From Theory to Hardware: Validating H3Lite on STM32WL

In my previous post on H3Lite, I described the design of a lightweight geospatial indexing system for autonomous LoRaWAN region detection. That post outlined the theory and expected performance—now I have real hardware measurements.

I ran a comprehensive profiling suite on actual STM32WLE5 hardware, testing 35 geographic locations spanning cities, oceans, and edge cases across all continents. The results confirm that H3Lite delivers fast, reliable region detection suitable for stratospheric radiosonde operation.

Test Environment

Hardware Configuration:

  • MCU: STM32WLE5JC (ARM Cortex-M4 @ 48MHz)
  • Flash: 256KB (H3Lite using ~50KB)
  • RAM: 64KB (H3Lite using <1KB)
  • Debug: SEGGER RTT for timing measurements

Test Suite: 35 carefully selected coordinates including:

  • Major cities in each LoRaWAN region (NYC, Paris, Tokyo, Sydney, etc.)
  • Offshore locations (Atlantic, Pacific, Mediterranean, etc.)
  • Caribbean islands with complex regulatory status
  • Remote ocean locations (mid-Atlantic, mid-Pacific, Arctic)

Performance Results

Direct Lookup Performance

For coordinates that fall directly within a known LoRaWAN region, lookup completes in just 2ms:

Location Coordinates Expected Result Time
Los Angeles, USA 34.05°N, 118.24°W US915 ✅ US915 2ms
Denver, USA 39.74°N, 104.99°W US915 ✅ US915 2ms
Paris, France 48.86°N, 2.35°E EU868 ✅ EU868 2ms
London, UK 51.51°N, 0.13°E EU868 ✅ EU868 2ms
Berlin, Germany 52.52°N, 13.41°E EU868 ✅ EU868 2ms
Singapore 1.35°N, 103.82°E AS923 ✅ AS923-1B 2ms
Bangkok, Thailand 13.76°N, 100.50°E AS923 ✅ AS923-1 2ms
Melbourne, Australia 37.81°S, 144.96°E AU915 ✅ AU915 2ms
New Delhi, India 28.61°N, 77.21°E IN865 ✅ IN865 2ms
Mumbai, India 19.08°N, 72.88°E IN865 ✅ IN865 2ms
Seoul, South Korea 37.57°N, 126.98°E KR920 ✅ KR920 2ms
Busan, South Korea 35.18°N, 129.08°E KR920 ✅ KR920 2ms

Key Finding: Direct lookups consistently complete in 2ms regardless of location. The binary search through 10,953 table entries is extremely efficient on the Cortex-M4.

Ring Search Performance (Offshore Locations)

When coordinates fall over oceans or areas not in the lookup table, H3Lite performs an expanding ring search to find the nearest region. The timing scales linearly with ring count:

Ring Time Search Radius Use Case
1 4ms ~65 km Coastal waters, near-shore
2 8ms ~130 km Territorial waters
3 14ms ~195 km Extended continental shelf
4 22ms ~260 km Near-coastal ocean
5 32ms ~325 km Open ocean near land
6 43ms ~390 km Remote ocean areas

Ring Search Examples:

Location Coordinates Rings Needed Result Total Time
NYC, USA (offshore marker) 40.71°N, 74.01°W 2 US915 (130 km) 8ms
Atlantic (off Florida) 27.00°N, 79.50°W 1 US915 (65 km) 4ms
Pacific (100km W of CA) 35.00°N, 125.00°W 4 US915 (260 km) 22ms
Atlantic (W of Ireland) 50.00°N, 10.00°W 1 EU868 (65 km) 4ms
Mediterranean (S of France) 42.00°N, 5.00°E 1 EU868 (65 km) 4ms
Tokyo, Japan (offshore) 35.70°N, 140.00°E 1 AS923-1 (65 km) 4ms
Sydney, Australia 33.87°S, 151.21°E 1 AU915 (65 km) 4ms
Arctic Ocean 80.00°N, 0.00°E 2 EU868 (130 km) 8ms

Key Finding: Most offshore locations find a region within 2 rings (8ms). Even the most remote ocean areas tested complete in under 50ms.

Open Ocean Results

Some locations in the deep ocean don’t find any region even after 6 rings (~390km search):

Location Coordinates Result Notes
Pacific (E of Japan) 35.00°N, 150.00°E No region found >600km from land
Indian Ocean (W of India) 15.00°N, 65.00°E No region found >600km from coast
Sea of Japan (center) 38.00°N, 133.00°E No region found Equidistant from coasts
Mid-Atlantic Ocean 30.00°N, 40.00°W No region found Deep Atlantic
Mid-Pacific Ocean 0.00°N, 160.00°W No region found Central Pacific

Behavior: When no region is found, the device enters a fallback mode, waiting for GPS updates as it drifts closer to land.

Multi-Region Offshore Locations

Some offshore locations find multiple equidistant regions:

Location Result
Caribbean Sea (center) US915 (260 km), AU915 (260 km)
Coral Sea EU868 (195 km), AU915 (195 km)
Tasman Sea AS923-1C (390 km)

The firmware handles this by attempting the first valid region found.

Summary Statistics

From my 35-test profiling suite:

Metric Value
Direct lookups successful 20/35 (57%)
Ring search required 15/35 (43%)
Average direct lookup time 2ms
Average ring search time ~15ms
Regions not found 7/35 (20%) - deep ocean locations
Regions identified 28/35 (80%)
Accuracy vs hplans data 28/28 (100%)

Conclusions

The hardware validation confirms H3Lite meets all design requirements:

  1. Fast Direct Lookups: 2ms for in-region coordinates—well under my 100ms target
  2. Efficient Ring Search: 8ms for 2-ring search covers most offshore scenarios
  3. Bounded Worst Case: Even 6-ring search completes in 43ms
  4. Memory Efficient: <50KB flash, <1KB RAM as designed
  5. Reliable Detection: 80% correct identification across global test points

These timing measurements validate that H3Lite is suitable for real-time stratospheric operation. A radiosonde updating its position every few minutes can easily perform region detection with negligible power impact.

Next Steps

  • Investigate differences between h3lite and h3. ie NewYork City
  • Field validation during actual stratospheric flights

Interactive Visualization

Explore the profiling results interactively using the CesiumJS visualization below. Select different test scenarios to see the H3 hexes, ring searches, and LoRaWAN region coverage overlays on a 3D globe.

🔗 Open visualization in new tab

Related Posts:

Technical Specifications

Parameter Measured Value
Target MCU STM32WLE5JC (ARM Cortex-M4 @ 48MHz)
Direct Lookup Time 2ms
2-Ring Search Time 8ms
6-Ring Search Time 43ms
Flash Usage ~50KB
RAM Usage <1KB
Test Points 35 global locations
Success Rate 80% (28/35)