Finite-alphabet bit allocation for C-RAN fronthaul compression

Develop a finite-alphabet counterpart to reverse waterfilling for allocating fronthaul bits across transformed coefficients while accounting for the saturation of mutual information under finite-alphabet signaling in C-RAN fronthaul compression.

Background

Classical reverse waterfilling solves bit allocation for Gaussian sources, but its unbounded logarithmic mutual-information growth does not reflect finite-alphabet communication systems, where mutual information saturates at the constellation-dependent limit of log₂ M. The paper identifies the absence of a corresponding finite-alphabet bit-allocation rule as an unresolved problem in the context of C-RAN transform-compress-forward architectures.

The paper proposes reverse mercury/waterfilling (RMWF) to address this gap by maximizing a finite-alphabet generalized mutual information objective and incorporating a constellation-dependent mercury level into the allocation rule. Thus, the problem is explicitly identified as open in the introduction but is addressed by the paper itself.