each time instant, the contribution to each tile definitely comes from no more than one component. A single reverse demodulation is then applied to all projected components. The resulting instantaneous frequency of each component in each tile is not constrained to a set polynomial form in time, and is readily calculated, as is the corresponding Hilbert energy spectrum. Two examples illustrate the method. In order better to understand the effect of additive noise, the approximate variance of the estimated instantaneous frequency in any tile has been formulated by starting with pure noise and studying its evolving covariance structure through each step of the algorithm. The validity and practical utility of the resulting expression for the variance of the estimated instantaneous frequency is demonstrated via a simulation experiment.

Orthogonal and biorthogonal multiwavelets for signal denoising and image compression

This paper presents new vector filter banks, in particular biorthogonal Hermite cubic multiwavelets with short, smooth duals. We study different preprocessing techniques and the covariance structure of corresponding transforms. Results of numerical experiments in signal denoising and image compression using multi-filters are discussed.We compare the performance of several multi-filters with the performance of standard scalar wavelets such as Daubechies orthogonal external phase and least asymmetric ones and biorthogonal 9- 7 pair. Often multiwavelet scheme turn out to be better. We analyze these results.

The phase–corrected undecimated discrete wavelet packet transform and its application to interpreting the timing of events

This paper is concerned with the development and application of the phase–corrected maximal overlap discrete wavelet packet transform (MODWPT). The discrete cyclic filtering steps of the MODWPT are fully explained. Energy preservation is proven. With filter coefficients chosen from Daubechies least asymmetric class, the optimum time shifts to apply to ensure approximate zero phase filtering at every level of the MODWPT are studied, and applied to the wavelet packet coefficients to give phase corrections which ensure alignment with the original time series. Also, the time series values at each time are decomposed into details associated with each frequency band, and these line up perfectly with features in the original time series since the details are shown to arise through exact zero phase filtering. The phase–corrected MODWPT is applied to a non–stationary time series of hourly averaged Southern Hemisphere solar magnetic field magnitude data acquired by the Ulysses spacecraft. The occurrence times of the shock waves previously determined via manual pattern matching on the raw data match those times in the time–frequency plot where a broadband spectrum is obtained; in other words, the phase–corrected MODWPT provides an approach to picking the location of complicated events. We demonstrate the superiority of the MODWPT in interpreting timing information over two competing methods, namely the cosine packet transform (or ‘local cosine transform’), and the short–time Fourier transform.

The variance of multitaper spectrum estimates for real Gaussian processes

Multitaper spectral estimation has proven very powerful as a spectral analysis method wherever the spectrum of interest is detailed and/or varies rapidly with a large dynamic range. In his original paper D.J. Thomson (1982) gave a simple approximation for the variance of a multitaper spectral estimate which is generally adequate when the spectrum is slowly varying over the taper bandwidth. The authors show that near zero or Nyquist frequency this approximation is poor even for white noise and derive the exact expression of the variance in the general case of a stationary real-valued time series. This expression is illustrated on an autoregressive time series and a convenient computational approach outlined. It is shown that this multitaper variance expression for real-valued processes is not derivable as a special case of the multitaper variance for complex-valued, circularly symmetric processes, as previously suggested in the literature. >

On Testing for Impropriety of Complex-Valued Gaussian Vectors

We consider the problem of testing whether a complex-valued random vector is proper, i.e., is uncorrelated with its complex conjugate. We formulate the testing problem in terms of real-valued Gaussian random vectors, so we can make use of some useful existing results which enable us to study the null distributions of two test statistics. The tests depend only on the sample-size n and the dimensionality of the vector p . The basic behaviors of the distributions of the test statistics are derived and critical values (thresholds) are calculated and presented for certain (n,p) values. For one of these tests we derive a distributional approximation for a transform of the statistic, potentially very useful in practice for rapid and simple testing. We also study the power (detection probability) of the tests. Our results mean that testing for propriety can be a practical and undaunting procedure.

A Statistical Study of Temporally Smoothed Wavelet Coherence

The use of the wavelet coherence of two series in hypothesis testing relies on some sort of smoothing being carried out in order that the coherence estimator is not simply unity. A previous study considered averaging via the use of multiple Morse wavelets. Here we consider time-domain smoothing and use of a single Morlet wavelet. Since the Morlet wavelet is complex-valued, we derive analytic results for the case of wavelet coherence calculated from complex-valued, jointly stationary and Gaussian time series. The temporally smoothed wavelet coherence can be written in terms of Welchs overlapping segment averaging (WOSA) spectrum estimators, and by using multitaper equivalent representations for the WOSA estimators we show that Goodmans distribution is appropriate asymptotically, and readily derive the appropriate degrees of freedom. The theoretical results are verified via simulations and illustrated using solar physics data.

Brain connectivity in positive and negative syndrome schizophrenia

If, as is widely believed, schizophrenia is characterized by abnormalities of brain functional connectivity, then it seems reasonable to expect that different subtypes of schizophrenia could be discriminated in the same way. However, evidence for differences in functional connectivity between the subtypes of schizophrenia is largely lacking and, where it exists, it could be accounted for by clinical differences between the patients (e.g. medication) or by the limitations of the measures used. In this study, we measured EEG functional connectivity in unmedicated male patients diagnosed with either positive or negative syndrome schizophrenia and compared them with age and sex matched healthy controls. Using new methodology (Medkour et al., 2009) based on partial coherence, brain connectivity plots were constructed for positive and negative syndrome patients and controls. Reliable differences in the pattern of functional connectivity were found with both syndromes showing not only an absence of some of the connections that were seen in controls but also the presence of connections that the controls did not show. Comparing connectivity graphs using the Hamming distance, the negative-syndrome patients were found to be more distant from the controls than were the positive syndrome patients. Bootstrap distributions of these distances were created which showed a significant difference in the mean distances that was consistent with the observation that negative-syndrome diagnosis is associated with a more severe form of schizophrenia. We conclude that schizophrenia is characterized by widespread changes in functional connectivity with negative syndrome patients showing a more extreme pattern of abnormality than positive syndrome patients.