A Large-Scale LoRa Measurement Campaign in Urban and Suburban Environments

Published: 14 April 2026| Version 1 | DOI: 10.17632/79vh7wnwhg.1
Contributors:
, Fábio Mendonça

Description

This dataset supports a large-scale LoRa measurement campaign conducted at 868 MHz across urban and suburban environments in the city of Funchal, Madeira Island, Portugal. The research hypothesis is that inconsistencies in measurement methodology, particularly the use of mobile platforms, uncalibrated hardware, and insufficient sampling per location, introduce systematic bias in path-loss estimation for LoRa-based Internet of Things deployments, and that a fixed-position, calibrated, multi-sample approach yields more accurate and reproducible path-loss characterization. The dataset contains received signal strength indicator (RSSI), signal-to-noise ratio (SNR), and estimated effective signal power (ESP) values for both uplink and downlink transmissions, collected at 1,203 fixed locations across an area of approximately 8 km², from February to November 2025. Three custom gateways, designated Gateway A, Gateway B, and Gateway C, operating at carrier frequencies of 868.1 MHz, 868.3 MHz, and 868.5 MHz respectively, received transmissions from four end devices arranged in a spatial diversity structure. Eight samples were acquired per measurement location, resulting in a total of 28,872 uplink packets transmitted. Each record includes GPS-verified geographic coordinates, timestamp, and per-packet RSSI and SNR values for each gateway link. The data reveal that small-scale fading induces signal fluctuations of up to 35 dB, with per-location standard deviations reaching approximately 12 dB, underscoring the need for multiple samples when estimating path loss. Uplink and downlink path-loss values were found to be statistically similar, with a mean difference of 0.9 dB and a standard deviation of 1.5 dB. Near the receiver sensitivity limit, packet loss reduces the number of available samples and degrades path-loss estimation accuracy; an order-statistics-based correction technique is described in the companion paper to address this effect. The data also demonstrate that conventional single-slope log-distance path-loss models achieve a root mean square error of approximately 12 dB across the full measurement area, indicating that more refined modeling approaches incorporating environmental parameters are necessary for accurate propagation prediction in complex urban scenarios.

Files

Steps to reproduce

The measurement campaign was conducted using custom-developed hardware based on the Adafruit Feather 32u4 RFM95 platform, which integrates an ATmega32u4 microcontroller and a Semtech SX1276 LoRa transceiver. Firmware was implemented using the Arduino IDE with the RadioHead library. All radios were configured with spreading factor 12, a bandwidth of 125 kHz, and a coding rate of 4/5, operating in the European ISM band at 868 MHz with a transmit power of 14 dBm. Omnidirectional antennas with a gain of 1 dBi were used at both transmitters and receivers. Prior to the campaign, RSSI calibration was performed on all devices following the procedure described in the companion publication to correct for hardware-induced systematic offsets. Three gateways were deployed at fixed locations representing distinct environmental conditions: Gateway A on the terrace of the University of Madeira at 27 m antenna height and approximately 173 m above sea level; Gateway B at a suburban residential property at 7 m antenna height and approximately 198 m above sea level; and Gateway C on a building rooftop at 20 m antenna height and approximately 90 m above sea level. Each gateway was powered by a lithium battery supplemented by a 1 W solar panel. The transmitter system consisted of four end devices arranged in a square grid with 1.5-wavelength spacing between adjacent devices, mounted on a tripod to maintain a stable antenna height of 3 m above ground. The four end devices were connected via a USB hub to a Raspberry Pi equipped with a display, used to control data acquisition and record measurements. Power was supplied by a pack of six 3.7 V, 5 Ah lithium batteries. At each fixed measurement location, each end device transmitted one packet per gateway at 20-second intervals, with successive transmissions on the same frequency separated by approximately 1.5 minutes to comply with duty-cycle regulations. After four packets were transmitted, the operator displaced the transmitter structure by approximately one to two wavelengths and repeated the procedure, yielding eight samples per location per gateway link. Upon receiving a packet, each gateway extracted the uplink RSSI and SNR and transmitted a reply containing this information; the end device then extracted the downlink RSSI and SNR. A Python 3 script running on the Raspberry Pi recorded all RSSI and SNR values for both link directions, together with GPS coordinates acquired using a Digilent Pmod GPS module and a timestamp generated at the start of each location's acquisition. Geographic coordinates were validated and manually corrected where GPS errors were identified, particularly in dense urban areas where multipath propagation degraded positioning accuracy.

Institutions

Categories

Electrical Engineering, Telecommunication

Funders

Licence