Stateful least-mean-squares adaptive filter

Adaptive filters estimate the part of each signal channel linearly predictable from its paired reference. The returned residual is not inherently a clean physiological signal; that interpretation requires a caller-owned reference-noise assumption.

Usage

lmsFilter(n_taps, step_size, leakage = 0, initial_weights = NULL)

Arguments

n_taps

Exact adaptive-filter order.

step_size

Positive LMS step size.

leakage

Weight leakage in [0, 1).

initial_weights

Optional finite n_taps by channel matrix.

Value

A mutable LMSFilter streaming processor.

Examples

# Cancel a reference-correlated component from a short synthetic signal.
set.seed(1)
reference <- sin(2 * pi * 5 * seq_len(200) / 100)
signal <- 0.4 * reference + rnorm(200, sd = 0.05)
filt <- lmsFilter(n_taps = 4L, step_size = 0.05)
result <- update(filt, matrix(signal, ncol = 1),
                 reference = matrix(reference, ncol = 1))
str(result$output)
#>  num [1:200, 1] 0.0923 0.2435 0.2727 0.4295 0.3312 ...
#>  - attr(*, "dimnames")=List of 2
#>   ..$ : NULL
#>   ..$ : chr "channel_1"