Exponentially weighted online Welch spectrum

Complete windows are converted to one-sided power spectral density and combined as alpha * newest + (1 - alpha) * previous.

Usage

welchOnline(
  sampling_rate,
  window_samples,
  hop_samples = floor(window_samples/2),
  n_fft = window_samples,
  alpha = 0.1,
  window = c("hann", "hamming", "rectangular"),
  detrend = c("mean", "none"),
  bands = NULL,
  causal_filter = NULL,
  n_features = NULL
)

Arguments

sampling_rate

Positive samples per second.

window_samples

Exact frame length.

hop_samples

Exact frame advance, no larger than the window.

n_fft

Exact FFT length, at least the frame length.

alpha

Weight assigned to the newest periodogram.

window

Exact governed window definition.

detrend

Per-frame mean removal or no detrending.

bands

Optional uniquely named two-column Hz matrix.

causal_filter

Optional independent PhysioPreprocess::StreamFilter objects or factory.

n_features

Optional channel count to bind at construction.

Value

A mutable OnlineWelch streaming processor.

Examples

sr <- 100
signal <- sin(2 * pi * 10 * seq_len(256) / sr)
welch <- welchOnline(sr, window_samples = 64L, hop_samples = 32L)
result <- update(welch, signal)
dim(result$output)
#> [1]  7 33  1