Seventeen Minutes to Move — Lessons from Nobeoka's Wifi HaLow field trial

October 2025

Nobeoka City, Miyazaki (Japan), operates under fast-tsunami assumptions and an aging population profile. The study documents weak indoor audibility for municipal disaster radio, low uptake of individual evacuation plans among assistance-eligible residents, and limited remote situational awareness for staff.¹

A neighborhood-scale pilot using Wifi HaLow (IEEE 802.11ah) tested whether communities could organize and complete evacuation inside a fixed time budget using local voice notices, periodic camera imaging, passive GPS tags, and digital drills with route simulation and “My Timeline” planning.

Deployment, summarized

Two 11ah access points covered the Kitamidorigaoka drill area. On this link ran IP voice notice terminals at community facilities, two scheduled-image cameras, and compact GPS tags carried by assistance-eligible residents. Training included route simulation and timeline-making workshops; unannounced guidance exercises used the neighborhood voice path.

Design targets emphasized intelligible local voice, 1-minute camera cadence, kilometer-class reach for tag reception, and stable operation across the full window. Concurrency was scoped to 2 cameras + 1 voice terminal + 40 GPS tags.

Measurement approach

Two evaluation levels:

1. Pilot — confirm 17-minute evacuation; increase understanding and creation of “My Timeline”; verify residents can set safer routes using simulation. Evidence combined app location logs, camera/voice telemetry, and resident surveys.

2. City-scale modeling — estimate reachable population within simulated flood zones from candidate 11ah placements across three neighborhoods.

Field results

Evacuation performance

Two drills were run. Drill 1: 33/34 (97%) reported on-time evacuation; the outlier was tied to a farther home location and different shelter site. Drill 2: 6/6 groups (100%) on time. The requirement “evacuation within 17 minutes can be completed” is marked achieved.

Communications over Wifi HaLow (802.11ah)

Voice. Audio stayed intelligible with no dropouts during the 17-minute window. To respect an effective ~10% duty ratio, messages were ≤10 s with ≥60 s spacing and a lower-bitrate codec.

Cameras. Two cameras transmitted images every minute. Imagery supported checks of run-up and evacuation flows; higher resolution or more cameras will require airtime and access-point re-budgeting.

Location tags. In city blocks near Kitamidorigaoka, tag reception reached ~490 m; access-point beacons were heard ~600 m. The ~1 km aspiration for tag reception was not met with this antenna/placement mix.

Simultaneous load. With two parallel Wifi HaLow networks—1 MHz (location) and 2 MHz (camera+voice)—the target mix (2 cams + 1 voice + 40 tags) held. A single 1 MHz network carrying all streams did not meet the target.

Planning and preparedness

Workshops tied to the drills moved intent into action. District-wide among assistance-eligible residents, 12/30 (40%) produced individual evacuation plans, above the 14.3% benchmark. In the drill’s south-side subarea, 6/10 completed plans during the pilot. Among drill participants, 62.5% reported strong understanding of “My Timeline”, and 60% created an individual plan.

Coverage modeling

Modeled placements across three neighborhoods yield 95.96% average population coverage within simulated flood zones: Kitamidorigaoka 100%, Ikata 96.6%, Kitaura 91.3%.

What this implies for node builders

The study substantiates a neighborhood communications layer that remains useful in daily life and holds under disaster constraints.

• Voice alerts: Short, periodic announcements ride cleanly within 11ah duty budgets. Treat voice as low-duty broadcast, sized to the evacuation window and repeated at minute intervals for indoor audibility.
• Imaging: One-minute stills from a small camera set provided actionable context. Image size and camera count should scale with access-point count and airtime budgets, not vice-versa.
  Periodic stills are easier to schedule than continuous video under duty limits while preserving situational value.
• Tags vs. AP beacons: Embedded-antenna tags under-reached AP beacons by ~100 m in dense blocks. Plan for placement and antenna options if 1 km receive range is mandatory; design for building-shadow pockets.
• Concurrency planning: Splitting traffic—e.g., a narrow channel for location and a wider one for media/voice—matched the target mix. A single narrow channel did not. Dimension networks to the load profile observed in drills, not nominal device counts.
• Behavior change matters: The communications fabric increased plan creation and route familiarity, which likely lowered last-minute coordination costs during drills.

Constraints and open work

Duty ratios cap continuous audio; the tested pattern of ≤10-second clips spaced a minute apart was reliable. Camera throughput met needs at low image modes but will trade off with coverage unless access-point density rises. Tag reception fell short of 1 km in city blocks and showed building-shadow nulls; antenna and placement choices are first-order factors to revisit in follow-on deployments.

Why this study is useful to practitioners

Baseline conditions included poor indoor audibility for municipal radio, ~1% plan coverage among ~4,000 assistance-eligible residents, and no remote real-time view for staff. Against that backdrop, the pilot delivered near-universal on-time evacuation in drills, a working concurrency recipe for neighborhood-scale nodes, and measurable gains in plan-making and route understanding. For builders of local alerting, sensing, and guidance nodes, the findings provide grounded parameters for airtime budgeting, antenna selection, and network partitioning.