Source S · Destination D
UAV
Relay route in use
Link in range
Red = cumulative delivery ratio (one packet every 0.1 s; share delivered). Blue = cumulative share of time an S–D path existed. A route, once found, is used until one of its links breaks, then rediscovered after 0.1 s (an AODV-like rule).
Things to try first
- Click Paper setting (range 100 m). With 50 UAVs spread over 1 km², each UAV sees only about 1.6 neighbours on average, so S and D are almost never connected under any model. How does this compare with the paper's 91% delivery ratio?
- Raise the range to 200 m and switch between the three models. Under SREM the UAVs gather on the ellipse rim and form two relay corridors, one above and one below the axis.
- Set the orbit scale to Fixed. The gap between SREM and SRCM grows. Why? (Look at the distribution in tab ②.)
Node distribution by Monte Carlo (paper §4.2, Figs. 5–6)
Time one UAV spends in each cell over a long flight (darker = longer). Red line = S–D relay axis; dashed = reference ellipse (circle).
Paper Fig. 5 · SREM Monte Carlo trajectories
Paper Fig. 6 · Centre-distance distribution (δ 30–90°, 15–30°, 7.5–15°)
Three models compared (20 seeds, 50 UAVs, 100 s)
Delivery ratio (PDR proxy, %)
Hop count of the route used
UAV–UAV link duration (s)
UAVs within ±R/2 of the relay axis (%)
Delivery ratio per seed (cf. paper Fig. 7)
The ranking changes with communication range
The ranking changes with the number of UAVs (range 150 m)
Side by side with paper Table 4
Values here come from unit-disk links and a simplified route-maintenance rule. The NS-3 802.11a and TCP layers are not modelled, so look at the ordering and the conditions under which it holds, not the absolute numbers.
Each SREM parameter varied one at a time (range 200 m, 20 seeds; other values at defaults a = 500 m, b/a = 0.5, δ = 30–90°, scale 0.5–1.0×).
Semi-major axis a
Axis ratio b/a
Angular step δ
Orbit scale range (SREM vs SRCM)
Paper Fig. 9 · Sensitivity to semi-major axis a (PDR, EED, CT)
SREM algorithm (paper Algorithm 1)
- Place the ellipse centre O at the S–D midpoint, with the major axis along the S–D line (the relay axis).
- Each UAV draws a random angle θ and a scale s ∈ [smin, smax], and starts at (a′cos θ, b′sin θ) on the ellipse with a′ = a·s, b′ = a′·(b/a).
- It adds δ ∈ [θmin, θmax] to θ and flies in a straight line to the new point at constant linear speed.
- Every fixed number of moves it redraws the scale and moves to a different ellipse.
The cost is O(1) per UAV. Unlike SRCM, which follows curves, SREM flies straight between waypoints, which resembles real autopilot waypoint navigation.
The simulator in this demo (sim.js)
- Area
- 1000×1000 m; S(0,500) and D(1000,500) fixed
- UAVs
- 50, speed 20–40 m/s (same for all models)
- Links
- Connected if distance ≤ R (unit disk)
- Routing
- BFS shortest-hop route, kept until a link breaks, rediscovered 0.1 s later
- Delivery
- One packet every 0.1 s, delivered if a maintained route exists
- Link duration
- Mean time a UAV pair stays in range
- Reproduce
node train/experiment.js > results.json(~40 s)
There are no MAC collisions, TCP congestion control, propagation loss or queueing delay, so end-to-end delay (EED) and throughput are not computed.
Review points from today's perspective
- A 100 m range does not fit the reported results. With 50 UAVs in 1 km² and R = 100 m, the mean degree is about 1.6, well below the ~4.5 needed for a random geometric graph to connect. Under unit-disk links RWP delivers almost 0%, while the paper reports 51.7%. The effective range in NS-3 was probably larger, so the propagation model (e.g., LogDistance) and transmit power must be stated for the results to be reproducible.
- The traffic settings are inconsistent. One hundred 64-byte packets total 6.4 KB, which at 1 Mbps take about 0.05 s to send. Yet throughput is reported as ~110 KB/s (about 11 MB over 100 s). CBR is normally sent over UDP, but the traffic type is listed as TCP. The generator and measurement window need to be specified.
- "Aligned with the relay axis" conflicts with "avoids the centre". The relay axis (S–D line) passes through the ellipse centre, but SREM avoids the centre and gathers on the rim. Relay routes form along the upper and lower arcs, not on the axis. Measured as the share within ±R/2 of the axis, SREM concentrates only slightly more than RWP (tab ③).
- The conclusion flips with conditions. With range ≥ 250 m or ≥ 90 UAVs, SRCM matches or beats SREM. "SREM is better" holds only for sparse networks, so results should be shown across densities.
- No confidence intervals or tests. Only means of 10 runs are tabulated. Standard deviations, 95% confidence intervals or paired tests should be reported.
- Fairness of the comparison. SREM knows where S and D are and aligns its orbit with them; RWP and SRCM do not. Adding an ideal baseline (UAVs fixed in a line between S and D) and a path-aware baseline (e.g., a Column model along the axis) would isolate the contribution of the mobility model itself. Only RWP's speed is listed as 40 m/s, so speed conditions also need checking.
- "Midpoint = unbiased estimator of dwell position" is an approximation. At constant speed on a straight segment, dwell is spread evenly along the whole segment, so positions should be time-sampled rather than represented by a single midpoint (tab ② compares both).
- Also missing: altitude (3D), acceleration and turn limits, collision avoidance and energy (the paper acknowledges these). Releasing the code and NS-3 scripts would make this a good reproduction exercise.
Questions for class
- Compute the mean degree λ = N·πR²/A, predict the λ at which S and D become connected, and check it in tab ①.
- What happens if the SREM ellipse is laid perpendicular to the S–D axis instead? (As b/a approaches 1 it becomes a circle.)
- Routes last only about one second, yet delivery is high. Why? Predict what happens if rediscovery takes 0.5 s.
- What do the hop counts in Table 4 (SREM 11.2, RWP 21.7) mean compared with the minimum of 10 hops at a 100 m range?
Fig. 1 · Categories of FANET mobility models
Fig. 2 · Structure of SREM
Fig. 3 · Trajectories after 10 / 50 / 200 moves
Table 3 · Simulation parameters
- Tool
- NS-3, 1000×1000 m², 50 UAVs, 100 s, 10 runs
- SREM
- a = 500 m, b = 250 m
- SRCM
- Max radius 500 m
- RWP
- Pause 1 s, speed 40 m/s
- S · D
- (0, 500) · (1000, 500)
- Routing
- AODV, HELLO 100 ms
- Traffic
- CBR, TCP, 64 B, 1 Mbps (100 packets)
- Radio
- IEEE 802.11a, omnidirectional, range 100 m
- Speed
- 20–40 m/s
Table 4 · Performance by model
| Model | PDR % | EED s | Throughput KB/s | CT s | Hops |
|---|---|---|---|---|---|
| SREM | 91.3 | 0.061 | 110.3 | 23.8 | 11.2 |
| SRCM | 81.4 | 0.089 | 98.1 | 18.1 | 14.5 |
| RWP | 51.7 | 0.175 | 61.0 | 10.4 | 21.7 |
Sensitivity (§5.3): best at a = 500 m and b/a ≈ 0.4; δ makes little difference.
Fig. 7 · PDR per test
Fig. 8 · End-to-end delay per test
Fig. 4 · Analytical SRCM distribution (Eq. 4)
Source: Choe, H.; Kang, D. A Semi-Random Elliptical Movement Model for Relay Nodes in Flying Ad Hoc Networks. Telecom 2025, 6, 56. CC BY 4.0. Figures are reproduced from the original paper.
Research and teaching demo · A simplified simulator, so numbers do not match the paper directly. Use it to compare trends and experimental design.