What about the case of zeros or poles with two real solutions?

[menzi11]

I have read your paper and code, and I observed that in operation from (z,p) to (b,a), you always assume that the filter consists of two conjugate complex solutions. However, in reality, the solutions could also be composed of two distinct real solutions. It seems that IIRNet deliberately ignores this form of biquad filter. May I ask what the consideration behind this is?

Thank you.