Weighted Non-stationary Averaging

Dynamically performs weighted averaging and SNR estimation under a non-stationary noise.

Given a NxM data matrix where N is the epoch length and M is the number of trials, WNSFMP returns (among other):

w- weight for each trials estimated by dynamically segmenting
the data matrix into regions where noise is locally stationary
ave- the final weight average
snr - the estimate SNR and residual noise levels under weighted
averaging and normal (uniform) averaging.

The function is implemented recursively and through the use of PERSISTENT variables. So that multiple calls to WNSFMP allows it to calculate a running statistics in case loading of entire dataset is not possible

See also: Silva,I, "Estimation of Postaverage SNR from Evoked Responses under Nonstationary Noise", 2009, 56(8)