Abstract
This paper benchmarks dedicated wireless power transfer (WPT) from FR3 (7.125–24.25 GHz) ultra-massive MIMO base stations to energy-harvesting IoT devices. The devices lie in the radiating near field, and an exact spherical-wave model shows that the Friis link budget holds only with near-field beam focusing, without which 9–25 dB is lost. Study 1 compares policies for scheduling one energy beam across many devices, with an analytical bound on how many devices a beam can sustain, communication quality-of-service metrics and 95% confidence intervals. A naive lowest-battery-first policy is worst at every population size. Under high contention, sharing the beam in proportion to harvesting surplus saves more devices than winner-take-all rules (40.4% versus 25.6% with 120 devices), and a perfect urgency predictor used greedily does no better. A PPO agent trained on device survival matches but does not exceed this surplus-weighted rule, and a DQN agent performs like the greedy rule. Study 2 compares element partitioning, full-aperture power splitting and time switching under an equal total-power constraint: element partitioning loses harvested power quadratically, time switching cannot be both energy-neutral and queue-stable beyond 4.6 m, and full-aperture power splitting meets both requirements up to nearly the 8.5-m break-even distance. A joint evaluation confirms that the aperture-sharing architecture should be chosen before the scheduling policy.