Self-adaptive contention window for WSN using Q-Learning / Mohamed Abderrahim Makhloufi

Public ISBD Titre : Self-adaptive contention window for WSN using Q-Learning Type de document : document multimédia Auteurs : Mohamed Abderrahim Makhloufi, Auteur ; Khadidja Makhloufi, Auteur ; Lakhdar Kamel Oulad Djedid, Directeur de thèse Editeur : Laghouat : Université Amar Telidji - Département d'informatique Année de publication : 2026 Importance : 73 p. Accompagnement : 1 disque optique numérique (CD-ROM) Note générale : Option : Networks, systems and distributed applications Langues : Anglais (eng) Mots-clés : Wireless sensor networks  IEEE 802.15.4  MAC  Q-Learning  Contention window  Reinforcement learning  Python simulation Résumé : Wireless Sensor Networks (WSNs) are an enabling technology for the Internet of Things (IoT). Efficient channel access at the Medium Access Control (MAC) layer is a major challenge, especially in dense networks where collisions increase packet loss, delay, and energy consumption. The Binary Exponential Backoff (BEB) algorithm used in IEEE 802.15.4 doubles the Contention Window (CW) after a collision and resets it after a successful transmission. Although simple, this behaviour does not adapt well to changes in node density and traffic load. This thesis proposes QL-CW, a Q-Learning-based contention window adaptation mechanism. The problem is formulated as a Markov Decision Process (MDP), where each sensor node acts as an independent agent using only local observations: recent collision rate, channel busy ratio, and queue occupancy. The action consists of selecting a CW value from the extended set f8; 16; 32; 64; 128; 256; 512g, providing coverage from small to large contention windows. The reward function considers three objectives: transmission success or failure, energy consumption, and access delay. The resulting Q-table contains only 189 entries (27 states7 actions), requiring approximately 378 bytes when stored as 16-bit fixed-point values, making the approach suitable for resource-constrained sensor nodes. The proposed mechanism is implemented in a Python-based discrete-time simulator that reproduces IEEE 802.15.4-like CSMA/CA behaviour and is compared with BEB. The results show that, under non-saturated conditions, both mechanisms maintain a high packet delivery ratio, while QL-CW reduces collisions and energy consumption. Under saturated conditions, QL-CW clearly outperforms BEB: collisions are reduced by about 38–57%, the Packet Delivery Ratio improves by up to 20:9 percentage points, average delay is reduced by 33–54%, and average per-node energy consumption decreases by 30–44%. These results show that a lightweight and fully distributed tabular Q-Learning approach can effectively improve contention behaviour in IEEE 802.15.4-based WSNs while respecting the memory and processing constraints of sensor nodes. note de thèses : Mémoire de master en informatique Self-adaptive contention window for WSN using Q-Learning [document multimédia] / Mohamed Abderrahim Makhloufi, Auteur ; Khadidja Makhloufi, Auteur ; Lakhdar Kamel Oulad Djedid, Directeur de thèse . - Laghouat : Université Amar Telidji - Département d'informatique, 2026 . - 73 p. + 1 disque optique numérique (CD-ROM). Option : Networks, systems and distributed applications Langues : Anglais (eng) Mots-clés : Wireless sensor networks  IEEE 802.15.4  MAC  Q-Learning  Contention window  Reinforcement learning  Python simulation Résumé : Wireless Sensor Networks (WSNs) are an enabling technology for the Internet of Things (IoT). Efficient channel access at the Medium Access Control (MAC) layer is a major challenge, especially in dense networks where collisions increase packet loss, delay, and energy consumption. The Binary Exponential Backoff (BEB) algorithm used in IEEE 802.15.4 doubles the Contention Window (CW) after a collision and resets it after a successful transmission. Although simple, this behaviour does not adapt well to changes in node density and traffic load. This thesis proposes QL-CW, a Q-Learning-based contention window adaptation mechanism. The problem is formulated as a Markov Decision Process (MDP), where each sensor node acts as an independent agent using only local observations: recent collision rate, channel busy ratio, and queue occupancy. The action consists of selecting a CW value from the extended set f8; 16; 32; 64; 128; 256; 512g, providing coverage from small to large contention windows. The reward function considers three objectives: transmission success or failure, energy consumption, and access delay. The resulting Q-table contains only 189 entries (27 states7 actions), requiring approximately 378 bytes when stored as 16-bit fixed-point values, making the approach suitable for resource-constrained sensor nodes. The proposed mechanism is implemented in a Python-based discrete-time simulator that reproduces IEEE 802.15.4-like CSMA/CA behaviour and is compared with BEB. The results show that, under non-saturated conditions, both mechanisms maintain a high packet delivery ratio, while QL-CW reduces collisions and energy consumption. Under saturated conditions, QL-CW clearly outperforms BEB: collisions are reduced by about 38–57%, the Packet Delivery Ratio improves by up to 20:9 percentage points, average delay is reduced by 33–54%, and average per-node energy consumption decreases by 30–44%. These results show that a lightweight and fully distributed tabular Q-Learning approach can effectively improve contention behaviour in IEEE 802.15.4-based WSNs while respecting the memory and processing constraints of sensor nodes. note de thèses : Mémoire de master en informatique

---

1 (1 - 1 / 1) Par page : 25 50 100 200