Source-linked AI summary
Dynamic Routing for Flying Ad Hoc Networks
S. Rosati, K. Kruzelecki, G. Heitz, D. Floreano, B. Rimoldi
TL;DR
FANETs need routing that can track topology changes caused by high UAV mobility, because existing MANET-oriented protocols can fail. The paper compares OLSR and GPS-informed P-OLSR using field experiments and MAC-layer emulations, finding that P-OLSR substantially improves routing performance under frequent topology changes.
Problem
High UAV mobility causes rapid FANET topology changes, while MANET routing protocols can fail to track the evolving topology.
Method
The paper compares OLSR with P-OLSR through field experiments using two UAVs and a ground node, plus MAC-layer emulations for larger networks.
Results
P-OLSR cuts outage time by at least 85% across tested configurations and achieves more stable goodput than OLSR.
Takeaways & Limitations
P-OLSR follows topology changes without interruptions in the reported experiments and emulations, unlike OLSR.
Abstract
from arXiv · showhide
This paper reports experimental results on self-organizing wireless networks carried by small flying robots. Flying ad hoc networks (FANETs) composed of small unmanned aerial vehicles (UAVs) are flexible, inexpensive and fast to deploy. This makes them a very attractive technology for many civilian and military applications. Due to the high mobility of the nodes, maintaining a communication link between the UAVs is a challenging task. The topology of these networks is more dynamic than that of typical mobile ad hoc networks (MANETs) and of typical vehicle ad hoc networks (VANETs). As a consequence, the existing routing protocols designed for MANETs partly fail in tracking network topology changes. In this work, we compare two different routing algorithms for ad hoc networks: optimized link-state routing (OLSR), and predictive-OLSR (P-OLSR). The latter is an OLSR extension that we designed for FANETs; it takes advantage of the GPS information available on board. To the best of our knowledge, P-OLSR is currently the only FANET-specific routing technique that has an available Linux implementation. We present results obtained by both Media Access Control (MAC) layer emulations and real-world experiments. In the experiments, we used a testbed composed of two autonomous fixed-wing UAVs and a node on the ground. Our experiments evaluate the link performance and the communication range, as well as the routing performance. Our emulation and experimental results show that P-OLSR significantly outperforms OLSR in routing in the presence of frequent network topology changes.
I. INTRODUCTION
FANETs offer rapidly deployable, self-managed connectivity, but their highly mobile topology challenges routing protocols designed for less dynamic networks. The paper compares OLSR with the GPS-informed P-OLSR extension through experiments and emulations.
- Motivation: FANETs can provide rapidly deployable, self-managed ad hoc Wi-Fi connectivity when ordinary infrastructure is unavailable.The paper describes applications including connecting and coordinating ground rescue teams after calamitous events.
- Motivation: Low-cost embedded computers and Wi-Fi interfaces have enabled inexpensive flying ad hoc networks, but networking remains challenging.
- Routing challenge: FANET topologies change more frequently than MANET and VANET topologies, making routing a crucial task.
- Routing challenge: Existing MANET routing algorithms, including BABEL and OLSR, fail to follow rapidly evolving FANET topology.
- Approach: P-OLSR extends OLSR for FANETs by using GPS information, while the paper evaluates it against OLSR in field experiments and larger-network emulations.The field test uses two autonomous fixed-wing UAVs and a ground station; the emulations consider 19 UAVs.
II. ROUTING FOR FLYING AD HOC NETWORKS
The paper contrasts infrastructure-dependent and ad hoc UAV networking architectures, then explains why FANET routing requires rapid topology adaptation and link-quality-aware selection. OLSR uses ETX, but its estimation can trade responsiveness against stability.
- Network architectures: Direct-link and satellite architectures simplify routing but require long-range links and can inefficiently route UAV-to-UAV traffic through control infrastructure.
- Network architectures: Cellular architectures avoid a single control center but limit operations to cellular coverage and require functioning infrastructure during deployment.
- Network architectures: Ad hoc FANETs lack central infrastructure, enabling rapid deployment and robustness against isolated attacks or node failures.
- Routing challenge: Rapid and erratic UAV movement requires routing procedures that automatically update tables as FANET topology changes.
- Link-quality estimation: OLSR selects routes using the smallest ETX, which estimates transmissions needed for delivery from forward and reverse receiving ratios.
- Link-quality estimation: The Hello Interval controls broadcast frequency, while link-quality aging balances receiving-ratio stability against reaction speed.
B. Speed-Weighted ETX
The paper modifies ETX for rapidly changing FANET links by incorporating UAV position, direction, and relative motion. The resulting speed-weighted metric favors links between approaching nodes and uses GPS-derived velocity estimates to anticipate link-quality changes.
- Motivation: Rapid topology changes and delayed ETX reactions can make wireless links break before routing adapts.The UAVs cruise at around 12 meters per second, and exponential moving averages introduce non-negligible detection delays.
- Metric design: The modified ETX metric weights each hop using the relative speed between neighboring UAVs.The weighting accounts for the relative motion of nodes i and j rather than relying only on conventional link-quality terms.
- Metric design: Approaching nodes receive preference over separating nodes when their other link-quality values are equal.Negative relative speed produces a weighting factor below 1, while positive relative speed produces a factor above 1.
- Estimation: Neighbor positions are distributed through added GPS-coordinate fields in Hello messages, enabling speed-weighted ETX computation.Instantaneous relative velocity is estimated from successive Hello-message arrival times and corresponding inter-node distances, then smoothed with an exponential moving average.
- Parameterization: P-OLSR parameters β and γ can be adjusted to the UAV cruising speed and the selected HI.These parameters control routing selection in the speed-weighted design.
- Prediction basis: Fixed-wing UAV direction helps predict near-future position because these aircraft require forward motion, minimum air-speed, and a turning radius.This prediction is used to foresee how link quality is likely to evolve.
III. IMPLEMENTATION DETAILS
P-OLSR was implemented by modifying an open-source OLSR implementation. Hello messages carry neighbor-position data so nodes can compute the modified ETX metric.
- Implementation: P-OLSR was implemented by forking the open-source OLSRd implementation of OLSR.
- Implementation: The modified Hello messages are augmented with position information.
- Metric computation: Each node uses neighbors’ positions to compute the corresponding ETX according to equation (4).
A. OLSRd with Link-Quality Extension
OLSRd uses link-quality information in Hello and Topology Control messages, while P-OLSRd adds GPS-derived coordinates and relative-speed information to support predictive routing.
- OLSRd link-quality extension: OLSRd replaces hysteresis with link-quality sensing and advertises receiving ratios φ and ρ in Hello and Topology Control messages.These ratios support ETX-based routing decisions.
- P-OLSRd modifications: P-OLSRd shares each node’s longitude, latitude, and altitude with neighbors through Hello messages.The coordinates are used to compute relative speeds between nodes.
- P-OLSRd modifications: P-OLSRd adds averaged relative speed to each neighbor block in Hello messages using the previously unused two bytes.The speed is encoded as a 16-bit fixed-point number.
- Message overhead: The modified Hello message is 8 bytes larger than the original, independently of the number of nodes.The paper describes this additional size as negligible for medium and large networks.
- P-OLSRd modifications: P-OLSRd modifies TC messages only by using two reserved bytes for averaged relative speed, so original and modified TC messages have the same size.The speed uses a 16-bit fixed-point format.
IV. UAV TESTBED
The testbed uses small autonomous fixed-wing eBee UAVs equipped with embedded computers, GPS access, cameras, and 802.11n radios for FANET experiments.
- UAV platform: The study focuses on fixed-wing UAVs, which require forward motion and can cover greater areas but produce rapidly changing network topologies.Their minimum speed and turning radius make ad hoc networking more challenging.
- UAV platform: The experiments use two autonomous SenseFly eBee planes made of expanded polypropylene and powered by a rear-mounted electric propeller.The eBee has an integrated autopilot, approximately 57 km/h cruising speed, and 45 minutes of autonomy.
- Onboard computing: Each eBee carries a Gumstix Overo Tide computer running customized Linux and connected to the autopilot, HD camera, GPS data source, and Wi-Fi interface.The computer accesses current GPS data through a serial connection to the autopilot.
- Wireless configuration: The Wi-Fi interface is a Linksys AE3000 802.11n dongle using MIMO, STBC, 20 MHz bandwidth in the 5 GHz band, QPSK, and one spatial stream.
V. EXPERIMENTS
The experiments characterize UAV-to-ground link performance and compare OLSR with P-OLSR routing. They use an eBee testbed, UDP datagrams, and distance-based datagram-loss analysis.
- Experimental aims: The first experiment measures end-to-end link performance and communication range between a UAV and a ground node.The second compares OLSR and P-OLSR protocols.
- Experimental setup: The UAV flew between checkpoints 450 meters apart at 75 meters height, while iperf measured link quality once per second.Each second, it sent 85 UDP datagrams totaling 1 Mbit to the ground node.
- Loss-rate measurement: Datagrams delayed more than 5 seconds were counted as lost, while out-of-order datagrams arriving sooner were accepted.This delay threshold reflects the tolerance assumed for video streaming with a 5-second playback delay.
- Loss-rate analysis: The analysis plots measured datagram loss rate against distance and fits a non-linear least-square regression function.The fitted coefficients are p1 = 8.9 and p2 = 0.025.
- Link-performance results: 250 meters marks the reliable region, where observed DLR remains below 0.2; beyond 300 meters, DLR often approaches 1.Average DLR is 0.5 at 350 meters, although sporadic good connections occur beyond 400 meters.
B. Routing Performance Assessment
Field experiments compared OLSR and P-OLSR on a three-node UAV-ground network undergoing repeated topology changes. P-OLSR reacted to broken links before service interruptions, whereas OLSR incurred delay during route switching.
- Configuration: The routing comparison used a 0.5-second Hello interval for both algorithms to balance signaling overhead against reactivity.The default 2-second OLSRd interval was considered too slow for the topology changes, while equal settings preserved fairness.
- Experimental setup: The testbed used two flying UAVs and one fixed ground destination, with the source and relay following distinct trajectories.The source traveled 600 meters west and back, while the relay flew a 30-meter-radius circle; both UAVs flew about 75 meters above ground.
- Results: Fig. 9 reports the evolution of the average DLR over 10 runs, comparing MAC-layer emulations with field experiments.Dashed lines denote emulations, while solid lines denote field experiments.
- Results: OLSR exhibited DLR peaks when routing switched from a direct link to a two-hop path.The peaks occurred because OLSR took several seconds to detect a broken wireless direct link, interrupting service.
- Results: P-OLSR reacted promptly to topology changes by predicting them before the previous link broke.Its remaining DLR peaks were attributed to wireless-channel fading rather than incorrect routing.
VI. MAC-LAYER EMULATIONS WITH LARGER NETWORKS
The study used a Linux-based emulation platform to assess P-OLSR in medium and large FANETs without relying exclusively on costly field experiments. The platform combines real routing software with emulated MAC and physical layers driven by UAV positions.
- Motivation: The platform enables routing analysis in medium and large FANETs because many-UAV field experiments are expensive.It integrates the testbed aspects into a network emulation environment.
- Platform architecture: The platform creates one Linux container for each network node and connects the containers through EMANE, a MAC-layer real-time emulator.EMANE is an open-source framework developed primarily by the Naval Research Laboratory.
- Platform architecture: MAC and physical layers are emulated, while the remaining layers use the real Linux software implementations.This preserves the routing software used by the testbed while modeling lower-layer network behavior.
- Mobility and propagation: The emulator imports UAV positions from log files and uses them to compute pathloss for each wireless link.The logs can come from real flights or a flight simulator reproducing realistic conditions.
B. Emulation Results
Larger-network emulations evaluated P-OLSR across network size, Hello intervals, and parameter settings. P-OLSR consistently reduced outage time and produced more stable goodput than OLSR, while parameter choices affected the best configuration.
- Experimental design: The larger-network campaign used 19 moving UAVs, with one UAV scanning a 1200-square-meter area in a 380-second trajectory.The campaign examined the Hello interval, link-quality aging α, and P-OLSR parameters β and γ.
- Metrics: Outage time was defined by intervals where the DLR exceeded 0.2 and was summed across each run.Each configuration was repeated 10 times for averaging.
- Outage results: At least 85%: P-OLSR reduced outage time by at least 85% across all tested configurations.For the best configurations, reductions were about 95%, 92%, and 90% at Hello intervals of 0.5, 1, and 2 seconds, respectively.
- Parameter effects: When the Hello interval was short, decreasing γ was preferable; when it was long, increasing β gave the speed term greater weight.With infrequent Hello messages, faster aging of speed estimates and greater speed weighting helped compensate for slower link-quality tracking.
- Parameter trade-offs: Shorter Hello intervals improved performance, but halving the interval from 1 to 0.5 seconds might not justify the additional signaling traffic.P-OLSR maintained topology tracking with lower Hello-message rates by using GPS information.
- Parameter trade-offs: 80%: P-OLSR with a 2-second Hello interval reduced outage time by 80% relative to OLSR with a 0.5-second interval.The P-OLSR configuration generated about one quarter as many Hello messages.
- Goodput results: 0.95 Mbit/s: P-OLSR’s average goodput exceeded OLSR’s 0.83 Mbit/s average in the reported emulation.P-OLSR never fell below 0.6 Mbit/s and was above 0.8 Mbit/s most of the time, whereas OLSR sometimes reached zero.
VII. CONCLUSION
The paper compares P-OLSR with OLSR for small fixed-wing UAV FANETs and reports that P-OLSR follows topology changes without interruptions. Emulation results examine goodput across different protocol parameters and Hello intervals.
- P-OLSR and OLSR were compared in a FANET composed of small fixed-wing UAVs.
- Average goodput was evaluated for OLSR and P-OLSR across different Hello intervals and parameter settings.The emulations varied HI, α, β, and γ in separate evaluations.
- P-OLSR uses GPS information to predict wireless-link quality as network topology evolves.
- P-OLSR routing followed topology changes without interruptions, unlike OLSR.