Low-Altitude Economic Networks (LAENets)

• LAENets are distributed network systems in low-altitude airspace that integrate UAVs, eVTOLs, terrestrial base stations, and space assets for diverse operational services.
• They leverage 3D placements, AI-driven orchestration, and digital twin management to enhance connectivity, sensing, and control across dynamic environments.
• LAENets support applications from logistics and emergency response to urban air mobility by combining communication, computation, and regulatory oversight into a unified service ecosystem.

Low-Altitude Economic Networks (LAENets) are distributed air–ground, and increasingly air–ground–space, network systems built to support economic and operational activity in low-altitude airspace. Across the literature, the term denotes integrated infrastructures that couple unmanned aerial vehicles (UAVs), eVTOL aircraft, terrestrial base stations, edge/cloud resources, and regulatory control functions for applications such as logistics, infrastructure inspection, emergency response, sensing, and aerial mobility. The altitude range is not fixed across papers: some works define LAENets below $1{,}000$ m, others below $3{,}000$ m, and some resource-control studies focus on environments below approximately $300$ m, reflecting distinct operational scopes rather than a single canonical boundary (Cai et al., 27 May 2025, Tekbıyık et al., 2 Jun 2025, Wang et al., 30 Apr 2025).

1. Definition, scope, and conceptual boundaries

LAENets are consistently described as more than conventional UAV communication networks. They integrate communication, sensing, computation, positioning, navigation, surveillance, flight control, and airspace management into a common low-altitude service system. In one formulation, a LAENet is “a distributed, multi-layer wireless system employing aerial platforms (at altitudes below 3 000 m) together with terrestrial infrastructure to support economic, social, and operational services,” while other works define it as a distributed wireless network of low-altitude UAVs “cooperating to provide transportation, communication, computation, and sensing services below 1 000 m” (Tekbıyık et al., 2 Jun 2025, Zhao et al., 25 Feb 2025).

A central distinction in the literature is between LAENets and earlier 2D terrestrial or purely non-terrestrial systems. Relative to conventional terrestrial cellular networks, LAENets exploit 3D placement, dominant line-of-sight links, and mission-driven mobility. Relative to pure non-terrestrial networks such as LEO constellations, they emphasize dense, highly mobile aerial nodes operating close to ground users and physical assets, often under stringent latency, safety, or regulation constraints (Jiang et al., 2023).

The service model is also broader than connectivity alone. LAENets are described as service-centric 6G ecosystems in which UAV fleets function as aerial base stations, relays, sensors, edge servers, and mission platforms. This service orientation appears in formulations centered on user hotspots, dynamic task allocation, multi-service deployment, computility sharing, and airspace assurance rather than only packet transport (Yang et al., 5 Mar 2026, Luo et al., 25 Aug 2025).

A recurrent misconception is that LAENets are synonymous with low-altitude drone access links. The surveyed architectures instead treat them as cyber-physical infrastructures in which networking, flight operations, and institutional oversight are tightly coupled. This is especially explicit in works that embed supervision systems, digital-twin airspace management, or control-plane separation directly into the architecture (Wang et al., 30 Apr 2025, Luo et al., 10 Jun 2025).

2. Architectural strata and infrastructural variants

The architectural literature presents several compatible stackings of LAENets, each emphasizing different operational layers.

Framework	Layers or tiers	Primary role
HAP-enabled LAENet	Global / Regional / Local	LEO backbone, HAP orchestration, edge-ground execution
AI-driven LAENet	Device / Edge / Control	UAVs and sensors, ground relays/servers, AI orchestration
Core LAENet architecture	Airborne Infrastructure / Digital Airspace / Service Assurance	vehicles and vertiports, computable airspace, supervision and service systems
UTICN	Sensing / Positioning / Communication / Management / Service	cellular-native low-altitude operation

In the HAP-centered formulation, the Global Tier is composed of LEO satellites at $500$–$2{,}000$ km serving as a global backbone for wide-area connectivity, GNSS augmentation, and intercontinental data aggregation. The Regional Tier places HAPs at approximately $20$ km as “regional orchestrators” with 200–500 km coverage radius, millisecond-scale round-trip latency control loops, edge computing and caching, integrated sensing–communication, and real-time UTM functions. The Local Tier contains terrestrial base stations, IoT gateways, ground computing nodes, and UAVs acting as both sensors and users (Huang et al., 23 Feb 2026).

A different but complementary AI-driven architecture uses a Device Tier, an Edge Tier, and a Control Tier. Here UAVs and sensors interact with ground stations, mobile relays, and edge-cloud servers, while federated learning, multi-agent reinforcement learning, and SDN/SON-like orchestration are placed in the control layer (Tekbıyık et al., 2 Jun 2025).

The more expansive core-architecture treatment divides LAENets into Layer 1 – Airborne Terminals & Physical Infrastructure, Layer 2 – Intelligent Collaboration & Digital Airspace, and Layer 3 – Multi-Party Coordination & Service Assurance. This stack includes UAVs, eVTOL aircraft, rooftop vertiports, charging or battery-swap stations, mobile repair-and-refuel units, a digital-twin airspace management system, a low-altitude supervision system for regulators, and a flight service system for operators (Wang et al., 30 Apr 2025).

UTICN extends this logic into a cellular-native management architecture with vertically coupled layers for sensing, positioning, communication and control-data decoupling, management/control, and service orchestration. It introduces dedicated UAV control-plane entities such as UARF, UAGF, UACF, UAMF, and UANF, and explicitly separates regulatory and control traffic from mission data (Luo et al., 10 Jun 2025).

Several papers further broaden the infrastructure boundary of LAENets. HAPs are treated as regional intelligence hubs rather than mere relays (Huang et al., 23 Feb 2026); LEO satellites are incorporated as communication, navigation, sensing, and computation assistants (He et al., 7 May 2025); O-RAN deployments add Non-RT and Near-RT RIC loops for mission-aware control (Abdalla et al., 1 Jan 2026); and blockchain-based “Low-Altitude Computility Networks” treat airborne compute resources as tokenized real-world assets coordinated through hybrid permissioned/permissionless chains (Luo et al., 25 Aug 2025). This suggests that LAENets are best understood not as a fixed topology but as a family of interoperable low-altitude infrastructures with varying degrees of space integration, network softwarization, and economic coordination.

3. Core technical primitives and formal models

At the physical and link levels, LAENet studies formalize wide-area aerial coverage, low-latency control, resource-constrained offloading, and predictive cross-layer adaptation. In the HAP model, the coverage footprint of one HAP with half-beam angle $\theta$ and altitude $h$ has horizon-limited radius

$R = h \cdot \tan \theta,$

and spherical-cap area

$A = 2\pi R^2(1-\cos\theta).$

For horizontal separation $3{,}000$0, slant distance $3{,}000$1, and propagation speed $3{,}000$2 m/s, the round-trip latency is approximated by

$3{,}000$3

The example $3{,}000$4 km and $3{,}000$5 km yields $3{,}000$6 km and $3{,}000$7 ms, which is the basis for claims of millisecond-scale regional control loops (Huang et al., 23 Feb 2026).

Communication and offloading models are equally explicit. The same HAP formulation gives the free-space path loss

$3{,}000$8

received power

$3{,}000$9

and Shannon capacity

$300$0

while computation offloading time is

$300$1

subject to cache and latency constraints such as $300$2 and $300$3 (Huang et al., 23 Feb 2026).

At the control level, Cai et al. formulate LAENet management as a Markov Decision Process

$300$4

with state

$300$5

action

$300$6

position and energy transitions

$300$7

and explicit resource constraints $300$8, $300$9, and $500$0 (Cai et al., 27 May 2025).

Resource-control works increasingly add semantic state abstractions. In the vision-aided ISAC framework, a masked De-Diffusion model maps raw image $500$1 to a structured textual token

$500$2

encoding only agent type, activity class, and heading orientation. Fused with radar-derived range, radial velocity, and azimuth, these tokens define a semantic risk heatmap

$500$3

which then guides radio access technology selection, beamforming, and power control under user-specific QoS constraints (Gao et al., 2 Jul 2025).

A further step is predictive communications. There, long-horizon strategic routing, mid-horizon tactical timing, and short-horizon operational power control are aligned with predictable trajectories and radio maps. The strategic layer chooses path $500$4, tactical control schedules hop transmission times $500$5, and operational control solves a convex power-allocation problem over $500$6 to minimize predicted ground interference while satisfying $500$7 (Chen et al., 1 Sep 2025). A plausible implication is that LAENet modeling is shifting from purely reactive radio optimization toward foresight-driven control in which mobility predictability is treated as a primary network resource.

4. Optimization, learning, and orchestration paradigms

A major strand of LAENet research centers on AI-driven orchestration under nonconvexity, uncertainty, and multi-objective coupling. Cai et al. identify three core obstacles—complex decision-making, resource constraints, and environmental uncertainty—and propose an LLM-enhanced RL framework in which the LLM serves as information processor, reward designer, decision-maker, and generator. In a case study for UAV-assisted IoT data collection over a $500$8 m $500$9 $2{,}000$0 m marine area with 10 randomly placed terminals, TD3 with GPT-4o-designed reward converges in approximately 150 episodes and reaches a final total energy 7.2% lower than TD3 with manual reward, while DDPG shows an approximately 5% improvement (Cai et al., 27 May 2025).

Generative models appear in several other optimization roles. For robust physical-layer design, a diffusion-based actor with a Mixture-of-Expert Transformer improves worst-case secrecy-rate optimization; in the reported secure beamforming study, the proposed method reaches $2{,}000$1 b/s/Hz at 200 epochs versus approximately $2{,}000$2 for diffusion RL without the MoE-Transformer and approximately $2{,}000$3 for SAC, corresponding to $2{,}000$4 versus GDM and $2{,}000$5 versus SAC (Zhao et al., 25 Feb 2025). For temporal spectrum cartography, RecMAE reconstructs RF power maps from sparse measurements, and MADP then optimizes multi-UAV sensing trajectories; under 10% sensing coverage, RecMAE reports MSE $2{,}000$6 versus $2{,}000$7 for AE and $2{,}000$8 for Kriging, while the full planning stack reduces cumulative reconstruction error by 57.35% and 88.68% relative to the cited interpolation and deep learning baselines (Zhao et al., 21 May 2025).

Distributed learning and competition-aware coordination are also prominent. In a multi-service deployment setting with $2{,}000$9 service providers, $20$0 services, and $20$1 hotspots, the DAPCR-FedPG method combines an authenticity-guaranteed auction with resilient federated policy gradient and Byzantine filtering. The associated game is shown to admit a Nash equilibrium via potential-game analysis, while simulations with two Byzantine providers show smooth loss toward zero, reward toward $20$2, and robustness under 0–33% Byzantine fractions (Yang et al., 5 Mar 2026).

Other works couple model formulation and optimization more explicitly. HybridRAG combines KeywordRAG, VectorRAG, and GraphRAG to generate optimization problems for low-carbon multi-UAV-assisted MEC networks, then solves them with $20$3DSAC. In the reported evaluation, the method improves precision from 44.6% to 46.2%, recall from 74.4% to 76.5%, and F1-score from 49.9% to 53.2% on RAGChecker, while the resulting control policy yields approximately 64% higher test reward than SAC and records training-time carbon of about 70.3 g CO$20$4 and per-inference carbon of about 0.025 g CO$20$5 (Wen et al., 19 Jun 2025).

Digital twins and hierarchical optimization provide a more engineering-centered orchestration path. DT-MOO constructs coupled spectrum, traffic, and interference twins for low-altitude communication networks and scores candidate beam settings by their combined effect on coverage, interference, handover, and sensing. In a real 6-cell, 5G testbed, it increases RSRP coverage $20$6 from 14.0% to 52.9% and SINR coverage $20$7 from 76.6% to 80.3% (Huang et al., 20 Apr 2026). For onboard vision-language inference, ARPO-LLaRA decomposes optimization into resolution/power control and LLM-augmented PPO trajectory design, achieving approximately 45% latency reduction versus random policy, approximately 25% versus ARPO-GH, and approximately 13.7% versus ARPO-PPO (Li et al., 11 Oct 2025).

5. Security, trust, regulation, and covert semantics

Security research treats LAENets as exposed physical-layer systems whose attack surface is enlarged by broadcast channels, mobility, and the safety-critical coupling of communication to flight operations. The secure-communications survey organizes the problem into confidentiality, availability, and integrity, emphasizing anti-eavesdropping, authentication, anti-jamming, spoofing defense, anomaly detection, and injection protection (Cai et al., 12 Apr 2025).

A recurrent misconception is that LAENet security is exhausted by cryptographic protection of data flows. The cited works instead combine communication security with sensing, tracking, and regulatory enforcement. In the HAP-assisted framework, for example, joint waveform design supports continuous 3D tracking of UAV trajectories, GNSS fixes are cross-checked with HAP ISAC measurements for spoofing and jamming detection, and each UAV may be assigned a probabilistic trust score $20$8. If

$20$9

the system raises a Level-I warning, while geofencing, conflict detection, and spectrum/altitude deconfliction are enforced through UTM functions (Huang et al., 23 Feb 2026).

The survey also highlights two standalone GPS-spoofing defenses. Ott et al.’s SPREE uses Auxiliary-Peak Tracking on a single-antenna GPS front end and constrains a seamless-takeover attacker with power advantage up to approximately 15 dB to no more than about 1 km undetected position shift. Sathaye and Ranganathan’s SemperFi combines an Adversarial Peak Identifier with a Legitimate Signal Retriever and serial interference cancellation; in the reported evaluation it achieves 100% spoofing detection and recovers a sub-100 m position fix for power advantages up to 15 dB (Cai et al., 12 Apr 2025).

A distinct security and control problem arises in covert communication. In covert-semantic LAENets, UAVs must deliver task-related information under low-probability-of-detection constraints monitored by a passive warden Willie, with $\theta$0 ideally no greater than 0.5. The proposed response is semantic communication with controllable semantic abstraction $\theta$1, where

$\theta$2

or, more generally,

$\theta$3

The ensuing contract-theoretic model addresses information asymmetry between base station and UAVs, uses Prospect Theory to capture subjective BS utility, and applies RDSAC for optimal contract design. Reported simulations with $\theta$4 UAVs and $\theta$5 types show RDSAC achieving approximately 3.41% higher average reward than vanilla SAC and approximately 31.44% higher than PPO, with near-zero standard deviation in test reward across diverse channel SNRs and type distributions (Liu et al., 2 Mar 2026).

Regulatory architectures generalize these ideas into operational control. UTICN specifies identity and credential management via UARF/UACF, electromagnetic-signature fingerprinting, jamming-resilient A2X with frequency hopping and spread spectrum, and control-plane hardening with end-to-end encryption such as AES-128 for regulatory commands (Luo et al., 10 Jun 2025). In this body of work, trust is not a separate overlay but an intrinsic property of sensing, positioning, networking, and airspace governance.

6. Applications, empirical evidence, and research trajectories

The application literature covers disaster response, swarm coordination, passenger eVTOL service, logistics, emergency rescue, infrastructure inspection, and urban air mobility. HAP-enabled narrative scenarios include a disaster-response flow in which a LEO constellation informs a HAP, the HAP forms a C-band cell over a disaster zone, UAVs send high-resolution imagery via HAP backhaul to the cloud, and ground teams receive processed overlays by HAP broadcast; a swarm-coordination flow in which cloud planning is refined at the HAP into grid partitions and channel assignments; and an eVTOL service flow in which a flight plan is relayed between the eVTOL, the HAP, and a regional air-traffic controller while the HAP simultaneously monitors position for conflict warnings (Huang et al., 23 Feb 2026).

The empirical evidence is substantial and heterogeneous. In the HAP roadmap, adding HAP offloading enlarges the feasible $\theta$6 task region by 2–3×, and HAP-coordinated spectrum assignment produces linear throughput scaling up to $\theta$7 UAVs, whereas uncoordinated throughput plateaus or degrades for $\theta$8 (Huang et al., 23 Feb 2026). In the RWA-based computility architecture, the urban logistics LACNet simulation reports that RWA-LACNet maintains nearly linear latency growth under rising load, degrades least as malicious-node ratio increases, and achieves the highest resource utilization, exceeding 85% (Luo et al., 25 Aug 2025). In DT-MOO, measured high-quality coverage rises from 14.0% to 52.9% across evaluated altitudes (Huang et al., 20 Apr 2026).

Privacy-preserving semantic control also shows strong quantitative results. DeDiff-VARARO reaches an average reward within 4% of the raw-image upper bound while preserving privacy by transmitting only compact semantic tokens rather than raw video (Gao et al., 2 Jul 2025). Rotatable-antenna deployments improve coverage probability below 100 m altitude from 85% to 95%, improve SCNR by 5–8 dB over 10–30 dBm transmit power, and deliver up to 25% sum-throughput gain under optimized boresight scheduling (Li et al., 1 Nov 2025). Predictive communications yields an order-of-magnitude reduction in cross-tier interference, with representative results showing about 12 dB improvement at offered loads of 20 and 30 flows (Chen et al., 1 Sep 2025).

Space-assisted extensions are similarly consequential for remote or sparse environments. In the distributed satellite-MIMO architecture, cooperative satellite reception yields up to 3× higher sum-rate than single-satellite reception under equal transmit power, reduces required LAV power by more than 85% for an 18 bps/Hz target, and, when combined with distributed LAV MIMO, extends service duration by up to 9× while reducing handovers. The two-timescale scheme incurs average rate loss of no more than about 2% for $\theta$9 slots while reducing beam-steering reports from approximately 1800 per hour to approximately 72 per hour (He et al., 7 May 2025).

The future-direction literature converges on several themes. These include modular LLM–RL agents, multi-agent coordination, safety-critical human-in-the-loop extensions, explainable AI, low-carbon optimization, digital twins, cross-jurisdictional policy for tokenized assets, standard benchmarks and telemetry schemas, O-RAN-native semantic control loops, quantum-driven coordination, generative governance agents, and 3D air–ground–space integration (Cai et al., 27 May 2025, Tekbıyık et al., 2 Jun 2025, Wang et al., 30 Apr 2025, Abdalla et al., 1 Jan 2026). This suggests that the field is moving from isolated UAV networking problems toward a general theory of low-altitude infrastructure in which communication, computation, sensing, regulation, and economic coordination are designed as a single operational system.