PaperlayerView exact revision on arXiv

Source-linked AI summary

UAV Communications Based on Non-Orthogonal Multiple Access

Yuanwei Liu, Zhijin Qin, Yunlong Cai, Yue Gao, Geoffrey Ye Li, Arumugam Nallanathan

arXiv:1809.05767v1eess.SPcs.IT

TL;DR

The paper addresses how to realize UAV networks with massive connectivity using NOMA while handling spatial randomness, trajectory and power decisions, and dynamic UAV placement. It develops stochastic-geometry models, joint power-allocation and trajectory design, and a machine-learning framework for dynamic deployment. Across these case studies, it provides a framework spanning analytical modeling through practical UAV placement and movement design.

  • Problem

    The paper addresses the design challenges of integrating NOMA into UAV networks to provide massive connectivity, including spatial modeling, resource and trajectory decisions, and dynamic placement.

  • Method

    The paper combines stochastic-geometry modeling, joint power allocation and trajectory design, and machine-learning-based dynamic UAV placement and movement.

  • Results

    The case studies provide NOMA-aided UAV network analysis for single- and multiple-UAV settings, show NOMA outperforming OMA in max-min rate, and improve UAV positioning accuracy through machine learning.

  • Takeaways & Limitations

    The framework supports understanding NOMA-enabled UAV systems from fundamental stochastic modeling to resource optimization and dynamic three-dimensional deployment.

Abstract

from arXiv · show

This article proposes a novel framework for unmaned aerial vehicle (UAV) networks with massive access capability supported by non-orthogonal multiple access (NOMA). In order to better understand NOMA enabled UAV networks, three case studies are carried out. We first provide performance evaluation of NOMA enabled UAV networks by adopting stochastic geometry to model the positions of UAVs and ground users. Then we investigate the joint trajectory design and power allocation for static NOMA users based on a simplified two-dimensional (2D) model that UAV is flying around at fixed height. As a further advance, we demonstrate the UAV placement issue with the aid of machine learning techniques when the ground users are roaming and the UAVs are capable of adjusting their positions in three-dimensions (3D) accordingly. With these case studies, we can comprehensively understand the UAV systems from fundamental theory to practical implementation.

I. INTRODUCTION

The article develops a NOMA-enabled UAV network framework spanning mathematical modeling, resource and trajectory optimization, and aerial base-station placement and movement. It addresses challenges arising from UAV-specific channel characteristics, user mobility, and dynamic three-dimensional network design.

  • Motivation: NOMA is proposed for UAV networks because it can support higher spectral efficiency and massive connectivity.NOMA multiplexes users on shared resource blocks using different power levels, with superposition coding and successive interference cancellation.
  • Contribution: The article’s goal is a potential NOMA-based UAV network solution spanning mathematical modeling, joint resource and trajectory optimization, and aerial base-station placement and movement.These topics connect network analysis with practical design of UAV positioning and motion.
  • Challenges: Applying NOMA to UAV networks requires redesigning user association because UAV-specific channel characteristics differ from conventional terrestrial NOMA settings.The article identifies user association as a research challenge for NOMA-enabled UAV networks.
  • Challenges: Dynamic user grouping and flexible UAV positioning motivate more effective three-dimensional aerial base-station placement and movement algorithms.The design must account for roaming users and interference mitigation.
  • Article roadmap: The article studies stochastic-geometry modeling, joint power allocation and trajectory design, and machine-learning-based dynamic aerial base-station movement.These case studies progress from spatial analysis to optimization and adaptive deployment.

II. NON-ORTHOGONAL MULTIPLE ACCESS AIDED UAV NETWORKS

NOMA-aided UAV networks exploit UAV agility and shared-resource transmission but require designs tailored to aerial channels, mobility, interference, and dynamic deployment. The section identifies stochastic modeling, joint trajectory and resource optimization, and machine-learning-based movement as corresponding design directions.

  • UAV characteristics: UAV agility enables rapid deployment and flexible three-dimensional positioning for on-demand service to ground users.The article contrasts this flexibility with terrestrial base stations and associates it with lower costs.
  • NOMA operation: Power-domain NOMA supports massive connectivity by allowing different users to share the same time/frequency slot with different power levels.The downlink uses superposition coding and receiver-side successive interference cancellation.
  • NOMA operation: Because UAV-ground links are mainly line-of-sight, NOMA users may lack distinct channel-gain differences, requiring pairing and power designs based on large-scale fading.This differs from the channel separation commonly leveraged in conventional terrestrial NOMA.
  • NOMA operation: In a two-user example, the poorer-channel user is decoded first with higher transmit power, while successive interference cancellation then facilitates detection of the better-channel user.The first-decoded user requires sufficient power because its interference and detection error can affect the other user.
  • Design issues: Stochastic geometry is proposed to quantify NOMA-aided UAV networks with line-of-sight and Nakagami-m fading before network design and implementation.The approach addresses random spatial modeling and user grouping or association.
  • Design issues: The article proposes joint resource allocation and trajectory design for line-of-sight links, and a machine-learning framework for dynamic multi-UAV placement and movement.These directions target coverage, power allocation, roaming users, and complex environments involving line-of-sight and non-line-of-sight links.

III. NOMA-AIDED UAV NETWORKS

This section motivates stochastic geometry for analyzing average NOMA-aided UAV network performance under spatial randomness. It introduces a two-tier model with fixed-altitude UAVs and Nakagami-m small-scale fading to represent predominantly line-of-sight channels.

  • Modeling motivation: Stochastic geometry is used to capture the spatial randomness of wireless networks and obtain insights for NOMA-aided UAV network design.The section frames average-performance analysis as a step before implementation.
  • Spatial model: The proposed framework models randomly roaming NOMA users served by UAVs at a fixed altitude using a two-tier spatial model.The spatial model can be extended to UAVs flying at varied altitudes.
  • Channel model: Nakagami-m fading is applied to model small-scale fading because dominant line-of-sight UAV links make conventional Rayleigh fading improper for these channel characteristics.The modeling choice follows the aerial propagation setting described in the section.

A. Single-UAV case

The single-UAV case models user pairing within a disc-shaped coverage area and compares pairing strategies for NOMA users. Pairing the nearest cell-center user with the farthest cell-edge user provides the largest NOMA gain over OMA, but optimizing single pairs does not ensure multi-user network improvement.

  • System model: The single-UAV cell is modeled as a disc D, with the UAV at its center and 2M users divided into M orthogonal pairs.Each pair is randomly allocated to an orthogonal resource block, and the disc is divided into cell-center and cell-edge regions.
  • User pairing: Cell-center users are close to the connected UAV, whereas cell-edge users are farther away.
  • User pairing: Randomly pairing a cell-center user with a cell-edge user is the benchmark strategy and does not require full CSI of all users.
  • User pairing: Pairing the nearest cell-center user with the nearest cell-edge user has the best system performance among the considered strategies.
  • User pairing: Pairing the nearest cell-center user with the farthest cell-edge user provides the largest NOMA gain over conventional OMA.
  • Scope: Single-pair optimization does not necessarily improve multi-user NOMA-aided UAV networks, so resource management across resource blocks requires further investigation.

B. Multiple-UAV case

The multiple-UAV case uses stochastic geometry and considers flexible user association under heterogeneous UAV features. Distance-based association is simple but inflexible, while SINR-based association requires instantaneous CSI and incurs heavy overhead.

  • Motivation: Multiple-UAV NOMA networks are more challenging to analyze than the single-UAV case because UAV and user distributions are independent and numerous.
  • System model: UAVs and NOMA users are modeled as independent homogeneous Poisson point processes on planes, with UAV altitude h and densities λ_v and λ_u.Each UAV is assumed to communicate with K NOMA users.
  • User association: The simplest association policy assigns each UAV its K nearest NOMA users, but this is not flexible when UAV features differ.Relevant differing features include transmit power and multiple-antenna beamforming gain.
  • User association: Average-received-power association is more flexible, but each user's average received power depends on the power allocation scheme.Consequently, power allocation among NOMA users is important for network performance.
  • User association: Maximum-instantaneous-SINR association accounts for UAV mobility and agility but requires instantaneous CSI at both UAVs and NOMA users, causing heavy overhead.

IV. JOINT POWER ALLOCATION AND TRAJECTORY DESIGN

The joint-design case studies power allocation and UAV trajectory for static ground users in a fixed-height flight model. NOMA consistently outperforms OMA in max-min rate, with further gains from carefully designed trajectories and user scheduling.

  • Model and objective: A UAV flies from initial point A to destination B at constant height during flying time T while serving ground users in downlink.
  • NOMA transmission: Power-domain NOMA simultaneously serves multiple ground users on the same channel using superimposed signals and SIC ordered by channel gains.
  • Model and objective: The design jointly optimizes transmit power and UAV trajectory to maximize the minimum average rate under speed, power, endpoint, and SIC-order constraints.
  • Results: NOMA-aided UAV networks always outperform OMA-aided networks in max-min rate across different flight durations T.
  • Results: Carefully designed trajectory and user scheduling further improve NOMA performance.
  • Results: At T = 25 seconds, NOMA and OMA produce different trajectories: OMA approaches users successively, whereas NOMA slows near U2 and allocates more power to remote users.

V. DYNAMIC UAV PLACEMENT AND MOVEMENT DESIGN: A MACHINE LEARNING APPROACH

The paper addresses dynamic UAV placement and movement in 3D NOMA networks, where roaming users, multiple UAVs, and altitude-dependent channel trade-offs complicate trajectory design. It proposes a three-step machine-learning framework combining clustering, Q-learning-based placement, and Q-learning-based movement.

  • Motivation: Lower UAV altitudes reduce path loss but also reduce line-of-sight probability, creating a placement and trajectory trade-off.This makes altitude selection part of the dynamic design problem.
  • Dynamic access and movement: Dense user areas can be partitioned into clusters, applying NOMA within clusters and TDMA among clusters while UAVs move across clusters.Altitude can also be adjusted when users require high data rates, such as for high-dimension video transfers.
  • Motivation: Roaming users and varying NOMA group sizes require UAV positions to be adjusted in real time.The framework targets dynamic environments rather than fixed user deployments.
  • Three-step framework: The framework uses K-means to initialize user-cell positioning and NOMA group sizes, followed by Q-learning for 3D UAV placement and real-time movement.Placement treats users as static, whereas movement follows users modeled by a random walk.
  • Q-learning operation: Q-learning represents UAV coordinates as states, selects movement directions as actions, and updates decisions using rewards and Q-values.The state-action-reward cycle repeats to adjust UAV placement and movement, with training and testing supporting offline policy decisions.

VI. FUTURE CHALLENGES AND CONCLUSION REMARKS

The article presents NOMA-enabled UAV networks through stochastic-geometry analysis, joint power allocation and trajectory design, and machine-learning-based 3D movement design. It also identifies open challenges involving unified spatial modeling, MIMO-NOMA, latency, and practical implementation.

  • Conclusion remarks: The framework models and analyzes NOMA-aided UAV networks for both single-UAV and multiple-UAV cases using stochastic geometry.The analysis addresses network performance before subsequent optimization and deployment designs.
  • Conclusion remarks: Joint power allocation and trajectory design are introduced for the NOMA-aided single-UAV case.This addresses resource allocation and movement design together for line-of-sight links.
  • Conclusion remarks: A machine-learning framework addresses dynamic UAV placement and movement in 3D space.The data-driven design is intended to adjust UAV positions more accurately as user mobility changes.
  • Future challenges: A unified spatial model remains needed across diverse NOMA-aided UAV communication scenarios.The desired model should be readily adaptable to different practical applications.
  • Future challenges: Future work includes MIMO-NOMA channel ordering and beamforming or clustering designs that account for UAVs’ three-dimensional characteristics.Multiple-antenna NOMA uses vector or matrix channels, making ordering more difficult than in the single-antenna case.
  • Future challenges: Large NOMA user groups can create receiver delays through successive-interference-cancellation decoding, motivating hybrid multiple access.Hybrid access divides users into orthogonal groups, with a small number using NOMA within each group.
Loading 1809.05767v1…