The alphafold 2 uses channels in pairwise representation for protein residues. But why are they used? Shouldn't there be only one value, not a cannel array?

The pairwise representation has shape (r,r,c). Why is it like that? Shouldn't it be (r,r)