How the NARX delay setting and working ?

As indicated by the title,

can NARX be trained like below case ?

y(t) = f{ y(t-1) , x(t-3) }

If it can do this, is that the model like :

d1 = [ 3 ] ;
d2 = [ 1 ] ;
NARX = narxnet(d1,d2,10) ;