Channel-Geometry-Driven Optimization for Multi-LED VLC Secrecy with Mobile Eavesdroppers

Indoor visible-light communication (VLC) confines light within a room, yet any photodetector there can eavesdrop. Existing multi-LED physical-layer security studies often assume static eavesdroppers and line-of-sight channels. For a 2×2 ceiling LED array, we model a mobile eavesdropper following the random waypoint model, with wall reflections and, as a sensitivity analysis, human-body blockage. We derive a semi-analytical secrecy outage probability (SOP) in which the per-LED power vector appears explicitly; it agrees with a 106-sample Monte Carlo reference (mean relative error of 0.12% over non-saturated allocations) and, at the default point, with a time-stepped trajectory simulation. Under total-power and per-LED limits, we then optimize the per-LED powers without eavesdropper channel state information (CSI). In the default blockage-free setup, uniform allocation over four LEDs has a higher SOP floor than a single central LED. Relative to uniform allocation, the optimized allocation reduces the SOP by about 86% at the default operating point and the mean SOP by 38% over 10,000 simulated deployment scenarios. Against a simplified artificial-noise (AN) baseline with swept nullspace leakage ξ∈[0,0.2], AN is better at ξ≤0.01 and the proposed allocation at ξ≥0.05, so the crossover lies in (0.01,0.05).