Not sure what you have in mind. Can you write the equation whose coefficients you are trying to estimate using the recursive polynomial block?
A hint: Suppose the equation is:
y(t) = a1*y(t-1) + a2*y(t-2) + b0*y(t) + b1*u(t-1) + b2*u(t-2) + b3*u(t-2)
that relates the output y(t) to the input u(t). Here, the the unknowns are z1, z2, b0, b1, b2, b3. However, the number of inputs is still only one (i.e., u(t)), since the notion of "inputs" is in the physical sense. So when you use a recursive polynomial block, you use a scalar signal for input (u(t)) and another scalar signal for the output (y(t)). In the example above the number of lags in the output is 2 and the number of lags in the input is 3. The input to output delay (nk) is 0 since the term u(t) appears directly in the equation.
So if you are using:
- An ARX structure (A(q) y(t) = B(q) u(t) + e(t)) , you would set na = 2, nb = 3, nk = 0.
- An OE structure (y(t) = B(q)/F(q) u(t) + e(t)), you would set nf = 2, nb = 3, nk = 0.
(Bottomline: do not confuse the number of terms in the equation with the dimension of the input signal).
