Decompose EDA into Tonic and Phasic Components

Separates an electrodermal activity (EDA) signal into its slow-varying tonic component (skin conductance level, SCL) and fast-varying phasic component (skin conductance responses, SCR).

Usage

edaDecompose(
  x,
  method = c("highpass", "median", "cda", "dda", "cvxeda", "cvxeda_fast"),
  cutoff = 0.05,
  window_sec = 4,
  tau1 = 0.75,
  tau2 = 2,
  alpha = 0.01,
  gamma = 0.1,
  delta_knot = 10,
  qp_sr = 4,
  optimize_tau = FALSE,
  assay_name = NULL,
  output_tonic = "tonic",
  output_phasic = "phasic"
)

Arguments

x: A PhysioExperiment object containing EDA data.
method: Decomposition method: "highpass", "median", "cda", "dda", "cvxeda", or "cvxeda_fast". Default is "highpass".
cutoff: Cutoff frequency in Hz for the highpass method (default: 0.05).
window_sec: Window length in seconds for the median method (default: 4).
tau1: SCR rise time constant in seconds for CDA/cvxEDA (default: 0.75).
tau2: SCR decay time constant in seconds for CDA/cvxEDA (default: 2.0).
alpha: L1 sparsity penalty for cvxEDA (default: 0.01). Ignored by other methods.
gamma: Smoothness weight for cvxEDA tonic component (default: 0.1). Ignored by other methods.
delta_knot: Tonic spline knot spacing in seconds for the "cvxeda" QP (default: 10).
qp_sr: Target rate (Hz) at which the "cvxeda" QP is solved; the signal is decimated towards this rate for tractability (default: 4).
optimize_tau: Logical; for "cda" / "dda", optimise the decay constant tau2 to minimise the driver's negative energy (default: FALSE).
assay_name: Name of the input assay. If NULL, uses defaultAssay(x).
output_tonic: Name for the tonic output assay (default: "tonic").
output_phasic: Name for the phasic output assay (default: "phasic").

Value

A modified PhysioExperiment with new assays:

tonic: The slow-varying skin conductance level (SCL) component.
phasic: The fast-varying skin conductance response (SCR) component.
driver: (CDA and cvxEDA variants only) The sudomotor nerve activity driver signal.

Decomposition parameters are stored in metadata(x)$eda_decompose; for method = "cvxeda" this also records the per-channel solver engine ("qp" or "wiener_fallback") and final QP objective.

Details

Five methods are available:

highpass: FFT-based highpass/lowpass separation at a cutoff frequency.
median: Sliding median filter for tonic extraction.
cda: Continuous Decomposition Analysis (Benedek & Kaernbach, 2010). Deconvolution with a Bateman impulse response, Gaussian smoothing, and a non-negativity constraint on the driver signal; optional tau2 optimisation.
dda: Discrete Decomposition Analysis (Benedek & Kaernbach, 2010): the continuous CDA driver is reduced to a sparse impulse train (one impulse per significant SCR) and reconvolved.
cvxeda: Exact convex optimization decomposition (Greco et al., 2016), solved as a quadratic program: a cubic-knot spline tonic basis and a sparse, non-negative sudomotor driver through a biexponential system, with L1 sparsity on the driver and L2 smoothness on the tonic. Requires the quadprog solver; degrades gracefully to the Wiener approximation if it is unavailable.
cvxeda_fast: Fast Wiener/ADMM approximation of cvxEDA (no solver dependency); the original iterative approximation.

References

Benedek, M., & Kaernbach, C. (2010). "A continuous measure of phasic electrodermal activity." Journal of Neuroscience Methods, 190(1), 80-91. doi:10.1016/j.jneumeth.2010.04.028

Greco, A., et al. (2016). "cvxEDA: A convex optimization approach to electrodermal activity processing." IEEE Transactions on Biomedical Engineering, 63(4), 797-804. doi:10.1109/TBME.2015.2474131