GeoSetPPO is proposed, a geometry-aware set-attention proximal policy optimization (PPO) method that outputs per-UAM discrete association and subband decisions with permutation-invariant representations that demonstrates stable convergence, higher returns, and favorable reward and schedule-feasibility performance relative to algorithm-based and distance-based schedulers.
Abstract
In this paper, we investigate downlink scheduling for urban air mobility (UAM) in a cooperative space-air-ground integrated network. Multiple ground stations (GSs) employ narrow three-dimensional beams and share spectrum across multiple subbands, while a satellite provides an orthogonal-band service option. Rapidly time-varying geometry and directional interference require joint decisions on base station association, GS subband assignment, and transmit powers. We formulate a finite-horizon mixed discrete-continuous problem that maximizes sum rate while penalizing handovers and GS overload, using only UAM positions and velocities. To address the combinatorial scheduling problem, we propose GeoSetPPO, a geometry-aware set-attention proximal policy optimization (PPO) method that outputs per-UAM discrete association and subband decisions with permutation-invariant representations. Conditioned on each schedule, GS powers are computed by a per-slot successive convex approximation (SCA) module under per-GS power budgets and minimum signal-to-interference-plus-noise ratio (SINR) constraints. To reduce training cost and improve stability, we adopt a two-stage training strategy that transitions reward evaluation from uniform power to SCA-based power allocation. Simulations demonstrate stable convergence, higher returns than multi-layer perceptron (MLP)- and Transformer-based PPO under the considered training setting, and favorable reward and schedule-feasibility performance relative to algorithm-based and distance-based schedulers. In the larger evaluated network, GeoSetPPO also reduces the scheduling latency from 40.84 ms to 2.90 ms relative to the previous algorithm-based method.
IRS has received much attention recently and is envisioned as a revolutionary technology for 6G communication networks. In this paper, we consider a multicast traffic pattern in an IRS-aided single-cell MISO communication system, where an IRS is deployed to assist a multi-antenna AP in transmitting independent data streams to each group of multiple single-antenna users. We formulate and solve a new MMF problem by jointly optimizing the beamforming matrix at the AP and the phase shift matrix at the IRS, subject to both the power constraint and the unit-modulus constraints on the phase shifts. To tackle this highly nonconvex fractional problem, efficient algorithms are proposed based on GFP and AO. In each alternating step, a closed-form suboptimal solution for the transmit beamforming matrix is derived using SCA and LDD. For the phase shift optimization of the IRS, a penalty mechanism is introduced to efficiently handle the nonconvex unit-modulus constraints. Simulation results demonstrate the guaranteed convergence and the superiority of the proposed scheme in improving the minimum weighted SINR at the users, as well as the contribution of the IRS in reducing the transmit power consumption and the number of active transmit antennas at the AP.
The capacity and reliability of low Earth orbit (LEO) mega constellations is restricted by traditional point-to-point relaying due to resource constraints and frequent link interruptions. To address these challenges, we propose a multi-satellite coordinated relaying (MSCR) scheme. The ground station determines when to sample for resilient multi-link broadcasting via the introduced utility of information (UoI). To combat error-and interference-prone downlink, we design a dual-layer error control (DEC) mechanism integrating intra-packet physical-layer network coding (PNC) with inter-packet long erasure coding (LEC). Moreover, the relay satellites adaptively switch between diversity and multiplexing modes based on channel state to further ensure reliability. To maximize the UoI of MSCR, we first derive the block error rate (BLER) bounds under varying policies, and then formulate a joint scheduling problem encompassing the sampling decision at the source, the relaying modes and power allocation at the relays, and the number of LEC packets at both ends. A Lyapunov-guided game-theoretic alternating iterative (GAI) approach is proposed to solve this problem. Simulation results underscore the superiority of our MSCR scheme over state-ofthe-art benchmarks in terms of UoI, delay and throughput.
Jian-Hao Huang, J. Jiao, Qing-Xi Liu et al.· 2026 IEEE/CIC International...· 0 citations
This paper investigates a fluid active reconfigurable intelligent surface (FARIS)-aided downlink non-orthogonal multiple access (NOMA) system. We formulate a network sum-rate maximization problem that jointly optimizes user clustering, NOMA power allocation, FARIS amplification gains, discrete phase shifts, and fluid element selection under quality-of-service, reflected-power, and hardware constraints. To address the resulting nonconvex mixed-integer problem, we develop a two-stage framework comprising distance-based interleaved clustering for constructing successive-interference-cancellation (SIC)-friendly user groups and per-cluster alternating optimization. The resulting subproblems are handled using geometric programming (GP), fractional programming (FP), majorization-minimization (MM) with mixed-integer phase optimization, and the cross-entropy method (CEM). Numerical results demonstrate rapid convergence, near-optimal performance relative to brute-force search (BFS)-based optimum, and consistently outperforms the benchmarks. These results verify the effectiveness of jointly integrating FARIS and NOMA for high-rate downlink transmission.
This paper investigates joint subcarrier and power allocation for a multi-user Orthogonal Frequency Division Multiplexing (OFDM)-based Integrated Sensing and Communication (ISAC) system in Vehicle-to-Everything (V2X) environments. The goal is to maximize a weighted sum of the capped effective radar Signal-to-Noise Ratio (SNR) and aggregate communication rate, under per-user constraints on minimum effective radar SNR, communication rate, and range resolution. The problem is formulated as a Mixed-Integer Nonlinear Programming (MINLP) model. To address its non-convexity, we develop a Block Coordinate Descent–based Joint Resource Allocation (BCD-JRA) algorithm that alternates between nonlinear power allocation and mixed-integer subcarrier assignment and is used as a benchmark in our study. To support real-time V2X operation, we further propose a low-complexity two-stage heuristic, termed Phased Constraint Satisfaction and Greedy Allocation (PSGA). PSGA first allocates the minimum resources needed to satisfy the Quality of Service (QoS) constraints, and then greedily assigns remaining resources based on marginal utility gains while accounting for effective radar SNR capping. The simulation results show that PSGA attains utility close to the BCD-JRA benchmark with millisecond-level latency and satisfies all QoS constraints in the reported experiments.
Jia-Hao Zheng, Xinhao Chen, Linyu Huang et al.· IEEE Transactions on Wireles...· 0 citations
In rapidly expanding low Earth orbit (LEO) satellite networks, the transmission capacity of satellite-to-ground downlinks has become a critical bottleneck for massive data backhaul. In practical systems, the severe supply-demand imbalance between limited gateway antennas and numerous passing satellites makes it impossible to establish links simultaneously, severely limiting the transmission capacity of the downlinks. To address this challenge, we propose a multi-antenna cooperative dynamic scheduling method. First, the antenna-satellite matching strategy is formulated as a mixed-integer nonlinear programming (MINLP) problem aimed at maximizing total system throughput. Second, given the NP-hard nature of this problem, it is further transformed into a Markov decision process (MDP), and a Cooperative Antenna Scheduling Reinforcement Learning (CAS-RL) algorithm is proposed. Finally, simulation results demonstrate that the proposed algorithm increases the average system throughput by more than $\mathbf{1 6}$% compared to benchmarks.
