extricate {DecomposeR}

Extricate a signal: an EEMD algorithm

Description

Performes EEMD

Usage

extricate(
  xy,
  dt,
  nimf,
  ini = NULL,
  repl = 1,
  comb = 100,
  mirror_noise = TRUE,
  factor_noise = 3,
  unit_noise = "1stdiff",
  sifting = 1,
  output_sifting = FALSE,
  remove = "lin.trend",
  bind = FALSE,
  speak = FALSE,
  plot_process = FALSE,
  pdf = TRUE,
  name = "extricate",
  ext = ".pdf",
  dir = tempdir(),
  width = 10,
  height = 20,
  track = TRUE,
  openfile = TRUE
)

Arguments

xy

signal, maybe linearly interpolated to have regular sampling interval

dt

depth/time

nimf

number of modes/components/intrinsic mode functions to decompose the signal into

ini

an optional vector of length n of the eventual initial Intrinsic Mode Function xy would be a demodulation of, if it is a demodulation. In that case the mode indexes will start at 2.

repl

the amount of decompositions to output

comb

the amount of decompositions each output decomposition will be a combination of. Has to be a multiple of 2 (even and odd extension stacks have to be combined in any case)

mirror_noise

whether to generate a mirrored noise signal (for even and odd extension) that will cancel perfectly when combining the decompositions

factor_noise

a factor for the amplitude of white noise (finite amplitude obtained via runif). By default it will be multiplied with the mean of the lagged-one difference to define the noise amplitude

unit_noise

whether to multiply factor_noise by the mean of the lagged-one difference (unit_noise = "1stdiff") or not (unit_noise = "native")

sifting

amount of iterations of the sifting process

output_sifting

whether to output each sifting

remove

whether to remove the linear trend (remove = "lin.trend") or the mean (remove = "mean") prior to decomposition. The removed part will be added back after the decomposition. If remove is anything else, nothing will be removed, which can be problematic for the even and odd extension scheme used.

bind

whether to bind the removed linear trend or mean to the last component (T), or to add it as another component (F)

speak

whether to print a sentence at each sifting: it gives the stack (even or odd), the mode number and sifting number

plot_process

whether to have a plot of the entire sifting process. This slows down the algorithm, use with low 'repl' and 'comb' values for visualisation purposes

pdf

whether the plot be directly set as a pdf file

name, ext, dir, width, height, track, openfile

arguments to provide to pdfDisplay if plot_process and pdf are TRUE

Value

a list made of $xy (original signal), $dt (depth/time), $m (a matrix of the decomposition), $repl (the replicate id of each point) and $mode (the mode id of each point). If output_sifting is TRUE, additional $even_sifting and $odd_sifting data.tables are provided, giving the condensed siftings for the even and odd extensions.

Examples

set.seed(42)

n <- 600
t <- seq_len(n)

p1 <- 30
p2 <- 240

xy <- (1 + 0.6 * sin(t*2*pi/p2)) * sin(t*2*pi/p1)  + 2 * sin(t*2*pi/p2) +
  rnorm(n, sd = 0.5) + t * 0.01

inter_dt <- round(runif(length(xy), min = 0.5, max = 1.5),1)

dt <- cumsum(inter_dt)

dec <- extricate(xy, dt, nimf = 7, repl = 1, comb = 40, factor_noise = 10,
                 sifting = 10, speak = TRUE, output_sifting = TRUE)

integrity(xy, dec)

parsimony(dec)

plot_emd(dec, select = c(4, 6), pdf = FALSE)
## Not run: 
plot_emd(dec, li = list(v = 0), dir = tempdir())
## End(Not run)