why conj in ILRMA about pyroomacoustics HOT 4 OPEN

In the paper, the rows of W are defined as w^H, so the algorithm is just consistent with this definition.
To see it, you can breakdown line 162 in two steps

w = np.linalg.solve(WV, eyes[:, :, s])
W[:, s, :] = np.conj(w)

If you want to store the non-conjugated demixing filters in W, then you will need to add a conjugate operation around W in all other equation, e.g., (25).

from pyroomacoustics.

OK, got it.
two more questions please:

1. About the scale restoration of the signal. The paper cited this paper "An Approach to Blind Source Separation Based on Temporal Structure of Speech Signals" but I don't see in the paper the projection_back function implemented here(

   pyroomacoustics/pyroomacoustics/bss/common.py

   Line 29 in fa064c2

   def projection_back(Y, ref, clip_up=None, clip_down=None):

   ). Can you direct to the paper of this implementation?
2. About the normalization to prevent ambiguity(

   pyroomacoustics/pyroomacoustics/bss/ilrma.py

   Line 176 in fa064c2

   W[:, :, s] *= lambda_aux[s]

   ). I've encountered another set of normalization equations in the paper titled "Determined Blind Source Separation Unifying Independent Vector Analysis and Nonnegative Matrix Factorization." While I suspect they are identical, my query is regarding their origin—specifically, which paper served as their source?
   Thanks

from pyroomacoustics.