Online and Decentralized Spreading Factor (SF) Allocation in Multi-Gateway LoRaWAN Networks

The Research Institute in Computer Science, Mathematics, Automation, and Signal Processing (IRIMAS) is a research team (EA 7499) at the University of Haute-Alsace (UHA).

This interdisciplinary institute brings together all research activities related to the fields of mathematics, computer science, electronics, electrical engineering, automation, and signal and image processing at the University of Haute-Alsace.

IRIMAS is affiliated with the Doctoral School 269: Mathematics, Information Sciences, and Engineering (MSII) of the University of Haute-Alsace and the University of Strasbourg.

Environment:
The intern will be supervised by members of the RT and OMEGA teams at the IRIMAS laboratory, UHA, Mulhouse. IRIMAS provides a high-level scientific environment, hosting numerous researchers working across various fields, ranging from artificial intelligence and optimization to communication theory.

LoRaWAN networks, typical of LPWAN technologies, rely on a star topology in which end devices (sensors) communicate over the air with one or multiple gateways. Each transmission uses a Spreading Factor (SF) that determines the data rate and communication range: a higher SF increases range at the cost of a longer Time-on-Air (ToA).

In multi-gateway deployments, a packet may be received by several gateways, and the network server typically selects the copy with the highest RSSI. However, in dense deployments, this ALOHA-based architecture without coordination leads to significant packet collisions. Optimizing radio parameters (SF and transmission power) is therefore crucial to improve reliability (Packet Delivery Ratio – PDR) and fairness among nodes.

The LoRaWAN standard proposes an Adaptive Data Rate (ADR) mechanism managed by the network server, which adjusts SF and transmission power based on feedback (SNR, RSSI) from previous transmissions. However, classical ADR is centralized and reactive, which limits its responsiveness and adaptability. Several studies have shown that ADR converges slowly, as it requires many frames to stabilize. In addition, ADR relies on the analysis of the last 20 transmissions and network feedback, which is effective in stable scenarios but inefficient under rapid channel variations or mobility. In dynamic or multi-gateway environments, centralized reactive approaches may result in outdated configurations and increased packet loss.

Furthermore, recent studies highlight the importance of fairness in LoRaWAN networks. Without proper allocation, some nodes may monopolize the channel, degrading overall network performance. Research on energy fairness (e.g., max-min fairness approaches) shows that fair resource distribution can extend network lifetime.

Finally, the coexistence of multiple gateways enables cooperative strategies. For example, nodes can be grouped according to the gateway providing the strongest signal, or SF assignments can be coordinated to mitigate local congestion.

This internship aims to study and develop online and decentralized SF allocation algorithms to maximize PDR and fairness among nodes in multi-gateway LoRaWAN networks.

The work will include several components:

1. Analysis of ADR-Based Approaches

• Review existing ADR-based methods.
• Highlight their centralized and reactive nature.
• Analyze convergence time and inefficiency under rapid channel variations.
• Identify limitations in multi-gateway contexts.

2. Investigation of Online Decentralized Approaches

Explore approaches in which each node (or gateway) adjusts its SF locally based on its own observations (RSSI, SNR, transmission history, success/failure rate). In such blind or distributed approaches, no central server is required for parameter adaptation, improving responsiveness.

Relevant approaches include:

• Multi-Armed Bandit (MAB) algorithms such as EXP3, where each sensor autonomously selects and updates its transmission parameters.
• Expert-driven algorithms (e.g., EXP3.P variants).
• Lightweight heuristic-based adaptive methods.
• Hybrid statistical-learning approaches combining reactive rules and learning mechanisms.

3. Development of an Online SF Allocation Strategy. Two main research directions will be explored:

• Adaptive Heuristics: Simple rules where nodes change SF based on quality thresholds (RSSI, success rate). For example decrease SF when RSSI exceeds a predefined threshold, or increase SF when repeated transmission failures occur.
• Online Learning Approaches: Lightweight learning models such as MAB algorithms (EXP3, EXP4),  Simplified Q-learning over SF values {7..12}, Reinforcement learning strategies that maximize local PDR.

Previous studies have shown that MAB-based approaches improve PDR while maintaining reasonable energy consumption.

4. Local vs Cooperative Strategies

Evaluate potential gains when information exchange is introduced. In local (non-cooperative) mode: Each node/gateway relies solely on its own measurements. In cooperative mode: Gateways exchange statistics (e.g., global PDR, channel load) to jointly adjust SF allocation.

For example, zone-based clustering approaches can group nodes according to their strongest gateway and assign optimized parameters locally.

The impact of cooperation on PDR and fairness will be analyzed, considering communication overhead versus performance gains.

5. Implementation and Simulation

• Implement the proposed strategies in a simulation framework, typically the NS-3 with the LoRaWAN module for realistic multi-gateway simulations.
• Evaluate performance using metrics such as PDR, fairness index, convergence time, computational complexity, and coordination overhead.

We are looking for a Master’s student in Computer Networks (or equivalent) with the following qualities:

• Strong background in computer and wireless networks.
• Programming skills (Python, C, C++).
• Strong knowledge of optimization techniques (metaheuristics) and/or machine learning methods.
• Familiarity with Linux environments.
• Experience with NS-3 is highly appreciated.
• Good writing skills.
• Good English proficiency.
• Curiosity and creativity.