In this paper, we investigate the potential of integrating optical communications within a cellular based V2X (vehicle-to-everything) system, in order to tackle spectrum outage in congested scenarios. We envision a system where small devices, powered via Energy Harvesting (EH) from vehicular communications and acting as simple Road Side Units (RSUs), can relay part of the Radio Frequency (RF) vehicular communications via optical links to suitably deployed Optical Access Points (OAPs). The usage of directional antennas for the RF access links makes it possible to leverage spatial diversity and reuse spectrum resources, while the optical backhaul link grants benefits in terms of energy expense and communication security. The effectiveness of the proposed approach is assessed by means of extensive simulations which utilize realistic traffic patterns at a crossroads: a proper parameters tuning and resource allocation are shown to be able to improve aggregated throughput and spectral efficiency by up to 27% and 30%, respectively.
The integration of terrestrial networks (TN) and non-terrestrial networks (NTN) is emerging as a key enabler for providing seamless connectivity and extended coverage in 6 G systems. In multi-connectivity scenarios, efficient traffic steering across heterogeneous access segments becomes essential to balance network load while satisfying stringent real-time delay requirements. To this end, a delay-aware algorithm is conceived hereby: it dynamically distributes traffic between the two segments according to instantaneous delay conditions, thus enabling adaptive load balancing under time-varying traffic and channel behaviors. Simulation results show significant gains in terms of temporal variability of the steering procedures and stability of traffic allocation between TN and NTN segments, while maintaining effective load balancing during congestion.
This paper develops a marginal-utility framework for bottleneck analysis in the Maximum Concurrent Flow Problem (MCFP), motivated by capacity planning in UAV backhaul networks. To address the resource allocation problem for a hovering UAV formation serving ground users, we tackle the non-uniqueness of optimal dual solutions using directional derivatives to obtain robust sensitivity characterizations. We further give a necessary-and-sufficient condition for nonnegative capacity upgrades and a simpler graph-based sufficient condition. Simulations on an illustrative UAV mesh topology corroborate the analysis and show that the identified criteria distinguish effective from ineffective capacity upgrades, providing a theoretical basis for dimensioning aerial-terrestrial integrated networks.
In end-to-end autonomous driving systems, the separation of trajectory generation from control can produce reference trajectories that are incompatible with the downstream controller and vehicle dynamics, resulting in degraded tracking performance and dynamic feasibility. To address this challenge, we propose the Dynamics-Aware Integrated Trajectory Planning and Control (DIPC) framework, which enables a differentiable interface between explicit planning and control modules. Specifically, the planning component generates reference trajectories based on interactions with surrounding traffic participants and transmits them to the control component. The control component uses a deep-learning-based model predictive control (MPC) solver trained within a physics-informed framework that incorporates vehicle dynamics and stability requirements. To accommodate the different temporal resolutions of planning and control, trajectory interpolation and receding-horizon truncation are introduced as a multi-rate temporal-alignment mechanism. This integrated planning–control training framework leverages gradients backpropagated through the vehicle dynamics model to refine the planner, enabling more dynamics-aware and controller-compatible trajectory generation. Open- and closed-loop evaluations demonstrate improvements in planning quality, tracking performance, and dynamic feasibility over the corresponding modular planning–control baselines, while showing applicability across optimization-based and learning-based planners.
Direct-to-Satellite Internet of Things (DtS-IoT) networks deploy Low Earth Orbit (LEO) nanosatellites as fast-moving gateways for sensors on stationary, mobile, and vehicular platforms, creating intermittent contact windows that fundamentally reshape the uplink design trade-off: should the device transmit blindly and accept a non-zero loss probability, or first detect an overhead satellite via a downlink beacon? The blind strategy wastes transmit energy on uplinks made outside coverage, whereas satellite-detect spends energy listening for beacons. We derive a first-order analytical design rule for event-driven uplink under intermittent LEO coverage, characterising the crossover constellation size $N^*$ below which satellite-detect is preferable and above which blind transmission becomes the energy-efficient choice. The main result is a closed-form threshold, $N^*= \log \alpha \,/\, \log (1-f_{1})$, where $\alpha = \kappa _{\mathrm{eff}}\,T_{b}/(2\tau _{\mathrm{tx}})$ is a dimensionless beacon-overhead parameter and $f_{1}$ is the single-satellite instantaneous coverage fraction. We validate the coverage and waiting-time models against Walker Star propagation simulations (near-polar inclination $86.4^\circ$, globally uniform device distribution), provide a sensitivity analysis of $N^*$ with respect to orbital altitude, elevation mask, and device duty cycle, and extend the baseline model with a lightweight ALOHA collision/retransmission analysis cross-checked by Monte Carlo. For typical LoRa-over-satellite parameters ($h{=}600$ km, $D{=}1\%$, $\kappa {=}0.1$, $\tau _{\mathrm{tx}}{=}1$ s) the threshold falls at $N^*\approx 220$ satellites; symmetric ALOHA loads preserve this threshold, while asymmetric loads (higher blind contention) shift it upward or eliminate it entirely. Applying the model to 3GPP NB-IoT over Non-Terrestrial Networks (IoT-NTN) under a chipset-averaged $\kappa \approx 0.3$ derived from public NB-IoT module datasheets yields $N^*\approx 546$, identifying the constellation size beyond which a hypothetical lightweight-access mode would outperform today's acquisition-mandatory protocol.
