arXiv2026
Multi-aperture reception yields a coherent-combining gain only if the aperture fields are aligned in phase and polarization, which in a downlink must track the atmosphere. We show that the exact gradients of the combined power with respect to the phase and polarization angle of each branch can be measured optically, before photodetection: a passive tap of the combined field, split one-to-$N$, is the common self-reference and, interfered with a branch tap on a 3 dB coupler, yields the phase gradient; with a $90^\circ$ Faraday rotator and a static $π/2$ offset it yields the polarization gradient. Each branch then runs fixed-gain phase and polarization loops, without division, perturbation, dither, reference laser, or per-branch receiver, as analog integrators or a low-rate processor. We derive the gradients, their optical readout, and the loop stability condition. A phase-only simulation with identical disturbances, a common 100 MS/s clock, and retuning of every scheme compares the loop with centralized and distributed stochastic parallel gradient descent and with dither locking at the OSNR where ideal four-beam combining reaches the 28 GBd DP-QPSK FEC threshold. Up to $10^5$ rad/s of multi-tone disturbance the loop keeps the lowest residual loss, at most $5.6\times10^{-4}$ dB for four beams and 13 to 160 times below the alternatives, and reaches $90\%$ efficiency in 0.6 $μ$s for up to 64 beams. The taps leave the receiver OSNR unchanged and every retuned scheme stays within 0.06 dB of ideal combining, whereas brief dither misalignments cost 0.14 dB. Up to $10^7$ rad/s it preserves the ideal-combining bit-error ratio for four, sixteen, and sixty-four beams while every baseline crosses the threshold and slips carrier recovery. In fades it loses $4\times10^{-3}$ dB where the others lose 0.2 to 0.9 dB. Assumptions and limitations are stated; experimental validation remains open.