⚡ electrical engineering

Dataset and UAV Propagation Channel Modeling for LoRa in the 860 MHz ISM Band

This paper presents a publicly available dataset of LoRa IQ samples collected via an SDR-based testbed in a campus environment, which is used to derive empirical propagation channel models for UAV and pedestrian scenarios in the 860 MHz ISM band.

Original authors: Joachim Tapparel, Andreas Burg

Published 2026-04-10

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Joachim Tapparel, Andreas Burg

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to shout a secret message to a friend across a busy, noisy city park. You want to do it without shouting too loud (to save your voice) and without using a fancy megaphone (to keep costs low). This is essentially what LoRa technology does for the Internet of Things (IoT). It's a special way of sending data that is great for long distances and low power, but it's very sensitive to how the "air" carries the sound.

This paper is like a team of scientists going into that park with high-tech recording equipment to figure out exactly how the wind, trees, and buildings mess up your message. They did this not just for people walking on the ground, but also for drones flying overhead.

Here is the breakdown of their adventure, explained simply:

1. The Mission: Mapping the "Air"

The researchers wanted to know: How does a LoRa signal change as it travels from a drone or a person to a receiver?

Most people just check if the signal is "strong" or "weak" (like checking if you can hear a whisper). But this team wanted the full recording. They used a special tool called a Software-Defined Radio (SDR). Think of this as a super-sensitive microphone that doesn't just record volume; it records the exact shape of every sound wave. This allows them to see the "fingerprint" of the signal, including tiny wobbles and delays that normal devices miss.

2. The Experiment: The Park and the Drones

They set up a "test park" on the EPFL campus in Switzerland.

The Receivers (The Listeners): They placed four listening stations on different rooftops. Imagine four friends standing on different balconies, all trying to hear the same shout.
The Transmitters (The Shouters): They had two types of shouters:
1. The Pedestrian: A person walking on the ground, carrying a LoRa device.
2. The Drone (UAV): A flying robot carrying a similar device.
The Scenarios: They tested three situations:
- Drone with a clear view (LoS): The drone flies high with no buildings blocking the path.
- Drone with obstacles (NLoS): The drone flies behind buildings or trees.
- Pedestrian with obstacles (NLoS): The person walks through the busy, cluttered campus.

They collected over 75,000 messages (frames) and made this data public so other scientists can use it. It's like publishing a massive library of recordings so everyone can study how sound travels in a city.

3. The Discovery: How the Signal Fades

When you shout across a park, your voice gets quieter the further away you are. In radio terms, this is called Path Loss.

The Rule of Thumb: They found that for drones, the signal gets weaker in a predictable way, similar to how sound fades in a city.
The "Noise" Factor: But it's not just about getting quieter. The signal also gets "jittery."
- Small-Scale Jitter: Imagine the signal is a smooth wave, but the wind makes it ripple instantly. This happens fast and is caused by things like reflections off a nearby window. The researchers found that when you are on the ground (pedestrian), there is more of this jitter because there are more things to bounce the signal off.
- Large-Scale Drift: This is the slow, steady change in signal strength as you move further away.

4. The Big Insight: Drones vs. People

The most interesting part of the paper is the difference between the drone and the person.

The Pedestrian: When walking on the ground, the signal is chaotic. It bounces off walls, cars, and people constantly. It's like shouting in a canyon full of echoes. The signal strength changes wildly and quickly.
The Drone: When the drone flies high, the signal is much smoother. It's like shouting from a hilltop. The signal doesn't change as wildly from one step to the next. The "jitter" is more consistent over longer distances.

The Analogy:
Think of the signal as a drunk person walking home.

Pedestrian Scenario: The drunk person is walking through a crowded bar. They bump into tables, chairs, and people. Their path is zig-zaggy and unpredictable every second.
Drone Scenario: The drunk person is walking down a long, empty highway. They still stumble a bit, but their path is much straighter and more predictable over a long distance.

5. The Result: A New "Map" for Engineers

The team didn't just collect data; they built a mathematical map (a model) based on their findings.

This map tells engineers exactly how to expect a signal to behave if they are designing a network for drones.
It accounts for the fact that a drone's signal behaves differently than a person's signal.
It also helps engineers design "smart" receivers that can predict and fix these signal wobbles, making the network more reliable.

Why Does This Matter?

As we start putting sensors on everything—from delivery drones to smart streetlights—we need to know how they talk to each other. If we use old maps (models designed for cell phones or ground-level devices) to plan a network for drones, it might fail.

This paper provides the first detailed "weather report" for LoRa signals specifically for drones and ground users. It gives engineers the tools to build better, more reliable networks for the future of flying robots and smart cities.

In short: They recorded thousands of radio messages, figured out exactly how buildings and height change the signal, and wrote a new rulebook for how to build better networks for the sky and the street.

1. Problem Statement

Low-Power Wide-Area Networks (LPWANs), particularly LoRa, are critical for the Internet of Things (IoT) due to their long-range capabilities and low power consumption. However, as IoT networks densify, accurate channel modeling becomes essential for evaluating network performance (coverage, capacity, interference).

The Gap: Existing channel models for cellular technologies do not adequately address LoRa's specific characteristics: long transmission times, narrow bandwidths (~250 kHz), and sensitivity to channel variations.
The Specific Challenge: There is a lack of empirical data and models for Unmanned Aerial Vehicles (UAVs) operating at different altitudes compared to pedestrian-level devices. UAV channels often differ significantly due to Line-of-Sight (LoS) conditions and different scattering environments, yet few datasets capture the raw signal details (IQ samples) required to model these variations accurately.
Limitations of Current Data: Commercial off-the-shelf devices typically only provide Received Signal Strength Indicator (RSSI) values (one per frame), which lack the temporal resolution and phase information necessary for advanced receiver algorithm testing (e.g., interference cancellation, multi-user detection).

2. Methodology

The authors conducted a comprehensive measurement campaign on the EPFL campus in Lausanne, Switzerland, using a custom Software-Defined Radio (SDR) testbed.

Hardware Setup:
- Transmitters: Two mobile transmitters were used:
  1. Pedestrian: Raspberry Pi 5 + Semtech SX1276 module.
  2. UAV: Lightweight HTCC-AB02S board + SX1262 transceiver.
- Receivers: Four distributed gateways (fixed on campus roofs) consisting of Raspberry Pi 5, NI USRP-2920 SDR, and Quectel L80 GPS for precise timestamping and synchronization.
Scenarios: Measurements were taken in three distinct scenarios:
1. UAV Line-of-Sight (LoS).
2. UAV Non-Line-of-Sight (NLoS).
3. Pedestrian NLoS.
Data Collection:
- Over 75,000 LoRa frames were transmitted (approx. 17,000 per scenario across all receivers).
- Parameters included Spreading Factors (SF) 7 and 10, Bandwidths of 125/250 kHz, and 862.5 MHz carrier frequency.
- Raw IQ Samples: Unlike standard RSSI logs, the system captured raw In-phase/Quadrature (IQ) samples of valid demodulated frames in SigMF format, annotated with GPS position, velocity, and synchronization data.

3. Key Contributions

Open-Source Dataset: The creation and publication of the first dataset containing raw IQ samples from outdoor LoRa transmissions in UAV and pedestrian scenarios. This enables the evaluation of advanced receiver algorithms beyond simple RSSI-based modeling.
Empirical Channel Modeling: Development of a propagation model that captures spatial correlation of large-scale fading for distributed receivers, specifically tailored for UAV and pedestrian heights.
Differentiation of Fading Effects: A rigorous separation of small-scale fading (fast, uncorrelated variations) and large-scale fading (slow, spatially correlated variations) to provide a more accurate statistical representation of the channel.

4. Results and Findings

A. Log-Distance Path Loss Modeling

The authors fitted the Signal-to-Noise Ratio (SNR) to a standard log-distance path loss model:
$\rho(d) = \rho_0 - 10\gamma \log_{10}\left(\frac{d}{d_0}\right) + Z$

Path Loss Exponents ( $\gamma$ ):
- UAV LoS: 2.79 (Lower loss due to direct path).
- UAV NLoS: 3.62.
- Pedestrian NLoS: 3.25.
- Observation: The NLoS exponents align with urban environment models (e.g., Hata-Okumura), though the pedestrian scenario showed higher variance due to increased scattering at ground level.
Shadowing Distribution: The deviation from the path loss model ( $Z$ ) was found to follow a Normal distribution for all scenarios.

B. Spatial Correlation and Fading Decomposition

The study decomposed the total fading ( $Z$ ) into small-scale ( $X_r$ ) and large-scale ( $Y_r$ ) components:
$Z_r = X_r + Y_r$

Small-Scale Fading ( $X_r$ ): Modeled as $N(0, \sigma_x^2)$ . The standard deviation $\sigma_x$ was higher in NLoS scenarios (more multipath) and consistent across receivers within a scenario.
Large-Scale Fading ( $Y_r$ ): Modeled as $N(0, \sigma_y^2)$ $N (0, σ_{y}^{2})$ with spatial correlation.
- An Autoregressive (AR) model of order 1 was used to describe how fading correlates over distance:
  $Y_r[(n+1)\Delta d] = \phi Y_r[n\Delta d] + \epsilon$
- Key Finding: UAV scenarios exhibited higher spatial correlation ( $\phi \approx 0.97$ for LoS) compared to pedestrian scenarios ( $\phi \approx 0.75$ ). This indicates that UAV channel variations occur over longer distances, whereas pedestrian channels fluctuate more rapidly due to local obstacles.

C. Model Validation

The proposed model (Multivariate Normal Distribution with a specific covariance matrix) was validated against experimental residuals.

The residuals were found to be uncorrelated (white noise) and Gaussian distributed, confirming that the autoregressive model effectively captures the spatial correlation of the large-scale fading.

5. Significance

Advancement in IoT Research: This work moves beyond simple link-budget analysis by providing the raw data (IQ samples) necessary to test next-generation receiver algorithms like interference cancellation and joint processing in Cloud Radio Access Networks (C-RAN).
UAV-Specific Insights: It highlights that UAV communication channels are statistically distinct from ground-level IoT, particularly regarding spatial correlation. This is crucial for designing reliable telemetry links for drones.
Reproducibility: By releasing the dataset and the derived empirical parameters, the authors enable the community to simulate LoRa networks with high fidelity, accounting for realistic spatial fading correlations rather than assuming independent fading across receivers.