PaperlayerView exact revision on arXiv

Source-linked AI summary

Cellular-Connected UAV: Uplink Association, Power Control and Interference Coordination

Weidong Mei, Qingqing Wu, Rui Zhang

arXiv:1807.08218v2cs.IT

TL;DR

Cellular-connected UAVs must exploit LoS macro-diversity while controlling the stronger interference they create for ground users. The paper jointly optimizes UAV association and power, proposes centralized SCA and decentralized clustered ICIC, and finds near-optimal performance with practical design trade-offs.

  • Problem

    LoS channels make more BSs visible to UAVs while causing stronger aerial-ground interference, motivating uplink ICIC under spectrum sharing with ground users.

  • Method

    The paper jointly optimizes UAV cell associations and multi-RB power allocations, using centralized SCA and a decentralized BS-cluster design with lower signaling overhead.

  • Results

    Both proposed ICIC designs achieve near-optimal rate performance, with a practically small centralized-decentralized gap and effective interference mitigation.

  • Takeaways & Limitations

    Moderate UAV altitude and tunable directional antennas maximize network throughput and improve the achievable UAV-ground rate trade-off.

Abstract

from arXiv · show

The line-of-sight (LoS) air-to-ground channel brings both opportunities and challenges in cellular-connected unmanned aerial vehicle (UAV) communications. On one hand, the LoS channels make more cellular base stations (BSs) visible to a UAV as compared to the ground users, which leads to a higher macro-diversity gain for UAV-BS communications. On the other hand, they also render the UAV to impose/suffer more severe uplink/downlink interference to/from the BSs, thus requiring more sophisticated inter-cell interference coordination (ICIC) techniques with more BSs involved. In this paper, we consider the uplink transmission from a UAV to cellular BSs, under spectrum sharing with the existing ground users. To investigate the optimal ICIC design and air-ground performance trade-off, we maximize the weighted sum-rate of the UAV and existing ground users by jointly optimizing the UAV's uplink cell associations and power allocations over multiple resource blocks. However, this problem is non-convex and difficult to be solved optimally. We first propose a centralized ICIC design to obtain a locally optimal solution based on the successive convex approximation (SCA) method. As the centralized ICIC requires global information of the network and substantial information exchange among an excessively large number of BSs, we further propose a decentralized ICIC scheme of significantly lower complexity and signaling overhead for implementation, by dividing the cellular BSs into small-size clusters and exploiting the LoS macro-diversity for exchanging information between the UAV and cluster-head BSs only. Numerical results show that the proposed centralized and decentralized ICIC schemes both achieve a near-optimal performance, and draw important design insights based on practical system setups.

I. INTRODUCTION

Cellular-connected UAVs can improve reliability, coverage, and throughput, but their LoS links expose more BSs and create difficult aerial-ground interference-management challenges. This paper formulates uplink ICIC as a joint cell-association and power-allocation problem, then develops centralized and lower-overhead decentralized solutions.

  • Motivation: Cellular-connected UAVs are proposed to improve the limited rate, reliability, security, interference resilience, and VLoS range of point-to-point UAV-ground links.The approach integrates UAVs into cellular networks as aerial UEs served by ground BSs.
  • Challenges: UAV altitude creates coverage challenges because down-tilted BS antennas may serve aerial UEs mainly through weak side-lobes.Existing work analyzes aerial-ground coverage using BS height, antenna pattern, UAV altitude, and association rules.
  • Challenges: LoS UAV-BS channels provide more reliable links and macro-diversity, but expose more BSs and intensify interference with ground-user transmissions.The resulting interference-management problem can involve many BSs across a large ICIC region.
  • Problem formulation: The paper maximizes the weighted sum-rate of the UAV and ground UEs by jointly optimizing UAV cell associations and transmit powers across multiple RBs.The uplink model allows the UAV to access multiple RBs while ground UEs use RBs assigned under a terrestrial ICIC criterion.
  • Proposed designs: The centralized design uses global BS information and SCA to obtain a locally optimal solution, while the decentralized design reduces complexity and signaling through BS clustering.The system includes ground-user SINR and achievable sum-rate calculations under terrestrial inter-cell interference coordination.
  • Scope: The study also considers extensions to multiple UAVs with orthogonal RB allocations and directional UAV antennas, while non-orthogonal multi-UAV allocation remains future work.The baseline model assumes a downward-pointing isotropic UAV antenna and one RB per ground UE.

B. Cellular Network with New UAV User Added

The model jointly optimizes UAV cell associations and power allocations across resource blocks to maximize a weighted sum-rate while managing interference to ground UEs. LoS macro-diversity enables per-resource-block association with the best available BS, while extreme weighting schemes expose UAV-rate and ground-rate trade-offs.

  • Cell association: The UAV may associate with different cells on different RBs, exploiting frequency-flat LoS channel gains and macro-diversity.The serving BS must have an available RB under the terrestrial ICIC constraints.
  • Problem formulation: The UAV jointly selects uplink serving BSs and transmit powers across RBs to maximize the weighted sum-rate of UAV and ground UEs.The formulation assigns weights μu and μg to the UAV rate and ground-UE sum-rate.
  • Cell association: For each RB, the optimal serving BS is the available BS with the maximum UAV channel gain, independently of UAV transmit-power allocations.This reduces the joint design to power allocation after cell association is determined.
  • Extreme weighting schemes: Egoistic water-filling maximizes UAV rate but can cause significant network sum-rate loss through strong interference to ground UEs.This scheme corresponds to μg = 0.
  • Extreme weighting schemes: The altruistic scheme protects ground-UE sum-rate by restricting UAV transmission to unoccupied RBs, but can severely compromise UAV rate under heavy loading.If no RB is globally unoccupied, the UAV is denied network access.

IV. CENTRALIZED ICIC

The centralized design gathers network-wide channel and interference information, then uses SCA to solve the non-convex power-allocation problem through successive convex approximations. The resulting iterations converge monotonically to a locally optimal solution, with power levels reflecting interference costs across BSs.

  • Centralized architecture: A central scheduler collects information from all involved BSs, computes cell associations and power allocations, and informs the serving BSs.This centralized architecture requires network-wide coordination.
  • Successive convex approximation: SCA handles the non-convex objective by replacing it with a concave approximation at each local point and solving successive convex problems.The ground-UE sum-rate term is not concave in UAV power.
  • Successive convex approximation: The approximated objective gives each RB an interference price that is iteratively updated to maximize the network weighted sum-rate.The price represents the cost per unit UAV power caused by co-channel interference.
  • Power allocation: The per-iteration convex problem has a water-filling-like solution whose water levels depend on all UAV-to-BS gains, ground-UE SINRs, and rate weights.Unlike ordinary water-filling, the levels incorporate network-wide interference conditions.
  • Convergence: The SCA objective converges monotonically, with each iteration no worse than the previous one.The centralized protocol stops when the objective improvement falls below a small tolerance.

B. Primal-Dual Based Upper Bound

The primal-dual procedure provides an efficiently computable upper bound for evaluating the locally optimal SCA solution. It decomposes the dualized power problem across RBs, using closed-form or monotonic optimization methods depending on RB occupancy.

  • Upper-bound construction: The dual problem supplies an upper bound on the non-convex primal objective, enabling numerical assessment of SCA performance.The bound is tight if strong duality holds.
  • Centralized implementation: The central scheduler collects channel gains and ground-UE SINRs, determines optimal associations, initializes powers, and iterates the SCA updates.It then sends the final RB assignments and UAV powers to the serving BSs and UAV.
  • Dual decomposition: Dualizing the total-power constraint decomposes the Lagrangian maximization into N parallel RB subproblems.The dual function is convex in its multiplier, which is optimized by bisection.
  • RB subproblems: When an RB has no occupied ground-UE cell, its dual subproblem admits a solution obtained from the first-order derivative.This is the simpler of the two occupancy cases considered.
  • RB subproblems: When an RB has at least one occupied cell, no closed-form solution is available, so monotonic optimization with outer polyblock approximation is used.The subproblem is transformed into maximizing a monotonically increasing function over a normal set.

V. DECENTRALIZED ICIC

The decentralized design targets lower implementation complexity than centralized ICIC by clustering BSs and using UAV LoS macro-diversity to limit information exchange to cluster heads.

  • Motivation: The centralized scheme requires extensive exchange with all involved BSs, causing overhead and delay as the ICIC region grows or changes with UAV movement.This motivates a decentralized design based on BS clustering.
  • Decentralized design: The decentralized approach divides BSs into small clusters and exploits LoS macro-diversity for information exchange between the UAV and cluster-head BSs.The design aims to reduce implementation complexity and signaling overhead.

A. BS Clustering

The decentralized ICIC design clusters BSs under cluster heads, allowing the UAV to coordinate cell association and power allocation using limited exchanged information. A one-round SCA variant further reduces signaling while retaining performance close to iterative SCA.

  • BS clustering: BSs are divided into non-overlapping, intra-connected clusters with one LoS-connected cluster head coordinating each cluster.The cluster head may be selected as the BS with the best UAV channel condition, and clustering is static with uniform cluster size.
  • Information aggregation: The UAV constructs each SCA iteration using cluster-head parameters that summarize local BS information across resource blocks.Cluster heads aggregate local parameters and report the resulting values to the UAV, avoiding direct global information exchange.
  • Decentralized protocol: The decentralized protocol selects a serving cluster head and serving BS for each resource block, then computes and broadcasts the UAV power allocation.The UAV initiates the exchange, cluster heads report local quantities, and selected cluster heads notify their serving BSs to begin uplink transmission.
  • Iterative SCA: Iterative decentralized SCA requires repeated UAV–cluster-head exchanges until convergence and yields a locally optimal solution to (P2).The protocol reports N parameters per cluster head in subsequent iterations and broadcasts updated power allocations.
  • One-round SCA: The one-round SCA starts from zero UAV power, updates the allocation once, and exchanges at most 2MN + 2N parameters.It maximizes a first-order approximation of network sum-rate and achieves performance close to iterative SCA in the reported simulations.

VI. SIMULATION RESULTS

The simulations evaluate centralized and decentralized ICIC in an OFDMA cellular network using 30 resource blocks and a five-tier, 91-cell UAV interference region. Benchmark schemes include egoistic and altruistic UAV transmission strategies.

  • Simulation setup: The evaluation uses an OFDMA system with N = 30 resource blocks and K = 60 active UEs.Each resource block contains 12 consecutive OFDM subcarriers with 15 kHz spacing.
  • Benchmark setting: Under q = 2, terrestrial inter-cell interference attenuates below background noise with high probability in the considered settings.This supports the use of the specified two-tier terrestrial ICIC benchmark in the simulations.
  • Simulation setup: The UAV interference region contains J = 91 cells across five tiers, with maximum UAV transmit power Pmax = 23 dBm.The UAV is positioned at (150 m, 420 m) in the central cell, and UAV-BS channels follow a probabilistic LoS/NLoS UMa model.
  • Benchmarks: The benchmarks compare egoistic transmission, which maximizes the UAV rate, with altruistic transmission, which preserves the ground UEs’ maximum sum-rate.The altruistic scheme restricts UAV transmission to resource blocks unused by ground UEs in all cells.

A. Network Rate Performance versus UAV Transmit Power

The proposed centralized and decentralized ICIC schemes achieve near-optimal network sum-rate while balancing UAV and ground-user performance more effectively than the terrestrial and heuristic benchmarks. Higher UAV rate demands make ICIC more important, and decentralized coordination can conservatively reduce UAV rate.

  • Network sum-rate: The centralized and decentralized ICIC schemes achieve almost the same network sum-rate as the primal-dual upper bound.This indicates near-optimal performance across the evaluated UAV transmit-power range.
  • Benchmark comparison: Terrestrial ICIC and altruistic transmission perform worse because their resource-block restrictions leave fewer opportunities for UAV transmission.The terrestrial benchmark can also degrade network sum-rate at high UAV transmit power, while altruistic transmission performs worst among the considered schemes.
  • UAV rate: The egoistic scheme provides the highest UAV achievable rate, whereas the proposed schemes sacrifice some UAV rate to improve network sum-rate.The decentralized design yields lower UAV rates than centralized ICIC because its approximation exaggerates UAV interference to ground UEs.
  • Ground-user rate: The altruistic scheme gives ground UEs the highest sum-rate, while the egoistic scheme gives them the lowest.The proposed ICIC schemes produce lower ground-UE rate loss than the terrestrial benchmark in the high-power regime.
  • Rate-region trade-off: Increasing Pmax from 13 dBm to 23 dBm enlarges the achievable rate region, while differences between boundaries become larger at high UAV rates.This indicates that ICIC is more crucial when UAV rate demand is high.

B. Network Rate Performance versus Number of Ground UEs

The rate-region analysis shows how ground-user loading, UAV altitude, and antenna beamwidth shape the trade-off between UAV and ground-user rates. Moderate altitude and tunable directional antennas improve the achievable network trade-off.

  • Number of Ground UEs: As the number of ground UEs increases, ground-user sum-rate grows through spatial RB reuse, while the UAV’s achievable rate decreases because average per-RB interference rises.For altruistic access, the UAV receives no rate at K = 140 and 180 because no unoccupied RBs remain.
  • UAV Altitude: Increasing UAV altitude creates a non-trivial rate trade-off because lower altitude improves distance and antenna gains but increases NLoS probability and path-loss exponent.The UAV reaches its maximum rate at the moderate altitude H = 60 m.
  • UAV Altitude: At H = 200 m, the UAV is more likely to fall into nearby BS antenna nulls, requiring more distant serving BSs and increasing association overhead.The UAV is generally associated with 4-6 BSs at high altitude versus at most 2 at low-to-moderate altitude.
  • UAV Altitude: At H = 1.5 m, ground UEs suffer the smallest rate loss at the UAV’s maximum rate, whereas at H = 200 m their rate loss rises with UAV interference.The paper therefore identifies moderate UAV altitude as preferable from the network-rate perspective.
  • UAV Antenna Beamwidth: The beamwidth model uses a downward directional antenna with half-power beamwidths 2Φu and main-lobe coverage radius rc = (H − HB) tan Φu.Increasing beamwidth reduces main-lobe gain but expands the projected coverage region; Φu = 90° is isotropic.
  • UAV Antenna Beamwidth: Widening the UAV antenna beamwidth from Φu = 80° to 85° increases the egoistic UAV maximum rate, while Φu = 85° yields a larger rate region than Φu = 90°.Wider beams increase macro-diversity, and directional antennas provide an additional network-rate design degree of freedom.
  • Overall Findings: The proposed ICIC designs jointly exploit interference mitigation and macro-diversity, achieving near-optimal rate performance with a practically small centralized–decentralized gap.The conclusion reports benefits especially under high UAV transmit power or heavy ground traffic, with moderate altitude and tunable directional antennas further improving rate trade-offs.

APPENDIX SOLUTION TO PROBLEM (18) VIA OPA ALGORITHM

The appendix reformulates problem (18) using slack variables so it becomes optimization of a strictly increasing objective over a normal feasible set. The OPA method then approaches global optimality by iteratively refining polyblock approximations and solving feasibility subproblems by bisection.

  • The appendix derives an upper bound on the optimal solution using monotonicity, establishing pD_n ≤ p̂_n.
  • Problem (18) is equivalently reformulated by introducing slack variables z1 and z2.
  • The reformulated objective is strictly increasing in z, and its feasible region G is a normal set.
  • Because the objective is strictly increasing over a normal set, the problem can be solved with global optimality using the OPA algorithm.
  • OPA iteratively constructs shrinking polyblocks that approximate G with increasing accuracy, starting from a box [0, z(0)].
  • Each iteration selects an optimal vertex, computes its Pareto-boundary intersection through a bisection-based feasibility search, and updates the vertex set.
  • The intersection point is r(q) = (δ(q)z1(q), δ(q)z2(q)), where δ(q) is obtained by updating upper and lower bounds until the feasibility condition is resolved.

Algorithm 3 OPA Algorithm for Solving Problem (18)

Algorithm 3 implements OPA by iteratively selecting the best vertex, finding a Pareto-boundary intersection through bisection, updating the vertex set, and stopping at ε-accuracy.

  • Algorithm 3 initializes q = 1 and the vertex set Z(1) = {z(0)}, then repeats iterations until ε-accuracy is reached.
  • Each iteration selects z(q) = arg max z∈Z(q) z1z2 as the optimal vertex in the current vertex set.
  • The algorithm computes δ(q) and r(q) by solving feasibility problem (30) with bisection search.
  • If U(z(q)) − U(r̃(q)) ≤ ε, the current r̃(q) is accepted as an ε-optimal solution; otherwise, the vertex set is updated.
  • After each nonterminal iteration, q is incremented and the loop continues until the stopping condition is satisfied.
Loading 1807.08218v2…