Fact-checked by Grok 2 weeks ago

Multitaper

The multitaper method is a nonparametric spectral estimation technique in used to compute the power () of a by applying multiple orthogonal tapers to the data and averaging the resulting , thereby reducing bias and variance compared to single-taper approaches. Developed by . Thomson in , it addresses limitations in traditional methods like the by minimizing through the use of discrete prolate spheroidal sequences (DPSS), also known as Slepian sequences, which are optimal windows that concentrate energy within a specified . This approach enables high-resolution estimates suitable for distinguishing continuous spectra from narrowband or line components in noisy data. Key advantages of the multitaper method include its robustness to finite-length observations and its ability to provide adaptive weighting schemes, such as Thomson's adaptive weighting, which further suppresses broadband noise while preserving signal peaks. The number of tapers, typically 2TW - 1 where is the time-bandwidth product, balances and , with common values like TW = 2 or 3 used in practice. Extensions of the method, including multi-taper cross-spectral analysis and estimation, expand its utility for multivariate signals. The technique has been widely adopted across diverse fields, including for seismic and climate data analysis, for (EEG) studies of and , and , and speech analysis, due to its superior performance in handling non-stationarities and low signal-to-noise ratios. Influential implementations and refinements appear in software libraries like MATLAB's Signal Processing Toolbox and Python's , facilitating its accessibility for research and engineering applications.

Introduction

Definition and Overview

The multitaper method is a non-parametric technique designed to estimate the () of stationary ergodic random processes from finite samples of data. Developed by . Thomson, it serves as a key tool in and analysis for producing reliable spectral estimates under limited data conditions. The core purpose of multitapering is to enhance the accuracy of estimates by addressing common issues in transform-based methods, such as and high variance, through the use of multiple orthogonal data windows. By averaging the resulting periodograms, the method achieves a more stable and consistent spectral representation without relying on parametric assumptions about the underlying process. In its basic workflow, the multitaper approach applies K orthogonal tapers to the input data, computes the individual periodograms from each tapered version, and averages them to yield the final estimate. Slepian sequences are typically employed as these tapers for their favorable properties in concentrating energy within a specified .

Historical Development

The multitaper method traces its origins to the foundational work on prolate spheroidal wave functions (PSWFs) by Slepian and colleagues during the and 1970s. In a series of papers, Slepian explored the properties of PSWFs, demonstrating their optimality for concentrating a function's energy within a finite while maintaining band-limited characteristics, which minimizes leakage in . This insight into concentration inspired the use of such functions as data tapers for . Slepian's extension to the case in 1978 provided the mathematical framework for applying these sequences to finite-length , laying the groundwork for practical implementations in . Building on Slepian's contributions, David J. Thomson developed the multitaper spectral estimation technique in the early 1980s while working at Bell Laboratories. Thomson recognized the limitations of single-taper periodograms and proposed averaging multiple orthogonal tapers derived from discrete prolate spheroidal sequences (DPSS) to achieve unbiased low-variance spectral estimates. This culminated in his landmark 1982 publication, "Spectrum Estimation and Harmonic Analysis," which formally introduced the method and demonstrated its superiority for harmonic analysis in noisy data. Post-1982, the multitaper method saw initial adoption in and fields, where it proved effective for analyzing nonstationary seismic signals and reducing in studies. Thomson further extended the approach in the and to multitaper estimates of —measuring linear relationships between —and techniques for line fitting in spectra, enhancing its utility for multivariate analysis. Key milestones in the method's evolution include its integration into mainstream software tools by the 1990s, such as MATLAB's Signal Processing Toolbox, which implemented Thomson's multitaper power spectral density estimator via the pmtm function, facilitating broader accessibility for researchers. The technique's profound applied impact was acknowledged in 2013 when Thomson received the Statistical Society of Canada's Impact Award for his multitaper innovations and their widespread influence across disciplines.

Motivation and Principles

Limitations of Traditional Spectral Estimation

Traditional spectral estimation methods, such as the , suffer from inherent bias due to the finite length of data records, which causes as energy from one spreads into adjacent frequencies through the with the window's , often a with significant . This leakage is particularly problematic for spectra with sharp features or large dynamic ranges, where the main lobe width limits to approximately 1/N cycles per sampling interval for N data points, smearing closely spaced peaks and rendering estimates unreliable even for large datasets. The also exhibits high variance that does not diminish with increasing data length, making it an inconsistent of the power (), as successive estimates remain uncorrelated and fluctuate erratically around the true regardless of sample size. For Gaussian , this variance equals the square of the true , while for general stationary processes, it leads to poor statistical reliability, especially in short time series common in geophysical or biomedical applications. These issues arise because the relies on a single without smoothing, violating consistency requirements for estimation. Methods attempting to mitigate variance, such as Bartlett's approach of averaging periodograms over non-overlapping segments, introduce trade-offs by reducing variance proportionally to the number of segments but at the expense of frequency resolution, as shorter segments broaden the effective and increase through additional . This segmentation sacrifices the ability to resolve fine details, limiting utility for processes with closely spaced frequencies, while the rectangular exacerbates leakage compared to tapered alternatives. Classical methods assume the underlying process is and ergodic, allowing the to approximate the under infinite data conditions, but real-world signals often violate these assumptions due to non-stationarities or noise, leading to distorted estimates that fail to capture true spectral structure in noisy, finite observations.

Benefits of Multitapering

The multitaper method addresses key challenges in spectral estimation by leveraging multiple orthogonal tapers, such as Slepian sequences, to produce more reliable power spectral density estimates. This approach combines the strengths of and non-parametric techniques, offering reduced variance and bias compared to traditional single-taper periodograms. A primary benefit is the significant reduction in variance achieved by averaging K independent estimates derived from orthogonal tapers, which yields an effective approximately equal to 2K and approaches near-optimal efficiency for a given time-bandwidth product (2NW). This averaging process stabilizes the spectral estimate, particularly in noisy environments, by mitigating the high variability inherent in single-window methods. Multitapering also minimizes through the use of tapers that concentrate nearly all their energy within a specified band, thereby reducing far more effectively than single-window techniques. The is bounded by the eigenvalue-related term (1 - λ_k) times the variance, ensuring that the estimate remains concentrated and accurate even for finite data lengths. Unlike averaged segment methods such as , multitapering preserves the full length of the data series without requiring segmentation, thereby maintaining higher frequency while avoiding the resolution trade-offs associated with window overlap or shortening. This preservation is especially advantageous for applications demanding fine spectral detail. Additionally, the method adapts well to unevenly spaced or short datasets, providing robust confidence intervals based on the approximately 2K effective , which enhances its utility in real-world scenarios with limited observations.

Mathematical Formulation

Slepian Sequences and DPSS

Discrete prolate spheroidal sequences (DPSS), also known as Slepian sequences, are a family of finite-length sequences designed to maximize the concentration of their within a specified band, given a time-bandwidth product NW, where N is the sequence length and W is the half-bandwidth (in normalized units). These sequences arise as the solutions to a variational eigenvalue problem that optimizes the ratio of within the band [-W, W] to the total across the full range [-1/2, 1/2]. The core formulation of this concentration problem is given by maximizing the eigenvalue \lambda in \frac{\int_{-W}^{W} \left| \sum_{n=-N/2}^{N/2} u_n e^{-i 2\pi f n} \right|^2 df}{\int_{-1/2}^{1/2} \left| \sum_{n=-N/2}^{N/2} u_n e^{-i 2\pi f n} \right|^2 df} = \lambda, where u_n are the coefficients of the sequence, and the sums assume N even for simplicity. The eigenfunctions corresponding to the largest eigenvalues provide the sequences with the highest spectral concentration, making them ideal for applications requiring low leakage outside the desired band. A key property of DPSS is that the first $2NW sequences exhibit eigenvalues \lambda_k close to 1, indicating near-perfect energy concentration within the band [-W, W], while subsequent eigenvalues drop sharply toward 0, marking a transition to poor concentration. These sequences are nearly to each other, with the orthogonality becoming exact in the limit of large N. In practice, the tapers used in are derived from these sequences as v_k(n) = \sqrt{\lambda_k / N} \, u_k(n), which normalizes them to account for the eigenvalue and sequence length, ensuring consistent energy scaling. The computation of DPSS leverages a formulation derived from the properties, as detailed in Slepian's analysis, which transforms the eigenvalue problem into a solvable linear task via a symmetric tridiagonal Jacobi . This approach guarantees and efficiency, while the inherent concentration minimizes sidelobe leakage in the , a critical advantage over simpler window functions. These sequences form the basis for the tapers in multitaper spectral estimation techniques.

Multitaper Spectral Estimator

The multitaper estimator provides an estimate of the () of a stationary by averaging the squared transforms obtained using multiple orthogonal tapers. For a x_n of length N, the estimator is given by \hat{S}(f) = \frac{ \sum_{k=0}^{K-1} \lambda_k \left| \sum_{n=0}^{N-1} x_n v_k(n) e^{-i 2\pi f n} \right|^2 }{ \sum_{k=0}^{K-1} \lambda_k }, where v_k(n) are the prolate spheroidal sequences (DPSS) serving as tapers, \lambda_k are the corresponding eigenvalues measuring their concentration, and K is the number of tapers used (typically K \approx 2NW - 1, with W the half-bandwidth). This formulation incorporates eigenvalue weighting via the \lambda_k (direct method), which downweights tapers with poorer concentration to minimize in the estimate. Thomson's adaptive extension further optimizes frequency-dependent weights for spectra that vary rapidly. An extension to cross-spectral estimates between two series x_n^{(l)} and x_n^{(m)} yields the and cross-PSD as \hat{S}^{lm}(f) = \frac{1}{K} \sum_{k=0}^{K-1} Y_k^l(f) \left[ Y_k^m(f) \right]^*, where Y_k^l(f) = \sum_{n=0}^{N-1} x_n^{(l)} v_k(n) e^{-i 2\pi f n} and similarly for Y_k^m(f). This averaged form reduces leakage and variance compared to single-taper cross-periodograms. The of \hat{S}(f) is reduced through the spectral concentration of the tapers, with local bias proportional to the second of the true S(f) and bounded bias terms involving (1 - \lambda_k). The variance is approximately \frac{[S(f)]^2}{K} (i.e., \frac{1}{K} times that of the ) for slowly varying spectra. For inputs, the estimator follows a scaled with $2K , \hat{S}(f) \sim \frac{S(f)}{2K} \chi^2_{2K}, enabling intervals. In Thomson's extension for harmonic analysis, an F-test detects line components by comparing the multitaper estimate to a baseline, using an F-statistic with 2 and $2K-2 degrees of freedom to assess significance against the chi-squared null.

Implementation and Computation

Generating Tapers

The generation of discrete prolate spheroidal sequences (DPSS), also known as Slepian tapers, for multitaper spectral analysis involves solving a discrete eigenvalue problem derived from the time-domain concentration criterion. In the discrete case, this is formulated as finding the eigenvectors of a symmetric Toeplitz matrix that approximates the integral operator for bandlimiting, but for efficient computation, Slepian approximated this matrix by a real symmetric tridiagonal (Jacobi) matrix whose elements are derived from the time-bandwidth parameters. The eigenvalue decomposition of this tridiagonal matrix yields the DPSS tapers as the eigenvectors corresponding to the largest eigenvalues, with the decomposition typically performed using QL or QR algorithms for stability and speed. Key parameters in taper generation include the sequence length N and the time-bandwidth product $2NW, where W is the normalized half-bandwidth (with W < 0.5). The product $2NW determines the effective and is typically selected in the range of 2 to 4 for many applications, resulting in K \approx 2NW tapers; specifically, the first K eigenvectors with the highest eigenvalues \lambda_k (ideally those exceeding 0.95 for good concentration) are retained, as these provide the optimal balance of time and localization. For numerical implementation, direct eigendecomposition of the N \times N tridiagonal matrix is feasible and stable for N up to several thousand using optimized linear algebra routines like those in LAPACK, exploiting the matrix's sparsity and symmetry to achieve O(N) complexity per eigenvector. In scenarios approximating the continuous-time case or for validation, fast Fourier transforms (FFT) can evaluate the bandlimiting integrals underlying the Toeplitz form, though discrete matrix methods predominate for finite N. Established libraries, such as SciPy's signal.dpss function, implement this tridiagonal approach, returning the tapers normalized to unit energy along with their eigenvalues. To address edge cases like small N (e.g., N < 100) or large NW (where eigenvalue clusters may degrade separation), post-processing via iterative refinement—such as Gram-Schmidt orthogonalization or subspace iteration—can enforce the tapers' required and unit norm, preventing accumulation of rounding errors in finite-precision arithmetic. These resulting sequences possess near-optimal energy concentration within the specified while remaining approximately timelimited.

Estimation Procedure

The estimation procedure for the multitaper method begins with preprocessing the input data. The data is typically detrended by removing the or a linear trend to eliminate low-frequency es, and the time-bandwidth product NW is selected based on the desired , with the number of tapers set to K = 2NW - 1 or approximately $2NW to balance and variance. Tapers are generated as described in the prior section on generating tapers. For each taper index k = 1, \dots, K, the tapered is computed as \tilde{x}_k(f) = \sum_{n=0}^{N-1} x_n v_k(n) e^{-i 2\pi f n}, where x_n are the detrended data samples, v_k(n) is the k-th taper (normalized such that \sum_n v_k^2(n) = 1), N is the data length, and f is the . The corresponding eigenspectrum, or tapered , is then formed as I_k(f) = |\tilde{x}_k(f)|^2. This step is repeated for all K tapers, often using (FFT) implementations for efficiency, with zero-padding to the next power of 2 if needed to improve . The estimate is obtained by averaging the eigenspectra, typically with weights that account for the eigenvalues to minimize variance. In the non-adaptive case, equal weights are applied: \hat{S}(f) = \frac{1}{K} \sum_{k=1}^K I_k(f). For adaptivity, which improves performance in the presence of line components or non-white , the weights are w_k = \lambda_k / \sum_{j=1}^K \lambda_j, yielding \hat{S}(f) = \sum_{k=1}^K w_k |\tilde{x}_k(f)|^2, or more precisely, an iterative adjustment based on estimated to downweight tapers affected by spectral lines. This weighted average reduces bias while suppressing leakage from discrete spectrum components. Post-processing may include additional smoothing across adjacent frequencies if higher resolution is not critical, though the inherent averaging in multitapering often suffices. Confidence intervals are derived from the chi-squared distribution of the estimate, assuming a locally white or Gaussian process; for large K, the 95% bounds are approximately \hat{S}(f) \left(1 \pm \frac{2.98}{\sqrt{2K}}\right), scaled by the degrees of freedom $2K. For handling coherent structures or regression of line components, Thomson's F-test is applied to detect and adaptively regress harmonics, using the eigenspectra to estimate line amplitudes and phases.

Applications

In Geophysics and Seismology

In , the multitaper method has been applied to estimate earthquake source time functions through deconvolution techniques, providing robust recovery of rupture characteristics from noisy seismograms. This approach leverages multiple orthogonal tapers to minimize leakage and variance, enabling accurate inversion of source spectra even for short-duration events. Additionally, multitaper analysis facilitates the detection of harmonic modes in Earth's free oscillations following major s, such as those after , by employing Thomson's for identifying lines amid . This test assesses the significance of peaks, allowing precise of spheroidal and toroidal modes that reveal and structure. In , multitaper power (PSD) estimation analyzes nonstationary from bottom records to study wave propagation and internal tide generation, effectively reducing bias in short, gappy datasets. For instance, it identifies high-Q peaks indicative of resonant modes, distinguishing them from red backgrounds. Similarly, in geomagnetism, the method computes PSDs of field variations from data, capturing nonstationary fluctuations in the geomagnetic and mitigating leakage from short observation windows. These applications highlight multitaper's utility in handling temporally varying signals, such as those influenced by core dynamics or solar interactions. A notable extension is the multidimensional multitaper approach developed by Simons et al. (2006), which adapts Slepian sequences to spherical geometries for estimating fields from satellite observations like those from . This spatiospectral localization concentrates energy in both spatial and domains, enabling high-resolution recovery of local anomalies with reduced sidelobe contamination. The multitaper method excels in geophysical contexts with noisy, unevenly sampled data, offering superior for detecting weak signals. By averaging eigenspectra from discrete prolate spheroidal sequences, it suppresses variance and , allowing identification of subtle features that single-taper methods obscure.

In Neuroscience and Biomedical Signals

In (EEG) analysis of , multitaper methods enable dynamic spectral estimation to capture nonstationary transitions across sleep stages, such as from to non-rapid (NREM) sleep, by producing high-resolution spectrograms that reveal evolving neural oscillations like and rhythms. The Chronux toolbox, a widely adopted open-source platform for neural , implements multitaper spectrograms tailored for EEG data, facilitating the quantification of microstructure and fragmentation in clinical contexts. For instance, work from the Prerau lab has applied these techniques to dissect neurophysiology, demonstrating how multitaper analysis uncovers subtle dynamic changes in power spectra during NREM that traditional methods overlook. Extensions of multitaper methods in biomedical include estimation to assess between neural signals, which helps map interactions across brain regions during cognitive tasks or pathological states like . This approach mitigates bias from common signals, such as volume conduction in EEG, providing more reliable measures of neural . Additionally, multitaper is particularly suited for handling short trials in event-related potentials (ERPs), where limited data length challenges traditional estimators; by using multiple orthogonal tapers, it yields stable power estimates for transient responses like the P300 component evoked by auditory stimuli. A specific application involves multivariate multitaper methods for detecting directional influences in neural time series, as implemented in toolkits like Chronux, which extend univariate techniques to cross-spectral analysis for inferring causal relationships in multi-channel recordings from cortical arrays. This has been used to estimate harmonic responses in optical imaging data, revealing directed neural interactions during . In broader biomedical contexts, multitaper's benefits include robustness to artifacts, such as motion-induced noise in EEG or irregular sampling in physiological recordings, and its ability to handle variability in rhythms like (HRV), where it provides unbiased spectral estimates of autonomic balance without requiring data .

Comparisons and Extensions

Comparison with Other Methods

The multitaper method addresses key limitations of the raw by reducing the variance of the spectral estimate by a factor of approximately $1/K, where K is the number of tapers, through averaging orthogonal periodogram estimates, while preserving the same . In contrast, the periodogram is computationally simple but suffers from high variance and statistical inconsistency, particularly in noisy environments, leading to unreliable estimates. This improvement in multitaper comes at the expense of higher computational demands for taper generation and averaging. Compared to , multitaper leverages the entire data length for estimation, yielding superior , and uses orthogonal tapers to eliminate the from segment overlaps inherent in Welch's overlapping windows. Welch's approach, while efficient for long due to its segmented averaging, shortens the effective data length per estimate, compromising and introducing potential leakage . Research on short electroencephalogram (EEG) signals has shown multitaper to provide a better than , resulting in smoother and more stable power estimates. Quantitatively, for a fixed time-bandwidth product NW, multitaper employs K \approx 2NW tapers, achieving roughly $2K in the spectral estimate, which enhances reliability under the assumption. , using 50% overlap and a , attains approximately $8/3 per segment, limiting its effective relative to multitaper for equivalent bandwidths. Simulations detailed in Percival and Walden (1993) highlight multitaper's advantages in variance reduction and resolution across diverse noise conditions. Multitaper is ideally suited for scenarios demanding high resolution in short or noisy datasets, such as biomedical , where its low-bias, low-variance properties outperform alternatives. Conversely, the or may be selected for simpler, uniform spectra where computational efficiency takes precedence over optimal statistical performance.

Modern Variants and Developments

Since the early , adaptive variants of the multitaper method have emerged to address biases in estimation, particularly through curvature corrections that adjust for the non-flatness of the spectral window. A key advancement is the adaptive multitaper , which reduces by tapers based on signal characteristics, as detailed in a published in the Geophysical Journal International. For handling nonstationary data, state-space multitaper methods integrate dynamic models to track time-varying spectra, enabling robust estimation in signals with evolving statistics. This approach, developed in a 2016 MIT by Michael K. Behr, combines Kalman filtering with multitaper techniques to model nonstationarities in geophysical . Extensions to multidimensional data have expanded multitaper applicability beyond one-dimensional time series. The multidimensional multitaper method, introduced by Simons in 2006, applies discrete prolate spheroidal sequences (DPSS) to spatial or spatiotemporal datasets, improving resolution in fields like oceanography and geophysics. Recent developments address challenges in irregular sampling and data gaps. For nonuniformly sampled data, an integration of multitaper with the generalized periodogram spectral estimation (GPSS) of Bronez, as in a 2024 arXiv preprint on fast multitaper power spectrum estimation, allows direct computation of power spectra without interpolation, preserving phase information. Similarly, a 2025 paper in Earth and Space Science extends magnitude-squared coherence to multitaper frameworks for handling missing data, using adaptive weighting to mitigate artifacts in incomplete records. Software implementations have democratized access to these advanced techniques. In , the multitaper package by Prieto (2022) provides tools for power spectral density () and estimation, supporting adaptive weighting and parallel computation for large datasets. The 'multitaper' on CRAN offers similar functionality, including support for Thomson's method and extensions for analysis. For applications, the MATLAB-based Chronux toolbox incorporates multitaper routines tailored for spike train and analysis. In recent applications, multitaper variants have been adapted for astronomy, where the mtNUFFT algorithm (2024, Astronomical Journal) accelerates power spectrum computation for non-uniform data from telescope arrays. In , multitaper methods analyze long-term for variability detection, as in studies of ENSO patterns using adaptive bias-corrected estimators. Ongoing research focuses on fast algorithms for , such as GPU-accelerated taper generation and approximate DPSS computation, to scale multitaper analysis to petabyte-scale datasets in monitoring.

References

  1. [1]
  2. [2]
    pmtm - Multitaper power spectral density estimate - MATLAB
    Returns Thomson's multitaper power spectral density (PSD) estimate, pxx, of the input signal x using Discrete Prolate Spheroidal (Slepian) Sequences as tapers.
  3. [3]
  4. [4]
    Multi-Taper Method (MTM) | Theoretical Climate Dynamics - ucla.edu
    Spectral Estimation. MTM can provide estimates of both the singular components (ie, the lines'') and the continuous component (ie, the background) of the ...
  5. [5]
    A review of multitaper spectral analysis - PubMed - NIH
    Among the techniques that are used to estimate the spectral representation of a system based on finite observations, multitaper spectral estimation has many ...<|control11|><|separator|>
  6. [6]
  7. [7]
    [PDF] Prolate Spheroidal Wave Functions, Fourier Analysis and Uncertainty
    I. By D. SLEPIAN and H. O. POLLAK. (Manuscript received August 1, 1960). –. A complete ...Missing: David 1970s
  8. [8]
    [PDF] PSWF-V.pdf - UC Davis Mathematics
    This paper investigates the extent to which a time series can be concentrated on a finite index set and also have its spectrum concentrated on a subinterval of ...
  9. [9]
    Multitaper spectral analysis of high‐frequency seismograms - Park
    Nov 10, 1987 · Slepian, D., Prolate spheroidal wave functions, Fourier analysis, and uncertainty, V, The discrete case, Bell Syst. Tech. J., 57, 1371–1429 ...
  10. [10]
    David Thomson, SSC Award for Impact of Applied and Collaborative ...
    The key accomplishment of this paper was to explain why multiple data tapers are needed for spectrum estimation. This work gave a consistent estimator for power ...
  11. [11]
    [PDF] Spectrum Estimation and Harmonic Analysis - UC Davis Math
    Spectrum estimation involves direct and indirect methods, with a new method developed for finite samples, and a goal of non-bias and consistency.
  12. [12]
    [PDF] SPECTRAL ANALYSIS OF SIGNALS
    Jun 4, 2015 · ... spectral estimation in the main part of the text. However, several advanced topics are considered in the complements that appear at the end ...
  13. [13]
    [PDF] Spectral Estimation - Wharton Statistics and Data Science
    All phase (relative location/time origin) information is lost. The pe- riodogram would be the same if all of the data were circularly rotated.
  14. [14]
    A Review of Multitaper Spectral Analysis - IEEE EMBS
    Apr 13, 2014 · ... multitaper spectral estimation technique developed by David Thomson in 1982. In recent years, multitaper spectral analysis has gained ...Missing: history | Show results with:history
  15. [15]
  16. [16]
    Prolate spheroidal wave functions, fourier analysis, and uncertainty
    This paper investigates discrete prolate spheroidal sequences and functions, and their use in signal analysis of discrete time series.
  17. [17]
    [PDF] Thomson's Multitaper Method Revisited - arXiv
    Mar 22, 2021 · Abstract. Thomson's multitaper method estimates the power spectrum of a signal from N equally spaced sam-.
  18. [18]
    dpss — SciPy v1.16.2 Manual
    Compute the Discrete Prolate Spheroidal Sequences (DPSS). DPSS (or Slepian sequences) are often used in multitaper power spectral density estimation.Missing: implementation | Show results with:implementation
  19. [19]
    [PDF] Spectral Analysis Tools using the Multitaper Method - CRAN
    This package implements multitaper spectral analysis using Slepian sequences and sine tapers, including weighted estimates, coherence, and complex demodulation.
  20. [20]
    Improving earthquake source spectrum estimation using multitaper ...
    Improving earthquake source spectrum estimation using multitaper techniques.
  21. [21]
    Multiple-taper spectral analysis of terrestrial free oscillations: part I
    Summary. We present a new method for estimating the frequencies of the Earth's free oscillations. This method is an extension of the techniques of Thomson.Missing: density | Show results with:density
  22. [22]
    High‐Q Spectral Peaks and Nonstationarity in the Deep Ocean ...
    Feb 20, 2019 · The power spectral density exhibits frequent narrowband peaks at 6-11 standard deviations above the noise level. Some of the peak frequencies ...
  23. [23]
    Power Spectral Density Background Estimate and Signal Detection ...
    Dec 31, 2020 · We evaluate the power spectral density with the adaptive multitaper method, a nonparametric spectral analysis technique suitable for time ...
  24. [24]
    Sleep Neurophysiological Dynamics Through the Lens of Multitaper ...
    In this paper, we review the neurophysiology of sleep through this lens of dynamic spectral analysis. We begin by reviewing spectral estimation techniques ...
  25. [25]
    Chronux: A Platform for Analyzing Neural Signals - PMC
    Chronux is an open-source software project that aims to fill this void by providing a comprehensive software platform for the analysis of neural signals.
  26. [26]
    [PDF] Sleep Neurophysiological Dynamics Through the Lens of Multitaper ...
    Jan 18, 2017 · As an example of potential clinical utility, we use the multitaper sleep EEG spectrogram to illustrate frag- mentation of NREM during ...
  27. [27]
    The Effect of Common Signals on Power, Coherence and Granger ...
    As a result, our understanding of how common signals affect neural connectivity estimation is incomplete. For example, little is known about recording ...
  28. [28]
    The Effect of Common Signals on Power, Coherence and Granger ...
    As a result, our understanding of how common signals affect neural connectivity estimation is incomplete. For example, little is known about recording ...
  29. [29]
    A Multivariate, Multitaper Approach to Detecting and Estimating ...
    Multitaper methods have become common statistical tools in the analysis of univariate time series that can give improved estimates. Here, we extend univariate ...
  30. [30]
    A multivariate, multitaper approach to detecting and estimating ...
    Multitaper methods have become common statistical tools in the analysis of univariate time series that can give improved estimates. Here, we extend univariate ...
  31. [31]
    Approximate minimum bias multichannel spectral estimation for ...
    Spectral decomposition of variations in heart rate permits noninvasive measurement of autonomic nervous activity in humans and animals.
  32. [32]
    [PDF] Multitaper Spectral Estimation of Power Law Processes
    Walden, E.J. McCoy and D.B. Percival, \The variance of multitaper spectrum estimates for real Gaussian processes," IEEE Trans. Signal Processing, vol. 42 ...Missing: formula | Show results with:formula
  33. [33]
    [PDF] Factor Analysis for Spectral Estimation
    The multitaper method for spectral estimation was intro- duced by D. J. Thomson as an alternative to the periodogram with reduced variance at the cost of ...
  34. [34]
    A Multitaper Frequency-Domain Bootstrap Method - PMC - NIH
    The multitaper method is optimal for minimizing the bias-variance tradeoff in spectral estimation. ... multitaper method always performs better than Welch's ...
  35. [35]
    Spectral Proper Orthogonal Decomposition using Multitaper Estimates
    Dec 20, 2021 · At additional computational cost but for the same data, Multitaper-Welch estimators provide lower variance estimates at fixed frequency ...
  36. [36]
    [PDF] A Review of Multitaper Spectral Analysis
    Multitaper spectral estimation theory was developed through the seminal work of David J. Thomson in 1982 [7]. The goal of this theory is to address the issues ...