ARFIT2 estimates multivariate autoregressive parameters
of the MVAR process Y

Y(t,:)' = w' + A1*Y(t-1,:)' + ... + Ap*Y(t-p,:)' + x(t,:)'

ARFIT2 uses the Nutall-Strand method (multivariate Burg algorithm)
which provides better estimates the ARFIT [1], and uses the
same arguments. Moreover, ARFIT2 is faster and can deal with
missing values encoded as NaNs.

[w, A, C, sbc, fpe] = arfit2(v, pmin, pmax, selector, no_const)

INPUT:
v		data - each channel in a column
pmin, pmax 	minimum and maximum model order
selector	'fpe' or 'sbc' [default]
no_const	'zero' indicates no bias/offset need to be estimated
in this case is w = [0, 0, ..., 0]';

OUTPUT:
w		mean of innovation noise
A		[A1,A2,...,Ap] MVAR estimates
C		covariance matrix of innovation noise
sbc, fpe	criteria for model order selection

see also: ARFIT, MVAR

REFERENCES:
[1] A. Schloegl, 2006, Comparison of Multivariate Autoregressive Estimators.
Signal processing, p. 2426-9.
[2] T. Schneider and A. Neumaier, 2001.
Algorithm 808: ARFIT-a Matlab package for the estimation of parameters and eigenmodes
of multivariate autoregressive models. ACM-Transactions on Mathematical Software. 27, (Mar.), 58-65.