Magnitude of the backward-map phase shift in practical encoders

Determine whether practical encoders with real-linear, non-holomorphic backward maps can produce a large phase shift rg c] in the descent condition for scalar channel aging.

Background

The scalar conditional-mean analysis assumes a complex-linear local backward map, but practical encoders may use real parameters or non-holomorphic normalization and therefore induce only a real-linear map. In that setting, the descent criterion depends not only on the channel correlation coefficient ρ\rho, but also on the complex quantity c=∥uP∥2+j⟨uP,vP⟩Rc=\|u_P\|^2+j\langle u_P,v_P\rangle_{\mathbb R}, whose phase shifts the boundary for mean descent.

The paper reports that the tested real-linear example has a very small phase, ∣arg⁡c∣≤0.11∘|\arg c|\le 0.11^\circ, but leaves unresolved whether practical encoder architectures can generate substantially larger values.

References

In our real-linear test $|\arg c|0.11\circ$; whether practical encoders produce large $\arg c$ is open.

— Coherent but Discordant: Rotate, Don't Invert, to Backpropagate Through an Aging Channel  (2610.23137 - Li et al., 16 Oct 2026) in Remark [Non-circular backward maps], Section 3, “The Phase Criterion and Its Scope”