Intelligent Bandwidth Allocation for Cooperative Perception in Edge–iot Vehicular Systems: Learning What to Share Under Communication Budgets

Cooperative perception lets connected vehicles share intermediate neural features, but vehicle-to-everything (V2X) links carry far less than a modern detector produces, and existing work reports transmitted bytes without their energy cost. Our objective is to decide what to share under a single hard ego-side communication budget. As methodology, we predict, from metadata available before any feature is transmitted, how many objects a candidate block would add to what the ego alone detects, using an ensemble of gradient-boosted trees and a multilayer perceptron, and allocate the budget across helpers, spatial blocks and numerical fidelity (fp16/int8/int4) with a greedy knapsack. Energy and latency are modelled without hardware from measured 5G modem power, a per-operation model calibrated on measured Jetson TX2 data, and replayed 5G driving traces. On the OPV2V benchmark the allocator reaches 99% of full-fusion AP@0.5 while transmitting 20.0 kB per frame instead of 282.9 kB; with int4 fidelity it matches full-fusion accuracy at 10.0 kB (3.5% of the volume), where a per-agent top-k baseline needs 50.0 kB; it improves on that baseline by +0.3 to +0.9 AP points at equal budgets of 10-50 kB; and sharing the helpers' box lists adds two to three points more for under 1 kB. The allocator's own inference costs 0.02% of the total energy. We conclude that, on the radio energy that selection controls, it reaches 97% of full-fusion accuracy at lower energy than the baseline in all 243 parameterisations tested, and in 240 of 243 on total energy, where the helpers' compute dominates.